Using Radial Basis Function Neural Network With Weight Memory Mechanism: Nonlinear System Identification

The radial basis function neural network (RBFNN) is a learning model with better generalization ability, which attracts much attention in nonlinear system identification. Compared with the other NNs however, RBFNN does not do well in learning accuracy and approximation ability. In this article, we propose an efficient RBFNN with weight memory mechanism (RBFNN-WMM). The major advantage of RBFNN-WMM is that it can make full use of weight parameter information from historical moments during its learning process. Moreover, RBFNN-WMM can also dynamically adjust the learning speed via the efficient adaptive hyperparameters. Finally, two simulation examples of nonlinear system identifications are used to test the effectiveness of the proposed RBFNN-WMM. The simulation results show that the proposed RBFNN-WMM achieves higher identification accuracy and better generalization ability.