Solar radiation forecasting using clustering methods and probabilistic forecasting models

<p>In recent years, the problem of climate change caused by anthropogenic greenhouse gases has become increasingly serious. Therefore, interest in renewable energy is growing. The advantage of using renewable energy is that there is no depletion of resources and no greenhouse gas emissions because the energy is derived from nature. Among renewable energies, photovoltaic power generation is easier to introduce than other power generation systems, and its introduction is expected to accelerate in Japan as the country moves toward a carbon-neutral policy by 2050. Accordingly, photovoltaic power generation forecasting is becoming extremely important in terms of power system planning and operation. Solar radiation forecasting research is being conducted because photovoltaic power generation can be calculated from solar radiation and is versatile. It is widely known that solar radiation is affected by weather factors and thus contains uncertainties. In this study, natural gradient boosting which is a probabilistic forecasting model is used to make forecasts that take into account the uncertainty of solar radiation. In addition, we aim to build a model that improves accuracy and reduces the amount of data by combining a clustering method which is k-means++ with a time-series forecasting model.</p>