Computerized management support for swine breeding farms

Abstract

1. INTRODUCTIONThe investigations described in this thesis have been directed towards computerized management support for swine breeding farms, focused on sow productivity and profitability. The study is composed of three basic parts: (1) basic description and definition of farm management and management information systems (MIS), (2) individual farm analysis, and (3) sow replacement optimization. As part of the study, a MIS on the personal computer named CHESS ( C omputerized H erd E valuation S ystem for S ows) has been developed. CHESS is primarily intended to support farm managers and other livestock specialists in analyzing the economic situation of individual sow-herds. Data, knowledge and expertise from many sources are required to find solutions to problems in these fields. As most problems cannot be solved with mathematical problem-solving techniques only (such as dynamic programming), CHESS consists of both decision support systems (DSS) and expert systems (ES). This study should also give insight into the underlying question of whether the combination of DSS and ES is of any advantage in formulating and solving major problems in the abovementioned fields.Basic concepts of computerized support for farm management decisions are described in Chapter 1. Good management is essential for efficient and effective operation of any farm, but the concept of management is nebulous and difficult to define. Therefore, scope and definition of farm management is discussed first, by paying attention to the three major management functions: planning, implementation and control. Then, impact of recent advances in computer technology on farm management and concept of MIS are described. Two potential applications of MIS are presented: DSS and ES. A DSS can be described as an information system that supports the process of making decisions. A DSS allows the decision maker to retrieve data, generate and test alternative solutions during the process of problem solving, and incorporates a variety of models. An ES can be defined as a computer program using expert knowledge to attain high levels of performance in a narrow problem area, and thus can be considered as a modelling of the human reasoning process, making the same decisions as its human counterpart. The current challenge in building MIS is to incorporate DSS and ES to create an effective tool for farm managers. This should take place within a general framework for integrated information systems. In the Netherlands, such a framework is called an information model.2. INDIVIDUAL FARM ANALYSISThe second basic part of this study is concentrated on individual farm analysis, which can briefly be described as the global analysis of technical and economic farm records in order to find strong and weak elements in management. In Chapter 2, a systematic and objective methodology for individual farm analysis has been developed and discussed. The methodology includes three types of analysis: (1) trend analysis, comparing the actual herd performance against predictions based on the herd's historical data, (2) comparative analysis, comparing the actual herd performance with the average performance of similar herds, and (3) comparative trend analysis, comparing the historical development of the herd performance with the average development of similar herds. In each type of individual farm analysis the following four stages are being considered: tracing deviations, weighting deviations, further analysis of deviations, and evaluation of individual farm performance.This methodology for analyzing the economic and technical records of individual swine breeding herds has been incorporated into CHESS. The individual farm analysis part of CHESS consists of both one DSS and three ES, designed in a modular manner. The DSS identifies and assesses the importance of relevant deviations between performance and standards. Its output is used in the ES that try to find strengths and weaknesses by combining and evaluating the previously identified deviations. As a supporting technique for the farm manager, the system can also be used for determining the maximum amount that could be spent in exploiting or improving farm performance.In Chapter 3, the validation procedure used for the individual farm analysis part of CHESS is outlined. Validation can be described as comparing CHESS with the observed world. It is generally considered as an important and difficult stage in developing computerized systems. This is especially the case for ES where symbolic problem-solving techniques and heuristics are used to draw conclusions. A methodology for validating this part of CHESS has been developed and described. Both the DSS and ES components of CHESS have been validated using, among others, a sensitivity analysis. Furthermore, a field test of the integrated system as a whole has been carried out. The field test resulted in a test agreement between CHESS and experts of about 60%. The percent mis-classification error turned out to be 4% only. The knowledge of the experts has thus successfully been incorporated into CHESS. So, CHESS shows to be a promising tool in performing - theoretically sounded but widely accessible - individual farm analyses.3. SOW REPLACEMENT OPTIMIZATIONIn the third basic part of the study, attention is focused on one of the strong or weak elements resulting from the individual farm analysis: the sow replacement policy. In Chapter 4, a (stochastic) dynamic programming model (DP-model) on the PC, which has been added to CHESS, is introduced to determine the economic optimal replacement policy in swine breeding herds. This optimal policy maximizes the present value of expected annual net returns from sows present in the herd and from subsequent replacement gilts over a given planning horizon. Sows are described in terms of parity number, production level in previous parity, production level in the next to previous parity, and number of unsuccessful breedings in the current parity. As the DP-model includes a large number of state and decision variables, a major issue in this chapter is how to cope with the resulting curse of dimensionality. This results in an alternative DP-model structure. Furthermore, a sensitivity analysis has been carried out to achieve insight into possibilities for further reduction of the model size. In this process, the quality of the results obtained are weighed against the amount of computation time needed to carry out the calculations on the PC. These two variables are especially sensitive to reductions in the maximum number of parities considered. In the basic situation, 7 minutes and 18 seconds of central processor time are needed for performing the optimization on a PC with an 80286 micro-processor and an 80287 math-processor.The zootechnical-economic aspects of the DP-model are described in Chapter 5. Besides determining the economic optimal replacement policy, the model calculates the total extra profit to be expected from attempting to retain an individual sow until her optimal lifespan and not replacing her immediately. This total extra profit, called Retention Pay-Off (RPO), is an economic index which makes it possible to rank sows within a herd on future profitability and, therefore, can be used as a management guide in culling decisions. Given the probabilities of involuntary replacement, the optimal replacement policy results in an average herd life of 5.5 parities. The corresponding present value of expected annual net returns per sow equals Dfl. 852. The optimal economic life of average producing sows emerges as 9 parities. The optimal policy is most sensitive to the difference between the cost of a replacement gilt and the slaughter value of culled sows. Moreover, conception rates have a considerable effect on net returns, replacement rate, and average herd fife.Usually, replacement decisions are not based on productive and reproductive sow performance only, but also on more qualitative sow characteristics such as lameness and leg weakness, mothering characteristics and udder quality (Chapter 6). Therefore, CHESS has been extended with an ES for the economic optimization of sow replacement decisions focused on such qualitative characteristics. The ES is integrated with the DP-model by using its output as input. The reasoning process of the ES is focused on three major elements: (1) clinical sow deviations, (2) deviations in number of pigs weaned per litter, in relation to (3) deviations in uniformity of the litter. Deviations of the first type always result in the advice to replace the sow by a replacement gilt. After deviations of the second and/or third type have been found, they are weighted economically by the ES. The RPO values determined by the DP-model are then adjusted by the ES based on these economic weights. The method of acquiring the knowledge (rules of thumb, heuristics) for the ES involved direct interviews with a domain expert. Real and example problems have been used in these interviews. Formal validation results indicate a high correspondence between the system's ranking of 30 blind test scenarios with the rankings provided by both the domain expert and a second expert. The adjusted RPO values turned out to be more complete management guides in cuffing decisions.4. MAIN CONCLUSIONSThe main conclusions of the present study are:- Recent advances in computer technology enable the support of realistic farm management decisions on the personal computer. As applications on the personal computer are suitable for use in the field, the relevance of on-farm decision support increases.- Integrated decision support and expert systems make it possible to formulate and solve major problems in the fields of individual farm analysis and sow replacement optimization.- In finding strengths and weaknesses in the management of individual - swine breeding - farms, three types of analysis are to be recommended: (1) trend analysis, comparing the actual her

Cite this work

No public copy is stored for this record, so only the metadata is cited.

Stored only when both papers are in the corpus — 28 citations exist upstream.

Papers that cite the same references — the relational fallback while embeddings are still backfilling.