This study develops climate-resilient energy policies for China's Loess Plateau, a region plagued by severe land degradation and economic challenges. It aims to balance ecological restoration with economic development under SSP2-4.5 and SSP5-8.5 climate scenarios, hypothesizing that integrating Artificial Intelligence (AI) and Multi-Criteria Decision Analysis (MCDA) can effectively manage these issues. The goal is to formulate sustainable policies that reduce degradation while promoting growth. Utilizing Sentinel-2 imagery (2015-2022), CMIP6 projections (2025-2050), socioeconomic data (2010-2022), and energy infrastructure details, the methodology involves three phases: predictive modeling via deep learning (CNN for land degradation classification at 92.3% accuracy; LSTM for energy demand forecasting), policy generation using reinforcement learning, and evaluation with a hybrid fuzzy-VIKOR framework. Results feature a land degradation map showing severe issues in central and northern areas, energy demand rises (74% under SSP2-4.5; 90% under SSP5-8.5 by 2050), and five policy scenarios. Scenario 4, ranked highest (Q(t) = 0.12), allocates 30% budget to solar, 20% to wind, and 7200 km(2) to afforestation, yielding 22% degradation reduction, 2.2% annual GDP growth, 18% GHG emissions cut by 2030, and ecosystem recovery (94.10% carbon fixation; 87.59% sand fixation). It supports SDGs 7 and 15.3, enhances social equity via community cooperatives, and aligns with China's 14th Five-Year Plan (15% non-fossil energy by 2025).