Microgrid short-term electrical load forecasting using machine learning models

Predicting electrical load is crucial for microgrid energy management. Short-term load forecasting (STLF) helps in optimizing energy management and load balancing within microgrids. It enables microgrid operators to balance energy supply and demand, utilize renewable energy sources and energy storage systems efficiently, and reduce energy costs. In this paper, two machine learning-based models for STLF are proposed. The first is an Artificial Neural Network (ANN) model with a single hidden layer and five features as inputs. The second is a Long Short-Term Memory (LSTM) model using only load data as time series input to forecast the next 24 hours load. The dataset used for training contains real data that comes from London households that participated in the UK Power Networks Led Low Carbon London project. The models were evaluated using four metrics: Root Mean Square Error (RMSE), Mean Square Error (MSE), Mean Absolute Error (MAE), and Mean Absolute Percentage Error (MAPE). The simulation results indicated that the proposed LSTM model demonstrated better performance compared to the ANN model. It achieved a lower MAPE of 7.52% and a smaller RMSE of 0.3843 kW. In comparison, the ANN model had a higher MAPE of 8.34% and a larger RMSE of 0.4454 kW. Unlike the LSTM models in the literature that use multiple input variables, such as time and weather variables, the LSTM model presented here only uses historical consumption data to predict the next 24 hours' consumption accurately. Additionally, the results demonstrate the LSTM model's ability to accurately predict the peak load value.