Interpretable Short-Term Electrical Load Forecasting Scheme Using Cubist

Jihoon Moon,Seungmin Rho,Eenjun Hwang,Sungwoo Park,Heng Liu

doi:10.1155/2022/6892995

Jihoon Moon, Seungmin Rho + Show 3 more

Open Access

https://doi.org/10.1155/2022/6892995

Copy DOI

Abstract

Daily peak load forecasting (DPLF) and total daily load forecasting (TDLF) are essential for optimal power system operation from one day to one week later. This study develops a Cubist-based incremental learning model to perform accurate and interpretable DPLF and TDLF. To this end, we employ time-series cross-validation to effectively reflect recent electrical load trends and patterns when constructing the model. We also analyze variable importance to identify the most crucial factors in the Cubist model. In the experiments, we used two publicly available building datasets and three educational building cluster datasets. The results showed that the proposed model yielded averages of 7.77 and 10.06 in mean absolute percentage error and coefficient of variation of the root mean square error, respectively. We also confirmed that temperature and holiday information are significant external factors, and electrical loads one day and one week ago are significant internal factors.

Highlights

A building energy management systems (BEMSs) is a computer-aided tool that improves energy efficiency between the grid operator and consumers through bidirectional interaction [7, 8]
To demonstrate the validity of the proposed model, we considered a total of 12 machine learning methods including multiple linear regression (MLR), partial least squares (PLS), multivariate adaptive regression splines (MARS), K-nearest neighbor (KNN), support vector regression (SVR), decision trees (DTs), bootstrap aggregating (Bagging), RF, gradient boosting machine (GBM), XGB, and CatBoost
For DA forecasting, we considered the MLR model in [20], which achieved better prediction performance than Korea Power Exchange (KPX)’s forecasting model, as a baseline model in the evaluation of the proposed model. e MLR model adapts a rolling procedure using a dataset from one day to one year before the prediction time point and is specified in the following equation: YD β0 + β1YD−1 + β2YD−2 + β3YD−7 + β4WD + β5SD + β6TD + β7SDTD + ε, (12)

Summary

Introduction

A BEMS is a computer-aided tool that improves energy efficiency between the grid operator and consumers through bidirectional interaction [7, 8]. Short-term load forecasting (STLF) has been widely used to determine the amount of power necessary for the reliable operation of electric utility grids from the hour to the week [9, 11]. Ey collected daily peak electrical load data in South Korea from 2012 to 2016 through the Korea Power Exchange (KPX) and constructed an MLR model using the days of the week, seasons, average temperature, and historical loads from one day to one year before the prediction time point. Ey collected hourly electrical load data for one year from a residential building with an RE system and configured five input variables, such as days of the week, hours of the day, isolation, temperature, and historical electricity consumption, to train the AANN model. Fan et al [25] proposed an SVR-based STLF model, namely, EMD-SVR-PSO-AR-GARCH, by hybridizing with empirical mode decomposition (EMD), particle swarm optimization (PSO), and autoregressive-generalized autoregressive conditional heteroscedasticity (AR-GARCH)

Methods

Results

Conclusion