基于非线性主成分分析和GA-RBF的高速公路交通量预测方法

为了提高高速公路交通量的预测精度, 综合考虑高速公路交通量的高度非线性和受多因素影响的特征, 提出一种基于非线性主成分分析和GA-RBF神经网络(NPCA-GA-RBF) 的高速公路交通量预测方法; 确定了高速公路交通量的主要影响指标, 运用非线性主成分分析法降低高速公路交通量影响指标的维数及其相关性, 用少数主成分代替原有的多指标, 以简化神经网络结构; 利用GA优化RBF神经网络的参数, 进一步提高交通量的预测精度; 以普洱市某高速公路为例, 对交通量预测方法进行实例验证。分析结果表明: 2组试验GA-RBF和NPCA-GA-RBF方法的平均相对误差分别比RBF方法降低1.62%、3.53%和2.27%、3.32%, 说明GA优化RBF神经网络能提高RBF方法的交通量预测精度; 与GA-RBF方法相比, 2组试验NPCA-GA-RBF方法的平均相对误差分别降低了1.91%、1.05%, 其交通量预测值更接近实际交通量, 预测结果更为可靠, 表明非线性主成分分析法消除了指标的相关性, 进一步提高了交通量预测精度, 减少了交通量预测复杂度。可见, NPCA-GA-RBF方法具有更高的交通量预测精度, 能为高速公路的良好管理提供可靠的决策依据, 满足高速公路合理运营管理的客观需求。 

Forecasting method of expressway traffic volume based on NPCA and GA-RBF

To improve expressway traffic volume forecasting accuracy, on the basis of the features of expressway traffic volume which was obviously nonlinear and affected by many factors, a forecasting method based on the nonlinear principal component analysis and GA-optimized RBF neural network (NPCA-GA-RBF) method was presented. The main influencing index of expressway traffic volume was determined, and the nonlinear principal component analysis method was used to reduce the dimensionality and correlation of influencing factors. The original multiple indexes were replaced by a few principal components to simplify the structure of the neural network. The GA was used to optimize the parameters of the RBF neural network and further improve the forecasting accuracy of traffic volume. An expressway in Puer City was taken as the example to verify the traffic volume forecasting method. Analysis result shows that theaverage relative errors of the GA-RBF and NPCA-GA-RBF methods in two experiments are 1.62%, 3.53% and 2.27%, 3.32% smaller than that of the RBF model, respectively, so the GA-optimized RBF neural network improves the traffic volume forecasting accuracy of the RBF method. Compared to the GA-RBF method, the average relative error of the NPCA-GA-RBF method decreases by 1.91% and 1.05% in two experiments, respectively, its traffic volume forecasting result is closer to the actual traffic volume and more reliable, which demonstrates that the nonlinear principal component analysis eliminates the correlation of the indexes to further improve the traffic volume forecasting accuracy and reduces the forecasting complexity of traffic volume. So, the NPCA-GA-RBF method has higher traffic volume forecasting accuracy, can provide the reliable decision basis for expressway operations management, and satisfies the objective requirements of the reasonable management for expressway.