• The Transactions of the Korean Institute of Power Electronics
• /
• v.27 no.4
• /
• pp.316-324
• /
• 2022

This paper uses seasonal auto-regressive integrated moving average (S-ARIMA), which is efficient in seasonality between time-series models, to predict the degradation tendency for lithium-ion batteries and study a method for improving the predictive performance. The proposed method analyzes the degradation tendency and extracted factors through an electrical characteristic experiment of lithium-ion batteries, and verifies whether time-series data are suitable for the S-ARIMA model through several statistical analysis techniques. Finally, prediction of battery aging is performed through S-ARIMA, and performance of the model is verified through error comparison of predictions through mean absolute error.

• The KIPS Transactions:PartB
• /
• v.9B no.3
• /
• pp.277-286
• /
• 2002

Genetic programming (GP) has been combined with quantum mechanical perturbation theory to make a new algorithm to construct mathematical models and perform predictions for chaotic time series from real world. Procedural similarities between time series modeling and perturbation theory to solve quantum mechanical wave equations are discussed, and the exemplary GP approach for implementing them is proposed. The approach is based on multiple populations and uses orthogonal functions for GP function set. GP is applied to original time series to get the first mathematical model. Numerical values of the model are subtracted from the original time series data to form a residual time series which is again subject to GP modeling procedure. The process is repeated until predetermined terminating conditions are met. The algorithm has been successfully applied to construct highly effective mathematical models for many real world chaotic time series. Comparisons with other methodologies and topics for further study are also introduced.

• Proceedings of the KIPE Conference
• /
• 2000.07a
• /
• pp.735-738
• /
• 2000

In this paper we propose a series active power filter and a simple calculation method acquiring the reference voltage. A series active power filter is suitable to suppress harmonics produced by voltage type harmonic source such as a diode rectifier with filter capacitor on the DC side The proposed series active power filter system is applied to 3-phase 3-wire power system including the voltage type harmonic source. Experimental result obtained from a laboratory model are shown to verify the viability and effectiveness of the proposed system.

• Journal of Korean Institute of Industrial Engineers
• /
• v.36 no.2
• /
• pp.101-107
• /
• 2010

Among a variety of asset dynamics models in order to explain the common properties of financial underlying assets, parametric models are meaningful when their parameters are set reliably. There are two main methods from which we can obtain them. They are to use time-series data of an underlying price or the market option prices of the underlying at one time. Based on the Girsanov theorem, in the pure diffusion models, the parameters calibrated from the option prices should be partially equivalent to those from time-series underling prices. We call this phenomenon model consistency. In this paper, we verify that the two-state regime switching Black-Scholes model is superior in the sense of model consistency, comparing with two popular conventional models, the Black-Scholes model and Heston model.

• The Korean Journal of Applied Statistics
• /
• v.28 no.5
• /
• pp.1013-1024
• /
• 2015

We compare the performance of two representative estimation methods for the dynamic conditional correlation (DCC) GARCH model. The first method is the pairwise estimation which exploits partial information from the paired series, irrespective to the time series dimension. The second is the multi-dimensional estimation that uses full information of the time series. As a simulation for the comparison, we generate a multivariate time series similar to those observed in real markets and construct a DCC GARCH model. As an empirical example, we constitute various portfolios using real KOSPI 200 sector indices and estimate volatility and VaR of the portfolios. Through the estimated dynamic correlations from the simulation and the estimated volatility and value at risk (VaR) of the portfolios, we evaluate the performance of the estimations. We observe that the multi-dimensional estimation tends to be superior to pairwise estimation; in addition, relatively-uncorrelated series can improve the performance of the multi-dimensional estimation.

• The KIPS Transactions:PartB
• /
• v.9B no.4
• /
• pp.421-428
• /
• 2002

The local prediction method utilizing embedding vector loses the prediction power when the parameter r estimation is not exact for predicting the chaos time series induced from the high order differential equation. In spite of the fact that there have been a lot of suggestions regarding how to estimate the delay time ($\tau$), no specific method is proposed to apply to any time series. The inclinded linear model, which utilizes inclinded netter, yields satisfying degree of prediction power without estimating exact delay time ($\tau$). The usefulness of this approach has been indicated not only theoretically but also in practical situation when the method w8s applied to economical time series analysis.

• Water for future
• /
• v.25 no.2
• /
• pp.89-97
• /
• 1992

This study is to verify the applicability of statistical models for predicting flood frequency at the stage gaging stations selected by considering whether the flow is natural condition in the Han River basin. From the result of verification, this statistical flood frequency models showed that is fairly reasonable to apply in practice, and also were compared with sampling variance to calibrate the statistical dfficiency of the estimate of the T year flood Q(T) by two different flood frequency models. As a result, it was showed that for return periods greater than about T=10 years the annual exceedence series estimate of Q(T) has smaller sampling variance than the annual maximum series estimate. It was showed that for the range of return periods the partial duration series estimate of Q(T) has smaller sampling varianed than the annual maximum series estimate only if the POT model contains at least 2N(N:record length)items or more in order to estimate Q(T) more efficiently than the ANNMAX model.

• Korean Chemical Engineering Research
• /
• v.59 no.4
• /
• pp.532-541
• /
• 2021

Recently, research on the application of artificial intelligence in the chemical process has been increasing rapidly. However, overfitting is a significant problem that prevents the model from being generalized well to predict unseen data on test data, as well as observed training data. Cross validation is one of the ways to solve the overfitting problem. In this study, the time-series cross validation method was applied to optimize the number of batch and epoch in the hyperparameters of the prediction model for the 2,3-BDO distillation process, and it compared with K-fold cross validation generally used. As a result, the RMSE of the model with time-series cross validation was lower by 9.06%, and the MAPE was higher by 0.61% than the model with K-fold cross validation. Also, the calculation time was 198.29 sec less than the K-fold cross validation method.

• Journal of the Korea Society of Computer and Information
• /
• v.27 no.5
• /
• pp.55-67
• /
• 2022

Sensor data can provide fault diagnosis for equipment. However, the cause analysis for fault results of equipment is not often provided. In this study, we propose an explainable convolutional neural network framework for the sensor-based time series classification model. We used sensor-based time series dataset, acquired from vehicles equipped with sensors, and the Wafer dataset, acquired from manufacturing process. Moreover, we used Cycle Signal dataset, acquired from real world mechanical equipment, and for Data augmentation methods, scaling and jittering were used to train our deep learning models. In addition, our proposed classification models are convolutional neural network based models, FCN, 1D-CNN, and ResNet, to compare evaluations for each model. Our experimental results show that the ResNet provides promising results in the context of time series classification with accuracy and F1 Score reaching 95%, improved by 3% compared to the previous study. Furthermore, we propose XAI methods, Class Activation Map and Layer Visualization, to interpret the experiment result. XAI methods can visualize the time series interval that shows important factors for sensor data classification.

• Journal of the Korea Society of Computer and Information
• /
• v.27 no.11
• /
• pp.19-27
• /
• 2022

In this study, we propose a language and platform to describe and manage the MLOps(Machine Learning Operations) workflow for time series data anomaly detection. Time series data is collected in many fields, such as IoT sensors, system performance indicators, and user access. In addition, it is used in many applications such as system monitoring and anomaly detection. In order to perform prediction and anomaly detection of time series data, the MLOps platform that can quickly and flexibly apply the analyzed model to the production environment is required. Thus, we developed Python-based AI/ML Modeling Language (AMML) to easily configure and execute MLOps workflows. Python is widely used in data analysis. The proposed MLOps platform can extract and preprocess time series data from various data sources (R-DB, NoSql DB, Log File, etc.) using AMML and predict it through a deep learning model. To verify the applicability of AMML, the workflow for generating a transformer oil temperature prediction deep learning model was configured with AMML and it was confirmed that the training was performed normally.