Marketing Management Support in Slaughter Pig Production
Ph.D. Thesis
Henrik Kure
The Royal Veterinary and Agricultural University, Copenhagen, January 1997
Go to:
Summary
Sammendrag (Danish Summary)
Table of contents (chapter and section headings of the thesis)
Download the thesisSummary
In Denmark approximately 19 million pigs are marketed for slaughtering yearly. The majority of these pigs are priced based on carcass merits and the manager of the slaughter pig operation determines when to market the individual pigs and/or groups of pigs. This problem of how to select and when to market (groups of) pigs is here defined as the "slaughter pig marketing management problem". Most managers solve this problem by simple calculations or rules of thumb and only very few applicable and operational tools for supporting the manager in this process exits. At the same a large amount of herd-specific data are produced every day, but only few of these data are, at present, utilized in a formalized way in the management of the operation in general, and in the management of the slaughter pig marketing in particular. The objective of the research project leading to this thesis was to investigate and/or develop models and methods for handling and solving the "slaughter pig marketing management problem", with special emphasis on utilizing the increasing (in quality and quantity) herd specific registrations. The overall aim was to develop the fundament of an applicable and operational Marketing Management Support Tool (MMST) to be used in the individual slaughter pig operation.
The MMST is represented logically and functionally by two subsystems: A general Belief Management System (BMS) used for updating the "belief in the state of the herd" based on herd specific registrations and a more specific Decision Support System (DSS) for optimizing marketing decisions. A stochastic animal growth model based on multi variate normal distributions is introduced and used as a representation of the state of the herd and as an interface between the BMS and the DSS.
The BMS is based on the EM-algorithm and other methods for reducing the bias in data caused by the selection of individual pigs for marketing and on the Kalman filter for updating the belief in the herd as new data are produced. Assuming that pigs are identified by batch-#, it is demonstrated how the BMS even in the case of very sparse data, returns fine and updated estimates of the state of the herd. One of the main features of the Kalman filter is demonstrated by examples: The quick adaption to changes in the system modeled. However, the results are based on simulated data and should not be interpreted without precautions.
"The slaughter pig marketing management problem" can be partitioned into two subproblems: (i) How to select and when to market individual pigs from batches and (ii) when to terminate (market the remainder of) the batch and insert a new batch of weaners. The two problems can be solved independently of each other and based on basic production economics and asset replacement theory. The DSS is based on these methods, but special methods are applied in order to cope with different selection criteria and with the uncertainty and variation that exists in biological systems as the slaughter pig operation. The results show that selection on carcass leanness as well as live weight is only slightly superior to selection on live weight only and very little financial room is left for performing the on-farm leanness measuring. The selection criteria is quite unaffected by changes in model parameters (e.g. prices and growth rates), while the optimal terminal marketing stage is more affected by such changes. As is the case for the BMS, the results are based on simulated data.
Some of the optimization algorithms in the DSS are based on Recursive Dynamic Programming; an optimization method originating from an alternative approach to Dynamic Programming: Recursions. The method is interesting from a computational as well as a theoretical point of view; it is efficient and general. These features are demonstrated by examples and by the DSS.
The thesis is concluded by a general discussion of theoretical and applicable aspect of the research and the proposed models and methods, with special emphasis on the aspects of implementation as a MMST. It is concluded that the models and methods have the potential for implementation as tools for producing general guidelines and, if or when more and better registrations on individual animals are available, as MMSTs.
Sammendrag (Danish summary)
I Danmark leveres der årligt ca. 19 millioner slagtesvin til slagtning. Størstedelen af disse bliver afregnet på basis af slagtekroppens egenskaber og driftslederen på slagtesvinebedriften bestemmer hvornår enkelte svin og/eller grupper af svin skal leveres. Dette problem om hvordan svin skal udvælges og hvornår (grupper af) svin skal leveres defineres her som "Slagte-svine-leveringsproblemet". De fleste driftsledere løser dette problem ved hjælp af simple beregninger og tommelfingerregler og der eksisterer kun få anvendelige og operationelle værktøjer der kan støtte driftslederen i denne proces. Samtidig produceres der hver dag en stor mængde besætningsspecifikke data, men kun en lille del af disse data udnyttes pt. på en formaliseret måde i styringen af bedriften i almindelighed og i styringen af leveringen af slagtesvin i særdeleshed. Formålet med det forskningsprojekt som resulterede i denne afhandling var, at undersøge og/eller udvikle modeller og metoder til håndtering og løsning af "Slagtesvine-leveringsproblemet", med speciel vægt på udnyttelsen af den stigende (i kvalitet og kvantitet) mængde besætningsspecifikke registreringer. Det overordnede mål var at skabe fundamentet for et anvendeligt og operationelt LeveringsStyrings-StøtteSystem (LSSS) som kan anvendes på den enkelte bedrift.
LSSS'et repræsenteres logisk og funktionelt ved to undersystemer: (i) Et generelt Forventnings StyringsSystem (FSS) som bruges til opdatering af "forventningen om bedriftens tilstand" baseret på besætningsspecifikke registreringer og et mere specifikt BeslutningsStøtteSystem (BSS) til optimering af beslutningerne vedrørende levering. En stokastisk dyrevækstmodel baseret på multivariable normalfordelinger introduceres og bruges som repræsentation af besætningens tilstand og som et mellemled mellem FSS'et og BSS'et.
FSS'et er baseret på EM-algoritmen og andre metoder til reduktion af bias (skævhed) i data, forårsaget af udvælgelsen af dyr til slagtning og på Kalman-filteret til opdatering af forventningen om besætningens tilstand, efterhånden som nye data bliver produceret. Under antagelsen om at holdnummeret er kendt på alle svin, demonstreres det hvordan FSS'et selv i tilfælde af meget "tynde" datasæt, giver gode og opdaterede estimater af besætningens tilstand. En af de væsentligste egenskaber ved Kalman-filteret demonstreres ved eksempler: Den hurtige tilpasning til ændringer i det modellerede system. Det bør dog anføres at resultaterne er baseret på simulerede data og derfor skal fortolkes med varsomhed.
Slagtesvine-leveringsproblemet kan opdeles i to delproblemer: (i) Hvordan skal dyr udvælges fra holdet og hvornår skal de leveres og (ii) hvornår skal et hold afsluttes (de resterende svin leveres) og et nyt hold smågrise indsættes? De to problemer kan løses uafhængigt og baseret på basal produktionsøkonomi og udskiftningsteori. BSS'et er baseret på disse metoder, men specielle metoder er anvendt for at kunne håndtere forskellige udvælgelseskriterier og usikkerheden og variationen forbundet med biologiske systemer som slagtesvinebesætningen. Resultaterne viser at udvælgelse på baggrund af både kødprocent og levendevægt kun er svagt fordelagtigt i forhold til udvælgelse på baggrund af levendevægt alene og overskuddet til betalingen af målingen af kødprocenten på bedriften er meget lille. Selektionskriterierne er relativt upåvirkede af ændringer i modelparametre (som f.eks. priser og vækstrate), mens den optimale holdlængde påvirkes mere af sådanne ændringer. Som for FSS'et gælder det at resultaterne er baseret på simulerede data.
Nogle af optimeringsalgoritmerne der anvendes i BSS'et er baseret på Rekursiv Dynamisk Programmering; en optimerings algoritme som har sit udspring i en alternativ indgangsvinkel til Dynamisk Programmering: Rekursioner. Metoden er interessant udfra en beregningsmæssig såvel som en teoretisk synsvinkel: Den er effektiv og general. Disse egenskaber demonstreres ved eksempler og via BSS'et.
Afhandlingen afsluttes med en generel diskussion af de teoretiske og anvendelsesmæssige aspekter af forskningsarbejdet og de foreslåede modeller og metoder, med speciel vægt på implementering som et LSSS. Det konkluderes at modeller og metoder har potentiale til at kunne implementeres som værktøjer til udarbejdelse af generelle retningslinjer og, hvis eller når flere og bedre registreringer på enkeltdyrsniveau bliver tilgængelige, som et LSSS.
Table of contents
1. General introduction
1.1 Objectives of the project, 1.2 Outline of the thesis
2. Basic concepts of Computer Based Information Systems in slaughter pig marketing
2.1 Introduction, 2.2 The need of information, 2.2.1 The management cycle, 2.2.2 The value of information, 2.3 Computer-Based Information Systems, 2.3.1 The fundamentals of a CBIS, 2.4 A computer based slaughter pig marketing information system, 2.4.1 Coping with biological variance and uncertainty, 2.4.2 The model
3. A slaughter pig marketing belief management system
3.1 Introduction, 3.1.1 The slaughter pig finishing operation, 3.1.2 Incomplete, biased and imprecise data, 3.1.3 Modeling and estimating the "Belief in the state of the herd", 3.2 Models and methods, 3.2.1 The animal growth model, 3.2.2 Herd model, 3.2.3 Belief updating - the Kalman filter, 3.2.4 Bias reduction, 3.2.5 Models and methods testing, 3.3 Results, 3.3.1 Sensitivity of TMT and EM3, 3.3.2 Adaption to changes in growth rate, 3.3.3 Covariance estimation, 3.4 Discussion
4. Optimal Slaughter Pig Marketing
4.1 Introduction, 4.1.1 Slaughter pig marketing defined, 4.1.2 Optimal marketing, 4.1.3 Objectives, 4.2 Material and methods, 4.2.1 Animal growth model, 4.2.2 General assumptions of optimizations, 4.2.3 Selection of individual pigs, 4.2.4 Terminal marketing. Constant and inflexible weaner supply, 4.2.6 Terminal marketing. Varying and flexible weaner supply, 4.2.7 Probabilistic Optimization, 4.2.8 Model testing, 4.3 Results, 4.3.1 Selection of individual pigs, 4.3.2 Terminal marketing, 4.4 Discussion
5. Recursive Dynamic Programming
5.1 Introduction, 5.2 Recursive Dynamic Programming, 5.2.1 The deterministic data model, 5.2.2 The Problem, 5.2.3 Bellman's Principle of Optimality, 5.2.4 solution - a direct derivation, 5.2.5 Memory functions, 5.2.6 Function based model, 5.3 Generalized RDP, 5.3.1 The model, 5.3.2 The solution function, 5.3.3 Definitions of the type Gain and of the operators, 5.4 Implications, 5.4.1 Heuristic rules, 5.4.2 Bounding, 5.4.3 Value Iteration, 5.4.4 Hierarchic Markov Processes, 5.5 Conclusion
6. General discussion
6.1 General limitations, 6.2 The animal growth model, 6.3 Optimization methods, 6.4 Recursive Dynamic Programming, 6.5 Implementation as a Marketing Management Support Tool, 6.6 Other applicable issues
Download the thesis
Her you can download the thesis (116 pages, A4 format and prepared for 2-page printing) in these formats:
- PostScript (6.6 MB)
- PKZipped PostScript (1.4 MB)
- Adobe Acrobat (2.5 MB)
- PKZipped Acrobat (1.6 MB)
If have you haven't got a viewer for Acrobat files you can get a free copy here .