• Transactions on Control, Automation and Systems Engineering
• /
• 제4권2호
• /
• pp.124-129
• /
• 2002

This paper demonstrates that the largest Lyapunov exponent λ of recurrent neural networks can be controlled efficiently by a stochastic gradient method. An essential core of the proposed method is a novel stochastic approximate formulation of the Lyapunov exponent λ as a function of the network parameters such as connection weights and thresholds of neural activation functions. By a gradient method, a direct calculation to minimize a square error (λ - λ$\^$obj/)$^2$, where λ$\^$obj/ is a desired exponent value, needs gradients collection through time which are given by a recursive calculation from past to present values. The collection is computationally expensive and causes unstable control of the exponent for networks with chaotic dynamics because of chaotic instability. The stochastic formulation derived in this paper gives us an approximation of the gradients collection in a fashion without the recursive calculation. This approximation can realize not only a faster calculation of the gradient, but also stable control for chaotic dynamics. Due to the non-recursive calculation. without respect to the time evolutions, the running times of this approximation grow only about as N$^2$ compared to as N$\^$5/T that is of the direct calculation method. It is also shown by simulation studies that the approximation is a robust formulation for the network size and that proposed method can control the chaos dynamics in recurrent neural networks efficiently.

• 전자공학회논문지C
• /
• 제36C권8호
• /
• pp.1-9
• /
• 1999

빠르고 정확한 결과를 얻기 위해서 타이밍 수준에서의 회로 해석이 이루어지며, 게이트와 연결선에서의 신호 지연 해석은 회로의 설계 검증을 위하여 필수적이다. 본 논문에서는 CMOS 회로 게이트에서의 지연 시간과 연결선의 지연 해석을 위한 초기 천이 시간을 동시에 계산할 수 있는 방법을 제시한다. 회로 연결선의 유효 커패시턴스 개념을 이용하여 게이트의 지연 시간과 게이트에서의 구동 저항을 고려한 연결선 선형 전압원의 천이 시간을 계산한다. 게이트 지연과 연결선 선형 전압원의 천이 시간을 구하는 과정은 예비 특성화된 게이트 타이밍 데이터를 이용하여 반복적인 연산과정을 통하여 동시에 구하게 된다. 기존의 게이트 지연 계산 알고리즘은 연결선 선형 전압원의 천이 시간을 위해 별도의 게이트 특성 데이터를 필요로 하였으나, 본 논문에서 제시하는 방법은 계산 과정 중에 생성된 데이터를 이용함으로써 현재의 예비 특성화 방법을 수정하지 않고서도 효율적인 타이밍 수준의 게이트 및 연결선 지연 시간 예측이 가능하도록 하였다.

• 한국항행학회논문지
• /
• 제24권5호
• /
• pp.374-381
• /
• 2020

수직위치 정확도를 개선하기 위해 GPS와 기압고도계의 장점을 융합하여 사용하지만, 두 센서를 융합하기 위해서는 지오이드 고도를 보상해 주어야 한다. 본 논문에서는 드론이나 자율주행 차량에 적용되는 저가의 내장형 시스템에서도 실시간으로 지오이드 고도를 계산할 수 있는 기법을 제안하였다. 기준이 되는 EGM08은 2160차의 다항식으로 결정되기 때문에 내장형 시스템에서 실시간 계산이 불가능하다. 때문에 선형 보간 기법을 도입하여 계산량을 줄이고, 정수형 지오이드 고도를 격자점으로 사용하여 저장공간을 75%절감하였다. 시뮬레이션을 통해 제안된 기법의 정확도를 평가하였으며, 지오이드 변화가 급격한 지역에서도 최대 오차 -1.215 m의 정확도를 갖는 것을 확인하였다.

• Nuclear Engineering and Technology
• /
• 제53권2호
• /
• pp.368-375
• /
• 2021

The performance time of human operators has been recognized as a key aspect of human reliability in socio-complex systems, including nuclear industries. Because of the importance of the time factor, most existing human reliability assessment methods provide ways to quantify human error probabilities (HEPs) that are associated with the performance time. To quantify such kinds of HEPs, it is crucial to rationally predict the length of time required and time available and compare them. However, there have not been detailed guidelines that identify the critical cue presentation time or initial time of human performance, which is important to calculate the time information. In this paper, we introduce a time-related HEP calculation technique with a decision algorithm that determines the critical cue and its timing. The calculation process is presented with the application examples. It is expected that the proposed algorithm will reduce the variability in the time-related reliability assessment and strengthen the scientific evidence of the assessment process. The detailed description is provided in the technical report KAERI/TR-7607/2019.

• 한국철도학회:학술대회논문집
• /
• 한국철도학회 2010년도 춘계학술대회 논문집
• /
• pp.1685-1690
• /
• 2010

This paper describes the calculation and application of the speed profile for the train speed control when the train is operated by ATO(Automatic Train Operation). We propose a speed profile calculation method considering the speed limitation and the jerk limitation, in order to maintain the quality of the train automatic operation. In previous works, ATO calculates a desired speed profile along the time, and controls the train to follow the profile. In this case, it may be hard to follow the restrictive speed along the location or to stop the train with a low location error, because of the difference between the desired location at each time and the real location of the train due to the control error. In the proposed method in this paper, we calculate a desired speed profile along the time considering the speed and jerk limitation first, and derive a speed profile along the location using it. If the restrictive speed profile is changed or the train speed strayed from the speed profile, ATO system calculates new speed profile and applies it immediately. Because ATO system controls the train speed based on the train location, the accuracy of the train location control can be improved. A simulation system for the test of the automatic train operation using this method is designed.

• 한국통신학회논문지
• /
• 제41권9호
• /
• pp.1141-1145
• /
• 2016

일반적으로 RLS 알고리즘에서 비정재성(non-stationary) 환경에서 시간에 따라 변하는 파라메터를 좀 더 잘 추정하기 위해서 가변 망각인자를 사용한다. RLS 알고리즘에서 가변 망각인자를 사용할 때는 연산량이 많이 증가하는 단점이 수반된다. 본 논문에서는 연산량이 적은 가우스 뉴턴 가변망각인자 RLS 알고리즘을 제안한다. 본 방법은 기존 가우스 뉴턴 가변망각인자 RLS와 거의 유사한 성능을 보유하고 있을 뿐만 아니라 부가로 요구되는 연산량을 $O(N^2)$에서 O(N)으로 줄이는 효과도 준다.

• 한국공간구조학회논문집
• /
• 제22권2호
• /
• pp.39-46
• /
• 2022

Recently, machine learning is widely used to solve optimization problems in various engineering fields. In this study, machine learning is applied to development of a control algorithm for a smart control device for reduction of seismic responses. For this purpose, Deep Q-network (DQN) out of reinforcement learning algorithms was employed to develop control algorithm. A single degree of freedom (SDOF) structure with a smart tuned mass damper (TMD) was used as an example structure. A smart TMD system was composed of MR (magnetorheological) damper instead of passive damper. Reward design of reinforcement learning mainly affects the control performance of the smart TMD. Various hyper-parameters were investigated to optimize the control performance of DQN-based control algorithm. Usually, decrease of the time step for numerical simulation is desirable to increase the accuracy of simulation results. However, the numerical simulation results presented that decrease of the time step for reward calculation might decrease the control performance of DQN-based control algorithm. Therefore, a proper time step for reward calculation should be selected in a DQN training process.

• 설비공학논문집
• /
• 제28권8호
• /
• pp.331-336
• /
• 2016

This study presents a coolant density calculation device and its corresponding method by using a mass flowmeter and the LabVIEW program. The method can be easily measured with a mixture of coolant and by calculating the percentage of ethylene-glycol without additional investment. The cooling water is very important in a vehicle to protect the engine, and the cooling performance is affected by the mixture concentration and coolant density. The coolant density calculation device measures the mixed concentration in the anti-freeze cooling mixture made from distilled water and ethylene-glycol in real time with the mass flowmeter that is commonly attached to the radiator or heater core. The calculation program for the mixture concentration percentage was developed using the LabVIEW software. The correlation between experimental results and the calculation was conducted for a range of temperature from 40 to $90^{\circ}C$ and by varying the mixture ratio of distilled water and ethylene-glycol. As a result, the anti-freeze coolant concentration in the volume percentage is able to monitor the coolant density in a timely basis by implementing a mixture concentration calculation program without the need for additional equipment investment. The results of the calculation for the mixture concentration level show a maximum 2.7% deviation compared to the experimental results.

• 소음진동
• /
• 제9권4호
• /
• pp.778-784
• /
• 1999

A method for obtaining the motion of an impact system whose primary and secondary system are composed of lumped masses, springs and dampers, and all the contacts are made through spring and damping elements is presented. The frequency response functions derived from the equations of motion and the impulse response functions obtained from the inverse Fourier transform of the derived frequency response functions are used for the calculation of the system responses. The procedure developed for the calculation of displacements and force time-histories was based on the convolution integrals of impulse response functions and forces applied to the systems. Time histories of displacements and contact forces are obtained for the case where a random excitation is applied to a point in the system. Impact statistics such as contact forces and the time between impacts calculated from those time histories is presented.

• 한국멀티미디어학회:학술대회논문집
• /
• 한국멀티미디어학회 2003년도 추계학술발표대회(하)
• /
• pp.1006-1009
• /
• 2003

In a motion estimation method for use in encoding a moving picture, a full-pixel motion vector is estimated by stochastically sampling a pixel to be processed in a predetermined-sized block of a previous frame or a next frame as a reference frame for each of a plurality of equal-sized blocks in a current frame. Then, a half-pixel motion vector is estimated based on the full-pixel motion vector. Accordingly, both the calculation amount and the calculation time required for the motion estimation are effectively reduced. Further, it can be prevented that the hardware becomes complicated. .