상세 보기
초록
To handle the issues related to climate change and the energy crisis, smart grid and microgrid technologies based on renewable energy and energy storage systems are receivingl ots of attention. In particular, accurate short-term load forecasting (STLF) is essential in establishing energy strategies in the energy management system, a significant element of the smart grid; herein, external factors, such as time-stamps and weather information, and historical electric load are utilized as input variables for the construction of STLF models. However, various internal and external aspects may cause missing values in smart meter data, which deteriorate the performance of the forecasting model due to a lack of training data. To address this issue, this study proposes a missing value imputation method for smart meter data. We first collect building electrical energy consumption data fromcommercial and industrial buildings in South Korea and configure various input variables related to the building electrical energy consumption. We then construct a missing value imputation model based on LightGBM (short for Light Gradient Boosting Machine) to replace missing values via 5-fold and 10-fold cross-validations, assuming that 20% and 10% of the missing values exist. We finally apply SHAP (short for SHapley Additive exPlanations) to the LightGBM model for different buildings and cross-validations to represent the interpretability of these models. Through various experiments, the LightGBM model derived better performance than other famous machine learning models in missing value imputation, and the input variables related to the days of the week were the significant variables in constructing the missing value imputation model.
키워드
- 제목
- 강건하고 설명 가능한 스마트 미터 데이터 결측치 대체 기법
- 제목 (타언어)
- A Robust and Explainable Missing Value Imputation Method for Smart Meter Data
- 저자
- 한유나; 문지훈; 노승민; 장항배
- 발행일
- 2022-08
- 저널명
- 한국전자거래학회지
- 권
- 27
- 호
- 3
- 페이지
- 21 ~ 43
- 언어
- KOR
- 출판사
- 한국전자거래학회
- 발행국가
- 대한민국
- 분량
- 23 페이지
- ISSN
- E 2765-3846
P 2288-3908