• 대한전기학회
Mobile QR Code QR CODE : The Transactions of the Korean Institute of Electrical Engineers
  • COPE
  • kcse
  • 한국과학기술단체총연합회
  • 한국학술지인용색인
  • Scopus
  • crossref
  • orcid

입력 데이터 전처리 과정을 통한 태양광 발전량 예측 모델 성능 향상 Enhancing Accuracy of Solar Power Forecasting by Input data Preprocessing and Competitive Model Selection Methods


박세준(Se-Jun Park) ; 최원석(Won-Seok Choi) ; 이두희(Duehee Lee)

This paper compares various prediction models and preprocessing methods based on data from the Kaggle competition "AMS 2013-2014 Solar Energy Prediction Contest". Four predictive models are used: Linear Regression (LR), Random Forest (RF), Gradient Boost Machine (GBM), and Extreme Gradient Boost (XGBOST). The forecasting accuracy of these four prediction models was compared by changing the preprocessing methods. There are four preprocessing methods proposed in this paper. First, training data is designed by averaging closest four points using the weighted average. Furthermore, training data is designed by averaging points within a circle using the weighted average. Second, various prediction intervals are tested. Third, we propose a data selection method by analyzing the correlation of each parameter. Fourth, the data interpolation is tested. Forecasting accuracy is measured by the mean absolute error

분산전원 수용 확대를 위한 계통보강 및 출력 감발 간 경제성 분석 Economic Analysis of Power System Reinforcement and Flexible Interconnection to Increase Hosting Capacity


김현진(Hyeonjin Kim) ; 조종민(Jongmin Jo) ; 백자현(Ja-hyun Baek) ; 박현곤(Hyeongon Park) ; 황평익(Pyeongik Hwang) ; 조성수(Seongsoo Cho)

This paper studies and proposes an economic analysis of power system reinforcement and flexible interconnection to increase hosting capacity from the grid operator's point of view. Recently, as DER(Distributed Energy Resources) has rapidly increased, the grid operator must perform system reinforcement in order to stably connect DER to the grid. However, in order to reinforce the system, it is essential to analyze the economic analysis because a large initial investment cost is incurred. Therefore, in this paper, considering the cost compensation of power curtailment, an algorithm is proposed that allows the grid operator to perform economic analysis between grid reinforcement and Flexible Interconnection and determine the connection plan when operating the grid for DER interconnection. In particular, this paper analyzed the economic impact of factors affecting economic efficiency through sensitivity analysis.

출력 향상을 위한 고압 유도전동기의 다중목적 형상 최적 설계 Multi-Objective Geometric Optimum Design of High-Voltage Induction Motor for Output Improvement


김민석(Min-Seok Kim) ; 김창업(Chang-Eob Kim)

In this paper, a multi-purpose geometrical optimal design of a 1000 kW high-voltage induction motor was proposed for improving the output of the motor and reducing the temperature using a high-grade iron core and magnetic wedge. The Global Response Surface Method (GRSM) was used to satisfy the multi-objective optimization in complex design problems related to power and temperature. GRSM performs parallel analysis for accurate and efficient optimization searches, including local and global search capabilities. In addition, Hyperstudy a commercial optimization program, and an electromagnetic FEM solver were used to analyze the motor. As a result of optimum motor design, the output increased by 8% and the temperature decreased by about 10 % compared to the base model. The reliability of the proposed method was verified through experiments.

유도 가열 시스템의 실시간 임피던스 계측을 통한 용기의 온도 추정 기법 Temperature Estimation Techniques of a Pot through Real-time Impedance Measurement of an Induction Heating System


김현지(Hyun-Ji Kim) ; 허경욱(Kyung-Wook Heo) ; 정지훈(Jee-Hoon Jung)

Recently, induction heating (IH) technologies are widely applied in home cooking appliances due to characteristics such as fast heating, energy-saving, and high efficiency. In the IH systems, load impedance variations depend on switching frequency, pot material, operating temperature, and alignment between the pot and the induction coil. The pot’s temperature estimation using the impedance measurement of the resonant network can be used to improve the efficiency of the IH system and to make the IH cooking safe without additional temperature sensors. In this paper, a time-split method is proposed to measure a real-time impedance and to estimate the pot’s temperature by using the regression model of the impedance. The proposed temperature estimation technique is verified using a 2-kW series resonant inverter prototype.

보조 정류회로를 활용한 통합된 5 kW 급 소형 제논 램프 전원장치 Integrated 5 kW Xenon Flash Lamp Power Supply Using the Auxiliary Rectifier


최민규(Min-Kyu Choi) ; 송승호(Seung-Ho Song) ; 류홍제(Hong-Je Ryoo)

This paper deals with the integrated 5 kW xenon flash lamp power supply. In order to drive a xenon lamp used in an application field of light sintering of printed electronics, the power supply in consideration of load characteristics is required. The simmer power supply that maintains the xenon lamp's ignition is designed based on an LCC resonant converter with both voltage and current source characteristics. Also, an auxiliary rectifier and trigger applying high voltage are required to ignite the xenon lamp. And the power supply is designed based on a boost-type resonance circuit and a Cockcroft Walton voltage multiplier, respectively, in consideration of sharing the inverter with the simmer. This paper analyzes the topology of the proposed power supplies and selects the parameters and confirms them through simulation. Finally it was verified through experiments that the output voltages of the simmer, the auxiliary rectifier, and the trigger were 20 V, 400 V, and 10 kV, respectively.

고전계 노화 SiC-MOSFET 소자를 가지는 양방향 DC-DC 컨버터 특성 분석 연구 Performance Analysis of Bidirectional DC-DC Converter With SiC-MOSFETs Aged by High Electric Field


정재윤(Jae-Yoon Jeong) ; 곽상신(Sang-Shin Kwak)

Recently, the use of Silicon-carbide Metal Oxide Semiconductor (SiC-MOSFET) with high frequency and low loss characteristics is increasing in power conversion systems. SiC-MOSFET is aged due to the kinetic energy of electrons and thermal stress. In this case, the aged SiC-MOSFET may not perform as expected and may cause a failure of the power conversion system. This can lead to personal injury and property damage, so it is very important to increase reliability by identifying and analysing the aging process of SiC-MOSFET. In this paper, changes in electrical characteristics due to SiC-MOSFET aging and changes in loss and efficiency of the power conversion system are verified through experiments and simulations. A high electric field aging methodology are used to accelerate the SiC-MOSFET aging. Thereafter, a change in loss occurring during device aging are measured through a double pulse test, and this is modeled. By applying the loss modeling result to a bidirectional DC-DC converter, loss changes and efficiency changes due to converter aging are analyzed to determine the effect of the SiC-MOSFET aging on the converter.

DPWM 기법을 적용한 비엔나 정류기의 L 필터 설계 L-filter Design of Vienna Rectifier with DPWM Method


김수현(Su-hyeon Kim) ; 고영민(Young-Min Go) ; 이준석(June-Seok Lee)

This paper proposes the L-filter design method of Vienna rectifier with the DPWM(Discontinuous Pulse Width Modulation) method for satisfying the desired input current THD(Total Harmonic Distortion) specification. The input voltage waveform of Vienna rectifier with DPWM method is analyzed in this paper. The input current ripple is decided by the voltage applied to both end-points of the L-filter, which is voltage difference between the gird voltage and the input voltage of Vienna rectifier. In addition, the RMS(Root Mean Square) value of the input current ripple is derived by using on the waveform of input current ripple. In this paper, the RMS value of the input current ripple is used to design the L-filter satisfying the current THD specification of Vienna rectifier with DPWM method. The effectiveness of the proposed L-filter design method for Vienna rectifier with the DPWM method is verified through simulations and experiments.

데이터 선별을 통한 EfficientNetV2-L 기반 조기 위암 컴퓨터 보조 진단 시스템 연구 A Study on the Computer Aided Diagnosis System for Early Gastric Cancer Lesion Based on EfficientNetV2-L through Data Filtering


이한성(Han-sung Lee) ; 조현종(Hyun-chong Cho)

Gastric cancer is a common cancer worldwide, especially in Korea. Early diagnosis is very important to increase the full recovery rate. However, early gastric cancer has no special symptoms and is a disease that even experts find difficult to diagnose in gastroscopy. Therefore, in this paper proposed a computer-aided diagnosis(CADx) for early gastric cancer diagnosis using EfficientNetV2-L. Due to the nature of medical data, it is difficult to collect a large amount of data. The data used for training was augmented using Cifar10 policy of the Google's AutoAugment. Additionally, the augmented image was used as an input to the model trained with the original dataset and filtered according to the classification threshold. EfficientNetV2 is a classification network designed Training-NAS that can learn the feature of lesions with a small number of parameters. As a result, EfficientNetV2 set to the threshold value of 0.9 achieved the performance of accuracy 0.943 for early gastric cancer and abnormal image classification. The AUC value also increases from 0.972 to 0.991, showing that the data filtering method of this study was effective for improvement of classification performance.

소음 환경에서 상용 음성인식 API의 성능 비교 A Performance Comparison of Commercial Speech Recognition APIs in Noisy Environments


이건희(Geonhui Lee) ; 이상화(Sanghwa Lee) ; 지수환(Suhwan Ji) ; 김아욱(Auk Kim) ; 임현승(Hyeonseung Im)

This paper compares the performance of five commercial speech recognition APIs under noisy environments, namely those provided by Amazon AWS, Microsoft Azure, Google, Kakao, and Naver. To this end, we used an open dataset for development and evaluation of multi-channel noise processing technology provided in AI Hub. We tested each API’s performance with respect to the speaker’s gender and location and the speech content, and measured their error rate using both word error rate (WER) and character error rate (CER). Except for the AWS API, the error rate was higher when tested with female’s data than male’s one, and when tested with the data recorded from the side than the front. The error rate was also relatively high when the test sentences contained proper nouns such as person’s names and local names, and the shorter the sentences, the higher the error rate. Moreover, the Google API outperformed all the others in terms of both WER and CER, with 53% and 18% of error rate, respectively

스펙트로그램을 이용한 기계의 이상상태 탐지 모델 An Abnormalities Detection Model of a Machine Using Spectrogram


김연준(Youngjun Kim) ; 이석필(Seok-pil Lee)

There are many methods for diagnosing abnormal conditions of machines. Among them we use a method using sound for detecting abnormalities of machines. Experimental data sets were collected at approximately 30 minutes intervals for 2 weeks. The collected data sets are converted into spectrogram images expressed by time, frequency and amplitude with a 5 second time step. In this paper, we propose a learning model created by combining Conv1D for image processing and LSTM for time series data processing to detect abnormal conditions of machines. The comparison test with the existing model combining CNN, Conv1D and GRU shows our method has a promising result.

Non-Local Means 잡음 제거와 데이터 증강을 이용한 YOLO 기반 객체 특징 탐색 YOLO based Object Features Detection using Non-Local Means Denoising and Data Augmentation


박건(Geon Park) ; 강예연(Ye-Yeon Kang) ; 김규일(Gyu-il Kim) ; 정경용(Kyungyong Chung)

When fruits are harvested in farms, most of them go through a manual sorting process and classify and distribute decomposed fruits. However, there is a limit to manually classifying large amounts in a situation where the number of workers is decreasing in farms. To solve this problem, it is important to divide normal and decomposed fruits in real time to minimize the proportion of manpower used in the screening process. We propose a method of YOLO based Object Features Detection Using Non-Local Means Denoising and Data Augmentation. The proposed method collects desired data through the Crawling method, and the preprocessing minimizes image noise through Non-Local Means Denoising. The built image dataset uses YOLOv3, an object detection algorithm, to distinguish and detect normal and decayed fruits. As a result of performance evaluation, object detection of YOLOv3 objects in a proposed method rather than detection results shows that Recall increases 10% performance and increases 9% in the remaining Recall and IoU. Therefore, the proposed method can increase screening efficiency by detecting decayed fruits well.

스토리 영상 콘텐츠 분석을 위한 장면 그래프 생성 모델 기반의 자동 온톨로지 구축 프레임워크 A Framework of Automatic Ontology Construction based on Scene Graph Generation Model for Analysis of Story Video Contents


강동구(Donggu Kang) ; 김지연(Jiyeon Kim) ; 정종진(Jongjin Jung)

In this paper, we propose an extended ontology automatic construction framework based on multiple deeplearning models as part of an effective analysis of story video content. Semi-automatic techniques usingimage processing techniques have been the mainstream for the existing ontology construction methods formoving pictures, but there is a problem that human intervention is required and the accuracy is low due tothe limitations of image processing techniques. To overcome this, in this paper, we propose an automatedmethod based on the deep learning scene graph generation technique. In particular, in the case of videocontent with a story, the relationship between characters is a very important factor in understanding thescene, so deep learning-based object relationship creation model, character identification model, andimportant area caption generation model are applied to extract objects and recognize their relationships.And design a framework that automatically builds a domain ontology dependent on the story throughprocedural fusion between each model and module function. In addition, the proposed framework suggests amethod for efficiently processing system requirements and system resources through meta control in thecondition that requires simultaneous operation of multiple deep learning models to analyze story videocontent. Through this, the proposed framework effectively identifies critical region captions and objectrelationships in a scene in story-telling video content, and executes three types of models simultaneously.Finally, we conduct an experiment to automatically build an ontology by applying the proposed framework tospecific video content, and check the effectiveness of the proposed framework.

동적시스템 제어를 위한 효율적인 강화학습 방법 Efficient Reinforcement Learning Method for Dynamic System Control


박채훈(Chaehun Park) ; 정철민(Cheolmin Jeong) ; 유재현(Jaehyun Yoo) ; 강창묵(Chang Mook Kang)

Reinforcement learning is a method in which the controller interacts with the target system to collect data and utilizes it to evaluate and update itself. The common disadvantages of reinforcement learning are that it takes a considerable amount of time from initial random controls to valid controls, and that it can fall into local optima. For this reason, a method for improving learning efficiency by applying verified external control is being studied. In this paper, external control is utilized for data collection, making it easier to access data that is helpful for learning. This method not only improved learning efficiency, but also was able to derive a more stable controller than the external control by learning. To prove this, we applied the proposed method to RIP(Rotary Inverted Pendulum), which is used for controller stability experiments, and as external controls, swing-up and balance controls, which are commonly used for RIP control, were utilized.

이미지 증대 기법을 이용한 노이즈에 강인한 사과 질병 분류 Noise-robust Apple Disease Classification with Image Augmentation Methods


김장연(Jang-yeon Kim) ; 김태경(Tae-kyeong Kim) ; 조현종(Hyun-chong Cho)

When the apple disease occurs, accurate and rapid control must be carried out. If appropriate measures are not taken, the spread of the disease and secondary damage such as soil contamination caused by pesticides may occur. In this paper, the apple disease classification system that can classify the type of disease as well as normal from image is proposed. The apple disease classes consists of Marssonina blotch, Fire Blight, Valsa cacker, Alernaria blotch, and Bitter rot. Xception network was used to extract and learn features from image. Google's AutoAugment CIFAR-10 policy is used to increase apple disease data to increase network’s classification performance. Then, in order to increase the reliability of data, the augmented data was selected by model trained only with original data. Gaussian, Salt-and-pepper, Speckle and Poisson noise were added to the test data to show good performance for noisy input data. We compared the performance of the model trained with original data and augmented data selected by threshold value 0.9. As a result, the proposed study showed a performance improvement of up to 6% in F1-Score.

YOLOv4를 이용한 산양 탐지 시스템 Goral Detection System using YOLOv4 Object Detection Algorithm


이한성(Han-sung Lee) ; 오유정(Yu-jeong Oh) ; 박영철(Yung-chul Park) ; 임상진(Sang Jin Lim) ; 조현종(Hyun-chong Cho)

Global industrialization has made human life comfortable, while causing environmental pollution. Environmental pollution has destroyed wildlife habitats, and as a result, several animals are endangered. In particular, long-tailed goral (Naemorhaedus caudatus) is an international endangered species that must be restored. A huge amount of photographic data has been accumulated through camera trap research for the restoration of endangered species. Until now, analysis of such photo data has only relied on experts, so it took a lot of time to analyze camera trap data. The goral detection system developed in this study, based on goral camera trap data, can enable goral data analysis at a faster than conventional methods. YOLOv4 was used as an object detection algorithm for goral detection. YOLOv4 is a network based on CSPDarknet53 and has improved performance through Bos (Bag of Specials) and BoF (Bag of Freebies). As a result, goral detection was possible with high accuracy and high speed. The performance of the goral detection system can be decreased by luminance at night. In this work, we propose a light invariant goral detection system by appropriately distributing day and night datasets for training YOLOv4. Test results on the night dataset showed an mAP of 0.907 and test results on the day dataset showed an mAP of 0.927. As a result, the goral detection system showed robust detection performance at night.

볼 베어링 고장진단 기법 비교 및 XAI Grad-CAM을 이용한 분류결과 해석 연구 A Comparison Study of Ball Bearing Fault Diagnosis and Classification Analysis Using XAI Grad-CAM


김예진(Yejin Kim) ; 전현직(Hyeonjick Jeon) ; 김영근(Young-Keun Kim)

Various machine learning and deep learning methods were proposed to monitor and classify the bearing's health state using vibration signals since bearing faults are one of the most causes of failure of rotationary machine. The process of diagnosing bearing faults using machine learning is as follows. First, the features, including the fault characteristic of the vibration signals, are extracted, and these features are selected to reduce the dimension of the features. These features are input into the machine learning classifier to diagnose the system's health. In addition to machine learning methods, CNN, one of the deep learning methods, is widely used. Since the deep learning model extracts features by itself, only the preprocessing process of converting the bearing signals into 2D is needed. The fault classification accuracy of two vibration signal transformation methods as preprocessing methods for the CNN model was compared. This paper compares the bearing fault classification performance of several machine learning commonly used and the CNN model for the lab-made wind turbine machinery testbed. By comparing different feature extraction, feature selection, and classification methods, the most appropriate pipeline is selected for the testbed. Also, grad-cam, an explainable AI(XAI) technique, is applied to interpret the CNN based classification in terms of interested frequency bandwidth. The XAI analysis was verified by designing preprocessing filters based on the grad-cam outputs for enhancing classification performance.

오손과 절연체 손상에 의한 폴리머 현수애자 교체기준에 관한 연구 A Study on Replacement Guidelines for Polymer Suspension Insulators due to the Pollution and Damage


김준오(Jun Oh Kim) ; 김승완(Seung Wan Kim)

20 years have passed since the first use of Korea-made polymer suspension insulators in distribution lines, and 24 million units have been installed in Korea. However, there are no inspection methods and replacement guidelines that can be practically used in the field. In this paper, to check the insulation condition of insulators and establish replacement guidelines, the lifespan of insulators was calculated by testing the insulators by used years. In addition, the effects of pollution in coastal and industrial areas, and insulation deterioration by shed damages were tested and analyzed, and replacement guidelines were suggested according to the shed damage. As a result of the test, polymer suspension insulators can be used for more than 30 years in the Korean power system environment, and replacement guidelines were proposed when 20 % of the shed is damaged because the damage to the shed has a greater effect on insulation deterioration than contamination. It is expected that the proposed replacement guidelines will be applied to the field to prevent power outages in the distribution system and contribute to improving power quality

1,500 ㎾ 유도발전기를 배전선로에 연결할 때 발생하는 전압강하에 대한 분석 Analysis of Voltage Drop Occurring When 1,500 kW Induction Generator is Connected to Distribution Line


김종겸(Jong-Gyeum Kim)

Induction generators, like induction motors, have a disadvantage in that a voltage drop occurs due to a high current during starting. Induction generators are often connected to the ends of distribution lines. When the induction generator is started and connected to the distribution line through the generated transformer, a voltage drop occurs in the distribution line as well as the transformer terminals. When an induction generator is applied to a hydroelectric power plant, it is necessary to accurately consider substations, distribution lines, loads, transformers for generators, and induction generators to know whether the voltage drop allowed by the system is satisfied. In this paper, we analyzed the application of induction generators with accurate data on substations, underground and overhead distribution lines, customer loads, transformers, and generators for power plants currently installed and operating in Korea. As a result of the analysis, the voltage drop was high in the premises of the power plant, but the voltage drop was low in the distribution line, confirming that it is possible to apply the induction generator to the distribution line.

오프셋 전압 인가 방식의 펄스 크기 변조 풀브릿지 다이오드 클램프 3레벨 LLC 공진형 컨버터 Pulse-Amplitude-Modulation Full-Bridge Diode-Clamped Three-Level LLC Resonant Converter With Offset Voltage Injection Method


송민섭(Min-Sup Song) ; 정호성(Hosung Jung) ; 김형철(Hyungchul Kim) ; 김재원(Jaewon Kim) ; 조규정(Gyu-Jung Cho) ; 조환희(Hwan-Hee Cho)

In this paper, a full-bridge diode-clamped three-level LLC resonant converter with pulse-amplitude-modulation (PAM) suitable for a high-input voltage system has been studied. Unlike the existing frequency sweep converter, the proposed PAM method controls the output voltage by modulating the magnitude of fundamental component of leg voltage input to the resonance tank while fixing the operating frequency at the resonance point. In addition, the PWM switching pattern was easily implemented by applying an offset voltage while reducing the switching loss by using the ±180° discontinuous PWM (DPWM) technique. In this study, the control algorithm and operation characteristics were analyzed in detail, and the feasibility was confirmed through simulation.