Short-term Freeway Traffic Flow Prediction Combining Seasonal Autoregressive Integrated Moving Average and Support Vector Machines

摘要

Short-term traffic volume predictions support proactive transportation management andtraveler information services. To cope with periodicity, nonlinearity, uncertainty and complexity ofshort-term traffic, the seasonal autoregressive integrated moving average (SARIMA) and supportvector machines (SVM) models are often employed separately to forecast traffic flow time seriesin the previous studies. SARIMA can discover the intrinsic relations (correlations) among the timeseries data, especially fit for seasonal, stochastic time series modeling. On the other hand, SVMhas a strong nonlinear mapping ability for input and output, appropriate for solving nonlinear andcomplex problems. A few combining schemes have been presented previously, but they tend to becomplicated. So as to establish a simple and effective hybrid model, a novel hybrid methodologythat combines both SARIMA and SVM models is proposed to take advantage of the uniquestrengths of SARIMA and SVM models in this paper. The selection of input features in SVM notonly refers to "parsimony, efficiency" modeling ideology of SARIMA model which setups anequation of correlations among the time series data, but also SARIMA forecast results takes as oneof input features. Two key issues in building a hybrid model process, the identification of inputdimension of SVM via SARIMA and parameters optimization of the hybrid model with particleswarm optimization (PSO), are discussed in this article. Experimental results with real-worlddatasets indicate that the hybrid model is superior to the individual model (SARIMA or SVMmodel) in terms of RMSE and MAPE for the prediction of the short term traffic flow.