Research On Hydrogen Fuel Cell Prognostics And Energy Management Strategy

The development of hydrogen energy is of great significance in reducing dependence on fossil fuels,ensuring energy security,and addressing climate change.Therefore,hydrogen energy has become a key direction for China’s energy industry transformation.As one of the key technologies for hydrogen utilization,hydrogen fuel cells have advantages such as zero emissions and high efficiency.They are currently widely used in areas such as automobiles,trains,and ships.However,the large-scale commercial application of hydrogen fuel cell systems still faces the challenge of insufficient durability.As the core of hydrogen fuel cell systems,the membrane electrode assembly(MEA)undergoes irreversible performance degradation during long-term operation,which affects the durability of hydrogen fuel cells.Prognostics can provide information support for the design of control and maintenance strategies for hydrogen fuel cells.In addition,during the service of hydrogen fuel cell systems,reversible performance degradation caused by platinum oxidation and other factors not only reduces the efficiency of hydrogen fuel cells but also transforms into irreversible performance degradation,thereby impacting the service life of hydrogen fuel cells.By establishing a reasonable energy management strategy and optimizing the timing of reversible performance recovery,reversible performance recovery of hydrogen fuel cells can be achieved without affecting the normal operation of the system.Therefore,researching methods for predicting performance degradation and energy management strategies for hydrogen fuel cells is of great significance for enhancing the durability of hydrogen fuel cell systems.Focusing on performance degradation prediction and energy management strategies for hydrogen fuel cells,this thesis conducts systematic research in the following three aspects:1.Under static load conditions,existing performance prediction approaches based on historical voltage data do not consider the impact of operating conditions,resulting in low prediction accuracy.To solve this problem,this thesis studies the effectiveness of operating conditions on hydrogen fuel cell prognosis,and based on this,selects key operating condition parameters to achieve accurate prediction of hydrogen fuel cell performance.Firstly,neural networks are constructed using historical voltage data and operating condition parameter data as inputs respectively,and then predict the future output voltage of the hydrogen fuel cell.Then,the fusion weight is determined based on the prediction error,and the weight of operating conditions parameters in the performance prediction results is analyzed to clarify the correlation between operating conditions and hydrogen fuel cell prognosis.Based on this,the redundant information of the operating conditions parameters is eliminated based on the LASSO algorithm,and the operating conditions parameters highly related to hydrogen fuel cell aging are selected.Based on this,a fuel cell performance prediction method based on the ESN ensemble model is constructed.The experimental results show that the prediction accuracy of fuel cell performance can be improved by selecting operating conditions parameters.2.Under dynamic load conditions,existing fuel cell health indicators cannot fully reflect the health state of fuel cells,making it difficult to predict performance accurately.To solve this problem,this thesis proposes an auto-encoder based method for extracting fuel cell health indicators.Furthermore,a performance prediction method based on auto-encoder and LSTM neural network under dynamic load conditions is proposed.Firstly,by using an auto-encoder to extract deep-level information from fuel cell voltage,a mapping relationship between hydrogen fuel cell voltage and health indicators is established.Then,health indicators are predicted based on LSTM network model,and health indicators are reconstructed into fuel cell voltage prediction results using the information reconstruction ability of auto-encoder.Experimental results show that compared with existing health indicators(including power changes,etc.),the proposed health indicator can retain original voltage information more completely,thereby can reflect the health status of fuel cells more accurately.Compared with existing prognosis approaches(including Attention-GRU,etc.),the proposed prognosis approach can predict future performance changes of fuel cells more accurately.3.Most of the existing energy management strategies ignore reversible performance degradation,which will result in the reduction of hydrogen fuel cell hybrid energy systems efficiency.This thesis integrates reversible performance recovery into energy management strategy to optimize energy flow and timing of reversible performance recovery in hybrid energy systems.Firstly,based on equivalent consumption theory,reversible performance degradation is converted into equivalent hydrogen consumption,and the cost function of hybrid system energy management strategy is updated.To ensure the stability of long-term operation of hybrid system,lithium battery SOC changes are taken as constraints,and performance recovery is introduced into the action space of energy management strategy.Based on this,a deep deterministic policy gradient reinforcement learning algorithm is constructed to optimize the energy distribution of hybrid systems and the execution time of fuel cell reversible recovery procedures.Based on the analysis of experimental results,compared with traditional energy management strategies,the proposed strategy can improve the efficiency of fuel cells,reduce the total cost of system operation,and extend the remaining useful life of fuel cells.