Research On Traffic Congestion Prediction Method Based On Multi-Source Data Fusion

In recent years,with the development of intelligent transportation technology,many cities in China have established road traffic operation evaluation index system for traffic management departments to monitor and evaluate urban traffic operation.After years of application,the sensitivity and reliability of the evaluation method have been proved,and the accurate evaluation of traffic operation has been realized.Firstly,the paper summarizes the research status of urban traffic state evaluation and traffic multi-source data fusion at home and abroad.According to the real-time,accuracy and efficiency of urban road traffic operation evaluation,the vehicle travel time index（TTI）is determined as the main data to judge the traffic operation status,and the short-term traffic operation status is predicted;According to the research of relevant literature at home and abroad,the short-term traffic prediction in this paper is divided into traffic operation condition classification prediction and travel time index prediction.The classification prediction selects XGBoost algorithm and Light GBM algorithm to build the model,and the numerical prediction selects LSTM algorithm and ARIMA algorithm to build the model.Secondly,in the data preparation stage,Baidu travel data,Didi travel data,traffic police platform data and other data sources are collected as the basis of the whole article.Standardize the data,preprocess the data through data smoothing,data interpolation,data integration and data fusion,and draw the road network map in Arc Map through the coordinate point set in Didi travel data and Baidu travel data,so as to realize the visualization of the road network.Thirdly,on the basis of data preprocessing,the time and road grade characteristics of the data are described,and the impact of traffic accidents on TTI is analyzed to provide basis for model feature selection.The time characteristics of data adopt continuous Fourier transform to determine the data cycle and the difference between working days and weekends;The analysis of road grade characteristics shows that the TTI of different road grades in peak periods is different,and the main road is the highest;The analysis of accident data shows that the TTI of this section within an hour after the accident is 5.4% higher than the average value when there is no accident.According to the national standard code,different traffic operation conditions are divided into 1-5 levels,and the TTI threshold of each level is converted accordingly.Then,the training set and verification set are constructed by using the fused multi-source data to train the constructed model.The accuracy and accuracy of XGBoost in the traffic operation classification prediction model are 2.67% and 4.91% higher than that of Light GBM respectively.By comparing the evaluation indexes in the travel time index prediction model,the RMSE and MSE of ARIMA model are 22% and 37.8% lower than that of LSTM model respectively;Comparing the predicted road network map with the real road network map can intuitively view the prediction differences of different models.Finally,in order to provide basic data for the traffic management,ARIMA model is selected as the travel time index prediction model to provide data support for traffic congestion early warning.According to the predicted value of TTI,the hierarchical early warning of congestion is advanced,and the Box-plot is drawn according to the TTI of different road grades,and its upper limit is determined as the warning threshold.When the predicted value of two consecutive time intervals reaches the warning threshold and is in the rising stage,the command center needs to send police.The predicted value and the actual value are tested by MAPE and SMAPE,which are 4.72% and 4.69%,verifying the accuracy of warning.