Research On Evaluation And Prediction Of Urban Road Traffic Congestion Considering Weather And Spatio-temporal Feature

To make timely and accurate evaluation and prediction of urban road traffic congestion is of great significance for formulating measures to control and prevent congestion and improving urban congestion.Urban road traffic flow parameters can directly reflect the state of congestion,so the core of road congestion identification and prediction is the analysis of traffic parameters’ characteristics.By comparing and analyzing the content of model algorithm and feature selection in congestion assessment and prediction research at home and abroad,this paper proposes a SVR-RBFNN traffic parameter prediction model based on multi-factor fusion,and a traffic congestion assessment model based on entropy weight-cloud fuzzy comprehensive evaluation.The details are as follows:Firstly,based on the traffic parameter data obtained in this paper,through qualitative analysis and quantitative calculation,time-varying feature and periodicity,the spatial correlation with other roads,and the correlation with some environmental factors are analyzed.What’s more important,semantic spatial correlation analysis is introduced in spatial dimension.The POI data is used to determine the land use types of roads,and the similarity of traffic flow patterns between roads with the same land use properties is studied,which is integrated into the prediction model.Secondly,based on the above feature analysis,the temporal,spatial and environmental features of traffic flow are extracted.Considering that SVR has strong generalization ability,rigorous theoretical basis and good interpretability,while RBFNN has the advantages of selfadaptive,self-learning and data parallel processing,a SVR-RBFNN combination prediction model of traffic parameters is constructed,so that the effective fusion of the three features is realized.Next,when proposing the congestion evaluation model based on traffic flow parameters,the cloud model is introduced to realize the mutual transformation of quantitative index and qualitative concept between traffic parameters and traffic congestion state,which improves the problem of fuzzy loss caused by the accuracy of membership degree in traditional evaluation method,and makes the membership degree become a random number with stable law.It well describes the fuzziness of the boundary between congestion states and the difference of travelers’ psychological feelings,and the results are more in line with the reality.The congestion prediction model and evaluation model are verified by actual data.Results show that:(1)the combined predicting model based on SVR and RBFNN performs well,whose error is about 6%.And its accuracy and the degree of data fitting are both better than the single model(SVR or RBFNN);(2)the introduction of semantic spatial features and the consideration of environmental factors make the prediction accuracy of parameters significantly improved,and further analysis shows that the adjacent spatial features have the greatest impact on the model’s performance,followed by the semantic spatial correlation,and the environmental features have no significant impact;(3)compared with the traditional evaluation method based on threshold division,the traffic congestion evaluation model based on cloud model theory to determine the membership degree make the results more consistent with the reality,which can solve the problem that the congestion level is not clear when the index value is near the threshold.To sum up,whether it is the extraction of semantic spatial features and the combination of models in the prediction model,or the introduction of cloud model theory in the congestion assessment model to achieve qualitative and quantitative conversion,the model’s feasibility and applicability are excellent.Moreover,the research results provide theoretical support for traffic management,and the research perspective also has reference significance for the follow-up traffic congestion research.