Machine Learning-Based Renewable Energy Forecasting and Priority Load Management for Smart Energy Systems

Reliable electricity is not just a convenience; it is a basic human necessity. However, a huge number of people in developing countries face daily power cuts that disrupt their everyday lives. When load demand exceeds the available energy supply, high-priority services such as emergency hospitals (particularly ICUs, ventilators, and surgical units), emergency services, street lighting, railways, airport operations, and critical household appliances such as refrigerators, water pumps, and heating or cooling systems may fail to function properly. This challenge is further worsened by the heavy reliance on costly and environmentally damaging fossil fuels. Although solar and wind energy are cleaner and more affordable, managing them is still a challenge because they do not produce a steady supply of power—their output rises and falls with the weather. This paper introduces a smart energy management system that uses a Long Short-Term Memory (LSTM) model to predict renewable energy output and an XGBoost model to predict electricity consumption by the load. The forecasting models were developed and tested using four real-world datasets that capture solar generation, wind generation, and the electricity needs of both high-priority and low-priority loads. Using these predicted values, a rule-based control system was designed and simulated in MATLAB/Simulink to efficiently manage the distribution of power. The system classifies electrical loads into two groups: high-priority and low-priority loads, and guarantees that high-priority loads are always supplied with power before others. The backup generator kicks in automatically only when renewable power generation is insufficient to supply high-priority loads, which helps keep operational costs low and reduces reliance on non-renewable energy sources. The proposed forecasting models have shown strong and consistent performance on the four datasets. For solar generation forecasting, the LSTM model achieved an R² of 0.9577, MSE of 29.72, RMSE of 5.45, and MAE of 2.75, improving over baseline values of R²=0.94, MSE=45.00, RMSE=20.1, and MAE=15.2. For wind generation forecasting, the LSTM model achieved an R² of 0.9537, MSE of 143.20, RMSE of 11.97, and MAE of 7.83, compared to baseline values of R²=0.94, MSE=180.00, RMSE=20.1, and MAE=15.2. The XGBoost model's results for high priority load demand forecasting were R² =0.9007, MSE=10.32, RMSE=3.21 and MAE=2.61, compared to the baseline R²=0.85, MSE=56.72, RMSE=7.48 and MAE=5.90. For low priority load demand forecasting, the XGBoost model achieved an R² of 0.9004, MSE of 126.78, RMSE of 11.26, and MAE of 9.13, compared to baseline values of R²=0.86, MSE=130.00, RMSE=13.00, and MAE=13.00. The simulation outcomes further validate that the proposed system reliably delivers an uninterrupted power supply to high-priority loads under varying operating conditions, whether renewable generation is high, limited, or completely insufficient, while considerably reducing the unnecessary activation of backup generators