モデル規範形適応制御系のパラメータ調整法

抄録

To adjust the parameter in the model reference adaptive control system, the gradient method is often employed making use of the actual error between the outputs of the basic control system and the reference model. But, in this method, it is difficult to determine the coefficient of the gradient so as to ensure stability and quick response.
In this paper, a parameter adjusting method is proposed that minimizes the index of performance described as the integral of the squared value of the equation error with respect to the basic control system and the reference model. For the configuration of the basic control system considered here, this proposed method is approximately equivalent to the Gauss-Newton's method based on the actual output error in the neighbourhood of the optimum point. In this method, the construction of the adaptive control system is comparatively simple and the parameter in the controller can be brought to the neighbourhood of its optimum value by only one trial regardless of the initial value of the parameter. But, when the dynamic characteristics of the basic control system can be made only approximately equal to that of the reference model, the parameter fluctuates around its optimum value in the steady state, depending on the wave form of the input signal. The fluctuation can be made small by selecting a long integration time in the index of performance but the time required for one trial of adjustment increases.
This method is also investigated by the computer simulation and its effectiveness is ascertained experimentally.