Elsevier

Systems & Control Letters

Volume 44, Issue 3, 23 October 2001, Pages 233-243
Systems & Control Letters

Minimizing a convex combination of the overshoot and undershoot

https://doi.org/10.1016/S0167-6911(01)00157-8Get rights and content

Abstract

We study the problem of minimizing a parameterized convex combination of the overshoot and undershoot of SISO continuous time system in response to a known input. From a dual formulation, we develop a condition for solution existence and specify the structure of optimal solution. In addition, an interrelation between the overshoot and undershoot in controller synthesis is analytically explained in our framework.

Introduction

As before, it is still a hard task to check the feasibility of given time-domain design specifications and to merge them into controller synthesis, even though we are now armed with more advanced results in the area of time-domain optimal control techniques such as L1 [1] and L [6], [11], [12]. In a tracking system design with a fixed reference, for example, we have nontrivial questions:

  • Is there any controller such that the resulting tracking error has no overshoot (positive error)? If many, which one gives the smallest value? If none, what is the best result we can achieve?

  • Often controllers designed for a small overshoot give a large undershoot and vice versa. What are the quantitative aspects of this phenomenon?

Deodhare and Vidyasagar [2] studied the first problem for a restricted class of discrete time tracking systems. They also presented several control applications where positive and negative errors have different engineering importance. However, their design aim was to minimize ℓ1 norm of positive tracking errors, not the peak values. Hill and Halpern [4], [5] and Yoon and Kang [13] dealt with the minimization of one-sided peak; overshoot or undershoot, for discrete and continuous time systems, respectively. In their framework, however, optimal solution does not exist and thus their results are not sufficiently abundant for the above questions. From existing results of L optimal theory [6], [12], we can compute the minimal absolute magnitude of tracking errors, but we cannot deal with the positive and negative errors in a certain optimal way.

The motivation and objective of our study come from the two open questions above. We choose a new cost function that discriminates positive and negative errors under the L framework developed in [6]. Having an asymmetric cost function, we can obtain useful answers to the above questions against increased mathematical difficulties.

This paper is divided into five sections. Section 2 gives a mathematical preliminary. We define our problem in Section 3.1 and develop its dual formulation in Section 3.2. In Section 3.3, we study on the existence of the optimal solution and specify its structure. In Section 3.4, we examine an interrelation of the overshoot and undershoot. We give a numerical example in Section 4 and the conclusion in Section 5.

Section snippets

Mathematical preliminary

Let Lp, 1⩽p⩽∞, denote the space of Lebesque measurable functions from R+[0,∞) to R, the set of real number, with bounded ∥·∥p norms. For a xLp, we define its positive and negative projections;x+(·)max{x(·),0},x(·)min{x(·),0}.Clearly, it holds that 0⩽∥x±∥⩽∥x1, ∥x+1+∥x1=∥x1 and ∥x1=∥x+1x⩾0.

Let RH denote the set of proper and stable transfer functions. Throughout this paper, with a function f:R+↦R, its Laplace transform is denoted by f̂. Similarly, given a ĝ∈RH, g denotes its

Problem definition

Overall our problem formulation comes from that of L optimal control problem developed in [6]. Remarkable differences include a new cost function and corresponding modification in rational approximation results (Lemma 1 below). First, we shortly summarize L problem formulation.

Consider a generalized system configuration in Fig. 1, where G is a finite dimensional linear time-invariant plant and all signals r̂f, φ̂, û, ŷ are assumed to be scalar valued.

From the YJBK parameterization, every

An illustrative example

Our problem is to minimize fλ(φ) where φ̂=[1+GK]−1r̂f, G=NM−1=(s−2/s+1)/(s−1/s+1) and r̂f=1/s, the unit step, and K is the controller to be designed.

Following standard procedures we have the next SPM problem:μ̃λinfq∈RHfλ3(s−1)s(s+1)(s−1)(s−2)s(s+1)2q.Since r.d.(m̂)=1, Lemma 1 implies that there exist a rational controller whose performance is arbitrary close to the optimal value.

With two real zeros s1=1 and s2=2, we haveψ=(e−te−2t)t∈L12,β=(ω(1)ω(2))t=(00.5)tand the dual SPM problem, with αt

Conclusion

We have dealt with the problem of minimizing a convex combination of the overshoot and undershoot of SISO continuous time system due to a fixed input. It was shown that every optimal solution, if existing, is a sum of delayed step functions whose positive and negative magnitude are generally different. In addition, we analytically explained the interplay of the overshoot and undershoot in controller synthesis. We believe our results are significant generalizations of the L optimal problem.

We

References (13)

  • M.A. Dahleh et al.

    L1-optimal compensators for continuous-time systems

    IEEE Trans. Automat. Control

    (1987)
  • G. Deodhare, M. Vidyasagar, Design of non-overshooting feedback control systems, in: Proceedings of the 29th IEEE...
  • B.A. Francis

    A Course in H Control Theory, Lecture Notes in Control and Information Science

    (1987)
  • R.D. Hill, M.E. Halpern, Minimum overshoot design for SISO discrete-time systems, IEEE Trans. Automat. Control 38...
  • R.D. Hill, M.E. Halpern, Minimization of signed peak and absolute peak values of signals in feedback systems, in:...
  • D.E. Miller, Minimizing the maximum value of the regulated output to a fixed input, in: Proceedings of the 31st IEEE...
There are more references available in the full text version of this article.

Cited by (2)

View full text