This paper describes a method for predicts future behavior of generating power of photovoltaic power generation apparatus, PV for short, using time-series datamining technique. Introducing the renewable energy apparatus, such as PV, and then operating pieces of apparatus by renewable power are essential for reducing the total emission of carbon dioxide. The point here is harmonization among the whole pieces of apparatus, we thus are necessary making a plan to share the limited PV power. The generating power of PV varies depending on meteorological factors, which means the optimized harmonization plan must be dynamically changed in accordance with the generated power. For this reason, we need a method to predict the future generation power of PV in making a plan. In this paper, we develop a prediction method based on a combination of clustering, decision tree learning and dynamic time warping. The utilization data is time series from five sensors that are the temperature, the atmospheric pressure, humidity, the velocity of the wind, and the solar radiation. The output of the system as prediction is a sequence of daily generating power of PV at intervals of five minutes. We carry out to install and experiment with a PV of 25 [m
2] and several sensors on a detached house. As a result of comparing measurements to prediction values, the mean absolute error per day is 0.34 [kW].
View full abstract