| Title |
An Ensemble Forecasting Method of Winter and Summer Peak Load Using Deep Neural Networks |
| Authors |
권보성(Bo-Sung Kwon) ; 송경빈(Kyung-Bin Song) |
| DOI |
https://doi.org/10.5370/KIEE.2020.69.6.765 |
| Keywords |
Peak Load Forecasting; Deep Neural Networks; Ensemble Forecasting Method |
| Abstract |
It is important to accurately forecast the peak load for winter and summer to maintain an adequate reserve power and operate a stable power system. Deep neural networks are used to forecast the peak load for winter and summer, which reflect the periodic features of load and factors that have nonlinear relationship with load. For the mid-term load forecasting model using deep neural networks, the model structure is improved to a deep neural network forecasting model in which the long short-term memory(LSTM) layer and the fully-connected(FC) layer are connected in parallel. The forecast models using deep neural networks are classified into an average temperature model, a highest temperature model, and a lowest temperature model according to the type of input data.
The proposed ensemble model is a model in which the structure of deep neural network is improved and the monthly optimal weight is applied to the three models according to the type of the temperature input. The proposed ensemble model has improved the mean absolute percentage error(MAPE) and the variance of MAPE than the multiple linear regression model and the forecast model which are trained using various temperature. |