Prompted by the global carbon neutralization strategy, the electric vehicle (EV) has become the core carrier of green transformation in the transportation field by virtue of their advantages such as zero exhaust emission and high energy conversion efficiency. However, faults in core components such as IGBT, high-voltage cables, DC bus capacitors, and link capacitors in their driving system can easily cause power transmission interruptions and system overvoltage, threatening driving safety and system reliability. Traditional fault diagnosis methods based on threshold judgment and spectrum analysis have limitations in extracting complex nonlinear fault features and identifying multiple fault modes. Hence, an immediate need exists to advance sophisticated, accurate fault-detection methodologies. In this study, a novel fault diagnosis model for electric vehicle drive system is developed by fusing a convolutional neural network (CNN), a squeeze excitation network (SE), and a bidirectional long short-term memory network (BiLSTM). The test results reveal that compared with the traditional diagnosis methods and other single or simple combined deep learning models, the CNN-SE-BiLSTM model has an accuracy rate of over 99% for ripple current faults in various operating conditions, and still exhibits good diagnostic performance in noisy environments, with stronger anti-interference ability. This integrated model demonstrates remarkable efficacy in diagnosing faults within electric vehicle driving system. It offers an innovative technical approach and theoretical foundation, improving the security and dependability of EV performance.