Research On Online Multi-objective Energy Management Strategy Based On BP Neural Network For Fuel Cell Vehicle

Fuel cell hybrid electric vehicle,which takes fuel cell as one of the power sources,not only has the characteristics of high fuel efficiency,zero emission and low noise,but also represents the development trend of using advanced technology,new energy and pursuing environmental protection in the future.Energy management is a key technology of fuel cell hybrid electric vehicle.Its core lies in ensuring the power demand of the whole vehicle and optimizing the working performance of each energy unit by rationally distributing the output power of fuel cell and auxiliary energy,thus improving the economy of the vehicle.In this paper,the fuel cell hybrid electric vehicle is taken as the research object,and the online multi-objective energy management strategy is proposed,which mainly includes the following contents:(1)The modeling of fuel cell hybrid electric vehicle.Firstly,the vehicle dynamics model is established.Then,the mathematical model and decline model of lithium battery are established.Finally,the thermal model of fuel cell system including fuel cell stack,radiator,humidifier,reservoir and condenser is established.In addition,the decline model of the fuel cell is established and it lays a foundation for solving the energy management problem later.(2)The research on energy management strategy based on Pontryagin’s Minimum Principle.Firstly,the mathematical model of energy management problem is established,and the constraint on the rate of change of output power of fuel cell is considered to improve the durability of fuel cell.Then,on the basis of the model built in the section II,the single-objective energy management strategy and the multi-objective energy management strategy are proposed,respectively.Finally,the proposed strategies are simulated to verify their effectiveness.(3)The establishment of driving cycle identification model based on simulated annealing and K-means clustering.Firstly,25 representative driving cycles are taken as the sample driving cycles,and 8 typical characteristic parameters are selected to describe the driving cycles.Then the principal component analysis method is used to reduce the dimension of the characteristic parameters,and K-means clustering algorithm based on simulated annealing is used to cluster the sample driving cycles into four categories.Finally,on this basis,through the analysis of historical driving data of comprehensive driving cycle,the driving cycle is identified based on K-means clustering algorithm.(4)The research on on-line multi-objective energy management strategy based on BP neural network.Firstly,the relevant parameters and the structure of the neural network are determined.Then,the simulation data obtained by PMP-based multi-objective optimization energy management strategy is classified according to the results of driving cycle identification.Four neural network sub-models are obtained after training the classified data,and then the neural network sub-models are combined into a multi-neural network model.On this basis,an on-line multi-objective energy management strategy based on driving cycle recognition and BP neural network is established.Finally,the software simulation and hardware-in-the-loop simulation are carried out,respectively.The results of hardware-in-the-loop simulation are basically consistent with those of software simulation,which provides sufficient verification for the proposed strategy can run on real vehicles.