Case study of outlier event of solar irradiance forecasts from a numerical prediction model

摘要

A significant number of photovoltaic (PV) power systems have been installed in Japan since the introduction of a feed-in-tariff program in 2012. PV power generations are directly affected by solar irradiance and have also large variations in spatial and time scales. For stability control of power grids with system operation in energy management systems (EMS), one day ahead forecasts of solar irradiance with numerical weather prediction model (NWP) are important. However, solar irradiance forecasts by the NWP have often yield errors. In particular, large forecast errors of PV power or solar irradiance forecasts are undesirable in the operation of EMS. In this study, we investigate, as a case study, outliers events of solar irradiance forecasts (i.e, cases when relatively large forecast errors occurred). The forecasts are based on the NWP from the Japan Meteorological Agency, and they were performed for a spatially large area (the operation area of a power utility) around Tokyo in Japan. From validation results, we found that, in some cases of one day ahead forecasts, it was difficult to forecast cold front locations with moving synoptic-scale low pressure zone. However, forecasts in early morning (06 LST initialization time in Japan) were able to detect the location of cold front and/or cloud fields associated with low pressure zones. It is seems that the improvement of solar irradiance happened due to data assimilation using latest observation data. These results indicated that updated forecast data of solar irradiance proved to be useful to comprehend large errors in one day ahead forecasts.