Adaptive search and the management of logistics systems. Base models for learning agents (Q1964480)

The book is divided into two parts, the fundamentals and the application of evolutionary adaptive systems. The first part gives a general overview and is an introduction for a reader not familiar with the topic. The second part is focussed on a case study. Starting with an introduction into basic logistic problems several methods and techniques related to those used in the context of that case study are discussed. The first chapter is a detailed introduction on ``Computational Intelligence'' (CI) and its use as a learning technique in artificial systems. ``Artificial intelligence'' (AI), one of CI's roots, is examined first to show the difference between CI and AI. Chapter two introduces adaptive systems as a subclass of feedback control systems. Especially the figures used allow for an easy understanding of the different models. Methods to coordinate decentralised systems close chapter two. Chapter three outlines the basic ideas of genetic algorithms in the context of evolutionary algorithms, then deals with some modifications and extensions of the basic principles. The adaptive behaviour of genetic algorithms, especially the fitness of search spaces, in static environments is scrutinized in chapter four. The aim of that chapter is to allow for a prediction of the performance of evolutionary algorithms. Especially the convergence of several approaches is considered. In Chapter five the use of interacting computational systems which are able to adapt to changing environments -- so called ``adaptive agents'' -- is discussed. Their sensing, acting, communicating and (machine) learning are reflected. Chapter six (the first chapter of part 2) deals with some basic logistic problems and whether they have to be enhanced by an encoding-decoding system or learning operators to be feasible for evolutionary algorithms. Several ways to encode and decode solutions to scheduling problems for evolutionary algorithms are discussed in chapter seven. Later on the job shop problem -- due to its complexity -- serves as a benchmark for solving scheduling problems. The author admits that adaptive scheduling is dominated by neighbourhood-search algorithms when using the job shop problem. Chapter eight re-examines the abilities of the adaptive scheduling approach in less restrictive environments than the job shop problem. According to the authors experiments, in those scenarios oriented ``towards real world scheduling'', the adaptive search algorithm ``turned out consistently superior'', after some improvements detailed in that chapter were made. Some drawbacks of the genetic algorithms are mentioned, especially their trouble in handling complex problems. Even when some improvements are made, they may not succeed in finding an optimal or near optimal solution, but they are able to find local optima in a short runtime. Two more applications for adaptive scheduling agents are presented in chapter nine: The scheduling of customer jobs in an existing firm and a control problem. The former emphasizes planning over control, the latter vice versa. The design of the adaptive agents for the respective case is outlined, then a simulation study is done. Discussions of the results follow. In these cases, the author is pleased with the success of his algorithm.