Data Generation and Preliminary Fault Detection of Power Battery Based on GAN Algorithm

<div class="section abstract"><div class="htmlview paragraph">The safety of power batteries is an important issue that has attracted widespread attention in new energy vehicle technology. In this paper, Generative Adversarial Networks (GAN) are introduced, and the data generation and fault diagnosis of power battery life-cycle data are carried out. GAN is composed of a pair of generators and discriminators, combining signal processing with neural networks, using the discriminator architecture based on Fourier transform and the generator architecture based on wavelet transform, so that the neural network can learn the characteristics of power battery life-cycle data from the perspective of time and frequency domain, and use the good performance of wavelet transform in data denoising and repair to generate high-quality and low-noise data, and use Fourier transform to target the characteristics of periodicity. Identify and distinguish the periodic characteristics and time-frequency domain data characteristics in the generated data and laboratory data. The results show that the GAN architecture adopted in this paper can generate high-quality power battery charge and discharge cycle data, and can observe the location of power battery fault data.</div></div>