Research Article A Location Prediction Algorithm with...

Research ArticleA Location Prediction Algorithm with Daily Routines inLocation-Based Participatory Sensing Systems

Ruiyun Yu1 Xingyou Xia1 Shiyang Liao1 and Xingwei Wang2

1Software College Northeastern University Shenyang 110819 China2College of Information Science and Engineering Northeastern University Shenyang 110819 China

Correspondence should be addressed to Ruiyun Yu yurymailneueducn

Received 21 November 2014 Accepted 11 January 2015

Academic Editor Joel Rodrigues

Copyright copy 2015 Ruiyun Yu et al This is an open access article distributed under the Creative Commons Attribution Licensewhich permits unrestricted use distribution and reproduction in any medium provided the original work is properly cited

Mobile node location predication is critical to efficient data acquisition and message forwarding in participatory sensing systemsThis paper proposes a social-relationship-based mobile node location prediction algorithm using daily routines (SMLPR) TheSMLPR algorithm models application scenarios based on geographic locations and extracts social relationships of mobile nodesfrom nodesrsquo mobility After considering the dynamism of usersrsquo behavior resulting from their daily routines the SMLPR algorithmpreliminarily predicts nodersquos mobility based on the hidden Markov model in different daily periods of time and then amends theprediction results using location information of other nodes which have strong relationship with the node Finally the UCSDWTDdataset are exploited for simulations Simulation results show that SMLPR acquires higher prediction accuracy than proposals basedon the Markov model

1 Introduction

Participatory sensing is a recent appearing sensing technol-ogy which emphasize that people participate in the sens-ing process Participatory sensing enables individuals andcommunities to gather analyze and share local knowledgeand to subsequently make intelligent decisions and alsooffer social services The earliest research of participatorysensing is executed by Srivastava et al who have proposedthe conception of urban sensing in a technical report [1] inthe year 2006 to discuss the system architecture and technicalmethods of urban sensingTheMSG (Mobile Sensing Group)laboratory at Dartmouth College also conducts research onthis area including BikeNet [2] SoundSense [3] CenceMe[4 5] MetroSense [6] and Bubble-Sensing [7]

Participatory sensing systems rely on mobile phone usersto sense and transmit data with diverse purposes in the pro-cess of monitoring or solving a particular problem Based onthe large number of users participatory sensing systems havethe potential to acquire large amounts of data from variousplaces and address large-scale location-based problems [8ndash10] A typical example of location-based participatory sensingsystems collects and records air quality measurements to

monitor the pollution of a particular location What ismore the smartphone with Internet connectivity can alsocontribute to the participatory sensing systemsrsquo growth In[11] a solution based on web services is proposed to permitthe interaction between a mobile application and the IPv6compliant WSNs scenario

In participatory sensing systems mobile devices areusually weakly connected Due to uncertainty of connectionnodes sometimes need encounter opportunities to accom-plish data communication and transmission If the location ofmobile nodes can be predicted ahead of several time slots theservice quality and efficiency of the systemwill be remarkablyimproved

In order to solve the location prediction problems humanmobility has been analyzed at different geographic scales [12ndash14] In [12] the limits of predictability provided by humansrsquomobility patterns are examined The data that is collected by50000 anonymousmobile phone users over a period of threemonths has been used to study the humansrsquomobility patternsand different entropy measures are adopted in this researchto estimate the potential predictability in human dynamicsBased on their analysis a 93 potential predictability in usermobility across the whole user base is found in their report

Hindawi Publishing CorporationInternational Journal of Distributed Sensor NetworksVolume 2015 Article ID 481705 12 pageshttpdxdoiorg1011552015481705

2 International Journal of Distributed Sensor Networks

In [13] the location data from 100000 mobile phone userscollected by tracking each personrsquos position for six monthsis investigated This research shows that individual mobilitypatterns of the test persons can be described by a single spatialprobability distribution after some corrective preprocessingThus the results of this study suggest that humans generallyfollow simple reproducible patterns

Based on the human mobility analysis different tech-niques based on the Markov model have been applied tolocation prediction problems of human individuals In [15]a set of discrete locations has been defined using the WIFIcells on a university campus Two different kinds of locationpredictors the 119896th order Markov predictor as well as aLZ-based predictor are tested in predictions to the nextlocation Based on the test results the research shows thatthe second order Markov predictor with a certain fallbackfeature performs best and provided a median accuracy of72 Literature [16] presents a similar extended Markovpredictor which takes the arrival time and residence timeinto account Specifically delay embedding is used to extractlocation sequences of a certain length from time seriesThen these sequences are directly used to predict a userrsquoslocation and the prediction result is obtained by comparingthe last observed locations to all embedded location sequen-ces

Meanwhile hidden Markov models (HMMs) are alsoconsidered to predict human mobility Literature [17] pre-sents a hybridmethod on the basis of hiddenMarkovmodelsThe proposed approach clusters location histories accordingto their characteristics and the latter trains an HMM foreach cluster Based on HMMs location characteristics areconsidered as unobservable parameters and the effects of eachindividualrsquos previous actions are also accounted in the processof predictions Finally a prediction accuracy of 1385 canbe achieved when considering regions of roughly 1280 squaremeters

Except for the Markov model and HMM there are someother location prediction algorithms that are used to analyzelocation information in literature [18ndash26] such as the artifi-cial neural network-based algorithm [18] Bayesian network-based methods [19 20] mobile-sink-based methods [2122] the regression-based method [23] and mobile anchorassisted localization algorithms [24 25] These methodspredict future positions of nodes from different perspectiveswhich focus on the behavior of each single node

However in fact the behavior of a node is decidednot onlyby location state of the former period but also by the socialrelationship of mobile users

In this paper a social-relationship-based mobile nodelocation prediction algorithm using daily routines (SMLPR)is proposed In this approach the social relationship of nodesis used to optimize the location prediction algorithm sothat it can better adapt to participatory sensing applicationsand promote the prediction accuracy The remainder of thispaper is organized as follows the networkmodel is illustratedin Section 2 Section 3 specifies the mobile node locationprediction algorithm Extensive simulations have been donefor performance evaluation in Section 4 Section 5 concludesthe paper

2 Network Model

Human movements often exhibit a high degree of repetitionincluding regular visits to certain places and regular contactsduring daily activities In this paper a hybrid urban networkmodel is proposed for participatory sensing systems asillustrated in Figure 1 The map 119872 is partitioned into smallregions In other words 119872 is represented as a finite set ofregions 119886

1 119886

119899 such that ⋃119899

119894=1119886119894= 119872 with 119886

119894⋂119886119895= 0

(119894 = 119895)Themovement possibility of the user from one regionto another is represented by a directed graph In these regionsthe sensed data needs to be aggregated to reduce networkoverhead and to enhance its usefulness among consumers

As mentioned above the network has been reinforcedas a mixture of an opportunistic network and a centralizedinfrastructure which is shown in Figure 1 The centralizedinfrastructure consists of a number of wireless access points(APs) and a backbone connecting the APs The purpose ofthis model is to collect data from a peer-to-peer network(scenario 2 in Figure 1) or WIFI APs which is in the vicinityof the consumers (scenarios 1 and 3 in Figure 1) rather thancollecting it from any 3G4G server Mobile nodes that arecarrying smart device can only access to the network whenthey are walking into the transmission range of any AP anddata transmission can only occur between peer counterpartswhen they fall into each otherrsquos transmission range as innormal opportunistic networks

Definition 1 (location) Inside a geographical region amobiledevice closest to a fixed location is selected to perform datacollection and the collected data is sent to a set of boundingAPs where it is stored and pulled by the consumersThe fixedlocation is termed as the aggregation location

Definition 2 (location granularity) An aggregation locationincludes a set of APs in adjacent position The granularityof the location represents the locationrsquos transmission rangewhich is decided by the number and the scope of the APs inthe same geographical region

In this study historical trajectories of users connectingwith APs are recorded and they are divided by time granu-larity Trajectory of a moving user is defined as a sequence ofpoints (id119895 119897

1 V1199051) (id119895 119897

2 V1199052) (id119895 119897

119898 V119905119898) where id119895

is the userrsquos identifier and location 1198971is represented by a

set of adjacent APs which users are connecting to at timeslot V119905

119894 1 le 119894 le 119898 In this paper a mechanism to

construct aggregation location based on the information ofAPs is proposed which divides APs into different locationsin different granularities

The urban scenario is transformed into a graph 119866 =

119864 119881119882 where each AP is replaced by a vertex V isin 119881 andthe relation between two APs is replaced by an edge 119890 isin 119864where119864 sub 119881times119881The relationshipmatrix ofAPs is replaced by119882 = [119908

119894119895] 119894 isin 119873 119895 isin 119873 where119908

119894119895represents the relationship

between AP 119894 and AP 119895 which can be calculated by

119903119894119895=

2119899119894119895

119899119894+ 119899119895

(1)

International Journal of Distributed Sensor Networks 3

Opportunistic

Data

collection

Contact

Data

distribution

Map M

WIFI AP WIFI AP

Mobile nodeLocations and path

User movementData communication

Scenario 1 Scenario 2 Scenario 3

Figure 1 Participatory sensing scenario

086091

092

091064

077

062

071

083

086

095074

061

082

096

077069

062062

086

075 063067

088

011

04705

058

056

026

018

(a)

077 091 091 092

026

083

011

047 096 075

095 053 067 077

086

086

077 091 091 092

083 096 075

095 053

067

077

086

086

(c)(b)

`C

CA

A

D

D

F

F FG

G G

E

E

B

BB C

A

D E

M

MJ

J J

M N

OL

N

N

OO

LLP P

K K

Q

Q

Q

K

P

H

H H

I

I I

120582 ge 06

Figure 2 Process of location construction

In formula (1) the frequency of AP 119894 and AP 119895 appearing onall usersrsquo devices in the same period is counted denoted as 119899

119894119895

and the number of times that AP 119894 appears in total is denotedas 119899119894(the same to 119899

119895)

Using the greedy algorithm of Kruskal [27] the maxi-mum spanning tree from the graph119866 is easily got denoted as119879 After choosing a weight 120582 as the location granularity theedge in 119879 whose weight is less than 120582 will be cut down and


leave the tree into some separated connected componentsOne connected component is regarded as a location Figure 2shows the process of constructing the aggregation locationsand a construction with granularity 06 is represented inFigure 2(c)

In most WLAN datasets connection is recorded by theformat (node contact time APs and signal strength) andone or more APs which a user connects to may appear inone item at the same time which may cause usersrsquo locationconfusion Therefore after getting the set of locations in themobility scenario estimating which location the user belongsto in the same period is also needed The signal strengthbetween a userrsquos mobile device and a WIFI AP in the WLANdataset can help to solve this problem A weight between theuser and location can be calculated by

weight119894=sum119895=1

119899strength

119895

119899 (2)

In formula (2) 119899 represents the number of aps in location119894 and strength 119895 represents the signal strength between theuser 119860 and AP 119895 (AP

119895isin location

119894) The user 119860 is considered

to be at location 119894 in the time period if 119894meets the conditiondenoted in

119894 = argmax weight119894 (3)

3 Algorithm Design

Thispaper proposes a simplemethod for predicting the futurelocations of mobile nodes on the basis of their previous waysto other locationsThe proposed approach considers differentdaily time periods which relates to the fact that users presentdifferent behaviors and visit different places during their dailyroutines Therefore the hidden Markov model is introducedto capture the dynamism of usersrsquo behavior resulting from thedaily routines

What ismore users experience a combination of periodicmovement that is geographically limited and seeminglyrandom jumps correlated with their social networks Socialrelationships can explain about 10 to 30 of all humanmovement while periodic behavior explains 50 to 70 [28]On the basis of this theory prerequisite social relationshipbetween nodes is also exploited in this paper for optimizationand amendment of location prediction result

31 Hidden Markov Prediction Model In urban scenariousers adopt different behaviors during different periods ofdaily time and the usersrsquo daily routines may influence usersrsquotrajectories Thus the different daily time periods (same asdaily sample) should be considered in order to guaranteea more realistic representation With filtering and hiddenMarkov model this can be done in a simple way

HiddenMarkovmodel (HMM) is awell-known approachfor the analysis of sequential data in which the sequences areassumed to be generated by a Markov process with hiddenstates

Figure 3 shows the general architecture of an instantiatedHMM Each shape in the diagram represents a random

X1 X2 X3

E1 E2 E3 E4

b11 b12 b13 b14 b21 b22 b23 b24 b31 b32 b33 b34

a11

a31

a22

a12

a21a13

a33

a23

a32

Figure 3 Example of hidden Markov model

variable that can adopt any number of values The randomvariable 119909(119905) is the location state at time 119905 The randomvariable 119890(119905) is the daily sample state (in this paper theobserved state is called evidence) at time 119905 The arrows inthe diagram denote conditional dependencies From the dia-gram it is clear that the conditional probability distributionof the location variable 119909(119905) at time 119905 given by the values ofthe location variable 119909 at all times depends only on the valueof the location variable 119909(119905minus1) and thus the value at time 119905minus2and the values before it have no influence This is called theMarkov property Similarly the value of the evidence 119890(119905) onlydepends on the value of the location variable 119909(119905) at time 119905

Given the result of filtering up to time 119905 one can easilycompute the result for 119905 + 1 from the new evidence 119890

119905+1The

calculation can be viewed as actually being composed of twoparts first the current state distribution is projected forwardfrom 119905 to 119905 + 1 Second it is updated using the new evidence119890119905+1

This two-part process emerges using

119875 (119883119905+1

| 1198901119905+1

)

= 120572119875 (119890119905+1

| 119883119905+1)sum

119883119905

119875 (119883119905+1

| 119883119905) 119875 (119883

119905| 1198901119905)

(4)

where 120572 is a normalizing constant used to make probabilitiessum up to 1 Within the summation 119875(119883

119905+1| 119883119905) is the

common transitionmodel and 119875(119883119905| 1198901119905) is the current state

distribution 119875(119890119905+1

| 119883119905+1) is used to update the transition

model and it is obtainable directly from the statistical dataIn participatory sensing system for each application

scenario the hiddenMarkovmodel can be used to predict thefuture location state of each mobile node Prediction processincludes the following steps

311 Preparatory Stage At the beginning the system needsto collect enough information of user movement trajectoriesto construct the Markov chain and therefore a ldquowarm-uprdquo stage is assumed in the prediction system Duringpreparatory stage the system only collects historical data andit cannot provide any predicted information The warm-upstage can last for one day or one week depending on theamount of information collected

312 Determination of State Set The location elements in thecollecting data are extracted and they are denoted as set 119871 As


set 119871 contains a number of location elements the location ofhigher visiting frequency is chosen as state set of the systemdenoted as set 119864 If there are119898 locations in the current scenethe state space can be denoted as 119864 = 119883

1 1198832 119883

119898 and

the location 119894 is the 119894th status119883119894of Markov process

313 Discretization of Data Set Statistical data of all usersrelated to state set 119864 is made Then the data set of each useris processed to be discrete set of the fixed time period so theset after discretization is denoted as follows

(119905119896 119883119894) 119896 = 1 2 3 119894 isin 1 2 3 119898 (5)

314 Calculation of 1-Order Transition Probability Matrix119899119894119895is the frequency that node 119860 departs from location 119894 for

location 119895 then the probability of node 119860 departing fromlocation 119894 for location 119895 is denoted as

119901119894119895=119899119894119895

119899 (6)

where 119899 is the total number of time node119860 departed location119894 to visit other locations in data set

Therefore suppose that there are 119898 locations in the setan119898 times 119898 transition probability matrix is generated as

119875 =

[[[[

[

11990111

11990112

sdot sdot sdot 1199011119898

11990121

11990122


d

1199011198981

1199011198982

sdot sdot sdot 119901119898119898

]]]]

]

(7)

Given 119901(119897)119895

as the probability of node on state 119883119895at initial

moment 119897 and computing the probability of each state theinitial distribution of Markov chain can be obtained as

119875 (119897) = (119901(119897)

1 119901(119897)

2 119901

(119897)

119898) (8)

For example assume that the initial state is1198832 the initial

distribution is as 119875(119897) = (0 1 0 0) and the absolutedistribution at time 119897 + 1 is as

119875 (119897 + 1) = 119875 (119897) 119875 = (119901(119897+1)

1 119901(119897+1)

2 119901(119897+1)

3 119901(119897+1)

4 119901(119897+1)

5) (9)

315 Update the Result from the New Evidence The differentdaily time periods (daily sample) is regarded as the evidencevariable 119890 so the set of evidences can be defined as EV =

119890119903 119903 = 1 2 119889 where 119889 is the total number of daily

time period state For example break the day into four dailysamples (119889 = 4) and the EV is denoted as

EV = am noon pm evening (10)

For each daily sample 119903 a diagonalmatrix119874(119890119903) is definedas

119874 (119890119903) = [119900

119894119895] 119900

119894119895=

0 if 119894 = 119895

119901 (119890119903| 119883 = 119894) if 119894 = 119895

(11)

where 119875(119890119903 | 119883 = 119894) = 119875(119890119903 119883 = 119894)119875(119883 = 119894) = 119899

119903

119894119899119894

119899119903

119894represents the frequency of node arriving at location 119894 in

the daily sample 119903 and 119899119894represents the number of times that

node arrives at location 119894Based on all of the above calculate the probability of node

arriving on location 119894 at next time slot 119897 + 1 using

119875 (119897 + 1)1015840= 120572119874 (119890

119903

119897+1) 119875 (119897 + 1) = 120572119874 (119890

119903

119897+1) 119875 (119897) 119875 (12)

It can be considered that the state 119883119895obtained by the

system at time 119897 + 1 is119883119895= argmax119901(119897+1)

119895

The formula above incorporates a one-step predictionand it is easy to derive the following recursive computationfor prediction of the state at 119905 + 119896 + 1 from a prediction for119905 + 119896 therefore the state119883

119905+119896+1can be obtained by

119883119905+119896+1

= argmax 119875 (119883119905+119896+1

) (13)

32 Social-Aware Prediction Optimization In participatorysensing system amobile node can be the social node carryingdata acquisition equipment Thus the social relationship isused to estimate the future locations of mobile nodes andoptimize the prediction result of hidden Markov model

In this paper capturing the evolution of social interac-tions in the different periods of time (daily sample) overconsecutive days is the aim by computing social strengthbased on the average duration of contacts

Figure 4 shows how social interaction (from the point ofview of user 119860) varies during a day For instance it indicatesa daily sample (8 amndash12 pm) over which the social strengthof user119860 to users 119861 and 119862 is much stronger (less intermittentline) than the strength to users 119863 119864 and 119865 Figure 4 aims toshow the dynamics of a social network over a one-day periodwhere different social structures lead to different behaviorwhen a user moves towards the social community that theuser is related to

As illustrated in Figure 4 the total contact time of mobilenodes119860 and 119861 during a daily sampleΔ119879

119894in a day 119896 is denoted

as

119872119896

119894=

119899

sum

119888=1

(119905119890

119888minus 119905119904

119888) 119888 = 1 2 3 119899 (14)

where 119899 is the number of contact times inΔ119879119894 119905119904119888indicates the

start time of the 119888th contact of mobile node 119860 and 119861 and 119905119890119888

indicates the terminate time of the 119888th contact ofmobile node119860 and 119861

Hence the social strength between any pair of nodes 119860and 119861 in Δ119879

119894is denoted as

119882(119860 119861)119894=sum119898

119896=1119872119896

119894

119898 times Δ119879119894

(15)

where119898 is the total number of days in the historical recordAccording to formula (15) the social relationship matrix

of nodes in Δ119879119894can be obtained On the basis of relation

matrix mobile nodes can be partitioned as communitieswhich determine the closer relation nodes as a subgroupSince the usersrsquo proximity is only taken into accountpartition-based clustering methods such as 119896-means andfuzzy 119888-means are not applicableTherefore use a hierarchical


Daily sample ΔTi

A

Contact (A B) Contact (A B)Contact (A B)

Contact (A B)Contact (A D)Contact (A E)

Contact (A F)Contact (A C)Contact (A C)

A A A AA

B

B

BD

F

E

C C

B

C

W(A B)

800 am 400 pm1200 pm 800 pm 1200 am 400 am 800 am

middot middot middot


Contact (A B)Contact (A C)

Figure 4 Contacts a user 119860 has with a set of users in different daily samples Δ119879119894

clusteringmethod namely complete linkage clustering [29] asthe community partition algorithm

Suppose that it used social relationship to calculatethe probability of node 119860 arriving at the location 119894 (119894 =

1 2 119898) at next period Given that node 119860 belongs tocommunity 119862 and the set of other nodes belonging to119862 on location 119894 at current time slot is denoted as 119878 =

1198781 119878

119895 119878

119899 where 119878 sube 119862 according to conditional

probability then the following formula is proposed

119875119894(119860 | 119878

119895) =

119875119894(119860 119878119895)

119875119894(119878119895) 119895 = 1 119899 (16)

where 119875119894(119860 | 119878

119895) represents the probability of node arriving

at location 119894 on the condition that node 119878119895has already been

on the 119894 location 119875119894(119878119895) represents the probability that node

119878119895keeps on staying at location which can be obtained by

Markovmodel calculation119875119894(119860 119878119895) represents the encounter

probability of node 119860 and node 119878119895on location 119894 and the

formula is defined as

119875119894(119860 119878119895) =

119891119894(119860 119878119895)

sum119898

119894=1119891119894(119860 119878119895) (17)

where 119891119894(119860 119878119895) represents number of encounter times on

location 119894Given the relationship weight of node 119860 and node 119878

119895as

119882(119860 119878119895) = 120588119895 the probability of node 119860 arriving on location

119894 at next time slot is

119875119894(119860) =

119899

sum

119895=1

120582119895119875119894(119860 | 119878

119895) 120582

119895=

120588119895

sum119899

119895=1120588119895

(18)

where 120582119895is the weight of each conditional probability which

is calculated by normalization method sum119899119895=1

120582119895= 1

According to the location distribution of all the nodesbelonging to119862 the probability of node119860 arriving at differentlocation can be obtained And combined with the predictionresult from hidden Markov model and using weight formula(19) to calculate the probability distribution of node 119860 arriv-ing at all the location in the location set the location havingthemaximum of the visiting probabilities is considered as theoutput of the prediction algorithm

119875119894= 119875119894

HMM+ 119889 (119875

social119894

minus 119875119894

HMM) (19)

01 02 03 04 05 06 07 08 09 1

300

350

400

450

500

550

Loca

tion

quan

tity

Mean445

The total number of APs

The value used in the experiments

120582 value

Figure 5 Quantity of locations

where 119875119894

HMM is location prediction probability of state 119883119894

using hidden Markov model and 119875social119894

is the predictionprobability of location 119894 based on social relationship and 119889 isthe damping factorwhich is defined as the probability that thesocial relation between the nodes helps improve the accuracyof the prediction This means that the higher the value of 119889is the more the algorithm accounts for the social relationbetween the nodes

It is beneficial to use social relationship to optimize theprediction result making the transition probability matrixsparse and improve the accuracy of the prediction model

4 Experimental Analyses

41 Simulation Configuration In this paper the experimentdata is from the dataset provided by Wireless TopologyDiscovery (WTD) [30] from which two-month-period datatotal 13215412 items is chosen to simulate the predictionalgorithm There are 275 nodes and 524 APs (access points)in the dataset According to the vicinity of AP positions thenumber of locations at which APs are clustered is shown inFigure 5

Figure 5 shows that when the defined granularity 120582

becomes bigger the quantity of the locations in the gained


183

246

148

235

253

163

89

156

127

206 203

128

123

263

2

186

126

178

192

189134

132155

146

101

212

257

35

237

61

153

65

4

108

165

262

52121

133

47

17734

33

66

68

191

184

49

232

(a) am257

212

132155

189

47

186123

128

203

206

133

156

16552

262121

68 192

253

235

134

232246

191

49

184

183

146

101

148

237

153

61

35

177

178

108

263

126127

16334

66

89

33

2

4

65

(b) Noon

183

184

191

49

246

232

61

153

35257

212

132

148

203

155235

189146

101134

24

19247

68

186 34

178

177

253

126

163

33

89

108127

165

133

52

206

156

121

262

123

263

66

65 237

128

(c) pm

Figure 6 Continued


128

203

257

21265

132

155

235

189

101

146148

237

1922

4

6847

186178

177 263

66253

123

262

108206

52

156

183

184

49

246

191

232

153

61134

35

133

121

165

163

34

12789

33

126

(d) Evening

Figure 6 Social network structures of the dataset in different daily samples

scenario will also become larger When 120582 is defined as 1the quantity of location is equal to the total number of APsThe location granularity 120582 has been given as 05 in followingexperiment

42 Similar User Clustering In order to predict the furtherlocation of mobile nodes using social relationship the socialnetwork structure in the system should be primarily consid-ered Based on the quantization formula (15) we calculatethe relation strength between any pair of nodes 119860 and 119861 indifferent daily sample (am noon pm and evening) andthe social network structures of the dataset are achievedillustrated in Figure 6

A hierarchical clustering method complete linkage clus-tering has been used to cluster mobile users Figure 6(a)shows the social network clustering result in the am periodand the clustering structures in the period of noon pm andevening are respectively illustrated in Figures 6(b) 6(c) and6(d)

43 Prediction Accuracy In order to evaluate the accuracyof prediction model the processed node locations can bedivided into two parts using the 50 that has been chosenfrom the original information to train theMarkovmodel andusing the rest as the test case of the prediction model Theprediction precision 119875result is denoted as

119875result =sum119899

119894=1accuracy

119894

119899 (20)

In formula (20) 119899 represents prediction times andaccuracy

119894is the prediction result of location 119894 denoted as

accuracy119894=

1 when result is right0 when result is wrong

(21)

Firstly the training data set is used to train the predictionmodel which includes standard Markov model (SMM) anddaily-routine-based prediction model (MLPR) Afterwardthe test cases are used respectively to verify the abovementioned two models The prediction accuracies of the twoprediction models are shown in Figure 7 where Figure 7(a)shows the prediction accuracy of nodes from 1 to 92 Figure7(b) shows the prediction accuracy of nodes from 92 to 184and Figure 7(c) shows the prediction accuracy of nodes from185 to 275 From Figure 7 it indicates that the daily-routine-based mobile node location prediction algorithm (MLPR)gains a better performance than standard Markov modelThis shows that daily routines can promote the accuracy andimprove the algorithmrsquos performance

Then make a comparison among the proposed social-relationship-based mobile node location prediction algo-rithmusing daily routines (SMLPR) O2MMand the SMLP

119873

Among these algorithms second order Markov predictor(O2MM) has the best performance among Markov order-119896 predictors [15] and social-relationship-based mobile nodelocation prediction algorithm (SMLP

119873) has the same even

better performance thanO2MMwhich can be obtained fromthe previous work in the paper [31]The comparative result isshown in Figure 8 from which it indicates that SMLPR hasbetter prediction effects after combining with daily routinesand social relationship and gains a higher accuracy thanO2MM and SMLP

119873 Figure 9 shows the number of users in

different precision range among SMM O2MM SMLP119873 and

SMLPR and it illustrates that SMLPR obtained the largestnumber of node distribution in a higher precision range Forinstance the number of nodes with accuracy greater than90 in SMLPR is 198 and in O2MM is 114 and SMM onlyachieves 55 nodes

Lastly the performance of these algorithms is shownas Table 1 The accuracy of SMLPR is 30 higher than the


10 20 30 40 50 60 70 80 900

02

04

06

08

12

1

Node number

Pred

ictio

n ac

cura

cy

SMMMLPR

(a) Nodes 1ndash92

100 110 120 130 140 150 160 170 1800

02

04

06

08

1

Node number

Pred

ictio

n ac

cura

cy

SMMMLPR

12

(b) Nodes 93ndash184

190 200 210 220 230 240 250 260 2700

02

04

06

08

1

Node number

Pred

ictio

n ac

cura

cy

SMMMLPR

12

(c) Nodes 185ndash275

Figure 7 Prediction precision of SMM and MLPR

Table 1 The algorithm performance comparison

SMM O2MM SMLP119873

SMLPRPrediction accuracy 06164 08275 08488 09014Time complexity 119874(119873) 119874(119873

2) 119874(119873) 119874(119873)

Storage space 119874(1198732) 119874(119873

3) 119874(119873

2) 119874(119873

2)

standard Markov model and nearly 10 higher than thesecond order Markov model Then a better result could alsobe obtained in the comparison between SMLPR and SMLP

119873

In the aspects of space cost from Table 1 the complexityof SMLPR is 119874(119873) while O2MM is 119874(1198732) and the memorydemand of SMLPR is 119874(1198732) while O2MM is 119874(1198733) Thus itis proved that the SMLPRgets better performance than order-2Markov predictor atmuch lower expense and the SMLPR is

more practical than order-2 Markov predictor in the WLANscenario

44 Impact of Location Granularity In location-basedmobil-ity scenario location granularity may have a significantinfluence on the prediction accuracy In order to evaluate theimpact of location granularity the algorithmsrsquo performanceis tested by adjusting the granularity value 120582 and the result isshown in Figure 10

As shown in Figure 10 with the increasing of the locationgranularity 120582 due to the number of locations in the scenariothe average accuracies of these four algorithms are relativelydecreasing In these algorithms SMM and O2MM meeta more significant impact on the factor of location andthe accuracy reduces approximately to 25 For SMLPR itshows a relativelymoderate downward trend and the locationgranularity effect to SMLPR is not very obvious


10 20 30 40 50 60 70 80 9005

06

07

08

09

1

Node number

Pred

ictio

n ac

cura

cy

SMLPRO2MM

SMLPN

(a) Nodes 1ndash92

100 110 120 130 140 150 160 170 18005

06

07

08

09

1

Node number

Pred

ictio

n ac

cura

cy

SMLPRO2MM

SMLPN


190 200 210 220 230 240 250 260 27005

06

07

08

09

1

Node number

Pred

ictio

n ac

cura

cy

SMLPRO2MM

SMLPN


Figure 8 Prediction precision of O2MM SMLPR119873 and SMLPR

5 Conclusion

In this paper the influence of opportunistic characteristic inparticipatory sensing system is introduced and the problemsof sensing nodes such as intermittent connection limitedcommunication period and heterogeneous distribution areanalyzed This paper focuses on the mobility model ofnodes in participatory sensing systems and proposes themobile node location prediction algorithm with usersrsquo dailyroutines based on social relationship between mobile nodesAccording to the historical information of mobile nodestrajectories the state transition matrix is constructed by thelocation as the transition state and hidden Markov model isused to predict the mobile node location with the certainduration Meanwhile social relationship between nodes is

exploited for optimization and amendment of the predictionmodelThepredictionmodel is tested based on theWTDdataset and proved to be effective

Conflict of Interests

The authors declare that there is no conflict of interestsregarding the publication of this paper

Acknowledgments

This work is supported by the National Natural ScienceFoundation of China under Grant no 61272529 the NationalScience Foundation for Distinguished Young Scholars of


gt10 gt20 gt30 gt40 gt50 gt60 gt70 gt80 gt900

50

100

150

200

250

300

Prediction accuracy ()

Num

ber o

f use

rs

SMMO2MM SMLPR

SMLPN

Figure 9 The number of users in different precision range

01 02 03 04 05 06 07 08 09 1

05

06

07

08

09

1

Accu

racy

rate

SMMO2MM SMLPR

120582 value

SMLPN

Figure 10 The influence of location granularity to predictionaccuracy

China under Grant nos 61225012 and 71325002 Ministryof Education-China Mobile Research Fund under Grantno MCM20130391 the Specialized Research Fund of theDoctoral Program of Higher Education for the PriorityDevelopment Areas under Grant no 20120042130003 theFundamental Research Funds for the Central Universitiesunder Grant nos N120104001 and N130817003 and LiaoningBaiQianWan Talents Program under Grant no 2013921068

References

[1] M Srivastava M Hansen J Burke et al ldquoWireless urban sens-ing systemsrdquo Tech Rep 65 Center for Embedded NetworkedSensing at UCLA 2006

[2] S B Eisenman E Miluzzo N D Lane R A Peterson G-SAhn and A T Campbell ldquoBikeNet a mobile sensing systemfor cyclist experience mappingrdquo ACM Transactions on SensorNetworks vol 6 no 1 article 6 2009

[3] H Lu W Pan N D Lane T Choudhury and A T Camp-bell ldquoSoundSense scalable sound sensing for people-centricapplications on mobile phonesrdquo in Proceedings of the 7th ACMInternational Conference on Mobile Systems Applications andServices (MobiSys rsquo09) pp 165ndash178 Krakov Poland June 2009

[4] E Miluzzo N D Lane K Fodor et al ldquoSensing meets mobilesocial networks the design implementation and evaluationof the CenceMe applicationrdquo in Proceedings of the 6th ACMConference on Embedded Networked Sensor Systems (SenSysrsquo08) pp 337ndash350 Raleigh NC USA November 2008

[5] EMiluzzoN Lane S Eisenman andACampbell ldquoCenceMe ainjecting sensing presence into social networking applicationsrdquoin Smart Sensing andContext G Kortuem J Finney R Lea andV Sundramoorthy Eds vol 4793 of Smart Sensing andContextpp 1ndash28 2007

[6] S B Eisenman N D Lane E Miluzzo et al ldquoMetroSenseproject people-centric sensing at scalerdquo in Proceedings of theWorkshop on World-Sensor-Web pp 6ndash11 Boulder Colo USA2006

[7] H Lu N D Lane S B Eisenman and A T Campbell ldquoBubble-sensing binding sensing tasks to the physical worldrdquo Pervasiveand Mobile Computing vol 6 no 1 pp 58ndash71 2010

[8] L Deng and L P Cox ldquoLive compare grocery bargain huntingthrough participatory sensingrdquo in Proceedings of the 10thWork-shop onMobile Computing Systems and Applications (HotMobilersquo09) Santa Cruz Calif USA February 2009

[9] E Kanjo ldquoNoiseSPY a real-time mobile phone platform forurban noise monitoring and mappingrdquo Mobile Networks andApplications vol 15 no 4 pp 562ndash574 2010

[10] A J Perez M A Labrador and S J Barbeau ldquoG-Sense ascalable architecture for global sensing and monitoringrdquo IEEENetwork vol 24 no 4 pp 57ndash64 2010

[11] L M L Oliveira J J P C Rodrigues A G F Elias and G HanldquoWireless sensor networks in IPv4IPv6 transition scenariosrdquoWireless Personal Communications vol 78 no 4 pp 1849ndash18622014

[12] C Song Z Qu N Blumm and A-L Barabasi ldquoLimits of pre-dictability in human mobilityrdquo Science vol 327 no 5968 pp1018ndash1021 2010

[13] M C Gonzalez C A Hidalgo and A-L Barabasi ldquoUnder-standing individual human mobility patternsrdquo Nature vol 453no 7196 pp 779ndash782 2008

[14] S-M Qin H Verkasalo MMohtaschemi T Hartonen andMAlava ldquoPatterns entropy and predictability of human mobilityand liferdquo PLoS ONE vol 7 no 12 Article ID e51353 2012

[15] L Song D Kotz R Jain et al ldquoEvaluating location predictorswith extensive Wi-Fi mobility datardquo in Proceedings of the 23rdAnnual Joint Conference of the IEEE Computer and Communi-cations Societies (INFOCOM rsquo04) vol 2 pp 1414ndash1424 2004

[16] S Scellato M Musolesi C Mascolo V Latora and A TCampbell ldquoNextPlace a spatio-temporal prediction frameworkfor pervasive systemsrdquo in Pervasive Computing vol 6696 ofLecture Notes in Computer Science pp 152ndash169 Springer BerlinGermany 2011

[17] W Mathew R Raposo and B Martins ldquoPredicting future loca-tions with hidden Markov modelsrdquo in Proceedings of the 14thInternational Conference on Ubiquitous Computing (UbiComprsquo12) pp 911ndash918 September 2012


[18] M C Mozer ldquoThe neural network house an environment thatadapts to its inhabitantsrdquo inProceedings of theAAAI Spring Sym-posium pp 110ndash114 Stanford Calif USA 1998

[19] H A Karimi and X Liu ldquoA predictive location model forlocation-based servicesrdquo inProceedings of the 11th ACM Interna-tional Symposium on Advances in Geographic Information Sys-tems (GIS rsquo03) pp 126ndash133 New Orleans La USA November2003

[20] J D Patterson L Liao D Fox et al ldquoInferring high-levelbehavior from low-level sensorsrdquo in Proceedings of the 5thAnnual Conference on Ubiquitous Computing (UbiComp rsquo03)pp 73ndash89 Seattle Wash USA 2003

[21] C Zhu Y Wang G Han J J P C Rodrigues and J LloretldquoLPTA location predictive and time adaptive data gatheringscheme with mobile sink for wireless sensor networksrdquo TheScientific World Journal vol 2014 Article ID 476253 13 pages2014

[22] C Zhu Y Wang G Han J J P C Rodrigues and H Guo ldquoAlocation prediction based data gathering protocol for wirelesssensor networks using a mobile sinkrdquo in Proceedings of the 2ndSmart Sensor Networks and Algorithms (SSPA rsquo14) Co-Locatedwith 13th International Conference on Ad Hoc Mobile andWoreless Networks (Ad Hoc rsquo14) Benidorm Spain June 2014

[23] Y-B He S-D Fan and Z-X Hao ldquoWhole trajectory modelingof moving objects based onMOSTmodelrdquo Computer Engineer-ing vol 34 no 16 pp 41ndash43 2008

[24] G Han C Zhang J Lloret L Shu and J J P C Rodrigues ldquoAmobile anchor assisted localization algorithm based on regularhexagon in wireless sensor networksrdquo The Scientific WorldJournal vol 2014 Article ID 219371 13 pages 2014

[25] G Han H Xu J Jiang L Shu and N Chilamkurti ldquoTheinsights of localization throughmobile anchor nodes inwirelesssensor networks with irregular radiordquo KSII Transactions onInternet and Information Systems vol 6 no 11 pp 2992ndash30072012

[26] G Han C Zhang T Liu and L Shu ldquoMANCL a multi-anchor nodes cooperative localization algorithm for underwa-ter acoustic sensor networksrdquo Wireless Communications andMobile Computing In press

[27] J Kruskal ldquoOn the shortest spanning subtree of a graph andthe traveling salesman problemrdquo Proceedings of the AmericanMathematical Society vol 7 pp 48ndash50 1956

[28] E Cho S A Myers and J Leskovec ldquoFriendship and mobilityuser movement in location-based social networksrdquo in Proceed-ings of the 17th ACM SIGKDD International Conference onKnowledgeDiscovery andDataMining (KDD rsquo11) pp 1082ndash1090ACM August 2011

[29] G Punj and D W Stewart ldquoCluster analysis in marketingresearch review and suggestions for applicationrdquo Journal ofMarketing Research vol 20 no 2 pp 134ndash148 1983

[30] M McNett and G M Voelker ldquoUCSDWireless Topology Dis-covery Project [EBOL]rdquo 2013 httpwwwsysnetucsdeduwtdwtdhtml

[31] R Ru and X Xia ldquoSocial-relationship-based mobile nodelocation prediction algorithm in participatory sensing systemsrdquoChinese Journal of Computers vol 35 no 6 2014

International Journal of

AerospaceEngineeringHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

RoboticsJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014


Active and Passive Electronic Components

Control Scienceand Engineering

Journal of



RotatingMachinery


Hindawi Publishing Corporation httpwwwhindawicom

Journal ofEngineeringVolume 2014

Submit your manuscripts athttpwwwhindawicom

VLSI Design



Shock and Vibration


Civil EngineeringAdvances in

Acoustics and VibrationAdvances in



Electrical and Computer Engineering

Journal of

Advances inOptoElectronics


Volume 2014

The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014

SensorsJournal of


Modelling amp Simulation in EngineeringHindawi Publishing Corporation httpwwwhindawicom Volume 2014


Chemical EngineeringInternational Journal of Antennas and

Propagation




Navigation and Observation



DistributedSensor Networks



In [13] the location data from 100000 mobile phone userscollected by tracking each personrsquos position for six monthsis investigated This research shows that individual mobilitypatterns of the test persons can be described by a single spatialprobability distribution after some corrective preprocessingThus the results of this study suggest that humans generallyfollow simple reproducible patterns

Based on the human mobility analysis different tech-niques based on the Markov model have been applied tolocation prediction problems of human individuals In [15]a set of discrete locations has been defined using the WIFIcells on a university campus Two different kinds of locationpredictors the 119896th order Markov predictor as well as aLZ-based predictor are tested in predictions to the nextlocation Based on the test results the research shows thatthe second order Markov predictor with a certain fallbackfeature performs best and provided a median accuracy of72 Literature [16] presents a similar extended Markovpredictor which takes the arrival time and residence timeinto account Specifically delay embedding is used to extractlocation sequences of a certain length from time seriesThen these sequences are directly used to predict a userrsquoslocation and the prediction result is obtained by comparingthe last observed locations to all embedded location sequen-ces

Meanwhile hidden Markov models (HMMs) are alsoconsidered to predict human mobility Literature [17] pre-sents a hybridmethod on the basis of hiddenMarkovmodelsThe proposed approach clusters location histories accordingto their characteristics and the latter trains an HMM foreach cluster Based on HMMs location characteristics areconsidered as unobservable parameters and the effects of eachindividualrsquos previous actions are also accounted in the processof predictions Finally a prediction accuracy of 1385 canbe achieved when considering regions of roughly 1280 squaremeters

Except for the Markov model and HMM there are someother location prediction algorithms that are used to analyzelocation information in literature [18ndash26] such as the artifi-cial neural network-based algorithm [18] Bayesian network-based methods [19 20] mobile-sink-based methods [2122] the regression-based method [23] and mobile anchorassisted localization algorithms [24 25] These methodspredict future positions of nodes from different perspectiveswhich focus on the behavior of each single node

However in fact the behavior of a node is decidednot onlyby location state of the former period but also by the socialrelationship of mobile users

In this paper a social-relationship-based mobile nodelocation prediction algorithm using daily routines (SMLPR)is proposed In this approach the social relationship of nodesis used to optimize the location prediction algorithm sothat it can better adapt to participatory sensing applicationsand promote the prediction accuracy The remainder of thispaper is organized as follows the networkmodel is illustratedin Section 2 Section 3 specifies the mobile node locationprediction algorithm Extensive simulations have been donefor performance evaluation in Section 4 Section 5 concludesthe paper

2 Network Model

Human movements often exhibit a high degree of repetitionincluding regular visits to certain places and regular contactsduring daily activities In this paper a hybrid urban networkmodel is proposed for participatory sensing systems asillustrated in Figure 1 The map 119872 is partitioned into smallregions In other words 119872 is represented as a finite set ofregions 119886

1 119886

119899 such that ⋃119899

119894=1119886119894= 119872 with 119886

119894⋂119886119895= 0

(119894 = 119895)Themovement possibility of the user from one regionto another is represented by a directed graph In these regionsthe sensed data needs to be aggregated to reduce networkoverhead and to enhance its usefulness among consumers

As mentioned above the network has been reinforcedas a mixture of an opportunistic network and a centralizedinfrastructure which is shown in Figure 1 The centralizedinfrastructure consists of a number of wireless access points(APs) and a backbone connecting the APs The purpose ofthis model is to collect data from a peer-to-peer network(scenario 2 in Figure 1) or WIFI APs which is in the vicinityof the consumers (scenarios 1 and 3 in Figure 1) rather thancollecting it from any 3G4G server Mobile nodes that arecarrying smart device can only access to the network whenthey are walking into the transmission range of any AP anddata transmission can only occur between peer counterpartswhen they fall into each otherrsquos transmission range as innormal opportunistic networks

Definition 1 (location) Inside a geographical region amobiledevice closest to a fixed location is selected to perform datacollection and the collected data is sent to a set of boundingAPs where it is stored and pulled by the consumersThe fixedlocation is termed as the aggregation location

Definition 2 (location granularity) An aggregation locationincludes a set of APs in adjacent position The granularityof the location represents the locationrsquos transmission rangewhich is decided by the number and the scope of the APs inthe same geographical region

In this study historical trajectories of users connectingwith APs are recorded and they are divided by time granu-larity Trajectory of a moving user is defined as a sequence ofpoints (id119895 119897

1 V1199051) (id119895 119897

2 V1199052) (id119895 119897

119898 V119905119898) where id119895

is the userrsquos identifier and location 1198971is represented by a

set of adjacent APs which users are connecting to at timeslot V119905

119894 1 le 119894 le 119898 In this paper a mechanism to

construct aggregation location based on the information ofAPs is proposed which divides APs into different locationsin different granularities

The urban scenario is transformed into a graph 119866 =

119864 119881119882 where each AP is replaced by a vertex V isin 119881 andthe relation between two APs is replaced by an edge 119890 isin 119864where119864 sub 119881times119881The relationshipmatrix ofAPs is replaced by119882 = [119908

119894119895] 119894 isin 119873 119895 isin 119873 where119908

119894119895represents the relationship

between AP 119894 and AP 119895 which can be calculated by

119903119894119895=

2119899119894119895

119899119894+ 119899119895

(1)


Opportunistic

Data

collection

Contact

Data

distribution

Map M

WIFI AP WIFI AP





086091

092

091064

077

062

071

083

086

095074

061

082

096

077069

062062

086

075 063067

088

011

04705

058

056

026

018

(a)

077 091 091 092

026

083

011

047 096 075

095 053 067 077

086

086

077 091 091 092

083 096 075

095 053

067

077

086

086

(c)(b)

`C

CA

A

D

D

F

F FG

G G

E

E

B

BB C

A

D E

M

MJ

J J

M N

OL

N

N

OO

LLP P

K K

Q

Q

Q

K

P

H

H H

I

I I

120582 ge 06



119894119895


119895)





weight119894=sum119895=1

119899strength

119895

119899 (2)


119895isin location



119894 = argmax weight119894 (3)

3 Algorithm Design






X1 X2 X3

E1 E2 E3 E4

b11 b12 b13 b14 b21 b22 b23 b24 b31 b32 b33 b34

a11

a31

a22

a12

a21a13

a33

a23

a32




119905+1The



119875 (119883119905+1

| 1198901119905+1

)

= 120572119875 (119890119905+1

| 119883119905+1)sum

119883119905

119875 (119883119905+1

| 119883119905) 119875 (119883

119905| 1198901119905)

(4)


119905+1| 119883119905) is the


distribution 119875(119890119905+1








1 1198832 119883

119898 and



(119905119896 119883119894) 119896 = 1 2 3 119894 isin 1 2 3 119898 (5)



119901119894119895=119899119894119895

119899 (6)



119875 =

[[[[

[

11990111

11990112


11990121

11990122


d

1199011198981

1199011198982

sdot sdot sdot 119901119898119898

]]]]

]

(7)

Given 119901(119897)119895



119875 (119897) = (119901(119897)

1 119901(119897)

2 119901

(119897)

119898) (8)



119875 (119897 + 1) = 119875 (119897) 119875 = (119901(119897+1)

1 119901(119897+1)

2 119901(119897+1)

3 119901(119897+1)

4 119901(119897+1)

5) (9)






119874 (119890119903) = [119900

119894119895] 119900

119894119895=

0 if 119894 = 119895

119901 (119890119903| 119883 = 119894) if 119894 = 119895

(11)

where 119875(119890119903 | 119883 = 119894) = 119875(119890119903 119883 = 119894)119875(119883 = 119894) = 119899

119903

119894119899119894

119899119903





119875 (119897 + 1)1015840= 120572119874 (119890

119903

119897+1) 119875 (119897 + 1) = 120572119874 (119890

119903

119897+1) 119875 (119897) 119875 (12)



119895



119883119905+119896+1

= argmax 119875 (119883119905+119896+1

) (13)






as

119872119896

119894=

119899

sum

119888=1

(119905119890

119888minus 119905119904

119888) 119888 = 1 2 3 119899 (14)





119894is denoted as

119882(119860 119861)119894=sum119898

119896=1119872119896

119894

119898 times Δ119879119894

(15)





Daily sample ΔTi

A




A A A AA

B

B

BD

F

E

C C

B

C

W(A B)

800 am 400 pm1200 pm 800 pm 1200 am 400 am 800 am








1198781 119878

119895 119878



119875119894(119860 | 119878

119895) =

119875119894(119860 119878119895)

119875119894(119878119895) 119895 = 1 119899 (16)

where 119875119894(119860 | 119878








119875119894(119860 119878119895) =

119891119894(119860 119878119895)

sum119898

119894=1119891119894(119860 119878119895) (17)



119895as



119875119894(119860) =

119899

sum

119895=1

120582119895119875119894(119860 | 119878

119895) 120582

119895=

120588119895

sum119899

119895=1120588119895

(18)



120582119895= 1


119875119894= 119875119894

HMM+ 119889 (119875

social119894

minus 119875119894

HMM) (19)

01 02 03 04 05 06 07 08 09 1

300

350

400

450

500

550

Loca

tion

quan

tity

Mean445



120582 value


where 119875119894










183

246

148

235

253

163

89

156

127

206 203

128

123

263

2

186

126

178

192

189134

132155

146

101

212

257

35

237

61

153

65

4

108

165

262

52121

133

47

17734

33

66

68

191

184

49

232

(a) am257

212

132155

189

47

186123

128

203

206

133

156

16552

262121

68 192

253

235

134

232246

191

49

184

183

146

101

148

237

153

61

35

177

178

108

263

126127

16334

66

89

33

2

4

65

(b) Noon

183

184

191

49

246

232

61

153

35257

212

132

148

203

155235

189146

101134

24

19247

68

186 34

178

177

253

126

163

33

89

108127

165

133

52

206

156

121

262

123

263

66

65 237

128

(c) pm

Figure 6 Continued


128

203

257

21265

132

155

235

189

101

146148

237

1922

4

6847

186178

177 263

66253

123

262

108206

52

156

183

184

49

246

191

232

153

61134

35

133

121

165

163

34

12789

33

126

(d) Evening






119875result =sum119899

119894=1accuracy

119894

119899 (20)



accuracy119894=


(21)



119873









10 20 30 40 50 60 70 80 900

02

04

06

08

12

1

Node number

Pred

ictio

n ac

cura

cy

SMMMLPR

(a) Nodes 1ndash92

100 110 120 130 140 150 160 170 1800

02

04

06

08

1

Node number

Pred

ictio

n ac

cura

cy

SMMMLPR

12


190 200 210 220 230 240 250 260 2700

02

04

06

08

1

Node number

Pred

ictio

n ac

cura

cy

SMMMLPR

12




SMM O2MM SMLP119873


2) 119874(119873) 119874(119873)

Storage space 119874(1198732) 119874(119873

3) 119874(119873

2) 119874(119873

2)


119873






10 20 30 40 50 60 70 80 9005

06

07

08

09

1

Node number

Pred

ictio

n ac

cura

cy

SMLPRO2MM

SMLPN

(a) Nodes 1ndash92

100 110 120 130 140 150 160 170 18005

06

07

08

09

1

Node number

Pred

ictio

n ac

cura

cy

SMLPRO2MM

SMLPN


190 200 210 220 230 240 250 260 27005

06

07

08

09

1

Node number

Pred

ictio

n ac

cura

cy

SMLPRO2MM

SMLPN



5 Conclusion





Acknowledgments




50

100

150

200

250

300


Num

ber o

f use

rs

SMMO2MM SMLPR

SMLPN


01 02 03 04 05 06 07 08 09 1

05

06

07

08

09

1

Accu

racy

rate

SMMO2MM SMLPR

120582 value

SMLPN



References



































RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation










Opportunistic

Data

collection

Contact

Data

distribution

Map M

WIFI AP WIFI AP





086091

092

091064

077

062

071

083

086

095074

061

082

096

077069

062062

086

075 063067

088

011

04705

058

056

026

018

(a)

077 091 091 092

026

083

011

047 096 075

095 053 067 077

086

086

077 091 091 092

083 096 075

095 053

067

077

086

086

(c)(b)

`C

CA

A

D

D

F

F FG

G G

E

E

B

BB C

A

D E

M

MJ

J J

M N

OL

N

N

OO

LLP P

K K

Q

Q

Q

K

P

H

H H

I

I I

120582 ge 06



119894119895


119895)





weight119894=sum119895=1

119899strength

119895

119899 (2)


119895isin location



119894 = argmax weight119894 (3)

3 Algorithm Design






X1 X2 X3

E1 E2 E3 E4

b11 b12 b13 b14 b21 b22 b23 b24 b31 b32 b33 b34

a11

a31

a22

a12

a21a13

a33

a23

a32




119905+1The



119875 (119883119905+1

| 1198901119905+1

)

= 120572119875 (119890119905+1

| 119883119905+1)sum

119883119905

119875 (119883119905+1

| 119883119905) 119875 (119883

119905| 1198901119905)

(4)


119905+1| 119883119905) is the


distribution 119875(119890119905+1








1 1198832 119883

119898 and



(119905119896 119883119894) 119896 = 1 2 3 119894 isin 1 2 3 119898 (5)



119901119894119895=119899119894119895

119899 (6)



119875 =

[[[[

[

11990111

11990112


11990121

11990122


d

1199011198981

1199011198982

sdot sdot sdot 119901119898119898

]]]]

]

(7)

Given 119901(119897)119895



119875 (119897) = (119901(119897)

1 119901(119897)

2 119901

(119897)

119898) (8)



119875 (119897 + 1) = 119875 (119897) 119875 = (119901(119897+1)

1 119901(119897+1)

2 119901(119897+1)

3 119901(119897+1)

4 119901(119897+1)

5) (9)






119874 (119890119903) = [119900

119894119895] 119900

119894119895=

0 if 119894 = 119895

119901 (119890119903| 119883 = 119894) if 119894 = 119895

(11)

where 119875(119890119903 | 119883 = 119894) = 119875(119890119903 119883 = 119894)119875(119883 = 119894) = 119899

119903

119894119899119894

119899119903





119875 (119897 + 1)1015840= 120572119874 (119890

119903

119897+1) 119875 (119897 + 1) = 120572119874 (119890

119903

119897+1) 119875 (119897) 119875 (12)



119895



119883119905+119896+1

= argmax 119875 (119883119905+119896+1

) (13)






as

119872119896

119894=

119899

sum

119888=1

(119905119890

119888minus 119905119904

119888) 119888 = 1 2 3 119899 (14)





119894is denoted as

119882(119860 119861)119894=sum119898

119896=1119872119896

119894

119898 times Δ119879119894

(15)





Daily sample ΔTi

A




A A A AA

B

B

BD

F

E

C C

B

C

W(A B)

800 am 400 pm1200 pm 800 pm 1200 am 400 am 800 am








1198781 119878

119895 119878



119875119894(119860 | 119878

119895) =

119875119894(119860 119878119895)

119875119894(119878119895) 119895 = 1 119899 (16)

where 119875119894(119860 | 119878








119875119894(119860 119878119895) =

119891119894(119860 119878119895)

sum119898

119894=1119891119894(119860 119878119895) (17)



119895as



119875119894(119860) =

119899

sum

119895=1

120582119895119875119894(119860 | 119878

119895) 120582

119895=

120588119895

sum119899

119895=1120588119895

(18)



120582119895= 1


119875119894= 119875119894

HMM+ 119889 (119875

social119894

minus 119875119894

HMM) (19)

01 02 03 04 05 06 07 08 09 1

300

350

400

450

500

550

Loca

tion

quan

tity

Mean445



120582 value


where 119875119894










183

246

148

235

253

163

89

156

127

206 203

128

123

263

2

186

126

178

192

189134

132155

146

101

212

257

35

237

61

153

65

4

108

165

262

52121

133

47

17734

33

66

68

191

184

49

232

(a) am257

212

132155

189

47

186123

128

203

206

133

156

16552

262121

68 192

253

235

134

232246

191

49

184

183

146

101

148

237

153

61

35

177

178

108

263

126127

16334

66

89

33

2

4

65

(b) Noon

183

184

191

49

246

232

61

153

35257

212

132

148

203

155235

189146

101134

24

19247

68

186 34

178

177

253

126

163

33

89

108127

165

133

52

206

156

121

262

123

263

66

65 237

128

(c) pm

Figure 6 Continued


128

203

257

21265

132

155

235

189

101

146148

237

1922

4

6847

186178

177 263

66253

123

262

108206

52

156

183

184

49

246

191

232

153

61134

35

133

121

165

163

34

12789

33

126

(d) Evening






119875result =sum119899

119894=1accuracy

119894

119899 (20)



accuracy119894=


(21)



119873









10 20 30 40 50 60 70 80 900

02

04

06

08

12

1

Node number

Pred

ictio

n ac

cura

cy

SMMMLPR

(a) Nodes 1ndash92

100 110 120 130 140 150 160 170 1800

02

04

06

08

1

Node number

Pred

ictio

n ac

cura

cy

SMMMLPR

12


190 200 210 220 230 240 250 260 2700

02

04

06

08

1

Node number

Pred

ictio

n ac

cura

cy

SMMMLPR

12




SMM O2MM SMLP119873


2) 119874(119873) 119874(119873)

Storage space 119874(1198732) 119874(119873

3) 119874(119873

2) 119874(119873

2)


119873






10 20 30 40 50 60 70 80 9005

06

07

08

09

1

Node number

Pred

ictio

n ac

cura

cy

SMLPRO2MM

SMLPN

(a) Nodes 1ndash92

100 110 120 130 140 150 160 170 18005

06

07

08

09

1

Node number

Pred

ictio

n ac

cura

cy

SMLPRO2MM

SMLPN


190 200 210 220 230 240 250 260 27005

06

07

08

09

1

Node number

Pred

ictio

n ac

cura

cy

SMLPRO2MM

SMLPN



5 Conclusion





Acknowledgments




50

100

150

200

250

300


Num

ber o

f use

rs

SMMO2MM SMLPR

SMLPN


01 02 03 04 05 06 07 08 09 1

05

06

07

08

09

1

Accu

racy

rate

SMMO2MM SMLPR

120582 value

SMLPN



References



































RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation












weight119894=sum119895=1

119899strength

119895

119899 (2)


119895isin location



119894 = argmax weight119894 (3)

3 Algorithm Design






X1 X2 X3

E1 E2 E3 E4

b11 b12 b13 b14 b21 b22 b23 b24 b31 b32 b33 b34

a11

a31

a22

a12

a21a13

a33

a23

a32




119905+1The



119875 (119883119905+1

| 1198901119905+1

)

= 120572119875 (119890119905+1

| 119883119905+1)sum

119883119905

119875 (119883119905+1

| 119883119905) 119875 (119883

119905| 1198901119905)

(4)


119905+1| 119883119905) is the


distribution 119875(119890119905+1








1 1198832 119883

119898 and



(119905119896 119883119894) 119896 = 1 2 3 119894 isin 1 2 3 119898 (5)



119901119894119895=119899119894119895

119899 (6)



119875 =

[[[[

[

11990111

11990112


11990121

11990122


d

1199011198981

1199011198982

sdot sdot sdot 119901119898119898

]]]]

]

(7)

Given 119901(119897)119895



119875 (119897) = (119901(119897)

1 119901(119897)

2 119901

(119897)

119898) (8)



119875 (119897 + 1) = 119875 (119897) 119875 = (119901(119897+1)

1 119901(119897+1)

2 119901(119897+1)

3 119901(119897+1)

4 119901(119897+1)

5) (9)






119874 (119890119903) = [119900

119894119895] 119900

119894119895=

0 if 119894 = 119895

119901 (119890119903| 119883 = 119894) if 119894 = 119895

(11)

where 119875(119890119903 | 119883 = 119894) = 119875(119890119903 119883 = 119894)119875(119883 = 119894) = 119899

119903

119894119899119894

119899119903





119875 (119897 + 1)1015840= 120572119874 (119890

119903

119897+1) 119875 (119897 + 1) = 120572119874 (119890

119903

119897+1) 119875 (119897) 119875 (12)



119895



119883119905+119896+1

= argmax 119875 (119883119905+119896+1

) (13)






as

119872119896

119894=

119899

sum

119888=1

(119905119890

119888minus 119905119904

119888) 119888 = 1 2 3 119899 (14)





119894is denoted as

119882(119860 119861)119894=sum119898

119896=1119872119896

119894

119898 times Δ119879119894

(15)





Daily sample ΔTi

A




A A A AA

B

B

BD

F

E

C C

B

C

W(A B)

800 am 400 pm1200 pm 800 pm 1200 am 400 am 800 am








1198781 119878

119895 119878



119875119894(119860 | 119878

119895) =

119875119894(119860 119878119895)

119875119894(119878119895) 119895 = 1 119899 (16)

where 119875119894(119860 | 119878








119875119894(119860 119878119895) =

119891119894(119860 119878119895)

sum119898

119894=1119891119894(119860 119878119895) (17)



119895as



119875119894(119860) =

119899

sum

119895=1

120582119895119875119894(119860 | 119878

119895) 120582

119895=

120588119895

sum119899

119895=1120588119895

(18)



120582119895= 1


119875119894= 119875119894

HMM+ 119889 (119875

social119894

minus 119875119894

HMM) (19)

01 02 03 04 05 06 07 08 09 1

300

350

400

450

500

550

Loca

tion

quan

tity

Mean445



120582 value


where 119875119894










183

246

148

235

253

163

89

156

127

206 203

128

123

263

2

186

126

178

192

189134

132155

146

101

212

257

35

237

61

153

65

4

108

165

262

52121

133

47

17734

33

66

68

191

184

49

232

(a) am257

212

132155

189

47

186123

128

203

206

133

156

16552

262121

68 192

253

235

134

232246

191

49

184

183

146

101

148

237

153

61

35

177

178

108

263

126127

16334

66

89

33

2

4

65

(b) Noon

183

184

191

49

246

232

61

153

35257

212

132

148

203

155235

189146

101134

24

19247

68

186 34

178

177

253

126

163

33

89

108127

165

133

52

206

156

121

262

123

263

66

65 237

128

(c) pm

Figure 6 Continued


128

203

257

21265

132

155

235

189

101

146148

237

1922

4

6847

186178

177 263

66253

123

262

108206

52

156

183

184

49

246

191

232

153

61134

35

133

121

165

163

34

12789

33

126

(d) Evening






119875result =sum119899

119894=1accuracy

119894

119899 (20)



accuracy119894=


(21)



119873









10 20 30 40 50 60 70 80 900

02

04

06

08

12

1

Node number

Pred

ictio

n ac

cura

cy

SMMMLPR

(a) Nodes 1ndash92

100 110 120 130 140 150 160 170 1800

02

04

06

08

1

Node number

Pred

ictio

n ac

cura

cy

SMMMLPR

12


190 200 210 220 230 240 250 260 2700

02

04

06

08

1

Node number

Pred

ictio

n ac

cura

cy

SMMMLPR

12




SMM O2MM SMLP119873


2) 119874(119873) 119874(119873)

Storage space 119874(1198732) 119874(119873

3) 119874(119873

2) 119874(119873

2)


119873






10 20 30 40 50 60 70 80 9005

06

07

08

09

1

Node number

Pred

ictio

n ac

cura

cy

SMLPRO2MM

SMLPN

(a) Nodes 1ndash92

100 110 120 130 140 150 160 170 18005

06

07

08

09

1

Node number

Pred

ictio

n ac

cura

cy

SMLPRO2MM

SMLPN


190 200 210 220 230 240 250 260 27005

06

07

08

09

1

Node number

Pred

ictio

n ac

cura

cy

SMLPRO2MM

SMLPN



5 Conclusion





Acknowledgments




50

100

150

200

250

300


Num

ber o

f use

rs

SMMO2MM SMLPR

SMLPN


01 02 03 04 05 06 07 08 09 1

05

06

07

08

09

1

Accu

racy

rate

SMMO2MM SMLPR

120582 value

SMLPN



References



































RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation











1 1198832 119883

119898 and



(119905119896 119883119894) 119896 = 1 2 3 119894 isin 1 2 3 119898 (5)



119901119894119895=119899119894119895

119899 (6)



119875 =

[[[[

[

11990111

11990112


11990121

11990122


d

1199011198981

1199011198982

sdot sdot sdot 119901119898119898

]]]]

]

(7)

Given 119901(119897)119895



119875 (119897) = (119901(119897)

1 119901(119897)

2 119901

(119897)

119898) (8)



119875 (119897 + 1) = 119875 (119897) 119875 = (119901(119897+1)

1 119901(119897+1)

2 119901(119897+1)

3 119901(119897+1)

4 119901(119897+1)

5) (9)






119874 (119890119903) = [119900

119894119895] 119900

119894119895=

0 if 119894 = 119895

119901 (119890119903| 119883 = 119894) if 119894 = 119895

(11)

where 119875(119890119903 | 119883 = 119894) = 119875(119890119903 119883 = 119894)119875(119883 = 119894) = 119899

119903

119894119899119894

119899119903





119875 (119897 + 1)1015840= 120572119874 (119890

119903

119897+1) 119875 (119897 + 1) = 120572119874 (119890

119903

119897+1) 119875 (119897) 119875 (12)



119895



119883119905+119896+1

= argmax 119875 (119883119905+119896+1

) (13)






as

119872119896

119894=

119899

sum

119888=1

(119905119890

119888minus 119905119904

119888) 119888 = 1 2 3 119899 (14)





119894is denoted as

119882(119860 119861)119894=sum119898

119896=1119872119896

119894

119898 times Δ119879119894

(15)





Daily sample ΔTi

A




A A A AA

B

B

BD

F

E

C C

B

C

W(A B)

800 am 400 pm1200 pm 800 pm 1200 am 400 am 800 am








1198781 119878

119895 119878



119875119894(119860 | 119878

119895) =

119875119894(119860 119878119895)

119875119894(119878119895) 119895 = 1 119899 (16)

where 119875119894(119860 | 119878








119875119894(119860 119878119895) =

119891119894(119860 119878119895)

sum119898

119894=1119891119894(119860 119878119895) (17)



119895as



119875119894(119860) =

119899

sum

119895=1

120582119895119875119894(119860 | 119878

119895) 120582

119895=

120588119895

sum119899

119895=1120588119895

(18)



120582119895= 1


119875119894= 119875119894

HMM+ 119889 (119875

social119894

minus 119875119894

HMM) (19)

01 02 03 04 05 06 07 08 09 1

300

350

400

450

500

550

Loca

tion

quan

tity

Mean445



120582 value


where 119875119894










183

246

148

235

253

163

89

156

127

206 203

128

123

263

2

186

126

178

192

189134

132155

146

101

212

257

35

237

61

153

65

4

108

165

262

52121

133

47

17734

33

66

68

191

184

49

232

(a) am257

212

132155

189

47

186123

128

203

206

133

156

16552

262121

68 192

253

235

134

232246

191

49

184

183

146

101

148

237

153

61

35

177

178

108

263

126127

16334

66

89

33

2

4

65

(b) Noon

183

184

191

49

246

232

61

153

35257

212

132

148

203

155235

189146

101134

24

19247

68

186 34

178

177

253

126

163

33

89

108127

165

133

52

206

156

121

262

123

263

66

65 237

128

(c) pm

Figure 6 Continued


128

203

257

21265

132

155

235

189

101

146148

237

1922

4

6847

186178

177 263

66253

123

262

108206

52

156

183

184

49

246

191

232

153

61134

35

133

121

165

163

34

12789

33

126

(d) Evening






119875result =sum119899

119894=1accuracy

119894

119899 (20)



accuracy119894=


(21)



119873









10 20 30 40 50 60 70 80 900

02

04

06

08

12

1

Node number

Pred

ictio

n ac

cura

cy

SMMMLPR

(a) Nodes 1ndash92

100 110 120 130 140 150 160 170 1800

02

04

06

08

1

Node number

Pred

ictio

n ac

cura

cy

SMMMLPR

12


190 200 210 220 230 240 250 260 2700

02

04

06

08

1

Node number

Pred

ictio

n ac

cura

cy

SMMMLPR

12




SMM O2MM SMLP119873


2) 119874(119873) 119874(119873)

Storage space 119874(1198732) 119874(119873

3) 119874(119873

2) 119874(119873

2)


119873






10 20 30 40 50 60 70 80 9005

06

07

08

09

1

Node number

Pred

ictio

n ac

cura

cy

SMLPRO2MM

SMLPN

(a) Nodes 1ndash92

100 110 120 130 140 150 160 170 18005

06

07

08

09

1

Node number

Pred

ictio

n ac

cura

cy

SMLPRO2MM

SMLPN


190 200 210 220 230 240 250 260 27005

06

07

08

09

1

Node number

Pred

ictio

n ac

cura

cy

SMLPRO2MM

SMLPN



5 Conclusion





Acknowledgments




50

100

150

200

250

300


Num

ber o

f use

rs

SMMO2MM SMLPR

SMLPN


01 02 03 04 05 06 07 08 09 1

05

06

07

08

09

1

Accu

racy

rate

SMMO2MM SMLPR

120582 value

SMLPN



References



































RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation










Daily sample ΔTi

A




A A A AA

B

B

BD

F

E

C C

B

C

W(A B)

800 am 400 pm1200 pm 800 pm 1200 am 400 am 800 am








1198781 119878

119895 119878



119875119894(119860 | 119878

119895) =

119875119894(119860 119878119895)

119875119894(119878119895) 119895 = 1 119899 (16)

where 119875119894(119860 | 119878








119875119894(119860 119878119895) =

119891119894(119860 119878119895)

sum119898

119894=1119891119894(119860 119878119895) (17)



119895as



119875119894(119860) =

119899

sum

119895=1

120582119895119875119894(119860 | 119878

119895) 120582

119895=

120588119895

sum119899

119895=1120588119895

(18)



120582119895= 1


119875119894= 119875119894

HMM+ 119889 (119875

social119894

minus 119875119894

HMM) (19)

01 02 03 04 05 06 07 08 09 1

300

350

400

450

500

550

Loca

tion

quan

tity

Mean445



120582 value


where 119875119894










183

246

148

235

253

163

89

156

127

206 203

128

123

263

2

186

126

178

192

189134

132155

146

101

212

257

35

237

61

153

65

4

108

165

262

52121

133

47

17734

33

66

68

191

184

49

232

(a) am257

212

132155

189

47

186123

128

203

206

133

156

16552

262121

68 192

253

235

134

232246

191

49

184

183

146

101

148

237

153

61

35

177

178

108

263

126127

16334

66

89

33

2

4

65

(b) Noon

183

184

191

49

246

232

61

153

35257

212

132

148

203

155235

189146

101134

24

19247

68

186 34

178

177

253

126

163

33

89

108127

165

133

52

206

156

121

262

123

263

66

65 237

128

(c) pm

Figure 6 Continued


128

203

257

21265

132

155

235

189

101

146148

237

1922

4

6847

186178

177 263

66253

123

262

108206

52

156

183

184

49

246

191

232

153

61134

35

133

121

165

163

34

12789

33

126

(d) Evening






119875result =sum119899

119894=1accuracy

119894

119899 (20)



accuracy119894=


(21)



119873









10 20 30 40 50 60 70 80 900

02

04

06

08

12

1

Node number

Pred

ictio

n ac

cura

cy

SMMMLPR

(a) Nodes 1ndash92

100 110 120 130 140 150 160 170 1800

02

04

06

08

1

Node number

Pred

ictio

n ac

cura

cy

SMMMLPR

12


190 200 210 220 230 240 250 260 2700

02

04

06

08

1

Node number

Pred

ictio

n ac

cura

cy

SMMMLPR

12




SMM O2MM SMLP119873


2) 119874(119873) 119874(119873)

Storage space 119874(1198732) 119874(119873

3) 119874(119873

2) 119874(119873

2)


119873






10 20 30 40 50 60 70 80 9005

06

07

08

09

1

Node number

Pred

ictio

n ac

cura

cy

SMLPRO2MM

SMLPN

(a) Nodes 1ndash92

100 110 120 130 140 150 160 170 18005

06

07

08

09

1

Node number

Pred

ictio

n ac

cura

cy

SMLPRO2MM

SMLPN


190 200 210 220 230 240 250 260 27005

06

07

08

09

1

Node number

Pred

ictio

n ac

cura

cy

SMLPRO2MM

SMLPN



5 Conclusion





Acknowledgments




50

100

150

200

250

300


Num

ber o

f use

rs

SMMO2MM SMLPR

SMLPN


01 02 03 04 05 06 07 08 09 1

05

06

07

08

09

1

Accu

racy

rate

SMMO2MM SMLPR

120582 value

SMLPN



References



































RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation










183

246

148

235

253

163

89

156

127

206 203

128

123

263

2

186

126

178

192

189134

132155

146

101

212

257

35

237

61

153

65

4

108

165

262

52121

133

47

17734

33

66

68

191

184

49

232

(a) am257

212

132155

189

47

186123

128

203

206

133

156

16552

262121

68 192

253

235

134

232246

191

49

184

183

146

101

148

237

153

61

35

177

178

108

263

126127

16334

66

89

33

2

4

65

(b) Noon

183

184

191

49

246

232

61

153

35257

212

132

148

203

155235

189146

101134

24

19247

68

186 34

178

177

253

126

163

33

89

108127

165

133

52

206

156

121

262

123

263

66

65 237

128

(c) pm

Figure 6 Continued


128

203

257

21265

132

155

235

189

101

146148

237

1922

4

6847

186178

177 263

66253

123

262

108206

52

156

183

184

49

246

191

232

153

61134

35

133

121

165

163

34

12789

33

126

(d) Evening






119875result =sum119899

119894=1accuracy

119894

119899 (20)



accuracy119894=


(21)



119873









10 20 30 40 50 60 70 80 900

02

04

06

08

12

1

Node number

Pred

ictio

n ac

cura

cy

SMMMLPR

(a) Nodes 1ndash92

100 110 120 130 140 150 160 170 1800

02

04

06

08

1

Node number

Pred

ictio

n ac

cura

cy

SMMMLPR

12


190 200 210 220 230 240 250 260 2700

02

04

06

08

1

Node number

Pred

ictio

n ac

cura

cy

SMMMLPR

12




SMM O2MM SMLP119873


2) 119874(119873) 119874(119873)

Storage space 119874(1198732) 119874(119873

3) 119874(119873

2) 119874(119873

2)


119873






10 20 30 40 50 60 70 80 9005

06

07

08

09

1

Node number

Pred

ictio

n ac

cura

cy

SMLPRO2MM

SMLPN

(a) Nodes 1ndash92

100 110 120 130 140 150 160 170 18005

06

07

08

09

1

Node number

Pred

ictio

n ac

cura

cy

SMLPRO2MM

SMLPN


190 200 210 220 230 240 250 260 27005

06

07

08

09

1

Node number

Pred

ictio

n ac

cura

cy

SMLPRO2MM

SMLPN



5 Conclusion





Acknowledgments




50

100

150

200

250

300


Num

ber o

f use

rs

SMMO2MM SMLPR

SMLPN


01 02 03 04 05 06 07 08 09 1

05

06

07

08

09

1

Accu

racy

rate

SMMO2MM SMLPR

120582 value

SMLPN



References



































RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation










128

203

257

21265

132

155

235

189

101

146148

237

1922

4

6847

186178

177 263

66253

123

262

108206

52

156

183

184

49

246

191

232

153

61134

35

133

121

165

163

34

12789

33

126

(d) Evening






119875result =sum119899

119894=1accuracy

119894

119899 (20)



accuracy119894=


(21)



119873









10 20 30 40 50 60 70 80 900

02

04

06

08

12

1

Node number

Pred

ictio

n ac

cura

cy

SMMMLPR

(a) Nodes 1ndash92

100 110 120 130 140 150 160 170 1800

02

04

06

08

1

Node number

Pred

ictio

n ac

cura

cy

SMMMLPR

12


190 200 210 220 230 240 250 260 2700

02

04

06

08

1

Node number

Pred

ictio

n ac

cura

cy

SMMMLPR

12




SMM O2MM SMLP119873


2) 119874(119873) 119874(119873)

Storage space 119874(1198732) 119874(119873

3) 119874(119873

2) 119874(119873

2)


119873






10 20 30 40 50 60 70 80 9005

06

07

08

09

1

Node number

Pred

ictio

n ac

cura

cy

SMLPRO2MM

SMLPN

(a) Nodes 1ndash92

100 110 120 130 140 150 160 170 18005

06

07

08

09

1

Node number

Pred

ictio

n ac

cura

cy

SMLPRO2MM

SMLPN


190 200 210 220 230 240 250 260 27005

06

07

08

09

1

Node number

Pred

ictio

n ac

cura

cy

SMLPRO2MM

SMLPN



5 Conclusion





Acknowledgments




50

100

150

200

250

300


Num

ber o

f use

rs

SMMO2MM SMLPR

SMLPN


01 02 03 04 05 06 07 08 09 1

05

06

07

08

09

1

Accu

racy

rate

SMMO2MM SMLPR

120582 value

SMLPN



References



































RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation










10 20 30 40 50 60 70 80 900

02

04

06

08

12

1

Node number

Pred

ictio

n ac

cura

cy

SMMMLPR

(a) Nodes 1ndash92

100 110 120 130 140 150 160 170 1800

02

04

06

08

1

Node number

Pred

ictio

n ac

cura

cy

SMMMLPR

12


190 200 210 220 230 240 250 260 2700

02

04

06

08

1

Node number

Pred

ictio

n ac

cura

cy

SMMMLPR

12




SMM O2MM SMLP119873


2) 119874(119873) 119874(119873)

Storage space 119874(1198732) 119874(119873

3) 119874(119873

2) 119874(119873

2)


119873






10 20 30 40 50 60 70 80 9005

06

07

08

09

1

Node number

Pred

ictio

n ac

cura

cy

SMLPRO2MM

SMLPN

(a) Nodes 1ndash92

100 110 120 130 140 150 160 170 18005

06

07

08

09

1

Node number

Pred

ictio

n ac

cura

cy

SMLPRO2MM

SMLPN


190 200 210 220 230 240 250 260 27005

06

07

08

09

1

Node number

Pred

ictio

n ac

cura

cy

SMLPRO2MM

SMLPN



5 Conclusion





Acknowledgments




50

100

150

200

250

300


Num

ber o

f use

rs

SMMO2MM SMLPR

SMLPN


01 02 03 04 05 06 07 08 09 1

05

06

07

08

09

1

Accu

racy

rate

SMMO2MM SMLPR

120582 value

SMLPN



References



































RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation










10 20 30 40 50 60 70 80 9005

06

07

08

09

1

Node number

Pred

ictio

n ac

cura

cy

SMLPRO2MM

SMLPN

(a) Nodes 1ndash92

100 110 120 130 140 150 160 170 18005

06

07

08

09

1

Node number

Pred

ictio

n ac

cura

cy

SMLPRO2MM

SMLPN


190 200 210 220 230 240 250 260 27005

06

07

08

09

1

Node number

Pred

ictio

n ac

cura

cy

SMLPRO2MM

SMLPN



5 Conclusion





Acknowledgments




50

100

150

200

250

300


Num

ber o

f use

rs

SMMO2MM SMLPR

SMLPN


01 02 03 04 05 06 07 08 09 1

05

06

07

08

09

1

Accu

racy

rate

SMMO2MM SMLPR

120582 value

SMLPN



References



































RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation











50

100

150

200

250

300


Num

ber o

f use

rs

SMMO2MM SMLPR

SMLPN


01 02 03 04 05 06 07 08 09 1

05

06

07

08

09

1

Accu

racy

rate

SMMO2MM SMLPR

120582 value

SMLPN



References



































RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation


























RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation









Research Article A Location Prediction Algorithm with...

Documents

Transcript of Research Article A Location Prediction Algorithm with...