| Since the reform and opening up,with the continuous improvement of national per capita GDP,people’s demand for travel has increased,and the number of private vehicles has expanded sharply.The rapid growth of motor vehicles has also caused a series of traffic problems,leading to the increase of traffic accident rate and frequent traffic jams.The core of intelligent traffic control system(ITS)and traffic flow guidance system is to analyze the traffic flow of road vehicles in real time and give early warning of traffic congestion in the future.In order to control the traffic flow in time,its system needs to be able to obtain accurate information of road traffic flow and make short-term prediction and analysis of traffic flow information.Through its system,the traffic management department can accurately obtain the real-time traffic flow and dredge the possible congestion in the future lanes in advance.The key point of intelligent transportation system is to monitor and predict the short-term traffic flow in real time.The prediction of short-term traffic flow is the future development direction of its system.After research at home and abroad,there are many methods for short-term traffic flow prediction,including periodic time series prediction based on traffic flow data,signal decomposition and combination based on nonlinear characteristics of traffic flow data and neural network prediction.Because the road traffic flow is easily affected by time period,surrounding bayonets and weather conditions,it is not comprehensive to only consider that the traffic flow has a single characteristic in one aspect.Therefore,the prediction of traffic flow needs to comprehensively consider the multiple characteristics of traffic flow.According to the previous research on short-term traffic flow,and combined with the advantages and disadvantages of different methods.In this paper,a multi-component combination model based on adaptive noise complete set empirical mode decomposition(CEEMDAN)is proposed to predict short-term traffic flow.Firstly,the original traffic flow data is obtained and preprocessed.The processed data is divided into multiple IMF components by CEEMDAN signal decomposition method;Then,the permutation entropy algorithm is used to calculate the time series complexity of different components,and multiple components are divided into high-frequency components,intermediate frequency components and low-frequency components;Then,the LSTM model optimized by the small batch gradient descent method is used to predict the high-frequency component,the LSSVM model based on genetic algorithm is used to predict the intermediate frequency component,and the ARIMA model is used to predict the low-frequency component;The final predicted value of traffic flow is equal to the sum of the predicted values of each component.This paper uses the traffic flow data of the south to north section of Wujia Jucheng Road Station in Yichang,Hubei Province as the experimental data.Mean square error(MSE),mean absolute error(MAE)and mean absolute percentage error(MAPE)are used as evaluation indexes of prediction effect.The bayonet data is pre processed into the final predicted value of the constructed model output,and compared with the results of a single model(long-term and short-term memory neural network,least squares support vector machine,differential integration moving average Autoregression).Through comparative experiments,it can be concluded that the average absolute error,average absolute percentage error and mean square error of the multi-component combined traffic flow prediction model based on CEEMDAN decomposition used in this paper are 8.2431,7.4285 and 114.1549,which are lower than those of the control model,indicating that this composite model makes up for the deficiency of a single model only for one aspect of the characteristics of traffic flow data and improves the prediction accuracy,It can meet the prediction demand of short-term traffic flow. |
PDF Full Text Request