Distributed control of second-order nonlinear
time-delayed multiagent systems with
disturbance using neural networks
Hongwen Ma, Derong Liu, and Ding Wang
Abstract—In this paper, a class of second-order nonlinear
time-delayed multiagent systems with disturbance is investigated.
In order to improve the adaptivity, neural networks are used
to learn the unknown dynamics. Then, by utilizing Lyapunov-
Krasovskii functional, time delays can be eliminated. Moreover,
a robustifying term is introduced to constrain external distur-
bance. With divide-and-conquer idea, the distributed controller
is divided into five different parts to make the multiagent systems
reach consensus. To circumvent the singularity induced by the
time-delay elimination part, a σ-function is developed. Finally,
the simulation results demonstrate the validity of the distributed
controller.
I. INTRODUCTION
Distributed control is an important technique in multiagent
systems. It can be traced back to Boid model [1] and Vicsek
model [2], which are derived from natural phenomena. Variety
of problems investigated include optimal control problems
[3]–[5], output-based control problems [6]–[8], event-triggered
control problems [9], [10] and time-delayed control problems
[11]–[13]. For more details, please refer to the survey papers
[14]–[18] and the references therein. In [19], a decentralized
adaptive control with neural networks (NNs) was established
for multiagent systems with unknown dynamics. In [12], a
class of first-order nonlinear time-delayed multiagent systems
with external noises is studied. In [11], a Lyapunov-Krasovskii
functional and Young’s inequality were used for the consen-
sus of time-delayed multiagent systems. Thus, it is of great
significance to investigate how to apply the distributed control
technique to second-order nonlinear time-delayed multiagent
systems.
The technique of NNs is a powerful tool for learning
the unknown dynamics [20]–[25]. In [26], adaptive neural
control was introduced to solve the uncertain MIMO nonlinear
systems. In [27], an adaptive neural control protocol was
This research was funded by the IEEE Computational Intelligence Society
Graduate Student Research Grant 2015. This work was also supported in part
by the National Natural Science Foundation of China under Grants 61233001,
61273140, 61304086, 61533017, and U1501251, in part by Beijing Natural
Science Foundation under Grant 4132078, in part by China Scholarship
Council under the State Scholarship Fund, and in part by the Early Career
Development Award of SKLMCCS.
Hongwen Ma and Ding Wang are with The State Key Laboratory of
Management and Control for Complex Systems, Institute of Automation,
Chinese Academy of Sciences, Beijing 100190, China (e-mail: mahong-
wen2012@ia.ac.cn, ding.wang@ia.ac.cn). Derong Liu is with the School of
Automation and Electrical Engineering, University of Science and Technology
Beijing, Beijing 100083, China (e-mail: derong@ustb.edu.cn).
utilized for a class of strict-feedback nonlinear systems with
unknown time delays. We utilize the technique of Lyapunov-
Krasovskii functional from [27] and [11] to eliminate the
negative effects of time delays. However, this technique will
induce singularities in the distributed controller and a σ-
function is established to deal with it.
To the best of our knowledge, it is the first time to
investigate second-order time-delayed nonlinear multiagent
systems with the developed σ-function. A reference signal
which can reduce the difficulty of achieving consensus is
also applied. Furthermore, by using the property of hyperbolic
tangent function, a robustifying term is utilized to constrain the
disturbance.
The rest of this paper is organized as follows. Preliminaries
for graph theory and radial basis function neural networks
(RBFNNs) are given in Section II. Main results are given in
Section III. Simulation example is conducted to demonstrate
the effectiveness of the developed method in Section IV.
Conclusion is given in Section V.
Notations: (·)
T
denotes the transpose of a given matrix.
tr (·) is the trace of a given square matrix. ·is the Frobenius
norm or Euclidian norm. ⊗ stands for the Kronecker product.
λ
min
(·) and λ
max
(·) are the smallest nonzero eigenvalue
and the largest eigenvalue of a given real symmetric matrix,
respectively. diag(·) represents a diagonal matrix.
II. P
RELIMINARIES
A. Graph Theory
A triplet G = {V, E, A} is called a graph if V =
{1, 2,...,N} is the set of nodes, E⊆V×Vis the set of
edges, and A =(A
ij
) ∈ R
N×N
is the adjacency matrix of G.
Denote A
ij
as the element of the ith row and jth column of
the matrix A. The ith node represents the ith agent, and an
ordered pair (i, j) ∈Emeans that agent i can directly transfer
its information to agent j. No self-loop will be considered.
Laplacian matrix L of graph G is given as follows:
L
ij
=
k∈N
i
A
ik
, if i = j;
−A
ij
, if i = j.
(1)
B. Radial Basis Function Neural Networks
In practice, we usually employ a neural network as the func-
tion approximator to model an unknown function. RBFNN is a
potential candidate for approximating the unknown dynamics
5263
978-1-5090-0620-5/16/$31.00
c
2016 IEEE