Application research of multi-agent GLM-4 based on generative adversarial network (GAN) dynamic denoising and transfer reinforcement learning