• 스마트미디어저널
• /
• 제10권3호
• /
• pp.23-30
• /
• 2021

제조 시계열 데이터 클러스터링 기법은 제조 대용량 데이터 기반 군집화를 통한 설비 및 공정 이상 탐지 분류를 위한 중요한 솔루션이지만 기존 정적 데이터 대상 클러스터링 기법을 시계열 데이터에 적용함에 있어 낮은 정확도를 가지는 단점이 있다. 본 논문에서는 진화 연산 기반 시계열 군집 분석 접근 방식을 제시하여 기존 클러스터링 기술에 대한 정합성 향상하고자 한다. 이를 위하여 먼저 제조 공정 결과 이미지 형상을 선형 스캐닝을 활용하여 1차원 시계열 데이터로 변환하고 해당 변환 데이터 대상으로 Pearson 거리 매트릭을 기반으로 계층적 군집 분석 및 분할 군집 분석에 대한 최적 하위클러스터를 도출한다. 해당 최적 하위클러스터 대상 유전 알고리즘을 활용하여 유사도가 최소화되는 최적의 군집 조합을 도출한다. 그리고 실제 제조 과정 이미지 대상으로 기존 클러스터링 기법과 성능 비교를 통하여 제안된 클러스터링 기법의 성능 우수성을 검증한다.

• 대한토목학회논문집
• /
• 제43권3호
• /
• pp.363-370
• /
• 2023

COVID-19 (Coronavirus disease 2019) 확산으로 2020년 초부터 도시철도 등 대중교통수단의 이용량이 크게 변동하였다. 이에 본 연구에서는 COVID-19 이전과 COVID-19 확산 이후, 3년 동안 도시철도 역별 일별 시계열 자료를 수집하여 DTW (Dynamic Time Warping) 거리법을 통해 시계열 군집분석 유사도를 평가하여 군집 별 회귀 중앙치를 도출하고, COVID-19 등 여러 외부 사건이 이용객 수의 변동에 미치는 영향을 시계열 충격 탐지 함수(Outlier Detection)로 진단하였다. 또한 도시철도 역의 군집 별 이용 특성을 분석하고 또한 외부 충격에 따른 승객량의 변동을 파악하였다. 향후 COVID-19 재확산 시 이용량의 유지와 회복에 대한 방안을 검토하는 데 목적을 두었다.

• 대한환경공학회지
• /
• 제31권6호
• /
• pp.442-448
• /
• 2009

예혼합 분무화염의 유적군 연소 기구를 평가하기 위하여 레이저 가시화법, MICRO (multi-color integrated Cassegrain receiving optics) 및 PDA (phase Doppler anemometer) 광학계측 시스템을 이용하여 동시 시계열 계측을 실시하였다. 또한 유적군의 군연소수를 실험적으로 산출하여 이론해석에 적용하였을 경우 실제로 관찰되는 군연소 형태와 일치하는지에 대하여 검토하였다. 유적군 단면화상에 의해 확인된 모든 유적군에 대하여 실험적으로 군연소수 $G_c$를 산출한 결과 주로 내부 군연소와 외부 군연소로 분류되었으며 이론해석과 일치하였다. 또한 실제 관찰된 군연소 형태와 그 유적군의 군연소수를 이론해석에 적용한 경우에 군연소 형태가 일치하는 경우와 일치하지 않는 경우가 있었다. 일치하지 않는 원인은 군연소수를 유적의 기하학적 배치만으로 결정한 것이나 현상의 3차원성이 측정 결과에 영향을 미친 것이라고 생각되어진다.

• 한국컴퓨터정보학회논문지
• /
• 제23권5호
• /
• pp.9-14
• /
• 2018

For an efficient management of electricity market and power systems, accurate forecasts for electricity demand are essential. Since there are many factors, either known or unknown, determining the realized loads, it is difficult to forecast the demands with the past time series only. In this paper we perform a cluster analysis on electricity demand data collected from Jan. 2000 to Dec. 2017. Our purpose of clustering on electricity demand data is that each cluster is expected to consist of data whose latent variables are same or similar values. Then, if properly clustered, it is possible to develop an accurate forecasting model for each cluster separately. To validate the feasibility of this approach for building better forecasting models, we clustered data with t-SNE. To apply t-SNE to time series data effectively, we adopt the dynamic time warping as a similarity measure. From the result of experiments, we found that several clusters are well observed and each cluster can be interpreted as a mix of well-known factors such as trends, seasonality and holiday effects and other unknown factors. These findings can motivate the approaches which build forecasting models with respect to each cluster independently.

• 한국측량학회지
• /
• 제35권5호
• /
• pp.329-338
• /
• 2017

This study attempts to classify 1008 alley markets in Seoul through cluster analysis using Dynamic Time Warping, one of the methods used to analyze the similarity of time series, and evaluate the possibility of opening new stores. The sequence of the gross sales of an alley market and that of gross sales per store stand for the potential of growth and profitability of the market, respectively and are used as variables for cluster analysis. Five clusters are obtained for the gross sales and four clusters for the gross sales per store. These two types of clusters are again classified as rising and falling trends, respectively, and the combination of these trends produces four categories. These categories are used to evaluate the possibility of opening new stores in alley markets. The results show that the southeast which is relatively wealthy inferior to other regions in opening new stores. Alley markets in the northeast and the southwest are better than other regions such that opening a new store is justified. In the northwest, there are many markets with trend of gross sales and that of gross sales per store moving in opposite directions, and new store openings in these markets should be postponed.

• Communications for Statistical Applications and Methods
• /
• 제27권6호
• /
• pp.589-602
• /
• 2020

The development of smart grids has enabled the easy collection of a large amount of power data. There are some common patterns that make it useful to cluster power consumption patterns when analyzing s power big data. In this paper, clustering analysis is based on distance functions for time series and clustering algorithms to discover patterns for power consumption data. In clustering, we use 10 distance measures to find the clusters that consider the characteristics of time series data. A simulation study is done to compare the distance measures for clustering. Cluster validity measures are also calculated and compared such as error rate, similarity index, Dunn index and silhouette values. Real power consumption data are used for clustering, with five distance measures whose performances are better than others in the simulation.

• Communications for Statistical Applications and Methods
• /
• 제29권1호
• /
• pp.103-125
• /
• 2022

In this paper, we analyze the time series data of the case and death counts of COVID-19 that broke out in China in December, 2019. The study period is during the lockdown of Wuhan. We exploit functional data analysis methods to analyze the collected time series data. The analysis is divided into three parts. First, the functional principal component analysis is conducted to investigate the modes of variation. Second, we carry out the functional canonical correlation analysis to explore the relationship between confirmed and death cases. Finally, we utilize a clustering method based on the Expectation-Maximization (EM) algorithm to run the cluster analysis on the counts of confirmed cases, where the number of clusters is determined via a cross-validation approach. Besides, we compare the clustering results with some migration data available to the public.

• 응용통계연구
• /
• 제32권1호
• /
• pp.149-160
• /
• 2019

본 연구는 2016년 7월부터 2017년 6월까지 인천 소재 A 대학교의 15분 단위의 일일 전기 사용량 시계열 데이터에 대해 functional data analysis 기법을 적용하여 군집화하고 각 군집의 특성을 파악하고 예측에 활용하고자 한다. 하루동안의 A 대학교의 전기 사용량은 패턴은 주중과 주말 에 큰 차이를 보이며 스플라인 기저함수로 FPCA 구한 후 이들에 대한 가우시안 분포의 혼합모형 기반 군집분석으로 3개의 군집화가 적절해 보인다. 각 군집에 대해 평균 함수, 확률밀도함수, 일들의 분포 등을 정리해 각 군집에 대한 정보와 특징을 보여준다.

• Smart Structures and Systems
• /
• 제26권1호
• /
• pp.23-33
• /
• 2020

Majority of the damage in engineering structures is nonlinear. Damage sensitive features (DSFs) extracted by traditional methods from linear time series models cannot effectively handle nonlinearity induced by structural damage. A new DSF is proposed based on vector space cosine similarity (VSCS), which combines K-means cluster analysis and Bayesian discrimination to detect nonlinear structural damage. A reference autoregressive moving average (ARMA) model is built based on measured acceleration data. This study first considers an existing DSF, residual standard deviation (RSD). The DSF is further advanced using the VSCS, and then the advanced VSCS is classified using K-means cluster analysis and Bayes discriminant analysis, respectively. The performance of the proposed approach is then verified using experimental data from a three-story shear building structure, and compared with the results of existing RSD. It is demonstrated that combining the linear ARMA model and the advanced VSCS, with cluster analysis and Bayes discriminant analysis, respectively, is an effective approach for detection of nonlinear damage. This approach improves the reliability and accuracy of the nonlinear damage detection using the linear model and significantly reduces the computational cost. The results indicate that the proposed approach is potential to be a promising damage detection technique.

• 한국정보시스템학회지:정보시스템연구
• /
• 제30권2호
• /
• pp.147-163
• /
• 2021

Purpose The study aims to predict real estate prices by utilizing regional characteristics. Since real estate has the characteristic of immobility, the characteristics of a region have a great influence on the price of real estate. In addition, real estate prices are closely related to economic development and are a major concern for policy makers and investors. Accurate house price forecasting is necessary to prepare for the impact of house price fluctuations. To improve the performance of our predictive models, we applied LSTM, a widely used deep learning technique for predicting time series data. Design/methodology/approach This study used time series data on real estate prices provided by the Ministry of Land, Infrastructure and Transport. For time series data preprocessing, HP filters were applied to decompose trends and SOM was used to cluster regions with similar price directions. To build a real estate price prediction model, SVR and LSTM were applied, and the prices of regions classified into similar clusters by SOM were used as input variables. Findings The clustering results showed that the region of the same cluster was geographically close, and it was possible to confirm the characteristics of being classified as the same cluster even if there was a price level and a similar industry group. As a result of predicting real estate prices in 1, 2, and 3 months, LSTM showed better predictive performance than SVR, and LSTM showed better predictive performance in long-term forecasting 3 months later than in 1-month short-term forecasting.