Title: Comparison of machine learning methods using time series data: focusing on inverter data

Authors: Sang-Ha Sung; Chang Sung Seo; Min Ho Ryu; Sangjin Kim

Addresses: Department of Management Information Systems, Dong-A University, Busan 49236, South Korea ' SCT, Busan 48059, South Korea ' Department of Management Information Systems, Dong-A University, Busan 49236, South Korea ' Department of Management Information Systems, Dong-A University, Busan 49236, South Korea

Abstract: In this study, we use inverter data to understand the inverter status and present a predictive model for the future status. Data was collected from the inverter through the sensor, and was collected for about two months from July to August 2020, for a total of 8,954,665 time points. The used data consists of frequency and leak level, and when the value of data increases significantly, it is classified as having an abnormality in the inverter. In this study, we present a time series prediction model that can predict inverter status abnormalities by comparing various machine learning techniques. In this study, the inverter state was predicted using the boosting method, the tree method, the SVR method, and the deep learning method. As a result of the experiment, the error rate of the deep learning technique was the lowest.

Keywords: time series data; machine learning; boosting methods; tree methods; SVR methods; deep learning; regression; inverter data; predict; failure predict.

DOI: 10.1504/IJEWE.2023.132428

International Journal of Environment, Workplace and Employment, 2023 Vol.7 No.1, pp.13 - 33

Received: 08 Mar 2022
Accepted: 12 Jan 2023
Published online: 19 Jul 2023 *