The MIMO (Multiple-input multiple-output) systems are ubiquitous in real-world engineering applications [1]. Due to the variety of variables and complex mechanisms in actual engineering, it is very difficult to describe the MIMO systems accurately, and it is inevitable to model and identify parameters of the system in the actual engineering control process. In this paper, an improved parameter estimation algorithm based on VAE (Variational Auto-encoder) and GAN (Generative Adversarial Network) is proposed to solve the problem of parameter identification in MIMO systems. In this method, the input and network architecture of the GAN are improved, the unknown parameters are transformed into network weights, and the iterative update algorithm for parameter identification is derived by using the cross-entropy loss and gradient descent method, and the effectiveness and stability of the algorithm are analyzed.