基于QPSO-RBF神经网络的混合交通流车速预测模型

Road Speed Prediction Model Based on QPSO-RBF Neural Network in Mixed Traffic Flow

为解决传统道路车速标定以单一车型浮动车而忽略城市道路车型复杂导致实际应用误差较大等问题,根据车型分类标准将道路运行车辆分为小客车、出租车、公交车、大型客货车四类,从车辆性能、运输要求两方面对全车型车速进行因素分析,以浮动车数据为基础,利用粒子群算法优化RBF神经网络并对其余车型车速进行预测,以粒子的维度分量作为RBF网络的权值和阈值,以神经网络均方误差的倒数作为粒子群算法的适应度函数并训练神经网络使其达到均方误差最小化。实例分析表明:粒子群算法有效降低了RBF神经网络的预测误差,预测模型对小客车、出租车、大型客货车的预测平均相对误差分别为9.21%、10.83%、12.78%,经算法计算出的道路车速精度优于基于浮动车的道路车速,且平均绝对误差控制在5km/h以内,达到实际应用精度要求。

In order to solve the problem that the traditional road speed calibration uses a single model of floating vehicles and ignores the complexity of urban road vehicles which leads to large error during the practical application,according to the vehicle classification standard,the road vehicles are divided into four categories:car,taxi,bus,and large passenger and freight cars.Analysis of full vehicles speed is conducted from two aspects of vehicle performance and transportation requirements,and based on the floating car data,the RBF neural network is optimized by particle swarm optimization,and the speed of other vehicles is predicted,the dimension component of the particle is used as the weight and threshold of the RBF network,and the mean square error of the neural network is used as the fitness function of the particle swarm optimization algorithm,and the neural network is trained to minimize the mean square error.Example analysis shows that:particle swarm optimization effectively reduces the prediction error of RBF neural network,the average relative errors of prediction models for car,taxi and large passenger and freight cars are 9.21%,10.83% and 12.78%,the accuracy of road speed calculated by the algorithm is better than that based on the floating vehicle,and the average absolute error is less than 5 km/h,which satisfies the requirement of practical application.