Prediction Of Short-term Passenger Flow In Urban Rail Transit Based On GPR And KRR Model

With the rapid development of cloud computing and big data technology and the gradual advancement of China's urban rail transit construction,the demand for urban public transport intelligence by governments and urban residents is increasingly urgent.Through the analysis of the massive travel data,mining the important information behind the data can help the operation and management department to adjust the work in time.At the same time,it can also guide the traveler to choose a reasonable travel time and travel path.It plays an important role in promoting the development of smart cities.Therefore,based on the AFC(Automatic Fare Collection System)data of urban rail transit,this thesis analyzes the temporal and spatial characteristics of residents' travel based on the Hadoop platform.At the same time,based on the shortcomings of the current short-term passenger flow forecasting model,a hybrid forecasting model is constructed.The main research contents and innovations of this article are as follows:(1)In view of the defect that most of the feature extraction algorithms cannot extract the deep features of data,the SAE(Stacked Auto-encoder)network is introduced to build a feature extraction model based on deep network,which can effectively extract the feature of the site's passenger flow,and apply the K-means algorithm on this basis.Implement the division of site types.After comparing the model division effect is far better than the existing site type division method.(2)In terms of feature selection,the limitations of traditional time series in the selection of features,the vulnerability of passenger flow data to external factors,and the susceptibility of many existing feature selection algorithms to disturbances in the training set are introduced.This is based on the introduction of LARS(Least Angle Regression).The algorithm's stability feature selection algorithm makes the error of limited samples controlled,extracting more stable features and improving the prediction accuracy of the model.(3)In the aspect of predictive model construction,GPR(Gaussian Process Regression)and KRR(Kernel Ridge Regression)are introduced into the field ofshort-distance passenger flow forecasting in urban rail transit.Based on the passenger flow forecasting by GPR,the forecast results of GPR,variance estimation of forecast results,holiday information,and site category information are integrated,and KRR is used.The algorithm implements the correction of the GPR prediction result and effectively improves the prediction accuracy on the basis of ensuring the time efficiency.(4)Based on the concept of road traffic index,a calculation method for the subway line network congestion index is designed.The principal component analysis method was used to extract the characteristics of cross-section passenger flow and assign corresponding weights to each section.The weighting index of the final line network was obtained,and the congestion of the line network was divided according to the line network congestion index.The line network congestion index can effectively reflect the overall congestion status of the current and future line networks,and has important reference significance for managers and travelers.