Prediction Modeling and Optimization of Power Grid Energy Storage System Based on Machine Learning

Abstract

The rise of renewable energy, like solar panels and wind turbines, stems from price hikes and emissions rules. Weather affects energy production from solar farms. The power operations like pneumatic transports. Challenges with grid capacity arise from renewables’ growth. The power market shifts from government control to deregulation. Regulations affect energy expenses, efficiency, and environmental impact benefits. To address this issue, the authors propose a technical solution that allows the EPGCCS system combines Deep Reinforcement Learning (DRL) with solar photovoltaic (PV), battery storage, and grid management to enhance pneumatic transport plant efficiency. It involves designing the plant, simulating processes, optimizing mathematically, developing control systems, initializing renewable energy sources, and optimizing scheduling through DRL. The plant integrates PV modules, inverters, a battery system, grid connection, and an air transport system for energy-efficient material handling. Simulation involves random arrivals and airflow calculations using the Hudson formula. The PV system uses performance mappings and cloud forecasting, while battery management employs a flexible state-of-charge estimation method. A model combining Multi-Agent Deep Deterministic Policy Gradient (MADDPG) and Model Predictive Control (MPC) was created to optimize costs under set constraints. Enhancements like Priority Experience Replay (PER), adaptive noise exploration, and improved Adadelta optimization methods were added to enhance convergence rate, scheduling, and computational efficiency for automated day-ahead scheduling and real-time operational efficacy.