Abstract: In recent years, intelligent methods represented by multi-source data fusion, machine learning and digital twins have been increasingly applied to geothermal exploration, drilling and completion, heat extraction optimization and operational regulation, providing technical support for the transition of middle-deep geothermal development from experience-based design and offline evaluation to data sensing, dynamic prediction and collaborative decision-making. To clarify the applicable conditions, major bottlenecks and development trends of intelligent exploitation technologies for middle-deep geothermal energy, and to provide references for efficient, low-risk and sustainable development, research progress in middle-deep geothermal development in China and abroad was systematically analyzed through data synthesis, case summarization and technical comparison. Progress in intelligent exploration and target-area optimization, intelligent drilling and development optimization, heat extraction by pumping and reinjection, closed-loop heat extraction, resource utilization of existing underground engineering carriers, and intelligent operational regulation and sustainable optimization was reviewed. The applicable conditions, engineering constraints and intelligent adaptation characteristics of different heat extraction methods were compared. The results show that multi-source data fusion, machine learning and deep learning can improve the integrated utilization of geological, geophysical, geochemical, remote-sensing, well-logging and historical well data, thereby supporting the identification of hidden geothermal reservoirs and target-area optimization. Real-time drilling data sensing, rate of penetration prediction, drilling condition identification, surrogate models, data assimilation and physics-constrained models can be used for drilling parameter optimization, geothermal reservoir parameter prediction and rapid comparison of heat extraction schemes. Heat extraction by pumping and reinjection is suitable for hydrothermal reservoirs with relatively clear water productivity, reinjection conditions and inter-well connectivity, but its long-term operation is constrained by reinjection clogging, thermal breakthrough, hydro chemical changes and production-reinjection balance. Closed-loop heat extraction is less dependent on natural geothermal fluids and reinjection capacity, and is more suitable for formations with limited reinjection conditions, high groundwater protection requirements or dominant conductive heat transfer in rock masses. However, its performance is controlled by well depth, geothermal gradient, rock thermal properties, well configuration, circulation parameters and wellbore thermal resistance. Enhanced geothermal systems, CO₂-based heat extraction, integration of geothermal energy with carbon capture, utilization and storage, and underground thermal energy storage and recovery further extend the development scope of middle-deep geothermal energy. Existing underground engineering carriers, including abandoned oil and gas wells, abandoned mines, mine roadways and goaf areas, can be utilized according to geothermal reservoir conditions, spatial integrity, reconstruction cost, operational safety and surface heat demand. During operational regulation, geothermal reservoir pressure and temperature fields, reinjection capacity, thermal breakthrough risk, wellbore heat transfer, equipment status and user-side heat load should be comprehensively considered, so as to promote the application of intelligent prediction and scheduling, predictive maintenance, closed-loop well control optimization and digital twins in the collaborative control of the wellbore-reservoir-surface system. The key to intelligent exploitation of middle-deep geothermal energy lies in establishing a continuous feedback mechanism across resource identification, drilling and completion, heat extraction scheme optimization and operational regulation. Further efforts are needed in whole-process data standardization, integration of physical models and data-driven methods, long-term field monitoring and engineering verification.