Applicational Perspectives of recent Developments in Dynamic Programming Methods for Herd M.pdfVIP

Applicational Perspectives of recent Developments in Dynamic Programming Methods for Herd Management Support Dina Notat No. 33 1995 Anders R. Kristensen1 and Erik Jørgensen2 Introduction During the last decade several methodological studies have contributed to the development of dynamic programming techniques as a framework for herd management support. A survey of such studies has been given by Kristensen (1994a). The main contributions have been the introduction of hierarchic Markov processes as a method for circumventing the curse of dimensionality (i.e. that models representing problems of the real world tend to be very large and, therefore, computationally prohibitive), the combination of dynamic programming techniques with methods for Bayesian updating, and methods for dealing with herd restraints like the milk quota or a limited supply of heifers. The hierarchic technique was originally presented by Kristensen (1988; 1991) and has been applied by Kristensen (1987; 1989), Broekmans (1992), Jørgensen (1993), Verstegen et al. (1994) and most recently by Houben et al. (1994; 1995a,b) in a very comprehensive study on economic optimization of decisions with respect to dairy cow health management (Houben, 1995). The combination of dynamic programming and techniques for Bayesian updating of traits has been introduced by Kristensen (1993) for the single trait case and later extended by Kristensen (1994b) to a multi-dimensional formulation using Kalman filter techniques. Independently of these studies, Kennedy Stott (1993) seem to have used similar methods in a dairy cow replacement model. Herd restraints in the form of a limited supply of heifers have been studied using an approximative method called parameter iteration by Ben-Ari and Gal (1986) and later on by Kristensen (1992). A recent study by Houben et al. (1995c) uses genetic algorithm