• Proceedings of the Korea Water Resources Association Conference
• /
• 2010.05a
• /
• pp.679-683
• /
• 2010

본 연구에서는 공간적 변수인 조도계수를 기지의 수위값을 이용하여 최적값을 결정하는 방법에 대해서 검토하고자 한다. 최적화 기법에 의한 조도계수는 기지의 수위값과 수치모의에 의한 결과 값의 전체 오차를 최소화하는 값으로 결정된다. 본 연구에서는 3가지 최적화 기법을 이용하였으며 가상 수로에 대해서 적용하였다. 수위계산은 표준축차법에 의해 수행하였으며 사용된 최적화 기법은 quasi-Newton 방법이다. 1차원 모형은 Matlab을 이용하여 표준축자법으로 구성하였으며 BFGS 기법, L-BFGS 기법, Steepest Gradient Descent 기법 등도 Matlab으로 구성하였다. 표준축차법은 조도계수가 입력되면 기지의 수위값과의 2-norm을 계산하도록 구성하였다. 계산 결과에 의하면 세가 기법 모두 20 23회 정도의 반복계산을 수행하고 값이 수렴되었는데, L-BFGS의 경우에는 정확하게 음수의 조도계수로 수렴하였으며, BFGS기법과 Steepest Gradient 기법의 경우에는 양의 값으로 정확하게 수렴하였다.

• Journal of the Korean Association of Geographic Information Studies
• /
• v.6 no.3
• /
• pp.129-138
• /
• 2003

A new methodology for selecting spatially variable model control parameter values through consideration of inference models within a Hydroinformatic system has been developed to overcome problems associated with determination of spatially variable control parameter values for both ungauged and gauged catchment. The adopted Hydroinformatic tools for determination of control parameter values were a GIS(Arc/Info) to handle spatial and non-spatial attribute information, the SWMM(stormwater management model) to simulate catchment response to hydrologic events, and lastly, L_BFGS_B(a limited memory quasi-Newton algorithm) to assist in the calibration process. As a result, high accuracy of control parameter estimation was obtained by considering the spatial variations of the control parameters based on landuse characteristics. Also, considerable time and effort necessary for estimating a large number of control parameters were reduced from the new calibration approach.

• The Journal of the Convergence on Culture Technology
• /
• v.10 no.5
• /
• pp.163-171
• /
• 2024

This study applies neural network analysis techniques to examine the factors influencing the purchasing decisions of fashion eyewear among women in their 30s and 40s, comparing these findings with traditional parametric analysis methods. In the fashion area, machine learning techniques are utilized for personalized fashion recommendation systems. However, research on such applications in Korea remains insufficient. By reanalyzing a study conducted in 2017 using traditional quantitative methods with these new techniques, this study aims to confirm the utility of neural network methods. Notably, the study finds that the classification accuracy of preferred sunglasses design is highest, at 86.2%, when the L-BFGS-B neural network is activated using the hyperbolic tangent function. The most critical factors influencing purchasing decisions were consumers' occupations and their pursuit of new styles. It is interpreted that Korean sunglasses consumers prefer "safe changes." These findings are consistent for selecting both the frames and lenses of sunglasses. Traditional quantitative analysis suggests that the type of sunglasses preferred varies according to the group to which a consumer belongs. In contrast, neural network analysis predicts the preferred sunglasses for each individual, thereby facilitating the development of personalized sunglasses recommendation systems.