Intelligent Proportional Differential Neural Network Control for Unknown Nonlinear System

Abstract

This paper presents an intelligent proportion-differential neural network (iPDNN) controller for unknown nonlinear systems. This controller is based on the intelligent proportion integration differentiation (iPID) controller. In an iPID controller system, a unknown nonlinear SISO system is regarded as an ultra-local two-order or one-order model and a lumped unknown dynamics (LUD) disturbance which contains the high-term and parametric uncertainties by the differential algebra and estimation method online. However, its performance of an iPID control depends on the precision and rapidity for estimating the LUD disturbance. Besides, it also influences the parameter in the ultra-local model. In order to compensate the estimation error of LUD disturbance, we put forward an extra radial basis function (RBF) neural network observer to estimate it. This extra observer cannot only ensure to acquire the estimation error rapidly, but also has an ability of self-learning. In addition, this iPDNN method can ensure the closed-loop system stability under the Lyapunov stability theory. Finally, in order to demonstrate its performance, an inverted pendulum plant has been applied and the results indicate this method is of efficiency.

Robust Model-Free Adaptive Interval Type-2 Fuzzy Sliding Mode Control for PEMFC System Using Disturbance Observer

“…MFC-based iPID is well-known an attractive approach which has gained more attention in recent years [29][30][31][32]. To achieve excellent performance, MFC-based iPID was integrated with other techniques [33][34][35]; for instance, to guarantee the robustness of the control synthesis, time-delay estimation and neural network control were applied in Ref [36], while to ensure the stability of the unknown nonlinear system under uncertainties and external disturbance, adaptive fuzzy logic control, adaptive SMC and fractional-order SMC were proposed [35,37,38]. However, the robustness of these strategies relies on the approach used to estimate the unmodeled system dynamics, such as algebraic estimator [39,40], extended state observer [29,31,35] and timedelay estimation [14,30,36].…”

Machine learning and control engineering: The model-free case

“…However, it should be impossible for a control engineer not to be impressed by the recent successes of the RL community such as solving Go [66]." Many concrete case-studies have already been investigated: see, e.g., [3,9,13,14,16,17,21,22,31,34,39,42,43,44,46,51,52,53,54,57,58,59,68,74,75,77,78,81,83,85,86,87]. Although those works are most promising, they show that ANNs and RL have perhaps not provided in this field such stunning successes as they did elsewhere.…”

Adaptive Sliding Mode Control for Semi-Active Structural Vibration Control

“…The other class called the intelligent class is operating algorithm without requiringthe mathematical model presentation of the system (Zizouni et al, 2019). In many cases a combination of the pervious classes is necessary, this is called hybrid algorithm (Wang et al, 2016).…”