In this paper, a combined model of ARIMA and RBF neural network is proposed by combined the good linear fit ability of ARIMA and the strong dynamic nonlinear mapping ability of RBF neural network. The velocity of microwave is predicted in real time with the consideration of the temporal characteristics of traffic flow by the models.The results indicate that the Mean Absolute Percentage Error of combined model is lower, and the goodness of fit of combined model is higher. Keywords: traffic engineering, traffic flow prediction, Combined Model, ARIMA model, RBF neural network model
1.IntroductionIn order to solve the traffic problems such as traffic congestion and traffic accidents, Intelligent Transportation System (ITS) has been developed and applied to practice. The core research on ITS is traffic control and guidance, and short-term traffic flow prediction is the data foundation and decision support of traffic control and guidance system.In recent years, domestic and overseas experts and scholars have proposed a variety of combinatorial models or improved models. Xiaomo Jiang et al.[1] proposed a neural network combined with wavelet analysis for traffic flow prediction; Yanru Zhang et al.[2] used a modified gradient boosting regression tree to predict travel time; Yisheng Lv et al. [3] used a deep learning model for traffic flow prediction, which combines BP neural network and stack automatic encoders; Wu Wei [4] proposed a PSO-PLS (particle swarm optimization-partial least squares regression) combination forecasting method. Sun Liguang et al. [5] proposed a double-layer update mechanism which used a recursive regression method to update the sub-model coefficients and weighting coefficient; Zhang Jinglei et al.[6] used a three-tier structure of the RBF neural network for non-linear combination of RBF and ARIMA; Chen Gang [7] used BP neural network to amend error of the gray prediction model. Qian Wei et al [8] combined BP neural network model and GM (1,1) model with a variable weight coefficients form.In many prediction models, ARIMA model can reflect the trend of traffic flow time series, and it is easy to apply to traffic flow forecasting, while RBF neural network model has strong dynamic non-linear mapping ability, and it has high satisfaction and accuracy to stochastic sample. Based on the good linear fitting ability of ARIMA model and the powerful dynamic nonlinear mapping ability of RBF neural network model, this paper builds models and uses the speed-based traffic flow state identification