類神經控制器於高效能感應馬達控制系統之設計

Design of Neural Network Controllers for High Efficiency Induction Motor Vector Control System

李文雄(Wen-Shyong Lee)

指導教授 ： 周錦惠 ； 王順源

國立臺北科技大學/機電學院/電機工程系所/碩士(2005年)

https://doi.org/10.6841/NTUT.2005.00103

摘要

傳統類神經網路架構應用於控制方面，都是在離線(Off-line)
之下學習與訓練，才能達到參數的最佳化，但往往訓練的資料取之
不易，造成所設計之控制系統響應不如預期之結果。有鑑於此，為
了改善傳統類神經網路線上(On-line)學習與調整之缺點，本研究利
用適應性理論投射演算法(Projection Algorithm)於類神經網路，
設計出類神經比例積分控制器(Neural Network PI Controller,
NNPIC)，來達到具有即時線上學習的類神經網路架構，可適時的反
映出系統的動態性能，來提供系統的適應性與強健性。
本研究根據投射演算法，設計出類神經速度控制器，並搭配虛
擬降階型磁通估測器(Adaptive Pseudo-reduced-Order Flux
Observer)，來植入無轉速量測器之三相感應電動機向量控制系統中
，同時應用高效能法則，提高電動機之運轉效能，可獲得良好的運
轉特性與節約能源之優點。
經實驗結果證明，將類神經控制器加入無轉速量測器之感應馬
達向量控制系統中，在暫態和穩態之速度響應方面都保有良好特性
，且所設計的適應性磁通估測器與轉子電阻估測器，在轉子電阻等
參數變化20%條件下，仍具有良好的速度響應與強健性。

關鍵字

類神經比例積分控制器 ； 適應理論 ； 投射演算法 ； 參考模型適應性系統 ； 無轉速量測器

並列摘要

In general, applying traditional artificial neural
network structure to control systems requires off-line
learning and training to obtain the optimal parameters.
However, it is not easy to get suitable training data
and as a result the performance of response of designed
control system is not as expected in such circumstances.
Therefore,to improve the drawbacks of on-line learning and
adaptation properties under the traditional Artificial
Neural Network architecture, this paper adopts the
projection algorithm in adaptive theory to the Neural
Network and a so-called Neural Network PI Controller(NNPIC)
is designed to construct a new Neural Network with online
learning ability, which can properly reflect dynamical
characteristics of the control system and hence provide the
adaptation ability and robustness.
Based on the projection algorithm, this paper also
proposes the Neural Network speed controller, which is
combined with the Adaptive Pseudo-Reduced-Order Flux
Observer, for the sensorless of Induction Motor Vector
Control System.Accommodating with the high power efficiency
control algorithm,the controller increases the motor
operation efficiency, and gains the advantages of superior
dynamical property and power-saving.
From the experimental results, the speed sensorless
adaptive vector control systems with the proposed
Artificial Neural Network Controller shows excellent
performance in both transient and steady-state responses.
In addition, the Adaptive Flux Observer and rotor
resistance estimator still keep the desired speed responses
and robustness within the +-20% variation range of parameters.

並列關鍵字

artificial neural network PI controller ； adaptive theory ； projection algorithm ； model reference adaptive system

參考文獻

[2] J. A. Santisteban and R. M. Stephan, “Vector Control Methods for Induction Machines: An Overview,” IEEE Transaction on Education, Vol. 44, No. 2, MAY 2001.

[3] S. Ogasawara, H. Akagi and A. Nabae, “The Generalized Theory of Indirect Vector Control for AC Machines,” IEEE Transaction on Industrial Applications, Vol. 24, No. 3, May/June 1988, pp. 470-478.

[4] X. Zou, P. Zhu, K. Yong and J. Chen, “Speed Identification for Vector Control of Induction Motors with Voltage Decoupling Control Principle,” Industry Applications Conference, 2003. 38th IAS Annual Meeting, Oct. 2003, pp. 111-116.

[5] R. D. Lorenz and D. B. Laeson, “Flux and Torque Decoupling Control for Field-Weakened Operation of Field-Oriented Induction Machines,” IEEE Transaction on Industrial Applications, Vol. 26, No. 2, March/April 1990, pp. 290-295.

[6] I. Miki, O. Nakao and S. Nishiyama, “A New Simplified Current Control Method for Field-Oriented Induction Motor Drives,” IEEE Transaction on Industrial Applications, Vol. 27, No. 6, Nov/Dec., 1991.