Abstract ▼ AI-Helper

We propose the novel solar power forecasting algorithm by using the Extreme Gradient Boosting (XGBoost) machine based on the 2-stage forecasting structure. Our algorithm is implemented to solve three problems. First, the solar power is linearly proportional to the solar irradiation on a target solar...

We propose the novel solar power forecasting algorithm by using the Extreme Gradient Boosting (XGBoost) machine based on the 2-stage forecasting structure. Our algorithm is implemented to solve three problems. First, the solar power is linearly proportional to the solar irradiation on a target solar panel, but it is hard to obtain the target solar irradiation. Therefore, in the first stage, we predict the target solar irradiation by using the XGBoost based on numerical weather prediction, which is measured on a different location but modified for the target location. Second, the forecasting errors on the predicted solar irradiation can be transferred to the second stage when the predicted solar irradiation is used to predict the solar power. We forecast the conditional error distribution of predicted irradiation by collecting forecasting errors, and we sample solar irradiation scenarios, which are converted to the solar power scenarios. Then, the final point forecast of solar power is estimated by calculating the median of scenarios so that we can improve the forecasting accuracy. Third, in this process, the quality of numerical weather prediction deteriorates as the target hour is farther. Therefore, we build forecasting models for each target hour in parallel to minimize the forecasting accuracy deterioration from the quality deterioration. Finally, we verify our proposed algorithm by participating in the solar power forecasting competition hosted by KPX.

Keyword