| Title |
Optimal Model Selection of Long Short-Term Memory for Mid-Term Electric Load Forecasting |
| Authors |
엄호용(Ho-Yong Eom) ; 백승묵(Seung-Mook Baek) |
| DOI |
https://doi.org/10.5370/KIEE.2022.71.11.1658 |
| Keywords |
Activation function; electric load; long short-term memory; optimization method |
| Abstract |
This paper proposes an optimal electric load forecasting technique by applying several activation functions and optimization methods in the internal structure of the Long Short-Term Memory (LSTM) algorithm to forecast mid-term electric load in Korea. As the activation function, the rectified linear unit (ReLU) and hyperbolic tangent (Tanh) are used. Optimization methods such as stochastic gradient descent (SGD) and adaptive moment optimization (Adam) are compared in the paper. In the input data, daily peak load over past seven days, maximum and minimum temperatures are used. In order to increase the efficiency of learning, special weeks and special day load data are replaced. The proposed method shows stable and great performance with small average errors, compared to the RANN based forecasting method which is a previous study. Especially, the estimated error is improved in the summer and winter where the temperature changes significantly. |