Development of forecasting method of time variation of net solar output over wide area using grand data based neural network

Abstract

This paper focuses on the short-term forecasting of the temporal variation in the net output of photovoltaic power generation across a wide area. Due to the unstable output fluctuations of photovoltaic power generation, thermal power generation is necessary. However, to handle unpredictable power fluctuations, thermal power often operates in a no-load standby mode, resulting in wasteful energy consumption. To address this issue, we have developed a novel prediction method that utilizes neural networks for short-term forecasting of the net output of photovoltaic power generation in a wide area. The key aspect of this method is the utilization of the distributed solar power generation itself as a sensor within the target area, enabling the use of BIG DATA derived from the sensor to predict future net output of solar power generation using a neural network. To expedite calculations, we have incorporated an autoencoder and a decoder. We applied this methodology to northern Kyushu and conducted thorough verification. Furthermore, we compared the persistent model with the smart persistent model and demonstrated their effectiveness as viable solutions.