The Energy Management Strategy Research For Plug-in Hybrid Electric Vehicles Based On Intelligent Optimization

This paper focus on plug-in hybrid electric vehicle,and embarks on the energy management study for a power-split plug-in hybrid electric vehicle and a multi-mode plug-in hybrid electric vehicle,respectively.Quadratic programming,dynamic programming,Pontryagin's minimum principle,neural network and model predictive control are employed to the investigation,and energy management strategy considering engine on-off sequence,based on rule fitting and based on multi-neural networks are proposed in this paper.The main research works can be summarized as follows.(1)Plug-in hybrid electric vehicle model in a power-split type and a multi-mode type are established and analyzed in detail.The power flow relationships and state functions in different model type and working scenario are deduced,so that the corresponding simplified calculation-oriented battery model and engine optimal operating line can be built up,providing convenience for the subsequent deduction and calculation.(2)Concerning the power-split plug-in hybrid electric vehicle,a global energy management strategy based on quadratic programming and genetic algorithm is proposed.According to the simplified vehicle model,the quadratic relationship between battery power and instant fuel rate under different vehicle velocity and power demand are built up,and the approximation precise is validated.In terms of global fuel consumption optimization,the genetic algorithm is deployed to optimize the on-off sequence by treating the sequence as a time series and engaging the quadratic programming to optimize the battery power.Simulation results show that the proposed approach can effectively achieve eco-driving and improve the fuel economy with the saving percentage of 10.46% in maximum.(3)In regard to Chevrolet Voltec,the multi-mode type plug-in hybrid electric vehicle,an optimal rule extraction-based real time energy management strategy is proposed for online application.Firstly,dynamic programming is adopted under a mass of driving cycles in different cycle types and cycle lengths to solve the global optimal operating mode and battery power,and the optimal results are employed as the training datasets for rule extraction.Subsequentially,back propagation neural network is utilized to learn the implicit mode distribution regulations concealing in the optimal information.Combining driving pattern recognition,battery power fitting under different driving pattern is also conducted,as a result,the battery power rule is extracted as an indexed Tab.which can be directly used for online power allocation.The simulation results manifests that the total cost can be reduced by 10.92% in maximum comparing with charge-depletion/charge-sustaining strategy,and the proposed online rules result in a similar battery state of charge dropping trajectory,comparing with what dynamic programming dose.(4)Referring to the multi-model plug-in hybrid electric vehicle,on the basis of aforementioned study,a multi-neural network predictive online energy management strategy is proposed.Pontryagain's minimum principle is conducted to explore the optimal co-state under various of driving cycles.Recurrence neural network is firstly employed to learn the co-state variation,so that the co-state in an uncovered driving scenario can be predicted in real time.Moreover,the vehicle speed prediction based on back propagation neural network and radial basis function neural network are compared and analyzed to find the propriate one on speed prediction,and the radial basis function neural network is consequently selected.Equivalent consumption minimum strategy and general minimum principle-based model predictive control are employed as the comparison benchmark.Simulation results illustrate that the introduced training character can improve the recognition accuracy from 96.65% to 99.08%;online battery power calculation can be achieved by co-state estimation,of which the computational time can shorten by 57.3% of general minimum principle-based model predictive control;combining the slacked state reference and online co-state correction,the reference traction is realized,and the total cost can be reduced by 7.43% comparing with charge depletion/ charge sustaining strategy,reaching 34.8% of dynamic programming's result;in addition,the total cost can be reduced from 12.2% to 34.68% comparing with equivalent consumption minimum strategy.