A resource-allocating network based on local conditions and its application in prediction of nonlinear systems

In this paper, a resource-allocating network based on local conditions (RAN-LC) is proposed to avoid the existing problems of RAN. This method gets the initial hidden nodes by using K-means clustering algorithm and the characteristics of activation function, and it utilizes new Novelty Criterion based on local conditions instead of the old one to keep the network neat and efficient. Moreover, it adopts Multi-patterns to enhance the generalization ability of network in the state of parameters adjustment. The simulation results show that this method can generate network quickly and more reasonable. The network generated finally has good performance and also works well in the prediction of nonlinear systems.