Qizhi Wang
PID, resource allocation network, neural network, learning algorithm
An improved resource allocation network (RAN)-based online modelling method is proposed. RAN is a hidden layer network whose hidden units are Gaussian radial basis functions. The improved RAN can suppress the interference of the new data on the weights of the previous training links. The influence of initial parameters on the network is thus described and the property values of these parameters are given. The proportional integral derivative (PID) parameters are adjusted by the RAN algorithm. The control system is mainly controlled by the object, dynamic RAN identifier dynamic neural network identifier and single neural network controller. According to studying the input data by minimum resource allocation network (MRAN) algorithm, the structure and parameters of RAN are optimised in real- time, so that the network output can well approximate the controlled object. The simulated results show that the algorithm has fast learning speed, optimal generalisation ability and high approximation precision.
Important Links:
Go Back
