Research Of Road Travel-time Prediction Based On Spark

In recent years,with the rapid growth of urban transportation travel demand,the supply of road resources has gradually become a bottleneck in traffic operations,leading to traffic congestion as an increasingly serious problem.Road travel time is an important parameter reflecting the running state of road traffic,whose prediction can provide reference for traffic management measures and public travel.In order to alleviate the contradiction between fastgrowing travel demand and limited road resource supply,and improve the efficiency of road network operation and overcome the low efficiency problem of massive trajectory mining of single-machine serial,this paper takes road travel time as the research object,proposes parallel road travel time extraction and road travel time prediction methods by using Spark big data processing platform,and realizes a timely and accurate road travel time acquisition and prediction,which may help to formulate more scientific and effective traffic management measures and provide reference for traveler's travel route decision.Due to the limited data storage capacity and data processing capability of a single machine,coupled with the difficulty to expand,the serial trajectory data mining is inefficient.Therefore,based on the high availability,low cost and easy expansion of HDFS and data reuse and parallel execution mechanism of Spark,this paper firstly builds a big data processing platform from three levels,that is,data storage,data processing and data application.Secondly,a long shortterm memory network(LSTM)-based travel time prediction method is proposed,for the shortcomings of the traditional road travel time prediction method in the efficiency to consider the influencing factors of road travel time and the ability to fully exploit the internal correlation of road travel time series.In this paper,the main random influencing factors of travel time(precipitation type,precipitation,wind speed,air temperature,visibility)are selected to construct the feature vector,and LSTM network parameters are optimized to construct the road travel time prediction model based on LSTM.Finally,taking the online car-hailing trajectory data released by Didi as the experimental data,this study uses Spark big data processing platform through the strategy of data parallelization and task parallelization to mine and analyze the massive trajectory data,and realizes the parallelization method of travel time prediction.The experimental results show that the Spark big data processing platform constructed in this paper can quickly and accurately extract the road travel time from the massive trajectory data and predict the road travel time.The average relative error of the travel time prediction model proposed in this paper is 0.071,which is significantly better than the ARIMA model and the random forest model.The parallel processing efficiency of road travel time extraction and prediction is significantly improved compared with serial processing,and the platform has good scalability and acceleration ratio.The method proposed in this paper can quickly and accurately extract and predict the road travel time,which may help solve the road congestion problem and has certain value for creating a efficient and environment-friendly urban traffic environment.