General regression neural network-based data-driven model-free predictive functional control for a class of discrete-time nonlinear systems

Design of Model-Free Adaptive Predictive Controllers with Low Online Computational Load for a Class of Discrete-Time Nonlinear Multiple-Input Multiple-Output Systems

“…As analyzed in literature [27], in the GPC-MFAPC, current system inputs are computed by minimizing a vector including predictive tracking errors at all sampling time points (STPs) within the predictive periods. Thus, hardwares must handle complicated mathematical operations among highdimensional matrices whose dimensions are closely relative to the predictive periods.…”

“…Increasing the predictive periods would result in dramatical growth of the OCL. Although the defect was pointed out in studying a class of single-input single-output (SISO) systems in [27], it also exists in studying the multiple-input multiple-output (MIMO) systems [28,29] since it is a common drawback in the GPC framework-based predictive control approaches. Besides, the defect in the MIMO systems become especially obvious since the predictive periods appear in form of multiplying by system dimensions such that slight increasing of the predictive periods is quite easy to cause OCL surging [28,29].…”

“…In this paper, input structuralization where system inputs are expressed by linear combination of a few basis functions [27] is integrated into the MFAPC design for nonlinear MIMO systems such that only tracking errors at the terminal STPs instead of all STPs within the predictive periods are minimized. Thus, the OCL in the proposed approach is only determined by the number of the basis functions and irrelevent to the predictive periods.…”

“…Thus, the OCL in the proposed approach is only determined by the number of the basis functions and irrelevent to the predictive periods. Besides, as shown in [27], the number of the basis functions is only two or three even the systems are complicated. Thus, the mathematical operations among high-dimensional matrices are avoided successfully such that the OCL in the proposed approach is low compared with the GPC-MFAPC approach.…”

“…Besides, the defect in the MIMO systems become especially obvious since the predictive periods appear in form of multiplying by system dimensions such that slight increasing of the predictive periods is quite easy to cause OCL surging [28,29]. So far, only one literature [27] has studied the OCL reduction problem for MFAPC while the control objects only belong to the SISO types. (2) There is a severe shortage of studies on MIMO systems.…”