Fuel Cell Experimental Data Anomaly Detection and Polarization Curve Fitting Method

In recent years, fuel cell vehicles have gained widespread attention due to their environmental benefits. How to quickly predict the lifespan of fuel cells has become a research focus in both academia and industry. The polarization curve, described typically by semi-empirical formula, is an important tool for describing the fundamental characteristics of fuel cells. By studying and analyzing the polarization curves of fuel cells throughout their lifespan, we can gain insights into their performance under different operating conditions, providing valuable insight for the design and optimization of fuel cells. Parameters in semi-empirical equations are typically fitted using raw experimental data, but outlier points can severely impact the results of data analysis. Therefore, the rapid and effective automatic identification and removal of these outliers is a crucial step in the fitting process of polarization curve parameters. This article explores the use of data cleansing methods based on the Isolation Forest and Local Outlier Factor (LOF) algorithms to eliminate these outliers. For both algorithms, thresholds were set to identify and remove the outliers, and then the parameters in the semi-empirical formula for the polarization curve were fitted. For evaluation, we employed the coefficient of determination and root mean square error. The results showed that after removing the outliers, the fitting effect of the polarization curve improved significantly in both evaluation indicators. In addition, we conducted a comparative analysis between the Isolation Forest and LOF algorithms. Overall, the LOF algorithm was found to have higher accuracy and stability in detecting outliers compared to the Isolation Forest algorithm.