To enhance the efficiency of the surrogate-assisted evolutionary algorithm (SAEA) for solving high-dimensional expensive optimization problems with multiple local optima and multivariate coupling, an SAEA incorporating two machine learning methods is proposed, named the adaptive search strategy selection based on the guide of two learning mechanisms (ASSS-GTLM). The proposal is based on three heterogeneous surrogate modeling techniques integrating four search strategies with different merit search orientations. In ASSS-GTLM, reinforcement learning is adopted to adaptively adjust the selection of the search strategies according to the online feedback information of the optimal solution during optimization so as to achieve a balance between global exploration and local exploitation in the solution space. In addition, an Autoencoder-based subspace surrogate-assisted search strategy is designed to reconstruct the original space features to incrementally search the space for potentially optimal sample information. Experimental results show that the ASSS-GTLM algorithm performs significantly better than others in high-dimensional benchmark problems up to 300 dimensions and has stronger convergence performance and optimization efficiency.