Short-term Traffic Volume Prediction On Extreme Learning Machine And Improved KDE Method

With the rapid development of Intelligent Transportation Systems(ITS)in China,short-term traffic volume forecasting technology has attracted widespread attention in the practical field and academia.Efficient and accurate short-term traffic volume forecasting can provide the basis for real-time,dynamic and efficient traffic management and control systems.Short-term traffic volume forecasting has affected the development of intelligent transportation systems to a certain extent.Therefore,more accurate short-term traffic volume forecasting is needed to maintain the safe and stable operation of ITS.However,it is quite difficult to predict the traffic flow accurately considering the existing prediction models and the characteristics of traffic itself,such as nonlinearity,complexity and uncertainty.Therefore,a more scientific and effective prediction model is needed to describe the complex characteristics of short-term traffic data and trend prediction to help improve urban traffic management.In this paper,the basic characteristic parameters of traffic flow are firstly described,the characteristics of short-term traffic volume are analyzed,the data sources and data pre-processing methods are introduced,the advantages and disadvantages of different short-term traffic volume prediction models and the advantages and disadvantages of different probabilistic prediction models are compared and analyzed;then,the theoretical knowledge of extreme learning machine(ELM)method and improved kernel density estimation(AKDE-CKDE)is introduced,and the ELM-AKDE-CKDE hybrid forecasting model and explains the construction process of this model for deterministic forecasting;secondly,for the problem that the deterministic forecasting model cannot reflect the randomness of traffic volume,ELM-AKDE-CKDE probabilistic forecasting model is constructed and the solution method of this probabilistic model is given;then,the indicators of the mean absolute error,mean relative percentage error,root mean square error and root mean square relative error are chosen to evaluate the prediction accuracy of the hybrid model,and the average confidence interval width and kickoff percentage are chosen to evaluate the reliability of the model;finally,the traffic volume data of the A and B intersections of the main road in the main urban area of Chongqing are utilized for prediction,and the the prediction results of the short-term traffic volume hybrid prediction model ELM-AKDE-CKDE are compared with other prediction models.The results show that the short-term traffic volume hybrid prediction model ELM-AKDE-CKDE can not only achieve high accuracy deterministic prediction,but also reliably reflect the randomness of short-term traffic volume.The innovation of this paper makes full use of the advantages of the ELM model and AKDE-CKDE method to establish a novel combined forecasting method ELMAKDE-CKDE,which can take into account nonlinear and stochastic characteristics embedded in traffic volume data and exhibit a satisfactory performance,and provide strong support for urban traffic management.