Highly-accurate short-term forecasting photovoltaic output power architecture without meteorological observations in smart grid

We propose a forecasting architecture of near future photovoltaic output power based on the multipoint output power data via smart meter. The conventional forecasting methods are based on the analysis of meteorological observation data, and need the implementation of dedicated meters and the connection to them. Moreover, highly-accurate forecasting(in second-scale, or meter-scale) is difficult in the conventional methods. Our proposed method is based on not meteorological observation data but the actual measured output power data by using the solar panels connected with a smart meter as sensing units. A forecasting calculation server interpolate spatially the actual measured data collected from multipoint, and forecasts near future output power in each point using optical flow estimation. Virtual sampling technique involves the forecast performance when the sampling point is sparse. We show the forecasting method achieves high accuracy of less than 5% error rate by the computer simulation.