State Space Modeling and Short-Term Traffic Speed Prediction Using Kalman Filter Based on ANFIS

Abstract—Speed is an important component in any traffic control or monitoring systems especially for road safety. This paper presents a novel short-term traffic speed prediction model using Adaptive Neuro Fuzzy Inference System (ANFIS). By applying this method, despite nonlinear nature of traffic, linear time variant state space model will be presented. Kalman Filter (KF), based on this model, will reduce modeling error and modify prediction accuracy. Using this method, KF will be applied to the nonlinear system so Jacobian computations of Extended Kalman Filter (EKF) that is essential for nonlinear systems are not needed. Another advantage of suggested method is that there is no need to design different ANFIS structure for different predict horizons in order to obtain acceptable prediction accuracy, because the error due to the model structure is to some extent reduced by filter. Simulation results with real data sets indicate that this model is an efficient way which surpasses a common multilayer feed forward Neural Network (MLFNN) and an ANFIS predictive model.