Regional Traffic Condition Prediction Method Based On Deep Spatio-temporal Residual Network

With the continuous and rapid development of society and economy,the pace of urbanization is accelerating and the amount of vehicles is increasing,resulting in increased traffic congestion problems in urban cities.The transportation problem has become one of the urgent problems in both developing and developed countries.Therefore,as a means to effectively solve transportation problems,intelligent transportation systems have become a hot research topic.More and more countries have begun to attach importance to develop intelligent transportation systems.Among them,the prediction of road traffic conditions is one of the core parts of intelligent transportation system research.In general,the existing prediction models can be summarized into two categories: traditional statistical learning models and deep learning models.For traditional statistical learning models,common methods include vector autoregression(VAR),autoregressive moving average(ARIMA),K nearest neighbor(KNN).However,the simple structures of models make them unable to capture the complex spatial-temporal correlation characteristics of urban transportation.Compared with traditional model learning methods,deep learning has deeper and more complex model structures,and thus they can derive more attractive results.However,the existing deep learning base methods mainly focus on the independent prediction problem of one or several road sections.Once the predictions are scaled to the whole city,existing works will be trapped in huge system overhead by generating a large number of complex models.In order to efficiently predict the traffic state information of the entire road network,this paper proposes a regional-level traffic condition prediction model based on a deep spatial-temporal residual network.To describe the regional traffic conditions and capture their spatial-temporal dependencies,we present a deep learning based model-Deep RTP.Different from previous studies,the method in this paper is to make predictions for each region,so it can evaluate the entire city with low overhead.Secondly,for the area where traffic congestions frequently occur,we use the LSTM model to predict each section of the area to ensure the high accuracy of the final result.Specifically,we use a novel metric called Traffic State Index(TSI)to measure regional traffic conditions,and carefully classify traffic data into three categories that are used to capture hourly,daily,and weekly traffic patterns.Furthermore,we employ the convolutional and residual neural networks to model both spatial and temporal dependencies.Experimental results from real-world traffic data demonstrate that Deep RTP outperforms five baseline methods and can achieve higher prediction accuracy.