Jc. Spall et Ja. Cristion, MODEL-FREE CONTROL OF NONLINEAR STOCHASTIC-SYSTEMS WITH DISCRETE-TIMEMEASUREMENTS, IEEE transactions on automatic control, 43(9), 1998, pp. 1198-1210
Consider the problem of developing a controller for general (nonlinear
and stochastic) systems where the equations governing the system are
unknown. Using discrete-time measurements, this paper presents an appr
oach for estimating a controller without building or assuming a model
for the system (including such general models as differential/differen
ce equations, neural networks, fuzzy logic rules, etc.), Such an appro
ach has potential advantages in accommodating complex systems with pos
sibly time-varying dynamics. Since control requires some mapping, taki
ng system information, and producing control actions, the controller i
s constructed through use of a function approximator (FA) such as a ne
ural network or polynomial (no FA is used for the unmodeled system equ
ations), Creating the controller involves the estimation of the unknow
n parameters within the FA. However, since no functional form is being
assumed for the system equations, the gradient of the loss function f
or use in standard optimization algorithms is not available. Therefore
, this paper considers the use of the simultaneous perturbation stocha
stic approximation algorithm, which requires only system measurements
(not a system model), Related to this, a convergence result for stocha
stic approximation algorithms with time-varying objective functions an
d feedback is established. It is shown that this algorithm can greatly
enhance the efficiency over more standard stochastic approximation al
gorithms based on finite-difference gradient approximations.