• 대한전기학회:학술대회논문집
• /
• 대한전기학회 2007년도 제38회 하계학술대회
• /
• pp.1680-1681
• /
• 2007

This paper deals with RLS algorithm using Newton-Raphson method based adaptive forgetting factor for a passive telemetry RF sensor system in order to estimate the time variant parameter to be included in RF sensor model.

• 전기전자학회논문지
• /
• 제25권3호
• /
• pp.412-419
• /
• 2021

본 논문에서는, 지속적 학습 환경에서 효율적 경로 선택에 의한 LwF방법의 성능향상을 제안한다. 이를 위해 콘볼루션 레이어를 분리하는 방법을 사용하여 기존의 LwF와 성능 및 구조를 비교한다. 비교를 위해 복잡도가 다른 구성을 가진 MNIST, EMNIST, Fashion MNIST, CIFAR10 데이터를 사용하여 성능을 실험하였다. 실험결과, 각 태스크 별 정확도가 최대 20% 향상되었으며, LwF 기반의 지속적 학습 환경에서 치명적 망각 현상이 개선되었다.

• 드라이브 ㆍ 컨트롤
• /
• 제14권2호
• /
• pp.16-22
• /
• 2017

This paper presents the algorithm of an adaptive model-free-control-based steering control for multi-axle all-terrain cranes for which the recursive least squares with forgetting are applied. To optimally control the actual system in the real world, the linear or nonlinear mathematical model of the system should be given for the determination of the optimal control inputs; however, it is difficult to derive the mathematical model due to the actual system's complexity and nonlinearity. To address this problem, the proposed adaptive model-free controller is used to control the steering angle of a multi-axle crane. The proposed model-free control algorithm uses only the input and output signals of the system to determine the optimal inputs. The recursive least-squares algorithm identifies first-order systems. The uncertainty between the identified system and the actual system was estimated based on the disturbance observer. The proposed control algorithm was used for the steering control of a multi-axle crane, where only the steering input and the desired yaw rate were employed, to track the reference path. The controller and performance evaluations were constructed and conducted in the Matlab/Simulink environment. The evaluation results show that the proposed adaptive model-free-control-based steering-control algorithm produces a sound path-tracking performance.

• 한국정보통신학회논문지
• /
• 제26권3호
• /
• pp.347-354
• /
• 2022

지속적 학습에서의 망각현상을 완화시키기 위해, 본 논문에서는 지식전달 방법에 기반한 개선된 LwF 모델을 제안하고, 이의 효율성을 실험 결과로 보인다. LwF에 지속적 학습을 적용할 경우, 학습되는 데이터의 도메인이 달라지거나 데이터의 복잡도가 달라지면, 이전에 학습된 결과는 망각현상에 의해 정확도가 떨어지게 된다. 특히 복잡한 데이터에서 단순한 데이터로 학습이 이어질 경우 그 현상이 더 심해지는 경향이 있다. 본 논문에서는 이전 학습 결과가 충분히 LwF 모델에 전달되게 하기 위해 지식전달 방법을 적용하고, 효율적인 사용을 위한 알고리즘을 제안한다. 그 결과 기존 LwF의 결과보다 평균 8% 정도의 망각현상 완화를 보였으며, 학습 태스크가 길어지는 경우에도 효과가 있었다. 특히, 복잡한 데이터가 먼저 학습된 경우에는 LwF 대비 최대 30% 이상 효율이 향상되었다.

• 한국게임학회 논문지
• /
• 제15권3호
• /
• pp.151-160
• /
• 2015

본 논문에서는 망각곡선을 응용한 반복학습시기 설정 및 학습자 수준과 단어 난이도 자동계산 방법이 적용된 영어단어 암기시스템을 소개한다. 우리 시스템은 망각곡선을 사용해서 학습자의 단어암기 횟수에 따라 적절한 반복 주기를 정하고 그 시기에 복습을 요구한다. 학습자가 알고 있는 단어들에 대한 복습시간을 없애고 잊어버릴 확률이 가장 높은 단어들을 우선 적으로 복습하는 것으로 시간을 절약 할 수 있는 것이다. 또 수준에 맞는 난이도의 단어들을 제공함으로써 학습 흥미와 성취도 유지에 기여할 수 있다. 학습자의 수준을 고려하지 않은 난이도의 단어를 무작위로 제공하거나 이미 다른 사람들의 평가에 맞춰진 난이도의 단어들을 제공하는 기존의 시스템과 차별되도록 학습자와 단어 난이도 설정에 온라인 체스게임 랭킹시스템에서 사용하고 있는 "Glicko" 시스템을 적용시켰다. 플레이어간의 대결을 통해서 서로의 실력이 결정되고 매칭되는 이 시스템을 우리는 단어와 사람간의 대결로 시스템에 적용하였다. 그것으로 인해 학습하는 사람의 수준과 단어들의 난이도가 실시간으로 측정되고 학습과정에 반영이 된다. 이 외에 부가적으로 분산학습, 시험형식의 문제풀이의 즉각적인 피드백을 활용하여 영어 단어 암기의 효율성을 극대화 한다.

• 전기학회논문지
• /
• 제59권11호
• /
• pp.2058-2065
• /
• 2010

This paper presents the sensor location and step estimation algorithm for personal navigation system (PNS). PNS has the disadvantage in that the position of the sensor must be fixed on a human body. Three-axis acceleration sensor is used to solve the disadvantage and to consider the real situation. We simplify the measurement data by using the band pass filter, witch It has the advantage in the detection of characteristic point. Through the detected characteristic points, it is possible to setup the parameter for the pattern detection. Depending on the sensor location, the parameters have the different type of noise covariance. Particularly, when the position of the sensor is changed, the impulse noise shows up. Considering the noise, we apply the recursive least square algorithm using the variable forgetting factors, which can classify the sensor location based on the estimated parameters. We performed the experiment for the verification of the proposed algorithm in the various environments. Through the experimental results, the effectiveness of the proposed method is verified.

• 대한전기학회:학술대회논문집
• /
• 대한전기학회 1999년도 하계학술대회 논문집 C
• /
• pp.1063-1066
• /
• 1999

This paper presents an extended integral control with the PID controller by introducing the delay and decaying factors. The convolution integral control scheme is developed by substituting proportional convolution integral controls for the proportional-integral control. So far, the integral part of the PI controller produces a signal that is proportional to the time integral of the input of the controller. The steady-state operation points are affected forever by the errors in the past due to the input signal containing the information of the errors in the past. These phenomina may cause some disturbances for other control purposes related to the given PI control. Introduction of forgetting factors of the error in the past can resolve the disturbance problems. Various forgetting factors are developed using the delay, the decaying factors, and the combination of the delay and the decaying factors. The proposed various extended integral control schemes can be applicable to corresponding PI control designs in which the error in the past may badly affect to the current steady-state operation points and may cause some disturbances for other control purposes.

• 대한전기학회논문지:전력기술부문A
• /
• 제49권7호
• /
• pp.345-349
• /
• 2000

This paper presents an extended integral control with the PI controller by introducing the delay and decaying factors. The extended integral control scheme is developed by substituting the proportional convolution integral control for the PI(Proportional Integral) control. So far, the integral part of PI controller produces a signal that is proportional to the time integral of the input signal to the controller. The steady-state operation points are affected forever by errors in the past due to the input signal containing the information of the error in the past. These phenomena may cause some disturbances for other control purposes related to the given PI control. Introduction of forgetting factors to the error in the past can resolve the disturbance problems. Various forgetting factors are developed using the delay elements, the decaying factors, and the combination of the delay and decaying factors. The proposed various extended integral control schemes can be applicable to the corresponding PI control designs in which the error in the past may badly affect the current steady-state operation points and may cause some disturbances for other control purposes.

• 전기학회논문지
• /
• 제60권12호
• /
• pp.2246-2253
• /
• 2011

This paper presents the error correction method of magnetic position sensor using recursive least square method (RLSM) with forgetting factor. Magnetic position sensor is proposed for linear position detection of the linear motor which has tooth shape stator, consists of permanent magnet, iron core and linear hall sensor, and generates sine and cosine waveforms according to the movement of the mover of the linear motor. From the output of magnetic position sensor, the position of the linear motor can be detected using arc-tan function. But the variation of the air gap between magnetic position sensor and the stator and the error in manufacturing process can cause the variation in offset, phase and amplitude of the generated waveforms when the linear motor moves. These variations in sine and cosine waveforms are changed according to the current linear motor position, and it is very difficult to compensate the errors using constant value. In this paper, the generated sine and cosine waveforms from the magnetic position sensor are compensated on-line using the RLSM with forgetting factor. And the speed observer is introduced to reduce the effect of uncompensated harmonic component. The approaches are verified by some simulations and experiments.

• The Journal of the Acoustical Society of Korea
• /
• 제25권4E호
• /
• pp.154-158
• /
• 2006

In this paper, we suggest a noise estimation method for speech enhancement in nonstationary noisy environments. The proposed method consists of the following two main processes. First, in order to receive fewer affect of variable signals, a best fitting regression line is used, which is obtained by applying a least squares method to coefficient magnitudes in a node with a uniform wavelet packet transform. Next, in order to update the noise estimation efficiently, a differential forgetting factor and a correlation coefficient per subband are used, where subband is employed for applying the weighted value according to the change of signals. In particular, this method has the ability to update the noise estimation by using the estimated noise at the previous frame only, without utilizing the statistical information of long past frames and explicit nonspeech frames by voice activity detector. In objective assessments, it was observed that the performance of the proposed method was better than that of the compared (minima controlled recursive averaging, weighted average) methods. Furthermore, the method showed a reliable result even at low SNR.