Spatial-temporal Analysis And Prediction For Urban Ground Bus Transit Based On Artificial Intelligence Algorithm

In order to achieve the goal of healthy and sustainable development of the urban public transportation system,it is necessary to continuously improve and improve the urban public transportation system,improve the service level of urban public transportation,and achieve the goal of convenient,efficient and comfortable public transportation system.Urban ground bus service is an important part of the urban bus transportation system.With the emergency of big data,a new idea shoule be provided for optimizing the urban ground bus system.Efficient and accurate information mining for large-scale and multi-source public transportation spatial-temporal big data and timely and scientific operational decision-making have become major problems to be solved in the construction of smart public transportation system under the“public transit priority development”strategy.There are a large number of bus stations and operation lines in citys,and there are passenger flow corridors that undertake the main passenger flow.Wether can we use the public time and space big data to accurately analyze the key indicators in the bus research such as passenger flow demand and headway distance?Bus line network layout and reconstruction,urban public transportation operation management and dispatch control,passenger flow emergency and support are of key significance.This paper firstly establishes the flow of the multi-source data analysis system for public transportation based on the pre-processing and analysis of bus operation data.Then,based on the spatial-temporal characteristics of public transportation big data,a method for processing grid based public transit spatial-temporal data is established.Then based on the data pre-processing and analysis,combined with the time and spatial characteristics of the ground passenger flow and the influencing factors such as holidays and weather,a spatial-temporal deep learning method that can comprehensively consider the above factors is proposed to predict the urban bus transit passenger flow.The method consolidates,stacks,and uses historical bus transit data,holidays data,and weather conditions data for the Conv-LSTM layer,LSTM layer,and convolution layer.This paper applies the method to the spatial-temporal passenger flow prediction of Changzhou ground bus system,and compares the method with ARIMA,DNN,CNN,LSTM,Conv-LSTM and other prediction methods based on evalutation with RMSE,R~2.The results show that the spatial-temporal deep learning method has better prediction performance,and it also proves that a good passenger flow prediction model needs to consider both time,space and exogenous dependence to obtain better prediction performance.Then the thesis addresses the practical problems such as the frequent bus rides and the long waiting time of passengers in the urban ground bus passenger flow corridor,as well as the bus line in the passenger traffic corridor in the urban traffic system,the complex and diverse service levels,and the concentrated operation of the passenger flow.With the goal of optimizing bus dispatching system,scientific decision-making and improving bus service level,the prediction of ground bus headway distance on passenger flow corridor is carried out.Based on the prediction of the headway of the passenger flow corridor with the GBRT algorithm,the results show that GBRT is superior to KNN,KF,ANN and LS-SVM in terms of robustness and accuracy in the headway prediction scheme mentioned in the paper comparison experiment.Predicted bus headway information can help passengers better adjust their travel plans to reduce waiting times of bus station level.Bus operation management departments such as bus companies can use real-time control strategies based on the predicted headway time to maintain a stable bus headway and provide real-time bus headway distance information.