Dutch researchers have developed a PV prediction method that uses the XGBoost algorithm. They claim that their approach predicts the level of electricity generation an hour ahead for large fleets of residential solar panels.
They described their findings as “Individual return on now-casting residential solar systems”, published recently Solar energy. The researchers said the new approach can predict the individual output of large solar PV systems.
Their new method is based on a single XGBoost algorithm, which is a decision tree ensemble, open access algorithm that uses a gradient boosting framework.
“Since XGBoost is made from a combination of decision trees, the importance of each feature is relatively simple to calculate,” the researchers said. “The decision tree makes predictions by dividing decisions into branches.”
They applied the algorithm to 1,102 residential PV systems in the Netherlands and Belgium. They looked at global horizontal radiation (GHI), cloud cover, wind speed, precipitation and ambient temperature based on data provided by the Royal Netherlands Meteorological Institute.
Their methods also take into account the PV system’s size, age, panel types, latitudes, longitudes, panel tilt, orientation, inverter annual decay efficiency, and operating cell temperatures.
“The descriptive parameters consist of the day of the year and the historical yield 24, 48, 72, 96 and 120 hours earlier,” the researchers said, noting that Dutch startup Solar Monkey provided all system data. “The data must be properly prepared for the machine learning model to get the best results.”
With an average PV system size of 4.4 kW, the algorithm achieved a mean absolute error (MAE) of 0.877 kWh and a mean absolute percentage error (MAPE) of 23% for hourly data aggregated to daily values.
“XGBoost predictions for individual PV systems are, on average, two times better than currently used commercial software,” the researchers said. “Despite the issues presented, XGBoost provides a two-fold improvement over a commercially available analytical model.”
