• Proceedings of the Korea Inteligent Information System Society Conference
• /
• 1999.03a
• /
• pp.271-280
• /
• 1999

인공신경망을 시계열예측에 적용하는 경우에 고려되어야 할 문제중, 특히 모형에 적합한 입력변수의 생성이 중요시되고 있는데, 이러한 분야는 인공신경망의 모형생성과정에서 입력변수에 대한 전처리기법으로써 다양하게 제시되어 왔다. 가장 최근의 입력변수 전처리기법으로써 제시되고 있는 신호처리기법은 전통적 주기분할처리방법인 푸리에변환기법(Fourier transforms)을 비롯하여 이를 확장시킨 개념인 웨이블릿변환기법(wavelet transforms) 등으로 대별될 수 있다. 이는 기본적으로 시계열이 다수의 주기(cycle)들로 구성된 상이한 시계열들의 집합이라는 가정에서 출발하고 있다. 전통적으로 이러한 시계열은 전기 또는 전자공학에서 주파수영역분할, 즉 고주파 및 저주파수를 분할하기 위한 기법에 적용되어 왔다. 그러나, 최근에는 이러한 연구가 다양한 분야에 활발하게 응용되기 시작하였으며, 그 중의 대표적인 예가 바로 경영분야의 재무시계열에 대한 분석이다. 전통적으로 재무시계열은 장, 단기의사결정을 가진 시장참여자들간의 거래특성이 시계열에 각기 달리 가격으로 반영되기 때문에 이러한 상이한 집단들의 고요한 거래움직임으로 말미암아 예를 들어, 주식시장이 프랙탈구조를 가지고 있다고 보기도 한다. 이처럼 재무시계열은 다양한 사회현상의 집합체라고 볼 수 있으며, 그만큼 예측모형을 구축하는데 어려움이 따른다. 본 연구는 이러한 시계열의 주기적 특성에 기반을 둔 신호처리분석으로서 기존의 시계열로부터 노이즈를 줄여 주면서 보다 의미있는 정보로 변환시켜줄 수 있는 웨이블릿분석 방법론을 새로운 필터링기법으로 사용하여 현재 많은 연구가 진행되고 있는 인공신경망의 모형결합을 통해 기존연구과는 다른 새로운 통합예측방법론을 제시하고자 한다. 본 연구에서는 제시하는 통합방법론은 크게 2단계 과정을 거쳐 예측모형으로 완성이 된다. 즉, 1차 모형단계에서 원시 재무시계열은 먼저 웨이브릿분석을 통해서 노이즈가 필터링 되는 동시에, 과거 재무시계열의 프랙탈 구조, 즉 비선형적인 움직임을 보다 잘 반영시켜 주는 다차원 주기요소를 가지는 시계열로 분해, 생성되며, 이렇게 주기에 따라 장단기로 분할된 시계열들은 2차 모형단계에서 신경망의 새로운 입력변수로서 사용되어 최종적인 인공 신경망모델을 구축하는 데 반영된다. 기존의 주기분할방법론은 모형개발자입장에서 여러 가지 통계기준치중에서 최적의 기준치를 합리적으로 선택해야 하는 문제가 추가적으로 발생하며, 본 연구에서는 이상의 제반 문제들을 개선시키기 위해 통합방법론으로서 기존의 인공신경망모형을 구조적으로 확장시켰다. 이 모형에서 기존의 입력층 이전단계에 새로운 층이 정의된다. 이렇게 해서 생성된 새로운 통합모형은 기존모형에서 생성되는 기본적인 학습파라미터와 더불어, 본 연구에서 새롭게 제시된 주기분할층의 파라미터들이 모형의 학습성과를 높이기 위해 함께 고려된다. 한편, 이러한 학습과정에서 추가적으로 고려해야 할 파라미터 갯수가 증가함에 따라서, 본 모델의 학습성과가 local minimum에 빠지는 문제점이 발생될 수 있다. 즉, 웨이블릿분석과 인공신경망모형을 모두 전역적으로 최적화시켜야 하는 문제가 발생한다. 본 연구에서는 이 문제를 해결하기 위해서, 최근 local minimum의 가능성을 최소화하여 전역적인 학습성과를 높여 주는 인공지능기법으로서 유전자알고리즘기법을 본 연구이 통합모델에 반영하였다. 이에 대한 실증사례 분석결과는 일일 환율예측문제를 적용하였을 경우, 기존의 방법론보다 더 나운 예측성과를 타나내었다.

• The KIPS Transactions:PartD
• /
• v.13D no.7 s.110
• /
• pp.889-900
• /
• 2006

Even though sensor networks are different in user requests and data characteristics depending on each application area, the existing researches on stream data transmission problem focus on the performance improvement of their methods rather than considering the original characteristic of stream data. In this paper, we introduce a hierarchical or distributed sensor network architecture and data model, and then evaluate the multivariate data reduction methods suitable for user requirements and data features so as to apply reduction methods alternatively. To assess the relative performance of the proposed multivariate data reduction methods, we used the conventional techniques, such as Wavelet, HCL(Hierarchical Clustering), Sampling and SVD (Singular Value Decomposition) as well as the experimental data sets, such as multivariate time series, synthetic data and robot execution failure data. The experimental results shows that SVD and Sampling method are superior to Wavelet and HCL ia respect to the relative error ratio and execution time. Especially, since relative error ratio of each data reduction method is different according to data characteristic, it shows a good performance using the selective data reduction method for the experimental data set. The findings reported in this paper can serve as a useful guideline for sensor network application design and construction including multivariate stream data.

• Journal of Internet Computing and Services
• /
• v.25 no.1
• /
• pp.109-121
• /
• 2024

In various manufacturing processes such as textiles and automobiles, when equipment breaks down or stops, the machines do not work, which leads to time and financial losses for the company. Therefore, it is important to detect equipment abnormalities in advance so that equipment failures can be predicted and repaired before they occur. Most equipment failures are caused by bearing failures, which are essential parts of equipment, and detection bearing anomaly is the essence of PHM(Prognostics and Health Management) research. In this paper, we propose a preprocessing algorithm called SWT-SVD, which analyzes vibration signals from bearings and apply it to an anomaly transformer, one of the time series anomaly detection model networks, to implement bearing anomaly detection model. Vibration signals from the bearing manufacturing process contain noise due to the real-time generation of sensor values. To reduce noise in vibration signals, we use the Stationary Wavelet Transform to extract frequency components and perform preprocessing to extract meaningful features through the Singular Value Decomposition algorithm. For experimental validation of the proposed SWT-SVD preprocessing method in the bearing anomaly detection model, we utilize the PHM-2012-Challenge dataset provided by the IEEE PHM Conference. The experimental results demonstrate significant performance with an accuracy of 0.98 and an F1-Score of 0.97. Additionally, to substantiate performance improvement, we conduct a comparative analysis with previous studies, confirming that the proposed preprocessing method outperforms previous preprocessing methods in terms of performance.

• Journal of Intelligence and Information Systems
• /
• v.5 no.1
• /
• pp.103-123
• /
• 1999

Input filtering as a preprocessing method is so much crucial to get good performance in time series forecasting. There are a few preprocessing methods (i.e. ARMA outputs as time domain filters, and Fourier transform or wavelet transform as time-frequency domain filters) for handling time series. Specially, the time-frequency domain filters describe the fractal structure of financial markets better than the time domain filters due to theoretically additional frequency information. Therefore, we, first of all, try to describe and analyze specially some issues on the effectiveness of different filtering methods from viewpoint of the performance of a neural network based forecasting. And then we discuss about neural network model architecture issues, for example, what type of neural network learning architecture is selected for our time series forecasting, and what input size should be applied to a model. In this study an input selection problem is limited to a size selection of the lagged input variables. To solve this problem, we simulate on analyzing and comparing a few neural networks having different model architecture and also use an embedding dimension measure as chaotic time series analysis or nonlinear dynamic analysis to reduce the dimensionality (i.e. the size of time delayed input variables) of the models. Throughout our study, experiments for integration methods of joint time-frequency analysis and neural network techniques are applied to a case study of daily Korean won / U. S dollar exchange returns and finally we suggest an integration framework for future research from our experimental results.

• The Journal of Korean Institute of Communications and Information Sciences
• /
• v.33 no.9C
• /
• pp.720-724
• /
• 2008

Channel estimation in OFDM systems is usually carried out in frequency domain based on the least-squares (LS) method and the minimum mean-square error (MMSE) method with known pilot symbols. The LS estimator has a merit of low complexity but may suffer from the noise because it does not consider any noise effect in obtaining its solution. To enhance the noise immunity of the LS estimator, we consider estimation noise in time domain. Residual noise existing at the estimated channel coefficients in time domain could be reduced by reasonable selection of a threshold value. To achieve this, we propose a channel-estimation method based on a time-domain threshold which is a standard deviation of noise obtained by wavelet decomposition. Computer simulation shows that the estimation performance of the proposed method approaches to that of the known-channel case in terms of bit-error rates after the Viterbi decoder in overall SNRs.

• Journal of Biomedical Engineering Research
• /
• v.25 no.1
• /
• pp.5-10
• /
• 2004

Body movement is one of the important factors in sleep analysis. In this study, a thermopile detector with four channels was implemented as a non-contacting detector of body movement in sleep. Using a thermopile mathematical model and several frames of thermal images, the possibility of detecting body movement was evaluated. Instant body movement signals were evaluated for the upper, lower, and entire body using the Haar wavelet. This decomposition shows the points in time when the upper-body or lower-body movement occurred and the level of body movement. Additionally, partial body movement was decomposed in head-only, whole body, and leg-only movement using the ANFIS algorithm. Finally, three subject's data were evaluated for 60 minutes, and the detection rates of instant and partial body movement, on average, were 96.3％ and 89.2％, respectively.

• The Journal of Korean Institute of Communications and Information Sciences
• /
• v.29 no.11C
• /
• pp.1518-1526
• /
• 2004

In this paper, an algorithm using wavelet transform for detecting a cut that is a radical scene transition point, and fade and dissolve that are gradual scene transition points is proposed. The conventional methods using wavelet transform for this purpose is using features in both spatial and frequency domain. But in the proposed algorithm, the color space of an input image is converted to YUV and then luminance component Y is transformed in frequency domain using 2-level lifting. Then, the histogram of only low frequency subband that may contain some spatial domain features is compared with the previous one. Edges obtained from other higher bands can be divided into global, semi-global and local regions and the histogram of each edge region is compared. The experimental results show the performance improvement of about 17% in recall and 18% in precision and also show a good performance in fade and dissolve detection.

• Journal of the Korean Institute of Intelligent Systems
• /
• v.16 no.6
• /
• pp.691-697
• /
• 2006

Induction motors are critical components in industrial process. So there are many research in the condition based maintenance, online monitoring system, and fault detection. This paper presents a scheme on the detection and diagnosis of the three-phase squirrel induction motor under unbalanced voltage, broken rotor bar, and a combination of these two faults. Actually one fault happen in operation, it influence other component in motor or cause another faults. Accordingly it is useful to diagnose and detect a combination fault in induction motor as well as each fault. The proposed fault detection and diagnosis algorithm is based on the stator currents from the squirrel induction motor and simulated with the aid of Matlab Simulink.

• Journal of Broadcast Engineering
• /
• v.24 no.5
• /
• pp.791-801
• /
• 2019

The serial arc is one of factors causing electrical fires. Over past decades, various researches have been carried out to detect arc occurrences. Even though frequency analysis, wavelet and statistical features have been used, arc detection performance is degraded due to diverse arc waveforms. Therefore, there is a need to develop a method that could increase the feature dimension, thereby improving the detection performance. In this paper, we use variational mode decomposition (VMD) to obtain multiple decomposed signals and then extract statistical features from them. The features from VMD outperform those from no-VMD in terms of detection performance. Further, artificial neural network is employed as an arc classifier. Experiments validated that the use of VMD improves the classification accuracy by up to 4 percent, based on 14,000 training data.

• Journal of the Korea Society of Computer and Information
• /
• v.11 no.4 s.42
• /
• pp.197-206
• /
• 2006

One of the most significant characteristics of JPEG2000. the emerging still image standards. is the ROI (Region of Interest) coding. JPEG2000 provides a number of ROI coding mechanisms and ROI parameters. To apply them to an application, it must select the applicable values. In this paper, we evaluate how the ROI coding mechanisms and the ROI parameters influencing JPEG2000 qualify affect the ROI quality and the whole image quality. The ROI coding mechanisms are Maxshift and Implicit. and the parameters are tile size and ROI size, codeblock size, number of DWT decomposition levels and ROI importance. The bigger the tile size, the better the quality. The bigger the ROI size, the ROI importance and the number of DWT decomposition levels, the worse the qualify. In code block $32{\times}32$ of Maxshift and Implicit, it has the best qualify.