Chaos identification based on CMAC with replacing eligibility learning

In the conventional CMAC learning scheme, the correcting amounts of errors are equally distributed into all addressed weight, regardless the temporal credibility of those weights. In order to solve the temporal credit assignment problem of the CMAC, an improved CMAC neural network based on replacing eligibility learning concept was designed. The proposed improved leaning approach uses the replacing eligibility learning concept of the reinforcement learning to improve the prediction capability. The simulations for chaotic system identification show that the improved CMAC neural network is effective.