Research on Navigation Risk Assessment of Unmanned Ship Under Complex Navigation Conditions

Abstract

With the research on unmanned ships conducted by the International Maritime Organization, the European Union and China, the navigation risk of unmanned ships has become a hot topic around the world. Based on the research and development history and current situation of unmanned ships at home and abroad, focusing on the three key elements of “Unmanned Ship, shore-based control personnel, external navigation environment”, this study establishes a navigation risk index system for unmanned ships under complex navigation conditions, considering the particularity of the unmanned ship, its perception ability, environmental understanding, situation judgment, communication ability and other indicators. The analytic hierarchy process is used to solve the weight of all kinds of navigation risk factors, and a consistency test of the established index system is carried out. Through expert investigation, the fuzzy membership set of risk indexes is established, and a fuzzy comprehensive evaluation model is established to evaluate the navigation risk of unmanned ships under complex navigation conditions. To avoid the situation of single-factor information being inundated, this study adopts the method of fuzzy comprehensive evaluation from a low level to a high level, and it verifies the algorithm through cases, which proves the validity and rationality of the proposed risk assessment method.

1. Introduction

At present, the research of unmanned ships is a hotspot around the world. Unmanned ships are taken as a strategic development direction in relevant research institutions in different regions or countries, such as the European Union, the United Kingdom, Norway, Finland, Japan, South Korea, etc., who carried out the development of unmanned ships, classified unmanned ships, formulated corresponding guidelines, and established maritime test sites. The International Maritime Organization defined unmanned ships as maritime Autonomous Surface ships (MASSs), which began to establish the laws and regulations for unmanned ships and created a preliminary classification of unmanned ships.

In recent years, China has been actively promoting the research of unmanned ships. In May 2015, the “Made in China 2025” was released. It aims to accelerate the development of marine engineering equipment and high-tech ships, which is one of the ten key development areas. In March 2016, the China Classification Society issued the “Code for Intelligent Ships” [1], which clarified the technical development direction of future intelligent ships. In July 2017, the State Council issued the New Generation of Artificial Intelligence Development Plan [2]. The plan proposed that common technologies, such as the computing architecture of autonomous unmanned systems, perception and understanding of complex dynamic scenes, real-time accurate positioning, and adaptive intelligent navigation for complex environments, should be emphasized. In addition, intelligent technologies, such as autonomous control of drones and automatic driving of cars, ships and rail transit, can support the application of unmanned systems and industrial development.

The risk of autonomous ships under sail is an inevitable problem during operation, especially in complex navigation conditions, such as narrow waterways, waters with high traffic flow density, wind, waves, and bad natural conditions. Thus, the safe state of unmanned ships is of great concern for researchers, managers, ship owners, and insurance companies. The combined effect of various risk factors under complex navigation conditions may cause incalculable impacts and losses to unmanned ships, cargoes, and the marine environment. Therefore, in order to ensure the safe navigation of unmanned ships and avoid accidents, it is urgent to analyze and study the risks of unmanned ships under complex navigation conditions.

For ship-related risk assessment, recent research has focused on comprehensive safety assessments of container ships [3,4], risk assessment of ship navigation safety [5,6], risk assessment of a ship capsizing [7], risk assessment of ship oil spills [8], etc. Jiang Dan [9] from Wuhan University of Technology takes the complex weather conditions of the Three Gorges Reservoir as a background and analyzes the influencing factors of ship navigation safety. The index system is established, and the early warning and assessment of ship navigation risk are realized by using chromatography analysis and fuzzy comprehensive evaluation. There are also some typical risk warning results, such as the collision risk assessment of ships sailing in the Strait of Malacca based on Automatic Identification System (AIS) data [10,11], navigation risk prevention and control in crowded waters near shore [12], ship collision risk assessment based on traffic flow and ship field [8], and comparative study of various collision risk assessment methods [13,14]. In the field of risk assessment, the risk analysis system model of PAWSA (Port and Waterway Safety Assessment), which is widely used at present, was established and developed by the United States Coast Guard. PAWSA consists of four parts: risk source identification, risk event estimation, risk control methods, and risk control implementation measures. In addition, Zhen et al. [15] propose a novel collision risk assessment method for quantifying the risk of certain regions, which considers the aggregation density of ships. Li and Zhang [16] propose a collision avoidance decision method for unmanned surface vehicles, which is based on the improved velocity obstacle algorithm. Huang et al. [17] established a novel regional collision risk assessment framework for multi-ship encounters in order to identify the collision hotspots in complex waters.

Unmanned vessels present unique risk factors, such as the completeness of ship perception systems, the effectiveness of ship-to-shore communication, and the situation of shore-based personnel. These limitations pose challenges in adopting existing risk assessment methods or technologies designed for traditional ships. Meanwhile, the increasing complexity of navigational environments introduced by unmanned ships, particularly in congested ports or narrow waterways characterized by high traffic complexity and limited navigation capabilities, has heightened the challenges associated with risk assessment in intelligent ship navigation. This study refers to the traditional navigational risk assessment method of manned ships and fully considers the particularity of unmanned ships under complex navigational conditions, which not only considers the channel environment and the influencing factors of the natural environment but also adds the ship’s own influencing factors and human control factors into the entire evaluation system. In addition, this study determines the entire navigational safety assessment index system in a more comprehensive way. The research results will play a good role in promoting the development and application of unmanned ships.

2. Risk Assessment Index System for Unmanned Ships

2.1. Analysis of Navigation Risk Assessment Index of Unmanned Ships

The International Maritime Organization (IMO) classifies unmanned ships into four classes [18], termed as ‘Degree of Autonomy’: Class L1—ships with automated procedure and operation and decision-making assistance; Class L2—ships with remote control function and a crew on board; L3 Class—ships with remote control function but no crew on board; Class L4 ships that are completely autonomous in navigation. For Class L1, the primary risks are associated with automation and decision-making processes. For Class L2, the main risks arise from perception and communication aspects, encompassing the adequacy of ship perception systems and the efficacy of ship-to-shore communication channels. For Class L3, in addition to the aforementioned risks at Class L2, there are also concerns regarding onboard equipment reliability. For Class L4, navigation risks primarily stem from both the completeness and reliability of autonomous navigation systems themselves. Due to varying approaches and degrees of autonomy, different levels of unmanned ships encounter distinct navigation risks. Consequently, an assessment should be conducted on their specific navigation risks to ensure navigational safety for ships with differing levels of autonomy.

2.1.1. Factors Affecting Navigation Safety of Unmanned Ships

The safe navigation of unmanned ships is mainly affected by the following four factors: shore-based control personnel [19,20], ship factors, environmental factors, and management factors [1]. Based on the analysis of factors affecting the navigation safety of unmanned ships under a complex navigation environment, the navigation risk factor set of unmanned ships is sorted out, as shown in Figure 1, which is mainly divided into four aspects: shore-based control personnel, ship, environment, and management.

Figure 1. Set of risk factors for navigation of unmanned ships.

In order to more accurately calculate and evaluate the navigation risks of unmanned ships under complex navigation conditions, this study adopts the expert questionnaire method [21] to screen, simplify, and synthesize the indexes of the navigation risk factor set of unmanned ships mentioned above. The expert questionnaire method is used because it is an efficient method to collect professional and representative evaluation information with high flexibility. Through expert questionnaires, this study identifies factors with the same system or influencing characteristics among various risk sources and simplifies and unifies them. Factors which are unreasonable or not applicable to unmanned ships shall be removed or replaced according to expert opinions. For example, navigation notices, navigation warnings, etc., are classified into the navigation aid conditions, and all management factors are deleted.

2.1.2. Index Screening of Navigation Risk of Unmanned Ships

According to the investigation results of the expert opinions on navigation risk indexes of unmanned ships in Section 2.1.1, navigation risk assessment indexes of unmanned ships under complex navigation conditions can be divided into 3 first-level indexes, 8 second-level indexes, and 18 third-level indexes. The details are as follows:

The first-level indexes: shore-based control personnel, ship factors, environmental factors.

The second-level indexes: professional skills, education and training, situational awareness, ship itself, perception and understanding, maritime communication technology, hydrometeorology, navigation environment.

The third-level indexes: ship size, ship maneuverability, ship age, ship loading condition, target recognition ability, navigation environment understanding, situation judgment, semantic understanding, communication bandwidth, communication type, communication reliability, wind level, rainfall, surge wave, visibility, traffic flow, obstruction, navigation aid conditions.

2.2. Assessment Index Weight Calculation

Based on the research content and summary experience of the MUNIN project [22] and the actual situation of relevant domestic unmanned ship R&D and operation companies, this study investigates and consulates corresponding cooperative institutions and colleges. The analytic hierarchy process (AHP) is used to solve the weight of all kinds of navigation risk factors in this study. The AHP can decompose complex problems into multiple levels, criteria and sub-criteria, establish a judgment matrix, and finally obtain the optimal decision scheme through the comparison and weight calculation of each level. The steps of AHP are listed as follows:

• Divide complex indexing system into multiple hierarchical levels.
• Based on expert judgment, express the relative importance of factors at higher levels according to the arrangement of factors at each level. This step is conducted by constructing the discriminant matrix.
• Then, determine the order weights of each factor’s relative importance within each level based on the maximum eigenvalue and its eigenvector derived from this matrix.
• Finally, by analyzing each level, obtain the overall ranking weight for the entire system.

The discriminant matrix is established through an expert questionnaire survey, and then the importance of the preliminary index is calculated. This section mainly carries out a consistency test and index weight calculation for the navigation risk assessment index system of unmanned ships. The values in the discriminant matrix represent the relative importance ratios between two factors and the higher-level objective factor. Typically, these ratios are assigned on a scale ranging from 1 to 9, with their reciprocals ranging from 1 to 1/9.

(1) Establish and analyze the navigation risk hierarchy diagram of unmanned ships under complex navigation conditions.

(2) Based on expert evaluation, the importance of the indexes is inferred through the mutual discriminant matrix of the first-level indexes, and the discriminant matrix of the first-level indexes is obtained, as shown in Table 1:

Table 1. Discriminant matrix of the first-level indexes.

First-Level Indexes	Shore-Based Control Personnel A1	Ship Factors A2	Environmental Factors A3
Shore-based control personnel $A_1$	1	6/8	8/4
Ship factors $A_2$	8/6	1	6/4
Environmental factors $A_3$	4/8	4/6	1

Specifically:

$$P=\left[\begin{array}{ccc}1& 6/8& 8/4\\ 8/6& 1& 6/4\\ 4/8& 4/6& 1\end{array}\right]$$

Similarly, the factors of each level and their corresponding sub-index factors are scored by experts, and the discriminant matrix of each level is obtained.

(3) For the calculation of the weight vector [23,24] of the first-level indexes, it is necessary to normalize and geometrically average the sequence of different primary index vectors. The geometrical average is less affected by extremes than the arithmetic mean. The corresponding calculation formula is shown in Equation (1):

$$T{R}_i={(\underset{j=1}{\overset{3}{\Pi }}{a}_{ij})}^{\raisebox{1ex}{$1$}\!\left/ \!\raisebox{-1ex}{$3$}\right.}$$

(1)

$$\begin{array}{c}T{R}_1={[1\cdot (6/8)\cdot (8/4)]}^{1/3}=1.145\\ T{R}_2={[(8/6)\cdot 1\cdot (6/4)]}^{1/3}=1.260\\ T{R}_3={[(4/8)\cdot (4/6)\cdot 1]}^{1/3}=0.693\end{array}$$

where $TR$ refers to the geometrically averaged row vectors.

The normalization calculation of the ranking index vector $\omega$ is shown in Equation (2):

$${\omega }_1=T{R}_1/{\displaystyle \sum _{i=1}^{3}T{R}_i}$$

(2)

$${\displaystyle \sum _{i=1}^{3}T{R}_i=T{R}_1+T{R}_2+T{R}_3=1.145+1.260+0.693=3.098}$$

$${\omega }_1=T{R}_1/{\displaystyle \sum _{i=1}^{3}T{R}_i=0.3695}$$

(3)

Similarly, the following results can be obtained: $w_2=0.4067$, $w_3=0.2238$

(4) Consistency test

Given the intricate nature of unmanned ship risk systems and the limited comprehension of individuals, experts’ evaluations sometimes result in an inconsistent discriminant matrix that fails to meet satisfaction, thereby leading to contradictions in expressing factor importance. Consequently, it becomes imperative to conduct a consistency test for the judgment matrix. Otherwise, the weights calculated would be unreasonable to some extent.

(1) We need to calculate the maximum eigenvalue of the discriminant matrix and then carry out the consistency test of the discriminant matrix according to the corresponding calculation formula. See Equation (4) for the calculation formula of the maximum eigenvalue ${\lambda }_{\max }$:

$${\lambda }_{\max }=\sum _{i=1}^3\frac{{(A\omega )}_i}{3{\omega }_i}=\frac{1}{3}\sum _{i=1}^3\frac{{\displaystyle \sum _{j=1}^3{a}_{ij}{\omega }_{ij}}}{{\omega }_i}$$

(4)

$$\begin{array}{c}{A}_{\omega }=\left(\begin{array}{ccc}1& 6/8& 8/4\\ 8/6& 1& 6/4\\ 4/8& 4/6& 1\end{array}\right)\left(\begin{array}{c}0.370\\ 0.406\\ 0.224\end{array}\right)=\left(\begin{array}{c}1.1225\\ 1.2353\\ 0.6797\end{array}\right)\\ {\lambda }_{\max }={\displaystyle \frac{1}{3}}[{\displaystyle \frac{1.1225}{0.3700}}+{\displaystyle \frac{1.2353}{0.4060}}+{\displaystyle \frac{0.6797}{0.2240}}]=3.0369\end{array}$$

(2) Calculation of CI and CR:

$$CI={\displaystyle \frac{{\lambda }_{\max }-n}{n-1}}=0.0184$$

(5)

$$CR={\displaystyle \frac{CI}{RI}}=0.0355<0.1$$

(6)

where CI refers to the consistency index, RI refers to the average random consistency index of the same order, and CR refers to the consistency test result. $A_{\omega }$ is the judgment matrix.

The CR value, which is less than or equal to 0.1, meets the corresponding criteria and meets the requirements of the consistency test.

According to the above hierarchical analysis method for obtaining the risk impact factor index weight, combining the navigation characteristics of unmanned ships and referring to the prior knowledge of manned ship risk assessment, the weight value of each assessment index of unmanned ship navigation risk is obtained. The weights of each risk index are shown in Table 2. The weights of each index of the navigation risk assessment of unmanned ships are obtained through calculation. When calculating the weights, CI and CR values are calculated to test the effectiveness of the discriminant matrix of weights. It is found that CR is less than 0.1, so the index system in Table 2 can be used.

Table 2. Weight value of navigation safety evaluation index of unmanned ships.

Destination Layer	First-Level Indicators Ai	Weight	Second-Level Indicators Aij	Weight	Third-Level Indicators Aijk	Weight
Navigation risk assessment index system of unmanned ships	Shore-based control personnel A1	0.3695	Professional skills A11	0.2808
			Education and training A12	0.1350
			Situational awareness A13	0.5842
	Ship factors A2	0.4067	Ship itself A21	0.1952	Ship size A211	0.3178
					Ship maneuverability A212	0.2200
					Ship age A213	0.1837
					Ship loading condition A214	0.2785
			Perception and understanding A22	0.4592	Target recognition ability A221	0.3289
					Navigation environment understanding A222	0.1813
					Situation judgment A223	0.3086
					Semantic understanding A224	0.1813
			Maritime communication technology A23	0.3457	Communication bandwidth A231	0.1571
					Communication type A232	0.2493
					Communication reliability A233	0.5936
	Environmental factors A3	0.2238	Hydrometeorology A31	0.5000	Wind level A311	0.3038
					Rainfall A312	0.1201
					Surge wave A313	0.2190
					Visibility A314	0.3571
			Navigation environment A32	0.5000	Traffic flow A321	0.1667
					Obstruction A322	0.3333
					Navigation aid conditions A323	0.5000

By calculating all the consistency indexes and the average random consistency indexes of the same order, the total CI and RI can be obtained, and the consistency test of the above table can be realized. Through the above consistency test calculation and analysis, where the value of CR meets the requirement, it is concluded that the navigation risk assessment index system of unmanned ships established in this section meets the consistency test requirements. From the perspective of weight, factors, such as situational awareness, target recognition ability, communication reliability, and navigation aid conditions of shore-based control personnel, have a great impact on safe navigation under complex navigation conditions for unmanned ships.

3. Navigation Risk Assessment of Unmanned Ships Based on Fuzzy Comprehensive Evaluation

Based on the index system of the navigation risk assessment and risk index weights determined in Section 2, the main work of this section is to evaluate the navigation risk of unmanned ships of highly autonomous level under complex navigation conditions by using the fuzzy comprehensive evaluation method. The fuzzy comprehensive evaluation method is widely used in assessing the navigational risk with multiple influencing factors.

3.1. Risk Assessment of Unmanned Ships Under Complex Navigation Conditions

This study mainly evaluates the navigation risk of unmanned ships under complex navigation conditions based on the theory of fuzzy comprehensive evaluation [25]. The navigation risk assessment model of unmanned ships is constructed by establishing a risk factor set, risk assessment set, and fuzzy subset table of the membership degree of indexes at each level. Combined with Section 2, the weights of indexes at each level obtained by the index system are verified using cases.

3.1.1. Establishment of Navigation Risk Factor Set for Unmanned Ships Under Complex Navigation Conditions

According to the analysis of factors affecting the navigation risk of unmanned ships under complex navigation conditions in Section 2, the navigation risk set of unmanned ships under complex navigation conditions is established, as shown in Table 3 below. It should be noticed that the risk factors set for autonomous ships in this study were screened by the experts who have worked on autonomous ships or have specialized knowledge in the research of autonomous ships through a questionnaire.

Table 3. A set of risk factors for unmanned ship navigation under complex navigation conditions.

The first-level indicators
$A^{(1)}$	$\left\{A_1^{\left(1\right)}{A}_2^{\left(1\right)} A_3^{\left(1\right)}\right\}=\left\{\mathrm{shore}-\mathrm{based}\mathrm{control}\mathrm{personnel},\mathrm{ship}\mathrm{factors},\mathrm{environmental}\mathrm{factors}\right\}$
The second-level indicators
$A_1^{(2)}$	$\left\{A_{11}^{\left(2\right)}{A}_{12}^{\left(2\right)} A_{13}^{\left(2\right)}\right\}=\left\{\mathrm{professional}\mathrm{skills},\mathrm{education}\mathrm{and}\mathrm{training},\mathrm{situational}\mathrm{awareness}\right\}$
$A_2^{(2)}$	$\left\{A_{21}^{\left(2\right)}{A}_{22}^{\left(2\right)} A_{23}^{\left(2\right)}\right\}=\left\{\mathrm{ship}\mathrm{itself},\mathrm{perception}\mathrm{and}\mathrm{understanding},\mathrm{maritime}\mathrm{communication}\mathrm{technology}\right\}$
$A_3^{(2)}$	$\left\{A_{31}^{\left(2\right)}{A}_{32}^{\left(2\right)}\right\}=\left\{\mathrm{hydrometeorology},\mathrm{navigation}\mathrm{environment}\right\}$
The third-level indicators
$A_{21}^{(3)}$	$\left\{A_{211}^{\left(3\right)}{A}_{212}^{\left(3\right)} A_{213}^{\left(3\right)} A_{214}^{\left(3\right)}\right\}=\left\{\mathrm{ship}\mathrm{size},\mathrm{ship}\mathrm{maneuverability},\mathrm{ship}\mathrm{age},\mathrm{ship}\mathrm{loading}\mathrm{condition}\right\}$
$A_{22}^{(3)}$	$\left\{A_{221}^{\left(3\right)}{A}_{222}^{\left(3\right)} A_{223}^{\left(3\right)} A_{224}^{\left(3\right)}\right\}=\{\mathrm{target}\mathrm{recognition}\mathrm{ability},\mathrm{navigation}\mathrm{environment}\mathrm{understanding},\mathrm{situation}\mathrm{judgment},\mathrm{semantic}$ $\mathrm{understanding}\}$
$A_{23}^{(3)}$	$\{A_{231}^{\left(3\right)}{A}_{232}^{\left(3\right)} A_{233}^{\left(3\right)}=\left\{\mathrm{communication}\mathrm{bandwidth},\mathrm{communication}\mathrm{type},\mathrm{communication}\mathrm{reliability}\right\}$
$A_{31}^{(3)}$	$\left\{A_{311}^{\left(3\right)}{A}_{312}^{\left(3\right)} A_{313}^{\left(3\right)} A_{314}^{\left(3\right)}\right\}=\left\{\mathrm{wind}\mathrm{level},\mathrm{rainfall},\mathrm{surge}\mathrm{wave},\mathrm{visibility}\right\}$
$A_{32}^{(3)}$	$\left\{A_{321}^{\left(3\right)}{A}_{322}^{\left(3\right)} A_{323}^{\left(3\right)}\right\}=\left\{\mathrm{traffic}\mathrm{flow},\mathrm{obstruction},\mathrm{navigation}\mathrm{aid}\mathrm{conditions}\right\}$

3.1.2. Establishment of Navigation Risk Assessment Set for Unmanned Ships Under Complex Navigation Conditions

In this study, according to the different degree of navigation risk of unmanned ships under complex navigation conditions, the assessment set is designed as follows:

$$V=\left\{\mathrm{Relatively}\mathrm{safe},\mathrm{Criticality},\mathrm{Low}\mathrm{risk},\mathrm{Medium}\mathrm{risk},\mathrm{High}\mathrm{risk}\right\}$$

3.1.3. Construction of a Fuzzy Subset Table for Membership Degree of Indexes at Each Level

Membership Degree of Shore-Based Control Personnel

The personnel factors of shore-based operation mainly include professional skills, education and training, and situational awareness.

(1)

Membership degree of professional skills

According to the International Convention on Standards of Training, Certification and Watchkeeping for Seafarers, International Convention on Standards of Training, Certification and Watchkeeping for Seafarers, STCW, and the description of the Autonomous Cargo Transport Ships Guide and the unmanned ships industry guidelines issued by the United Kingdom, professional skills are classified into three categories: skilled, medium, and poor.

(2)

Membership degree of education and training

The skill requirements of the shore-based control personnel of unmanned ships must be higher than that of the senior pilots of manned ships. According to the existing assessment mechanism of crew training, education and training levels are classified as adequate, average, and insufficient.

(3)

Membership degree of situational awareness

Situational awareness refers to the ability of the crew or the shore-based control personnel to understand and judge the navigation environment where the unmanned ship is trapped and the ability to effectively warn of the navigation risk in the future navigation state. The situational awareness is specifically divided into strong, moderate, and weak.

Membership Degree of Ship Factors

Ship factors mainly include the ship itself, perception and understanding ability, and maritime communication technology.

(1)

Membership degree of ship itself

For the safe navigation of unmanned ships under complex conditions, the ship’s own influence on safe navigation is mainly reflected in four aspects: ship size, maneuverability, ship age, and ship loading.

• Membership degree of ship size
  Since there is no international specification and document for the size limit of unmanned ships, this study temporarily divides the size of ships into large, medium, and small.
• Membership degree of maneuverability
  Ship maneuverability is mainly reflected in the ship following and controlling effect after trajectory planning. Therefore, the maneuverability of unmanned ships is divided into three categories: good, medium, and poor.
• Membership degree of ship age
  In this study, according to the requirements of ship machinery repair and ship inspection and other relevant documents, the ship age membership is divided into more than 15 years, 6 years to 15 years, and less than 6 years.
• Membership degree of ship loading
  At present, the unmanned ships for the autonomous navigation test in some typical sea areas in the world are all container ships, so this study divides the ship loading situation into full load, half load, and no load according to manned container ships.

(2)

Membership degree of perception and understanding ability

The perception and understanding ability of unmanned ships can be divided into four types: target recognition ability, navigation environment understanding, situation judgment, and semantic understanding.

• Membership degree of target recognition ability
  According to the proportion of target recognition, this study divides the target recognition ability into full recognition, majority recognition, and partial recognition.
• Membership degree of navigation environment understanding
  Navigation environment understanding mainly refers to the estimation of navigation situation on the basis of perception, and the judgment of the status of the unmanned ship, including the number of obstacles, the position, size, relative orientation, relative speed, and other information. Therefore, this study divides the membership degree of navigation environment understanding index into good, medium, and poor.
• Membership degree of situation judgment
  The situation judgment mainly refers to the judgement of the encounter state between the unmanned ship and the target ship of an obstacle based on the understanding of the environment, such as the head-on situation, overtaking situation, and cross-encounter situation. In this study, the membership degree of situation judgment is divided into accurate, medium, and unclear.
• Membership degree of semantic understanding
  Semantic understanding is to convert the information of navigational environment understanding into binary attribute relationships among entities, such as unmanned ships, obstacles, charts, and environmental information (e.g., the location attribute and relationship attribute). Therefore, this study divides semantic understanding into three categories: strong, general, and weak.

(3)

Maritime communication technology

Maritime communication technology is one of the key technologies in unmanned ship technology, which affects the quality and efficiency of shore-based support services and the communication between two ships or between a ship and the shore. It is usually manifested in the aspects of communication bandwidth, communication type, and communication reliability, which can influence the transmission effect of the information between the shore and the ship.

• Membership degree of communication bandwidth
  The principle of communication and traffic is similar; each piece of information needs to be effectively transmitted on the communication channel. The narrower the bandwidth, the more likely it is for errors such as information serialization and cross-talk to occur between messages. Therefore, this study divides the membership degree of the influence of communication bandwidth on the safe navigation of unmanned ships into broadband and narrowband.
• Membership degree of communication type
  The more types of communication, the easier it is to traverse a variety of scenarios or ship communication under a variety of complex navigation conditions. Therefore, the types of communication are divided into more and less.
• Membership degree of communication reliability
  In addition to the impact of communication bandwidth and communication types, the reliability of communication is also crucial to the safe navigation of unmanned ships. In this study, the membership degree of communication reliability is divided into strong, medium, and weak.

Membership Degree of Environmental Factors

The environmental factors mainly include hydrometeorology and navigation environment.

(1)

Membership degree of hydrometeorology

For the safe navigation of unmanned ships under complex conditions, the impact of complex weather conditions on unmanned ships is especially prominent, which is reflected in four aspects: wind level, rainfall, surge wave, and visibility.

• Membership degree of wind level
  According to the concept and classification of meteorology, this study mainly divides the subordination degree of wind class into 0–4, 5–8, and 9 and considers the influence of wind class on the navigation safety of unmanned ships.
• Membership degree of rainfall
  Rainfall of significant size can cause poor visibility, free liquid surface influence, and other dynamic hazard sources. Therefore, the membership degree of rainfall is defined as maximum, medium, and small.
• Membership degree of surge wave
  The surge and wind level have similar principles to affect the safe navigation of unmanned ships. Therefore, the membership degree of a wave surge is divided into level 0–3, level 4–5, and above level 6.
• Membership degree of visibility
  According to the perception ability of the onboard instruments and equipment of unmanned ships, the visibility membership is divided into good and above, medium, and poor.

(2)

Membership degree of navigation environment

The navigation environment of unmanned ships is mainly divided into three types: traffic flow, navigation obstruction, and navigation aids.

• Membership degree of traffic flow
  The density of traffic flow has a significant impact on the safe navigation of unmanned ships. In this study, the membership degree of traffic flow is divided into high density, medium density, and low density.
• Membership degree of navigation obstruction
  The safety of unmanned ships in complex navigation conditions depends in part on the presence of navigational obstacles during the voyage. Therefore, this study divides the membership degree of obstructions into existence and non-existence.
• Membership degree of navigation aids
  Compared with manned ships, unmanned ships are more dependent on time-effective navigation service information set services, which are based on the ship’s own needs. Therefore, this study divides the membership degree of navigation aid service into two levels: sufficient and lacking.

According to the membership degree of each level of indexes above, this study issues a total of 60 questionnaires on the navigation risk assessment of unmanned ships under complex navigation conditions based on the consideration of population size, confidence level, margin of error, and population variance from a statistical perspective, 52 of which were valid, while the others were not completed. The respondents of the survey include teachers and students of navigation in maritime colleges and universities, experts in the field of unmanned ship technology research, and staff of ship enterprises and institutions. The questionnaire is conducted and collected on Wechat with the help of the Tencent questionnaire platform. As a result, a fuzzy subset table of the membership degree of each level index is determined.

3.1.4. Establishment of Weight

The weights of indexes at each level obtained in Section 2 can be obtained as follows:

The weights of the third-level indexes:

$$\begin{array}{l}{A}_{21}^{(3)}=\left[0.3178,0.2200,0.1837,0.2785\right]\\ A_{22}^{(3)}=\left[0.3289,0.1813,0.3086,0.1813\right]\\ A_{23}^{(3)}=\left[0.1571,0.2493,0.5936\right]\\ A_{31}^{(3)}=\left[0.3038,0.1201,0.2190,0.3571\right]\\ A_{32}^{(3)}=\left[0.1667,0.3333,0.5000\right]\end{array}$$

The weights of the second-level indicators:

$$\begin{array}{c}{A}_1^{(2)}=\left[0.2810,0.1350,0.5842\right]\\ A_2^{(2)}=\left[0.1952,0.4592,0.3457\right]\\ A_3^{(2)}=\left[0.5000,0.5000\right]\end{array}$$

The weights of the first-level indicators:

$$A^{(1)}=\left[0.3695,0.4067,0.2238\right]$$

3.1.5. Fuzzy Comprehensive Judgement

For the comprehensive assessment of the navigation risk of unmanned ships under complex navigation conditions, a large amount of single-index information will be lost if the first-level fuzzy comprehensive evaluation model is used alone, because the risk sources include many influencing factors. Therefore, this study adopts multi-level fuzzy comprehensive evaluation [26]. The fuzzy comprehensive evaluation is a multi-factor decision-making approach used to systematically assess entities with multiple attributes or influenced by numerous factors. The steps of fuzzy comprehensive evaluation are shown as follows:

• Given a set of evaluations.
• Given a set of factors.
• Each factor in the factor set corresponds to a weight coefficient, and all weight coefficients constitute a weight vector.
• Evaluate each factor in the factor set based on the evaluation set to obtain a single-factor evaluation vector.
• By evaluating all factors, an evaluation matrix can be obtained.
• The final fuzzy evaluation result is obtained by multiplying the weight vector with the evaluation matrix. The multiplying can be conducted by using different operators.
• The final evaluation vector can be obtained through the maximum membership degree method, centroid method, or weighted average method. The maximum membership degree method is used in this article.

The multi-level fuzzy comprehensive evaluation combines the analytical hierarchy process and the fuzzy comprehensive evaluation method to effectively evaluate the overall system risk. The navigation risk assessment factors under complex navigation conditions are classified and evaluated step by step from low to high levels.

(1)

Fuzzy comprehensive evaluation of the third-level indexes

According to the actual sailing situation of unmanned ships under complex sailing conditions, the third-level evaluation matrix can be obtained by using the weight for ship factors and environmental factors obtained from the single-factor evaluation calculated by the discriminant matrix in Section 2, as shown in Table 4. By multiplying the weight and the evaluation matrix, the comprehensive evaluation index set of ship itself, perception and understanding, maritime communication technology, hydrometeorology, and navigation environment can be obtained, respectively.

(2)

Fuzzy comprehensive evaluation of the second-level indexes

According to the actual sailing conditions of unmanned ships under complex sailing conditions, the evaluation matrix of the second level can be obtained by using the weights obtained from the single-factor evaluation calculated by the discriminant matrix in Section 2 and by using the fuzzy comprehensive evaluation results of the third level, as shown in Table 5. By multiplying the weight and the evaluation matrix, the comprehensive evaluation index set of shore-based control personnel, ship factors, and environmental factors can be obtained, respectively.

(3)

Fuzzy comprehensive evaluation of the first-level indexes

According to the actual sailing conditions of unmanned ships under complex sailing conditions, the evaluation matrix of the first level can be obtained by using the weights obtained from the single-factor evaluation calculated by the discriminant matrix in Section 2 and by using the fuzzy comprehensive evaluation results of the second level, as shown in Table 6. By multiplying the weight with the evaluation matrix of this level, the comprehensive evaluation index set of navigation risk assessment of unmanned ships under complex navigation conditions can be obtained. The evaluation index set is normalized, and the comprehensive evaluation result can be obtained according to the principle of the maximum membership degree.

Table 4. Matrix table of the evaluation for the third-level indexes.

Indicators of Evaluation	Judgment Matrix
	Relatively Safe	Criticality	Low Risk	Medium Risk	High Risk
Ship size Ship maneuverability Ship age Ship loading	$R_{21}^{(3)}$
Target recognition ability Navigation environment understanding Situation judgment Semantic understanding	$R_{22}^{(3)}$
Communication bandwidth Communication type Communication reliability	$R_{23}^{(3)}$
Wind level Rainfall Surge wave Visibility	$R_{31}^{(3)}$
Traffic flow Obstruction Navigation AIDS	$R_{32}^{(3)}$

Table 4. Matrix table of the evaluation for the third-level indexes.

Indicators of Evaluation	Judgment Matrix
	Relatively Safe	Criticality	Low Risk	Medium Risk	High Risk
Ship size Ship maneuverability Ship age Ship loading	$R_{21}^{(3)}$
Target recognition ability Navigation environment understanding Situation judgment Semantic understanding	$R_{22}^{(3)}$
Communication bandwidth Communication type Communication reliability	$R_{23}^{(3)}$
Wind level Rainfall Surge wave Visibility	$R_{31}^{(3)}$
Traffic flow Obstruction Navigation AIDS	$R_{32}^{(3)}$

Table 5. Matrix table of the evaluation for the second-level indicators.

Indicators of Evaluation	Judgment Matrix
	Relatively Safe	Criticality	Low Risk	Medium Risk	High Risk
Professional skills Education and training Situational awareness	$R_1^{(2)}=\left[\begin{array}{c}{R}_{11}^{(2)}\\ R_{12}^{(2)}\\ R_{13}^{(2)}\end{array}\right]$
Ship itself Perception and understanding Maritime communication technology	$R_2^{(2)}=\left[\begin{array}{c}{A}_{21}^{(3)}•R_{21}^{(3)}\\ A_{22}^{(3)}•R_{22}^{(3)}\\ A_{23}^{(3)}•R_{23}^{(3)}\end{array}\right]$
Hydrometeorology Navigation environment	$R_3^{(2)}=\left[\begin{array}{c}{A}_{31}^{(3)}•R_{31}^{(3)}\\ A_{32}^{(3)}•R_{32}^{(3)}\end{array}\right]$

Table 5. Matrix table of the evaluation for the second-level indicators.

Indicators of Evaluation	Judgment Matrix
	Relatively Safe	Criticality	Low Risk	Medium Risk	High Risk
Professional skills Education and training Situational awareness	$R_1^{(2)}=\left[\begin{array}{c}{R}_{11}^{(2)}\\ R_{12}^{(2)}\\ R_{13}^{(2)}\end{array}\right]$
Ship itself Perception and understanding Maritime communication technology	$R_2^{(2)}=\left[\begin{array}{c}{A}_{21}^{(3)}•R_{21}^{(3)}\\ A_{22}^{(3)}•R_{22}^{(3)}\\ A_{23}^{(3)}•R_{23}^{(3)}\end{array}\right]$
Hydrometeorology Navigation environment	$R_3^{(2)}=\left[\begin{array}{c}{A}_{31}^{(3)}•R_{31}^{(3)}\\ A_{32}^{(3)}•R_{32}^{(3)}\end{array}\right]$

Table 6. Matrix table of the evaluation for the first-level indicators.

Indicators of Evaluation	Judgment Matrix
	Relatively Safe	Criticality	Low Risk	Medium Risk	High Risk
Shore-based control personnel Ship factors Environmental factors	$R^{(1)}=\left[\begin{array}{c}{A}_1^{(2)}•R_1^{(2)}\\ A_2^{(2)}•R_2^{(2)}\\ A_3^{(2)}•R_3^{(2)}\end{array}\right]$

Table 6. Matrix table of the evaluation for the first-level indicators.

Indicators of Evaluation	Judgment Matrix
	Relatively Safe	Criticality	Low Risk	Medium Risk	High Risk
Shore-based control personnel Ship factors Environmental factors	$R^{(1)}=\left[\begin{array}{c}{A}_1^{(2)}•R_1^{(2)}\\ A_2^{(2)}•R_2^{(2)}\\ A_3^{(2)}•R_3^{(2)}\end{array}\right]$

3.2. Navigation Risk Assessment and Verification of Unmanned Ships Under Complex Navigation Conditions

This section aims to verify the half-load navigation safety of unmanned ships under complex navigation conditions by setting the situations of strong shore-based control personnel, 2-year ship age, good maneuverability, accurate situation judgment, low traffic flow density, and sufficient navigation aid conditions. However, the navigation environment of unmanned ships is at wind level 6, and the visibility is medium. Not all targets can be recognized, the communication bandwidth is narrowband, and the semantic understanding is general. By sending questionnaires and verifying by experts, it is considered that unmanned ships are safe to sail under these conditions. The navigation risk assessment method of unmanned ships under complex navigation conditions constructed in this study is used for risk assessment. The following can be calculated:

The evaluation matrix of the third-level indicators:

$$\begin{array}{c}{R}_{21}^{(3)}=\left[\begin{array}{ccccc}0.2692& 0.2692& 0.2308& 0.2308& 0.0000\\ 0.8462& 0.1154& 0.0385& 0.0000& 0.0000\\ 0.6923& 0.2308& 0.0769& 0.0000& 0.0000\\ 0.3846& 0.3462& 0.2308& 0.0385& 0.0000\end{array}\right]\\ R_{22}^{(3)}=\left[\begin{array}{ccccc}0.2692& 0.1923& 0.5000& 0.0385& 0.0000\\ 0.8846& 0.0769& 0.0385& 0.0000& 0.0000\\ 0.8846& 0.0769& 0.0385& 0.0000& 0.0000\\ 0.1923& 0.3462& 0.2308& 0.1923& 0.0385\end{array}\right]\\ R_{23}^{(3)}=\left[\begin{array}{ccccc}0.0769& 0.1538& 0.0769& 0.4231& 0.2692\\ 0.7692& 0.1154& 0.1154& 0.0000& 0.0000\\ 0.8462& 0.1154& 0.0385& 0.0000& 0.0000\end{array}\right]\\ R_{31}^{(3)}=\left[\begin{array}{ccccc}0.1154& 0.3846& 0.3846& 0.0769& 0.0385\\ 0.2308& 0.3846& 0.2692& 0.1154& 0.0000\\ 0.1538& 0.5385& 0.2692& 0.0385& 0.0000\\ 0.2308& 0.5000& 0.2692& 0.0000& 0.0000\end{array}\right]\\ R_{32}^{(3)}=\left[\begin{array}{ccccc}0.6923& 0.1923& 0.0769& 0.0000& 0.0385\\ 0.8077& 0.1154& 0.0769& 0.0000& 0.0000\\ 0.7308& 0.1923& 0.0769& 0.0000& 0.0000\end{array}\right]\end{array}$$

The evaluation matrix of the second-level indicators:

$$\begin{array}{c}{R}_1^{(2)}=\left[\begin{array}{c}{R}_{11}^{(2)}\\ R_{12}^{(2)}\\ R_{13}^{(2)}\end{array}\right]=\left[\begin{array}{ccccc}0.2692& 0.3846& 0.3462& 0.0000& 0.0000\\ 0.8077& 0.1154& 0.0769& 0.0000& 0.0000\\ 0.7308& 0.2308& 0.0385& 0.0000& 0.0000\end{array}\right]\hfill \\ R_2^{(2)}=\left[\begin{array}{c}{A}_{21}^{(3)}•R_{21}^{(3)}\\ A_{22}^{(3)}•R_{22}^{(3)}\\ A_{23}^{(3)}•R_{23}^{(3)}\end{array}\right]=\left[\begin{array}{ccccc}0.5060& 0.2498& 0.1602& 0.0841& 0.0000\\ 0.5568& 0.1637& 0.2252& 0.0475& 0.0070\\ 0.7061& 0.1214& 0.0637& 0.0665& 0.0423\end{array}\right]\hfill \\ R_3^{(2)}=\left[\begin{array}{c}{A}_{31}^{(3)}•R_{31}^{(3)}\\ A_{32}^{(3)}•R_{32}^{(3)}\end{array}\right]=\left[\begin{array}{ccccc}0.1789& 0.4595& 0.3043& 0.0457& 0.0117\\ 0.7500& 0.1667& 0.0769& 0.0000& 0.0064\end{array}\right]\hfill \end{array}$$

The evaluation matrix of the first-level indicators:

$$R^{(1)}=\left[\begin{array}{c}{A}_1^{(2)}•R_1^{(2)}\\ A_2^{(2)}•R_2^{(2)}\\ A_3^{(2)}•R_3^{(2)}\end{array}\right]=\left[\begin{array}{ccccc}0.6116& 0.2584& 0.1301& 0.0000& 0.0000\\ 0.5986& 0.1659& 0.1567& 0.0612& 0.0178\\ 0.4645& 0.3131& 0.1906& 0.0228& 0.0091\end{array}\right]$$

The navigation risk assessment results of unmanned ships in complex navigation environment are shown as follows:

$$A^{(1)}•R^{(1)}=\left[\begin{array}{ccccc}0.5734& 0.2330& 0.1545& 0.0300& 0.0093\end{array}\right]$$

According to the results of the above fuzzy comprehensive evaluation algorithm, the maximum value of risk assessment is 0.5734, and the corresponding risk assessment is “relatively safe”. Therefore, under the designed complex navigation conditions, the navigation risk assessment of the unmanned ship is relatively safe. We also conducted a comparative evaluation on the navigation risks of conventional manned ships under identical ship and environmental conditions. The validation specifically focused on a semi-loaded navigation safety case for manned ships operating in complex navigation conditions, with a ship age of 2 years, well-educated and -trained personnel onboard, excellent maneuverability of the ship, low traffic density, and sufficient navigational aids. The ship operated in an environment with wind force level 6 and moderate visibility, where not all targets could be identified. The communication bandwidth is limited while maintaining average semantic understanding. However, there exists inadequate situational awareness among personnel regarding encounter situations onboard. Through the distribution of questionnaires and expert verification, it has been determined that manned ships under these conditions are classified as having “high-risk” navigation. Subsequently, utilizing the risk assessment method developed in this article for unmanned ships under similar circumstances, an evaluation of the navigation risks for manned ships will be conducted. The results are shown as follows:

$$A^{\left(1\right)}•R^{\left(1\right)}=\left[\begin{array}{ccccc}0.1690& 0.0959& 0.1514& 0.2383& 0.3455\end{array}\right]$$

The fuzzy comprehensive evaluation algorithm mentioned above yielded a maximum risk assessment value of 0.3455, corresponding to the “high risk” category in the risk assessment set. Thus, under complex navigation conditions, manned ships are classified as “high risk” based on the risk assessment, which aligns with expert investigation findings. Simultaneously, it can be observed that under consistent conditions, a decrease in ship operators’ abilities corresponds to an increase in the risk of ship navigation. This alignment with reality that lowers the abilities and qualities of ship operators results in compromised safety conditions during navigation. Therefore, this comparison demonstrates that the method proposed in this study is both scientifically rigorous and effective.

In addition, in order to further validate the efficacy of the proposed methodology in this study, additional adverse hydrological and meteorological conditions, as well as a complex navigational environment, are introduced while keeping the shore-based operators and ship factors unchanged. The semi-loaded navigation safety case for unmanned autonomous ships is verified under these challenging conditions. The ship used has an age of 2 years, highly educated and trained personnel onboard, excellent maneuverability, accurate situational judgment skills, narrowband communication bandwidth, and average semantic understanding capability. However, there are obstacles due to high traffic flow density and insufficient navigational aids. The wind level in the sailing environment is grade 9 with poor visibility, leading to difficulties in identifying all targets. Through the distribution of questionnaires and expert validation, it is determined that the safety level of unmanned autonomous ships navigating under these conditions is considered “low” compared to the aforementioned case study. Next, we will employ the proposed methodology from this study to conduct a risk assessment for unmanned autonomous ship navigation under even more adverse hydrological and meteorological conditions along with a complex navigational environment. After performing calculations using this method, the risk assessment results of unmanned autonomous ships navigating under more adverse hydrological and meteorological conditions and navigational environments were obtained as follows:

$$A^{(1)}•R^{(1)}=\left[\begin{array}{ccccc}0.4728& 0.1771& 0.1434& 0.0987& 0.1068\end{array}\right]$$

According to the results obtained from the aforementioned fuzzy comprehensive evaluation algorithm, the maximum value attained in risk assessment is 0.4728, which is smaller than the previously calculated result of “0.5734” in a similar case. Consequently, it can be inferred that unmanned autonomous ships’ navigation safety diminishes under more intricate navigational conditions, corroborating expert surveys. Thus, through empirical validation across diverse environmental scenarios, this study substantiates the scientific and effective nature of the proposed methodology.

It can be seen that the navigation risk assessment results of unmanned ships under complex navigation conditions constructed in this study are consistent with the expert identification results. Unmanned ships can sense, understand, and resolve risks, using their own multiple on-board instruments, under complex navigation conditions through the strong situational awareness of shore-based control personnel, so as to achieve safe navigation.

4. Conclusions and Prospect

4.1. Conclusions

This study focuses on the navigation risk assessment of unmanned ships under complex navigation conditions. The conclusions are drawn as follows:

(1)

We reviewed the development status of unmanned ships worldwide, as well as the research status of risk assessment at home and abroad, to lay the foundation for establishing a risk assessment of unmanned ships under complex navigation conditions.

(2)

The risk assessment indexes of unmanned ships are analyzed and screened theoretically, the establishment process of the risk assessment index system of unmanned ships is expounded, and the risk factor indexes affecting the navigation safety of unmanned ships under complex navigation conditions are determined by using the analytic hierarchy process; a consistency test is also carried out.

(3)

The fuzzy comprehensive evaluation method is used to determine the membership degree of risk factors of unmanned ships under complex navigation conditions, and the algorithm in this study is verified through cases, where the navigation risk assessment results of unmanned ships under complex navigation conditions constructed in this study are consistent with the expert identification results, proving the effectiveness and applicability of the proposed risk assessment method.

(4)

The research also has some limitations. One of the limitations lies in the lack of empirical data for autonomous ships, resulting in a little arbitrary assignment of membership degrees to certain factors, such as shore-based control personnel. Another limitation is that, due to the lack of international standards for the size and maneuverability of unmanned vessels, this study temporarily classifies them based on conventional manned vessels. In addition, the fuzzy comprehensive evaluation method used in this study has subjectivity on member degrees to some extent. This is due to the nature of fuzzy comprehensive evaluation itself.

4.2. Prospects

This study focuses on the research of highly autonomous unmanned ships. However, in the development process of unmanned ships, ships with other autonomy degrees will also exist in large numbers. In future studies, the navigation risk assessment of the autonomous ships with other degrees according to the ‘Degree of Autonomy’ defined by IMO under complex navigation conditions could be carried out.

For overcoming the problem of data lacking during the evaluation process, the digital twin technology or building simulated environments can be used to construct navigation scenarios in order to collect data. In addition, it is suggested that the subjectivity in the research can be reduced with the support of a large amount of data in a future study.

The most important step in risk assessment and analysis is to select risk impact indicators. Since unmanned ships are not mature at present, there are still many unknown factors that may affect the safe navigation of unmanned ships. Future research could adjust different risk impact factors in real time according to the development trend of unmanned ships and then improve the navigational risk assessment results of unmanned ships.