• Structural Engineering and Mechanics
• /
• v.8 no.5
• /
• pp.513-529
• /
• 1999

A concept of hierarchical modeling, the newest modeling technology, has been introduced in early 1990's. This new technology has a great potential to advance the capabilities of current computational mechanics. A first step to implement this concept is to construct hierarchical models, a family of mathematical models sequentially connected by a key parameter of the problem under consideration and have different levels in modeling accuracy, and to investigate characteristics in their numerical simulation aspects. Among representative model problems to explore this concept are elastic structures such as beam-, arch-, plate- and shell-like structures because the mechanical behavior through the thickness can be approximated with sequential accuracy by varying the order of thickness polynomials in the displacement or stress fields. But, in the numerical, analysis of hierarchical models, two kinds of errors prevail, the modeling error and the numerical approximation error. To ensure numerical simulation quality, an accurate estimation of these two errors is definitely essential. Here, a local a posteriori error estimator for elastic structures with thin domain such as plate- and shell-like structures is derived using the element residuals and the flux balancing technique. This method guarantees upper bounds for the global error, and also provides accurate local error indicators for two types of errors, in the energy norm. Compared to the classical error estimators using the flux averaging technique, this shows considerably reliable and accurate effectivity indices. To illustrate the theoretical results and to verify the validity of the proposed error estimator, representative numerical examples are provided.

• International Journal of Fluid Machinery and Systems
• /
• v.10 no.3
• /
• pp.240-253
• /
• 2017

Latin hypercube sampling is widely used design-of-experiment technique to select design points for simulation which are then used to construct a surrogate model. The exploration/exploitation properties of surrogate models depend on the size and distribution of design points in the chosen design space. The present study aimed at evaluating the performance characteristics of various surrogate models depending on the Latin hypercube sampling (LHS) procedure (sample size and spatial distribution) for a diverse set of optimization problems. The analysis was carried out for two types of problems: (1) thermal-fluid design problems (optimizations of convergent-divergent micromixer coupled with pulsatile flow and boot-shaped ribs), and (2) analytical test functions (six-hump camel back, Branin-Hoo, Hartman 3, and Hartman 6 functions). The three surrogate models, namely, response surface approximation, Kriging, and radial basis neural networks were tested. The important findings are illustrated using Box-plots. The surrogate models were analyzed in terms of global exploration (accuracy over the domain space) and local exploitation (ease of finding the global optimum point). Radial basis neural networks showed the best overall performance in global exploration characteristics as well as tendency to find the approximate optimal solution for the majority of tested problems. To build a surrogate model, it is recommended to use an initial sample size equal to 15 times the number of design variables. The study will provide useful guidelines on the effect of initial sample size and distribution on surrogate construction and subsequent optimization using LHS sampling plan.

• Transactions of the Korean Society of Mechanical Engineers
• /
• v.12 no.5
• /
• pp.944-951
• /
• 1988

Global-Local Laminate Variational Model is utilized to investigate the characteristics of interlaminar stresses in thick composite laminates under uniform axial extension. Various laminates with different fiber orientation and stacking sequences are analyzed to observe the behavior of interlaminar normal stresses. From this result, the interlaminar normal stress distribution along the laminate interfaces is examined and discussed with an existing approximation model. The repeated stacking of Poisson's ratio symmetric sublaminates is found to be the best stacking method of thick composite laminates to reduce the interlaminar normal stresses for the prevention of the free-edge delamination.

• Journal of the Society of Naval Architects of Korea
• /
• v.34 no.1
• /
• pp.93-101
• /
• 1997

Generally the traditional optimization methods have possibilities not only to give a different optimum value according to their starting point, but also to get to local optima. On the other hand, Genetic Algorithm (GA) has an ability of robust global search. In this paper, a new optimization method - the combination of traditional optimization method and genetic algorithm - is presented so as to overcome the above disadvantage of traditional methods. In order to increase the efficiency of the hybrid optimization method, a knowledge-based reasoning is adopted in the part of mathematical modeling, algorithm selection, and process control. The validation of the developed knowledge-based hybrid optimization method was examined and verified applying the method to nonlinear mathematical models.

• Journal of the Korean Physical Society
• /
• v.73 no.10
• /
• pp.1589-1595
• /
• 2018

Similarity index measures the topological proximity of node pairs in a complex network. Numerous similarity indices have been defined and investigated, but the dependency of structure on the performance of similarity indices has not been sufficiently investigated. In this study, we investigated the relationship between the performance of similarity indices and structural properties of a network by employing a two-state random network. A node in a two-state network has binary types that are initially given, and a connection probability is determined from the state of the node pair. The performances of similarity indices are affected by the number of links and the ratio of intra-connections to inter-connections. Similarity indices have different characteristics depending on their type. Local indices perform well in small-size networks and do not depend on whether the structure is intra-dominant or inter-dominant. In contrast, global indices perform better in large-size networks, and some such indices do not perform well in an inter-dominant structure. We also found that link prediction performance and the performance of similarity are correlated in both model networks and empirical networks. This relationship implies that link prediction performance can be used as an approximation for the performance of the similarity index when information about node type is unavailable. This relationship may help to find the appropriate index for given networks.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.17 no.9
• /
• pp.2483-2504
• /
• 2023

Most existing low-light enhancement algorithms either use a large number of training parameters or lack generalization to real-world scenarios. This paper presents a novel lightweight and robust pixel-wise polynomial approximation-based deep network for low-light image enhancement. For mapping the low-light image to the enhanced image, pixel-wise higher-order polynomials are employed. A deep convolution network is used to estimate the coefficients of these higher-order polynomials. The proposed network uses multiple branches to estimate pixel values based on different receptive fields. With a smaller receptive field, the first branch enhanced local features, the second and third branches focused on medium-level features, and the last branch enhanced global features. The low-light image is downsampled by the factor of 2b-1 (b is the branch number) and fed as input to each branch. After combining the outputs of each branch, the final enhanced image is obtained. A comprehensive evaluation of our proposed network on six publicly available no-reference test datasets shows that it outperforms state-of-the-art methods on both quantitative and qualitative measures.

• Journal of Power Electronics
• /
• v.17 no.2
• /
• pp.476-489
• /
• 2017

Under partial shading conditions (PSCs), multiple maximums may be exhibited on the P-U curve of string inverter photovoltaic (PV) systems. Under such conditions, heuristic methods are invalid for extracting a global maximum power point (GMPP); intelligent algorithms are time-consuming; and model-based methods are complex and costly. To overcome these shortcomings, a novel hybrid MPPT (MPF-IP&O) based on a model-based peak forecasting (MPF) method and an improved perturbation and observation (IP&O) method is proposed. The MPF considers the influence of temperature and does not require solar radiation measurements. In addition, it can forecast all of the peak values of the PV string without complex computation under PSCs, and it can determine the candidate GMPP after a comparison. Hence, the MPF narrows the searching range tremendously and accelerates the convergence to the GMPP. Additionally, the IP&O with a successive approximation strategy searches for the real GMPP in the neighborhood of the candidate one, which can significantly enhance the tracking efficiency. Finally, simulation and experiment results show that the proposed method has a higher tracking speed and accuracy than the perturbation and observation (P&O) and particle swarm optimization (PSO) methods under PSCs.

• Journal of the Korean Institute of Telematics and Electronics C
• /
• v.35C no.12
• /
• pp.85-98
• /
• 1998

This paper proposes a new design method of neuro-FLC by the Lamarckian co-adaptation scheme that incorporates the backpropagation learning into the GA evolution in an attempt to find optimal design parameters (fuzzy rule base and membership functions) of application-specific FLC. The design parameters are determined by evolution and learning in a way that the evolution performs the global search and makes inter-FLC parameter adjustments in order to obtain both the optimal rule base having high covering value and small number of useful fuzzy rules and the optimal membership functions having small approximation error and good control performance while the learning performs the local search and makes intra-FLC parameter adjustments by interacting each FLC with its environment. The proposed co-adaptive design method produces better approximation ability because it includes the backpropagation learning in every generation of GA evolution, shows better control performance because the used COG defuzzifier computes the crisp value accurately, and requires small workspace because the optimization procedure of fuzzy rule base and membership functions is performed concurrently by an integrated fitness function on the same fuzzy partition. Simulation results show that the Lamarckian co-adapted FLC produces the most superior one among the differently generated FLCs in all aspects such as the number of fuzzy rules, the approximation ability, and the control performance.

• Journal of Information Processing Systems
• /
• v.9 no.4
• /
• pp.633-650
• /
• 2013

All the imputation techniques proposed so far in literature for data imputation are offline techniques as they require a number of iterations to learn the characteristics of data during training and they also consume a lot of computational time. Hence, these techniques are not suitable for applications that require the imputation to be performed on demand and near real-time. The paper proposes a computational intelligence based architecture for online data imputation and extended versions of an existing offline data imputation method as well. The proposed online imputation technique has 2 stages. In stage 1, Evolving Clustering Method (ECM) is used to replace the missing values with cluster centers, as part of the local learning strategy. Stage 2 refines the resultant approximate values using a General Regression Neural Network (GRNN) as part of the global approximation strategy. We also propose extended versions of an existing offline imputation technique. The offline imputation techniques employ K-Means or K-Medoids and Multi Layer Perceptron (MLP)or GRNN in Stage-1and Stage-2respectively. Several experiments were conducted on 8benchmark datasets and 4 bank related datasets to assess the effectiveness of the proposed online and offline imputation techniques. In terms of Mean Absolute Percentage Error (MAPE), the results indicate that the difference between the proposed best offline imputation method viz., K-Medoids+GRNN and the proposed online imputation method viz., ECM+GRNN is statistically insignificant at a 1% level of significance. Consequently, the proposed online technique, being less expensive and faster, can be employed for imputation instead of the existing and proposed offline imputation techniques. This is the significant outcome of the study. Furthermore, GRNN in stage-2 uniformly reduced MAPE values in both offline and online imputation methods on all datasets.

• The KIPS Transactions:PartB
• /
• v.11B no.2
• /
• pp.207-212
• /
• 2004

This paper presents a reinforcement learning framework for stock trading systems. Trading system parameters are optimized by Q-learning algorithm and neural networks are adopted for value approximation. In this framework, cooperative multiple agents are used to efficiently integrate global trend prediction and local trading strategy for obtaining better trading performance. Agents Communicate With Others Sharing training episodes and learned policies, while keeping the overall scheme of conventional Q-learning. Experimental results on KOSPI 200 show that a trading system based on the proposed framework outperforms the market average and makes appreciable profits. Furthermore, in view of risk management, the system is superior to a system trained by supervised learning.