Abstract—Accurate photovoltaic (PV) power forecasting repre- sents a critical component of modern microgrid energy manage- ment systems. This study presents a comprehensive evaluation and comparison of feedforward neural networks (FFNNs) and classical AutoRegressive Integrated Moving Average (ARIMA) models for PV forecasting, utilizing high-resolution data from the LAMBDA Microgrid Laboratory at Sapienza University of Rome. The investigation systematically examines the influence of environmental features, specifically temperature and irradi- ance measurements, across multiple forecasting horizons. Results indicate that FFNN models substantially outperform ARIMA approaches, with the optimal configuration, which combines PV output, temperature, and irradiance data, achieving a root mean square error (RMSE) of 0.0915 for one-step-ahead forecasting. Among the seven forecast horizons tested (8, 16, 24, 32, 48, 64, and 96 steps ahead), the configuration maintains robust generalization, achieving RMSE 0.189±0.005 at the 96-step (24- hour ahead) horizon. Although computationally efficient, the ARIMA benchmark exhibited limited predictive performance with an RMSE 0.2732 on test data. These findings establish FFNNs with environmental inputs as a reliable solution for microgrid forecasting, offering practical insights for deployment in an energy management system.
A Neural Network-Based PV Forecasting Approach with Environmental Features for the LAMBDA Microgrid EMS / Aslam Nat Dept., Of Computer; Control And Management Engg., University of Rome ”La Sapienza”; Rome, Properties Ltd; Italy Email:, Aslam. Nat@uniroma1. It Stefano Leonori Dept. Of Information Engg.; Electronics And Telecom., University of Rome ”La Sapienza”; Rome, Properties Ltd; Italy Email: Stefano. Leonori@uniroma1. It Cristina Moscatiello Dept., Of Astronautical; Electrical And Energy Engg., University of Rome ”La Sapienza”; Rome, Properties Ltd; Italy Email: Cristina. Moscatiello@uniroma1. It Luigi Martirano Dept., Of Astronautical; Electrical And Energy Engg., University of Rome ”La Sapienza”; Rome, Properties Ltd; Italy Email: Luigi. Martirano@uniroma1., It. - (2025). ( IEEE ISGT Europe 2025 – Innovative Smart Grid Technologies Conference Malta ).
A Neural Network-Based PV Forecasting Approach with Environmental Features for the LAMBDA Microgrid EMS
Rome;Rome;Rome;Rome;
2025
Abstract
Abstract—Accurate photovoltaic (PV) power forecasting repre- sents a critical component of modern microgrid energy manage- ment systems. This study presents a comprehensive evaluation and comparison of feedforward neural networks (FFNNs) and classical AutoRegressive Integrated Moving Average (ARIMA) models for PV forecasting, utilizing high-resolution data from the LAMBDA Microgrid Laboratory at Sapienza University of Rome. The investigation systematically examines the influence of environmental features, specifically temperature and irradi- ance measurements, across multiple forecasting horizons. Results indicate that FFNN models substantially outperform ARIMA approaches, with the optimal configuration, which combines PV output, temperature, and irradiance data, achieving a root mean square error (RMSE) of 0.0915 for one-step-ahead forecasting. Among the seven forecast horizons tested (8, 16, 24, 32, 48, 64, and 96 steps ahead), the configuration maintains robust generalization, achieving RMSE 0.189±0.005 at the 96-step (24- hour ahead) horizon. Although computationally efficient, the ARIMA benchmark exhibited limited predictive performance with an RMSE 0.2732 on test data. These findings establish FFNNs with environmental inputs as a reliable solution for microgrid forecasting, offering practical insights for deployment in an energy management system.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
