• Smart Media Journal
• /
• v.9 no.4
• /
• pp.26-35
• /
• 2020

Attempts to predict future stock prices have been studied steadily since the past. However, unlike general time-series data, financial time-series data has various obstacles to making predictions such as non-stationarity, long-term dependence, and non-linearity. In addition, variables of a wide range of data have limitations in the selection by humans, and the model should be able to automatically extract variables well. In this paper, we propose a 'sliding time step normalization' method that can normalize non-stationary data and LSTM autoencoder to compress variables from all variables. and 'moving transfer learning', which divides periods and performs transfer learning. In addition, the experiment shows that the performance is superior when using as many variables as possible through the neural network rather than using only 100 major financial variables and by using 'sliding time step normalization' to normalize the non-stationarity of data in all sections, it is shown to be effective in improving performance. 'moving transfer learning' shows that it is effective in improving the performance in long test intervals by evaluating the performance of the model and performing transfer learning in the test interval for each step.

• Journal of the Korean Operations Research and Management Science Society
• /
• v.38 no.2
• /
• pp.65-73
• /
• 2013

In the simulation output analysis, bootstrap method is an applicable resampling technique to insufficient data which are not significant statistically. The moving block bootstrap, the stationary bootstrap, and the threshold bootstrap are typical bootstrap methods to be used for autocorrelated time series data. They are nonparametric methods for stationary time series data, which correctly describe the original data. In the simulation output analysis, however, we may not use them because of the non-stationarity in the data set caused by the trend such as increasing or decreasing. In these cases, we can get rid of the trend by differencing the data, which guarantees the stationarity. We can get the bootstrapped data from the differenced stationary data. Taking a reverse transform to the bootstrapped data, finally, we get the pseudo-samples for the original data. In this paper, we introduce the applicability of bootstrap methods to the time series data having trend, and then verify it through the statistical analyses.

• Journal of the Korean Institute of Landscape Architecture
• /
• v.41 no.6
• /
• pp.43-51
• /
• 2013

This study aims to examine the presence of non-stationary relationship between water quality and land use in watersheds. In investigating the relationships between land use and water quality, most previous studies adopted OLS method which is assumed stationarity. However, this approach is difficult to capture the local variation of the relationships. We used 146 sampling data and land cover data of Korean Ministry of Environment to build conventional regressions and GWR models for BOD, TN and TP. Regression model and GWR models of BOD, TN, TP were compared with $R^2$, AICc and Moran's I. The results of comparisons and descriptive statistics of GWR models strongly indicated the presence of Non-Stationarity between water quality and land use.

• The Korean Journal of Applied Statistics
• /
• v.30 no.3
• /
• pp.323-334
• /
• 2017

We note that the dynamics of realized volatilities (RVs) are near the boundary between stationarity and non-stationarity because RVs have persistent long-memory and are often subject to fairly large outlying values. To forecast realized volatility, we consider a new method that adaptively use models with and without unit root according to the abnormality of observed RV: heterogeneous autoregressive (HAR) model and the Integrated HAR (IHAR) model. The resulting method is called the IHAR-O-HAR method. In an out-of-sample forecast comparison for the realized volatility datasets of the 3 major indexes of the S&P 500, the NASDAQ, and the Nikkei 225, the new IHAR-O-HAR method is shown superior to the existing HAR and IHAR method.

• Communications for Statistical Applications and Methods
• /
• v.29 no.2
• /
• pp.225-238
• /
• 2022

In this paper, we propose a clustering method for advanced metering infrastructure (AMI) data in Korea. As AMI data presents non-stationarity, we consider time-dependent frequency domain principal components analysis, which is a proper method for locally stationary time series data. We develop a new clustering method based on time-varying eigenvectors, and our method provides a meaningful result that is different from the clustering results obtained by employing conventional methods, such as K-means and K-centres functional clustering. Simulation study demonstrates the superiority of the proposed approach. We further apply the clustering results to the evaluation of the electricity price system in South Korea, and validate the reform of the progressive electricity tariff system.

• Atmosphere
• /
• v.23 no.3
• /
• pp.307-320
• /
• 2013

We performed comprehensive quality control for eddy-covariance measurements from 3 farmland sites and 1 industrial site adjacent to Nakdong river. The quality control program is based on Foken and Wichura (1996) and Vicker and Mahrt (1997) and we added criteria for trend and standard deviation for scalar variables and modified criteria for non-stationarity condition of Foken and Wichura (1996) to consider random error of fluxes. The classification of data quality is designed for the raw data and the processed flux data, separately. Use of added criteria efficiently reduces the number of outlier for water vapor and $CO_2$ fluxes and use of modified criteria for non-stationarity reduces the number of outlier for scalar fluxes and increases the number of data with accepted quality for further work. Energy balance ratio is higher in farmlands than industrial site, which is due to neglect of heat storage term in industrial site. Among farmland sites, C4 site shows higher energy balance ratio than other sites. This is due to more homogeneous surface of C4 site than other farmland sites. However, energy balance ratio is very low or even negative at night. Mismatch between the flux footprint and the other energy component footprint over the heterogeneous surface is main cause for energy imbalance at night. Other possible causes are also discussed.

• Earthquakes and Structures
• /
• v.8 no.6
• /
• pp.1463-1480
• /
• 2015

This research aims to develop an empirical model for simulation of time-varying frequency in earthquake ground motion so as to be used easily in engineering applications. Briefly, 10545 recordings of the Next Generation Attenuation (NGA) global database of accelerograms from shallow crustal earthquakes are selected and binned by magnitude, distance and site condition. Then the wavelet spectrum of each acceleration record is calculated by using one-dimensional continuous wavelet transform, and the frequencies corresponding to the maximum values of the wavelet spectrum at a series of sampling time, named predominant frequencies, are extracted to analyze the variation of frequency content of seismic ground motions in time. And the time-variation of the predominant frequencies of 178 magnitude-distance-site bins for different directions are obtained by calculating the mean square root of predominant frequencies within a bin. The exponential trigonometric function is then use to fit the data, which describes the predominant frequency of ground-motion as a function of time with model parameters given in tables for different magnitude, distance, site conditions and direction. Finally, a practical frequency-dependent amplitude envelope function is developed based on the time-varying frequency derived in this paper, which has clear statistical parameters and can emphasize the effect of low-frequency components on later seismic action. The results illustrate that the time-varying predominant frequency can preferably reflect the non-stationarity of the frequency content in earthquake ground motions and that empirical models given in this paper facilitates the simulation of ground motions.

• Journal of Environmental Impact Assessment
• /
• v.23 no.1
• /
• pp.39-50
• /
• 2014

In recent years, the frequency and scale of the natural disasters are growing rapidly due to the global climate change. In case of the urban flooding, high-density of population and infrastructure has caused the more intensive damages. In this study, we analyzed the spatial characteristics of urban flooding damage factors using GWR(Geographically Weighted Regression) for effective disaster prevention and then, classified the causes of the flood damage by spatial characteristics. The damage factors applied consists of natural variables such as the poor drainage area, the distance from the river, elevation and slope, and anthropogenic variables such as the impervious surface area, urbanized area, and infrastructure area, which are selected by literature review. This study carried out the comparative analysis between OLS(Ordinary Least Square) and GWR model for identifying spatial non-stationarity and spatial autocorrelation, and in the results, GWR model has higher explanation power than OLS model. As a result, it appears that there are some differences between each of the flood damage areas depending on the variables. We conclude that the establishment of disaster prevention plan for urban flooding area should reflect the spatial characteristics of the damaged areas. This study provides an improved understandings of the causes of urban flood damages, which can be diverse according to their own spatial characteristics.

• Wind and Structures
• /
• v.35 no.6
• /
• pp.369-383
• /
• 2022

Wind tunnel experiment was carried out to study the cross-wind layer forces on a square cross-section building model using a synchronous multi-pressure sensing system. The stationarity of measured wind loadings are firstly examined, revealing the non-stationary feature of cross-wind forces. By converting the measured non-stationary wind forces into an energetically equivalent stationary process, the characteristics of local wind forces are studied, such as power spectrum density and spanwise coherence function. Mathematical models to describe properties of cross-wind forces at different layers are thus established. Then, a conditional simulation method, which is able to ex-tend pressure measurements starting from experimentally measured points, is proposed for the cross-wind loading. The method can reproduce the non-stationary cross-wind force by simulating a stationary process and the corresponding time varying amplitudes independently; in this way the non-stationary wind forces can finally be obtained by combining the two parts together. The feasibility and reliability of the proposed method is highlighted by an ex-ample of across wind loading simulation, based on the experimental results analyzed in the first part of the paper.

• Wind and Structures
• /
• v.17 no.2
• /
• pp.135-162
• /
• 2013

The majority of experiments to characterize the turbulence in the surface layer have been performed in flat, open expanses. In order to characterize the turbulence in built-up terrain, two mobile towers were deployed during Hurricane Ike (2008) in close proximity, but downwind of different terrain conditions: suburban and open. Due to the significant non-stationarity of the data primarily caused by changes in wind direction, empirical mode decomposition was employed to de-trend the signal. Analysis of the data showed that the along-wind mean turbulence intensity of the suburban terrain was 37% higher than that of the open terrain. For the mean vertical turbulence intensity, the increase for the suburban terrain was as high as 74%, which may have important implications in structural engineering. The gust factor of the suburban terrain was also 16% higher than that of the open terrain. Compared to non-hurricane spectral models, the obtained spectra showed significantly higher energy in low frequencies especially for the open terrain.