<![CDATA[Accurate short-term power output forecasting for Photovoltaic (PV) systems is crucial for electricity grid management and energy trading. This study proposes and validates a Long Short-Term Memory (LSTM) model, a Deep Learning architecture, for forecasting PV power output 1-hour ahead using historical and real-time weather variables (irradiance, temperature, humidity, and wind speed). The model is compared against the conventional Autoregressive Integrated Moving Average (ARIMA) and Support Vector Machine (SVM) models. One year of 15-minute performance data from a 50 kWp rooftop PV system was utilized for model training and testing. Evaluation results demonstrated that the LSTM model significantly outperformed the ARIMA and SVM models in terms of accuracy metrics. The LSTM model achieved a Mean Absolute Error (MAE) of 5.5% and a Root Mean Square Error (RMSE) of 7.8% of the nominal capacity, substantially lower than the comparative models, especially under fluctuating weather conditions (partial cloudiness). The superiority of LSTM lies in its ability to capture the complex temporal dependencies between weather variables and power output, a major challenge for traditional statistical models. This research confirms that the integration of Deep Learning offers a more robust and accurate solution for PV power forecasting, supporting grid operators in achieving higher reliability and operational efficiency.]]>
Copyrights © 2026
