Abstract: Aiming at the problem that the internal short circuit fault characteristics of battery are not obvious, which makes it difficult to detect faults, a fault feature extraction method based on the correlation coefficient of voltage was proposed. Firstly, the improved recursive correlation coefficient between the voltages of adjacent cells in the battery pack is calculated; Then, the obtained correlation coefficient signals are decomposed into multidimensional feature indicators using an empirical modal decomposition algorithm, and these indicators are feature-dimensioned down by kernel principal component analysis to obtain the most representative fault features; Finally, the extracted components are modelled by using a support vector machine to detect and identify battery faults. In the experiment, an actual battery test platform was used to generate fault dataset. The experimental results show that the proposed method can achieve accurate detection of internal short circuit fault with a fault detection rate of 86.0465%.