Hybrid Electric Vehicle Energy Management Optimization Based On Prediction-planning Scenarios

In this thesis,the parallel hybrid electric vehicle(HEV)with CVT is taken as the research object.With the MPC frame as the principal energy management strategy,three different predictor models based on machine learning have been proposed to predict vehicle velocity in short-term time horizons,and a vehicle-to-road model incorporating traffic light has been established to realize path and velocity planning in advance for vehicle fleet in the abstract road.The main contributions are as follows:(1)The topology of parallel HEV has been mathematically modeled and different working modes of the parallel hybrid electric vehicle are analyzed,which lays the foundation for applications of energy management in parallel HEV.(2)The modeling of three different velocity predictors based on machine learning has been completed.The support vector machine(SVM)predictor has been autoregressively processed to enable it to predict velocity in short-term time horizon.The feedforward neural network(FNN)has been modified,and Dropout technology has been added to reduce the overfitting phenomenon during the training process.Due to the time series property of velocity,a long-short term memory neural network(LSTM)is also used to predict the future velocity.By using historical data of velocity as the training set,three predictors are trained,and the root mean squared error(RMSE)of different predictors is compared and analyzed.Among three predictors,LSTM-NN has the best prediction performance,followed by the feedforward neural network,and support vector machine has the worst performance.(3)The effect of prediction accuracy in short-term time horizon on vehicle fuel consumption during energy management is compared,and the results indicate that the higher prediction accuracy,the lower fuel consumption.Besides,the length of prediction horizon is adjusted to investigate its effect on fuel consumption.The best length of prediction horizon corresponding to minimum fuel consumption is summarized and,then,the optimal length in each predictor can be obtained.The velocity prediction methods in short-term horizons are verified to be effective in HEV energy management.(4)The abstract timing model of the traffic light signal points of in vehicle-to-road coordination scenario is realized.This model takes fuel economy,vehicle-to-vehicle spacing safety,and the passability during green light signal domain as the objectives of the cost function.The model predictive control framework is applied to solve the problem and obtain the optimal longitudinal velocity curve that completely avoids the red light signal domain.To ensure the vehicle-to-vehicle spacing safety of vehicle fleet,a lateral planning model for lane change operation based on the vehicle spacing is proposed,which applies the Pontryagin minimum principle(PMP)to obtain the optimal comfort velocity curve by employing the acceleration function as the cost function.The problem of longterm future velocity planning in vehicle-to-road collaboration scenario is resolved.(5)The visual trajectory of the vehicle is achieved by MATLAB Automated Driving Toolbox after combing longitudinal and horizontal velocity and the reasons for the separation of vehicle fleet are analyzed.The united velocity conditions of the first group and the second group are extracted,and the fuel consumption of the vehicle fleet is obtained by using the model predictive control-based energy management method.The fuel consumption of the two fleets of vehicles and the working status of the main vehicle components are analyzed and compared.Furthermore,future velocity conditions are demonstrated to be of great importance to efficient energy management strategies.