短时交通流预测特性及实例分析

Short-term Traffic Flow Characteristics and Instance Analysis

在分析短时交通流具有的非线性和不确定性的基础上,指出应用径向基神经网络(RBF)和非参数回归方法(NPR)进行短时交通流预测的适用性。提出将径向基神经网络的输入端和训练数据重新进行选择,并且对网络的权值和阈值重新进行训练是保证交通流特性的关键,对应此关键问题反映到算法中的2个重要参数:预测误差范围和最大神经元数目进行了预测结果分析说明。同样对于非参数回归方法中的近邻点个数和预测误差范围也做了预测结果分析。应用这2种方法对某一天进行了微观预测结果分析。预测结果说明,这2种方法都能够比较好地适应交通流特性,预测效果很好。

Based on the analysis of the nonlinearity and uncertainty characteristics of short-term traffic flow,the applicability of radial-basis function(RBF) neural network and non-parametric regression(NPR) method for short-term traffic flow forecasting was presented.The most important keys of RBF include re-selection of the inputs and training data,re-training of weights and threshold values inside network.Two important parameters in the corresponding algorithm including the prediction error scope and the largest neuron number were explained.Similarly,the most important parameters including neighbor number and forecasting error scope in NPR were also discussed.The microscopic forecasting analysis was made using these two methods for one day traffic.The forecasting result shows that they all deal with the characteristics of short-term traffic flow well and get perfect forecasting performance.