• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.18 no.1
• /
• 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.

• Journal of Korea Water Resources Association
• /
• v.37 no.2
• /
• pp.145-154
• /
• 2004

The purpose of this study is to develop a real-time forecasting model in order to predict the flood runoff which has the nature of non-linearity and to verify applicability of neural network model for flood warning system. Developed model based on neural network, NRDFM(Neural River Discharge-Stage Forecasting Model) is applied to predict the flood discharge on Waekwann and Jindong stations in Nakdong river basin. As a result of flood forecasting on these two stations, it can be concluded that NRDFM-II is the best predictive model for real-time operation. In addition, the results of forecasting used on NRDFM-I and NRDFM-II model are not bad and these models showed sufficient probability for real-time flood forecasting. Consequently, it is expected that NRDFM in this study can be utilized as suitable model for real-time flood warning system and this model can perform flood control and management efficiently.

• Proceedings of the Korean Operations and Management Science Society Conference
• /
• 2006.11a
• /
• pp.165-168
• /
• 2006

Exponential smoothing methods do not adapt well to unexpected changes in underlying process. Over the past few decades a number of adaptive smoothing models have been proposed which allow for the continuous adjustment of the smoothing constant value in order to provide a much earlier detection of unexpected changes. However, most of previous studies presented ad hoc procedure of adaptive forecasting without any theoretical background. In this paper, we propose a detection-adaptation procedure applied to simple and Holt's linear method. We derive level and slope change detection statistics based on Bayesian statistical theory and present distribution of the statistics by simulation method. The proposed procedure is compared with previous adaptive forecasting models using simulated data and economic time series data.

• Communications for Statistical Applications and Methods
• /
• v.10 no.3
• /
• pp.931-954
• /
• 2003

It is widely recognized that economic time series involved not only the linearities but also the non-linearities. In this paper, when the economic time series data have the nonlinear characteristics we propose the forecasts method using combinations of both forecasts from linear and nonlinear models. In empirical study, we compare the forecasting performance of 4 exchange rates models(AR, GARCH, AR+GARCH, Bilinear model) and combination of these forecasts for dairly Won/Dollar exchange rates returns. The combination method is selected by the estimated individual forecast errors using Monte Carlo simulations. And this study shows that the combined forecasts using unrestricted least squares method is performed substantially better than any other combined forecasts or individual forecasts.

• Proceedings of the Korea Society for Simulation Conference
• /
• 1998.03a
• /
• pp.101-105
• /
• 1998

A large system predictor, which can perform prediction of sales trend in a huge number of distribution centers, is presented using neural predictive model. There are 20,000 number of distribution centers, and each distribution center need to forecast future demand in order to establish a reasonable inventory policy. Therefore, the number of forecasting models corresponds to the number of distribution centers, which is not possible to estimate that kind of huge number of accurate models in ERP (Enterprise Resource Planning)module. Multilayer neural net as universal approximation is employed for fitting the prediction model. In order to improve prediction accuracy, a sequential simulation procedure is performed to get appropriate network structure and also to improve forecasting accuracy. The proposed simulation procedure includes neural structure identification and virtual predictive model generation. The predictive model generation consists of generating virtual signals and estimating predictive model. The virtual predictive model plays a key role in tuning the real model by absorbing the real model errors. The complement approach, based on real and virtual model, could forecast the future demands of various distribution centers.

• Journal of Wetlands Research
• /
• v.1 no.1
• /
• pp.17-28
• /
• 1999

The purpose of this study was to analysis water quality monitoring data and to estimate future trends using ARIMA model of time series analysis. Water quality data in Chin yang reservoir were used with monthly monitoring interval during past 7 years. The variations of water quality parameters with periodicity and trend could be estimated by multiplicative ARIMA models and the statistical tests showed a good agreement with the observed data. Therefore, the monthly values of water quality parameters could be forecasted using these models.

• Communications for Statistical Applications and Methods
• /
• v.28 no.5
• /
• pp.521-536
• /
• 2021

As life expectancies increase continuously over the world, the accuracy of forecasting mortality is more and more important to maintain social systems in the aging era. Currently, the most popular model used is the Lee-Carter model but various studies have been conducted to improve this model with one of them being 6-parametric factor model (6-PFM) which is introduced in this paper. To this new model, long short-term memory (LSTM) and regularized LSTM are applied in addition to vector autoregression (VAR), which is a traditional time-series method. Forecasting accuracies of several models, including the LC model, 4-PFM, 5-PFM, and 3 6-PFM's, are compared by using the U.S. and Korea life-tables. The results show that 6-PFM forecasts better than the other models (LC model, 4-PFM, and 5-PFM). Among the three 6-PFMs studied, regularized LSTM performs better than the other two methods for most of the tests.

• International conference on construction engineering and project management
• /
• 2005.10a
• /
• pp.913-916
• /
• 2005

This paper presents the development of Artificial Neural Network models for forecasting the cost and contract duration of school reconstruction projects to assist the planners' decision-making in the early stage of the projects. 132 schools reconstruction projects in central Taiwan, which received the most serious damage from the Chi-Chi Earthquake, were collected. The developed Artificial Neural Network prediction models demonstrate good prediction abilities with average error rates under 10% for school reconstruction projects. The analytical results indicate that the Artificial Neural Network model with back-propagation learning is a feasible method to produce accurate prediction results to assist planners' decision-making process.

• Proceedings of the Korean Society of Disaster Information Conference
• /
• 2023.11a
• /
• pp.259-260
• /
• 2023

The rapid urbanization and modernization observed in countries like Myanmar have led to significant concerns regarding traffic congestion, especially in urban areas. This study focuses on the analysis and revitalization of urban transport in selected areas of Myanmar. The core of urban transportation planning lies in travel forecasting, which employs models to predict future traffic patterns and guide decisions related to road capacity, transit services, and land use policies. Travel demand modeling involves a series of mathematical models that simulate traveler behavior and decision-making within a transportation system, including highways, transit options, and policies. The paper offers an overview of the traditional four-step transportation modeling system, utilizing a simplified transport network in the context of Mandalay City, Myanmar.

• Communications for Statistical Applications and Methods
• /
• v.31 no.1
• /
• pp.105-127
• /
• 2024

We propose recurrent neural network (RNN) methods for forecasting realized volatility (RV). The data are RVs of ten major stock price indices, four from the US, and six from the EU. Forecasts are made for relative ratio of adjacent RVs instead of the RV itself in order to avoid the out-of-scale issue. Forecasts of RV ratios distribution are first constructed from which those of RVs are computed which are shown to be better than forecasts constructed directly from RV. The apparent asymmetry of RV ratio is addressed by the Piecewise Min-max (PM) normalization. The serial dependence of the ratio data renders us to consider two architectures, long short-term memory (LSTM) and gated recurrent unit (GRU). The hyperparameters of LSTM and GRU are tuned by the nested cross validation. The RNN forecast with the PM normalization and ratio transformation is shown to outperform other forecasts by other RNN models and by benchmarking models of the AR model, the support vector machine (SVM), the deep neural network (DNN), and the convolutional neural network (CNN).