Research On Muti-Source Traffic Data Fusion Method Based On The Improved Neural Network Algorithm

With the increase of road mileage,the complexity of road network and the continuous growth of traffic demand,informatization and intelligent management have become essential to ensure the smooth flow of expressways.The progress of science and technology has made the collection methods of traffic information more diverse.At present,various types of traffic detectors are deployed on expressways.In addition,with the application of ETC technology on the main expressway,it has also become a new way to obtain the traffic operation status information by collecting vehicle OBU data.However,the data obtained by different collection methods have different characteristics in terms of field type,collection frequency,data accuracy and space-time coverage.How to make full use of multi-source data has become the primary problem for expressway intelligent management.Based on the abovementioned background,a series of researches are conducted on the multisource traffic data fusion.First of all,the study analyzed and summarized relevant literature about multi-source traffic data fusion technology,including the application direction and main methods.The paucity and limitation of previous research in data sources,data processing,and data fusion algorithms is pointed out.And then the research objective of this thesis is clarified to fill this gap.Secondly,the study determined the fusion data source of this thesis by comparing the characteristics of the various information collection technology.Then,the relevant theory of data fusion technology is introduced,and the importance and effectiveness of fusion technology in realizing the complementary verification of multi-source detector information is discussed,which lays theoretical foundation for the construction of multi-source traffic data fusion model.After that,an extraction and preprocessing method for multi-source traffic detection information is proposed.The extraction of traffic parameters is carried out based on the working principles of different collection methods,and the data preprocessing is based on the analysis of the spatial and temporal characteristics of the traffic data,and the Ha LRTC tensor completion method is introduced to repair the traffic data.This algorithm is able to obtain the global characteristics of the dataset and integrate the temporal and spatial information of traffic flow,so as to improve the recovery accuracy of traffic detection data.Then,a BSO-BP algorithm is proposed based on the beetle search technology and particle Swarm optimization theory.And then,a multi-source traffic data fusion model based on the BSO-BP algorithm was established.This model overcome the paucity of classical particle swarm optimization algorithm,such as lack of population richness so as to fall into local optimal solution.Therefore,the model can improve the accuracy of traffic data fusion.Finally,the method and model proposed in the thesis are verified by the data collected from the G50 expressway.In the experiment,more accurate traffic parameter estimation is achieved by fusing the data of the ETC system with the microwave detection data of RTMS.The experimental results show that in terms of data preprocessing,the Ha LRTC tensor completion method is very effective in recovering traffic abnormal data,and can achieve good accuracy for different types and degrees of missing;In terms of multisource traffic data fusion,the error analysis results show that the multi-source traffic data fusion model based on the BSO-BP neural network algorithm is superior than the existing models,and obtains more ideal and accurate results,which verifies that the model proposed by this thesis is effective.