基于混沌粒子群算法的神经网络短时交通流预测

Traffic Flow Forecasting Based on the Wavelet Neural Network with Particle Swarm Optimization Algorithm

针对短时交通流量具有复杂性、非线性等特点,提出基于粒子群算法的神经网络交叉路口短时交通流量预测方法;利用混沌粒子群算法对BP神经网络权值和阈值进行优化,克服易陷入局部极小和引起振荡效应现象,从而提高了网络的预测精度;实验仿真结果说明,与标准粒子群算法相比较,新算法可以有效提高预测精度,减少预测误差,最大绝对误差下降至12.15%,相对预测误差在10%以内的预测数据提高至57.5%,并且很好地反应了交通流的特点,是一种可行的预测方法。

The characteristic of short term traffic flow is complexity and nonlinear. A new algorithm for crossroads short term traffic flow forecasting based on neural network with Chaos Particle Swarm optimization algorithm is proposed. The algorithm makes use of chaos particle swarm optimization algorithm to optimize the structure parameters of neural network. The algorithm combined the search global optimization ability of chaos particle swarm optimization algorithm and the nice time--frequency local character of wavelet network, and overcomes the phenomenon of early falling into local minima and oscillation effects. The results of simulation show that , compared with the standard swarm optimization algorithm, the new algorithm can improve the prediction precision, reduce the forecast error, the maximum absolute error dropped to 12.15% , the relative forcast error that less than 10% of forecast data increased to 57.5%, and reflect the characteristic of traffic flow. It＇s a feasible forecast algorithm.