Fault Diagnosis of PEMFC Based on Fast EIS Measurement and Optimized Random Forest Algorithm

Electrochemical impedance spectroscopy (EIS) is often used for fault diagnosis as an important parameter to characterize the state of fuel cells. However, online diagnosis requires high real-time performance and usually can only measure single-frequency or dual-frequency impedance. Too few diagnostic features make it difficult for traditional fault diagnosis methods based on EIS to ensure high accuracy. Therefore, this paper proposes a fault diagnosis method based on fast EIS measurement and an optimized random forest algorithm. Firstly, using a multi-sine excitation signal to realize the simultaneous measurement of multi-frequency impedance, provides more health status information in a single measurement. To solve the problem of large signal peaks caused by the superimposed signals, the phase is optimized by the genetic algorithm, which reduces the crest factor of the excitation signal. Then, multi-frequency impedance is used as a training feature for the random forest (RF) algorithm to realize the diagnosis of flooding and drying faults. The particle swarm optimization (PSO) algorithm is used to optimize the algorithm's hyperparameters to improve the identification accuracy. Finally, experimental verification is carried out based on the fault dataset of an automotive fuel cell, and the results show that the accuracy of the proposed algorithm can reach 99%, which is better than other common methods.