Predictive Energy Management Of Hybrid Electric Vehicle Powertrains

Hybrid electric vehicle(HEV) technology is an effective way to resolve the currently fossil energy over-consuming, air pollution and high gasoline price problems caused or partly caused by traditional internal combustion engine(ICE) vehicles. This dissertation targets at power-split hybrid electric vehicle powertrains, and proposed a predictive energy management methodology concerning HEVs and plug-in HEVs, through a series of studies at online control decision optimization, real-time future velocity forecast and traffic information engaged global energy planning. The main contributions include:Established a predictive energy management topology for hybrid electric vehicle powertrains. Dynamic programming(DP) and model predictive control(MPC) are applied to realize real-time receding control of the main power components in HEVs. The sensitivity of fuel consumption to the discretization accuracy of control and state variables in DP is analyzed, which determines that the optimal variable discretization grid number is between 30 and 50. The sensitivity of online computation time in MPC to the prediction/control horizon length is also analyzed, which determines that the real-time implementable prediction/control horizon length is between 6 and 12 seconds. The globally optimal energy management results derived from DP and the short-term future velocity pre-known real-time energy management results derived from MPC are compared in terms of state variable trajectory evolving, torque splitting and fuel consumption. This study provides a fundamental result for further improvement of the predictive energy management formulation.Exhibited and quantified the improvement in fuel economy by introducing short-term velocity forecast techniques into the predictive energy management strategy, and proposed an artificial neural network(ANN) based short-term velocity forecast method for HEV powertrain control. Three different ANN models are formulated for velocity forecast purpose, namely the back-propagation NN, the layer-recurrent NN and the radial basis function NN. The existing exponentially varying and stochastic Markov-chain velocity forecast methods in literature are extended. All the above three velocity forecast methodologies are compared under predicting precision, computation time, constraint enforcement performance and the consequent fuel consumption when combined with the predictive energy management formulation. Simulation and hardware-in-the-loop(HIL) test results show that employing the radial basis function NN for online short-term velocity forecast improves the powertrain fuel economy by at least 8% compared with the other two approaches.Implemented the radial basis function NN short-term velocity forecast approach to the adaptive equivalent consumption minimization strategy(A-ECMS). Before that, a novel equivalent factor(EF) online adaptation law is proposed based on the bisection converging principle, which resolved the iteration vibration problem existed in the commonly used EF adaptation laws. Also the optimal recursion iteration frequency in the EF optimizing is determined to be 6~8 per time step through sensitivity analysis. Three situations of A-ECMS are compared and evaluated: when the short-term future velocity requests are fully known, when fully unknown(history velocities are used), and when forecasted by the radial basis function NN. Simulation results show that the ANN based short-term velocity forecast method is able to improve the fuel economy of A-ECMS by over 3% compared with the traditional approach. The effectiveness of ANN based short-term velocity forecast in improving the HEV fuel efficiency is thus further validated.Proposed a traffic data enabled predictive energy management strategy. Plug-in HEVs usually request the terminal battery SOC falls to the minimum by the end of the target driving task. In this case, a traffic information based global energy usage planning approach is proposed to adapt the predictive energy management strategy to the PHEVs. To catch and follow the traffic dynamics, a simplified PHEV powertrain model is proposed based on the energy balancing principle, with which DP is able to generate the globally optimal SOC trajectory with 80% computation time reduced, and maintain the average RMSE within 3%. The generated battery SOC trajectory is introduced to the predictive energy management strategy as terminal constraint references in each control horizon. The impacts of the traffic data updating pattern and the SOC reference indexing mode on the consequent fuel economy are also analyzed. Eventually, the global energy usage planning problem for PHEVs is systematically solved in this dissertation.