Publication Details
Abstract
The widespread growth of distributed renewable energy (DRE), mainly solar PV and wind generation, has completely changed the way electrical networks are operated. While the traditional load balancing algorithms are predominantly used for dispatchable generation, they are not adequate to deal with the stochastic and intermittent nature of renewables on a large scale. This research proposes a complete solution for an AI-based smart grid load balancing system, which implements the synergic control of heterogeneous renewable energy resources, energy storage systems (ESSs) and demand-response (DR) programs, in a single intelligent control architecture. It uses a hierarchical multi-agent reinforcement learning (MARL) algorithm coupled with long short-term memory (LSTM)-based forecasting for solar irradiance, wind speed and load demand prediction, with a forecast mean absolute percentage error (MAPE) of 2.31% for a 24-hour horizon. Compared to conventional proportional-integral (PI) controllers, the deep reinforcement learning (DRL) controller achieves a 34.7% reduction in peak-to-average ratio (PAR), 28.6% reduction in renewable curtailment losses, and 22.4% reduction in operational costs. The five different renewable penetration scenarios (from 20% to 80% RES share) of the IEEE 118-bus test system demonstrate the scalability and robustness of the algorithm under various meteorological and demand conditions. In addition, the paper analyzes grid stability indices such as frequency nadir, voltage stability margin and N-1 security compliance, showing that the AI-based framework keeps all the indices in the range of operational limits of the grid even at 80% renewable penetration. Moreover, the proposed AI framework is compared with five world-class load balancing solutions such as model predictive control (MPC), fuzzy logic control (FLC), single-agent deep Q-networks (DQN), and graph neural networks-based approaches. Results show better performance in terms of accuracy, computational efficiency, grid stability indices and techno-economic cost metrics. The research provides a novel technically sound and scalable artificial intelligence based solution which can help move renewable energy into the modern smart grid infrastructures and can be directly applied to utility companies who are aiming to achieve the decarbonization goals, under limited capital spender budgets.