Neural Network Delayed Control of An Idealized
Offshore Steel Jacket Platform
Zhihui Cai
School of Science
China Jiliang University
Hangzhou 310018, China
Email: zhcai81@163.com
Baolin Zhang
School of Science
China Jiliang University
Hangzhou 310018, China
Email: zhangbl2006@163.com
Xianhu Yu
Ningbo Radio and TV University
Ningbo 315016, China
Abstract—This paper deals with the problem of neural network
delay control for an idealized offshore platform system. Two
neural networks (feed-forward neural network and nonlinear
autoregressive neural network) delay controllers are proposed to
the vibration control of the offshore platform under external wave
force. It is shown through simulation results that the designed
neural network delay controllers can effectively improve the
stability of the offshore platform. Furthermore, the designed
neural network delay controllers show more robustness than the
delayed H
∞
control in terms of system parametric perturbations
and external wave loads.
Keywords—offshore platforms; feed-forward neural network;
nonlinear autoregressive neural network; delayed feedback.
I. INTRODUCTION
In the past decades various types of control schemes have
been proposed to attenuate the vibrations in offshore steel jack-
et platforms that generally suffered from external disturbances
such as earthquake, wave, wind, ice and the self-induced non-
linear hydrodynamic forces [1], [2]. Passive control schemes,
semi-active control schemes and active control schemes are
the mainly three types of control schemes [3]. For instance,
to reduce the vibration excited by the earthquake, an optimal
neural network controller was developed in [4], [5]. In [6],
a delayed H
∞
control (DHC), which was presented as an
active controller for an offshore platforms under external wave
force, shew its positive effects of the extra time-delay on the
controller.
In recent years, neural networks have shown advantage
on data-driven model-free control in various systems [7],
[8]. However, there are a few literatures applying neural
networks on the study of stability of offshore platforms with
external wave force. In this work, two structures of neural
networks, referred to as the feed-forward neural network and
the nonlinear autoregressive neural network, are introduced in
the vibration control of an idealized offshore platform model.
The main contributions are as follows:
1) Neuron network delay controllers, namely FFNC and
NARXNC, are presented for an idealized offshore plat-
form model with active mass damper (AMD) subject to
external wave force. These controllers have almost the
same control performance as the delayed H
∞
control
scheme in [6].
2) Robustness of the proposed controllers are illustrated in
terms of system parametric perturbations and external
wave force with small random disturbances.
This paper is organized as follows. In Section II, the
equations of the external wave force are given firstly. Then,
an idealized model of offshore steel jacket platform with
an AMD device are presented, and the structures of two
neural networks are also shown. The design of the neural
network delay controllers are proposed in Section III. Several
simulation results are shown in Section IV. Some discussions
are illustrated in Section V.
II. PROBLEM FORMULATION
In this section, the external wave force model will be
expressed first. Then, an idealized model of offshore platform
with an active mass damper device will be introduced, and
the structure of the artificial neural network used in this paper
well be described.
A. Wave force model
In this paper, the wave force is assumed to be same to the
one used in [6]. As stated in [6], the wave fore f(t) can be
described in the form as
f(t) =
d
0
p(z, t)ϕ(z)dz (1)
where d is the water depth, z is the vertical coordinate with
the origin, ϕ(z) is the shape function related to the offshore
platform and p(z, t) is the physical horizontal wave force per
unit length and can be approximated as follows
p(z, t) =
1
2
ρC
d
˜
D
8
π
σ
v
(z)v(z , t) +
1
4
ρπC
m
˜
D
2
˙v(z, t) (2)
where ρ,
˜
D, C
d
and C
m
are the water density, diameter of
the cylinder, drag, and inertia coefficients, respectively. v(z, t),
˙v(z, t) and σ
v
(z) are the water particle velocity, acceleration,
and the standard deviation of the velocity at direction z,
respectively. In addition,
v(z, t) = T
vη
(ω, z)η(t) (3)
˙
v
(
z, t
) =
T
˙vη
(
ω, z
)
η
(
t
)
(4)
σ
v
(z) =
ω
0
|T
˙vη
(ω, z)|
2
S
η
(ω)dω
1/2
(5)
Eighth International Conference on Intelligent Control and Information Processing
November 3-5, 2017; Hangzhou, China
978-1-5386-1168-5/17/$31.00 ©2017 IEEE