See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/330756548

Robust adaptive regulation of dynamically positioned ships with unknown

dynamics and unknown disturbances

Article in International Journal of Adaptive Control and Signal Processing · January 2019
DOI: 10.1002/acs.2972

CITATIONS READS

7 208

3 authors, including:

Hu Xin Miroslav Krstic

Ludong University University of California, San Diego
81 PUBLICATIONS 1,793 CITATIONS 739 PUBLICATIONS 30,884 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Hu Xin on 19 February 2019.

The user has requested enhancement of the downloaded file.

Received: 19 January 2018 Accepted: 5 January 2019
DOI: 10.1002/acs.2972

RESEARCH ARTICLE

Robust adaptive regulation of dynamically positioned ships

with unknown dynamics and unknown disturbances

Xin Hu1 Jialu Du1 Miroslav Krstić2

1
School of Marine Electrical Engineering,
Dalian Maritime University, Dalian, China Summary
2
Department of Mechanical and We develop a robust adaptive regulating control law for dynamically positioned
Aerospace Engineering, University of
ships subject to unknown dynamics and bounded unknown disturbances incor-
California, San Diego, California
porating the radial basis function (RBF) neural network (NN), the dead zone
Correspondence adaptive technique, and a robust control term into the vectorial backstepping
Jialu Du, School of Marine Electrical
approach. The RBF NNs with the dead zone adaptive laws approximate the ship
Engineering, Dalian Maritime University,
Dalian 116026, China. unknown dynamics. The adaptive law-based robust control term compensates
Email: [email protected] for unknown disturbances, NN approximation errors, and undesirable errors
Funding information arising from the design procedures. The developed dynamic positioning (DP)
National Natural Science Foundation of control law regulates the ship position and heading to the desired values with
China, Grant/Award Number: 51079013;
arbitrarily small errors, while guaranteeing the uniform ultimate boundedness
Program for Liaoning Excellent Talents in
University, Grant/Award Number: of all signals in the DP closed-loop control system of ships. High-fidelity simu-
LR2015007; Technology Foundation for lations on two supply ships and comparisons demonstrate the effectiveness and
Selected Overseas Chinese Scholar;
the superiority of the developed DP control law.
Fundamental Research Funds for the
Central Universities, Grant/Award
Number: 3132017302 K E Y WO R D S
robust adaptive control, ship dynamic positioning, unknown dynamics, unknown time-varying
disturbances, vectorial backstepping

1 I N T RO DU CT ION

With the ocean exploitation and exploration expanding toward the distant and deep sea, the ships installed with dynamic
positioning (DP) systems have been increasingly used in the offshore industries. The DP systems of ships are intended
to regulate the ships' horizontal motions by exclusive means of their own propellers and thrusters.1 The DP systems
make the ships work in the deep sea, have high positioning accuracy, and are flexible in operation. As such, they are the
indispensable technology for many offshore operations, such as drilling, diving support, and offloading.2
The first DP system came to existence in the 1960s for offshore drilling applications adopting the proportional-
integral-derivative control algorithm cascaded with the low-pass and/or the notch filter.3 More advanced DP control was
developed in the 1970s based on multivariable optimal control and Kalman filter theory,4-6 which requires the lineariza-
tion of the ship motion mathematical model. To overcome this problem, the DP nonlinear control methods emerged
from the 1990s, such as backstepping control7,8 and Takagi-Sugeno (T-S) fuzzy control,9 however ignoring the ocean dis-
turbances. Considering unknown constant disturbances, Godhavn et al10 proposed a nonlinear tracking control law for
the ships using the adaptive backstepping method, where adaptive laws were employed to estimate unknown constant
disturbances. Bertin et al11 developed a DP nonlinear control law for the underactuated ships using the feedback lineariza-
tion technique; Fossen and Strand12 proposed a DP weather optimal positioning control using the adaptive backstepping
method, where the ship heading automatically adjusts such that the minimum energy was used; Veksler et al13 presented
a DP control strategy using model predictive control method. Considering wave- and low-frequency (LF) disturbances,

Int J Adapt Control Signal Process. 2019;1–12. wileyonlinelibrary.com/journal/acs © 2019 John Wiley & Sons, Ltd. 1
2 HU ET AL.

Fossen and Strand14 presented a passive nonlinear observer as a wave filter to estimate the LF position, velocities, and
disturbances from noisy position measurements. Combining this passive observer with a proportional-derivative–type
controller, Loria et al15 proposed a DP output-feedback controller. However, this passive observer requires a priori knowl-
edge of ocean disturbances. Considering unknown disturbances, Hassani et al16 presented a multiple model adaptive
wave filtering estimator for the DP system design of vessels; Nguyen et al17 proposed a DP hybrid output-feedback control
strategy based on the supervisory control theory, where different controllers and observers can be online switched accord-
ing to varying sea states; Hu et al18 developed a robust adaptive control law using an observer and vectorial backstepping
method; Du et al19 proposed a DP robust control law for ships under input saturation incorporating an auxiliary dynamic
system and a disturbance observer into the dynamic surface control method. The aforementioned DP nonlinear control
schemes require a priori knowledge of the ship motion mathematical model parameters.
Considering unknown ship model parameters and unknown constant disturbances, Do20 designed a DP robust adaptive
output-feedback control law using the adaptive vectorial backstepping method and an adaptive velocity observer. Under
parameter perturbations and unknown time-varying disturbances in the ship motion mathematical model, Kjerstad et al21
proposed a DP disturbance rejection control scheme, where an observer provides the estimates of unknown time-varying
disturbances and model parameter perturbations according to acceleration measurements; Tannuri et al22 developed a
DP sliding mode control, whose performance was evaluated numerically and experimentally on a 1:150 scaled shuttle
tanker; Hassani et al23 developed a DP robust control scheme using H∞ and mixed-𝜇 technique, whose effectiveness was
verified by experiments with a 1:30 scaled supply ship.
In practice, there obviously exist uncertainties in both the ship dynamics and the environmental disturbances. In the
presence of unknown dynamics and unknown time-varying disturbances, Chang et al24 proposed a DP fuzzy control law
for ships with multiplicative noises under actuator saturation using the T-S fuzzy model, the linear matrix inequality
method, and the concept of parallel distributed compensation. Du et al25 developed a DP robust adaptive control law,
where the projection algorithm-based adaptive laws together with the constructed observer provide online estimates
of unknown time-varying disturbances and the other projection algorithm-based adaptive laws update the unknown
ship mathematical model parameters. Du et al26 proposed a DP robust adaptive neural controller, where the radial basis
function (RBF) neural networks (NNs) were used to approximate ship unknown dynamics together with unknown
time-varying disturbances. In the work of Du et al,26 the uniform ultimate boundedness of the ship positioning errors was
guaranteed and the NN approximation errors were not addressed.
In this paper, the DP control problem of ships simultaneously with unknown dynamics and bounded unknown
time-varying disturbances is solved. A robust adaptive control law is developed to regulate the ship to the desired position
and heading with arbitrarily small errors, while the uniform ultimate boundedness of all signals in the DP closed-loop
control system of ships is guaranteed. The main contributions of this paper are as follows.
1. Different from the work of Hu et al,18 the developed DP robust adaptive regulating control law does not require any
priori knowledge of ship motion mathematical model parameters due to employing adaptive RBF NNs.
2. Compared with the work of Du et al,26 the dead zone is introduced in the adaptive laws of RBF NNs, so that the
parameter drift and burst are prevented; and the adaptive law-based robust control term is inserted into the DP
control design to compensate the bounded unknown disturbances, NN approximation errors, and undesirable errors
arising from the design procedures, so that the robustness of the DP control law is improved.
The remainder of this paper is organized as follows. Section 2 presents the problem formulation and preliminaries.
Section 3 details the DP robust adaptive control design. Section 4 provides simulations involving two supply ships and
simulation comparisons with an existing DP control law. Section 5 concludes this paper.

2 PROBLEM FO RMULATION AND PRELIMINARIES

2.1 Problem formulation

Let the position (x, y) and heading 𝜓 of the ship in the north-east frame be expressed in the vector 𝜼 = [x, y, 𝜓]T and the
velocities in the ship-fixed frame be expressed in the vector 𝝊 = [u, v, r]T . These three modes are referred to surge, sway,
and yaw. The motion mathematical model of a dynamically positioned ship is given by14 :

𝜼̇ = J(𝜓)𝝊 (1)
M 𝝊̇ = −D𝝊 + 𝝉 + d(t), (2)
HU ET AL. 3

where J(𝜓) is the rotation matrix as follows

⎡ cos (𝜓) − sin (𝜓) 0⎤
J(𝜓) = ⎢ sin (𝜓) cos (𝜓) 0⎥, (3)
⎢ ⎥
⎣ 0 0 1⎦
with the property J T (𝜓) = J−1 (𝜓). M ∈ R3 × 3 being the inertia matrix including the added mass components, which is
positive definite and symmetric. D ∈ R3 × 3 is the linear damping matrix. 𝝉 = [𝜏 1 , 𝜏 2 , 𝜏 3 ]T is the control vector produced by
the propulsion system, consisting of surge force 𝜏 1 , sway force 𝜏 2 , and yaw moment 𝜏 3 . d(t) = [d1 (t), d2 (t), d3 (t)]T denotes
the LF disturbance forces and moment due to second-order waves, wind, currents in the ship-fixed frame.
Assumption 1. (1) The model parameter matrices M and D are unknown. (2) The LF disturbances di (t), i = 1, 2, 3
are bounded unknown, which means there exists
‖d‖ ≤ d∗ < ∞ (4)
with an unknown constant d* > 0 and || · || representing the two-norms a vector or a matrix.

Remark 1. The ship dynamic model parameters depend on the ship's mass, hydrodynamic derivatives, inertia
moment, etc., which are difficult to be accurately identified. On the other hand, the ocean environment is often
changing and unpredicted as well as has finite energy. As such, Assumption 1 is reasonable.
The control objective in this paper is to develop a DP robust adaptive control law for the ship (1)-(2) under Assumption 1,
so that the ship position (x, y) and heading 𝜓 are regulated to the desired value 𝜼d = [xd , yd , 𝜓 d ]T with arbitrarily small
errors, while all signals in the DP closed-loop control system are uniformly ultimately bounded.

2.2 RBF neural network

In the RBF NN, the hidden layer completes a fixed nonlinear transformation and the output layer linearly combines the
outputs of the hidden layer. Therefore, the output of the RBF NN is as follows27,28 :
𝑓𝑛𝑛 (𝝌) = wT 𝝓(𝝌), (5)
where 𝝌 = [𝜒 1 , 𝜒 2 , … , 𝜒 m ]T ∈ 𝜴𝜒 ⊂ Rm is the input vector and w = [w1 , w2 , … , wl ]T is the NN weight vector with l > 1
being the node number. 𝝓(𝝌) = [𝜙1 (𝝌), 𝜙2 (𝝌), … , 𝜙l (𝝌)]T is the basis function vector. In this paper, 𝜙j (𝝌) ( j = 1, 2, … , l)
is taken as the Gaussian function [ ]
‖𝝌 − 𝜿 𝑗 ‖2
𝜙𝑗 (𝝌) = exp − , (6)
h2𝑗
where 𝜿 𝑗 = [𝜅𝑗1 ,𝜅𝑗2 , … , 𝜅𝑗m ]T and hj > 0 denote the center and the width of the Gaussian function, respectively.
Universal approximation results show that if the node number l is sufficiently large, the NN wT 𝝓(𝝌) can approximate
any continuous function f (𝝌) defined on a compact set 𝜴𝜒 ⊂ Rm . This is described by
𝑓 (𝝌) = w∗T 𝝓(𝝌) + e(𝝌), ∀𝝌 ∈ 𝜴𝜒 ⊂ Rm (7)
where e(𝝌) is the NN approximation error and w* ∈ Rl is the ideal NN weight vector defined by
{ }
| |
w∗ ≔ arg min sup |𝑓 (𝝌) − wT 𝝓(𝝌)| . (8)
w∈Rl 𝝌∈𝛺𝜒 | |

Over the compact set 𝜴𝜒 , the ideal NN weight vector w* and the NN approximation error e(𝝌) are bounded, ie,
‖w∗ ‖ ≤ wm , |e(𝝌)| ≤ em ∀𝝌 ∈ 𝜴𝜒 , (9)
where wm and em are positive constants.

3 DP RO B UST ADAPT IVE CO NTRO L D ESIGN

A robust adaptive control law is derived incorporating the RBF NNs, the dead zone adaptive technique, and a robust
control term into the vectorial backstepping method29 for the DP of ships. The RBF NNs with the dead zone adaptive
4 HU ET AL.

laws30 approximate the ship unknown dynamics. The adaptive law-based robust control term compensates for bounded
unknown disturbances, NN approximation errors, and undesirable errors arising from the design procedures.
The DP control design consists of two steps based on the following change of coordinates:
z1 = 𝜼 − 𝜼d (10)

z2 = 𝝊 − 𝜶 1 , (11)
where 𝜶 1 ∈ R3 is the intermediate control function vector to be designed later.
Step 1: Differentiating (10) along (1) obtains
ż 1 = J(𝜓)𝝊, (12)
where 𝝊 is viewed as the virtual control vector.
Choose the Lyapunov function candidate for (12)
1 T
V1 = z z1 . (13)
2 1
According to (12) and (11), the time derivative of (13) is
V̇ 1 = zT1 J(𝜓) (z2 + 𝜶 1 ) . (14)

Design the intermediate control function vector 𝜶 1 for 𝝊 as follows:

𝜶 1 = −J T (𝜓)K 1 z1 , (15)
where K 1 = K T1 is the 3 × 3 positive-definite design matrix.
Substituting (15) into (14), according to the property JT (𝜓) = J−1 (𝜓), we obtain
V̇ 1 = −zT1 K 1 z1 + zT1 J(𝜓)z2 . (16)

Step 2: The time derivative of (11) along (2) is

ż 2 = M −1 (𝝉 + d − D𝝊 − M 𝜶̇ 1 ) . (17)

Choose the Lyapunov function candidate for (12) and (17)

1
V2 = V1 + zT2 Mz2 . (18)
2
The time derivative of (18) along the solutions of (16) and (17) is
V̇ 2 = −zT1 K 1 z1 + zT1 J(𝜓)z2 + zT2 (𝝉 + d − D𝝊 − M 𝜶̇ 1 ) . (19)

Since the dynamic model parameters M and D are unknown, the term D𝝊 + M 𝜶̇ 1 in (19) is unknown and is expressed
with (7) as follows:
D𝝊 + M 𝜶̇ 1 = 𝜣 ∗T 𝜷(𝝇) + 𝜺(𝝇), (20)
T
⎡ 𝜽∗T 01×l 01×l ⎤
⎢ 1 ⎥
where 𝝇 = [𝝊T , 𝜶̇ T1 ]T is the input vector, 𝜣 ∗ = ⎢ 01×l 𝜽∗T
2 01×l ⎥ ∈ R3l×3 with 𝜽∗i = [𝜽∗i1 , 𝜽∗i2 , … , 𝜽∗il ]T (i = 1, 2, 3) is
⎢0 01×l 𝜽∗T ⎥
⎣ 1×l 3 ⎦
the ideal neural-network weight matrix, l > 0 is the node number, 𝜷(𝝇) = [𝜷 T1 (𝝇), 𝜷 T2 (𝝇), 𝜷 T3 (𝝇)]T ∈ R3l with 𝜷 i (𝝇) =
[𝜷 i1 (𝝇), 𝜷 i2 (𝝇), … , 𝜷 il (𝝇)]T (i = 1, 2, 3) is the basis function vector, and 𝜺(𝝇) ∈ R3 is the approximation error vector. From
(9), it follows that ‖ ‖ ∗ ∗
‖𝜺(𝝇)‖ ≤ 𝜀 with 𝜀 > 0 being an unknown constant.
For the convenience of the DP control design later, notate the unknown parameter
∗
𝛿 = d∗ + 𝜀 + c∗ , (21)
where c* is any positive constant representing the upper bound of the undesirable errors arising from the design
procedures.
Substituting (20) into (19) obtains
V̇ 2 = −zT1 K 1 z1 + zT1 J(𝜓)z2 + zT2 [𝝉 + d − 𝜣 ∗T 𝜷(𝝇) − 𝜺(𝝇)]. (22)
HU ET AL. 5

Based on the RBF NNs and the robust control term, we design the DP robust adaptive control law
̂ 𝜷(𝝇) + hr ,
𝝉 = −J T (𝜓)z1 − K 2 z2 + 𝜣
T
(23)
T
⎡ 𝜽̂ 1
T
01×l 01×l ⎤
̂ = ⎢0⎢ ⎥
𝜽̂ 2 ∈ R3l×3 with 𝜽̂ i =
T
where K 2 = K 2 is the 3 × 3 positive definite design matrix, 𝜣
T
1×l 01×l ⎥
⎢ T ⎥
⎣ 01×l 01×l 𝜽̂ 3 ⎦
[𝜽̂ i1 ,𝜽̂ i2 , … , 𝜽̂ il ]T (i = 1, 2, 3) is the estimate matrix of 𝜣 * , and hr is the robust control term.
Motivated by Gutman,31 we design the adaptive law-based robust control term
z2 𝛿̂2
hr = − (24)
̂ 2 || + 𝜌
𝛿||z
with
𝛿̂̇ = 𝛾1 (||z2 || − 𝛾2 𝛿),
̂ 𝛿(0)
̂ > 0, (25)
where 𝛿̂ is the estimate of 𝛿, 𝜌 is any positive constant, and 𝛾 1 and 𝛾 2 are positive design constants.
To prevent the parameter drift and burst, the adaptive laws with the dead zone are designed as follows:
{
̇𝜽̂ = −𝜞 i [𝜷 i (𝝇)z2i + 𝜎i 𝜽̂ i ], if ∣ z2i ∣≥ bi , i = 1, 2, 3, (26)
i
0l ×1 , if ∣ z2i ∣< bi

where 𝜞 i = 𝜞 Ti is the l × l positive definite design matrix, 𝜎 i is the positive design constant, and bi > 0 is the dead zone
break point.
Remark 2. The effect of the dead zone is that the parameter adaptive laws ignore the smaller errors resulting from
the disturbances or the measurement noises and respond to the larger errors being related to the transient response
from large initial conditions. We introduce the dead zone in the adaptive laws to prevent the parameter drift and burst
due to the second-order wave disturbances, the measurement noises, etc. A much larger value would stop adaptation,
which brings the system to unstable, while a much smaller value would lead to bursting. The choice of dead zone
break point (threshold) depends on the characteristics of wave disturbances and measurement noises.
The main results are given in the following theorem.
Theorem 1. Consider the closed-loop system consisting of the ship (1)-(2) under Assumption 1 and the DP robust adap-
tive control law (23) with the robust control term (24)-(25) and the adaptive laws (26). The ship position (x, y) and heading
𝜓 are maintained at the desired value 𝜼d = [xd , yd , 𝜓 d ]T with arbitrarily small errors, while all signals in the DP closed-loop
control system are guaranteed to be uniformly ultimately bounded through appropriately choosing the positive design
constants 𝜌, 𝛾 1 , 𝛾 2 , 𝜎 i (i = 1, 2, 3) and the positive-definite symmetric design matrices 𝜞 i , K1 , K2 .

Proof. Choose the augmented Lyapunov function candidate

1 ∑ ̃T −1 ̃
3
1 ̃2
V2a = V2 + 𝜽i 𝜞 i 𝜽i + 𝛿 , (27)
2 i=1 2𝛾1

where 𝜽̃ i = 𝜽̂ i − 𝜽∗i (i = 1, 2, 3) and 𝛿̃ = 𝛿̂ − 𝛿.

In the light of (22), (23), and (25), the time derivative of (27) is
∑
3
V̇ 2a = V̇ 2 + ̃Ti 𝜞 −1 𝜽̂̇ i + 1 𝛿̃𝛿̂̇
𝜽
i=1
i
𝛾1
∑
3
= − zT1 K 1 z1 − zT2 K 2 z2 + zT2 [𝜣
T
̃ 𝜷(𝝇) + hr + d − 𝜺(𝝇)] + ̃iT 𝜞 −1 𝜽̂̇ i + 𝛿̃ ||z2 || − 𝛾2 𝛿̃𝛿.
𝜽 ̂ (28)
i
i=1

Subsequently, the stability analysis of the DP closed-loop control system is conducted in the following two cases.

Case 1. ∣ z2i ∣≥ bi (i = 1, 2, 3). Then, 𝜽̂̇ i = −𝜞 i [𝜷 i (𝝇)z2i + 𝜎i 𝜽̂ i ] from (26).

6 HU ET AL.

Using (24), (4), ‖𝜺(𝝇)‖ ≤ 𝜀 , (21) and 𝛿̃ = 𝛿̂ − 𝛿, we have

∗

∑
3
̃ T 𝜷(𝝇) + hr + d − 𝜺(𝝇)] +
zT2 [𝜣 ̃Ti 𝜞 −1 𝜽̂̇ i + 𝛿̃ ||z2 || − 𝛾2 𝛿̃𝛿̂
𝜽 i
i=1

T ∑
3
∑
3
̃ 𝜷(𝝇) + zT [hr + d − 𝜺(𝝇)] −
= zT2 𝜣 ̃Ti 𝜷 i (𝝇)z2 −
𝜽
T
̃i 𝜽̂ i + 𝛿̃ ||z2 || − 𝛾2 𝛿̃𝛿̂
𝜎i 𝜽
2 i
i=1 i=1

zT2 z2 𝛿̂2 ∑3
=− + zT2 [d − 𝜺(𝝇)] − ̃Ti 𝜽̂ i + 𝛿̃ ||z2 || − 𝛾2 𝛿̃𝛿̂
𝜎i 𝜽
̂ 2 || + 𝜌
𝛿||z i=1

zT z2 𝛿̂2 ∑
3
≤− 2
+ 𝛿||z2 || − ̃Ti 𝜽̂ i + (𝛿̂ − 𝛿)||z2 || − 𝛾2 𝛿̃𝛿̂
𝜎i 𝜽
̂ 2 || + 𝜌
𝛿||z i=1

𝛿̂2 ‖z2 ‖2 ̂ 2 || ∑
𝛿̂2 ‖z2 ‖2 + 𝜌𝛿||z
3
=− + − ̃Ti 𝜽̂ i − 𝛾2 𝛿̃𝛿̂
𝜎i 𝜽
̂𝛿||z2 || + 𝜌 ̂𝛿||z2 || + 𝜌 i=1

̂ 2 ||
𝛿||z ∑
3
=𝜌 − ̃Ti 𝜽̂ i − 𝛾2 𝛿̃𝛿̂
𝜎i 𝜽
̂ 2 || + 𝜌
𝛿||z i=1

∑
3
𝜎i 𝛾2
≤𝜌− (‖𝜽̃ i ‖2 + ‖𝜽̂ i ‖2 − ‖𝜽∗i ‖2 ) − (𝛿̃ 2 + 𝛿̂2 − 𝛿 2 )
i=1
2 2
∑ 𝜎i
3
𝛾2 ̃2
≤− ̃ i ‖2 −
‖𝜽 𝛿 + C1 , (29)
i=1
2 2
∑3 𝜎i 𝛾2 2
where C1 = 𝜌 + i=1 2 ‖𝜽i ‖
∗ 2
+ 2
𝛿 .

Case 2. ∣ z2i ∣< bi (i = 1, 2, 3). Then, 𝜽̂̇ i = 0l×1 from (26).

Since 𝜽̂ i is the constant vector due to 𝜽̂̇ i = 0l×1 and 𝜷(𝝇) is the bounded basis function vector, 𝜣
̃ T 𝜷(𝝇) is bounded. c*
in (21) is viewed as the upper bound of ||𝜣 ̃ T 𝜷(𝝇)|| in Case 2.
Using (24), (4), ‖𝜺(𝝇)‖ ≤ 𝜀 , (21) and 𝛿̃ = 𝛿̂ − 𝛿, we have
∗

∑
3
̃ T 𝜷(𝝇) + hr + d − 𝜺(𝝇)] +
zT2 [𝜣 ̃Ti 𝜞 −1 𝜽̂̇ i + 𝛿̃ ||z2 || − 𝛾2 𝛿̃𝛿̂
𝜽 i
i=1

zT2 z2 𝛿̂2 T
=− ̃ 𝜷(𝝇) + d − 𝜺(𝝇)] + 𝛿̃ ||z2 || − 𝛾2 𝛿̃𝛿̂
+ zT2 [𝜣
̂ 2 || + 𝜌
𝛿||z
zT2 z2 𝛿̂2
≤− + 𝛿||z2 || + (𝛿̂ − 𝛿)||z2 || − 𝛾2 𝛿̃𝛿̂
̂ 2 || + 𝜌
𝛿||z
𝛿̂2 ‖z2 ‖2 𝛿̂2 ‖z2 ‖2 + 𝜌𝛿||z
̂ 2 ||
=− + − 𝛾2 𝛿̃𝛿̂
̂ 2 || + 𝜌
𝛿||z ̂ 2 || + 𝜌
𝛿||z
̂ 2 ||
𝛿||z
=𝜌 − 𝛾2 𝛿̃𝛿̂
̂ 2 || + 𝜌
𝛿||z

𝛾2 ̃2 ̂2 ∑ 𝜎i 3
∑ 𝜎i 3
≤𝜌− (𝛿 + 𝛿 − 𝛿 2 ) − ̃ i ‖2 +
‖𝜽 ̃ i ‖2
‖𝜽
2 i=1
2 i=1
2

∑
3
𝜎i 𝛾2 ̃2
≤− ̃ i ‖2 −
‖𝜽 𝛿 + C2 , (30)
i=1
2 2

𝛾2 2 ∑3 𝜎i ̃ 2 ̃i || being a constant.
where C2 = 𝜌 + 2
𝛿 + i=1 2 ‖𝜽i ‖ with ||𝜽
HU ET AL. 7

Synthesizing (29) and (30), we rewrite (28) as follows:

∑
3
𝜎i 𝛾2 ̃2
V̇ 2a ≤ − zT1 K 1 z1 − zT2 K 2 z2 − ̃ i ‖2 −
‖𝜽 𝛿 +C
i=1
2 2
≤ − 2𝜇V2a + C, (31)
𝛾 𝛾
where 𝜇 = min{𝜆min (K 1 ), 𝜆min (K 2 M -1 ), mini=1,2,3 ( 12 𝜎i 𝜆min (𝜞 i )), 12 2 } with 𝜆min (·) denoting the minimum eigenvalue
of a matrix and C = max {C1 , C2 }.
Solving (31), we have [ ]
C C −2𝜇𝑡
0 ≤ V2a (t) ≤ + V2a (0) − e . (32)
2𝜇 2𝜇
It is obviously seen from (32) that V2a (t) is uniformly ultimately bounded. Then, it is known from (27) that ||z1 ||, ||z2 ||,
̃i || (i = 1, 2, 3) and ∣ 𝛿̃ ∣ are bounded; hence, ||𝜽̂ i || (i = 1, 2, 3) and ∣ 𝛿̂ ∣ are bounded. From (10) and the boundedness of
||𝜽
||𝜼d ||, ||𝜼|| is bounded; ||𝜶 1 || is then bounded from (15); hence, ||𝝊|| is bounded from (11). Therefore, uniform ultimate
boundedness of all signals in the DP closed-loop control system is achieved.
Furthermore, according to (27), (18), (13), and (32), we have
√ [ ]
C C −2𝜇𝑡
||z1 || ≤ + 2 V2a (0) − e . (33)
𝜇 2𝜇
√
It is obvious from (33) that, for any positive constant 𝜁z1 > C∕𝜇, there exists a constant Tz1 > 0 such that ||z1 || ≤ 𝜁z1
for all t > Tz1 . Thus, z1 settles within the compact set Ωz1 = {z1 ∈ R3 |||z1 || ≤ 𝜁z1 }, which can be made arbitrarily
small by appropriately selecting the positive design constants 𝜌, 𝛾 1 , 𝛾 2 , 𝜎 i (i = 1, 2, 3) and the positive-definite sym-
metric design matrices 𝜞 i , K1 , K2 . Therefore, the ship position (x, y) and heading 𝜓 are regulated to the desired value
𝜼d = [xd , yd , 𝜓 d ]T with arbitrarily small errors. Thus, Theorem 1 is proved.

4 SIMULATION AND COMPA RISON STUDIES

In this section, to verify the effectiveness and the superiority of the developed DP robust adaptive regulating con-
trol law, we carry out the high-fidelity simulations using the Marine Systems Simulator (MSS) toolbox,32 which is a
MATLAB/Simulink-based modular simulator developed in the Department of Marine Technology, Norwegian Univer-
sity of Science and Technology. The module MSS∖MSS Hydro∖Vessel Templates∖DP MotionRAO Model, which is a ship
motion simulator for station-keeping and low-speed maneuvering in six degrees of freedom, is adopted. Furthermore, the
simulation comparisons with an existing DP robust adaptive neural control law from the work of Du et al26 are presented.

4.1 Performance of proposed DP control law

In the following, to show the adaptability to the ship model parameter changes and the robustness against bounded
unknown time-varying disturbances of our proposed control law, the simulations on two ship examples are carried out
in two different disturbance cases.
Example 1. The ship example is a supply ship of 82.8 m in length, whose data is included in MSS.
In the simulations, the wave spectrum–type is JONSWAP, the peak frequency is 1.07 rad/s, the significant wave height
is 0.15 m, and the mean wave direction is (35◦ /180◦ ) × 𝜋 rad in the north-east frame, corresponding to the sea state 2.
The current speed is 0.1 m/s and the current direction is (350◦ /180◦ ) × 𝜋 rad. The mean wind angle is (30◦ /180◦ ) × 𝜋
rad and the wind speed is 2.65 m/s. Then, the corresponding LF disturbance curves are shown in Figure 1A. The desired
ship position in the north-east frame is taken as the coordinate origin, that is, 𝜼d = [0 m, 0 m, 0 rad]T . The dead zone
breakpoints are taken as b1 = 0.025, b2 = 0.008, and b3 = 0.0005. Let the NN 𝜽̂ i 𝜷(𝝇) (i = 1, 2, 3) contain 20 nodes with
T

centers 𝜅 j ( j = 1, 2, … , 20) evenly spaced in [−2, 20] × [−2, 20] × [−0.1, 0.2] × [−2, 1] × [−2, 1] × [−0.02, 0.01] and widths
hj = 5. The design parameters are chosen as follows: 𝜌 = 1 × 103 , 𝛾 1 = 3 × 104 , 𝛾 2 = 5 × 10−8 , 𝜎 1 = 𝜎 2 = 1 × 10−5 ,
𝜎 3 = 1 × 10−8 , 𝜞 1 = 𝜞 2 = 1 × 102 I20 × 20 , 𝜞 3 = 1 × 104 I20 × 20 , K1 = diag (1.2 × 10−2 , 1.5 × 10−2 , 2.0 × 10−2 ) and
K2 = diag(2 × 106 , 2 × 106 , 2 × 109 ). The initial conditions are taken as 𝜼(0) = [20 m, 20 m, (10◦ /180◦ ) × 𝜋 rad]T ,
8 HU ET AL.

(A) (B)

(C) (D)

(E)

FIGURE 1 Simulation results on Example 1 under 𝝉 and 𝝉 com . A, Low-frequency disturbances d1 , d2 , and d3 acting on the ship; B, Ship
position (x, y) and heading 𝜓; C, Ship surge velocity u, sway velocity v, and yaw rate r; D, Two-norms ||𝜽̂ 1 ||, ||𝜽̂ 2 ||, ||𝜽̂ 3 || and ||𝜽̂ com1 ||, ||𝜽̂ com2 ||,
||𝜽̂ com3 ||; E, Control inputs 𝜏 1 , 𝜏 2 , 𝜏 3 and 𝜏com1 , 𝜏com2 , 𝜏com3 [Colour figure can be viewed at wileyonlinelibrary.com]
HU ET AL. 9

̂ = 10 and 𝜽̂ 1 (0) = 𝜽̂ 2 (0) = 𝜽̂ 3 (0) = 020×1 . In addition,

𝝊(0) = [0 m/s, 0 m/s, 0 rad/s]T , b(0) = [10 kN, 10 kN, 100 kNm]T , 𝛿(0)
the measurement noises are taken as the zero-mean Gaussian white noises whose standard deviations are 0.05 m in north
position x, 0.05 m in east position y, and (0.1◦ /180◦ ) × 𝜋 rad in heading 𝜓, respectively.
To filter out the first-order wave-induced motions and the measurement noises of the ship position and heading, the
low-pass filter (34) and the notch filter (35) are chosen as3
1
h𝐿𝑃 = (34)
1 + 1.2s
s2 + 0.17s + 0.69
h𝑁𝑂 (s) = 2 . (35)
s + 1.66s + 0.69

The simulation results on Example 1 are depicted with solid lines in Figures 1B to 1E. To quantitatively evaluate the
control performance of the developed DP control law, the DP control performance indices are summarized in Table 1,
where the settling time ts (s) is the required time for all the ship north position x, east position y, and heading 𝜓 to arrive
at and remain within the specified error bands 3 m, 3 m, and (1◦ /180◦ ) × 𝜋 rad, respectively. It is observed from Figure 1B
and Table 1 that the ship position (x, y) and heading 𝜓 converge to the desired value 𝜼d = [0 m, 0 m, 0 rad]T in around
90 seconds. Figure 1C shows the boundedness of the ship surge velocity u, sway velocity v, and yaw rate r. Figure 1D
demonstrates the boundedness of the two-norms ||𝜽̂ 1 ||, ||𝜽̂ 2 ||, ||𝜽̂ 3 ||. Figure 1E shows that the control inputs 𝜏 1 , 𝜏 2 , 𝜏 3 are
bounded and reasonable.
Example 2. The ship example is a semisubmersible vessel of 115.0 m in length, whose data is included in MSS.
In the simulations, change the mean wave direction to (35◦ /180◦ ) × 𝜋 rad, the peak frequency to 0.92 rad/s, and the
significant wave height to 0.55 m, respectively, corresponding to the sea state 3. The current direction, current speed,
mean wind angle, and wind speed are the same as those of Example 1. Then, the corresponding LF disturbance curves
are shown in Figure 2A. In addition, the ship desired position vector, the remaining design parameters, and the initial
conditions are identical to the counterparts of Example 1. This means that the DP robust adaptive regulating control law
is kept unchanged for Example 1 and Example 2.
The simulation results on Example 2 are depicted with solid lines in Figures 2B to 2E and the DP control performance
indices are listed in Table 1. As seen from Figures 1B to 1E, 2B to 2E, and Table 1, our developed DP robust adaptive control
law exhibits satisfactory control performance for both Example 1 and Example 2 under the different environmental con-
ditions, which demonstrates that our developed DP robust adaptive control law has the robustness against environmental
disturbances and the adaptability to the model parameter changes.

4.2 Comparisons with existing DP control law

In this section, we compare our developed DP control law with the DP robust adaptive neural control law (36) with the
parameter adaptive laws (37) (see the work of Du et al26 )
̂ Tcom 𝜷(𝝇)
𝝉 com = −J T (𝜓)z1 − K 2 z2 + 𝜣 (36)

𝜽̂̇ comi = −𝜞 i [𝜷 i (𝝇)z2i + 𝜎i 𝜽̂ comi ], i = 1, 2, 3. (37)

In the simulations, the design parameter values in (40)-(41) are taken as the same as those of the counterparts of our
developed DP control law. The simulations are then carried out on Example 1 and Example 2 respectively, in Section 4.1.
The simulation results on Example 1 and Example 2 are depicted with dashed lines in Figures 1B to 1E and 2B to 2E,
respectively. The DP control performance indices are summarized in Table 1. It is observed from Figures 1D and 2D that

TABLE 1 Performance indices of two control laws on Example 1 and Example 2

Performance Indices Control Law 𝝉 Control Law 𝝉 com
Example 1 Example 2 Example 1 Example 2
Settling time 90 s 507 s 91 s ----
Mean deviation in north 0.5790 m 2.0685 m 1.2415 m 13.2900 m
Mean deviation in east 0.2766 m 0.6840 m 0.4529 m 3.5140 m
Mean deviation in yaw 0.2086◦ 0.8410◦ 0.2079◦ 0.8490◦
10 HU ET AL.

(A) (B)

(C) (D)

(E)

FIGURE 2 Simulation results on Example 2 under 𝝉 and 𝝉 com . A, Low-frequency disturbances d1 , d2 , and d3 acting on the ship; B, Ship
position (x, y) and heading 𝜓; C, Ship surge velocity u, sway velocity v, and yaw rate r; D, Two-norms ||𝜽̂ 1 ||, ||𝜽̂ 2 ||, ||𝜽̂ 3 || and ||𝜽̂ com1 ||, ||𝜽̂ com2 ||,
||𝜽̂ com3 ||; E, Control inputs 𝜏 1 , 𝜏 2 , 𝜏 3 and 𝜏com1 , 𝜏com2 , 𝜏com3 [Colour figure can be viewed at wileyonlinelibrary.com]
HU ET AL. 11

the two-norms ||𝜽̂ com1 ||, ||𝜽̂ com2 ||, and ||𝜽̂ com3 || of parameter estimates drift, which is caused by the first-order waves, the
measurement noises, etc. It is clearly seen from Figures 1B, 2B, and Table 1 that the control performance of our developed
DP control law is superior to 𝝉 com .
Remark 3. In the simulations, we firstly choose the design parameters K1 and K2 satisfying 𝜆min (K1 ) > 0 and
𝜆min (K2 ) > 0 through the trial and error to ensure that the system is stable. Furthermore, we properly regulate the
other design parameters satisfying 𝜆min (𝜞 i ) > 0, 𝛾 1 > 0, 𝛾 2 > 0, bi > 0, 𝜎 i > 0, and 𝜌 > 0 to get the satisfactory control
performance. The chosen design parameters could affect the results of simulations. An amount of simulations done
in many scenarios show that the larger the parameters K1 , K2 , 𝜞 i , and 𝛾 1 are, the better the control accuracy is, while
the larger the control signal 𝝉 is.

5 CO N C LUSION S

We have developed a robust adaptive regulating control law for the DP of ships with unknown dynamics and unknown
disturbances incorporating the RBF NNs, the dead zone adaptive technique, and a robust control term into the vectorial
backstepping method. The dead zone has been introduced in the NN weight adaptive laws to prevent the parameter
drift and burst. The ship position and heading are maintained at the desired values with arbitrarily small errors, while
all signals in the DP closed-loop control system are guaranteed to be uniformly ultimately bounded. The high-fidelity
simulation results on two supply ships and the simulation comparisons with an existing DP control law have validated
the developed DP control law.

ACKNOWLEDGEMENTS
This work was supported in part by the National Natural Science Foundation of China under Grant 51079013, in part by
the Program for Liaoning Excellent Talents in University under Grant LR2015007, in part by the Technology Foundation
for Selected Overseas Chinese Scholar, and in part by the Fundamental Research Funds for the Central Universities under
Grant 3132017302.

ORCID

Xin Hu https://orcid.org/0000-0001-9043-9639
Jialu Du https://orcid.org/0000-0002-9809-0862
Miroslav Krstić https://orcid.org/0000-0002-5523-941X

REFERENCES
1. Sørensen AJ. A survey of dynamic positioning control systems. Annu Rev Control. 2011;35:123-136.
2. Tannuri EA, Agostinho AC. Higher order sliding mode control applied to dynamic positioning systems. IFAC Proc Vol. 2010;43(20):132-137.
Part of special issue: 8th IFAC Conference on Control Applications in Marine Systems.
3. Fossen TI. Guidance and Control of Ocean Vehicles. Chichester, UK: John Wiley & Sons, Inc.; 1994.
4. Balchen JG, Jenssen NA, Salid S. Dynamic positioning using Kalman filtering and optimal control theory. In: Proceedings of IFAC/IFIP
Symposium on Automation in Offshore Oil Field Operation; 1976; Bergen, Norway.
5. Balchen JG, Jenssen NA, Salid S. Dynamic positioning of floating vessels based on Kalman filtering and optimal control. In: Proceedings
of the 19th IEEE Conference on Decision and Control including the Symposium on Adaptive; 1980; Albuquerque, NM.
6. Grimble MJ, Patton RJ, Wise DA. Use of Kalman filtering techniques in dynamic ship positioning systems. IEE Proc D-Control Theory
Appl. 1980;127:93-102.
7. Grøvlen A, Fossen TI. Nonlinear control of dynamic positioned ships using only position feedback: an observer backstepping approach.
In: Proceedings of the 35th IEEE Conference on Decision and Control; 1996; Kobe, Japan.
8. Fossen TI, Grøvlen A. Nonlinear output feedback control of dynamically positioned ships using vectorial observer backstepping. IEEE
Trans Control Syst Technol. 1998;6:121-128.
9. Chang WJ, Chen GJ, Yeh YL. Fuzzy control of dynamic positioning systems for ships. J Mar Sci Technol. 2002;10:47-53.
10. Godhavn, JM, Fossen, TI, Berge, SP. Non-linear and adaptive backstepping designs for tracking control of ships. Int J Adapt Control Signal
Process. 1998;12(8):649-670.
12 HU ET AL.

11. Bertin D, Bittanti S, Meroni S, Savaresi SM. Dynamic positioning of a “single-thruster” vessel by feedback linearization. IFAC Proc Vol.
2000;33(21):275-280. Part of special issue: Proceedings of the IFAC Conference on Manoeuvring and Control of Marine Craft.
12. Fossen TI, Strand JP. Nonlinear passive weather optimal positioning control (WOPC) system for ships and rigs: experimental results.
Automatica. 2001;37:701-715.
13. Veksler A, Johansen TA, Borrelli F, Realfsen B. Dynamic positioning with model predictive control. IEEE Trans Control Syst Technol.
2016;24:1340-1353.
14. Fossen TI, Strand JP. Passive nonlinear observer design for ships using Lyapunov methods: full-scale experiments with a supply vessel.
Automatica. 1999;35:3-16.
15. Loria A, Fossen TI, Panteley E. A separation principle for dynamic positioning of ships: theoretical and experimental results. IEEE Trans
Control Syst Technol. 2000;8:332-343.
16. Hassani V, Sørensen AJ, Pascoal AM, Aguiar AP. Multiple model adaptive wave filtering for dynamic positioning of marine vessels. Paper
presented at: 2012 American Control Conference (ACC); 2012; Montreal, Canada.
17. Nguyen TD, Sorensen AJ, Quek ST. Design of hybrid controller for dynamic positioning from calm to extreme sea conditions. Automatica.
2007;43:768-785.
18. Hu X, Du JL, Sun YQ. Robust adaptive control for dynamic positioning of ships. IEEE J Ocean Eng. 2017;42(4):826-835.
19. Du JL, Hu X, Krstić M, Sun YQ. Robust dynamic positioning of ships with disturbances under input saturation. Automatica.
2016;73:207-214.
20. Do KD. Global robust and adaptive output feedback dynamic positioning of surface ships. J Mar Sci Appl. 2011;10:325-332.
21. Kjerstad OK, Skjetne R, Jenssen NA. Disturbance rejection by acceleration feedforward: application to dynamic positioning. IFAC Proc
Vol. 2011;44(1):2523-2528. Part of special issue: 18th IFAC World Congress.
22. Tannuri EA, Agostinho AC, Morishita HM, Moratelli JL. Dynamic positioning systems: an experimental analysis of sliding mode control.
Control Eng Pract. 2010;18:1121-1132.
23. Hassani V, Sørensen AJ, Pascoal AM, Athans M. Robust dynamic positioning of offshore vessels using mixed-𝜇 synthesis modeling, design,
and practice. Ocean Eng. 2017;129:389-400.
24. Chang WJ, Liang HJ, Ku CC. Fuzzy controller design subject to actuator saturation for dynamic ship positioning systems with
multiplicative noises. Proc Inst Mech Eng Part I: J Syst Control Eng. 2010;224(6):725-736.
25. Du JL, Hu X, Krstić M, Sun YQ. Dynamic positioning of ships with unknown parameters and disturbances. Control Eng Pract.
2018;76:22-30.
26. Du JL, Yang Y, Wang DH, Guo C. A robust adaptive neural networks controller for maritime dynamic positioning system. Neurocomputing.
2013;110:128-136.
27. Ge SS, Hang CC, Lee TH, Zhang T. Stable Adaptive Neural Network Control. Boston, MA: Kluwer; 2002.
28. Mears MJ, Polycarpou MM. Stable neural control of uncertain multivariable systems. Int J Adapt Control Signal Process.
2010;17(6):447-466.
29. Krstić M, Kanellakopoulos I, Kokotović P. Nonlinear and Adaptive Control Design. New York, NY: John Wiley & Sons, Inc.; 1995.
30. Krstić M, Banaszuk A. Multivariable adaptive control of instabilities arising in jet engines. Control Eng Pract. 2006;833-842.
31. Gutman S. Uncertain dynamical systems—a Lyapunov min-max approach. IEEE Trans Autom Control. 1979;24:437-443.
32. Fossen TI. Handbook of Marine Craft Hydrodynamics and Motion Control. Chichester, UK: John Wiley & Sons, Inc.; 2011.

How to cite this article: Hu X, Du J, Krstić M. Robust adaptive regulation of dynamically positioned
ships with unknown dynamics and unknown disturbances. Int J Adapt Control Signal Process. 2019;1–12.
https://doi.org/10.1002/acs.2972

View publication stats