Competition on solar generation forecasting


Submission details

The competition will comprise a single submission for each participant, which consists of the hourly solar generation forecast for an entire week. Submission by 31 October 2022 New Deadline: 15th January 2023. In order to build and refine the forecasting models, a data set comprising three Years of historical data is already available, as well as the data set comprising the previous 4 months before the week to be forecasted (participants need to register and log in to have access to these data).

The deadline for the results submission has been postponed to 15th January 2023.

The participants that already made their submissions are invited to improve their results until the new deadline.

In summary:

  • Requested forecasts: Solar generation forecast for each hour of a full week (168 values).
  • Inputs/historical data (all referring to 5 minutes intervals): PV generation and weather information, including temperature, wind speed, and humidity, among other details.

The provided values of power, included in data sets, represent the mean of instant power during 5 minutes. To participate in this competition, it is requested the forecasted values representing the mean of instant power during 1 hour, leaving to the participant the required processing of data needed to provide results for 1 hour periods.

Competition Outline

Energy resources forecasting is increasingly important in current and future power and energy systems. Due to the high uncertainty of generation based on renewable energy sources, which results from their dependence on weather conditions, such as wind speed or solar intensity, the need to develop suitable solutions to deal with such variability increases considerably. Relevant effort is being put on the development of energy consumption and generation forecasting methods, able to deal with different forecasting circumstances, e.g., the prediction time horizon, the available data, the frequency of data, or even the quality of data measurements. The main conclusion is that different methods are more suitable for different prediction circumstances, and it is not clear that a certain method can outperform all others in all situations. This competition fosters the benchmarking of artificial intelligence methods for solar generation forecasting.