Energy Management Strategy Based On Stochastic Model Predictive Control For Plug-in Hybrid Electric Vehicles

Because of the variety of working modes and sources of energy for plug-in hybrid electric vechicles,they require energy management.This task can be exstracted to an optimization issue under constrains of characteristics of components and limitations for the sake of safety.Meanwhile,the high uncertainty of potential driving cycles in the future pose chanllenges for the online application of these strategies.Thus,this thesis proposes energy management strategy based on stochastic model predictive control and the academic or engineering details are as follows:(1)Firstly,to make clear the technical background for further strategy development,this thesis gives an investigation and introduction of the subject bus as well as various driving cycles,verifys the reasonability of parameters of powertrain systems according to the statistical power feature of the cycle which represents the practical running situation of the bus and constructs energy transfering and converting model;what should be mentioned is that state of charge(SOC)of the power battery is an essential but hidden state,whose estimation accuracy can affect the advantage of energy management directly.For this problem,this thesis estimates SOC based on extended Kalman filter and so on and the errors are within 8%;then by matching the possible sources of errors and the characteristics of algorithms,three methods are combined.As a result,the SOC estimation error is within 2.5% while excellent preliminary stability is ensured.(2)This thesis designs energy management strategy based on dynamic programming.Besides,critical performance indicators for the application of this methodology is analyzed: in order to speed up simulation process,the length of input cycles need to be reduced,and the conclusion is that by scaling cycles and power capacity simultaneously can maintain the similarity of fuel consumption for 100 km mileage between the figures before and after scaling,which means the simulation results after scaling can also be used for strategy estimation;another exploration is the simplification of control variables when taking specific driving cycles into consideration.(3)Before using model predictive control to manage energy,prediction model for driving cycles based on Markov Chian Monte Carlo Method is designed.Through combining multi-scale single-step prediction and post-processing for forecast results,predictions with certain accuracy for typical driving cycles and practical cycles are conducted.Furthermore,in the latter one,to cope with the difference between reference database and practical driving cycle,history velocities and adjacent states are adopted to update reference states thus overcome the deficiency of reference state;besides,online accuracy estimation method are designed to eliminate velocity sequence with relatively higher errors and in this way,adopted prediction cycles are with variable horizons.The accuracy is improved from level of fixed horizon 35s(12.8km/h)to level of fixed horizon 20s(9.6km/h).(3)Finally,with DP and MCMC as the rolling optimization and model prediction methods repectively,energy management strategy based on stochastic model predictive control which can run real-timely is proposed;when the strategy is adopted for prediction of typical driving cycles,the post-precessing of predicted velocity sequences can reduce fuel consumption by 1.9% while in forecast of practical cycles which are gotten by hardware-in-the-loop experiment,the setting of variable useful horizon can reduce fuel consumption by 3.9%.