Research on Navigation Risks in Waterway Tunnels Based on Measurement of the Cognitive Load of Ship Officers

Abstract

Ship waterway tunnels are a new and special type of navigation facility that has emerged in the construction of complex hubs in high mountain valleys and rivers, and they have demonstrated broad applications worldwide. Due to their characteristics of long length, a dim visual background, and enclosed space, waterway tunnels are prone to causing tension and cognitive fatigue in ship officers on watch, affecting their decision-making and control abilities. This study constructs the visual navigation environment of a typical waterway tunnel in China using a ship maneuvering simulator. By monitoring the physiological data of ship officers, such as through electroencephalograms (EEGs) and electrocardiograms (ECGs), the temporal and spatial patterns of their physiological and psychological characteristics are analyzed systematically. Based on this, a quantitative model of the cognitive load of a ship officer working in a waterway tunnel is constructed. At the same time, the navigation risk of waterway tunnels of different lengths is quantized based on the entropy weight TOPSIS method, and finally, high-risk sections in waterway tunnels are identified and visualized, providing theoretical support for the management of safety in waterway tunnels.

1. Introduction

With the rapid development of the shipping industry in recent years, the construction of waterway tunnels has been continuously promoted, becoming an important means of solving the problem of ship navigation in mountainous rivers under complex terrain conditions [1]. Waterway tunnels are a new and special type of ship navigation facility that provide one-way or two-way navigation conditions for ships navigating high mountain valleys and rivers, significantly increasing the channel size of restricted sections and shortening the distance traveled by ships. As such, they are receiving increasing attention all over the world; examples include the Goupitan Tunnel in China and the Stad Tunnel in Norway. Waterway tunnels are usually semi-enclosed environments [2]; for example, the typical navigation environment in the Goupitan Tunnel is significantly different to the navigation environment in open waters, which puts higher demands on the perception abilities and behavioral decision-making of ship officers. The design of the tunnel’s geometric structure and the layout of its internal lighting will directly affect the difficulty of the officer’s operation and their psychological state. The ship officer on watch not only needs to quickly adapt to environmental factors such as sudden changes in light and spatial compression inside the tunnel, but they also need to maintain precise control of the ship’s state in a limited space. At the same time, prolonged exposure to the dark and closed monotonous environment of the tunnel can easily lead to a lack of concentration and a decrease in alertness and reaction speed, causing physiological and mental fatigue which affects officers’ ability to analyze and process the navigational information needed to make decisions on how to manipulate the ship, directly affecting their driving behavior and threatening navigation safety.

As the core element of ship maneuvering, changes in the psychological and physiological characteristics of ship officers are important indicators reflecting their level of stress in response to the environment and their cognitive load [3], which can effectively reflect the risks of ship maneuvering. Cognitive load usually manifests as physiological response behavior, and physiological measuring instruments such as eye-trackers, chest straps, and electroencephalographs have been widely applied in the research field of road and aviation transportation in order to evaluate drivers’ attention distribution, cognitive load, and fatigue status. In terms of selecting psychological and physiological indicators for traffic drivers, changes in indicators such as electroencephalographs, electrocardiographs, and eye movement are usually selected to characterize changes in cognitive load [4,5,6]. In the study of driving risk assessment in transportation areas, some scholars use data analysis methods such as the entropy weight method and the TOPSIS method to determine the weights of various established indicators in the driving area [7,8,9] and evaluate the risk level based on ideal-solution ranking. Through this, researchers can scientifically identify high-risk sections in the driving area, systematically reveal the spatial distribution patterns of driving risks, and provide data support for segmented management. Based on existing research, the use of ship maneuvering simulators for simulating ship operations in tunnel navigation is becoming increasingly advanced [10,11]. For example, Deng et al. [12] conducted three types of simulation experiments using a ship handling simulator—tunnel navigation simulation, braking distance maneuvering, and small-rudder angle maneuvering—to analyze the risk characteristics of tunnel navigation. Alcaide [13] pointed out that human factors are a significant cause of maritime accidents in Hainan. Therefore, a ship handling simulator was employed to analyze workload and stress levels. Yusuf [14] pointed out that increased ship size leads to severe accident losses in narrow waterways. Therefore, ship maneuvering simulators were used to simulate navigation and analyze risk levels.

In summary, existing studies have primarily addressed the interaction between the navigation environment in tunnels and the ships themselves, such as the structural design of tunnels, ship traffic capacity, and other engineering and technical aspects [15,16]. Environmental elements such as lighting and tunnel length may directly affect the decision-making quality and maneuvering behavior of ship officers. There is still a lack of research on how the changes in the physiological characteristics of ship officers affect tunnel navigation safety, and the use of cognitive load to classify risk zones in waterway tunnels requires further development. At present, there are relatively advanced methods for assessing the physiological and psychological changes in drivers in other research fields, such as highways and aviation [17], which can provide some points of reference for the study of the behavioral impact on ship officers in waterway tunnels. However, considering the significant differences in environmental factors in waterway tunnels, the applicability of directly citing research results from other transportation fields still needs to be verified.

This study focuses on the potential safety impact of psychological factors during navigation in waterway tunnels. By integrating multimodal physiological data such as EEG and ECG indicators, the time and spatial changes in the cognitive load of ship officers during tunnel navigation are systematically analyzed based on a large-scale ship operation simulator. Moreover, a quantitative model of the cognitive load of ship officers in waterway tunnels is constructed, and the effectiveness of the quantitative model prediction is verified. At the same time, the behavior indicator of steering frequency is introduced as an explicit feature representing navigation tasks, and entropy weight TOPSIS is used to analyze the changes in cognitive load during tunnel navigation from the perspective of safety driving, ultimately achieving the identification and visualization of high-risk sections in waterway tunnels with different lengths, providing scientific support for the design, safety management, and safety training of ship officers. In this study, Section 2 illustrates the scene construction of the Silin Tunnel based on the ship maneuvering simulator, Section 3 gives details of the design and implementation of the psychological monitoring experiment, including the preparation of monitoring equipment and the experimenter, Section 4 analyzes the physiological characteristics and navigational behavior using the EEG and ECG indicator, and a cognitive load quantification model is constructed in Section 5 based on the simulation results. The risk zones of tunnels of different lengths are also determined using the cognitive load quantification model. Section 7 summarizes the conclusions and findings of this paper.

2. Construction of Ship Maneuvering Simulation Scene

2.1. Visual Scene Modeling of Waterway Tunnel

The prototype researched in this study is the Silin Tunnel, a planned navigation facility for the second line of the Wujiang River. The cross-sectional size of the Silin Tunnel is the largest among the navigation facilities along the entire Wujiang River, with a net width of 16 m, a channel depth of 5.5 m, a net height of 20 m, a dome top of 9.4 m, a designed navigation water level of 431 m, and a length of 2.2 km. Silin Tunnel has a straight layout, a flat bottom slope, a city-gate-shaped cross-section, and emergency escape routes on both sides. The tunnel only provides one-way navigation for ships. The cross-sectional structure and geometric model of the tunnel are shown in Figure 1 and Figure 2.

The scenarios construction of the waterway tunnel is based on Wartsila’s NT-PRO 5000 ship maneuvering simulator. The simulator is equipped with 40 desktop systems running NT-Expert V3.35 and complies with the requirements of the International Maritime Organization (IMO) STCW Convention as well as relevant regulations of the China Maritime Safety Administration. It is one of the largest and most advanced simulators in terms of ship simulation [18], navigation equipment simulation, and marine simulation imaging technology. The system features seven visual channels and a 270° three-dimensional visual environment with dynamic audio-visual capabilities. Details of the simulator are shown in Figure 3.

The construction of the tunnel model in this study maintains consistency with the cross-sectional parameters of the Silin Tunnel. Figure 4 shows the three-dimensional visual scene of the tunnel.

2.2. Representative Ship Model

The maximum vessel designed for navigation in this tunnel is a 1000 DWT bulk carrier, and the main parameters of representative ship types are shown in

Table 1. Main parameters of the experimental ship type.

Ship Type	Length	Beam	Full-Load Draft	Displacement Volume	Block Coefficient
1000 DWT bulk carrier	58 m	11 m	2.5 m	1033 m3	0.771

, below. This study constructs a three-dimensional model of a ship based on its scale and hydrodynamic performance of representative ship types, as shown in Figure 5.

3. Design and Implementation of Psychological Monitoring Experiment

3.1. Preparation of Monitoring Equipment

The EEG and ECG indicators of the ship officer are collected during navigation through tunnels of different lengths. EEG signals are acquired using the Emotive EPOC Flex2 wireless portable EEG system. The Emotiv EPOC Flex2 is a high-precision EEG acquisition device made by WYNTRON INC. (Rosario, Cavite, Philippines), which is suitable for monitoring brain activity, emotional responses, and psychological states, as shown in Figure 6a. Heart rate measurements utilize the Polar H10 ECG device from Polar Electro. The Polar H10 employs ECG technology capable of providing millisecond-level high-precision R-R interval data, with a correlation coefficient of 0.93 compared to medical-grade ECG systems, as shown in Figure 6b. The experimenter wearing the monitoring equipment is shown in Figure 7.

3.2. Preparation of Experimenters

Eight experienced inland waterway ship officers (all male) were recruited for the simulation experiments. All participants have operational experience on the representative ship type, holding valid captain or senior crew qualification certificates, and all of them demonstrate good psychological resilience and stress resistance, and have no psychiatric and cardiovascular disorders or other severe physical impairments that could affect physiological signal acquisition. All participants were required to abstain from alcohol and medications potentially affecting experimental results for 12 h prior to formal testing. Figure 8 presents the basic information of the participating ship officers.

3.3. Experimental Condition Settings

The impact of two main influencing factors, tunnel length and illumination, are mainly evaluated in this study. Upstream and downstream of the water area where the tunnel is located, there are blocking facilities such as locks and ship lifts, so the water flows relatively gently and is considered to be unidirectional flow with a velocity of 0.2 kn, and wind conditions are set to be no wind. Based on the existing tunnel design and experimental requirements, the simulation experiment tunnel lengths for this study are determined to be 500 m, 1500 m, 3000 m, and 5000 m, which can basically cover the various length changes from shorter tunnels to extra-long tunnels worldwide.

According to the Guide for the Lighting of Road Tunnels and Underpasses (CIE 88:2004), tunnel entrance luminance is set at 1/10 to 1/5 of external brightness. Internal segment illuminance ranges from 5 to 30 lx, while the exit segment slightly exceeds internal levels (1.5–3 times). Aligned with Silin Tunnel design specifications, experimental internal illuminance values are configured as 10 lx, 50 lx, 100 lx, and 200 lx, spanning low to high illumination ranges. Each experimenter experienced two experiments, as listed in

Table 2. Simulation scenarios and parameter settings.

Scenario	Water Depth (m)	Tunnel Width (m)	Sailing Speed (m/s)	Tunnel Length (m)	Tunnel Illuminance (lx)
Scenario 1	5.5	16	1.5	500	10
Scenario 2	5.5	16	1.5	500	50
Scenario 3	5.5	16	1.5	500	100
Scenario 4	5.5	16	1.5	500	200
Scenario 5	5.5	16	1.5	1500	10
Scenario 6	5.5	16	1.5	1500	50
Scenario 7	5.5	16	1.5	1500	100
Scenario 8	5.5	16	1.5	1500	200
Scenario 9	5.5	16	1.5	3000	10
Scenario 10	5.5	16	1.5	3000	50
Scenario 11	5.5	16	1.5	3000	100
Scenario 12	5.5	16	1.5	3000	200
Scenario 13	5.5	16	1.5	5000	10
Scenario 14	5.5	16	1.5	5000	50
Scenario 15	5.5	16	1.5	5000	100
Scenario 16	5.5	16	1.5	5000	200

.

4. Analysis of Physiological Characteristics and Navigational Behavior

Based on physiological data collected under various conditions, this study analyzes variations in officers’ EEG power ratios, heart rate (HR), and heart rate variability (HRV) under different tunnel lengths and illuminance levels. Combined with navigational behavior characteristics, these findings reveal behavioral patterns under stress, workload, and fatigue.

4.1. EEG Indicator Variations

EEG signals exhibit α-waves (8–13 Hz) and β-waves (13–30 Hz). Decreased α-wave activity indicates heightened alertness, while increased β-wave activity reflects elevated cognitive load [19]. Fatigue accumulation may also trigger θ-wave enhancement (4–8 Hz). Thus, the β/(α + θ) power ratio serves as the primary analytical target. Short-Time Fourier Transform (STFT) is identified as being particularly effective for non-stationary EEG signals, and is used in calculating the EEG power ratio as below [20]:

$${\displaystyle \frac{\mathsf{\beta }}{\mathsf{\alpha }+\mathsf{\theta }}}(\mathrm{t})={\displaystyle \frac{{\mathrm{P}}_{\mathsf{\beta }}(\mathrm{t})}{{\mathrm{P}}_{\mathsf{\alpha }}(\mathrm{t}){+\mathrm{P}}_{\mathsf{\theta }}(\mathrm{t})}}$$

(1)

where ${\mathrm{P}}_{\mathsf{\alpha }}(\mathrm{t})$, ${\mathrm{P}}_{\mathsf{\beta }}(\mathrm{t})$, and ${\mathrm{P}}_{\mathsf{\theta }}(\mathrm{t})$ represent integrated power spectral densities of the three wavebands.

Figure 9 shows the averaged β/(α + θ) ratios for eight experimenters under four illuminance conditions across different tunnel lengths. All illuminance levels demonstrate rising ratios at tunnel entrances, gradual stabilization during mid-sections, and secondary peaks at exits. Maximum ratios occur at entrance/early internal segments, indicating illuminance-induced alertness and stress responses. Mid-phase stabilization suggests environmental adaptation, with fatigue-driven minima appearing later.

4.2. ECG Indicator Variations

ECG signals reflect cardiac electrical activity, enabling the assessment of ship officers’ stress responses, emotional states, and physiological adaptability during tunnel navigation through HR and HRV measurements [21,22]. HRV primarily includes time-domain indicators Standard Deviation of NN Intervals (SDNN) and Root Mean Square of Successive Differences (RMSSD) [23]. SDNN represents the standard deviation of all R-R intervals over a period (i.e., the standard deviation of heartbeat intervals, measured in milliseconds), while RMSSD predominantly reflects parasympathetic nervous activity. Therefore, HR, SDNN, and RMSSD are selected as measurement targets for analyzing ECG physiological indicator variations.

(1)

Heart rate growth rate (HRGR)

By observing heart rate variations in ship officers during navigation, comparisons can be made regarding how different navigational environments influence psychological stress and tension levels. The analysis adopts the HRGR, which quantifies the rate of heart rate increase relative to baseline values, thereby enabling the precise estimation of environmental stress effects. The computational formula is defined as follows [24]:

$$N={\displaystyle \frac{n_1-n_0}{n_0}}\times 100\%$$

(2)

where $N$ denotes the ship officer’s heart rate growth rate (%), $n_0$ represents the mean resting heart rate of the ship officer (beats/min), and $n_1$ indicates the dynamic heart rate of the ship officer (beats/min).

Statistical analysis of HRGR for eight officers under four tunnel lengths and varying illuminance conditions revealed the mean HRGR trends across time points, as shown in Figure 10. All four illuminance levels exhibited consistent patterns: rapid initial increase to peak values, followed by gradual decrease with fluctuating stabilization, and terminal transient elevation. Significant HRGR increases occurred during tunnel entrance and early internal segments under all illuminance conditions, indicating adaptation of demands to abrupt lighting changes. The 10 lx condition demonstrated the highest peak HRGR, reflecting maximal stress response intensity.

(2)

Variations in SDNN

In time-domain analysis, continuous R-R interval sequences are first extracted from ECG signals after excluding ectopic beats and noise artifacts. The arithmetic mean of all R-R intervals is calculated, followed by computing of the sum of squared deviations between each R-R interval and the mean value. Finally, the unbiased estimator is applied to determine the mean deviation, whose square root yields SDNN. The formula for SDNN is defined as

$$SDNN=\sqrt{{\displaystyle \frac{1}{N-1}}\sum _{i=1}^N{\left(R{R}_i-\overline{RR}\right)}^2}$$

(3)

where $R{R}_i$ denotes the value of the i-th R-R interval (ms); $\overline{RR}$ represents the mean of all R-R intervals; and N indicates the total number of R-R intervals.

Taking the 100 lx illuminance condition as an example, the distribution of SDNN variations under four tunnel lengths is shown in Figure 11. The overall SDNN pattern exhibited an initial gradual decline, stabilization at lower levels in later phases, and a slight recovery post-task completion, reflecting diminished HRV due to cumulative fatigue. During the tunnel entrance phase, sympathetic activation to meet navigational demands suppressed parasympathetic activity, initiating SDNN reduction from baseline levels. Prolonged navigation time enhanced environmental adaptation, sustaining sympathetic dominance while fatigue accumulation weakened autonomic regulation capacity, further decreasing SDNN. In the late internal tunnel phase, peak fatigue severely suppressed parasympathetic tone, causing homogenized heart rhythms and minimal HRV, with SDNN reaching its lowest value. At the tunnel exit, a partial alleviation of sympathetic excitation allowed slight SDNN recovery, though it remained below baseline levels. For extra-long tunnels, low-value outliers may emerge in later phases due to individual differences in extreme fatigue responses.

(3)

Variations in RMSSD

The Root Mean Square of Successive Differences (RMSSD) calculates the square root of the mean squared differences between adjacent R-R intervals, with units in milliseconds, as defined by the following formula [25]:

$$RMSSD=\sqrt{{\displaystyle \frac{1}{N-1}}\sum _{i=1}^{N-1}{\left(R{R}_{i+1}-R{R}_i\right)}^2}$$

(4)

where $R{R}_i$ represents the value of the i-th R-R interval (ms); $\overline{RR}$ denotes the mean of all R-R intervals; and N indicates the total number of R-R intervals.

Taking the 100 lx illuminance condition as an example, the distribution of RMSSD variations under four tunnel lengths is shown in Figure 12. RMSSD demonstrates heightened sensitivity to initial stress responses: a rapid decrease in illuminance at the tunnel entrance reduced heart rate variability, leading to a sharp decline in RMSSD. During the early internal phase, RMSSD remained at low levels as initial fatigue accumulated. With prolonged navigation time and increasing fatigue, RMSSD continued to decrease gradually, reaching its lowest point in the late internal phase under maximum fatigue. At the tunnel exits, the gradual recovery of illuminance partially alleviates fatigue, resulting in a slight rebound in RMSSD though values remained below baseline. For the 500 m tunnel, closely spaced segments and short navigation time limited fatigue accumulation, resulting in stable RMSSD trends characterized by an initial decline, stabilization at low levels, and a slow recovery by the end. Boxplot distributions exhibit wider spreads at entrance, exit and narrower mid-tunnel sections, with maximal fatigue accumulation observed in longer tunnels.

5. Construction and Validation of Cognitive Load Quantification Model

The original variable data underwent Z-score normalization. For computational processing and visualization, the standardized variables of EEG power ratio β/(α + θ), HRGR, SDNN, and RMSSD were designated as ZEEG, ZHR, ZSDNN, and ZRMSSD, respectively. Factor Analysis (FA) is employed for dimensionality reduction, extracting commonalities from multiple indicators to simplify analysis. Based on experimental observations, Factor 1 is termed the Stress Load Factor and Factor 2 is named the Physiological Regulation Factor. Factor loadings are derived through Thompson regression analysis, yielding the following scoring functions:

$$F_1=0.614Z_1+0.359Z_2+0.072Z_3+0.588Z_4$$

(5)

$$F_2=0.258Z_1+0.767Z_2+0.504Z_3+0.224Z_4$$

(6)

where Z₁, Z₂, Z₃, and Z₄ represent standardized β/(α + θ) ratio, SDNN, RMSSD, and HRGR, respectively.

The cognitive load quantification model for tunnel navigation is expressed as

$$P={\displaystyle \frac{W_1}{W_1+W_2}}{F}_1+{\displaystyle \frac{W_2}{W_1+W_2}}{F}_2={\displaystyle \frac{1}{0.98732}}\left(0.52164F_1+0.46567F_2\right)$$

(7)

The weights of the two common factors in the formula are determined by their respective variance contribution rates. Calculations revealed that the weight of Common Factor 1 (Stress Load Factor) is $W_1=0.52164$, and the weight of Common Factor 2 (Physiological Regulation Factor) is $W_2=0.46567$.

To validate the effectiveness of the cognitive load quantification model for ship officers in waterway tunnels, this study is grounded in the theoretical foundations of officers’ physiological characteristics. The Low Frequency (LF) primarily reflects sympathetic activity modulated by parasympathetic influence, while the High Frequency (HF) component is chiefly associated with parasympathetic activity, so the LF/HF ratio represents sympathetic–parasympathetic balance. These metric variations are manifested in both the Stress Load Factor and Physiological Regulation Factor. Consequently, correlation analysis is conducted between empirically measured LF/HF ratios and model-calculated cognitive load values to verify quantification accuracy.

A comparative analysis is performed between variations in officers’ cognitive load levels and HRV frequency-domain data under different illuminance conditions across four tunnel lengths. Pearson correlation analysis between the LF/HF ratio and the model-calculated cognitive load values produced a correlation coefficient of 0.876 (p = 0.876, Sig = 0.000), indicating strong associations between HRV frequency-domain indicators and cognitive load. These results suggest that HRV metrics may serve as potential quantitative indicators of cognitive load, operating through mechanisms analogous to those captured by the Stress Load Factor. The findings demonstrate that inferences drawn from the cognitive load quantification model are consistent with variations in LF/HF characteristics, confirming the model’s effectiveness in characterizing ship officers’ cognitive load under typical navigational environmental factors such as tunnel length and illuminance.

6. Determination of Risk Zones Based on Cognitive Load Measurement

6.1. Analysis Methodology of Navigation Risk Zone Classification

This section classifies tunnel navigation risk zones from a ship operation perspective via the Entropy-Weighted TOPSIS method, utilizing the cognitive load quantification model for ship officers in waterway tunnels developed in this study. As a multi-attribute decision analysis approach [26,27], the Entropy-Weighted TOPSIS method comprehensively considers multiple influencing factors on navigation risks, including cognitive load, steering behavior, and environmental conditions. This methodology effectively integrates these indicators by objectively determining their weights through entropy value calculation, subsequently enabling risk ranking and zonal classification through TOPSIS. The computational procedure involves the following.

(1)

Data preparation and indicator selection

In the risk assessment indicator system of this study, both cognitive load and steering frequency are positive indicators, where higher values indicate greater driving pressure, more frequent operations, and increased potential risks. The entire tunnel navigation route is divided into segments based on 30 s intervals of navigation time. For each segment, the average values of these two indicators are calculated to form the original data matrix:

$$X=\left[\begin{array}{cc}{x}_{11}& x_{12}\\ x_{21}& x_{22}\\ ⋮& ⋮\\ x_{n1}& x_{n2}\end{array}\right]$$

(8)

where $x_{i1}$ represents the cognitive load score of the i-th segment and $x_{i2}$ denotes the steering frequency value.

(2)

Data standardization and entropy weight calculation

Standardization is performed on both indicators to construct a normalized matrix R for dimensionality elimination. The entropy method is employed to calculate information entropy and redundancy degree, determining objective indicator weights w₁ and w₂ to enhance evaluation accuracy and adaptability.

$$z_{ij}={\displaystyle \frac{x_{ij}}{\sqrt{{\sum }_{i=1}^m{x}_{ij}^2}}}$$

(9)

where $z_{ij}$ is the standardized value and $0\le z_{ij}\le$1.

To compute the proportion matrix,

$$p_{ij}={\displaystyle \frac{r_{ij}}{{\sum }_{i=1}^m{r}_{ij}}}$$

(10)

$$E_j=-{\displaystyle \frac{1}{\mathit{ln}(m)}}\sum _{i=1}^m{p}_{ij}\mathit{ln}\left(p_{ij}\right)$$

(11)

If $p_{ij}=0$, then $p_{ij}\ln (p_{ij})=0$.

To calculate the entropy value,

$$W_j={\displaystyle \frac{1-E_j}{{\sum }_{k=1}^n(1-E_k)}}$$

(12)

where $w_j$ is the weight of the j-th indicator, $\sum w_j=1$.

(3)

Closeness and risk score calculation

Based on the weighted normalized matrix, we can compute the distances between each evaluation object; two parameters are calculated, and one is the positive ideal solution (the maximum workload and highest steering frequency) and the other one is the negative ideal solution—minimum workload and smoothest operation. We can calculate the relative closeness Ci, which reflects the comprehensive risk level of each segment. Lower Ci values indicate higher proximity to high-risk states.

Construct the weighted normalized matrix:

$$V=\left[v_{ij}\right]=\left[w_j·r_{ij}\right]$$

(13)

Identify positive/negative ideal solutions $A^{+}$ and $A^{-}$, then calculate distances $S_i^{+}$ and $S_i^{-}$:

$$S_i^{+}=\sqrt{{\sum }_{j=1}^n{\left(v_{ij}-v_j^{+}\right)}^2},{ S}_i^{-}=\sqrt{{\sum }_{j=1}^n{\left(v_{ij}-v_j^{-}\right)}^2}$$

(14)

To compute closeness coefficient,

$$C_i={\displaystyle \frac{S_i^{-}}{{S_i^{+}+S}_i^{-}}}$$

(15)

where $0\le C_i\le 1$, with smaller $C_i$ values indicating higher risk levels.

(4)

Risk level classification and spatial visualization

All navigation segments are classified into three risk levels (high, medium, low) based on closeness coefficient distributions. By integrating the tunnel’s geometric configuration, risk levels are mapped to physical locations to generate risk distribution diagrams, enabling the spatial identification of high-pressure zones and providing early-warning support.

6.2. Calculation of Navigation Risk Values

Using the aforementioned calculation method, quantitative risk values are computed for tunnels of different lengths. Considering the relatively slow navigation speed of vessels within the tunnels, closeness coefficient values are calculated at 30 s intervals. The resulting navigation risk values for four distinct tunnel lengths are presented in

Table 3. TOPSIS calculation scores for 500 m tunnel.

Tunnel Length	Navigation Time (min)	Relative Position (m)	Closeness Coefficient Ci	Risk Value	Ranking
500 m	0	0	0.599	0.401	6
	0.5	50	0.221	0.779	2
	1.0	100	0.149	0.851	1
	…	…	…	…	…
	3.5	350	0.847	0.153	11
	4.0	400	0.711	0.289	9
	4.5	450	0.608	0.392	7
	5.0	500	0.544	0.456	5

,

Table 4. TOPSIS calculation scores for 1500 m tunnel.

Tunnel Length	Navigation Time (min)	Relative Position (m)	Closeness Coefficient Ci	Risk Value	Ranking
1500 m	0	0	0.608	0.392	16
	0.5	50	0.235	0.765	4
	1.0	100	0.202	0.798	2
	1.5	150	0.143	0.857	1
	…	…	…	…	…
	13.5	1350	0.599	0.401	15
	14.0	1400	0.620	0.380	18
	14.5	1450	0.633	0.367	20
	15.0	1500	0.637	0.363	21

,

Table 5. TOPSIS calculation scores for 3000 m tunnel.

Tunnel Length	Navigation Time (min)	Relative Position (m)	Closeness Coefficient Ci	Risk Value	Ranking
3000 m	0	0	0.579	0.421	28
	0.5	50	0.191	0.809	2
	1.0	100	0.113	0.887	1
	1.5	150	0.199	0.801	3
	…	…	…	…	…
	28.5	2850	0.470	0.530	9
	29.0	2900	0.487	0.513	12
	29.5	2950	0.514	0.486	19
	30.0	3000	0.547	0.453	23

and

Table 6. TOPSIS calculation scores for 5000 m tunnel.

Tunnel Length	Navigation Time (min)	Relative Position (m)	Closeness Coefficient Ci	Risk Value	Ranking
5000 m	0	0	0.617	0.383	63
	0.5	50	0.208	0.792	2
	1.0	100	0.116	0.884	1
	1.5	150	0.244	0.756	3
	…	…	…	…	…
	48.5	4850	0.474	0.526	26
	49.0	4900	0.438	0.562	19
	49.5	4950	0.383	0.617	14
	50.0	5000	0.351	0.649	12

below.

6.3. Classification of Navigation Risk Zone for Tunnels of Different Lengths

Based on the tunnel navigation risk values calculated via the TOPSIS method, K-means clustering analysis is applied for risk stratification. High-risk segments are defined by a closeness coefficient Ci > 0.72, medium-risk segments by 0.38 ≤ Ci ≤ 0.72, and low-risk segments by Ci < 0.38. The resulting navigation risk zone classifications for tunnels of different lengths are as follows.

(1)

Analysis of risk zones in 500 m tunnel

Figure 13 shows the risk zone distribution in the 500 m waterway tunnel. The proportions of high-, medium-, and low-risk zones along the entire route are 27.8%, 31.2%, and 41.0%, respectively. The high-risk zones appear 0.29–1.68 min after the ship enters the tunnel, that is, approximately 29–168 m from the entrance, and reach the maximum risk value of 0.851 at about 62.4 s when navigating to a position about 104 m from the entrance. The low-risk zones appear at 239–444 m positions in the tunnel, and reach the minimum risk value of 0.121 at about 3.3 min when navigating to a position about 330 m into the tunnel. Other areas are medium-risk zones, appearing at 29 m after entering the tunnel, the middle section of 168–239 m, and about 56 m from the exit. This indicates that the first 240 m of the 500 m tunnel belongs to the medium–high risk zones. Due to the relatively short length of the tunnel and relatively short navigation time, the overall risk value in the entrance section is high, and has a significant impact on ship officers and is likely to cause tension and operational errors.

(2)

Analysis of risk zones in 1500 m tunnel

Figure 14 presents the navigation risk zoning results for the 1500 m tunnel. The distribution shows that high-, medium-, and low-risk zones account for 13.7%, 43.5%, and 42.8% of the total navigation distance, respectively. The high-risk zone occurs between 0.36 and 2.42 min after entry (approximately 36–242 m from the entrance), with the maximum risk value of 0.859 appearing at about 1.46 min (146 m). Medium-risk zones are identified in three segments: the initial 0.36 min (36 m), 2.42–6.52 min (242–652 m), and 12–14 min (1200–1400 m). The low-risk zone spans 652–1200 m and the final 100 m near the exit, reaching its minimum risk value of 0.203 at 10.2 min (1020 m).

Comparative analysis with the 500 m tunnel reveals that the 1500 m tunnel exhibits an increased proportion of medium-risk zones (43.5%), making it the dominant risk category. While the high-risk proportion decreases, it remains concentrated in the entrance section and early interior segment. The first 242 m (approximately 16% of total length) maintains medium-to-high risk levels until about 43% of the tunnel length, where it transitions to low risk. During the later navigation phase, a secondary medium-risk peak of 0.414 occurs at 12.8 min (approximately 1300 m, 200 m before exit), which subsequently declines as the vessel approaches the exit. These findings highlight the extended dominance of medium-risk conditions in longer tunnels while maintaining critical high-risk concentrations near entry points.

(3)

Analysis of risk zones in 3000 m tunnel

Figure 15 presents the navigation risk zoning results for the 3000 m tunnel. The distribution shows that high-, medium-, and low-risk zones account for 5.3%, 43.1%, and 51.6% of the total navigation distance, respectively. The high-risk zone occurs between 0.27 and 1.85 min after entry (approximately 27–185 m from the entrance), with the maximum risk value of 0.889 appearing at 0.95 min (95 m). The low-risk zone spans 8.7–24.2 min (870–2420 m), reaching its minimum risk value of 0.158 at 2080 m, and constitutes the largest proportion (51.6%) of the entire route. The remaining sections are classified as medium risk, peaking at 0.539 at 27.7 min (approximately 2770 m, 230 m before the exit).

Compared to the 1500 m tunnel, the 3000 m tunnel maintains a similar proportion of medium-risk zones while significantly expanding the low-risk duration to over half of the total distance. The initial 185 m section maintains medium-to-high risk levels, with a slight risk rebound at 430 m before gradually declining. During the latter phase, after entering the final 1000 m, the overall risk decreases before fatigue effects emerge at 20.8 min, causing risk levels to rise again. This elevated medium-risk condition persists for the remaining 3.4 min until the exit, demanding heightened attention and maneuvering skills from pilots throughout this critical final phase. The extended low-risk dominance in mid-tunnel sections contrasts with concentrated high-risk segments near the entry and fatigue-induced medium-risk zones before the exit.

(4)

Analysis of risk zones in 5000 m tunnel

Figure 16 shows the division of navigation risk zones in the 1500 m tunnel. It can be seen that in the 5000 m tunnel, the proportions of high, medium, and low-risk zones across the entire navigation segment are 1.1%, 33.7%, and 65.2%, respectively. The high-risk zone occurs 0.38–1.7 min after the vessel enters the tunnel, corresponding to a distance of 38–170 m from the entrance, with the maximum navigation risk reaching 0.886 at the 90 m mark. The low-risk zone appears in the section between 550 m and 3810 m inside the tunnel, primarily concentrated in the middle segment, where the navigation risk reaches its minimum value of 0.25 when the vessel travels approximately 1550 m (after 15.5 min of navigation). The low-risk zone accounts for the largest proportion overall at 65.2%. Other sections are classified as medium-risk zones, occurring in the first 38 m of the tunnel, the 170–550 m segment, and the final 270 m near the exit.

Nearly two thirds of the 5000 m tunnel consists of low-risk segments, all located in the interior section. Compared to shorter tunnels, this tunnel has the smallest proportion of high-risk zones, which are still concentrated within the first 200 m of the entrance. The overall risk level trend shows a rapid initial increase, followed by a gradual decline, a slight fluctuation with a gentle rise during the middle phase, and a final short, rapid increase near the end. After 44.5 min of navigation, when reaching the mid-to-late section of the tunnel, navigation risk rises to 0.501, which is a high value within the medium-risk range due to accumulated fatigue. It then slightly decreases before rapidly climbing to 0.649 in the exit segment.

7. Conclusions and Discussion

This study collects physiological data of ship officers during navigation in waterway tunnels, constructs a quantitative model of cognitive load, and divides the navigation risk zones of tunnels of different lengths based on the measurement of cognitive load. The main conclusions are as follows.

(1)

The impact of changes in tunnel length on the physiological characteristics of a ship officer is reflected in the accumulation of fatigue and work load caused by changes in navigation time. Increasing the tunnel length prolongs the time navigating a monotonous environment in the internal section, and the fatigue accumulation effect initially appears, leading to an increase in EEG theta waves and cumulative changes in heart rate and heart rate variability. The impact of changes in tunnel illumination on the physiological characteristics of the ship officer is reflected in high-contrast visual stimuli and immediate stress responses during the entrance and exit sections of tunnel. Poor illumination leads to an increase in EEG beta waves in physiological indicators, rapid fluctuations in heart rate, inhibition of parasympathetic nervous activity, and short-term imbalance in heart rate variability. Especially in long tunnels over 3000 m and with dim illumination below 50 lx, the increase in cognitive load is particularly significant, reflecting the typical characteristics of induced risk effects in ultra-long tunnels.

(2)

Based on the structural segmentation of the tunnel entrance section, internal section, and exit section, combined with the time series evolution of risk indicators, it is found that ship officers had significant stress responses in the tunnel entrance section, with significant fluctuations in physiological indicators and a rapid increase in cognitive load. The cognitive load of the tunnel is relatively stable, but it remains consistently high in the ultra-long tunnel. The exit section of the tunnel exhibits a certain degree of relaxation. Under tunnels of different lengths, there are significant differences in the peak risk and cognitive load response time points of each segment, indicating that the design of navigation tunnel length has an important impact on the phased load distribution of the ship officer’s navigation process.

(3)

By calculating the cognitive load values of each time segment in four different lengths of tunnels, the entropy weight TOPSIS method is used to obtain the tunnel navigation risk value and risk section division. The analysis shows that the navigation risk levels of each tunnel are highest, respectively, at the first 168 m, first 242 m, first 185 m, and first 170 m segments. This indicates that under the experimental conditions, the ship officer should maintain high vigilance in the first 100–200 m segment of the tunnel, concentrate their attention in the first 1–2 min of tunnel navigation, and adjust the steering direction in a timely manner. The navigation risk within 300 m of the exit in a 1500 m tunnel slowly increased, while the navigation risk level in a 3000 m tunnel began to rise after the 2500 m section.

(4)

For an ultra-long tunnel like the 5000 m tunnel, the navigation risk level also increased to medium risk after the 4000 m section. Based on the analysis of the data results from this simulation experiment, it can be concluded that the ship officer may need to guard against navigation risks caused by fatigue accumulation and other factors during 4/5 of the sailing voyage when navigating in long tunnels with a navigation time exceeding tens of minutes. At this time, it is appropriate to increase the duty lookout to reduce the continuous cognitive load of the ship officer. Transitional light strips or visually flat areas can also be set up at this location to reduce navigation fatigue.

(5)

This study extracts key indicators reflecting the cognitive load and physiological response characteristics of ship officers, including EEG power ratio β/(α + θ), heart rate growth rate, and heart rate variability time-domain indicators SDNN and RMSSD. The concepts of Stress Load Factor and Physiological Regulation Factor are proposed, and a quantitative model for the cognitive load of a ship officer in waterway tunnels is constructed. The effectiveness of the model is verified by the changes in LF/HF indicators, and finally a comprehensive risk assessment method of entropy weight TOPSIS is used to calculate and visualize the risk values of each section in waterway tunnels. The results show that high-risk periods are mostly concentrated in the areas with sudden changes in illumination at the entrance and inside the tunnel, with certain stages and sudden patterns. The research results of this study provide a scientific basis for the assessment of the load status of officer in waterway tunnel scenarios, and are also of great significance for risk warning and risk control of waterway tunnels.

This study focused on the navigation of a single vessel within a tunnel, and conducted relevant analyses by comprehensively considering changes in EEG, heart rate, and HRV indicators. However, during the entry and exit phases of tunnel navigation, variations in lighting conditions, along with eye-tracking data such as blink frequency and pupil size, can be used to assess the immediate impact of visual stress. Additionally, electrodermal activity can serve as a sensitive indicator reflecting psychological stress and emotional fluctuations. Future research could incorporate such indicators to enhance the richness of the findings.