Multilayer Algorithm Improvement And Application For Multi Layer RBF Network

This paper presents the improved clustering algorithm for multilayer radial basis functionnetwork. Computer experiments revealed that this algorithm has higher accuracy of functionapproximation than the multilayer of radial basis function network obtained by the basic clusteringalgorithm and adaptive genetic algorithm. This algorithm is applied to Logistic and Mackey Glassof chaotic time series multi-step prediction, and the prediction step is greatly increased and theprecision is higher than the basic clustering algorithm and adaptive genetic algorithm.Finally, in this paper, the adaptive genetic algorithm of multilayer radial basis functionnetwork is improved by replacing the fitness function by generalized cross rate instead of thegeneralized information critierian, so that the network can be used in nonlinear regression. Thenthe improved adaptive genetic algorithm of radial basis function network is applied to35cities inour country house price index prediction witch shows that its accuracy is much more high thanlinear models.