Solving predictive control problem of fast‐varying multivariable systems by incorporating unknown active dynamics generated by real‐time adaptive learning machine

다국어 초록 (Multilingual Abstract)

Not only adaptive predictive control of switched systems is a computationally intensive procedure, it also involves various challenges in addressing the problems of robust stabilization and precise tracking. This study proposes new strategies to deal with the aforementioned issues (namely safe and precise control alongside with reduction of computational burden). The first contribution of this work is reduction of conservatism for described class of systems. Control of switched systems with undetectable switching signals is often conducted in worst case switching configuration to ensure robustness, which potentially results in conservative design. The issue of conservativeness is intensified in multi input‐multi output (MIMO) dynamical systems due to increased dimensions. However, attaining a robust control scheme for all switching configurations while ensuring precise response is inherently paradoxical. To overcome this issue, this study proposes a new dual‐mode algorithm where control modes corresponding to safety and precision are activated at appropriate stages of system response. This is conducted based on incorporation of an adaptive fuzzy‐wavelet neural network identification scheme in predictive control of MIMO switched systems. However, as convergence of the adaptive algorithm to actual system is attained after a finite period of time, a safe‐mode control algorithm is proposed to maintain quality of transient response in convergence period. In other words, the proposed algorithm operates in safe and precise control modes to ensure robust stability in the convergence period and non‐conservative design in steady‐state. Second major contribution of the work is reduction of calculation burden based on incorporation of a suboptimal control algorithm. To this end, we propose a predictive control scheme based on a suboptimal gradient‐descent based controller, calculating feasible stabilizing inputs instead of optimal inputs. Effects of dynamical variations are incorporated in the model predictive control framework for increased compatibility with high‐speed switching dynamics. Then, based on incorporation of dual‐mode algorithm, precise steady‐state performance is attained while preventing notable perturbations in dynamical discontinuities at switching.