순환신경망을 이용한 충주수상태양광 발전량 예측 (The Power Generation Prediction of Chungju Floating Photovoltaic Utilizing RNN)

서지정보

· 발행기관 : 전력전자학회
· 수록지 정보 : 전력전자학회 논문지 / 28권 / 4호 / 263 ~ 270페이지
· 저자명 : 권오극, 홍현표, 조현식, 차한주

초록

Wind and the sun are major sources of renewable energy; however, operating a power grid is difficult, as renewable energy can infiltrate the grid rapidly because of the volatility and intermittent nature of renewable energy. To reduce the volatility of renewable energy, predicting solar power generation is important. The solar radiation amount is an important factor in solar power prediction. Based on the data recorded by the Chungju floating photovoltaic system, a prediction model was created in this study, and the power generation amount was predicted hourly. The prediction method used a recurrent neural network LSTM model, and the prediction model for predicting solar power generation was presented as well as the hyperparameter selection process. In addition, the accuracy of the prediction models was presented based on the different parameters. The prediction error decreased, and the optimal model was selected based on the optimal setting of the different hyperparameters. Results with few errors were obtained through 24-hour bundled calculation, the Sigmoid activation function, the Nadam optimum, and a batch size of 30.

영어초록

Wind and the sun are major sources of renewable energy; however, operating a power grid is difficult, as renewable energy can infiltrate the grid rapidly because of the volatility and intermittent nature of renewable energy. To reduce the volatility of renewable energy, predicting solar power generation is important. The solar radiation amount is an important factor in solar power prediction. Based on the data recorded by the Chungju floating photovoltaic system, a prediction model was created in this study, and the power generation amount was predicted hourly. The prediction method used a recurrent neural network LSTM model, and the prediction model for predicting solar power generation was presented as well as the hyperparameter selection process. In addition, the accuracy of the prediction models was presented based on the different parameters. The prediction error decreased, and the optimal model was selected based on the optimal setting of the different hyperparameters. Results with few errors were obtained through 24-hour bundled calculation, the Sigmoid activation function, the Nadam optimum, and a batch size of 30.

참고자료

· 없음