Markov Chain Modeling and On-Board Identification for Automotive Vehicles

摘要

The paper considers issues and algorithmic approaches related to modeling and identification of Markov Chain type models for vehicle applications. The use of Markov Chain models in these applications is stimulated by their ability to reflect aggregate vehicle operating conditions and induce "best on average" control policies based on application of stochastic dynamic programming and stochastic Model Predictive Control. A novel fuzzy encoding approach of continuous signals is proposed in which a signal value is simultaneously associated with multiple cells, and it is shown to enhance identification and prediction accuracy of Markov Chain type models. A computationally simple identification algorithm, suitable for on-board applications, is proposed to learn Markov Chain transition probabilities in real-time. Examples of real-time learning models of vehicle speed and road grade are reported to illustrate the overall identification approach.