演化式合作投票決策系統

Abstract

Voting is a common method used to make a decision in the real word. Inthis thesis, we introduce a self-learning decision making system based onvoting. We make use of rule base and genetic algorithms to establish thesystem. When multiple experts are involved in a problem solving situation, votingis an effective method to make a decision. In an expert system, the rules areconsidered various experts, thus, with voting we will have a good decisionmaking system. In order to obtain good rules, we train the rule base by usingreinforcement learning and niche theory. The concept of niche can beimplemented by employing a sharing function. However, the traditional sharing function is ineffective for chromosomeswhich take the form of rules. We present a new sharing function called theshareholder sharing function for rule base design. After training, the chromosomes can organize themselves to establish an effective rule-basedsystem base with a voting mechanism. In this thesis, we tested our system in three experiments, includingmultiplexer simulation, iris classification and the prediction of stock price.According to the results, we conclude that our method is simple and effective.Furthermore, we designed a visual training program to observe the process ofself- organization of the rule base in an on-line manner.