Single Step Multivariate Solar Power Forecasting using Adaptive Learning Rate LSTM Model with Optimized Window Size

Accurate photovoltaic (PV) power forecasting is crucial for the successful integration of residential PV systems into the electrical grid. It enables grid operators to optimize grid operations, ensure stability, facilitate market operations and trading, and plan for future system expansion. In this study, we propose a new model that combines an adaptive learning rate Long Short-Term Memory (LSTM) with an optimized window size for improved PV power forecasting. The proposed model is trained and tested using historical time series data of projected PV power and weather conditions, considering the GPS location of the PV system. The model’s performance is compared against other commonly used forecasting models, including LSTM, Bi-LSTM, LSTM-Transformer, and CNN-LSTM, for single-step size forecasting, specifically predicting PV power for the next hour. The results demonstrate that the proposed model outperforms all other models in terms of accuracy for the single-step forecasting task. The adaptive learning rate LSTM with optimized window size demonstrates superior performance, indicating its effectiveness in capturing the temporal patterns and dependencies in PV power generation.