• Journal of the Korea Society of Computer and Information
• /
• v.14 no.8
• /
• pp.11-18
• /
• 2009

This paper proposes the way of improving learning speed in Levenberg-Marquardt algorithm using the principal submatrix of Jacobian matrix. The Levenberg-Marquardt learning uses Jacobian matrix for Hessian matrix to get the second derivative of an error function. To make the Jacobian matrix an invertible matrix. the Levenberg-Marquardt learning must increase or decrease ${\mu}$ and recalculate the inverse matrix of the Jacobian matrix due to these changes of ${\mu}$. Therefore, to have the proper ${\mu}$, we create the principal submatrix of Jacobian matrix and set the ${\mu}$ as the eigenvalues sum of the principal submatrix. which can make learning speed improve without calculating an additional inverse matrix. We also showed that our method was able to improve learning speed in both a generalized XOR problem and a handwritten digit recognition problem.

• Journal of the Korea Society of Computer and Information
• /
• v.15 no.4
• /
• pp.57-63
• /
• 2010

The damping parameter of Levenberg-Marquardt algorithm switches between error backpropagation and Gauss-Newton learning and affects learning speed. Fixing the damping parameter induces some oscillation of error and decreases learning speed. Therefore, we propose the way of a variable damping parameter with referring to the alternation of error. The proposed method makes the damping parameter increase if error rate is large and makes it decrease if error rate is small. This method so plays the role of momentum that it can improve learning speed. We tested both iris recognition and wine recognition for this paper. We found out that this method improved learning speed in 67% cases on iris recognition and in 78% cases on wine recognition. It was also showed that the oscillation of error by the proposed way was less than those of other algorithms.

• Proceedings of the Korean Operations and Management Science Society Conference
• /
• 2005.05a
• /
• pp.746-752
• /
• 2005

본 논문에서는 다층 퍼셉트론 신경망 학습을 위한 새로운 두 단계 학습방법을 제안하고 이를 옵션 가격결정 모형에 응용하였다. 제안된 신경망 학습 알고리즘의 첫번째 단계는 Levenberg-Marquardt 알고리즘을 이용하여 빠르게 국소최적해를 찾는 것이고 두 번째 단계는 첫 번째 단계에서 찾은 국소최적해가 원하는 수준에 미치지 못할 경우 선형탐색 터널링을 이용해서 더 나은 해를 찾는 것이다. 이 두 단계를 반복적으로 수행함으로써 연결가중치 공간에서 구하고자 하는 해를 빠르고 안정적으로 찾을 수 있다. 현재 옵션가격결정 모형으로 많이 이용되고 있는 Black-Scholes 모형의 문제점을 극복하기 위해서 제안된 신경망 모형을 옵션가격결정 문제에 사용하였다. 이 모형을 KOSPI200 옵션 데이터로 실험한 결과 Black-Scholes 모형에 비해 검증오차를 60% 가량 줄일 수 있었다.

• Journal of Korea Water Resources Association
• /
• v.34 no.6
• /
• pp.701-711
• /
• 2001

Forecasting of water quality variation is not an easy process due to the complicated nature of various water quality factors and their interrelationships. The objective of this study is to test the applicability of neural network models to the forecasting of the water quality at Gongju station in Geum River. This is done by forecasting monthly water qualities such as DO, BOD, and TN, and comparing with those obtained by ARIMA model. The neural network models of this study use BP(Back Propagation) algorithm for training. In order to improve the performance of the training, the models are tested in three different styles ; MANN model which uses the Moment-Adaptive learning rate method, LMNN model which uses the Levenberg-Marquardt method, and MNN model which separates the hidden layers for judgement factors from the hidden layers for water quality data. the results show that the forecasted water qualities are reasonably close to the observed data. And the MNN model shows the best results among the three models tested

• Journal of Korea Water Resources Association
• /
• v.33 no.4
• /
• pp.505-515
• /
• 2000

THe purpose of this study is to forecast of runoff hydrographs according to rainfall event in a stream. The neural network theory as a hydrologic blackbox model is used to solve hydrological problems. The Back-Propagation(BP) algorithm by the Levenberg-Marquardt(LM) techniques and Radial Basis Function(RBF) network in Neural Network(NN) models are used. Runoff hydrograph is forecasted in Bocheongstream basin which is a IHP the representative basin. The possibility of a simulation for runoff hydrographs about unlearned stations is considered. The results show that NN models are performed to effective learning for rainfall-runoff process of hydrologic system which involves a complexity and nonliner relationships. The RBF networks consist of 2 learning steps. The first step is an unsupervised learning in hidden layer and the next step is a supervised learning in output layer. Therefore, the RBF networks could provide rather time saved in the learning step than the BP algorithm. The peak discharge both BP algorithm and RBF network model in the estimation of an unlearned are a is trended to observed values.

• Journal of Korea Water Resources Association
• /
• v.31 no.1
• /
• pp.45-57
• /
• 1998

The purpose of this paper is to develop a neural network model in order to forecast flood inflow into the reservoir that has the nature of uncertainty and nonlinearity. The model has the features of multi-layered structure and parallel multi-connections. To develop the model. backpropagation learning algorithm was used with the Momentum and Levenberg-Marquardt techniques. The former technique uses gradient descent method and the later uses gradient descent and Gauss-Newton method respectively to solve the problems of local minima and for the speed of convergency. Used data for learning are continuous fixed real values of input as well as output to emulate the real physical aspects. after learning process. a reservoir inflows forecasting model at flood period was constructed. The data for learning were used to calibrate the developed model and the results were very satisfactory. applicability of the model to the Chungju Mlultipurpose Reservoir proved the availability of the developed model.

• The Journal of Korean Institute of Information Technology
• /
• v.18 no.5
• /
• pp.55-64
• /
• 2020

Estimating the state of charge (SOC) of a battery is essential for increasing the stability and reliability of a photovoltaic system. In this study, battery SOC estimation methods were proposed using artificial neural networks (ANNs) with gradient descent (GD), Levenberg-Marquardt (LM), and scaled conjugate gradient (SCG), and an adaptive neuro-fuzzy inference system (ANFIS). The charge start voltage and the integrated charge current were used as input data and the SOC was used as output data. Four models (ANN-GD, ANN-LM, ANN-SCG, and ANFIS) were implemented for battery SOC estimation and compared using MATLAB. The experimental results revealed that battery SOC estimation using the ANFIS model had both the highest accuracy and highest convergence speed.

• Journal of the Korean Institute of Intelligent Systems
• /
• v.16 no.1
• /
• pp.1-7
• /
• 2006

In this paper, adaptive Facial Emotional Recognition system based on Gabor Wavelet Neural Network, considering six feature Points in face image to extract specific features of facial expression, is proposed. Levenberg-Marquardt-based training methodology is used to formulate initial network, including feature extraction stage. Therefore, heuristics in determining feature extraction process can be excluded. Moreover, to make an adaptive network for new user, Q-learning which has enhanced reward function and unsupervised fuzzy neural network model are used. Q-learning enables the system to ge optimal Gabor filters' sets which are capable of obtaining separable features, and Fuzzy Neural Network enables it to adapt to the user's change. Therefore, proposed system has a good on-line adaptation capability, meaning that it can trace the change of user's face continuously.

• Journal of the Korean Society for Aeronautical & Space Sciences
• /
• v.37 no.5
• /
• pp.425-432
• /
• 2009

In this study, the ability of neural network in modeling and predicting of the unsteady aerodynamic force coefficients of 2D airfoil with the data obtained from Euler CFD code has been confirmed. Neural network models are constructed based on supervised training process using Levenberg-Marquardt algorithm, combining this into genetic algorithm, hybrid genetic algorithm and the efficiency of the two cases are analyzed and compared. It is shown that hybrid-genetic algorithm is more efficient for neural network of complex system and the predicted properties of the unsteady aerodynamic force coefficients of 2D airfoil by the neural network models are confirmed to be similar to that of the numerical results and verified as suitable representing reduced models.

• Journal of the Korean Society for Nondestructive Testing
• /
• v.29 no.5
• /
• pp.466-472
• /
• 2009

An acoustic leak monitoring system(ALMS) using acoustic emission(AE) technique was applied for leakage detection of nuclear power plant's pipeline which is operated in high temperature and pressure condition. Since this system only monitors the existence of leak using the root mean square(RMS) value of raw signal from AE sensor, the difficulty occurs when the characteristics of leak size and shape need to be evaluated. In this study, dual monitoring system using AE sensor and accelerometer was introduced in order to solve this problem. In addition, artificial neural network(ANN) with Levenberg.Marquardt(LM) training algorithm was also applied due to rapid training rate and gave the reliable classification performance. The input parameters of this ANN were extracted from varying signal received from experimental conditions such as the fluid pressure inside pipe, the shape and size of the leak area. Additional experiments were also carried out and with different objective which is to study the generation and characteristic of lamb and surface wave according to the pipe thickness.