Research On Bus Arrival Time Prediction Based On LSTM-PF Model

One of the important indicators for measuring the quality of public transport services is whether the passengers' needs are met.Among them,the accuracy of the arrival time of public transport vehicles is one of the most concerned information for urban residents.Accurately predicting the arrival time of public transportation vehicles and displaying them to passengers will not only help passengers to make corresponding travel plans,but also save travel time and travel costs.At the same time,it also enhance the attractiveness of public transportation systems.So,it is helpful for the relief of traffic congestion and the development of urban information.Firstly,based on the acquisition method and data format of public transportation data,this paper pre-interpolates and normalizes the collected data,and determines the input of the model built in this paper based on the analysis of the bus arrival process and influencing factors.The variable is the vehicle arrival time of the first 2-4 stations in the demand site,and the corresponding algorithm for calculating the inbound time is designed.It is realized by software such as Matlab and MapInfo.Finally,the feasibility and accuracy of the algorithm are verified by the vehicle survey method.Secondly,the method of predicting the arrival time of public transportation vehicles is summarized and compared.Based on the principle,advantages and disadvantages of LSTM and particle filter,the proposed method of predicting the arrival time is the combination of the two methods(Public Transportation Arrival Time Prediction Long Short Term-Memory and Particle Filter,LSTM-PF).The learning rate of the LSTM network model in the LSTM-PF model is improved,and it is optimized from two aspects:function selection and network structure.The historical data is used to train the optimized LSTM neural network to fit the nonlinear relationship between the demand site and the previous station arrival time.Then,the real-time arrival time is used to predict the arrival time of the demand site,and the particle filter dynamically adjusts based on the predicted results to find the optimal arrival time,and the LSTM-PF model study verification be completed.Finally,several typical bus lines be selected as experimental lines in Beijing.The LSTM-PF model be used to predict the arrival time of the bus on the working days and non-working days,and the accuracy of the prediction is measured by the MAE value.The MAE value is basically controlled within 1.5 minutes.In order to further verify the prediction accuracy of the LSTM-PF model,this paper compares the results with the particle filter and the optimal results predicted by different LSTM models.The results show that the prediction results of the LSTM-PF model are better than the other two Compared with the standard LSTM network model and the standard particle filter model,the prediction results of the LSTM-PF model on working day(average absolute error:41.64 seconds)were improved by 35.84%(mean absolute error:64.90 seconds)and 49.20%(mean absolute error:81.97 seconds),the predicted results of non-working days(average absolute error:55.96 seconds)improved by 27.42%(mean absolute error:77.10 seconds)and 38.82%(mean absolute error:91.47 seconds).