Forecast of Long-term Trend of System Marginal Price with Amounts of Machine Learning Train Data

The yearly system marginal prices (SMPs) in mainland Korea, from 2020 to 2030, were predicted using significant amounts of machine learning training data. The factors for deciding SMP were collected from public data portal sites. The factors included supply capacity, maximum power, supply reserve, liquefied natural gas (LNG), West Texas intermediate crude oil (WTI), and FOB Kalimatan. The best two factors for forecasting SMP, LNG, and WTI were selected through correlation analysis. The training data were divided into cases, A for 10 years and B for 5 years. The models, K-nearest neighbor (KNN), light gradient boost machine (LGBM), random forest (RF), and support vector regression (SVR) models were used for machine learning, and their accuracy was evaluated. Finally, long-term mainland SMPs were forecasted using Japanese LNG and WTI prices. The resultant model for the most accurate machine learning was LGBM which was used to forecast long-term SMPs. The mainland SMP was predicted to decrease from 2020 to 2022 and then maintain 72 KRW/kWh for Case A and 69 KRW/kWh for Case B until 2030.