Artificial Intelligence-Driven Carbon-Aware Energy Management for Sustainable Smart Grids

Abstract

Intelligent energy management techniques that can work together to enhance the grid's efficiency and environmental footprint are becoming increasingly important as intermittent renewable energy sources and the need for electricity continue to rise. The traditional methods focus almost exclusively on minimizing the operating expenses and balancing supply and demand and do not consider the variations in carbon intensity or the uncertainty of renewable energy. This paper presents an artificial intelligence-based carbon-aware energy management system, which includes load forecasting, renewable energy forecasting, carbon-intensity estimation, battery optimization, and adaptive demand-response scheduling. The framework shows how the grid conditions will change and how electricity will be used flexibly in times of high renewable energy and low carbon intensity. Experimental results showed that CarbonSenseNet achieved a forecasting accuracy of 98.74%, an MAE of 0.041, an RMSE of 0.068, and an $\mathrm{R}^{2}$ of 0.994 when compared to the other models, including ANN, Random Forest, XGBoost, LSTM, GRU, Transformer, GNN, and DRL. It also accomplished 95.82% renewable-energy utilization, 46.74% carbon-emission reduction, 33.48% cost savings, and 99.21% grid reliability. All the proposed modules have been confirmed by ablation and statistical analyses. The results indicate that CarbonSenseNet effectively operates a smart grid in a reliable, cost-effective, and environmentally friendly manner.