Research On Traffic State Identification And Prediction Method Of Expressway

With the continuous adjustment and optimization of China’s urban economic structure,the urbanization process is gradually deepened,and the traffic demand continues to grow.As a result,the contradiction between road supply and demand is becoming increasingly prominent,and the highway traffic congestion is gradually normalized.Highway traffic congestion not only reduces the road traffic efficiency,but also brings a series of problems such as traffic safety and environmental pollution,which hinders the process of realizing the strategic objectives of "traffic power" and "double carbon" to a certain extent.Therefore,how to effectively alleviate highway traffic congestion and improve road operation efficiency is an urgent problem to be solved.Adopting advanced traffic management and control technology and building intelligent transportation system is an effective means to improve road operation efficiency.Accurate and reliable traffic state identification and prediction method is not only the key core part of intelligent transportation system,but also the basis of using its system to improve expressway operation efficiency.The traditional model-driven traffic state identification and prediction methods can not fully meet the current complex and changeable traffic flow system and the highly information-based intelligent transportation system at this stage.Therefore,it is necessary to mine the complex traffic flow change situation from the massive traffic big data through big data analysis technology and intelligent means,so as to further improve the accuracy of traffic state discrimination and prediction.This study takes expressway traffic as the research object and aims to improve the accuracy and efficiency of traffic operation state identification and prediction.Combined with traffic flow theory,big data technology,machine learning method and intelligent optimization theory,this paper make a profound study on the identification and prediction methods of expressway traffic operation state.The main work and research results are summarized as follows:(1)The method to repair the traffic flow missing data under different data missing scenariosFor different traffic flow data missing modes,according to the characteristics of data missing,the traffic flow data missing is divided into simple random missing scenarios and complex continuous missing scenarios.On the data restoration in simple random missing scenario,an improved IWKNN missing data restoration model is proposed based on Pearson correlation distance and Gaussian weighted.The key point is to use Pearson correlation coefficient and Gaussian weighted distance to measure the similarity between missing data and complete data points.On the data repair in complex continuous missing scenes,SARIMADBN repair model is proposed.The SARIMA model is used to fit the linear part of continuous missing data,and the remaining nonlinear residual sequence is repaired by deep belief neural network with strong nonlinear fitting ability.Finally,the effectiveness of the two missing data repair models built in this paper is verified by designing different data missing scenarios and missing ratios.The missing data repair method proposed in this part can not only accurately and quickly imput the missing data,but also provide complete data support for traffic state identification and prediction.(2)Traffic state identification model based on FCM clustering method and BPSO-(RSM-DAG-SVM)methodBased on unsupervised FCM clustering method and traffic flow theory,the traffic state categories and the corresponding clustering centers of each category are obtained.Taking DAG-SVMs as an individual learner,the random subspace ensemble learning method is used to disturb the data attributes of expressway traffic flow,several feature subspaces are obtained,and the traffic state discrimination model based on RSM-DAG-SVMs is established.In order to improve the generalization performance of RSM-DAG-SVMs ensemble model and reduce the integration scale,BPSO optimization algorithm with global optimization ability is used to select and optimize the base classifier in RSM-DAG-SVMs ensemble model,screen the base classifier with large travel diversity and small error,and obtain the traffic state identification model based on BPSO-(RSM-DAG-SVMs)selective integration.Finally,the effectiveness and good generalization performance of the traffic state identification model are verified by the measured traffic flow data.(3)Traffic state combinational prediction model based on the methods of CEEMDAN-IWEP and GWO-LSSVMAiming at the nonlinearity and time variability of complex traffic flow series,the CEEMDAN method is used to decompose the original traffic flow time series into several relatively stable and simple eigenmode components and residual sequences in order to improve the predictability of traffic flow data.The improved weighted permutation entropy is used to reorganize the components with similar complexity,so as to reduce the input of the prediction model.The LSSVM prediction model is used to predict the reconstructed sequence and residual components,and then the prediction results of each component are combined to obtain the final traffic flow prediction results.For the super parameter value problem of LSSVM,the GWO intelligent optimization algorithm is used to optimize the important parameters of LSSVM model,and the global optimal parameter combination is obtained.Finally,the actual traffic flow data are used to verify the effectiveness of the proposed traffic state prediction model of CEEMDAN-IWEP and GWO-LSSVM.The results show that the proposed traffic state prediction method can effectively improve the prediction accuracy and reduce the complexity of the model.(4)Multi-step prediction model for traffic flow based on IMODE-SA multi-objective optimization of Echo State NetworkAiming at the error accumulation problem caused by multi-step iterative prediction of traffic state,MIMO prediction strategy is used for multi-step prediction of traffic flow.Firstly,the time autocorrelation coefficient and spatial nonlinear Spearman correlation coefficient of traffic flow series are calculated to determine the variables with the strongest correlation with traffic state prediction;Secondly,a multi-step prediction model of traffic flow is established based on echo state network.For echo state network parameter selection and reserve pool design,the prediction accuracy and model complexity of the prediction model are considered at the same time.The parameters of ESN prediction model are optimized by taking prediction error,reserve pool size and sparsity as three optimization objectives.The standard differential evolution algorithm is improved.Combined with the simulated annealing algorithm,the IMODE-SA multi-objective optimization algorithm is designed to optimize the parameters of the ESN traffic state prediction model,and the Pareto optimal frontier composed of nondominated solutions is obtained.According to the different requirements for the prediction accuracy and complexity of ESN model,the optimal parameter combination of ESN model is selected.Finally,the proposed IMODE-SA-ESN multi-step traffic flow prediction model is verified by the measured data.The results show that the proposed model can effectively improve the prediction accuracy and provide the diversified optimal parameter combination with different requirements for model performance and complexity.