Bayesian optimization method and system containing generative adversarial network

The invention discloses a Bayesian optimization method containing an adversarial generative network, and the method comprises the steps: carrying out the knowledge extraction from a real sample and a prediction sample, and carrying out the secondary judgment of a current model prediction result, thereby avoiding the reduction of prediction capability caused by the optimization direction deviation of an agent model in a high-dimensional expensive optimization problem. According to the basic data and the multi-target agent auxiliary model, obtaining a group of Pareto solution sets which are converged to a real Pareto leading edge as much as possible and are uniformly distributed; wherein the construction process of the Bayesian multi-objective optimization model containing the generative adversarial network comprises the step of using two basic algorithms as basic optimizers. And the historical data is marked and selected. Meanwhile, a new convergence criterion of the Lp normal form and the maximum and minimum