Research On Improved Fuzzy Neural Network Based Reinforcement Learning And Related Application

Fuzzy neural network is a new kind of artificial intelligence technique．and itcombines both advantages of artificial neural network and fuzzy logic inference systemto make fuzzy pattern recognition and fuzzy logic inference have self-learningability． It has become a hot topic of the machine learning community in recent years.Reinforcement learning is an important machine learning method．Reinforcementlearning could learn the optimal policy of the dynamic system through environmentstate observation and improve its behavior through trial and error with theenvironment．Reinforcement learning has the quality of low requirement for the priorknowledge. In this paper, a clustering algorithm is developed based on thereinforcement learning paradigm. Then the fuzzy neural network with this algorithm isapplied to Radar Emitter Recognition.On this basis, we choose reference-nets formodeling and deducing the process of platform recognition.Through doing some research on the theory of fuzzy neural network, we designand implement a five-layered feed-forward neural network; named PseudoOuter-Product based Fuzzy Neural Network (POPFNN). Then reinforcement learningbased on the optimal policy approach is employed to generate fuzzy membershipparameters in fuzzy neural network. In this foundation, an algorithm of gettinginconstant number of membership parameters is proposed. Finally the improvedgenerating membership parameters algorithm based on reinforcement learning is usedfor recognizing radar emitter. Through feature extraction and normalization processingon Radar Emitter data, the pretreatment radar simulation data is used as experimentaldata sets, then compare the improved algorithm and the original algorithm by using theradar data sets, experiment shows that the improved algorithm works better.On theprocsss of platform recognition,reference-nets are simple and effective in knowledgerepresentation.