Multi-scale Prediction Method Of Driving Cycle For HEV Energy Management Strategy

The supervisory control strategy of a HEV coordinates the operation of vehiclesub-systems to achieve performance targets such as maximizing fuel economy andreducing exhaust emissions. This control problem is commonly referred as the powermanagement problem. In the past, many supervisory control strategies weredeveloped on the basis of a few pre-defined driving cycles, using intuition andheuristics. The resulting control strategy was often inherently cycle-beating andlacked a guaranteed level of optimality.In this study, the power management problem is tackled from a stochasticviewpoint. An infinite-horizon stochastic dynamic optimization problem is formulated.The driving cycle is modeled as a random Markov process. The optimal controlstrategy is then obtained by using Stochastic Dynamic Programming (SDP). So thisthesis researched the Multi-scale Prediction Method of Driving Cycle for HEVEnergy Management Strategy. This work consisted by the following aspects:1. Scholars both in domestic and overseas began to use Markov chain to designor predict driving cycles for the past few years, but none of them gave the validityproof of Markov property of driving cycle.In this study, Vehicle Dynamics model, Car-Following model and SingularPerturbation theory were all used to verify the Markov property of driving cycle in thefirst time. It is the premise of the predicting model of driving cycle.2. A Markov chain model of driving cycle was built in the thesis. Then Markovchain Monte Carlo simulation was used to predict driving cycle.ECE and FTP cycles were used to make as historical information to predictdriving cycle. It used different time scales and was used to analyze different influenceof prediction outcomes of driving cycle which come from different time scales.3. In this thesis, it was demonstrated that the prediction results of driving cycleare in connection with scale which was chosen. Moreover, these results using differentscales have their own advantage, so that a data fusing method of multi-scale wasproposed.The comparison between the deviation of multi-scale prediction results anddeviation of a single scale prediction results illustrated effectiveness and robustness ofthe multi-scale prediction methods to predict the length of prediction time. In order toimprove the prediction accuracy, the data fusing method of multi-scale consideredweight was researched and the results were optimized by using it. On the basis of itthe prediction error has become less and the correlation can be controlled at99%or more, the prediction time also meets real-time requirements.Finally prediction methods of deviation which were mentioned in differentdriving cycle in this thesis were contrasted. It was showed that the multi-scaleprediction method is effectiveness and is stable in prediction accuracy. Otherwise, theprediction time of the method at one time was calculated. It illustrated that the methodcan conform to the demand of HEV real-time controlled and is worth to be spread.