Yung-Seop Lee

The design of memory semiconductors involves satisfying various customer demands and rapid provision of high-quality products. Therefore, manufacturers have developed high-quality memory semiconductors using various technological elements to ensure their performance reliability and operation in various environments. Major quality aspects, such as warpage, stress, and strain, have different effects. Through experimental investigations, appropriate technological elements are selected using different technological elements (thickness and materials) to deliver the quality desired by customers. However, experimental investigations may delay product deliveries and incur considerable costs. To overcome these limitations, this study proposes a new design method for determining optimal solutions based on the prepared technological elements for the quality desired by customers. The proposed method differs from traditional optimization methods in that it provides multiple solutions that satisfy the customer requirements. Traditional experimental designs that use orthogonal arrays fail to reflect the various constraints involved in memory semiconductor designs. Therefore, we aimed to solve this problem by applying an analysis methodology for mixture experimental designs based on orthogonal arrays using the coordinate exchange algorithm. Furthermore, a desirability function was used to assess the satisfaction of multiple quality characteristics (warpage, stress, strain), through which optimal packaging conditions were confirmed at approximately 12.5% of the total combination level. The results of this study are expected to improve the optimization and efficiency of semiconductor packaging processes.

Hyejoon Park, Hyunjoong Kim, Yung-Seop Lee. Korean J Appl Stat 2024;37:691-702. https://doi.org/10.5351/KJAS.2024.37.5.691

The Fourth Industrial Revolution offers an opportunity to companies to improve their competitiveness by utilizing data analytics. Particularly, real-time analysis of data gathered using various sensors is an area of interest for manufacturing companies aiming to use captured data for developing more robust monitoring systems. Therefore, trace data related to real-time processes have attracted attention in various fields. However, exploiting large amounts of trace data requires high-performance smart infrastructure. To this end, this study proposes statistics that incorporate the characteristics of trace data based on functional data analysis (FDA) and applies them to supervised learning. The empirical test results indicate that the functional principal component of FDA exhibits a significantly lower misclassification rate for the proposed model compared with that of the summary statistics-based model. Particularly, the FDA-based supervised model is less complex and exhibits less variability in terms of the number of explanatory variables based on the sample size of training data. When using summary statistics, the FDA variables were potentially selected as important variables in the least absolute shrinkage and selection operator model. The results of this study may assist various industries dealing with the aggregation of trace data for anomaly detection and intelligent factory management.

The selection of group features is a critical aspect in reducing model complexity by choosing the most essential group features, while eliminating the less significant ones. The existing group feature selection methods select a set of important group features, without providing the relative importance of all group features. Moreover, few methods consider the relative importance of group features in the selection process. This study introduces a permutation-based group feature selection approach specifically designed for high-dimensional multiclass datasets. Initially, the least absolute shrinkage and selection operator (lasso) method was applied to eliminate irrelevant individual features within each group feature. Subsequently, the relative importance of the group features was computed using a random-forest-based permutation method. Accordingly, the process selected the highly significant group features. The performance of the proposed method was evaluated using machine learning algorithms and compared with the performance of other approaches, such as group lasso. We used real-world, high-dimensional, multiclass microarray datasets to demonstrate its effectiveness. The results highlighted the capability of the proposed method, which not only selected significant group features but also provided the relative importance and ranking of all group features. Furthermore, the proposed method outperformed the existing method in terms of accuracy and F1 score.

As renewable energy generation prediction systems have been introduced into the energy trading market, accurate prediction of solar photovoltaic (PV) energy generation has become a crucial challenge for ensuring stable trade of variable energy. Therefore, it is essential to quantitatively investigate and analyze current prediction technology and develop more advanced prediction systems. In this study, three models for predicting PV energy generation were investigated. Multiple regression, random forest, and gradient boosting machine (GBM) models were con-structed for 15 utility-scale power plants throughout South Korea. Model performance was evaluated in terms of root mean square error (RMSE) and mean absolute error (MAE). The multiple regression-based model had an RMSE of 12.00% and MAE of 9.06%. The random forest model had an RMSE of 11.69% and MAE of 8.61%. The GBM model had an RMSE of 11.34% and MAE of 8.18%, indicating its superior accuracy among the three models in predicting PV energy generation. These findings provide valuable insights that may contribute to the development of advanced PV energy generation models and their application in grid operations.

As renewable energy generation prediction system has been introduced into the energy trading market, making a model to accurately predict the quantity of solar photovoltaic (PV) energy generation has become a significant problem. Moreover, to encourage an accurate prediction of the quantity of energy generation, an incentive system has been implemented for those who predict the quantity of solar PV energy under the error rate of 8%. Therefore, it has become more important to investigate and analyze current prediction technology numerically and develop more advanced prediction system. In this study, we tried to develop a better model to improve the accuracy of solar PV energy generation quantity by comparing three models made with gradient boosting machine (GBM), Model 1, Model 2, Model 3 respectively. Model 1 was built with the whole training data set without any additional preprocessing. After conducting some additional preprocessing procedure to predict solar energy generation more accurately, we made Model 2 with the whole training data set and Model 3 with only upper 10% of energy generation capacity. To compare the accuracy of three models, normalized mean absolute error (nMAE) was used as an evaluation index. The nMAE of Model 1 was 9.64% while the Model 2 showed 8.41%. Also, Model 3, which was constructed with the training set of upper 10% energy generation capacity, outperformed with the nMAE of 8.08%. For further study, to check the effectiveness of models constructed with GBM, a time series model, autoregressive integrated moving average (ARIMA), was also built and the nMAE was compared.

Choi YJ, et al. Allergy Asthma Immunol Res. 2023 Nov;15(6):825-836. https://doi.org/10.4168/aair.2023.15.6.825

Solar irradiance is a major meteorological factor affecting solar energy generation. In this study, we develop a short-term solar irradiance forecasting model with a high time resolution to accurately predict the amount of solar energy generated and apply it to the real-time energy trade market. Two types of irradiance (Global Horizontal Irradiance and Direct Normal Irradiance) data observed at Daejeon, South Korea, are predicted using two time series analysis models (the ARIMA model, which does not consider seasonality, and SARIMA model which considers seasonality), which we compared to determine the model that is better suited to performing predictions. Comparing the prediction errors from 15 to 120 minutes, using RMSE and nRMSE as evaluation indices, GHI was better predicted using the ARIMA model from 15 to 75 minutes, while the SARIMA model performed better from 90 to 120 minutes. For DNI, the ARIMA model showed higher accuracy than the SARIMA model during the entire prediction period.