1 INTRODUCTION

Maritime transportation plays an important role in

the global economy and nearly 90% of trade is

realized by maritime transportation. Route design is

one of the main tasks faced by navigators. The quality

of route is related to the navigation safety and

economic benefits of ships. Generally, the route

should be able to satisfy two main objectives, which

are avoiding dangerous areas and ensuring economic

benefits, respectively. For maritime transport, it is

important to design safe routes quickly and easily.

Ship route design is usually completed by the

captain and chief officer. Although this approach will

not cause problems in most cases, the manually

designed route is still affected by the professional

quality, sailing experience and personal emotion of

the crew. Therefore, is not necessarily the optimal

route.

With the development of computer technology,

mapping technology and information technology, the

current marine data can accurately and completely

describe the navigation environmental information. In

this case, the researchers began to use heuristic

algorithms to automatically generate ship route and

many methods and procedures have been developed

to determine the best route available (Dijkstra 1959, E.

Kobayashi 2011, Choi et al. 2015). However, in

complex situations, it is difficult to realize in the

actual navigation environment, and it is difficult to

accurately and completely represent the obstacles in

the complex environment with many obstacles. The

crew still rely more on sailing experience and

reference routes to complete the route design.

In this paper, a path optimization method based

on historical information generation is proposed,

which can be a planned route in a certain area of a

ship of different types, sizes and headings. The

maritime department or the shipping company can

use this method to obtain an optimized route and

provide reference for the route planning of the ship. A

path optimization method based on historical

Ship Route Planning Using Historical Trajectories

Derived from AIS Data

Y.K. He, D. Zhang, J.F. Zhang & M.Y. Zhang

Wuhan University of Technology, Wuhan, China

National Engineering Research Center for Water Transport Safety, Wuhan, China

T.W. Li

ChangJiang Waterway Transportation Monitoring and Emergency Center, Wuhan, China

ABSTRACT: Ship route planning is one of the key issues in enhancing traffic safety and efficiency. Many route

planning methods have been developed, but most of them are based on the information from charts. This paper

proposes a method to generate shipping routes based on historical ship tracks. The ship's historical route

information was obtained by processing the AIS data. From which the ship turning point was extracted and

clustered as nodes. The ant colony algorithm was used to generate the optimize route. The ship AIS data of the

Three Gorges dam area was selected as a case study. The ships’ optimized route was generated, and further

compared with the actual ship's navigation trajectory. The results indicate that there is space of improvement

for some of the trajectories, especially near the turning areas.

http://www.transnav.eu

the International Journal

on Marine Navigation

and Safety of Sea Transportation

Volume 13

Number 1

March 2019

DOI: 10.12716/1001.13.01.06

information generation is proposed, which can be a

planned route in a certain area of a ship of different

types, sizes and headings.

The paper is organized as follows: section one

mainly introduces the research on route planning, and

section two introduces the route generation method.

In the third section, the AIS data of the three gorges

region of China were used to generate air routes and

discuss. Finally, the fourth part is the conclusion of

this paper.

2 RELATED STUDIES

2.1 AIS data in maritime transportation

The ship AIS equipment transmits information at

different frequencies according to the equipment level

and the navigation status of the ship in accordance

with the regulations of the local maritime

administration. The format of the information is

strictly defined, including the actual position, speed,

heading, and the direction of the ship, as well as the

static or semi-static information such as ship's name,

call sign, draught, ship size (Tetreault 2005).

At present, all the ships above a certain tonnage

are forced to install the AIS system. And a lot of small

ships also installed the AIS system. A large amount of

AIS data has been used in many studies in maritime

transport, such as risk assessment analysis

(Montewka and Kujala 2014, Hänninen and Kujala

2014), ship collision avoidance (Wang et al. 2013,

Zhang et al. 2015), traffic flow analysis (Xiao et al.

2015), etc.

AIS data can be used to generate ship historical

routes. Zhang et al. deleted the abnormal AIS data by

determining the state of the ship and used the

regression model to smooth the route generated by

the AIS data (Zhang et al. 2018). Wang et al. proposed

a modified clustering algorithm to extract and analyze

ship routes (Wang Gao and Yang 2017). To solve the

problem of missing AIS data, Sang et al. proposed a

curve fitting method to restore the vessel trajectory

(Sang et al. 2015). Although there are some problems

in AIS data, such as noise and absence, it can still

generate the ship's historical route with a satisfactory

accuracy.

2.2 Ship route planning

There are many ship route design methods based on

heuristic algorithm, such as Dijkstra algorithm,

genetic algorithm, ant colony algorithm, A*

algorithm. In order to avoid problems such as local

optimal solution and long calculation time, many

researchers have proposed improved methods

combining other relevant algorithms (Lee et al. 2018,

Vettor and Guedes Soares 2016, Roh 2013).

A lot of the above methods are cell-based, which

means that obstacles need to be gridded. Therefore, it

is difficult to accurately and completely express

obstacles in the complex environment of obstacles

such as bridges, shallow waterways, ship anchorage,

reefs and navigation marks.

Xiao et al.'s study on the traffic flow pattern of

ships shows that different types and sizes of ships

have different route positions, which is especially

obvious in waterways with traffic separation

management (Xiao et al. 2015). Most existing methods

do not take into account the route differences of

different ships.

The method proposed in this paper avoids the grid

description of obstacles and can design different

routes according to ship type, size or other special

requirements, which overcomes the limitations of

existing methods.

3 METHOD FOR THE ROUTE PLANNING

In this section, the route design method proposed is

introduced. This method obtains the ship's historical

route through AIS data, and then clusters the

optimization to design the route.

3.1 Extraction of historical routes

This section describes how to get historical routes

from AIS data. The time interval at which the AIS

device sends the message is short, and the message

contains the location information of the ship.

Therefore, the historical route of the ship can be

obtained by processing the AIS data. The AIS data

includes the ship's MMSI and time. It can classify

historical routes and select different historical route

information according to actual needs.

3.1.1 AIS data selection and classification

This paper mainly uses dynamic and static AIS

data. Dynamic data provides the latitude and

longitude of the ship at each moment and can be used

to map the ship's historical routes. Static data is

mainly used to classify ships. It should be noted that

since the ship's static data only divides the ship into

passenger ships, cargo ships, oil tankers, tugs and

official ships, it is necessary to use the MMSI number

to inquire about the specific ship type.

The purpose of classifying AIS data is to obtain

different types of historical routes. The AIS data

includes the ship's MMSI. The ship can be classified

by the unique MMSI to obtain ship route information

of different types, sizes and draughts. At the same

time, according to the special requirements of the

local maritime administration, the route information

of special ships or dangerous goods ships can be

selected. Further, considering the hydrological

environment of the waters, the route information of

the dry season, the flat-water period and the flood

season can be obtained by selecting the AIS data of

different seasons.

3.1.2 Outliers removal

The outliers handled in this article is "particularly

egregious" error data. Since the turning points will be

clustered, the influence of random errors can be

ignored. The processing of the abnormal data is

divided into two steps, which are removal and

replacement.

For the abnormal data of the velocity, the

confidence range is set by determining its maximum

and minimum values. The scope of credibility is

determined based on actual conditions. The speed

range of ships in different navigation areas is

inconsistent, different types of ships have different

speed ranges as well. The choice of speed range

should be determined by reference to the ship's

performance of the same type and size and the

regulations of the local maritime administration.

Figure 1 shows the speed in the AIS data of a bulk

carrier in an inland area, and the confidence range is

set to be between 1 and 20 knots.

Figure 1. Analysis of abnormal velocity data

The position abnormal data, that is, the longitude

or latitude, is abnormal, and the position deviates

from the route. In this paper, a confidence range is

established, and it is considered that the abnormal

data is outside the current trusted range. The

confidence range is determined by the position of the

previous moment, the position of the latter moment,

and the speed.

Depending on the speed, the maximum distance

the ship can navigate in a fixed time span can be

calculated, during which time the ship will not move

to a further position. Since the starting and ending

points of the movement of the ship during this time

are certain (i.e. the position at the previous moment

and the position at the next moment), the credible

range should be the ellipse with the starting point and

the end point as the focus. The sum of the distance

from any point on the ellipse to the focus is the

maximum distance the ship can move.

Figure 2. Analysis of abnormal position data

The long axis a and the short axis b of this ellipse

are:



nn n

vt t







(1)



nn n

vt t



















(2)

where: t

n-1 is the time at the previous moment; tn+1 is

the time at the later moment; l is the distance between

the position of the ship at the previous moment and

later moment; v is the speed of the ship.

3.1.3 Outliers restoration

After removing the outliers, the missing speed

data and location data should be restored.

AIS devices send dynamic data at intervals

ranging from 2 seconds to 3 minutes. The time

interval of sending dynamic data is very short when

the ship's navigation state are changing greatly, such

as large angle steering and acceleration or

deceleration. On the contrary, the time interval of

sending dynamic data is longer when the change of

navigation status is small. Therefore, the missing data

can be obtained simply by linear interpolation.

Suppose that the point to be restored is n, the point

at the previous moment is n-1, and the point at the

latter moment is n+1, then the longitude, latitude and

velocity of the point are:



nn nn

vv tt











 







(3)



nn nn

La La

La La t t











 







(4)



nn nn

Lo Lo

Lo Lo t t











 







(5)

where: V is the velocity at this point, La is the latitude,

and Lo is the longitude.

3.1.4 Generation and screening of historical routes

The processed AIS data contains more accurate

longitude and latitude information at each moment of

the ship and can be used to describe the ship's

trajectory. Ship routes on short-haul flights lack

reference value and should be deleted. In addition,

due to the failure of AIS equipment or poor signal, the

ship has too few AIS position points in the route. Such

routes may lack key information and should be

deleted as well. The route identification method with

less AIS information is to calculate the frequency of

the AIS message of the route. When the frequency is

lower than a threshold, it is considered that the route

lacks key information. The interval at which AIS

devices send packets varies from 2 seconds to 3

minutes. This is determined by the ship's heading

status and is highly variable. Therefore, this paper

selects a number of ships with different speeds and

counts the time interval at which AIS equipment

sends messages.

However, in the same waters with the same

navigation environment, the time interval for the AIS

device to send packets should converge in a range.

Assuming that Xt is a specimen for the time interval

of sending packets in a certain area, according to

Chebyshev's theorem, the time interval for sending

packets can be determined as:



0, 5







(6)

where:

𝑥

𝑡

is the average value;

𝐷𝑋

𝑡

is the standard

deviation. Then, the minimum frequency of messages

sent by the AIS device every hour is:

3600





(7)

When the number of AIS packets in the route is

less than such threshold, it indicates that the route

may lack key location information and needs to be

deleted. It should be noted that the frequency of the

AIS message calculated above refers to the dynamic

data packet including the location information, and

does not include other AIS messages such as static

data and voyage data.

3.2 Determine the turning point of the ship

Turning point refers to the point at which the ship

turns. The main method of route design is to

determine the turning point of the ship, and the route

between the turning points constitutes the route. After

obtaining the historical route, the optimized route is

generated by selecting the turning point of the ship

therein. Due to the large amount of data, the route

generated by using all the steering points of the ship

is too complicated. The calculation amount is huge.

The local optimal solution is easily generated, which

is not in accordance with the actual situation. There-

fore, the steering points need to be clustered before

the route is generated.

In the database, the ship's historical route can be

selected according to special requirements, and then

their turning points are clustered. In this way, routes

with different characteristics can be designed. For

example, designing routes with different headings or

routes in different seasons.

3.2.1 Clustering method

Density Based Spatial Clustering of Applications

with Noise (DBSCAN) is a spatial clustering

algorithm that is widely used in many applications. It

is capable of finding arbitrarily shaped clusters in the

presence of noise data and performs well without the

prior knowledge about the number of clusters. The

characteristics of DBSCAN determine its suitability

for clustering spatial data. Therefore, the DBSCAN

algorithm is selected to cluster the turning points.

The clustering effect of the DBSCAN algorithm is

related to the Eps radius (Reps) and the neighborhood

density threshold (MinPts). Start at any point

(unvisited) and find all nearby points within the

distance of Reps. If the number of nearby points is not

less than MinPts, the current point forms a cluster

with its nearby points. The starting point is marked as

visited. If the number of nearby points is less than

MinPts, the point is marked as a noise point. When no

new points are added to any cluster, the clustering

terminates.

In Figure 3, when MinPts is 3, a, b, c, d are core

points, and e, f are noise points.

Figure 3. The schemes of DBSCAN cluster algorithm

3.2.2 Algorithm parameters

When clustering the turning points, three

parameters need to be considered: steering angle,

Reps and MinPts in the DBSCAN algorithm. The

steering angle and Reps are determined based on

historical route conditions. MinPts is determined

based on the generated turning point density. When

the angle of the historical route changes small, the

large ship steering angle and small Reps should be

selected. When the historical route angle changes

greatly, the small ship steering angle and the large

Reps should be selected.

It should be noted that the historical routes

generated in the previous section are connected by

intermittent points, so the steering angle of each point

cannot be obtained directly. The steering angle is

calculated by the heading difference between the

position point in the AIS data and the previous point

position.

After turning to point clustering, a cluster

containing multiple turning points is needed. The

cluster needs to be converted into points with certain

coordinates. The average longitude and average

latitude of the turning points in the cluster should be

calculated to determine the turning points

represented by the cluster:

La La





(8)

Lo Lo





(9)

3.2.3 Clustering effect adjustment

In the clustering process described above, the

clustering parameters selected for the entire route are

consistent. However, the angle changes in various

parts of the route may be inconsistent, or even vary

greatly. Using the same parameters may cause the

turning points in some areas to be too sparse, while

the turning points in some areas are too dense. So for

the area where the steering point density is small, the

steering angle should be reduced. In the area where

the turning point is dense, the steering angle should

increase.

3.3 Route generation

After obtaining the turning point of the ship, treating

it as a node, the Dijkstra algorithm and the ant colony

algorithm are used to generate the route.

When there are too many data points, the

traditional ant colony algorithm has the

disadvantages of weak global search ability, slow

convergence speed and low search efficiency. It is

easy to generate local optimal solutions. If only the

Dijkstra algorithm is used, the generated route will

always be close to the outside and turn frequently,

which is not in line with the actual situation.

Therefore, this paper first uses the Dijkstra algorithm

to plan the initial path and then uses the ant colony

algorithm to optimize the path.

3.3.1 Initial route generation

The Dijkstra algorithm is essentially a traversal

algorithm that calculates the shortest path by

calculating the length of many paths between two

points (Dijkstra 1959). The Dijkstra algorithm starts

from the starting point and selects the nearest node

each time, until it reaches the end point.

When using the Dijkstra algorithm, the problem in

the Figure-4 will appear. Due to no other restrictions,

the route generated by the Dijkstra algorithm may

pass through obstacles or beyond the deep water

channel. When calculating the distance between two

points, set the path length through the obstacle or

beyond the deep water channel to infinity. In this way

it is ensured that the designed route does not pass

through obstacles or outside the deep water channel.

Figure 4. Unsafe path

The steps of Dijkstra algorithm are as follows:

Step 1: Create a point set S= {v

0}, U = {v0, v1, …, vn}. <

i, vj> is recorded as the distance between any two

points. When the two points are crossing the obstacle

or exceed the deep water channel, < v

i, vj> is recorded

as infinity.

Step 2: Select v

k from U, so that <v0, vk> = min{<vo, vi>},

put v

i into S.

Step 3: Add v

k to U and let <v0, vk>= min{<vo, vi>, <vo,

k>+<vk, vi>}.

Step 4: Repeat Step 2 and 3 until U is an empty set.

3.3.2 Route optimization

The ant colony algorithm regards the walking path

of the ant as a feasible solution to the problem to be

optimized. All the paths of the entire ant group

constitute the solution space of the problem to be

optimized. The ants with shorter paths release more

pheromone. As time progresses, the concentration of

pheromone accumulated on the shorter path gradual

gradually increases, and the number of ants that select

the path increases. In the end, the whole ant will

concentrate on the best path under the action of

positive feedback, and the corresponding solution is

the optimal solution to be optimized.

When selecting a path, the ant needs to be

determined according to the transition probability p

ij.

The state transition probability can be calculated by

the following formula:

 

ij ij

j allowed

others



































(10)

where: allowed

k = {0, 1, …, n-1} represents the set of

nodes allowed in the next step; τ

ij(t) represents the

pheromone left by ant k on path (i, j) at time t; η

represents the distance heuristic function factor,

usually expressed by the reciprocal of the distance

between nodes; α is the relative importance coefficient

of the ant pheromone trajectory; β is the relative

importance coefficient of the heuristic function.

The traditional ant colony algorithm pheromone

update method is a simple accumulation. In order to

avoid the local optimal problem and speed up the

algorithm efficiency, this paper uses a new

pheromone update method.

Whenever an ant passes a path, the pheromone of

the path is updated according to the following

formula, which is a local pheromone update strategy.

 

ij ij



  (11)

where τ

0 is the pheromone under the initial condition;

ρ represents the pheromone volatilization coefficient,

and the range of the value is [0, 1].

When all the ants complete a hit path search, the

shortest path in the iteration is selected and the

pheromone on the path is updated. This is the global

update strategy for pheromone. The update method is

as follows:

   

11 ,1

ij ij ij

tttt



   (12)

where: Δτ

ij is the pheromone increment left by the ant

on the path (i, j) in the global update; Δτ

ij(t, t+1) is the

reciprocal of the optimal path length for this iteration.

By setting the local update strategy and the global

update strategy of the pheromone, the quality of the

optimal solution and the search efficiency of the

algorithm can be effectively improved, but the

algorithm will also prematurely converge. To avoid

this problem, the upper and lower bounds of the

pheromone threshold are set as:

min min

min max

max max

ij ij ij



 



















(13)

where: τ

max and τmin represent the maximum and

minimum values of pheromone concentration,

respectively.

The optimization procedure for the initial path

using the ant colony algorithm is:

Step 1: Plan the initial path using the Dijkstra

algorithm;

Step 2: The ant colony algorithm starts searching, and

the ant selects the next node j according to the current

node position i;

Step 3: After the next node j is determined, the ant

needs to locally update the pheromone on the path (i,

j) that has just passed until the ant reaches the

endpoint;

Step 4: Count the optimal path searched by the

current m ants, select the one with the shortest length,

update the global pheromone, and repeat step 2 to

iterate;

Step 5: The number of iterations satisfies the out-put

result after iter≤NC.

4 SIMULATION AND RESULTS

In this section, the route is generated in the Three

Gorges Dam area of China for simulation and com-

pared with the actual trajectory of a ship. The data

used in this section is the ship AIS data for the Three

Gorges Dam area from Dec. 18 2018 to Jan. 18 2019,

collected by nearby AIS base stations.

4.1 Simulations

In this paper, the route of the ship with a length of 85

meters to 120 meters and from the downstream to the

upstream is selected to plan the route.

Figure 5. Outliers of AIS data

After extracting the ships that meet the

requirements in the database, they process their AIS

data, delete the abnormal data and complete the data

by linear interpolation.

Then it is necessary to select the historical route.

Short routes can be deleted directly in the database.

For the routes lacking key information, the minimum

frequency of messages sent by the AIS device every

hour needs to be calculated. After statistics, the time

interval for sending AIS messages in the Three Gorges

Dam area is (2s, 38.04s). Calculate the minimum

transmission frequency of AIS messages according to

the formula (7) to delete the routes with low

frequency.

Figure 6. Sample of the time interval

After obtaining the historical route, the ship

turning points are clustered. The most important

influence on the clustering effect is the steering angle.

After deleting all the turning points with a steering

angle of 0°, the average steering angle of the ship in

the Three Gorges Dam area is calculated to be 4.2°. It

can be seen that the deflection angle of the ship route

in the Three Gorges Dam area is small. The steering

angle selected in this paper is 4°. The selected steering

angle is appropriately changed in some areas to

obtain better clustering effect.

Figure 7. Partial area clustering effect

Finally, the route is generated according to the

Dijkstra algorithm and the ant colony algorithm, the

red line in the Figure 8 is the generated route. Among

them, the red line is the ship route from west to east

(from upstream to downstream), and the blue line is

the ship route from east to west (from down- stream

to upstream). This is achieved by selecting different

historical ship trajectories. Due to the Three Gorges

ship lift equipment, there are large ship anchorages

and port facilities in front of the Three Gorges Dam

area, where the ships are anchored and waiting to

pass the ship lift. No ship can pass the ship directly,

they all need to line up at the anchorage or port.

Therefore, it is meaningless to generate routes in this

area. This is also the reason why the generated route

is incomplete.

4.2 Results and discussion

In this section, two different course routes are

planned by selecting different historical ship

trajectories, which is presented in Figure 8. It can be

seen from the figure that there are obvious differences

be- tween the two routes. The route from west to east

is close to the south bank, and the route from east to

west is close to the north bank, which is in line with

the actual situation. This indicates that the proposed

method can plan unique routes for ships of different

types and different headings by selecting different

types of historical ship trajectories.

At present, there is little research on the route de-

sign of the Three Gorges Dam area. Therefore, this

paper selects the navigation trajectory of the actual

ship and compares it with the generated route.

The black line in the Figure 9 is the real ship

trajectory, and the red line is the generated route.

After calculation, the length of the generated route is

slightly smaller than the actual ship's trajectory.

Obviously, the generated route has a small number of

turns and a small amplitude. By comparing with the

deep-water channel map, the generated routes are all

in the deep water channel and remain basically in the

middle position. Therefore, it can be considered that

the generated route is safer than the actual ship

trajectory, which also indicates that the existing

trajectories of some ships have space of improvement.

Figure 8. Route planning results

Figure 9. Compared with the actual route

5 CONCLUSION

This paper proposes a method for route planning

based on historical AIS data. This method obtains

historical route information by analyzing AIS data,

and seeks optimal path implementation using Dijkstra

algorithm and ant colony algorithm. The results show

that although this method cannot significantly reduce

the length of the route, it can reduce the complexity

and improve safety of the route. Compared with other

route design methods, this method does not need to

model obstacles, and can plan different routes for

different requirements, such as designing routes of

different types of ships, designing dry seasons, flat

water periods, and wet season routes. The method

proposed in this paper has some limitations. First of

all, the processing of AIS data is not accurate enough.

Secondly, the method selected in this paper is simple

in the way of whether the route crosses obstacles and

exceeds the deep water channel. In a complicated

environment, some routes may cross obstacles or be

in shallow waters. Nevertheless, the proposed

method is feasible in most waters and can be used as

a design reference for route planning.

ACKNOWLEDGEMENTS

This research was supported by National Key Technologies

Research & Development Program (2017YFC0804900,

2017YFC0804904) and International Cooperation Technical

Innovation Project of Hubei province (2018AHB003).

REFERENCES

Dijkstra, E.W. 1959. A note on two problems in connexion

with graphs. Numerische Mathematik 1: 269-271.

Choi, M., Chung, H., Yamaguchi, H. & Nagakawa, K. 2015.

Arctic sea route path planning based on an uncertain ice

prediction model. Cold Regions Science and Technology

109: 61-69.

Kobayashi, E., Asajima, T. & Sueyoshi, N. 2011. Advanced

Navigation Route Optimization for an Oceangoing

Vessel. Transnav, the International Journal on Marine

Navigation & Safety of Sea Transportation 5(3): 377-383.

Hänninen, M. & Kujala, P. 2014. Bayesian network

modeling of Port State Control inspection findings and

ship accident involvement. Expert Systems with

Applications 41: 1632- 1646.

Lee, S.M., Roh, M., Kim, K.S., Jung, H. & Park, J. J. 2018.

Method for a simultaneous determination of the path

and the speed for ship route planning problems. Ocean

Engineering 157: 301-312.

Mazaheri, A., Montewka, J. & Kujala, P. 2014. Modeling the

risk of ship grounding - a literature review from a risk

management perspective. WMU Journal of Maritime

Affairs 13: 269-297.

Roh, M. 2013. Determination of an economical shipping

route considering the effects of sea state for lower fuel

consumption. International Journal of Naval Architecture

and Ocean Engineering 5: 246-262.

Sang, L.Z., Wall, A., Mao, Z., Yan, X. & Wang, J. 2015. A

novel method for restoring the trajectory of the inland

waterway ship by using AIS data. Ocean Engineering 110:

183-194.

Tetreault, B.J. 2005. Use of the Automatic Identification

System (AIS) for maritime domain awareness (MDA).

Oceans. IEEE.

Vettor, R. & Guedes Soares, C. 2016. Development of a ship

weather routing system. Ocean Engineering 123: 1-14.

Wang, S., Gao, S. & Yang, W. 2017. Ship route extraction

and clustering analysis based on automatic identification

system data. 2017 Eighth International Conference on

Intelligent Control and Information Processing (ICICIP).

Wang, Y., Zhang, J., Chen, X., Chu, X., & Yan, X. (2013). A

spatial–temporal forensic analysis for inland–water ship

collisions using AIS data. Safety Science, 57(Complete),

187-202.

Wang, Y., Zhang, J., Chen, X., Chu, X. & Yan, X. 2013. A

spatial–temporal forensic analysis for inland–water ship

collisions using AIS data. Safety Science 57: 187-202.

Xiao, F., Ligteringen, H., Van Gulijk, C. & Ale, B. 2015.

Comparison study on AIS data of ship traffic behavior.

Ocean Engineering 95: 84-93.

Zhang, L., Meng, Q., Xiao, Z. & Fu, X. 2018. A novel ship

trajectory reconstruction approach using AIS data. Ocean

Engineering 159: 165-174.

Zhang, W., Goerlandt, F., Montewka J. & Kujala, P. 2015. A

method for detecting possible near miss ship collisions

from AIS data. Ocean Engineering 107: 60-69.