The accurate evaluation of the electricity output of geothermal power plants requires effective coupling between the geothermal reservoir and power plant. Existing coupling models integrate numerical simulation models of the reservoir and power plant; however, they are computationally expensive for electricity prediction (forward modeling) and integrated reservoir-power plant optimization. Therefore, this study aimed to enhance the efficiency of the coupled reservoir-power plant model for forward modeling and optimization by replacing simulation forward models with deep-learning-based surrogate models. Two independent surrogate models of the reservoir and power plant were trained and assembled into one coupled forward model. Moreover, a multiobjective optimizer was integrated with the coupled forward model to optimize reservoir operations and power plant designs to achieve the highest electricity output or the best economic outcome. Surrogate models for the reservoir and power plant accurately predicted the geothermal production temperature and electricity output while approximately achieving speedups of 1.23×105  and 1.77×105 times over those of the corresponding simulation models, respectively. Furthermore, optimization using our surrogate-based coupled model was 1.31 × 106 times faster than that using the simulation-based coupled models. Optimization results revealed that low injection temperature, large well distance, and stable reservoir injection and production rates contributed to better power plant performance. High design geothermal temperature, mass flow rate, and ambient temperature favored electricity generation, particularly in power plants located in hot regions. Our work remarkably accelerates the feasibility assessment and decision-making procedures for geothermal reservoirs and power plants.