• 한국컴퓨터정보학회논문지
• /
• 제27권11호
• /
• pp.19-27
• /
• 2022

본 연구에서는 시계열 데이터 이상 탐지 수행을 위한 MLOps(Machine Learning Operations) 워크플로를 기술하고 관리할 수 있는 언어와 플랫폼을 제안한다. 시계열 데이터는 IoT 센서, 시스템 성능 지표, 사용자 접속량 등 많은 분야에서 수집되고 있다. 또한, 시스템 모니터링 및 이상 탐지 등 많은 응용 분야에 활용 중이다. 시계열 데이터의 예측 및 이상 탐지를 수행하기 위해서는 분석된 모델을 빠르고 유연하게 운영 환경에 적용할 수 있는 MLOps 플랫폼이 필요하다. 이에, 최근 데이터 분석에 많이 활용되고 있는 Python 기반의 AMML(AI/ML Modeling Language)을 개발하여 손쉽게 MLOps 워크플로를 구성하고 실행할 수 있도록 제안한다. 제안하는 AI MLOps 플랫폼은 AMML을 이용하여 다양한 데이터 소스(R-DB, NoSql DB, Log File 등)에서 시계열 데이터를 추출, 전처리 및 예측을 수행할 수 있다. AMML의 적용 가능성을 검증하기 위해, 변압기 오일 온도 예측 딥러닝 모델을 생성하는 워크플로를 AMML로 구성하고 학습이 정상적으로 수행됨을 확인하였다.

• Journal of Information Processing Systems
• /
• 제14권6호
• /
• pp.1508-1520
• /
• 2018

In this paper, we propose an improved model to provide users with a better long-term prediction of waterworks operation data. The existing prediction models have been studied in various types of models such as multiple linear regression model while considering time, days and seasonal characteristics. But the existing model shows the rate of prediction for demand fluctuation and long-term prediction is insufficient. Particularly in the deep running model, the long-short-term memory (LSTM) model has been applied to predict data of water purification plant because its time series prediction is highly reliable. However, it is necessary to reflect the correlation among various related factors, and a supplementary model is needed to improve the long-term predictability. In this paper, convolutional neural network (CNN) model is introduced to select various input variables that have a necessary correlation and to improve long term prediction rate, thus increasing the prediction rate through the LSTM predictive value and the combined structure. In addition, a multiple linear regression model is applied to compile the predicted data of CNN and LSTM, which then confirms the data as the final predicted outcome.

• Journal of the Korean Statistical Society
• /
• 제17권2호
• /
• pp.101-120
• /
• 1988

A time series model with Laplacian (double-exponential) marginal distribution, NLAR(2), was proposed by Dewald and Lewis (1985). The special cases of NLAR(2) process and their properties are considered. Extensions to the NLAR(p) is discussed. It is shown that the NLAR(1) satisfies the strong-mixing conditions, hence the model-free prediction interval using the sample quantiles can be obtained.

• 통합자연과학논문집
• /
• 제7권1호
• /
• pp.67-73
• /
• 2014

We consider the mean value function for NHPP software reliability model and time series regression model in software failure data. We estimate parameters for the proposed models from two data sets. The values of SSE and MSE is presented from two data sets. We compare the predicted number of faults with the actual two data sets using the mean value function and regression curve.

• 한국병원경영학회지
• /
• 제11권3호
• /
• pp.19-41
• /
• 2006

The Objective of this study was to design the model which predict the future cash flow of hospitals and on the basis of designed model to support sound hospital management by the prediction of future cash flow. The five cash flow measurement variables discussed in financial accrual part were used as variables and these variables were defined as NI, NIDPR, CFO, CFAI, CC. To measure the cash flow B/S related variables, P/L related variables and financial ratio related variables were utilized in this study. To measure cash flow models were designed and to estimate the prediction ability of five cash flow models, the martingale model and the market model were utilized. To estimate relative prediction outcome of cash flow prediction model and simple market model, MAE and MER were used to compare and analyze relative prediction ability of the cash flow model and the market model and to prove superiority of the model of the cash flow prediction model, 32 Regional Public Hospital's cross-section data and 4 year time series data were combined and pooled cross-sectional time series regression model was used for GLS-analysis. To analyze this data, Firstly, each cash flow prediction model, martingale model and market model were made and MAE and MER were estimated. Secondly difference-test was conducted to find the difference between MAE and MER of cash flow prediction model. Thirdly after ranking by size the prediction of cash flow model, martingale model and market model, Friedman-test was evaluated to find prediction ability. The results of this study were as follows: when t-test was conducted to find prediction ability among each model, the error of prediction of cash flow model was smaller than that of martingale and market model, and the difference of prediction error cash flow was significant, so cash flow model was analyzed as excellent compare with other models. This research results can be considered conductive in that present the suitable prediction model of future cash flow to the hospital. This research can provide valuable information in policy-making of hospital's policy decision. This research provide effects as follows; (1) the research is useful to estimate the benefit of hospital, solvency and capital supply ability for substitution of fixed equipment. (2) the research is useful to estimate hospital's liqudity, solvency and financial ability. (3) the research is useful to estimate evaluation ability in hospital management. Furthermore, the research should be continued by sampling all hospitals and constructed advanced cash flow model in dimension, established type and continued by studying unified model which is related each cash flow model.

• 대한전기학회논문지:시스템및제어부문D
• /
• 제51권1호
• /
• pp.39-42
• /
• 2002

For the purpose of modeling EEG signal which has nonstationary and nonlinear dynamic characteristics, this paper propose a state feedback real time recurrent neural network model. The state feedback real time recurrent neural network is structured to have memory structure in the state of hidden layers so that it has arbitrary dynamics and ability to deal with time-varying input through its own temporal operation. For the model test, Mackey-Glass time series is used as a nonlinear dynamic system and the model is applied to the prediction of three types of EEG, alpha wave, beta wave and epileptic EEG. Experimental results show that the performance of the proposed model is better than that of other neural network models which are compared in this paper in some view points of the converging speed in learning stage and normalized mean square error for the test data set.

• 디지털융복합연구
• /
• 제11권10호
• /
• pp.305-311
• /
• 2013

경제시장의 규모가 지속적으로 발전함에 따라 올바른 의사결정을 위하여 경제시장을 정확하게 예측하는 문제가 중요한 문제로 떠오르고 있다. 현대 경제시스템을 표현하는 다양한 경제지표 중 가장 큰 축인 주식지표의 올바른 이해와 분석 그리고 의사결정문제에 적용을 위하여 시계열자료의 모델에 적합한 은닉마아코프모델과 이를 토대로 시계열자료의 시간 및 계산비용의 절감을 위한 차원감소기법들을 모델의 추정과 예측 문제에 적용하였으며 그 유효성을 확인하였다. 실험 결과, 은닉마아코프모델과 차원감소기법을 적용한 모델 모두에서 장기예측보다는 단기의 예측에서 최적의 모델 추정과 유사패턴 예측률이 모두 실제의 자료와 매우 유사함을 확인할 수 있었다.

• International journal of advanced smart convergence
• /
• 제11권1호
• /
• pp.19-27
• /
• 2022

Across the world, 'housing' comprises a significant portion of wealth and assets. For this reason, fluctuations in real estate prices are highly sensitive issues to individual households. In Korea, housing prices have steadily increased over the years, and thus many Koreans view the real estate market as an effective channel for their investments. However, if one purchases a real estate property for the purpose of investing, then there are several risks involved when prices begin to fluctuate. The purpose of this study is to design a real estate price 'return rate' prediction model to help mitigate the risks involved with real estate investments and promote reasonable real estate purchases. Various approaches are explored to develop a model capable of predicting real estate prices based on an understanding of the immovability of the real estate market. This study employs the LSTM method, which is based on artificial intelligence and deep learning, to predict real estate prices and validate the model. LSTM networks are based on recurrent neural networks (RNN) but add cell states (which act as a type of conveyer belt) to the hidden states. LSTM networks are able to obtain cell states and hidden states in a recursive manner. Data on the actual trading prices of apartments in autonomous districts between January 2006 and December 2019 are collected from the Actual Trading Price Disclosure System of the Ministry of Land, Infrastructure and Transport (MOLIT). Additionally, basic data on apartments and commercial buildings are collected from the Public Data Portal and Seoul Metropolitan Government's data portal. The collected actual trading price data are scaled to monthly average trading amounts, and each data entry is pre-processed according to address to produce 168 data entries. An LSTM model for return rate prediction is prepared based on a time series dataset where the training period is set as April 2015~August 2017 (29 months), the validation period is set as September 2017~September 2018 (13 months), and the test period is set as December 2018~December 2019 (13 months). The results of the return rate prediction study are as follows. First, the model achieved a prediction similarity level of almost 76%. After collecting time series data and preparing the final prediction model, it was confirmed that 76% of models could be achieved. All in all, the results demonstrate the reliability of the LSTM-based model for return rate prediction.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• 제18권3호
• /
• pp.528-550
• /
• 2024

Warehousing demand prediction is an essential part of the supply chain, providing a fundamental basis for product manufacturing, replenishment, warehouse planning, etc. Existing forecasting methods cannot produce accurate forecasts since warehouse demand is affected by external factors such as holidays and seasons. Some aspects, such as consumer psychology and producer reputation, are challenging to quantify. The data can fluctuate widely or do not show obvious trend cycles. We introduce a new model for warehouse demand prediction called MAGRU, which stands for Multi-layer Attention with GRU. In the model, firstly, we perform the embedding operation on the input sequence to quantify the external influences; after that, we implement an encoder using GRU and the attention mechanism. The hidden state of GRU captures essential time series. In the decoder, we use attention again to select the key hidden states among all-time slices as the data to be fed into the GRU network. Experimental results show that this model has higher accuracy than RNN, LSTM, GRU, Prophet, XGboost, and DARNN. Using mean absolute error (MAE) and symmetric mean absolute percentage error(SMAPE) to evaluate the experimental results, MAGRU's MAE, RMSE, and SMAPE decreased by 7.65%, 10.03%, and 8.87% over GRU-LSTM, the current best model for solving this type of problem.

• 유통과학연구
• /
• 제12권2호
• /
• pp.59-71
• /
• 2014

Purpose - In recent years, many firms have attempted various approaches to cope with the continual increase of aviation transportation. The previous research into freight charge forecasting models has focused on regression analyses using a few influence factors to calculate the future price. However, these approaches have limitations that make them difficult to apply into practice: They cannot respond promptly to small price changes and their predictive power is relatively low. Therefore, the current study proposes a freight charge-forecasting model using time series data instead a regression approach. The main purposes of this study can thus be summarized as follows. First, a proper model for freight charge using the autoregressive integrated moving average (ARIMA) model, which is mainly used for time series forecast, is presented. Second, a modified ARIMA model for freight charge prediction and the standard process of determining freight charge based on the model is presented. Third, a straightforward freight charge prediction model for practitioners to apply and utilize is presented. Research design, data, and methodology - To develop a new freight charge model, this study proposes the ARIMAC(p,q) model, which applies time difference constantly to address the correlation coefficient (autocorrelation function and partial autocorrelation function) problem as it appears in the ARIMA(p,q) model and materialize an error-adjusted ARIMAC(p,q). Cargo Account Settlement Systems (CASS) data from the International Air Transport Association (IATA) are used to predict the air freight charge. In the modeling, freight charge data for 72 months (from January 2006 to December 2011) are used for the training set, and a prediction interval of 23 months (from January 2012 to November 2013) is used for the validation set. The freight charge from November 2012 to November 2013 is predicted for three routes - Los Angeles, Miami, and Vienna - and the accuracy of the prediction interval is analyzed using mean absolute percentage error (MAPE). Results - The result of the proposed model shows better accuracy of prediction because the MAPE of the error-adjusted ARIMAC model is 10% and the MAPE of ARIMAC is 11.2% for the L.A. route. For the Miami route, the proposed model also shows slightly better accuracy in that the MAPE of the error-adjusted ARIMAC model is 3.5%, while that of ARIMAC is 3.7%. However, for the Vienna route, the accuracy of ARIMAC is better because the MAPE of ARIMAC is 14.5% and the MAPE of the error-adjusted ARIMAC model is 15.7%. Conclusions - The accuracy of the error-adjusted ARIMAC model appears better when a route's freight charge variance is large, and the accuracy of ARIMA is better when the freight charge variance is small or has a trend of ascent or descent. From the results, it can be concluded that the ARIMAC model, which uses moving averages, has less predictive power for small price changes, while the error-adjusted ARIMAC model, which uses error correction, has the advantage of being able to respond to price changes quickly.