Advanced Herd Management Co se int od ction Course ... fileTransfer methods to other herd management...
Transcript of Advanced Herd Management Co se int od ction Course ... fileTransfer methods to other herd management...
1
Advanced Herd ManagementCo se int od ction
Slide 1
Course introduction
Anders Ringgaard Kristensen
Outline
PreconditionsOutcome: What are you supposed to learn?The framework and definition of herd managementThe management cycleObjectives of production, utility theoryClassical production theoryCl i l l h
Slide 2
Classical replacement theoryLimitation of classical theoriesOutline of the courseTeachersThe concept of uncertainty
Preconditions
Courses• Animal production: ”Husdyrproduktion”• Mathematics: ”Matematik og
modeller”/”Matematik og planlægning”• Statistics ”Statistisk dataanalyse 2”• Mandatory first year (economics etc)
Slide 3
Mandatory first year (economics etc)
2
Brush-up courses …
The course will start up with brush-up courses of• Probability calculus and statistics• Linear algebra
Slide 4
Learning outcome
After attending the course students should be able to participate in the development and evaluation of new tools for management and control taking biological variation and observation uncertainty into account.
Slide 5
Outcome - knowledge
After completing the course the student should be able to:Describe the methods taught in the courseExplain the limitations and strengths of the methods in relation to herd management problems.Give an overview of typical application areas of the methods.
Slide 6
3
Outcome - skills
After completing the course the student should be able to:Construct models to be used for monitoring and decision support in animal production at herd level.Apply the software tools used in the course.
Slide 7
Outcome: Competencies:
After completing the course the student should be able to:Evaluate methods, models and software tools for herd management.Transfer methods to other herd management problems than those discussed in the course.Interpret results produced by models and software tools.
Slide 8
A pig (an animal)
Medicine
Slide 9
Feed
MeatManure
Piglets
Milk
4
Pig production = N × a pig?
Medicine
Slide 10
Feed
MeatManure
Piglets
Milk
Medicine
Pig production = N × a pig?
Slide 11
Feed
MeatManure
Piglets
Milk
Pigs:•Ages•Groups•Individuals
B ildi Fi ld N i hb i
Pig production
Slide 12 Feed
Buildings Fields
Farm hands
Owner
Neighbors, society, consumers
5
Elements of production I
The factors (input to production)• Animals• Feed• Buildings, inventory• Labor• Management• Veterinary services• Energy
Slide 13
• …
Elements of production II
Objectives: Maximization of the farmer’s welfare:• Income (personal)• Leisure time (personal)• Animal welfare (animals)• Working conditions (farm hands)• Environmental preservation (future generations)
Slide 14
• Prestige (personal)• Product quality (consumers)
Elements of production III
Constraints limiting production• Physical (land, housing capacity, storage capacity)• Economical (capital, prices)• Legal (laws)• Personal (skills, education)
Slide 15
6
Definition of herd management
Having discussed the three key elements:• Factors (input to production)• Objectives (farmer’s welfare)• Constraints (limitations)
we are now able to define what we mean by Herd Management:
Herd management is a discipline serving the purpose of
Slide 16
g p g p pconcurrently ensuring that the factors are combined in such a way that the welfare of the individual farmer is maximized subject to the constraintsimposed on his production.
A dynamic optimization problem under constraints.Decisions! We decide how to combine the factors.
The management cycle: A never ending story
Slide 17
The management cycle: Classical theories
UtilityTheory,Ch. 3.
N l i l
(ScarceResources)
Basic
Slide 18
Neo-classicalProductionTheory,Ch. 4.
(Animal science,Production function)
ProductionMonitoring,Ch. 5.
7
Herd Management Science
Basic level:• As we define the basic level, it consists of
• Utility theory• Neo-classical production theory• Basic production monitoring• (Animal nutrition, animal breeding, ethology,
fa m b ildings)
Slide 19
farm buildings)• What any animal scientist should know about
management• The starting level of this course• Briefly revised today
Utility theory
We need a criterion for comparison of plans (“ways” to produce).Several concerns:
• The farmer• The staff• Consumers
Animals
F b ildi Land i hb i
Slide 20
• The animals• Environment• …
Who decides the weighting?My answer: The farmer!
Slide 6 OwnerFarm handsFeed
Farm buildings Land Neighbors, society, consumers
Farmer’s preferences
The farmer has/defines a list of concerns:• Own direct concerns:
• Income, u1
• Leisure time, u2
• Prestige, u3
• ... • Indirect concerns (because he cares for others)
Slide 21
( )• Animal welfare, u4
• Working conditions, u5
• Environment, u6
• Product quality, u7
• …The farmer knows/decides the weightingThe “items” on the list (u1, u2, … , uk) are called
attributes of the farmer’s utility.
8
When is ”something” an attribute?
When it directly influences the subjective welfare of the farmer.
May NOT be an attribute:• Average milk yield of cows• Average daily gain of slaughter pigs• Animal welfare, ”because animals at a high level of
welfare also produce at a higher level”.
Slide 22
May be an attribute:• Monetary gain• Leisure time• Animal welfare, if the farmer is willing to accept that
it to some extent decreases the levels of other attributes.
Consequences measured by attributes
Attribute
Stage
1 2 … T
1 u11 u12 … u1T
2 u11 u22 … u2T
Slide 23
2 u11 u22 … u2T
… … … … …
k u11 uk2 … ukT
At any stage, the attributes will depend on the production Yt andthe factors xt. The relation is given by the attribute function h:
ut = h(Yt , xt)
Aggregation of attributes: Utility function
The utility function• Aggregation over time
• Monetary gain• Animal welfare
Slide 24
• …• Aggregation over attributes
Expected Utility Theorem: Maximization of U is all we need to care about!
Refer to Chapter 3 for details!
9
Production function
Slide 25
Production function
Slide 26
In classical production theory, the uncertainty represented by the e’s is ignored.
Neo-classical production theory
Answers 3 basic questions:• What to produce.• How to produce.• How much to produce.
Marginal considerationsBasic principle: Continue as long as the
Slide 27
marginal revenue, MR, exceeds marginal costs, MC. At optimum we have MR = MC.
10
How much to produce
One factor x and one product yPrices px and py
A production function y = f(x).Profit u(x) = ypy – xpx = f(x)py – xpx
Problem:
Slide 28
Problem: • Find the factor level that maximizes the profit
How much to produce
Maximum profit where u’(x) = 0.u(x) = f(x)py – xpx
u’(x) = f’(x)py – px
u’(x) = 0 ⇔ f’(x)py = px
Slide 29
Maximum profit where:• Marginal revenue = Marginal
cost!
How much to produce
0,8
1
Total revenue, f(x)py
Slide 30
-0,2
0
0,2
0,4
0,6
Average revenue, f(x)py/x
Marginal revenue, f’(x)py
11
How much to produce, logical bounds
0,8
1
Total revenue, f(x)py
Slide 31
-0,2
0
0,2
0,4
0,6
Average revenue, f(x)py/x
Marginal revenue, f’(x)py
How much to produce, optimum
0,8
1
Total revenue, f(x)py
Slide 32
-0,2
0
0,2
0,4
0,6
Average revenue, f(x)py/x
Marginal revenue, f’(x)py
Price of factor px
Classical replacement theory
The replacement problem in a broad sense is one of the most important decision problems in animal production.
Dynamics: What we decide at this stage (keep/replace) may influence production in many future stages.
Many other decision problems relate to the replacement
Slide 33
y p pproblem:• Insemination• Treatment for diseases• Feeding level• …
A correct handling of the other problems implies that the question of replacement must be taken into account.
12
Definition
Replacement:• When an existing asset is substituted by a new
one with (more or less) the same function.• Examples:
• Light bulbs• Cars
So s
Slide 34
• Sows• Milking robots
Female production animals:• (Ewes, mink, goats)• Sows• Dairy cows
• Two levels:• Optimal lactation to replace• Optimal stage of lactation to replace
Replacement problems in animal production
Slide 35
• Repeatability of milk yield over lactation rather high (as opposed to litter size in sows).
Replacement problems in animal production
Technical:• Examples:
• Farm buildings• Equipment & machinery
• Very similar to the sow replacement problem, except for the
Slide 36
y p p , p”biological” variation
• Technological improvements probably more important than the corresponding genetic improvement in cows.
• Marginal/average considerations apply well
13
Slaughter calves:• If housing capacity is limited and replacements are
available, the problem is in agreement with classical theory.• Marginal/average considerations
Slaughter pigs:
Replacement problems in animal production
Slide 37
g p g• Two levels:
• When to deliver individual pigs (animal level)• When to deliver the remaining pigs (batch level)
Broilers:• At batch level (no animal level) in agreement with classical
theory.• Contracts may limit the decisions of the farmer
A chain of assets
Asset 1 Asset nAsset 3Asset 2
Slide 38
tr
How do we determine an optimal value for tr - the length of the period to keep each asset in the chain?
Optimal time for replacement
Assume that the price of a new asset is SThe salvage value of the asset at time t is
st
The net returns from the asset in stage (time step) t is rt
Slide 39
The total net revenue T(t) from the asset if it is replaced at stage t is then
14
Optimal time for replacement
Average net revenue, if replaced at time t:
Marginal net revenue at time t
Slide 40
g
Optimal replacement time where
Replace where:marginal revenue = average revenue
Graphical illustration
The replacement problem
200
300
Slide 41
-200
-100
0
100
200
0 5 10 15 20
Time
Reve
nue
MarginalAverage
Marginal revenue
Typically decreasing because of decreasing productivity and increasing maintenance costs.
The net returns adjusted for change in salvage value.
The marginal curve crosses the average curve where the average is maximal.
Slide 42
15
Limitations of neo-classical theory
Static approach:• Immediate adjustment• Only one time stage
Deterministic approach• Ignores risk
Slide 43
Ignores risk• ”Biological variation”• Price uncertainty
Knowledge representation (knowledge considered as certain):• Unobservable traits• ”Production functions”• Detached from production: No information flow from
observations. • No updating of knowledge.
Limitations of classical replacement theory
Uncertainty: The classical replacement theory assumes full certainty about the marginal profit function, the investment costs and all prices. As discussed in details in Chapter 2, uncertainty is an inherit property of the decision making process in herd management. The uncertainty is partly a consequence of imperfect knowledge, and partly of random variation.
Uniqueness: The general theory implicitly assumes that the marginal and average profit functions are as h i Fi 4 5 ith i l d t i d
Slide 44
shown in Figure 4.5 with a uniquely determined intersection. For several applications the intersection is not unique. This is, for instance, the situation in dairy cows, where the average and marginal profits are as shown in Figure 4.7.
Availability: The theory assumes that a new asset for replacement is always available.
Background for course
Structural development in the sector• Increasing herd sizes• Decreasing labour input
Technological development• Sensors, automatic registrations
Slide 45
Sensors, automatic registrations• Computer power• Networking
Methodological development• Statistical methods• Operations Research
16
Outline of course - I
Part I: • Brush-up course on
• Probability calculus and statistics• Linear algebra• ”Advanced” topics from statistics
• Basic production monitoring
Slide 46
• Registrations and key figures• Analysis of production results
Outline of course - II
Part II: The problems to be solved• From registrations to information, value of
information, information as a factor, sources of information
• Decisions and strategies, definition and knowledge foundation
• Consequences of decisions and states
Slide 47
Consequences of decisions and states• Visualisation and user interfaces
Outline of course - III
Part III: The methods to be used• State of factors
• Monitoring and data filtering• Bayesian networks
• Decision support• Decision graphs
Slide 48
• Simulation (Monte Carlo)• Linear programming (low priority)• Markov decision processes (dynamic programming)
• Mandatory reports
17
Teachers
Part I:• Anders Ringgaard Kristensen• Cécile Cornou
Part II:• Anders Ringgaard Kristensen
Part III:
Slide 49
Part III:• Anders Ringgaard Kristensen• Cécile Cornou (Post Doc IPH)• Tina Birk Jensen (Post Doc IPH)• Nils Toft (associate professor, • Guests:
• Thomas Nejsum Madsen (TNM Consult)• Thomas Algot Søllested (Egebjerg)• Tage Ostersen (Master student)• Lars Relund Nielsen (University of Aarhus)• Others …
Mandatory reports
4 minor reports must be handed in• Based on the exercises
At least 3 must be approved in order to attend the oral exam
The 4 reports are distributed over the following methods:• Linear programming• Monitoring and data filtering
Slide 50
• Monitoring and data filtering• Markov decision processes• Bayesian networks (including decision graphs)
The web
AbsalonHome page of the course
• http://www.prodstyr.ihh.kvl.dk/vp/• Course description• Plan
• Pages for each lesson with a description of the li i f
Slide 51
contents, literature, exercises, software to use etc.
18
Exercise, uncertainty
Production function:
• - milk yield given energy, protein and fat
f x x xc x c x c x c x c x c x c x x c x x c x x
( , , )1 2 3
11 12
22 22
33 32
1 1 2 2 3 3 12 1 2 13 1 3 23 2 3
=
+ + + + + + + +
Slide 52
fat
Adding uncertainty, the actual milk yield is
Y = f(x1,x2,x3) + e
Uncertainty, II
Adding uncertainty to production function:• Considerable improvement, BUT• Significant uncertainty about true energy, protein
and fat content still ignored• Example, only considering energy
Slide 53
Uncertainty - III
Silage obs.* Silage true
Concentr.*
Ration Milk yield*
Herd size*
Slide 54
True energy content of silage is unknownThe precision of the observed content depends
heavily on the observation method (standard value from table, laboratory analysis etc.)
19
Uncertainty - IV
Effects of decisions will be over-estimated if uncertainty about• true state• factor characteristics• factor effects
is ignored
Slide 55
is ignored.Wrong decisions may be made.
Uncertainty, V
Baysian networks with decisions and utilities added (student project).
Silage obs.* Silage true Ration Milk yield*
Slide 56
g g
Concentr.*
y
Herd size*Method
MixPrice Cost Rev.
Uncertainty - VI
Uncertainty is not the opposite of knowledgeUncertainty is a property of knowledgeReduction of uncertainty is often possible at some
cost!Reducing uncertainty is not always profitable.
Slide 57