Research On Identification And Forecasting Methods Of Freeway Traffic Congestion Based On Mobile Signaling Data

With the rapid development of smart cities and the continuous expansion of freeway networks,the number of cars in China continues to grow and traffic congestion frequently occurs,that problem requires scientific monitoring and management of high-speed road networks.At present,the freeway traffic information acquisition is mainly through fixed detectors or onboard GPS,because of its narrow coverage and high cost,it cannot meet the needs of intelligent transportation systems.Using traffic signalling data to identify and predict traffic jams is a hot research direction of intelligent transportation systems because of the features that mobile signaling has such as low cost,full coverage and real-time,but low accuracy is always the problem hard to solve,In thesis,research on the issues related to congestion recognition and prediction of freeway based on mobile signaling data to achieve accurate congestion identification and prediction.Firstly,thesis analyzes the characteristics of mobile signalling data and the characteristics of the "noise data" that are incorporated into it.The commonly used methods for calculating the speed and density of high-speed vehicles are studied.The principles of BP neural network and SVM support vector machine prediction algorithm are selected and studied.Secondly,based on the idea of physical kinematics,the shorter the distance between two trajectory points containing a road grid,the greater the contribution to the road grid vehicle speed and the characteristics of parallel road noise data,a method for calculating the average speed of roads on roads that fuses distance weights and eliminates parallel road noise data is proposed;According to multiple characteristics of onboard mobile phone users,a clustering method is used to perform matching of people and vehicles to identify the number of vehicles on the road,and then calculate the vehicle density.Then,using vehicle speed and density as input parameters,a multi-parameter comprehensive threshold congestion recognition method is constructed to identify freeway congestion.Finally,taking into account the spatial and temporal dimensions of data and the advantages of multiple sub-prediction models that can be complemented by each other,an optimal weight combination method is used to construct a short-term traffic congestion combined forecasting model for congestion prediction.Finally,the model is verified using a mobile signaling data acquisition system and data processing platform provided by mobile operators,comparing model results with fixed detector data,comparing the calculation results of vehicle speed and vehicle density of the congestion identification model with the traditional method's calculation results by MAE,RMSE and MAPE,it can be seen that the proposed model method results are superior to the traditional methods,and the final congestion recognition accuracy rate reaches 91.19%.The results of the combined forecasting model presented in thesis are compared with the results of the sub-prediction model for MAE,RMSE and MAPE.The results are better than the sub-prediction model,and the final accuracy of the prediction of congestion is 81.25%.The results meet the design requirements of the model and have a certain practical value.