Real time adaptive non-linear estimator/predictor design for traffic systems with inadequate detectors

Traffic control and management needs accurate estimation and prediction of traffic variables such as flow, speed, volume, travel time etc. Linear stochastic time series methods are powerful analytical tools. The capability of state space reconstruction makes them popular in traffic prediction and estimation. In this study, to overcome the linearity assumption of these methods, a non-linear kernel-based stochastic time series method with state space reconstruction capability is proposed. To minimise the prediction error of the method, adaptive time variant transformation from primary space to reproducing kernel Hilbert space is proposed by employing extended Kalman filter. Owing to high costs of traffic detectors, not all the metropolitan areas are equipped with these sensors; therefore in this study, an extended Kalman observer based on the new dynamic-adaptive-non-linear predictor is designed and applied for traffic flow estimation and prediction in the areas that suffer from lack of detectors. Practical data simulations and evaluations justify the high strength and accuracy of the proposed method in prediction of traffic speed with incomplete data sources.