Research of Prediction Scheme Based on Adaptive Particle Swarm Wavelet Neural Network Model

Traffic flow prediction have the characteristics of complexity, non-linearity and randomness. Researchers are resorting to hybrid neural networks fusing more effective algorithms into forecast process. This paper advances a fresh algorithm with both the thought of variation and adaptive to conquer the poor global optimization capability and slow convergent rate of traditional back propagation neural network (BPNN). Mathematically, this algorithm constructs an adjustable formula to generate acceleration factors based on a logistic curve equation. Variable random initialization strategy has been introduced simultaneously. Simulations reveal that in contrast to auto-regressive integrated moving average model (ARIMA), back propagation neural network and wavelet neural network (WNN), the mean relative error reduced from about 10% to nearly 5%, the running time dropped from 5 seconds above to 2.5 seconds below, and the quality of the traffic flow prediction has a great improvement, which verifies the superiority of the brand new method proposed in this paper. Keywords-intelligent transportation system; traffic flow prediction; adaptive; variation; neural network; particle swarm optimization