改进粒子群优化XGBoost模型的高速公路服务区交通量预测

Traffic volume prediction for highway service areas based on XGBoost model with improved particle swarm optimization

为了科学有效地评估高速公路服务区交通通行服务能力和进行基础设施优化配置,提出一种基于改进粒子群算法(Improved Particle Swarm Optimization, IPSO)和XGBoost融合的高速公路服务区交通量预测模型.首先对粒子群算法的拓扑关系结构进行改进,将粒子群划分为主粒子和从粒子,获得多个改进粒子组,并给出改进后主从粒子速度和位置的更新方法;其次利用线性递增和递减函数对粒子的迭代寻优速度进行自适应调整;最后在模型训练过程中,引入交通量调查数据这一重要特征,并利用IPSO算法对XGBoost模型的主要超参数进行优化选择,建立了交通量调查数据和服务区各交通量之间的IPSO;GBoost预测模型.结果表明:提出的改进粒子群算法具有更强的参数寻优能力,迭代收敛速度更快,能够搜索出XGBoost模型的理想超参数,实现对服务区交通量的有效预测.同时,提出模型的预测性能明显优于LSTM、XGBoost、CNN-LSTM和PSO;GBoost,对小型车、大型车、货车、车当量和人当量数据的预测精度分别达到了0.913、0.815、0.872、0.931和0.924.

In order to scientifically and effectively evaluate the traffic service capacity in highway service areas and optimize the configuration of service facilities, a traffic volume prediction model for the highway service areas based on the fusion of improved particle swarm optimization(IPSO) and XGBoost is proposed.First, the topological relationship structure of the particle swarm algorithm is improved.On this basis, the particle swarm is divided into master particles and slave particles to obtain multiple improved particle groups. At the same time, the improved method of updating the speed and position of master and slave particles is given. Secondly, the iterative optimization speed of particles is adaptively adjusted by using linear increasing and decreasing functions. Finally, in the model training process, the important feature of traffic volume survey data is introduced, and the IPSO algorithm is used to optimize the selection of the main hyperparameters for the XGBoost model, and the IPSO_XGBoost prediction model between traffic volume survey data and each traffic volume in the service area is established. The results show that the proposed IPSO algorithm has stronger parameter optimizing ability and faster iterative convergence speed. This can help search for the ideal hyperparameters of the XGBoost model and realize the effective prediction of the traffic volume in the service area. Meanwhile, the prediction performance of the proposed model is significantly better than that of LSTM, XGBoost, CNN-LSTM and PSO_XGBoost, and the prediction accuracies for small cars, large cars, trucks, vehicle equivalent and tourist equivalent data reach 0.913, 0.815, 0.872, 0.931 and 0.924, respectively.