Hopero: Forecasting one day ahead solar panels energy supply to the grid

The aim of our collaboration  with Greenlogy within the EDIH HOPERO was to create a predictive model for one day ahead solar panels energy supply to the grid as recorded by smart meters.

Together with green transition, solar energy is a fast growing source of electric power in the whole world. However, solar energy is by its nature an unstable source and is highly dependent on meteorological factors. These instabilities may result in grid disruptions and difficulties in market electricity value estimation, which makes precise predictions of solar power production and its supply to the grid crucial.

We tested the state-of-the-art machine learning approaches and brought improvements for models to better fit the addressed task.

The first step of the project was to identify the factors influencing the households production and consumption. Using the statistical methods we identified the most important weather and solar position parameters together with static categorical parameters such as day of the week, national or school holiday.

As a prediction model we used artificial neural networks, extending the state-of-the-art architecture by new layers to better adapt to the addressed problem. The final model was trained to deal with the principal above-mentioned factors and the historical production and consumption. The robustness of the model was supported by extensive data augmentation.