• 한국컴퓨터정보학회논문지
• /
• 제28권4호
• /
• pp.1-11
• /
• 2023

본 논문은 순환 신경망 대신 합성곱 신경망을 사용하여 시계열 데이터 분류 성능을 분석한다. TSC(Time Series Community)에는 GAF(Gramian Angular Field), MTF(Markov Transition Field), RP(Recurrence Plot)와 같은 전통적인 시계열 데이터 이미지화 알고리즘들이 있다. 실험은 이미지화 알고리즘들에 필요한 하이퍼 파라미터들을 조정하면서 합성곱 신경망의 성능을 평가하는 방식으로 진행된다. UCR 아카이브의 GunPoint 데이터셋을 기준으로 성능을 평가했을 때, 본 논문에서 제안하는 STFT(Short Time Fourier Transform) 알고리즘이 최적화된 하이퍼 파라미터를 찾은 경우, 기존의 알고리즘들 대비 정확도가 높고, 동적으로 feature map 이미지의 크기도 조절가능하다는 장점이 있다. GAF 또한 98~99%의 높은 정확도를 보이지만, feature map 이미지의 크기를 동적으로 조절할 수 없어 크다는 단점이 존재한다.

• 자원환경지질
• /
• 제50권4호
• /
• pp.267-276
• /
• 2017

전라남도는 연안지역은 농업활동과 상수도의 미보급으로 인하여 지하수에 크게 의존하고 있다. 지하수의 과다사용은 지하수위 저하를 일으키며 그로 인한 해수침투가 발생할 가능성이 매우 높다. 따라서 지하수 사용에 따른 해수침투 관리가 매우 필요한 지역이다. 전라남도 무안군의 연안암반대수층에서 측정된 EC 자료를 이용하여 해안가 대수층에 적합한 시계열 모형을 구축하고, 해수침투의 지표인 EC를 예측하고자 시계열 분석을 수행하였다. 1년 이상 측정한 EC 시계열 자료는 짧은 주기적인 변동과 함께 추세적으로 증가하는 비정상 시계열의 특성을 보였다. 시계열 분석을 통해 시계열 모형 식별 결과 ARIMA 모형과 계절적인 요인을 고려 할 수 있는 SARIMA 모형 이 적합한 것으로 나타났다. 하지만 두 모형 적용한 결과, EC의 주기적인 변동으로 인해 ARIMA보다는 EC 자료의 변동 특성을 잘 반영한 SARIMA 모형이 예측에 있어서 유리한 것으로 나타났다. 위와 같이 시계열 분석은 암반 대수층에서 해수침투로 인한 EC의 변화를 예측하는데 있어 유용한 것으로 나타났다.

• 정보처리학회논문지:소프트웨어 및 데이터공학
• /
• 제11권2호
• /
• pp.81-86
• /
• 2022

시계열 데이터는 주식, IoT, 공장 자동화와 같은 다양한 실생활에서 수집되고 활용되고 있으며, 정확한 시계열 예측은 해당 분야에서 운영 효율성을 높일 수 있어서 전통적으로 중요한 연구 주제이다. 전반적인 시계열 데이터의 향상된 특징을 추출할 수 있는 대표적인 시계열 데이터 분석 방법인 다층 수평 예측은 최근 부가적 정보를 포함하는 시계열 데이터에 내재한 이질성(heterogeneity)까지 포괄적으로 분석에 활용하여 향상된 시계열 예측한다. 하지만 대부분의 심층 학습 기반 시계열 분석 모델들은 시계열 데이터의 이질성을 반영하지 못했다. 따라서 우리는 잘 알려진 temporal fusion transformers 방법을 사용하여 실생활과 밀접한 실제 데이터를 이질성을 고려한 다층 수평 예측에 적용하였다. 결과적으로 주식, 미세먼지, 전기 소비량과 같은 실생활 시계열 데이터에 적용한 방법이 기존 예측 모델보다 향상된 정확도를 가짐을 확인할 수 있었다.

• 대한전기학회논문지:전력기술부문A
• /
• 제48권2호
• /
• pp.168-175
• /
• 1999

In this paper, a modified RBF(Radial Basis Function) network structure is suggested for the prediction of a time-series with non-linear, non-stationary characteristics. Coventional RBF network predicting time series by using past outputs sense the trajectory of the time series and react when there exists strong relation between input and hidden activation function's RBF center. But this response is highly sensitive to level and trend of time serieses. In order to overcome such dependencies, hidden activation functions are modified to react to the increments of input variable and multiplied by increment(or dectement) for prediction. When the suggested structure is applied to prediction of Macyey-Glass chaotic time series, Lorenz equation, and Rossler equation, improved performances are obtained.

• 한국태양에너지학회 논문집
• /
• 제37권5호
• /
• pp.77-84
• /
• 2017

The purpose of this paper is to propose the way of computing conduction time series factors (CTSF) using numerical method. After the accuracy of the numerical solution procedure being verified, the method is applied to the wall type 24 and roof type 14 of ASHARE to find the conduction time series coefficients, so called conduction time series factors. The results agree well with the values presented in the ASHRAE handbook. The method proposed can be easily applied to find unknown CTSF for more complex structures. It provides information about the temperature changes at a given location and time, thus validity of generated CTSF can be checked easily.

• 대한원격탐사학회:학술대회논문집
• /
• 대한원격탐사학회 2005년도 Proceedings of ISRS 2005
• /
• pp.294-297
• /
• 2005

Time series are comprehensively appeared and developed in many applications, ranging from science and technology to business and entertainrilent. Similarity search under time warping has attracted much interest between the time series in the large sequence databases. DTW (Dynamic Time Warping) is a robust distance measure and is superior to Euclidean distance for time series, allowing similarity matching although one of the sequences can elastic shift along the time axis. Nevertheless, it is more unfortunate that DTW has a quadratic time. Simultaneously the false dismissals are come forth since DTW distance does not satisfy the triangular inequality. In this paper, we propose an efficient range query algorithmbased on a new similarity search method under time warping. When our range query applies for this method, it can remove the significant non-qualify time series as early as possible before computing the accuracy DTW distance. Hence, it speeds up the calculation time and reduces the number of scanning the time series. Guaranteeing no false dismissals, the lower bounding function is advised that consistently underestimate the DTW distance and satisfy the triangular inequality. Through the experimental result, our range query algorithm outperforms the existing others.

• International Journal of Contents
• /
• 제10권1호
• /
• pp.47-53
• /
• 2014

The paper considers the challenges involved in measuring the similarities between time series, such as time shifts and the mixture of frequencies. To improve recognition accuracy, we investigate an improved linear dynamical system for discovering prominent features by exploiting the evolving dynamics and correlations in a time series, as the quality of unsupervised pattern recognition relies strongly on the extracted features. The proposed approach yields a set of compact extracted features that boosts the accuracy and reliability of clustering for time series data. Experimental evaluations are carried out on time series applications from the scientific, socio-economic, and business domains. The results show that our method exhibits improved clustering performance compared to conventional methods. In addition, the computation time of the proposed approach increases linearly with the length of the time series.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• 제18권1호
• /
• pp.1-14
• /
• 2024

Time-series forecasting is extensively used in the actual world. Recent research has shown that Transformers with a self-attention mechanism at their core exhibit better performance when dealing with such problems. However, most of the existing Transformer models used for time series prediction use the traditional encoder-decoder architecture, which is complex and leads to low model processing efficiency, thus limiting the ability to mine deep time dependencies by increasing model depth. Secondly, the secondary computational complexity of the self-attention mechanism also increases computational overhead and reduces processing efficiency. To address these issues, the paper designs an efficient multi-layer attention-based time-series forecasting model. This model has the following characteristics: (i) It abandons the traditional encoder-decoder based Transformer architecture and constructs a time series prediction model based on multi-layer attention mechanism, improving the model's ability to mine deep time dependencies. (ii) A cross attention module based on cross attention mechanism was designed to enhance information exchange between historical and predictive sequences. (iii) Applying a recently proposed sparse attention mechanism to our model reduces computational overhead and improves processing efficiency. Experiments on multiple datasets have shown that our model can significantly increase the performance of current advanced Transformer methods in time series forecasting, including LogTrans, Reformer, and Informer.

• 한국컴퓨터정보학회논문지
• /
• 제20권2호
• /
• pp.113-120
• /
• 2015

Technology analysis is to analyze technological data such as patent and paper for a given technology field. From the results of technology analysis, we can get novel knowledge for R&D planing and management. For the technology analysis, we can use diverse methods of statistics. Time series analysis is one of efficient approaches for technology analysis, because most technologies have researched and developed depended on time. So many technological data are time series. Time series data are occurred through time. In this paper, we propose a methodology of technology forecasting using the dynamic time warping (DTW) of time series analysis. To illustrate how to apply our methodology to real problem, we perform a case study of patent documents in target technology field. This research will contribute to R&D planning and technology management.

• 대한전기학회논문지:시스템및제어부문D
• /
• 제50권5호
• /
• pp.233-240
• /
• 2001

This paper presents a time series prediction method using a fuzzy rule-based system. Extracting fuzzy rules by performing a simple one-pass operation on the training data is quite attractive because it is easy to understand, verify, and extend. The simplest method is probably to relate an estimate, x(n+k), with past data such as x(n), x(n-1), ..x(n-m), where k and m are prefixed positive integers. The relation is represented by fuzzy if-then rules, where the past data stand for premise part and the predicted value for consequence part. However, a serious problem of the method is that it cannot handle nonstationary data whose long-term mean is varying. To cope with this, a new training method is proposed, which utilizes the difference of consecutive data in a time series. In this paper, typical previous works relating time series prediction are briefly surveyed and a new method is proposed to overcome the difficulty of prediction nonstationary data. Finally, computer simulations are illustrated to show the improved results for various time series.