摘要
根据交通流量的非线性、时变性和复杂性等特点,提出基于混沌粒子群CPSO(Chaos Particle Swarm Optimization)优化小波神经网络WNN(Wavelet Neural Networks)的短时交通流预测。结合混沌的随机性和遍历性改进粒子群优化算法,改善粒子群优化算法容易陷入局部最优的问题。利用混沌粒子群算法优化小波神经网络的模型参数,克服传统小波神经网络采用梯度下降法易陷入局部极值和引起振荡效应现象缺陷。仿真结果表明,混沌粒子群优化小波神经网络与粒子群优化小波神经网络和小波神经网络两种方法相比,其提高了收敛速度和预测精度。
According to the characteristics of nonlinearity, time-varying and complexity of traffic flow, we present a short-term traffic flow forecast which is based on the wavelet neural network （WNN） optimised by chaos particle swarm optimisation （CPSO）. In combination with the randomness and ergodicity of chaos, we improve PSO to meliorate its problem of easy to falling into local optimum. Also we optimise the model parameters of WNN with CPSO to overcome the defects of the traditional WNN that it is easy to fall into local minimum and cause oscillation effect phenomenon when using gradient descent algorithm. Simulation results show that comparing WNN optimised by CPSO with optimised by PSO and WNN itself, the former improves the convergence speed and forecasting precision.
出处
《计算机应用与软件》
CSCD
北大核心
2014年第6期84-86,90,共4页
Computer Applications and Software
关键词
混沌
粒子群
小波神经网络
短时交通流预测
Chaos Particle swarm Wavelet neural network Short-time traffic flow prediction
