BLOG TEST Application for forecasting models of photovoltaic generation for the ZE PAK farm

Tags:

Application for prognostic models of photovoltaic generation on the example of the ZE PAK farm

Zespół Elektrowni Pątnów-Adamów-Konin is an energy company from Wielkopolska historically dealing with the production of electricity from hard coal. Currently, it is a leader in the energy transformation, investing in one of the largest photovoltaic installations in Poland – the Brudzew farm. The newly built farm with a capacity of 70 MW consists of 155 thousand. panels, 30 transformer stations, 336 inverters and covers an area of ​​100 ha. It is planned to expand the farm to 390 MW, and in the future to 1000 MW.

Challenges and needs

Photovoltaics is known for its volatility and limited controllability. Therefore, for producers and sellers of green energy, precise production forecasts are a huge advantage on the market. In the case of the Brudzew farm, the most important were the day-ahead forecast and the rolling update of the intra-day forecast in order to balance the company’s position. The critical factor was the accurate modeling of the relationship between the solar radiation forecast and the AC power fed into the grid.

Implemented solution

We implemented two parallel predictive models. The first one is based on historical measurements of power and insolation. We tested many statistical models and chose the one with the best predictive power. The results are continuously refined with measurements received from the plant.

The second model directly simulates the operation of real PV farm components – cells, panels, inverters. This approach does not require historical data and is perfect for forecasting newly built installations, the actual characteristics of which are not fully known.

Both models use the weather forecast provided by external providers. We use the forecast of irradiance (insolation), temperature and cloud cover to correlate atmospheric data with real power measurements as accurately as possible. Our system automatically downloads data from weather services with the frequency set by the user (up to 15 minutes),

Results obtained

The quality of PV generation forecasts for the pilot plant is promising – the average daily error of our prediction model for real weather data does not exceed 5%.