Research On Intelligent Energy Management Strategy Of Plug-in Hybrid Electric Vehicle Considering Traffic Information

With the increase of car ownership,environmental pollution and traffic pressure gradually intensify.The new energy vehicle has become an effective solution to the environmental crisis because of its low emission advantage.Plug-in Hybrid Electric Vehicle(plug-in hybrid electric vehicle,PHEV),as a representative model of new energy vehicles,can solve the mileage anxiety and emission problems of other models by cooperating with its large-capacity power battery and low-displacement engine.However,with the increasing complexity of road conditions,the economic performance of PHEV will be affected by traffic factors.In recent years,intelligent network connection technology has been widely used in the automobile industry,so it’s very important to explore the intelligent optimization control of PHEV multi-power sources under the network connection technology for improving the fuel economy.This paper focuses on the velocity prediction and energy management of PHEV based on traffic information.In this paper,we first analyze the working mode of PHEV,and build a physical model of the vehicle in MATLAB/Simulink based on the mathematical models of core components such as engine,electric motor,power battery and the longitudinal dynamics model of the whole vehicle,and construct a simulation and verification platform for the control strategy.Then,simple and complex traffic flow models are established in VIS SIM to reflect different types of road conditions.Under the two models,the network information of the target vehicle speed,the preceding vehicle speed and the distance between vehicles are obtained respectively.In order to investigate the influence of vehicle network information on future short-time speed prediction,this paper designs a cascaded neural network velocity predictor based on the interaction relationship between several types of information.Simulation results show that under simple road conditions,the prediction model has high prediction accuracy.Under complex road conditions,it’s more adaptable to random changes of traffic factors and has excellent network generalization ability.Based on the above contents,this paper further establishes the model predictive control strategy based on deep reinforcement learning method as a solver for rolling time-domain optimization.This article expounds the basic principles of deep reinforcement learning and presents the global optimization strategy by using this algorithm.To address the advantages of deep reinforcement learning and the shortcomings of model predictive control,the two methods are combined to achieve optimal torque allocation in the short-time prediction time domain.The proposed strategy,global optimization strategy and ECMS strategy are simulated under different standard driving cycles and actual driving conditions to compare and analyze the range of SOC variation and fuel economy of the three strategies,and to verify the effectiveness of the proposed strategy.On this basis,the optimization effects of the control strategies with different prediction methods and different vehicle network information are analyzed respectively.The results show that the fuel economy of the designed strategy is significantly better than that of the conventional method under the consideration of various vehicle networking information.