• ETRI Journal
• /
• v.44 no.2
• /
• pp.194-207
• /
• 2022

In the last decade, substantial progress has been achieved in intelligent traffic control technologies to overcome consistent difficulties of traffic congestion and its adverse effect on smart cities. Edge computing is one such advanced progress facilitating real-time data transmission among vehicles and roadside units to mitigate congestion. An edge computing-based deep reinforcement learning system is demonstrated in this study that appropriately designs a multiobjective reward function for optimizing different objectives. The system seeks to overcome the challenge of evaluating actions with a simple numerical reward. The selection of reward functions has a significant impact on agents' ability to acquire the ideal behavior for managing multiple traffic signals in a large-scale road network. To ascertain effective reward functions, the agent is trained withusing the proximal policy optimization method in several deep neural network models, including the state-of-the-art transformer network. The system is verified using both hypothetical scenarios and real-world traffic maps. The comprehensive simulation outcomes demonstrate the potency of the suggested reward functions.

• Journal of Internet of Things and Convergence
• /
• v.9 no.3
• /
• pp.93-98
• /
• 2023

In this paper, an artificial neural network based on simulated annealing was constructed. The mapping relationship between the parameters of micro-scaled flow channels electrochemical machining and the channel shape was established by training the samples. The depth and width of micro-scaled flow channels electrochemical machining on stainless steel surface were predicted, and the flow channels experiment was carried out with pulse power supply in NaNO3 solution to verify the established network model. The results show that the depth and width of the channel predicted by the simulated annealing artificial neural network with "4-7-2" structure are very close to the experimental values, and the error is less than 5.3%. The predicted and experimental data show that the etching degree in the process of channels electrochemical machining is closely related to voltage and current density. When the voltage is less than 5V, a "small island" is formed in the channel; When the voltage is greater than 40V, the lateral etching of the channel is relatively large, and the "dam" between the channels disappears. When the voltage is 25V, the machining morphology of the channel is the best.

• Journal of the Korean Society for Precision Engineering
• /
• v.28 no.4
• /
• pp.482-489
• /
• 2011

In practical design process, designer needs to find an optimal solution by using full factorial discrete combination, rather than by using optimization algorithm considering continuous design variables. So, ANOVA(Analysis of Variance) based on an orthogonal array, i.e. Taguchi method, has been widely used in most parts of industry area. However, the Taguchi method is limited for the shape optimization by using CAE, because the multi-level and multi-objective optimization can't be carried out simultaneously. In this study, a combined method was proposed taking into account of multi-level computational orthogonal array and TOPSIS(Technique for Order preference by Similarity to Ideal Solution), which is known as a classical method of multiple attribute decision making and enables to solve various decision making or selection problems in an aspect of multi-objective optimization. The proposed method was applied to a case study of the multi-level shape optimization of lower arm used to automobile parts, and the design space was explored via an efficient application of the related CAE tools. The multi-level shape optimization was performed sequentially by applying both of the neural network model generated from seven-level four-factor computational orthogonal array and the TOPSIS. The weight and maximum stress of the lower arm, as the objective functions for the multi-level shape optimization, showed an improvement of 0.07% and 17.89%, respectively. In addition, the number of CAE carried out for the shape optimization was only 55 times in comparison to full factorial method necessary to 2,401 times.

• Journal of the Korean Society for Aeronautical & Space Sciences
• /
• v.37 no.4
• /
• pp.350-358
• /
• 2009

This paper describes a result of the guidance and control for re-entry vehicle during TAEM phase. TAEM phase (Terminal Aerial Energy Management phase) has many conditions, such as density, velocity, and so on. Under these conditions, we have optimized trajectory and other states for guidance in TAEM phase. The optimized states consist of 7 variables, down-range, cross range, altitude, velocity, flight path angle, vehicle's azimuth and flight range. We obtained the optimized reference trajectory by DIDO tool, and used feedback linearization with neural network for control re-entry vehicle. By back propagation algorithm, vehicle dynamics is approximated to real one. New command can be decided using the approximated dynamics, delayed command input and plant output, NARMA-L2. The result by this control law shows a good performance of tracking onto the reference trajectory.

• The Journal of the Acoustical Society of Korea
• /
• v.21 no.6
• /
• pp.522-530
• /
• 2002

This paper introduces a neural network (NN) -based nonlinear predictor for the LP (Linear Prediction) residual. To evaluate the effectiveness of the NN-based nonlinear predictor for LP-residual, we first compared the average prediction gain of the linear long-term predictor with that of the NN-based nonlinear long-term predictor. Then, the effects on the quantization noise of the nonlinear prediction residuals were investigated for the NN-based nonlinear predictor A new NN predictor takes into consideration not only prediction error but also quantization effects. To increase robustness against the quantization noise of the nonlinear prediction residual, a constrained back propagation learning algorithm, which satisfies a Kuhn-Tucker inequality condition is proposed. Experimental results indicate that the prediction gain of the proposed NN predictor was not seriously decreased even when the constrained optimization algorithm was employed.

• Bulletin of the Korean Chemical Society
• /
• v.26 no.12
• /
• pp.2007-2016
• /
• 2005

An artificial neural network (ANN) is successfully presented for prediction acidity constant (pKa) of various benzoic acids and phenols with diverse chemical structures using a nonlinear quantitative structure-property relationship. A three-layered feed forward ANN with back-propagation of error was generated using six molecular descriptors appearing in the multi-parameter linear regression (MLR) model. The polarizability term $(\pi_1)$, most positive charge of acidic hydrogen atom $(q^+)$, molecular weight (MW), most negative charge of the acidic oxygen atom $(q^-)$, the hydrogen-bond accepting ability $(\epsilon_B)$ and partial charge weighted topological electronic (PCWTE) descriptors are inputs and its output is pKa. It was found that properly selected and trained neural network with 205 compounds could fairly represent dependence of the acidity constant on molecular descriptors. For evaluation of the predictive power of the generated ANN, an optimized network was applied for prediction pKa values of 37 compounds in the prediction set, which were not used in the optimization procedure. Squared correlation coefficient $(R^2)$ and root mean square error (RMSE) of 0.9147 and 0.9388 for prediction set by the MLR model should be compared with the values of 0.9939 and 0.2575 by the ANN model. These improvements are due to the fact that acidity constant of benzoic acids and phenols in water shows nonlinear correlations with the molecular descriptors.

• KIEAE Journal
• /
• v.16 no.3
• /
• pp.89-94
• /
• 2016

Purpose: This study aimed at developing an artificial neural network (ANN) model to predict the optimal start moment of the setback temperature during the normal occupied period of a building. Method: For achieving this objective, three major steps were conducted: the development of an initial ANN model, optimization of the initial model, and performance tests of the optimized model. The development and performance testing of the ANN model were conducted through numerical simulation methods using transient systems simulation (TRNSYS) and matrix laboratory (MATLAB) software. Result: The results analysis in the development and test processes revealed that the indoor temperature, outdoor temperature, and temperature difference from the setback temperature presented strong relationship with the optimal start moment of the setback temperature; thus, these variables were used as input neurons in the ANN model. The optimal values for the number of hidden layers, number of hidden neurons, learning rate, and moment were found to be 4, 9, 0.6, and 0.9, respectively, and these values were applied to the optimized ANN model. The optimized model proved its prediction accuracy with the very storing statistical correlation between the predicted values from the ANN model and the simulated values in the TRNSYS model. Thus, the optimized model showed its potential to be applied in the control algorithm.

• Transactions of the Korean Society of Mechanical Engineers A
• /
• v.37 no.11
• /
• pp.1397-1405
• /
• 2013

The optimal rib pattern design of B-pillar trim considering occupant head protection can be determined by two methods. One is the conventional approximate optimization method that uses the statistical design of experiments (DOE) and response surface method (RSM). Generally, approximated optimum results are obtained through the iterative process by trial-and-error. The quality of results strongly depends on the factors and levels assigned by a designer. The other is a methodology derived from previous work by the authors, called the sequential design of experiments (SDOE), to reduce the trial-and-error procedure and to find an appropriate condition for using artificial neural network (ANN) systematically. An appropriate condition is determined from the iterative process based on the analysis of means. With this new technique and ANN, it is possible to find an optimum design accurately and efficiently.

• Journal of Institute of Control, Robotics and Systems
• /
• v.3 no.6
• /
• pp.644-651
• /
• 1997

In the water purification plant, chemicals are injected for quick purification of raw water. It is clear that the amount of chemicals intrinsically depends on water quality such as turbidity, temperature, pH and alkalinity. However, the process of chemical reaction to improve water quality (e.g., turbidity) by chemicals is not yet fully clarified nor quantified. The feedback signal in the process of coagulant dosage, which should be measured (through the sensor of the plant) to compute the appropriate amount of chemicals, is also not available. Most traditional methods focus on judging the conditions of purifying reaction and determine the amounts of chemicals through manual operation of field experts using Jar-test data. In this paper, a systematic control strategy is proposed to derive the optimum dosage of coagulant, PAC(Polymerized Aluminium Chloride), using Jar-test results. A neural network model is developed for coagulant dosing and purifying process by means of six input variables (turbidity, temperature, pH, alkalinity of raw water, PAC feed rate, turbidity in flocculation) and one output variable, while considering the relationships to the reaction of coagulation and flocculation. The model is utilized to derive the optimum coagulant dosage (in the sense of minimizing turbidity of water in flocculator). The ability of the proposed control scheme validated through the field test has proved to be of considerable practical value.

• Journal of the Korean Institute of Intelligent Systems
• /
• v.27 no.1
• /
• pp.22-28
• /
• 2017

The Convolutional Neural Network(CNN) can be divided into two stages: feature extraction and classification. The hyperparameters such as kernel size, number of channels, and stride in the feature extraction step affect the overall performance of CNN as well as determining the structure of CNN. In this paper, we propose a method to optimize the hyperparameter in CNN feature extraction stage using Parameter-Setting-Free Harmony Search (PSF-HS) algorithm. After setting the overall structure of CNN, hyperparameter was set as a variable and the hyperparameter was optimized by applying PSF-HS algorithm. The simulation was conducted using MATLAB, and CNN learned and tested using mnist data. We update the parameters for a total of 500 times, and it is confirmed that the structure with the highest accuracy among the CNN structures obtained by the proposed method classifies the mnist data with an accuracy of 99.28%.