Artificial Intelligence Study Reveals Effective Framework for Predicting Photovoltaic Power

Strategic Insights -

A recent study on artificial intelligence has demonstrated a framework based on transfer learning to predict day-ahead photovoltaic power accurately. This research, conducted at the Polytechnic University Milan, aimed to investigate the effectiveness of advanced forecasting methods, such as machine learning and deep learning, in predicting PV power production. The study's findings suggest that the transfer learning method, which utilizes reliable trained deep learning models of old PV plants in newly installed PV plants, yields better performance in PV power prediction.

Key Takeaways:

  • The study presented a framework based on the transfer learning method to use reliable trained deep learning models of old PV plants in newly installed PV plants.
  • The numerical results showed the effectiveness of transfer learning in day-ahead PV prediction in newly established PV plants where a sizable historical dataset of them is unavailable.
  • Among the nine models presented in the study, the LSTM models had better performance in PV power prediction.
  • The new LSTM model using an inadequate dataset had a 0.55 mean square error (MSE) and 47.07% weighted mean absolute percentage error (wMAPE).
  • The transferred LSTM model improved prediction accuracy to 0.168 MSE and 32.04% wMAPE.
  • The study used a dataset of years of PV power outputs to capture hidden patterns between essential variables to predict day-ahead PV power production accurately.
  • The researchers utilized machine learning models to predict PV power accurately and developed a new LSTM model using an inadequate dataset.

Statistics:

  • The new LSTM model using an inadequate dataset had a 0.55 mean square error (MSE).
  • The transferred LSTM model had a 0.168 MSE.
  • The new LSTM model using an inadequate dataset had a 47.07% weighted mean absolute percentage error (wMAPE).
  • The transferred LSTM model had a 32.04% wMAPE.
  • The study included a dataset of years of PV power outputs.
  • The researchers developed a new LSTM model using an inadequate dataset.

Sources:

  • A Day-Ahead Photovoltaic Power Prediction via Transfer Learning and Deep Neural Networks. Forecasting, 2023, 5(1). (Forecasting - http://www.mdpi.com/journal/forecasting)
  • Polytechnic University Milan Researcher Furthers Understanding of Machine Learning (A Day-Ahead Photovoltaic Power Prediction via Transfer Learning and Deep Neural Networks). Journal of Engineering. March 6, 2023; p 2020.