PREDICTION METHOD FOR SOLAR POWER BUSINESS BASED ON FORECASTED GENERAL WEATHER CONDITIONS AND PERIODIC TRENDS BY WEATHER

With the introduction of photovoltaics rapidly accelerating and its in(cid:13)uence on the electric power system expanding, there is a growing demand for the prediction of solar power output and solar radiation. In this paper, we present a method to predict solar radiation and solar power output using an estimated trend and general weather forecasts reported by the national meteorological agency, taking particular note of a smooth periodic trend identi(cid:12)ed when dividing the measured value of solar radiation by the hourly time zone and weather. First, by constructing a generalized additive model (GAM) in which the periodic dummy variable and actual general weather conditions are used as explanatory variables, we extract the seasonal trends of solar radiation and solar power output for different general weather scenarios, such as sunny, rainy and cloudy. Next, we estimate the probability (conditional expected value) of actualizing each weather scenario given a forecasted weather condition by using a multinomial logit model, noting that the prediction method used in common practice, in which the forecast values are directly submitted as if they were actualized, possibly brings bias to the predicted values because it excludes the probabilities that the weather forecast is wrong. Then, in combination with seasonal trends estimated by GAM, we construct a new prediction model calculating prediction values of solar radiation and power output. Finally, this study also veri(cid:12)es the superiority of this proposed prediction method in the reduction of prediction error by comparing it with preceding methods and the prediction method that directly substitutes forecast scenarios.