A Cultural Algorithm based on multilayer belief spaces and its application in neural network fault classifier

This paper presents a cultural algorithm (CA) based on multilayer belief spaces that selects the best belief space from the multilayer belief spaces to guide the search of population space. The selected best belief space exploits knowledge extracted during the search to improve the performance of an evolutionary algorithm. We integrate cultural algorithm based on multilayer belief spaces and evolutionary programming to develop a more efficient algorithm, called CMAEP used for global optimization, and the knowledge sources in the belief spaces of the cultural algorithm are specifically designed. The tests of optimizing benchmark functions show the approach is valid and effective. Then we apply it to optimize the weights and thresholds of BP network as fault classifier to discriminate chemical process steady faults. The simulations on Tennessee Eastman process (TEP) show CMAEP can effectively escape from local optima to find the global optimal value comparing with other optimization methods and the optimized BP network based on CMAEP can obtain a satisfied diagnosis result. it is successful to apply the algorithm in neural network fault classifier.