Study On Metro Short-term Passenger Flow Forecasting Based On Deep Learning

With the concept of green travel deeply rooted in people's minds,subways are another important choice for people's travel styles because of their fast,convenient and comfortable features.In order to improve the operating efficiency of subways and ensure the safety of subway operations,the short-term passenger flow forecast of subway has become an important issue that needs to be solved urgently.Compared with the shallow machine learning model,the deep learning model obtains more abstract data feature expression through the depth of the network structure.Find out the characteristics of the data in the massive data that conform to the inherent variation of the data.It has been successfully applied in many fields,but there are few achievements in the field of short-term passenger flow forecasting.The Deep Belief Network is an important model in deep learning theory.The two-layer structure of the model gives it a great advantage when dealing with large amounts of data,non-linearity,and strongly random data.Therefore,this paper will design and implement a passenger flow forecasting model based on the DBN model,so as to effectively solve the short-term passenger traffic forecasting problem.This paper designs and implements a DBN-P/GSVM prediction model based on deep learning theory and support vector regression theory.The overall design of the model is divided into two parts: The top-level structure constructs a regression model with Support Vector Regression.In order to obtain the optimal combination of parameters,three parameter optimization methods(genetic algorithm,particle swarm optimization,and grid search algorithm)are used to perform parameter optimization for support vector regression.And through the comparison experiments,the kernel function of SVR is determined,and the ability of the model to map the short-term passenger flow data of subway is improved.The underlying structure constructs a DBN model to complete the feature extraction of short-term traffic data of subways.Through each layer of RBM model reconstruction of data,a more abstract data representation is formed,highlighting the inherent variation of subway short-term passenger flow data.,Provides a good data foundation for top-level regression machines.Through unsupervised weight generation and supervised overall fine tuning of the two-stage model training,the robustness and stability of the prediction model are enhanced,and the final network structure parameters are determined in combination with experiments.Taking the example of passenger flow forecasting at the Chengdu North Railway Station subway station,the advantage of the DBN-P/GSVM prediction model in dealing with the short-term passenger traffic forecasting problem is verified.The experimental results show that compared with the other three shallow prediction models,the prediction accuracy of DBN-P/GSVM model is higher,and the prediction stability is stronger.It shows that the DBN-P/GSVM predictive model can effectively extract the short-time passenger flow data from the subway through the underlying network structure,greatly reducing the noise in the data,and then obtaining better prediction results.