Renewable Energy Forecasting and Scheduling with PV Generation and EV Integration

Abstract

The integration of renewable energy sources such as solar and wind into microgrids remains challenging due to the intermittent and unpredictable nature of generation, which affects reliability, cost efficiency, and operational performance. Conventional energy management systems rely on traditional forecasting and scheduling approaches, resulting in inefficient coordination and increased operating costs, while deep learning models such as LSTM, BiLSTM, and standalone Temporal Convolutional Networks (TCN) struggle to capture complex temporal dependencies and dynamic environmental variations effectively. To address these limitations, this study proposes a two-stage microgrid energy management framework that integrates advanced forecasting with intelligent optimization. In the first stage, a high-accuracy solar photovoltaic (PV) forecasting model is developed using an attention-enhanced Temporal Convolutional Network (TCN- Attention) to improve long-term temporal representation and weather-aware learning. In the second stage, a Mixed-Integer Linear Programming (MILP) optimizer coordinates energy generation, storage, and grid interactions in a cost-effective and reliable manner. The framework is trained and validated using the Kaggle Renewable Energy Microgrid Dataset containing time-series variables such as solar irradiance, temperature, wind speed, humidity, and load demand. Experimental results demonstrate that the proposed TCN-Attention model outperforms classical TCN, BiLSTM, and LSTM models, achieving MAE of 7.2864, MSE of 101.8693, RMSE of 10.093, and an R² of 0.923. Furthermore, the MILP-based strategy yields superior economic performance with a net profit of £8.34, cost savings of 131.6%, and a 25% reduction in grid imports, providing a scalable and sustainable solution for future microgrid energy management systems.