Self-amending deep learning framework for enhanced multi-step photovoltaic power prediction

Abstract

Smart grids require precise solar power forecasts for efficient operation. Traditional forecasting methods often fall short due to limited data, complex time patterns, and scalability issues. This research presents a novel multi-task forecasting method, the Self-Amending Bi-LSTM-based Photovoltaic Power Forecasting model (SA-PPF-BiLSTM), designed to accurately predict Photovoltaic (PV) power output. The SA-PPF-BiLSTM incorporates a sophisticated architecture that combines advanced Bi-LSTM layers with self-adjustment mechanism that refines predictions by detecting and correcting deviations between observed PV energy production and the baseline forecast. The method facilitates improved data usage, rapid model adjustments, and significant savings in computational resources. real-world data is used to validate the model's performance, which is assessed using Mean Square Error (MSE), Root Mean Square Error (RMSE), and Mean Absolute Error (MAE) metrics. Substantial improvements in precision of amended forecasts for amorphous silicon (a-Si) a-Si and monocrystalline silicon (m-Si) solar technologies are revealed. An impressive 63% reduction in MSE is observed for the a-Si plant, while the m-Si plant shows a 47% decrease. In addition, significant improvements are seen in MAE values, with a 41% reduction for a-Si plant, while m-Si plant records a more moderate 28% decrease. These findings are poised to significantly shape strategic planning relying on artificial intelligence for PV power predictions, particularly in the scheduling of activities pertaining to solar PV energy.