An Adaptive Hybrid Wavelet Neural Network and Its Application

This paper proposes a novel strategy to construct wavelet neural network: hybrid use of multiple types of mother wavelet according to the pattern change of time series being studied instead of one single type. Training is performed through using the genetic algorithm at first to locate a good region in the parameter space and then employing the gradient descent algorithm to determine a near optimal solution in that region. A case study of traffic flow forecasting using proposed wavelet, neural network is demonstrated