Dynamic programming as a framework for decision support in animal production
Anders Ringgaard Kristensen, Dina Kvl, Royal Veterinary · 1994
The sequential approach and the stochastic nature of dynamic programming makes the method well suited as a framework for operational decision support in animal production. In practise, however, we face the general problem of how to represent the traits of an animal. A general trait model based on Bayesian updating and Kalman filter techniques is suggested as a way of representing and combining traits of different nature. An other problem in practise is "the curse of dimensionality", i.e. that models tend to reach prohibitive sizes. A notion of multi-level hierarchic Markov processes is introduced in order to circumvent this problem. The idea is to split up the state space according to the variability of the individual state variables. With two levels the model performance is remarkable increased. The benefit of additional levels is not yet known.