Advances in Intelligent Systems Research (AISR), volume 151

2018 International Conference on Computer Modeling, Simulation and Algorithm (CMSA 2018)

Research on Drift Path Prediction Technology of

Ships
Dongdong An1, Tianchi Zhang1, Jing Zhang1,2,*, Zhimin Liu1 and Jiasen Chen1
1
College of computer science and technology, Harbin Engineering University, Harbin, 150001, China
2
College of Engineering, Shantou University, Guangdong, 515063, China.
*
Corresponding author

Abstract—In order to improve the prediction fitting degree proposed a random particle simulation method, which describes
of the existing method to predict the drift path of the ship, the the influence of the uncertain and volatile marine environment
particle reloading simulation method was proposed to predict the and the wind pressure coefficient on the ship drifting.
drift path from the aspect of segment prediction. Particle Reference [4] proposed Monte Carlo prediction method, it
reloading simulation method divides the drift time into multiple based on classical probability theory and mathematical
segments, and then the information is updated after each segment. statistics theory and stochastic process. Reference [5] based on
The updating will continue the whole drift process, and then the the Leeway offshore drift theory and the Monte Carlo method.
drift path will be obtained. Experimental results indicate that our The experimental result was basically consisted with the actual
method is extremely promising for predicting drift paths of ships.
results.
This is due to the avoidance of the problem of the Monte Carlo
method, which prediction accuracy of the long-distance decreases In this paper, we shall study the Monte Carlo method and
with time. then propose the particle reloading simulation method to
predict the drift path of the ship. The rest of this paper is
Keywords—drift path; particle reload simulation; prediction organized as follows. In section 2, we give the basic principles
degree; Monte Carlo of Monte Carlo method. In section 3, we analyze the ship drift
state and put forward a novel method that we call particle
I. INTRODUCTION reloading simulation.In section4, the paper provides
With the prosperity of maritime transport, there is the sea experimental results to validate the novel method. Finally, what
accident.The incidence of maritime accidents has been is presented is the conclusion.
increasing gradually, which has attracted people's attention.
Because the accident occurs on the sea, it is very difficult to II. MONTE CARLO METHOD
carry out the rescue. After accidents happen, we cannot rescue Monte Carlo prediction is one of the most widely used
real-time at sea, then the wrecked ship will drift over time, and methods in the drift path prediction. In the using of ship drift
the rescue will not be easy to obtain good results. If we can model, Monte Carlo method [6] simulates the ship drift process
predict the drift path of the wrecked ship, we can gain more at first, and then calculates the initial probability distribution
time for the rescue, and minimize the loss as much as possible. during the ship drift. Finally it solves the problem of ship drift
The existing methods of ship drift path prediction only solve path. Monte Carlo method bases on probability statistics, it
the problem of complex calculation, and it is better to predict regards the problem as a random event, and uses the random
the drift path of short distance. But in a long distance, the test to simulate the random feature of the problem of digital
performance is not satisfactory. At the same time, the fitting features, and the numerical features that obtained from the
degree and accuracy are not high. Precise prediction of paths is simulation of random events are considered as approximate
very important in helping to rescue and reduce casualties and solutions to the problem.
there is a pressing need for improved methods to help solve
challenging the prediction of the drift path problems. The drift model of the ship on the sea can be expressed by
the following simplified formula (only considering wind drift
For the study of ship drift paths on the sea, we need to and flow drift)
consider influencing factors of ship drift. There are some
existing methods for the prediction of the drift paths and the
analysis of drift on the sea. Reference[1] used a probability t t
model to predict the location and velocity of the drifting objects x(t )  x 0   V (t )dt   [ L(t )  uw(t )]dt (1)
0 0
at each moment in the study of the drift trajectories of non-
powered ships, it obtained result that is the closest real speed of
drift at a time. Reference[2] considered the influence of the Where L(t ) is the wind drift, uw(t ) is the flow drift, x 0 is
natural environment factors such as the wind field and the flow initial drift distance.
field on the drift, and types of drifting objects were classified. The wind drift of the ship drift model is the drift that the
If final location of the drifting object and the type are known, ship is exposed to air and dragged or pushed by wind. Since the
finally they could obtain the drift trajectory. Reference [3] shape of the drift ship is not necessarily symmetrical, a wind

Copyright © 2018, the Authors. Published by Atlantis Press. 231

This is an open access article under the CC BY-NC license (http://creativecommons.org/licenses/by-nc/4.0/).
 Advances in Intelligent Systems Research (AISR), volume 151

pressure angle between the direction of the ship drift and the Through the studying of wind drift and wind speed data can
downward wind direction is caused by a lateral wind force, and be obtained between about wind induced drift and wind speed
the wind pressure angle is likely to occur on either side of the regulation, but there are errors and accidental factors in
wind direction. According to the empirical formula, wind drift obtaining wind speed and linear relationship, Therefore, the
L(t ) is obtained as follows: slope and intercept of the linear relationship need to add a
random perturbation function[9],which satisfies the distribution
N(0, σ ). Wind drift coefficient is:
L(t )  Ld  adW 10  bd (2)

Where ad and bd are a linear parameter that obtained by a  εn

large number of fitting experiments. They are related to the an  a  20
type of floating ships. W 10 is the wind speed at 10m of the sea  (5)
surface. bn  b  εn
 2
In this paper, the numerical prediction of surface flow is
realized by using the FVCOM model [7], it can be better to fit When W 10 is 10m/s, the slope and the intercept have the
the complex curvature of the coastline by using the triangular
same perturbation,
unstructured meshes. In order to protect the energy
conservation of the influencing factor, the finite volume
method is adopted to process the physical quantities. The εn εn
FVCOM model is not adequately considered on the L  (a  )W 10  (b  ) (6)
temperature, salinity, density and so on. The equation of the 20 2
FVCOM model is given:
Equation 6 is taken into Equation 4 to obtain the ship drift
of the Monte Carlo prediction method.
uD u 2 D uvD uω Monte Carlo prediction method is to solve the low fitting
    fvD 
t x y σ degree of between the predicted drift path and the real drift path.
It’s solution is to exclude the effects of uncertain factors in the
γ gD  0 D drift process. Firstly, it chooses two main factors (wind and
 gD  [ ( D  ρdσ )  σρ ] (3) flow fields), while the drift ship is treated as the particle, and
x ρε x σ x the disturbance analysis is performed when the method
1  calculates the wind field, flow field data and some uncertain
 ( Km σu )  DFx  uω(t ) factors are excluded as much as possible.
D σ
Although the Monte Carlo prediction method achieved
The equation (1) is calculated by the second-order Longge good results, Monte Carlo prediction method does not solve to
Kutta method, and the equation (2) and the equation (3) can be how to obtain real-time drift velocity in the drift process. While
taken to obtain the drift distance. the method improves the accuracy of the drift path, but with the
increase of the drift time, the error is still increasing. So we
In order to solve the problem that the drift trajectory and the proposed a particle reloading simulation method based on the
real trajectory drift with time, the Monte Carlo method is Monte Carlo method to solve the problem.
introduced to solve the uncertain factors and errors in the
calculation process. The floating ship is defined as a single
particle, and the drift trajectory of the particle meets the III. PARTICLE RELOADING SIMULATION METHOD
random process of probability distribution. Under the premise The first step of particle reloading simulation method is to
of no changing rules, it is assumed that motions of these determine the initial probability distribution of the drift ship in
particles do not affect each other, and the appropriate method is time and space, and then deduce the drift path. Different initial
used to copy particles simultaneously. This is the idea of the probability distributions come from different influencing
drift model, which can obtain the drift regular distribution area factors on surface of ocean. It leads to a different drift path that
of the particle over a period of time. This paper assumes that affects the judgment of the final drift area of the ship drift. The
the process of drift ship is in accordance with the Markov particle reloading simulation prediction method can well
process [8], the probability density distribution function of the describe the initial probability distribution of ship drift in
next state of the floating ship is related only to the current state complex environment. The path prediction of drift ship may
of the floating ship and not related to other factors. The random appear multiple results. We assign them to different scenes
drift process is expressed by as follows: according to certain rules, and obtain different initial
probability distributions according to the different scenes. Then,
we analyze the force state of particles and get the velocity of
dx  V ( x, t )dt  dε (4) particle drift. At last, we can acquire the drift path.

232
 Advances in Intelligent Systems Research (AISR), volume 151

A. Simulation Particle Drift State Analysis B. A New Path Generation Method Based on Particle
In this paper, drifting ship is transformed into a particle that Reloading Simulation
has no influence each other, and they have the properties of We know the speed v. Where t= t , it is small enough. The
floating ship and environmental forces. We need to analyze method can form the integral curve, this curve is the drift path.
properties and force situation of ships, and then get the ship's
force situation:
t
y  x  x 0   vdt (10)
(m  m )dV n 0

  Fi (7)
dt i 1
This method is predictive for short distance drift paths, but
for drift predictions can produce errors on long distance. In this
case, total time of the drift is divided into segments, and then
By studying the drift ship, there are four main forces that particles are reloaded for each period of time. Each period will
affect the drift of the ship. They are the drag force of the wind, update initial probability distribution of the particle and force
the radiation force of the waves, the drag force of the flow and state of the particle at this time, which will make the drift path
the thrust of the turbulence. In this paper, particles are only closer to the real state:
affected by the above four forces. As for the intrusion ratio and
the material of the ship, particles are automatically acquired,
because they included above forces. According to the drift of t1 t2 tn
the assumed condition, the acceleration of the ship drift model y  x  x 0   vdt  vdt     v n (11)
0 t1 tn 1
can be expressed as:
After particle updates the information in N-1th time,
tn n
 (m  m )dV n
tn1 v is based on the force condition and the initial
 dt
  Fi  Fa  Fw  Fc  Ft probability distribution to obtain the drift path in Nth time.
 i 1

 1 The particle reloading simulation prediction method is

 Fa  2CdρaAaWa 2 advanced for Monte Carlo prediction method. Firstly,
according to the initial information of drift particles (mainly
 wind field, flow field, stress situation and floating material), the
 1
 Fw  (8) Monte Carlo method is used to obtain the initial probability
 2CiwgρwLtA2 distribution of particles. We select the drift time t the particle,
 1 analyze the force and the state of the particle in time t , and
 FC  obtain the drift velocity and direction of the particle according
 2CcdρwAwL2 b to above information. If the time of the particle drift is short
 Ft  2 R 6 Kαt / t 2 enough, the method can obtain drift path in time t . If the
 time t is relatively long, it will be divided into N parts. We
 repeat the process. Finally, at the end of each t , it is
Where, Fa indicates the drag force of the wind, Fw determined whether the particle initial information needs to be
Represents the radiant force of the waves, Fc is the drag force updated, then a new particle probability distribution is obtained.
of the flow, Ft means the thrust of the turbulence, m  The particle drift analysis is performed again.
originates from the acceleration of water particles on the The algorithm of particle reloading simulation is as follows:
surface of the drift ship. Cd and Ccd denote the drag force of Input: initial information of the drift particle(including the
wind and flow, Ciw represents the coefficient of thrust of a wind field, flow field and its own attributes and other
wave, ρw and ρa respectively indicate the actual density of information); drift time;
seawater and air. Output: Drift path.
Under normal conditions, drift velocity of particles can be Step1: Enter or read the initial information of the drift
obtained according to the following formula: particle to obtain the initial probability distribution;
Step2: Enter the particle drift time and analyze the state of
t
v  v 0   adt (9) particle drift;
0
Step3: determine whether the need to split the particle run
time, if it is, it executes Step4, otherwise executes Step7;
Where, v 0 is the initial drift velocity of the particle, t is
the time of the drift,a is drift acceleration of the particle. Step4: the drift time is divided into N parts, sequentially
update the particles in these time periods

233
 Advances in Intelligent Systems Research (AISR), volume 151

Step5: Output sequentially the drift paths in each time

period;
Step6: determine whether the combination of output time is
complete, if it is, it executes Step7, otherwise executes Step8.
Step7: Output the complete drift path;
Step8: exit algorithm.
FIGURE II. THE PATHS OF THE TWO METHODS AND ORIGINAL
The particle reloading simulation method divides the total DRIFT
drift time into multiple parts, and immediately updates the new
drift state of the particle after generating a small drift path. In fig3, at the top the red frame is the initial stage of the
According to the new state of the particle information, it drift;the black solid line is the middle stage of the drift;The
generates a new drift path, which it will continue the entire drift dashed frame is the last stage of the drift.
process. The information constantly updated, the error
continues to decrease. This approach makes the prediction of
the entire drift path more and more close to the real drift path
and achieves good results.

IV. RESULTS AND DISCUSSION

In the experiment, the most typical inshore is chosen as a
drift location. The inshore drift is the most frequent occurrence FIGURE III. THE ANALYSIS CHART OF THE DRIFT TRAJECTORY
of accidents in recent years and it is representative.For
influencing factors of ship drift,we choose the vector In summary, it can be found that the drift path of method 1
decomposition method.This section uses the Monte Carlo and method 2 is basically consistent with the original drift path
prediction method and the particle reloading simulation method at the initial stage of drift. This is because of both the method 1
for comparative experiments in this paper. and the method 2 obtain the initial probability distribution of
The starting point for this experiment is the ship drift model according to the Monte Carlo method,
S(36.32N,120.73E) , it is the waters of the grass island. The which has a great influence on the ship drift. In the middle and
end point is E(30.27N,117.35E), It is the open sea of Mount late stages of the drift, the fitting degree of the method 1 higher
Feng. In the experiment, the original drift trajectory is than method 2. Therefore, the particle reloading simulation
obtained.In fig1, the path is wavy because of the influence of method is more suitable for obtaining the ship drift path.
periodic tides and ocean current.
V. CONCLUSION
In this paper, the technique of obtaining the drift path of the
ship is analyzed. Based on the analysis of the traditional
method, this paper puts forward the method of particle
reloading simulation. By decomposing the time in the drift
process, the information is updated at the beginning of each
time period to obtain the initial probability distribution of the
FIGURE I. ORIGINAL DRIFT TRAJECTORY ship drift. The drift path and the original drift path have a good
fit degree. Two different methods of obtaining the drift
Experiment: trajectory of the ship are compared by experiment. The
The vector decomposition method is the Monte Carlo experimental results show that the proposed method has a
prediction method and the particle reloading simulation method higher fit degree with the original drift path.
combined respectively.The influence factors of ship drift are
dealt with by vector decomposition method.Two methods ACKNOWLEDGEMENTS
obtain the drift trajectory of the ship.We compare their fitting This research is supported by
degree of drift trajectory with the original drift path fitting.
(1) 2017-2020 National Natural Science Foundation of
The red path is the original drift trajectory. The method1 is China under Grand No.51679058
the combination of vector decomposition and particle reloading
simulation method, its drift path is black.The method2 is the (2) 2013-2016 Higher Specialized Research Foundation
combination of vector decomposition and Monte Carlo (PhD supervisor category) of China under Grand No.
prediction method, its drift path is blue. 20132304110018

REFERENCES
[1] Allen A A, Plourde J V. Review of Leeway: Field Experiments and
Implementation[J]. Review of Leeway Field Experiments &
Implementation, 1999.

234
 Advances in Intelligent Systems Research (AISR), volume 151

[2] Breivik.Fast ray sorting and breadth-first packet traversal for gpu ray
tracing. Computer Graphics Forum, 2013,21(3):pp.132-137.
[3] Xiao F, Yin Y, Jin Y, Zhang X,. Determination of Maritime Search Area
Based on Stochastic Particle Simulation[J]. Navigation of China, 2011,
34(3):pp.34-39.
[4] Zhang, Jinfen.Probabilistic modelling of the drifting trajectory of an
object under the effect of wind and current for maritime search and
rescue.Centre for Marine Technology and Ocean
Engineering.2017,129:pp.253-264.
[5] Gao J, Lin M U, Wang G S, Cheng L I,Dong J X,Bao X W,et al. Drift
analysis and prediction of debris from Malaysia Airlines flight
MH370[J]. Chinese Science Bulletin, 2016.
[6] Zhong L, Sun Youchao. Research on System Maintenance Time
Simulation Model Based on Monte Carlo Method and Roulette Rule[J].
China Mechanical Engineering, 2006, 17(10):pp.999-1002.
[7] Changsheng C, Beardsley R C, Cowles G. An Unstructured-Grid Finite-
Volume Coastal Ocean Model (FVCOM) System[J]. Oceanography,
2006, 19(1):pp.78–89.
[8] Breivik ,Allen A.An operational search and model for Norcve ian Sea
and the Nonh Sea.Journal of Marine Systems.2008.69:pp.99-113.
[9] Hu, Jian .Experimental and numerical study on wave drift forces on a
semi-submersible platform in waves.Ships and Offshore
Structures.2017.12(1):pp.56-65

235