Research On Energy Management Strategy For Hybrid Electric Vehicle With Incorporation Of Traffic Information

With the massive consumption of fossil energy and the dramatic increase in demand for electricity,environmental pollution is becoming increasingly serious.In order to prevent the deterioration of global environmental problems,all industries are facing the challenge of green development.The transport sector accounts for about 31% of global fuel consumption,with road transport accounting for the largest share.As one of the major industries consuming energy and emitting pollution,the automotive industry has become a key area for green development.By adopting new energy vehicles,promoting energy-saving and emission reduction technologies,the automobile industry can reduce its negative impact on the environment and make contributions to achieving sustainable development.Hybrid electric vehicles have proven to be an effective way to solve environmental and energy problems.Compared with electric vehicles,hybrid vehicles can escape the limitations of battery technology,improve range with lower energy consumption and emissions,achieve the advantages of multiple power sources and improve the overall performance of the vehicle.Hybrid electric vehicles have both fuel and electric power systems,and realize the conversion of fuel and electric power through energy management strategies,so as to achieve higher fuel efficiency and lower emissions.Energy management strategies are often affected by driving conditions,and the transient characteristics of traffic information greatly affect the driving conditions of vehicles.Therefore,if environmental information can be incorporated into the strategy,fuel economy of hybrid electric vehicles can be greatly improved.Most of the energy management strategies that incorporate environmental information are based on the vehicle itself,and the incorporated environmental information is often obtained directly from the vehicle network,but the vehicle network technology is not yet widespread.Therefore,this paper starts from the vehicle’s own sensors to extract useful information from the raw image data and fuse it into the energy management strategy.In addition,the incorporated information should be representative,general and typical,it should not be mixed with the influence of other error factors.Therefore,it is more necessary to start from the most typical data of vehicle driving,in this paper the depth and relative speed of the vehicle ahead are selected as the typical information of the environment to be incorporated into the energy management strategy.The main research contents are as follows:(1)The basic configuration of the hybrid vehicle is analyzed,and the object of this paper is established as a hybrid-connected hybrid vehicle.Eight operating modes of the hybrid vehicle are analyzed as well as the modeling of key components such as engine,motor,battery and transmission.In order to be able to correspond to the scenery information and driving information during the hardware-in-the-loop test,this paper models the scenery based on the Unreal engine to realize the correspondence with the world.(2)A monocular depth-velocity estimation algorithm applicable to energy management strategy is proposed to merge two major tasks,depth estimation and velocity estimation,with energy management strategy.In this part,firstly,the principles and transformation relations of three coordinate systems,namely world coordinate system,camera coordinate system and image pixel coordinate system,are introduced to provide the basis of transformation relations for geometric cues.Considering that geometric cues,depth feature cues,optical flow feature cues,and multidimensional features need to be fused,it is proposed to use the attention mechanism to combine these features in a reasonable way so that the model can focus more on the features of interest in training.(3)A reward function parameter matching method for energy management strategies in a reinforcement learning framework is designed.Usually,researchers construct reward functions subjectively and empirically.Moreover,in most of the studies,the corresponding hyperparameters are often obtained using manual tuning methods,which directly leads to the setting of the reward function will contain too many human interference factors.In this paper,it is proposed to reverse the behavior of the smart body by using the weight coefficients of the calibrated battery and engine dual smart body to reverse the obtained weight coefficients.After that,the obtained weights are put into the forward reinforcement learning algorithm and the results show that better fuel saving can be obtained.The algorithm consists of four main parts:the first part represents the optimal state of the engine and the battery,which is the optimal operating point for the engine and the maintenance of a reasonably stable battery SOC value,i.e.,a low internal resistance and a stable electric potential for the battery;the second part is the algorithmic framework for inverse reinforcement learning,which defines maximum entropy inverse reinforcement learning;the third part represents the reinforcement learning environment,into which the parameters obtained by inversion are input;and the fourth part is the DQN algorithm for reinforcement learning,which aims to validate the algorithm with forward reinforcement learning.(4)A method is proposed to fuse the monocular depth velocity estimation algorithm with the basic algorithm of reinforcement learning,in which hyperparameters for inverse reinforcement learning training are used.Based on the basic theory of reinforcement learning,a reinforcement learning-based hybrid vehicle energy management task is constructed,and then the monocular depth velocity estimation results are fed into the constructed RBF torque prediction network to predict the demand torque at the next moment.Finally,the predicted torque and depth velocity data are input to the reinforcement learning network in the form of state values to obtain the basic model of reinforcement learning with demand torque.(5)The hardware-in-the-loop research is carried out for the energy management strategy of fusing environmental information in this paper.Based on the laboratory conditions,a hardware-in-the-loop experimental bench is built,and the experimental system consists of a hybrid model,a driver operating system,a virtual scenario system,a sensor system,a NI realtime system,and a vehicle control unit.From the simulation experiments and hardware-in-the-loop experimental results,it can be seen that the algorithms proposed in this paper both have good accuracy and fuel savings.The research in this paper has positive implications for promoting energy management strategies incorporating traffic environment information,parameter setting of reward functions for energy management strategies with inverse reinforcement learning,and monocular distance-velocity estimation from data cues,which can help further improve the fuel-saving performance of hybrid vehicles.