Optimizing Power Dispatch and Market Trading Under Renewable Energy Uncertainty Using OPRTDPG Deep Reinforcement Learning

Abstract

The integration of renewable energy sources into modern power systems introduces uncertainty that challenges efficient dispatch and market trading. This research presents a targeted optimization approach using Deep Reinforcement Learning (DRL) for real-time power system dispatch and electricity market trading operations. A proposed method employs the Optimal Power Reinforced Twin Deterministic Policy Gradient (OPRTDPG) algorithm, which combines power system-specific reinforcement mechanisms with deterministic policy gradients for precise and stable decision making under renewable generation and market volatility. The methodology incorporates direct control of Thermostatically Controlled Loads (TCLs) and indirect control of price-responsive demands, enabling flexible resource management. The algorithm was trained and evaluated using the Power System Dispatch and Market Trading dataset from Kaggle, containing 4,876 fifteen-minute interval records of system states, generation, storage, loads, market prices, and reward metrics. Data preprocessing applied Min-Max normalization to ensure stable learning. The algorithm was implemented in Python 3.11 using NumPy and PyTorch within a custom power system simulation environment, capturing generation, storage, load dynamics, and market behavior, without requiring external real-time platforms. Performance comparison with the baseline Deep Deterministic Policy Gradient (DDPG) method, OPRTDPG reduces market price volatility by 10% ($50/MWh-$45/MWh), improves energy conversion efficiency by 15% (65%-74.75%), and lowers daily operating cost by 12% ($100,000-$88,000). These results demonstrate the algorithm’s capacity to enhance system reliability, maximize renewable utilization, and minimize operational cost. The framework provides a scalable, simulation-tested solution for dynamic power system dispatch and market trading, highlighting the practical applicability of DRL in renewable-rich electricity networks.