ARCHIVES

Integration of renewable energy in Gujarat's continuously expanding microgrid infrastructure presents significant challenges for real-time battery energy storage optimization. This paper examines feasibility for least-cost energy optimization through a novel hybrid model combining the Frequency Enhanced Decomposed Transformer FED former with Soft Actor-Critic reinforcement learning [FED former-SAC] across four major Gujarat cities: Ahmedabad, Ankleshwar, Gandhinagar, and Surat. The study defines comprehensive city-specific datasets incorporating local solar irradiance patterns, industrial load profiles, and Gujarat Electricity Regulatory Commission tariffs. This multi-city approach enables both single-city optimization and cross-city transfer learning. Experimental results demonstrate 25.8% average cost reduction compared to rule-based baselines, with 94.6% renewable energy utilization and 0.73 battery cycles per day. Transfer learning experiments show that pre-training on Ahmedabad and Surat enables 18% faster convergence and 3.5% better performance when fine-tuning on Gandhinagar. The proposed architecture achieves 41ms average inference time, making it suitable for edge deployment. Cross-city generalization analysis reveals that industrial mix (50%-80%) significantly impacts optimization strategies, with implications for Gujarat's 500 GW renewable energy target by 2030.