Vehicle Speed Prediction Using a Markov Chain With Speed Constraints

Abstract

Modern vehicles are designed to improve fuel consumption while satisfying emissions regulations. As a result, powertrains are becoming increasingly complex and changing rapidly. Optimal control based on the future vehicle speed is one way to address these changes. In this approach, accurate prediction of velocity is closely related to the performance of optimal control. However, there exists uncertainty and complexity in predicting the driver’s behavior, which can be influenced by the surrounding driving environment. In order to overcome such limitations, we propose a prediction algorithm of vehicle speed based on a stochastic model using a Markov chain with speed constraints. The Markov chain, which forms the basis of the proposed algorithm, generates the velocity trajectory stochastically within speed constraints. The constraints are estimated by an empirical model that takes into account the road geometry and is organized by the intuitive form of matrices. To reduce the complexity of the vehicle position and roadway integration, a curvilinear coordinate system is presented using GPS information and a roadway geometry model. Based on these coordinates, the algorithm predicts the future velocity for each cycle from the current vehicle position up to the ahead distance. The proposed algorithm is evaluated through experiments in an urban area, and the test vehicle collects driving data on the roadway position domain. The algorithm was verified in the following conditions: free of surrounding vehicles and traffic lights, the average velocity of the test vehicle is 47.8 km/h, and the maximum velocity of the test vehicle is 66.6 km/h. The experimental results show that a 3.8041 km/h root-mean-square-error is achieved with a prediction horizon of up to 200 m.