• 제어로봇시스템학회:학술대회논문집
• /
• 1995.10a
• /
• pp.187-190
• /
• 1995

A method is proposed which searches for optimal structures of Neural Networks (NN) using Genetic Algorithm (GA). The purpose of the method lies in not only finding an optimal NN structure but also leading us to the goal of self-organized control system that acquires its structure and its functionality by itself depending on its environment.

• Korean Journal of Remote Sensing
• /
• v.37 no.5_1
• /
• pp.927-938
• /
• 2021

A similar magnitude of non-gravitational perturbations are act on the formation flying low earth orbit satellites with a certain time difference. Using this temporal correlation, the non-gravity acceleration of the low earth orbiting satellites can be transferred for the othersatellites. There is a period in which the accelerometer data of one satellite is unavailable for GRACE and GRACE-FO satellites. In this case, the accelerometer data transplant method described above is officially used to recover the accelerometer data at the Jet Propulsion Laboratory (JPL). In this paper, we proposed a model for predicting accelerometer data of formation flying low earth orbit satellites using a neural network (NN) model to improve the estimation accuracy of the transplant method. Although the transplant method cannot reflect the satellite's position and space environmental factors, the NN model can use them as model inputs to increase the prediction accuracy. A prediction test of an accelerometer data using NN model was performed for one month, and the prediction accuracy was compared with the transplant method. The NN model outperformsthe transplant method with 55.0% and 40.1% error reduction in the along-track and radial directions, respectively.

• Structural Engineering and Mechanics
• /
• v.46 no.5
• /
• pp.673-693
• /
• 2013

This paper presents a systematic design and training procedure for the feed-forward back-propagation neural network (NN) modeling of both forward and inverse behavior of a rotary magnetorheological (MR) damper based on experimental data. For the forward damper model, with damper force as output, an optimization procedure demonstrates accurate training of the NN architecture with only current and velocity as input states. For the inverse damper model, with current as output, the absolute value of velocity and force are used as input states to avoid negative current spikes when tracking a desired damper force. The forward and inverse damper models are trained and validated experimentally, combining a limited number of harmonic displacement records, and constant and half-sinusoidal current records. In general the validation shows accurate results for both forward and inverse damper models, where the observed modeling errors for the inverse model can be related to knocking effects in the measured force due to the bearing plays between hydraulic piston and MR damper rod. Finally, the validated models are used to emulate pure viscous damping. Comparison of numerical and experimental results demonstrates good agreement in the post-yield region of the MR damper, while the main error of the inverse NN occurs in the pre-yield region where the inverse NN overestimates the current to track the desired viscous force.

• The Transactions of The Korean Institute of Electrical Engineers
• /
• v.57 no.5
• /
• pp.852-857
• /
• 2008

A new adaptive neuro-control algorithm for a SISO strict-feedback nonlinear system is proposed. All the previous adaptive neural control algorithms for strict-feedback nonlinear systems are based on the backstepping scheme, which makes the control law and stability analysis very complicated. The main contribution of the proposed method is that it demonstrates that the state-feedback control of the strict-feedback system can be viewed as the output-feedback control problem of the system in the normal form. As a result, the proposed control algorithm is considerably simpler than the previous ones based on backstepping. Depending heavily on the universal approximation property of the neural network (NN), only one NN is employed to approximate the lumped uncertain system nonlinearity. The Lyapunov stability of the NN weights and filtered tracking error is guaranteed in the semi-global sense.

• International Journal of Control, Automation, and Systems
• /
• v.6 no.5
• /
• pp.639-650
• /
• 2008

Electricity price forecasting has become an integral part of power system operation and control. In this paper, a wavelet transform (WT) based neural network (NN) model to forecast price profile in a deregulated electricity market has been presented. The historical price data has been decomposed into wavelet domain constitutive sub series using WT and then combined with the other time domain variables to form the set of input variables for the proposed forecasting model. The behavior of the wavelet domain constitutive series has been studied based on statistical analysis. It has been observed that forecasting accuracy can be improved by the use of WT in a forecasting model. Multi-scale analysis from one to seven levels of decomposition has been performed and the empirical evidence suggests that accuracy improvement is highest at third level of decomposition. Forecasting performance of the proposed model has been compared with (i) a heuristic technique, (ii) a simulation model used by Ontario's Independent Electricity System Operator (IESO), (iii) a Multiple Linear Regression (MLR) model, (iv) NN model, (v) Auto Regressive Integrated Moving Average (ARIMA) model, (vi) Dynamic Regression (DR) model, and (vii) Transfer Function (TF) model. Forecasting results show that the performance of the proposed WT based NN model is satisfactory and it can be used by the participants to respond properly as it predicts price before closing of window for submission of initial bids.

• International Journal of Fuzzy Logic and Intelligent Systems
• /
• v.1 no.1
• /
• pp.95-103
• /
• 2001

The purpose of Introducing a combined cycle with gas turbine in power plants is to reduce losses of energy, by effectively using exhaust gases from the gas turbine to produce additional electricity or process. The efficiency of a combined power plant with the gas turbine increases, exceeding 50%, while the efficiency of traditional steam turbine plants is approximately 35% to 40%. Up to the present time, the PID controller has been used to operate this system. However, it is very difficult to achieve an optimal PID gain without any experience, since the gain of the PID controller has to be manually tuned by trial and error procedures. This paper focuses on the neural network tuning of the 2-DOF PID controller with a combined 2-DOF parameter (NN-Tuning 2-DOF PID controller), for optimal control of the Gun-san gas turbine generating plant in Seoul, Korea. In order to attain optimal control, transfer function and operating data from start-up, running, and stop procedures of the Gun-san gas turbine have been acquired and a designed controller has been applied to this system. The results of the NN-Tuning 2-DOF PID are compared with the PID controller and the conventional 2-DOF PID controller tuned by the Ziegler-Nichols method through experimentation. The experimental results of the NN-Tuning 2-DOF PID controller represent a more satisfactory response than those of the previously-mentioned two controllers.

• Computers and Concrete
• /
• v.28 no.6
• /
• pp.635-642
• /
• 2021

In this paper, soft learning techniques are used to predict the ultimate torsional capacity of Reinforced Concrete beams strengthened with Fiber Reinforced Polymer. Soft computing techniques, namely Artificial Neural Network, trained by various back propagation algorithms, and Particle Swarm Optimization (PSO) algorithm, have been used to model and predict the torsional strength of Reinforced Concrete beams strengthened with Fiber Reinforced Polymer. The performance of each model has been evaluated by using statistical parameters such as coefficient of determination (R²), Root Mean Square Error (RMSE), and Mean Absolute Percentage Error (MAPE). The hybrid PSO NN model resulted in an R² of 0.9292 with an RMSE of 5.35 for training and an R² of 0.9328 with an RMSE of 4.57 for testing. Another model, ANN BP, produced an R² of 0.9125 with an RMSE of 6.17 for training and an R² of 0.8951 with an RMSE of 5.79 for testing. The results of the PSO NN model were in close agreement with the experimental values. Thus, the PSO NN model can be used to predict the ultimate torsional capacity of RC beams strengthened with FRP with greater acceptable accuracy.

• Proceedings of the Korean Institute of Intelligent Systems Conference
• /
• 1993.06a
• /
• pp.1021-1024
• /
• 1993

System identification is a major component for a control system. In biosystems, which is nonlinear and dynamic, precise identification would be very helpful for implementing a control system. It is difficult to precisely identify such non-linear systems. The measurable data on products from 2,3-butanediol fermentation could not be included in a process model based on kinetic approach. Meanwhile, a predictive capability is required in developing a control system. A neural network (NN) dynamic identifier with a by/(1＋ t ) transfer function was therefore designed being able to predict this fermentation. This modified inverse NN identifier differs from traditional models in which it is not only able to see but also able to predict the system. A moving window, with a dimension of 11 and a fixed data size of seven, was properly designed. One-step ahead identification/prediction by an 11-3-1 BPNN is demonstrated. Even under process fault, this neural network is still able to perform several-step ahead prediction.

• Biotechnology and Bioprocess Engineering:BBE
• /
• v.9 no.3
• /
• pp.156-165
• /
• 2004

Concentrations of substrates, glucose, and ammionia in biological processes have been on-line monitored by using glucose-flow injection (FIA) and ammonia-FIA systems. Based on the on-line monitored data the concentrations of substrates have been controlled by an on-off controller, a PID controller, and a neural network (NN) based controller. A simulation program has been developed to test the control quality of each controller and to estimate the control parameters. The on-off controller often produced high oscillations at the set point due to its low robustness. The control quality of a PID controller could have been improved by a high analysis frequency and by a short residence time of sample in a FIA system. A NN-based controller with 3 layers has been developed, and a 3(input)-2(hidden)-1(output) network structure has been found to be optimal for the NN-based controller. The performance of the three controllers has been tested in a simulated process as well as in a cultivation process of Saccharomyces cerevisiae, and the performance has also been compared to simulation results. The NN-based controller with the 3-2-1 network structure was robust and stable against some disturbances, such as a sudden injection of distilled water into a biological process.

• Journal of the Korean Institute of Intelligent Systems
• /
• v.14 no.7
• /
• pp.830-837
• /
• 2004

Neural networks-based fault diagnosis algorithm to detect and isolate faults in the nonlinear systems is proposed. In the proposed method, the fault is detected when the errors between the system output and the multilayer neural network-based nominal model output cross a Predetermined threshold. Once a fault in the system is detected, the system outputs are transferred to the fault classifier by nultilayer/ART2 NN (adaptive resonance theory 2 neural network) for fault isolation. From the computer simulation results, it is verified that the proposed fault diagonal method can be performed successfully to detect and isolate faults in a nonlinear system.