Research On Short-term Bus Passenger Flow Prediction Model Based On Improved Kalman Filter Algorithm

With the continuous proce,eding of Reform and Opening,China has seen considerable progress in the development of its public transport system.The state has formulated policies and strategies to give priority to the development of public transport.Under this background,intelligent public transportation system emerges as the times require.The implementation of the system proved promising effects,yet during which time,as discovered,it will provide support to the regulation and administration of public transport when the information of short-term passenger flow is timely and accurately acquired,even predicted.Short-term bus passenger flow prediction is affected by many factors,for instance,policies,environment,weather and etc.Many experts and scholars at home and abroad have carried out researches on short-term passenger flow prediction.Their methods are mainly based on statistics,nonlinear and hybrid methods.In particular,methods that involve implementing nonlinear algorithms such as neural networks and Kalman filter manifest good results after experiment.However,the prediction model based on Kalman filter algorithm has divergent prediction results and large errors;while the prediction model based on neural networks has poor filtering performance for the noise of input value,and too much dependence on sample data,sometimes resulting in unwanted outcomes.This paper proposes a short-term passenger flow prediction model based on improved Kalman filter algorithm,after thoroughly revising the status quo of the relevant fields.This model,supported by BP neural networks,takes advantage of the principals of Kalman filter algorithm,makes full use of the merits of the two algorithms and compensates for their defects mutually.The parameters(noise covariance,estimation errors and Kalman gain)that are generated in the process of Kalman filtering are treated as input arguments to BP neural networks.Then the output is combined with that of the prediction of Kalman filtering,to eliminate estimation errors.This paper has adopted transaction data of IC cards and GPS data records of No.28 bus in Dalian.Data sample is acquired after cleansing comparison of the raw data.Through analysis,this paper examined the circumstance that the passenger flow data of No.28 bus within adjacent time periods of the same day,the same time period of adjacent days,weeks and months,and the average of the historical data of the same time period are strongly correlated.An improved Kalman filter model for short-term passenger flow prediction is established.The results showed that the error of the improved model is smaller than that of the conventional models.This paper has constructed a short-term bus passenger flow prediction model based Kalman filter algorithm,modified by BP neural networks.The model has solved the problems of conventional algorithms that easily produce divergent and error-prone results,and offers a novel thinking and direction for further researches on this field.