Research On Approach And Application Of Short-term Power Demand Prediction In Fuel Cell Vehicles

In the context of severe environmental challenges and sustainable energy development,hydrogen energy becomes increasingly important.Proton exchange membrane fuel cell(after referred to as fuel cell)is an energy conversion device with high efficiency and zero emission,which plays an important role in the hydrogen economy.Especially in the field of transportation,the vehicles equipped with fuel cell power systems will play a long-term positive role in the low-carbon energy transition,and help achieve the strategic goals of "carbon peak" and "carbon neutral".However,fuel cell is still facing many key technical problems in vehicular power system,especially in full-power architecture.For example,due to the slow dynamic response caused by the limited internal mass transport,the output power of fuel cell is compromised under the rapid load changing condition,and the phenomenon of fuel starvation is easy to occur,which affects the lifespan of fuel cell.Due to the high real-time requirements,multiple optimization objectives and high complexity of power system energy flow management,it is difficult to achieve the optimal or nearly-optimal fuel economy and maintain the stable state of charge(SOC)of small-capacity battery at the same time considering the limited dynamic performance of fuel cell.For supporting the large-scale commercial operating demonstration of the fuel cell vehicle,this thesis focuses on the major technical problems of poor dynamic performance and complex energy management in the application of fuel cells under full-power powertrain architecture.The application of the short-term prediction method of vehiclular power demand in the control of fuel cell system and the energy flow control of powertrain is investigated,so as to achieve the research objectives of enhancing the mass transferring quality inside the fuel cell to improve the dynamic performance,improving the fuel economy of the powertrain and improving the battery SOC sustaining ability.The involved research contents are listed as follows:(1)The short-term prediction method of power demand of fuel cell vechcle was studied.Based on the short-term change trend of time-series data,the input and output variables were extracted according to the linear regression relationship,and the regression prediction model based on machine learning algorithm was then established.According to the recursive conception,a recursive prediction model of time series based on recursive least-square support vector machine was established by using sliding window analysis and phase space reconstruction theory.By using the regression prediction model and the recursive prediction model,the time-phase mismatch phenomenon that the predicted data curve for the future instants deviates backward along the time axis from the real data curve was reproduced,and the relationship between the predicted time range and the time-phase mismatching degree was analyzed.The mechanism of time-phase mismatch phenomenon appeared in the results of the regression prediction model was clarified.Based on this,an iterative regression prediction model of power demand based on iterative learning framework was proposed.Through the analysis of a case study,the prediction performance of the proposed prediction model was compared with that of the regression prediction model and the recursive prediction model,and the effectiveness of the proposed prediction model was proved.The prediction effects of different machine learning algorithms as the learning cores of the iterative learning framework were compared.The influence of the parameters of iterative learning framework(iteration step size and iteration number)on the prediction error of the iterative regression prediction model was analyzed,and the prediction generalization ability of the iterative regression prediction model under multiple driving cycle was verified.(2)The improvement method of fuel cell dynamic performance based on power demand prediction was studied.The feasibility of controlling the fuel cell subsystem components to change multiple operating parameters for improving dynamic performance in real-time manner was analyzed from the perspectives of control response speed and parasitic power.The hydrogen supply pressure was selected as the control variable to improve dynamic performance of fuel cell in real-time.The sensitivity of dynamic performance of fuel cell to hydrogen supply pressure was studied experimentally,and a three-dimensional transient model of multi-physical field of single-channel fuel cell was established to analyze the mechanism of hydrogen pressure on the internal mass transport of fuel cell,so as to provide a reasonable explanation for the experimental phenomenon from the perspective of mass transfer.The influence of hydrogen supply pressure on hydrogen utilization rate of the system was analyzed,and the control requirement of self-adaptively switching hydrogen supply pressure according to the power load was proposed.A variable-pressure hydrogen supply system was then designed,and an intelligent switching control strategy of hydrogen supply pressure was developed by combining the vehicular power demand prediction method.The validity of the proposed method was verified via experiments.The potential influence of anodic pressure fluctuation produced by the proposed system on fuel cell durability was also discussed.The feasibility of reducing the pressure difference between anode and cathode at the moment of hydrogen pressure switching via the pressure coordinated control of cathode and anode was demonstrated through simulation.(3)The energy management strategy of the full-power fuel cell vehicle based on power demand prediction was studied.The vehicle powertrain was modeled,the basic optimization problem of equivalent consumption minimization strategy(ECMS)was deduced,and the influencing principle of equivalent factor of ECMS on battery SOC sustaining was analyzed.The performance of battery SOC sustaining and fuel economy of the adaptive ECMS based on feedback controller principle were analyzed through simulation under different battery initial SOCs,initial equivalent factors,calibration parameters of strategies and driving cycles.In order to reduce the calibration parameters,improve the adaptability under different driving cycles,and ensure the approximately optimal fuel economy and battery SOC sustaining effect,an optimization-oriented adaptive ECMS based on the vehicular power demand prediction was proposed.The energy management effect of the proposed strategy was simulated under a specific driving condition.The fluctuation ratio of battery SOC trajectory and hydrogen consumption of different adaptive ECMS strategies under different initial SOC conditions were compared.It is proved that the proposed strategy has the advantages of fuel economy and battery SOC sustaining performance.The average time of the proposed optimization-oriented predictive adaptive ECMS strategy to complete one round of equivalent factor optimization and update was counted,which proves the feasibility of the real-time application of the proposed strategy.By comparing the simulation results of the proposed strategy using the real and predicted short-term power demand data respectively under a specific driving condition,the robustness of the proposed strategy against the interference of power demand prediction error was demonstrated.Finally,the adaptability of the proposed strategy under multiple driving conditions was verified and analysed.