基于集成学习的交通流短时特性分析与神经网络预测方法

为揭示交通流的内在动态特性,利用分析法对交通流分形特性进行研究,表明该城市交通流序列具有长程相关性;为达到更精准的短期交通预测效果,同时提出一种基于思维进化算法(MEC)对神经网络最优初始参数的定向搜索,解决神经网络易陷入局部最优的问题;并用自适应增强算法(adaptive enhancement algorithm,Adaboost)对优化过的神经网络集成,弥补神经网络对新样本集的泛化性能差缺陷,在此基础上通过预测误差平方和倒数准则重新调整Adaboost算法对弱预测器权值分布,使每个预测器最大程度提高网络预测精度.验证结果表明,改进MEC-BP_Adaboost模型与BP模型相比,均方误差和平均绝对误差分别下降78.2％和46.4％,证明本文改进方法对交通流预测具有合理性,对不同的交通流状态具有较好的适应性.

Analysis of short-term characteristics of traffic flow based onensemble learning and neural network prediction method

In order to reveal the inherent dynamic characteristics of the traffic flow, the fractal characteristics of the traffic flow were studied by using analysis method. Shows that the urban traffic flow sequence has long ran -ge correlation; In order to achieve a more accurate predicted effect about short-term traffic,a directional search of the optimal initial parameters of the neural network based on a mind evolution algorithm is proposed to solve t- he problem that the neural network is liable to fall into a local optimum; To compensate for the poor generalizati- on performance of the new sample set by the neural network, the optimized neural network is integrated by using the Adaboost algorithm. On this basis,the prediction of error square and reciprocal criteria can readjust the weight distribution of the weak predictors by the Adaboost algorithm,so that each predictor can maximize the network prediction accuracy. The verification results of the PeMS system dataset show that compared with the BP model, the mean square error and the average absolute error of the improved MEC - BP Adaboost model are reduced by 78.2% and 46.4%,respectively. It is proved that the improved method in this paper is reasonable for traffic flow prediction and has good adaptability to different traffic flow states.