Adaptive Learning of RBF Network Based on Adaptation Complexity

Radial neural network can be used to decompose complex problems with good biological properties. Adaptive control for neural network is helpful to improve the efficiency of pattern classification. In order to solve pattern classification with different adaptive characteristics, multiple adaptive algorithms are embedded in radial neural network. Through the test of the objective function, it is found that not all of the combinational algorithms can get desired results. After systemic tests, it is found that shifting strategy in the later stage of learning process can get good effect with the appropriate change strategy. Keeping consistent major evolutionary strategy is correct and necessary. Only if the simulation stop optimizing, appropriate other strategy should be taken for reaching better effect.