Computer algorithms for solving optimization problems for discrete-time fractional systems Andrzej Dzieli´ nski and Przemyslaw M. Czyronis Abstract— Dynamic programming and discrete-time calculus of variations optimization problems for fractional discrete-time systems with quadratic performance index have been formu- lated and solved. A new methods for numerical computation of optimization problems have been presented. The efficiency of the methods have been demonstrated on numerical example and illustrated by graphs. Graphs also show the differences between the fractional and classical (standard) systems theory. Results for both methods have been obtained through a computer algorithms written for this purpose. I. INTRODUCTION Dynamic optimization problems for integer (not fractional) order systems have been widely considered in literature (see e.g. [1]–[3]). Mathematical fundamentals of the fractional calculus are given in the monographes [4]–[6] and fractional calculus of variations is given in [7]. Fractional differential equations and their applications have been addressed in [8], [9]. The numerical simulation of the fractional order control systems has been investigated in [10]. One of the fractional discretization method has been presented in [11]. Some optimal problems for fractional order systems have been investigated in [12]–[15]. Dynamic Programming problem for discrete-time fractional systems has been formulated and solved in [16]. Variational calculus for discrete-time frac- tional systems have been investigated in [17]–[22]. Fractional Kalman filter and its application have been addressed in [23], [24]. Some recent interesting results in fractional systems theory and its applications for standard and positive systems can be found in [25], [26]. In this paper optimization problems for fractional discrete- time systems with quadratic performance index will be formulated. A general solutions for dynamic programming and discrete-time calculus of variations problems will be presented. A new methods for numerical computation of optimization problems will be presented. The efficiency of the methods will be demonstrated on numerical example and illustrated by graphs. Graphs also show the differences between the fractional and classical (standard) systems the- ory. Results for both methods will be obtained through a computer algorithm written for this purpose. Computer al- gorithms will be discussed in details and the block diagrams will be presented. The paper is organized as follows. In section II some preliminaries are recalled and the optimization problem for This work was not supported by any organization The authors are with the Faculty of Electrical Engineering - The Institute of Control and Industrial Electronics, Warsaw University of Technology, Pl. Politechniki 1, 00-661 Warszawa, Poland; email: an- drzej.dzielinski@ee.pw.edu.pl, przemyslaw.czyronis@ee.pw.edu.pl dynamic programming method will be formulated. The gen- eral solutions of the above problem are presented also in that section. In section III the optimization problem for discrete- time calculus of variations method will be formulated. The general solutions of the above problem are presented also in that section. In section IV a computer algorithms for com- putation of the optimal control are proposed and numerical examples are presented. Conclusions of the paper are given in section V. II. DYNAMIC PROGRAMMING A. Problem Formulation Consider a fractional discrete-time system, obtained by use of Grunwald-Letnikov’a (shifted) approximation, described by equations x k+1 = k j=0 d j x k-j + Bu k , k Z + , (1a) where x R n , u R m are respectively the state and control vectors, A R n×n , B R n×m and d 0 = A α = A + αI n , 0 <α< 1 , (1b) d j = (-1) j α j +1 I n , j =1,...,k . (1c) Consider a performance index of the form J i (u)= G(x N )+ N-1 k=i F k (x k ,u k ) = x T N Sx N + N-1 k=i ( x T k Qx k + u T k Ru k ) , (2) where R R m×m , Q R n×n , S R n×n and S 0, Q 0 and R> 0. Optimal trajectory starting at the point x 0 and ending at the point x k has been divided into N elementary time intervals [0,N ]. It is desired to find optimal control sequence u 0 ,u 1 ,...,u N-1 , u U, which minimizes the performance index (2) and satisfies the differential equation (1). The solution of this task by searching for a conditional minimum of the performance index (2) requires the solution of N equations with N unknown variables of the form ∂J (u) ∂u k =0 , (k =0,...,N - 1) , where J (u) is the performance index (2) after substituting (1) for k =1, 2,...,N - 1. 2013 European Control Conference (ECC) July 17-19, 2013, Zürich, Switzerland. 978-3-952-41734-8/©2013 EUCA 4009