1 INTRODUCTION
With the development of modern industry, the plants and
processes become more and more complicated, which
makes the control tasks difficult to complete. The control
methods could be categorized into model based control
(MBC) method and data-driven control (DDC) method,
based on whether the plant model is required as a prior
knowledge for controller design.
In MBC methods, the first principle model of the plant
should be built and identified. The MBC controller will be
designed and analyzed based on the identified model. From
the controller design process of MBC, we could find out
that the control performance severely depends on the
accuracy of the identified model. However, for the modern
industry, the complicated systems make the modelling task
difficult, which results in the questionable control
performance of the MBC controller. Noted that, operating
data of the controlled system could reflect their real-time
dynamics. And thanks to the informatization development,
these data of the plant could be stored for either online or
offline controller design and analysis. Hence, it is intuitive
to study whether these plant data could be utilized for
controller design directly, and avoid the complicated and
annoying modeling process, which is the basic idea of DDC
method. In this kind of method, the controller structure
design and parameter tuning of DDC methods are merely
based on the input and output data of the plant, and the
physical model is not required anymore. Moreover,
This work was supported by National Natural Science
Foundation of China (61503138), Shanghai Sailing Program
(15YF1402700), China Postdoctoral Science Foundation funded
project and Fundamental Research Funds for the Central
Universities (22A201514051).
considering the unmodeled dynamics or internal
disturbance, DDC methods are also more suitable. A recent
survey, which discussed the relationship and differences
between MBC and DDC methods could be found in [1].
With the idea of utilizing data of the plant, MFAC (model
free adaptive control) was proposed as a DDC method for a
class of discrete-time nonlinear system. Instead of
identifying the global plant model, MFAC builds the local
data model to mimic the plant behavior. The data model is
built by equivalent dynamic linearization technique along
the operation points, which simplifies the complex global
modeling and the controller design process. With the local
data model, the controller will be designed to minimizing
the output regulating error. The equivalent dynamic
linearization data models can be summarized as three types,
i.e., compact form, partial form and full form dynamic
linearization, which could be chosen according to the
complexity degree of the plant. Right now, due to its
characteristics of independence to physical model, MFAC
has been widely studied and applied in control community.
The proof of the convergence property of the
corresponding closed-loop control system using partial
form dynamic linearization was given in [2]. The MFAC
method for MIMO system was developed in [3]. A
preliminary study about the robust issue of MFAC was
shown in [4]. The theoretical framework and typical
application of MFAC were introduced systematically in [5],
where the MFAC for complex interconnected systems and
the modularized and symmetric similarity design
conception were investigated. Later, a novel MFAC that is
based on the dynamic linearization of ideal controller has
been emerged to open up another promising research
direction [6, 7], which is expected to build up a unified
framework of DDC method. With the development of the
theory of MFAC, it has been successfully implemented in
Stability Analysis of Full Form Dynamic Linearization Controller Based
Data-driven Model Free Adaptive Control
Yuanming Zhu
1
, Member, IEEE, Shangtai Jin
2
1. Key Laboratory of Advanced Control and Optimization for Chemical Processes, Ministry of Education,
East China University of Science and Technology, Shanghai 200237, China
E-mail: yuanmingzhu@ecust.edu.cn
2. Advanced Control System Laboratory, Beijing Jiaotong University, Beijing 100044, China
E-mail: shtjin@bjtu.edu.cn
Abstract: Full form dynamic linearization controller based data-driven model free adaptive control algorithm has been
studied in control community. This method merely requires the I/O data of the plant to design the controller and is easy
for implementation. However, there still lacks necessary research on the conditions which can guarantee the stability of
the closed loop system. This paper presents a stability analysis result for a class of nonlinear system. With some mild
assumptions, the convergence of the output regulating error was derived by rigorous mathematical method, which
guarantees the correctness of the proposed method in theory.
Key Words: Dynamic Linearization Controller, Data-driven, Model Free Adaptive Control, Nonlinear System
5755
978-1-4673-9714-8/16/$31.00
c
2016 IEEE