Research On Passenger Traffic Forecast Of New Bus Payment System Based On Hadoop

As the main means of daily travel of the citizens,urban public transportation is one of the most direct and realistic life issues people care most about.Constantly improving the quality of public transport operation,improving public transport operation services and meeting the needs of the masses can effectively alleviate traffic congestion and promote social harmony.The rational operation and intelligent scheduling of urban public transport management departments are very helpful to the promotion of public transport services.The passenger flow data of public transport is the key data to predict passenger flow in a certain period of time and achieve reasonable operation and intelligent dispatching.So it is necessary to find a better way to improve the accuracy of the passenger flow forecast.With the expansion of urban scale,the increase of urban population and the rapid development of Internet technology,the era of "big data" has come.The study of passenger flow prediction has become more complex and demanding.In order to achieve accurate passenger flow prediction,two key factors are needed: complete passenger flow data and good prediction algorithm.However,in the past,the data obtained by the passenger data acquisition methods have some shortcomings such as small amount of information,incompleteness and difficulty of sorting;and the traditional forecasting model has its own defects,according to the current situation of passenger flow of large amount of data and complex changes,the traditional prediction model has been unable to meet the forecast requirements.So it is particularly important tofind a better way toimprove the accuracy of the passenger flowforecast.This thesis based on the actual research projects,firstly,the current research status of bus payment system and passenger flow forecast under Hadoop platform is analyzed,and the direction of the research is determined;the related technologies of the new bus payment system are analyzed,in view of the new payment system,this thesis analyzes the advantages of the complete,accurate and comprehensive passenger flow data,and provides reliable data for the passenger flow forecast.At the same time,the overall scheme of traffic flowforecasting based on Hadoop distributed processing platform is designed.Using the results of on-site investigation,this thesis analyzes in detail the factors that affect the change of passenger flow,from the quantitative point of view,determining the influential factors that should be considered in the study of passenger flow forecasting;developing a preprocessing method of passenger flow data;and then analyzing and comparing several commonly used algorithm for prediction of passenger flow,showing that the BP neural network with error back-propagation has a good effect on solving complex stochastic nonlinear mapping problems;but there are other short comings,such as long training time,generalization ability,and easy to fall into the minimum value,on the above problems,inthis thesis,the Hadoop parallel processing technology is introduced into the passenger flow prediction,proposing that the weights and thresholds of BP algorithm are optimized by using particle swarm optimization(PSO)at first,then the MapReduce programming model is applied to parallelize the optimized algorithm,and then the whole algorithmis improved.Finally,structural design of the data processing platform to complete the Hadoop distributed processing platform.Under this platform,the improved parallel PSO-BP algorithm is tested by using the preprocessed passenger flow data,and the experimental data are calculated and analyzed.The results of the test show that the convergence speed and prediction accuracy of the improved algorithm proposed in this thesis are significantly better than the original algorithm,and can achieve faster training speed on largedata sets.