With the ever-increasing demands of energy conservation and emission reduction,every country has formulated strict regulations on motor vehicle fuel consumption and emissions.Hybrid electric vehicle is one of the core directions in the competition field of new energy vehicles,and its corresponding energy management strategy is a key factor to determine the power,economic and emission performance of the entire vehicle.By taking a P2-parallel hybrid vehicle as the research platform and considering its hybrid system dynamics characteristics.Hybrid model predictive control strategy is adopted and explored,in order to improve the fuel-saving effect and algorithm computing efficiency.The main research contents include:(1)System model construction and validation.The P2-parallel hybrid vehicle model is constructed by forward simulation method,and the equivalent fuel consumption minimization algorithm is adopted.The simulation results are compared with the test results to verify the effectiveness of the simulation model,which lays a foundation for the subsequent control strategy research.(2)The nonlinear model predictive control problem is solved by nonlinear optimization algorithm.When the driving cycle is locally known,the algorithm with prediction herizon between 2 and 30 is simulated using dynamic programming method.The influence of prediction herizon on vehicle fuel economy is analyzed,and finally the appropriate prediction herizon is determined.The continuous generalized minimum residual algorithm is used to improve the calculation speed because of the slow calculation speed of dynamic programming,and the simulation results of the two algorithms are compared and analyzed.(3)The hybrid model predictive control algorithm is used to solve the energy management problem.The researched object vehicle model contains both continuous and discrete variables,which is a hybrid system.Firstly,the vehicle model is linearized;then,the model is transformed into a hybrid logic dynamic system model and optimized;finally,the optimization results and calculation speed of the hybrid model predictive control algorithm are compared and analyzed.(4)Hierarchical hybrid model predictive control algorithm is proposed and validated.Firstly,the hybrid nonlinear model is treated by convex optimization.Then,the model is decomposed into two parts: torque optimization and transmission ratio optimization.Finally,the optimization results and computing speed of the hierarchical hybrid model predictive control algorithm are compared and analyzed.(5)Application of model predictive control algorithm is explored,considering different sources of speed profile information.Fistly,the exponential prediction,neural network prediction and neural network prediction based on the vehicle operating condition identification are used to predict the driving cycle,and the influence of prediction error on fuel economy is analyzed.Thanks to the rapid-growing Vehicle-to-Everything(V2X)capabilities,the leading vehicle information,traffic light information and road slope information can be obtained.On this basis,the model predictive control is used to control the fleet,and the impact of recieved information on fuel economy of the fleet is analyzed.Based on the above work,the following conclusions are obtained:(1)By using the model predictive control framework,when the condition information is locally known,the vehicle fuel economy is continuously improved with the increase of the prediction herizon,and the calculation time increase.Considering vehicle fuel economy and algorithem computing time,the prediction herizon 15 is ideal for application.Compared with the dynamic programming,under the NEDC?LA92?UDDS and WLTC driving cycles,the fuel consumption of the continuous generalized minimum residual algorithm increases by 28.9%,42.3%,59.2% and 27.6%,respectively,which optimization effect is poor,can not deal with discrete variables.(2)Compared with the continuous generalized minimum residual algorithm,the vehicle fuel economy of hybrid model predictive control algorithm is improved by 9.1%,8.3% and 3.0%,respectively,under the NEDC,UDDS and WLTC cycles,and is decreased by 4.3% under the LA92 cycle.The hybrid model predictive control algorithm computing speed increase rapidly with the increase of the prediction step herizon.When the prediction herizon is 1,3 and 6,the single-step solution time is 0.14 s,1s and 11.7s.The solution speed is slow and can not meet the demand of real vehicle.(3)Under the NEDC,UDDS,LA92 and WLTC cycles,the vehicle fuel economy of the hierarchical hybrid model predictive control is improved by 5.5%,3.7%,12.2% and 7.8%,respectively,compared with the minimum equivalent fuel consumption algorithm.The single-step computing time of the hierarchical hybrid model predictive control is around 0.4s,which has the potential for practical application.The linear time-varying model predictive control method is tested on a real vehicle.By adjusting the control weights and engine starting condition,the three vehicle drum test results are 7.03L/100 km,6.4L/100 km and 6.2L/100 km,respectively,which validate the optimization effect and real-time performance of the algorithm.(4)The neural network prediction method based on the speed profile identification has the highest prediction accuracy,while the prediction error will deteriorate the fuel economy.Based on the vehicle network condition,the current vehicle information is known,which can make the vehicle fleet have a higher fuel economy,a shorter following distance and better riding comfort.When the traffic light information is known,the vehicle acceleration and deceleration can be smaller,the idling time is shorter,and the fuel economy of the fleet can be improved by 6.6%.When the road slope information is known,the fuel economy of the fleet can be further improved by 5.1%.For the P2-parallel hybrid vehicle energy management problem,a model predictive control framework,along with high fidelity model linearization,hierarchical layering and vehicle-to-everything networking,is explored and applied.The promising results of fuel economy improvements indicate that the proposed algorithm may provide important insights for future hybrid vehicle development. |