The invention discloses a dense oil reservoir fracturing horizontal well optimization method based on a self-adaptive agent model. The method includes the following steps that S1, an optimization parameter x is determined, a target function J(x) is established, and an initial design space D is determined; S2, an initialization parameter k is 1, a sample X0 is obtained in the space D, and sample data X0 is processed to obtain a sample; S3, the variable X0 is substituted into an oil reservoir numerical simulator, an initial variable objective function value J(X0) is obtained, (X0,J(X0)) is stored in a sample point database, and according to the corresponding relationship between X0 and J(X0), a kth gaussian process agent model is established; S4, the self-adaptive process of the kth gaussianprocess agent model is carried out through point addition and an important design space; S5, iteration is stopped when a convergence criterion is satisfied, and otherwise step 6 is executed; S6, k isequal to k+1, the important design space and the sample point database are updated, and then step 4 is repeated. The integrated optimization design of horizontal well spacing and fracturing is adopted, so that accurate optimal well spacing and fracturing parameters are obtained, and the method has great significance on the development of dense oil reservoirs.