Forecasting Production and Demand for PV-Integrated System With Lstm and Comparison With Ann Model
Abstract
A growing global population demands more efficient and sustainable energy systems. Accurately modeling their production has become essential with the increasing integration of renewable energy sources. Equally important is forecasting energy demand to ensure a stable and optimized energy network. Improved forecasting of both production and consumption enhances energy management and supports the development of robust optimization strategies. This study investigates the use of Long Short-Term Memory (LSTM) networks to forecast photovoltaic (PV) production and building energy demand, comparing their performance with a standard Artificial Neural Network (ANN) model. Using an hourly data set from 2020 to 201, for PV production, the LSTM achieved a MAE of 5.31 and RMSE of 13.78, compared to 23.41 and 47.97 for the ANN, respectively. Similarly, for energy demand forecasting, the LSTM reduced the MAE from 2.52 to 1.60 and the RMSE from 3.69 to 2.33, indicating a significant improvement in predictive accuracy. In terms of the coefficient of determination (R²), the LSTM achieved 0.86 for PV prediction and 0.74 for demand, compared to 0.93 and 0.69 for the ANN. The findings highlight the superior performance of LSTM networks in capturing temporal dependencies and improving forecast precision for both PV generation and energy demand.
Keywords
ANN, LSTM, forecasting, PV production, electricity demand