Research On Short-time Bus Passenger Flow Prediction Method Based On KAL Combination Model

As China’s economy continues to grow and the urban population continues to increase,people’s demand for urban transportation travel continues to grow,and traffic problems in cities are becoming increasingly serious.Short-time bus passenger flow prediction can effectively help public transportation system to manage scheduling,help passengers to plan their trips in advance,reduce the operation cost of public transportation and ease the travel pressure.It is extremely important to adopt suitable model algorithms to make accurate prediction of shorttime bus passenger flow.This thesis analyzes and summarizes some Chinese scholars’ research on passenger flow forecasting methods for China’s urban transportation system,and compares and refers to foreign scholars’ theoretical modeling methods.In view of the shortcomings of traditional single network models and partial combinatorial models in dealing with multiple scenarios of passenger flow data input,such as neglecting multiple influencing factors and insufficient training efficiency,which result in less accurate prediction,a short-time bus passenger flow prediction method based on the KAL combinatorial model is proposed,and the main research contents are as follows.(1)Research and construction of K-means-Attention-LSTM combination modelAccording to the structural characteristics of K-means clustering algorithm and LSTM neural network,the input layer and hidden layer of the combined K-means-Attention-LSTM model are determined,and the overall structural design and algorithm formulation of the KAL model are determined.(2)Short-time bus passenger flow prediction method of KAL combined modelThe paper proposes a short-time bus passenger flow prediction model based on KAL.The K-means clustering algorithm is used to select the clustering indicators with high correlation with passenger flow and perform K classification,based on which the Pearson(Pearson)correlation coefficient method is used to analyze the K subclass data set and eliminate irrelevant variables to achieve the purpose of simplifying the model structure and reducing the input dimensionality.The weight function W of the attention mechanism is continuously adjusted to give the model higher stability.(3)Experimental comparison and validation of the short-time bus passenger flow prediction method based on the KAL combination modelBus passenger flow data at different stations and under different weather conditions are used for prediction,and a total of three representative methods,linear,nonlinear and combined models,are selected for comparison,and three types of passenger flow prediction results evaluation indexes are used.Experiments and results analysis of the short-time passenger flow prediction method based on the KAL combined model are conducted,and multiple group comparisons of control variables are performed.This paper uses the bus data provided by the data platform of Dalian Bus Group as the research basis,and uses data mining technology and neural network prediction methods to solve the defects of traditional single network models and partial combination models that tend to ignore the information of multiple influencing factors and insufficient training efficiency when dealing with the input of passenger flow data of multiple scenarios,and has certain reference value for the scheduling of the intelligent bus system in Dalian,and the model running time The model runtime and prediction accuracy and stability have been significantly improved,and the research results are of some value to passenger flow prediction model research theory.