Short-Term Traffic Condition Prediction Based on Multi-Source Data Fusion and LSTM 2020-01-5137
Short term traffic condition forecast is a key area of concern in the field of transportation. Accurate forecasting of short - term road condition can provide the basis for traffic path guidance, traffic decision-making and traffic control, and improve traffic efficiency and safety. In order to improve accuracy of short term traffic condition forecast, we use multi source data to predict short term traffic condition. First build the correlation of road conditions between roads, then fuse vehicle status data and road correlation data, and then build short-term traffic condition prediction model based on LSTM, at last combine traffic accident data to predict short term traffic condition. In this article, we use the measured data in Wuxi to verity the model, and compare the typical road prediction models. The result shows the traffic condition prediction model in this article can predict the short-term traffic conditions effectively. Compared with ARIMA model, our model can reduce RMSE, MRE and MAE by 26%, 31.7% and 33%. This paper mainly includes the following parts: the introduction part mainly introduces the necessity of short-term traffic condition forecasting, the limitations of the data sources and models used in the current research, and the data sources and models used in this article; the related work part mainly introduces development of the short-term traffic condition forecasting algorithm model; the methodology part mainly introduces how to preprocess the data, build road correlation, how to build an LSTM-based algorithm model and the fusion with accident data; the experiment part introduces the data used for verification, the basic parameters of the model and comparison of the model Results; the conclusion part summarizes the paper and introduces future work plans.