The Research On Toll Station’s Short-term Exit Traffic Flow Prediction Based On LSTM

Intelligent transportation service system is an important application of data mining and analysis technology in the transportation field.By collecting,storing,processing and analyzing data generated in the transportation field,it helps to improve the transportation field’s intelligent systems and services.Traffic flow forecasting is one of the important research directions to improve the performance of intelligent transportation systems.Accurate and real-time traffic forecasting can provide a basis for the relevant planning and decision-making of the transportation authorities,and reasonably control and deploy traffic.The short-term exit traffic flow prediction of toll stations is a practical scenario application of traffic flow prediction,and it is also a key part of intelligent traffic control.Due to a series of reasons such as deceleration or parking fees,the exit of the toll station restricts the traffic capacity,which often causes traffic congestion in the expressway network.Predicting the exit traffic flow of the toll station in advance is conducive to people’s judgment of the congestion situation,so that corresponding measures can be taken to alleviate the exit traffic pressure of the station.In order to accurately sense the exit traffic of the toll station in real time,this paper proposes a short-term traffic flow prediction model based on the LSTM,which can extract the time series characteristics of the exit traffic flow of the target toll station and the spatial and temporal characteristics of the stations associated with the target,and combine the two characteristics to predict the short-term exit traffic flow of the target toll station.Considering the different influences of the associated stations on the target,in the process of extracting spatial and temporal features,the Pearson Correlation Coefficient is used to measure the impact of the associated stations on the target,and used as the combined weight of the hidden layer in the spatial and temporal model.In addition,in the time series feature model,considering the impact of multiple features on the traffic for a period of time in the future,using multiple features as input to the model improves the prediction performance of the model.Finally,this paper verifies the model based on one province’s toll station data.The experimental results show that the performance of our proposed prediction model is better than two classic statistical methods,machine learning and deep learning models.In future work,we will extend the model to other application scenarios with the same characteristics.At the same time,we will consider more adequate external conditions to extract model features.