• International Journal of Control, Automation, and Systems
• /
• 제3권2호
• /
• pp.202-210
• /
• 2005

This paper presents a new neural network control for nonlinear bridge systems with earthquake excitation. We design multi-layer neural network controllers with a single hidden layer. The selection of an optimal number of neurons in the hidden layer is an important design step for control performance. To select an optimal number of hidden neurons, we progressively add one hidden neuron and observe the change in a performance measure given by the weighted sum of the system error and the control force. The number of hidden neurons which minimizes the performance measure is selected for implementation. A neural network was trained for mitigating vibrations of bridge systems caused by El Centro earthquake. We applied the proposed control approach to a single-degree-of-freedom (SDOF) and a two-degree-of-freedom (TDOF) bridge system. We assessed the robustness of the control system using randomly generated earthquake excitations which were not used in training the neural network. Our results show that the neural network controller drastically mitigates the effect of the disturbance.

• 한국컴퓨터정보학회논문지
• /
• 제21권1호
• /
• pp.1-8
• /
• 2016

Convolutional neural network (CNN) consists of a few pairs of both convolution layer and subsampling layer. Thus it has more hidden layers than multi-layer perceptron. With the increased layers, the size of convolution mask ultimately determines the total number of weights in CNN because the mask is shared among input images. It also is an important learning factor which makes or breaks CNN's learning. Therefore, this paper proposes the best method to choose the convolution size and the number of layers for learning CNN successfully. Through our face recognition with vast learning examples, we found that the best size of convolution mask is 5 by 5 and 7 by 7, regardless of the number of layers. In addition, the CNN with two pairs of both convolution and subsampling layer is found to make the best performance as if the multi-layer perceptron having two hidden layers does.

• 융합신호처리학회논문지
• /
• 제2권1호
• /
• pp.88-94
• /
• 2001

In the position control system, the output of a controller is generally used as the input of a plant but the undesired noise is included in the output of a controller. Therefore, there is a need to use a precompensator for rejecting the undesired noise. In this paper, the expanded PID controller with a precompensator is constructed. The precompensator and PID controller are designed by a neural network with two-hidden layer and these coefficients are changed automatically to be a desired response of system when the response characteristic is changed under a condition.

• 한국항해항만학회:학술대회논문집
• /
• 한국항해항만학회 2006년도 International Symposium on GPS/GNSS Vol.2
• /
• pp.101-106
• /
• 2006

The geoidal undulations are needed for determining the orthometric heights from the Global Positioning System GPS-derived ellipsoidal heights. There are several methods for geoidal undulation determination. The paper presents a method employing a simple architecture Two Layer Multiquadric-Biharmonic Artificial Neural Network (TLMB-ANN) to approximate an area of 4200 square kilometres quasigeoid surface with GPS-levelling data. Hardy’s Multiquadric-Biharmonic functions is used as the hidden layer neurons’ activation function and Levenberg-Marquardt algorithm is used to train the artificial neural network. In numerical examples five surfaces were compared: the gravimetric geometry hybrid quasigeoid, Support Vector Machine (SVM) model, Hybrid Fuzzy Neural Network (HFNN) model, Traditional Three Layer Artificial Neural Network (ANN) with tanh activation function and TLMB-ANN surface approximation. The effectiveness of TLMB-ANN surface approximation depends on the number of control points. If the number of well-distributed control points is sufficiently large, the results are similar with those obtained by gravity and geometry hybrid method. Importantly, TLMB-ANN surface approximation model possesses good extrapolation performance with high precision.

• Industrial Engineering and Management Systems
• /
• 제15권4호
• /
• pp.324-334
• /
• 2016

Manufacturers and retailers are interested in how prices, promotions, discounts and other marketing variables can influence the sales and shares of the products that they produce or sell. Therefore, many models have been developed to predict the brand share. Since the customer choice models are usually used to predict the market share, here we use hybrid model of Probabilistic Neural Network and Artificial Bee colony Algorithm (PNN-ABC) that we have introduced to model consumer choice to predict brand share. The evaluation process is carried out using the same data set that we have used for modeling individual consumer choices in a retail coffee market. Then, to show good performance of this model we compare it with Artificial Neural Network with one hidden layer, Artificial Neural Network with two hidden layer, Artificial Neural Network trained with genetic algorithms (ANN-GA), and Probabilistic Neural Network. The evaluated results show that the offered model is outperforms better than other previous models, so it can be use as an effective tool for modeling consumer choice and predicting market share.

• International Journal of Quality Innovation
• /
• 제3권2호
• /
• pp.113-131
• /
• 2002

The relationships among multi-dimensional data (such as medical examination data) with ambiguity and variation are difficult to explore. The traditional approach to building a data classification system requires the formulation of rules by which the input data can be analyzed. The formulation of such rules is very difficult with large sets of input data. This paper first describes two classification approaches using back-propagation (BP) neural network and Mahalanobis distance (MD) classifier, and then proposes two classification approaches for multi-dimensional feature selection. The first one proposed is a feature selection procedure from the trained back-propagation (BP) neural network. The basic idea of this procedure is to compare the multiplication weights between input and hidden layer and hidden and output layer. In order to simplify the structure, only the multiplication weights of large absolute values are used. The second approach is Mahalanobis-Taguchi system (MTS) originally suggested by Dr. Taguchi. The MTS performs Taguchi's fractional factorial design based on the Mahalanobis distance as a performance metric. We combine the automatic thresholding with MD: it can deal with a reduced model, which is the focus of this paper In this work, two case studies will be used as examples to compare and discuss the complete and reduced models employing BP neural network and MD classifier. The implementation results show that proposed approaches are effective and powerful for the classification.

• Journal of Mechanical Science and Technology
• /
• 제18권3호
• /
• pp.432-442
• /
• 2004

This paper presents a neural network predictor for analysing rigidity variations of hydrostatic bearing system. The designed neural network has feedforward structure with three layers. The layers are input layer, hidden layer and output layer. Two main parameter could be considered for hydrostatic bearing system. These parameters are the size of bearing pocket and the orifice dimension. Due to importancy of these parameters, it is necessary to analyse with a suitable optimisation method such as neural network. As depicted from the results, the proposed neural predictor exactly follows experimental desired results.

• 전자공학회논문지B
• /
• 제28B권11호
• /
• pp.852-858
• /
• 1991

An adaptive hierarchical classifier(AHCL) for Korean character recognition using a neural net is designed. This classifier has two neural nets: USACL (Unsupervised Adaptive Classifier) and SACL (Supervised Adaptive Classifier). USACL has the input layer and the output layer. The input layer and the output layer are fully connected. The nodes in the output layer are generated by the unsupervised and nearest neighbor learning rule during learning. SACL has the input layer, the hidden layer and the output layer. The input layer and the hidden layer arefully connected, and the hidden layer and the output layer are partially connected. The nodes in the SACL are generated by the supervised and nearest neighbor learning rule during learning. USACL has pre-attentive effect, which perform partial search instead of full search during SACL classification to enhance processing speed. The input of USACL and SACL is a directional edge feature with a directional receptive field. In order to test the performance of the AHCL, various multi-font printed Hangul characters are used in learning and testing, and its processing its speed and and classification rate are compared with the conventional LVQ(Learning Vector Quantizer) which has the nearest neighbor learning rule.

• 한국지능시스템학회:학술대회논문집
• /
• 한국퍼지및지능시스템학회 1998년도 The Third Asian Fuzzy Systems Symposium
• /
• pp.762-766
• /
• 1998

Mechanical valve is one of the most widely used implantable artificial organs, Since its failure (mechanical failures and thrombosis to name two representative example) means the death of patient, its reliability is very important and early noninvasive detection is essential requirement . This paper will explain the method to detect the thrombosis formation by spectral analysis and neural network. In order quantitatively to distinguish peak of a normal valve from that of a thrombotic valve, a 3 layer backpropagation neural network, which contains 7,000 input nodes, 20 hidden layer and 1output , was employed. The trained neural network can distinguish normal and thrombotic valve with a probability that is higher than 90% . In conclusion, the acoustical spectrum analysis coupled with a neural network algorithm lent itself to the noninvasive monitoring of implanted mechanical valves. This method will be applied to be applied to the performance evaluation of other implantable rtificial organs.

• 환경영향평가
• /
• 제14권2호
• /
• pp.55-62
• /
• 2005

This study applied the Neural Network and Fuzzy theory to show water-purity control and preventive measure in water quality forecasting of the future river. This study picked out NAJU and HAMPYUNG as the subject of investigation and used monthly the water quality and the outflow data of KWANGJU2, NAJU, YOUNGSANNPO and HAMPYUNG from 1995 to 1999 to forecast BOD, COD, T-N, T-P water density. The datum from 1995 to 1999 are used for study and that of 2000 are used for verification. To develop model of water quality forecasting, firstly, this research formed Neural Network model and divided Neural Network model into two case - the case of considering lag and not considering. And this study selected optimal Neural Network model through changing the number of hidden layer based on input layer(n) from n to 3n. Through forecasting result, the case without considering lag showed more precise simulated result. Accordingly, this study intended to compare, analyse that Fuzzy model using the method without considering lag with Neural Network model. As a result, this study found that the model without considering lag in Neural Network Network shows the most excellent outcome. Thus this study examined a forecasting accuracy, analyzed result and verified propriety through appling the method of water quality forecasting using Neural Network and Fuzzy Algorithms to the actual case.