Wavelet neural network (WNN), combining with wavelet analysis and neural network, brings forth a highaccuracy performance in identification and approximation. Passenger °ow forecast plays an important role in transit scheduling and an improved WNN model is constructed to actualize dynamic forecast, in which Morlet wavelet is selected as the activation function. Input data series, i.e. historical data, traffic condition and weather information about passenger flows surveyed from No.63 line in Harbin, China, is pre-processed via a fuzzy operator before transferred to train and test the constructed network. A hybrid genetic algorithm and identical dimension recurrence idea are performed to optimize the structure and shape of WNN dynamically so as to enhance its forecast accuracy. The experimental result indicates the proposed WNN model can satisfy the precision request, accelerate the convergence speed, improve the global generalization ability and possess the practicality in transit dynamic scheduling.
展开▼
