Hybrid Machine learning models for PV output prediction: Harnessing Random Forest and LSTM-RNN for sustainable energy management in aquaponic system

Abstract

Accurately forecasting photovoltaic (PV) System output is vital for optimizing energy management in sustainable aquaponic systems, where fluctuating solar irradiance poses significant challenges. This study presents a hybrid Long Short-Term Memory Recurrent Neural Network (LSTM-RNN) and Random Forest (RF) model to address these challenges effectively. By integrating LSTM-RNN’s capability to model temporal dependencies with RF’s strength in feature selection and non-linear data handling, the model demonstrates superior predictive accuracy across parameters such as voltage, current, power, and irradiance. Advanced preprocessing steps, including normalization and sequence transformation, are employed to align datasets with temporal patterns, enhancing the model’s learning efficiency. Evaluation metrics, such as Root Mean Squared Error (RMSE) and Mean Absolute Error, validate the model’s precision, with RMSE values of 0.0768 for voltage, 0.037 for current, and 0.0363 for irradiance, outperforming standalone LSTM (RMSE > 5 %) and RF models. The RF component prioritizes critical predictors like solar irradiance and temperature, contributing 45 % and 22 % to accuracy, respectively. The hybrid model supports efficient energy storage during peak sunlight and consistent power distribution during low irradiance, ensuring reliable operation of aquaponic systems for water circulation and lighting. Its scalability and adaptability make it a promising tool for improving energy efficiency and reducing operational costs. Future research will explore its application in larger PV installations and integration with weather forecasts, enhancing performance under diverse environmental conditions. This study underscores the transformative potential of hybrid models in advancing renewable energy forecasting and promoting agricultural sustainability.