9634712 Bean This research involves the development of theory and algorithms to solve nonhomogeneous Markov decision problems. The application area to be considered results from just-in-time production and distribution practices. Massive data requirements and extremely long computation times make solving nonhomogeneous Markov a formidable task. This research will develop theory that isolates the data important to making good decisions from the large volume of data that have no direct effect. The primary approach will integrate forecast horizons, parametric Markov decision processes and genetic algorithms. This theory will be embedded in algorithms that will be implemented and evaluated against data from production and transportation problems. The relatively high-risk research objectives of (1) developing theory for solving nonhomogeneous Markov through treatment as parametric Markov decision processes and acceleration of parametric Markov decision processes algorithms using sampling information and (2) developing implementable algorithms to solve nonhomogeneous Markov using forecast horizon theory to reduce problem size, genetic algorithms to eliminate the bulk of the computation, and then the results of the genetic algorithms to accelerate parametric Markov decision processes algorithms to obtain a provably optimal solution have an excellent chance of being met. The result will be a unique hybrid of artificial intelligence-based algorithms that result in fast computation times and provable optimality.
