Search

The objective of this study is to propose and evaluate a set of modifications to enhance a machine-learning-based method for forecasting day-ahead solar irradiation. To assess the proposed modifications, they were implemented in an initial forecast method, and their effectiveness was analyzed using two years of data on a national scale in Japan. In addition, the accuracy of the modified method was compared with one of the forecast methods for solar irradiation used by the Japan Meteorological Agency (JMA), namely, the mesoscale model (MSM). Such forecasts were made publicly available only recently, which makes this study one of the first ones to compare them with machine-learning-based forecasts. The annual root-mean-square error (RMSE) of local forecasts of the JMA-MSM varied from 0.1 to 0.14 kW h m−2; the regional equivalent varied from 0.062 to 0.091 kW h m−2. In comparison with these results, the modified model achieved an average RMSE reduction of 7.5% on the local scale and 16% on the regional scale. The modified model also had a skill score that was 23% higher than that of the JMA model. Furthermore, the performance of the JMA model had strong spatial and seasonal dependencies, which were reduced in the machine-learning-based forecasts. The results show that the proposed modifications are effective in reducing large forecasts errors, but they cannot compensate for situations in which the input data used to make the forecasts are highly inaccurate. [ABSTRACT FROM AUTHOR]

Abstract: In this study, forecast characteristics of a global horizontal irradiance (GHI) forecasts by a meteorological mesoscale model (MSM) developed by the Japan Meteorological Agency (JMA) for the Kanto region, Japan were investigated during the period 2008–2010 for the next day based on forecast on different issue times on the previous day of photovoltaic (PV) power production. The evaluation of the GHI forecasting errors by MSM showed that the mean bias error (MBE) values of the GHI range from −50 to 50W/m2 in a year. The root mean square error (RMSE) values in winter were about 90–100W/m2, while the RMSE values in summer approached up to 150W/m2. The hourly GHI forecasting errors indicated that the GHI values were generally underestimated (overestimated) in summer (winter) compared with the observations. The dependence on different four initial times were found from RMSE values, suggesting that forecasts of initial time of 21UTC were generally better than those of the other initial times. Appearance frequency of cloud types based on the visual monitoring in cases of relatively large GHI forecasting errors, which cases were selected by normalized forecast error by the surface observations (the threshold of >0.4 or <−0.4 are given for the Kanto region), indicated that stratocumulus (cirrus) clouds for the overestimation (underestimation) of the GHI were occurred significantly, while altocumulus and cumulus clouds were observed as a whole. Verifications in GHI forecasts indicated that improvements of the treatment of these clouds in MSM were necessary for the GHI forecasting and the PV power production accurately. [Copyright &y& Elsevier]