Research And Implementation Of Railway Passenger Flow Prediction Algorithm

With the continuous improvement of railway technology innovation ability and the upgrading of operation management mode,China’s high-speed railway has developed rapidly in recent years,and the high-speed railway network has made brilliant achievements from "four vertical and four horizontal" to "eight vertical and eight horizontal".Benefit from the speed and convenience of railway passenger transport,high-speed railway has become one of the most popular ways of travel for Chinese people.But in the face of such a huge railway system,how to allocate resources reasonably,adjust operation strategies according to the dynamic changes of the market,and increase operation revenue have become the challenges faced by the railway department,and the solutions to these problems require a deeper understanding of the trend of passenger flow and accurate prediction of railway passenger flow.Based on this,this thesis studies the railway passenger flow prediction algorithms.The main work of this thesis includes:1.A feature extraction method of railway passenger flow based on machine learning is proposed.Firstly,the relationship between railway ticket booking and passenger flow on the final departure date,as well as the influencing factors of passenger flow,is analyzed and studied.Then,the characteristics of railway passenger flow are analyzed through historical data,and several feature extraction methods in line with the law of railway passenger flow are constructed,including positive and negative correlation features,periodic features,group features and so on.Finally,the feature comparison experiments and the comparison experiments of four machine learning algorithms are carried out to verify the effectiveness of the proposed algorithm.2.A railway passenger flow forecasting algorithm HANet model based on time series convolution attention is constructed to realize multi section train passenger flow forecasting.The linear part is composed of AR,which is used to extract the linear law of passenger flow data.The nonlinearity consists of CNN layer and TCAN network with hierarchical attention.First,local features are extracted by CNN convolution,that is,the spatial information of train passenger flow in different intervals.Then,TCAN is used to learn the temporal law of passenger flow,and the hidden long-term information in passenger flow data is extracted combined with a hierarchical attention mechanism.Finally,the optimal value of the super parameters in the model are determined by experiments,and a comparative experiment is carried out.The results show that the designed model has achieved good results in the multi section railway passenger flow forecasting task.3.The prototype system of railway passenger flow forecasting service is designed and implemented.First of all,using the proposed two train passenger flow prediction algorithms,the offline prediction model is trained by using the passenger flow data of Beijing-Shanghai line.Then,the trained model is deployed to the server,and the prediction results are automatically saved to the database through the scheduled task.Furthermore,users can query the prediction results directly through the service.Finally,Vue framework combined with ECharts technology is used to build a visualization service,which can provide support for scientific decision-making of railway researchers.