Intelligent ship navigation system based on space-time prediction risk field and deep reinforcement learning

Abstract

The invention discloses an intelligent ship navigation system based on a space-time prediction risk field and deep reinforcement learning, relates to the technical field of ship navigation, and aims to solve the problem that a conventional navigation system is insufficient in dynamic risk assessment and prospective planning in the prior art. The technical key points are that the system comprises a multi-source data acquisition module, a dynamic risk mapping module, an intelligent path planning module, a visual interaction module, a risk early warning module and a self-checking module. According to the system, multi-source data such as weather, ocean current and ship dynamics are collected in real time, a predictive risk field is generated by using a long-short-term memory network, a route is optimized in combination with an adaptive deep reinforcement learning algorithm, and risk distribution and a recommended route are visually presented in a multi-scale three-dimensional form; the system can early warn a high-risk area in real time, and the reliability is ensured through a self-checking function. According to the invention, the sailing safety and the intelligent level of the ship in a complex environment are obviously improved.

Description

Technical Field

[0001] This invention relates to the field of ship navigation technology, and more specifically, to a ship intelligent navigation system based on spatiotemporal prediction risk field and deep reinforcement learning. Background Technology

[0002] As the shipping industry moves towards intelligent and unmanned operations, the limitations of traditional ship navigation systems are becoming increasingly apparent. Existing technologies mainly suffer from the following technical bottlenecks:

[0003] 1. Limitations of static environment models Traditional ship navigation systems typically rely on static geographic information such as electronic charts for risk assessment. This approach can only handle unchanging factors like fixed channels and reefs, and cannot effectively address dynamic environmental threats such as typhoons and severe sea conditions. While existing real-time navigation systems can acquire current weather and sea state data, they lack the ability to predict future risk trends. This often results in ships reacting only after a risk has already materialized, missing the optimal opportunity for avoidance.

[0004] 2. Disconnect between path planning and environmental perception Current reinforcement learning-based path planning methods mostly simplify the environment as a discrete distribution of obstacles, ignoring the continuity and spatiotemporal evolution of risks. This simplification leads to poor performance of planning systems in complex dynamic environments, especially in scenarios requiring proactive avoidance. Environmental perception and path planning, as two relatively independent modules, lack deep information fusion and collaborative optimization.

[0005] 3. Insufficiency of multi-objective optimization Traditional path planning methods often optimize path length or energy consumption in isolation, lacking a collaborative optimization mechanism among multiple objectives such as safety, efficiency, and economy. When faced with complex decisions, the system struggles to comprehensively consider navigation efficiency and economic benefits while ensuring safety.

[0006] In summary, the existing technology lacks an environmental model that can quantitatively predict future spatiotemporal risks, as well as a mechanism to deeply embed this model into the path planning decision loop. This results in ships being unable to achieve truly intelligent and forward-looking navigation when facing rapidly changing marine environments.

[0007] To address the aforementioned issues, this application proposes a ship intelligent navigation system based on spatiotemporal prediction risk field and deep reinforcement learning. Summary of the Invention

[0008] The purpose of this invention is to solve the existing technical problems mentioned in the background section and to provide a ship intelligent navigation system based on spatiotemporal prediction risk field and deep reinforcement learning.

[0009] The present invention achieves the above objectives through the following solutions: A ship intelligent navigation system based on spatiotemporal prediction of risk field and deep reinforcement learning includes: The multi-source data acquisition module is used to collect multi-dimensional data related to ship navigation in real time, including ship position, meteorological parameters, ocean current speed and surrounding ship dynamics; The dynamic risk mapping module is used to generate a predictive risk field based on the multi-source data and update the risk distribution of the navigation area in real time. The intelligent route planning module is used to generate the optimal route based on the predictive risk field using an adaptive deep reinforcement learning algorithm. A visualization and interaction module is used to present the risk field and recommended route in a three-dimensional dynamic form; The predictive risk field is generated based on the following risk value calculation formula: In the formula, coordinates At any moment The dimensionless risk value, As static environmental factors, To predict environmental factors, These are dynamic ship density factors, all of which are dimensionless. The weights are dimensionless and satisfy the following conditions: The static environmental factors It includes the sum of standardized meteorological factors and other static factors.

[0010] Furthermore, the dynamic risk mapping module includes a prediction unit, which calculates the predicted environmental factors using a Long Short-Term Memory (LSTM) network. Its formula is: ; In the formula, It uses the sigmoid activation function, with an output range of [0, 1]. It is a dimensionless weight matrix. Let be the hidden state vector from the previous time step, initialized as a zero vector. The input vector consists of multi-source data at the current moment, including standardized wind speed, wave height, and ocean current velocity. The term is a dimensionless bias term; These are dimensionless values, used directly for calculations. .

[0011] Furthermore, the dynamic ship density factor Calculated using the following formula: ; In the formula, The number of ships in the region. For the first Ships and coordinates The distance, in meters. The standard deviation of the density distribution is expressed in meters. The exponential term is dimensionless. (Results) These are dimensionless values, used directly for calculations. .like 0, then .

[0012] Furthermore, the intelligent path planning module employs an adaptive deep reinforcement learning algorithm, and its reward function is defined as: ; In the formula, The reward value is dimensionless. This represents the average risk value for the route. The length of the flight path is in meters. For reference, the maximum route length is in meters. The estimated energy consumption is expressed in joules. For reference to maximum energy consumption, the unit is joules. It is a dimensionless adjustment coefficient, and satisfies The Through formula Calculation, where Energy consumption coefficient For ship speed.

[0013] Furthermore, the average risk value of the route The risk value is calculated by integrating the risk values ​​of each grid point along the path: ; In the formula, Path is the route output by the intelligent route planning module. Here is the arc length parameter along the path, in meters. The total length of the route. The risk value at the path point, the result It is a dimensionless value.

[0014] Furthermore, the multi-source data acquisition module includes a satellite remote sensing unit, an AIS receiving unit, and an underwater sonar unit, wherein the meteorological parameters are standardized to generate standardized meteorological factors using the following formula. ; In the formula, This is the raw meteorological data. These are the minimum and maximum values ​​of the data, respectively, with units of and . Consistent, The static environmental factor is a dimensionless value. ,in For other standardized static factors.

[0015] Furthermore, the visualization interaction module supports multi-scale switching, and its risk field display resolution is adaptively adjusted according to the following formula: ; In the formula, To display the resolution, the unit is 1 / meter. These are the maximum and minimum values ​​of the current risk field, respectively, both of which are dimensionless. This is a zoom level in meters. It is a dimensionless constant; the stated and based on The statistical value.

[0016] Furthermore, the system also includes a risk warning module, which detects when the risk value of a certain grid point... Exceeding the preset threshold When the alarm signal is triggered, the area of ​​the risk exceeding the standard is calculated using the following formula: ; In the formula, This is an indicator function, dimensionless, with a value of 0 or 1. For dimensionless risk threshold, The area of ​​a single grid cell is expressed in square meters. The unit is square meters; the aforementioned Used to assess the scope of risks and guide route adjustments.

[0017] Furthermore, the adaptive deep reinforcement learning algorithm introduces a dynamic learning rate during training, the calculation formula of which is: ; In the formula, The current dimensionless learning rate, The initial dimensionless learning rate, The dimensionless attenuation coefficient is The average risk value for the route is dimensionless; and Collaborative optimization of path planning.

[0018] Furthermore, when the system is applied to unmanned vessels, it further includes a self-testing module for real-time monitoring of hardware status and for evaluating system reliability using the following formula: ; In the formula, It is a dimensionless reliability index. For the first Fault flags for individual hardware components, dimensionless, with values ​​of 0 (normal) or 1 (fault), determined based on a comparison of sensor signals with preset thresholds. The total number of components; When the value is below the threshold, prompt to adjust dependencies. Navigation decisions.

[0019] The system also includes a communication encryption and data security module, which uses a combination of symmetric and asymmetric encryption algorithms to encrypt navigation data transmitted between the ship and the shore-based control center or other ships, and evaluates the security level using the following formula: ; In the formula, The safety factor is dimensionless. This is the strength value of the symmetric encryption algorithm, which is dimensionless and ranges from [0,1]. This represents the strength value of the asymmetric encryption algorithm, which is dimensionless and ranges from [0,1]. , The weighting coefficients are dimensionless and satisfy the following conditions: ,when When this happens, the system automatically triggers a communication security warning and switches to a high-strength encryption mode to ensure the confidentiality and integrity of navigation data in a dynamic and risky environment.

[0020] Compared with the prior art, the present invention has the following beneficial effects: 1. This invention significantly improves the safety of ship navigation by introducing predictive risk fields and dynamic risk mapping technology. The system of this invention can integrate multi-source environmental data in real time and rely on Long Short-Term Memory (LSTM) networks to predict future environmental evolution trends, constructing a continuously changing risk distribution field to achieve early identification of potential threats. For example, in high-risk scenarios (such as typhoon-affected areas or densely populated waters), the system can proactively issue warnings before the risk level escalates, guiding ships to avoid dangerous areas and effectively preventing accidents such as collisions and groundings. Compared to traditional navigation methods that rely on static information and human experience, this invention achieves dynamic and forward-looking risk assessment, providing more reliable safety guarantees for ships in complex weather and congested waterway environments.

[0021] 2. This invention employs an adaptive deep reinforcement learning algorithm to achieve multi-objective route optimization, significantly improving navigation efficiency while ensuring safety. Traditional route planning methods typically focus on finding the shortest path, while this system comprehensively considers multiple factors such as safety, voyage length, and energy consumption to collaboratively generate the Pareto optimal route. The reward function is carefully designed to balance various optimization objectives, enabling ships to select economical routes with controllable voyage and energy consumption while minimizing risks. For example, when it is necessary to detour around high-risk areas, the system can quickly plan alternative routes with lower risks and limited increases in total cost, avoiding excessive detours and resource waste, and providing shipping companies with more economical operational decision support.

[0022] 3. This invention achieves significant progress in system intelligence and operational reliability, laying a technological foundation for the large-scale application of unmanned vessels. The system possesses multi-scale risk visualization capabilities and a dynamic learning rate adjustment mechanism, enabling highly autonomous navigation decisions and significantly reducing reliance on human intervention. The self-testing module can monitor hardware status in real time, ensuring continuous and stable system operation in complex environments, such as timely detection of sensor anomalies and activation of fault-tolerant mechanisms. Furthermore, the system can adapt to various maritime scenarios, including near-shore channels and open sea areas, supporting fully automatic navigation functions. Its intelligence level and environmental adaptability far exceed those of traditional navigation systems. In summary, this invention provides crucial support for the intelligent development of modern shipping technology by synergistically enhancing safety, economy, and autonomy. Attached Figure Description

[0023] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0024] Figure 1 This is a system block diagram of a ship intelligent navigation system based on spatiotemporal prediction risk field and deep reinforcement learning according to the present invention. Figure 2 This is a flowchart of the predictive risk field generation process of the present invention (detailed description of the complete process from data acquisition to risk field fusion). Detailed Implementation

[0025] The following will refer to the appendices in the embodiments of the present invention. Figure 1-2The technical solutions in the embodiments of the present invention are clearly and completely described herein. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0026] The present invention proposes a ship intelligent navigation system based on spatiotemporal prediction of risk fields and deep reinforcement learning. Through multi-source data acquisition, dynamic risk assessment, intelligent path optimization, and visual interaction, it enables ships to navigate safely and efficiently in complex environments. The implementation methods and steps of the system are described in detail below to ensure that the technical content is clear and operable.

[0027] Furthermore, the system consists of multiple modules working collaboratively, and the functions and implementation details of each module are as follows: The first step is data acquisition and preprocessing. The system acquires navigation-related data in real time through various sensors and communication devices, including ship position, meteorological parameters, ocean current speed, and the dynamics of surrounding vessels. Specific acquisition equipment includes a satellite remote sensing unit for receiving meteorological information such as wind speed (m / s), wave height (m), and air pressure (hPa); and an AIS receiving unit for acquiring the latitude and longitude coordinates, speed, and heading of surrounding vessels.

[0028] An underwater sonar unit measures water depth and ocean current speed; a GPS module provides the vessel's precise position, with a positioning error controlled within ±5 meters. This data is updated every 10 seconds and stored in a local database. For ease of subsequent calculations, all raw data undergoes standardization. Taking wind speed as an example, the standardization formula is:

[0029] ; in, For real-time wind speed, m / s, m / s, result The values ​​are dimensionless, ranging from 0 to 1. Wave height, air pressure, and ocean current speed are also standardized using a similar method, with ranges set to 0-10m, 900-1100hPa, and 0-5m / s, respectively. This standardization process ensures the uniformity of different types of data and provides reliable input for risk assessment.

[0030] Next comes risk assessment and dynamic mapping. The system divides the navigation area into grids (each grid is 100m × 100m), calculates the risk value for each grid point, and generates a predictive risk field.

[0031] The formula for calculating the risk value is: in, This represents the overall risk of a grid point, ranging from 0 to 1. These are weighting coefficients, reflecting the degree of influence of different factors, including static environmental factors. It consists of meteorological factors and water depth factors, namely .in, It is a weighted average of wind speed, wave height, and air pressure, calculated for example with weights of 0.5, 0.3, and 0.2.

[0032] , m represents the risk of insufficient water depth and is a predicted environmental factor. Generated using a Long Short-Term Memory (LSTM) network to predict environmental changes over the next 30 minutes. The LSTM input vector includes standardized wind speed, wave height, air pressure, and ocean current velocity. The hidden layer dimension is 64, and the initial state is set to a zero vector. The calculation formula is as follows: ; in, For the sigmoid function, and The prediction is trained using historical meteorological data (approximately 10,000 records) to ensure dimensionless accuracy. Dynamic ship density factor. The formula for calculating the density of surrounding vessels is as follows:

[0033] ; in, For the number of ships, In order to be with the first Distance between ships (unit: m) m, if there are no ships The risk field is updated every 5 minutes to ensure real-time accuracy.

[0034] Next is path planning and optimization. The system uses a deep reinforcement learning (DQN) algorithm to generate the optimal route based on the risk field. The algorithm's state space includes the ship's position and the risk field, the action space is heading adjustment, and the reward function is... ; in, The average risk of a flight route is calculated as follows: ; Output by DQN The flight path length (in meters) is calculated by accumulating the steps at waypoints. Energy consumption. , 100kg / m, parameter set Reference value 1500km J, the learning rate is dynamically adjusted as follows:

[0035] ; The algorithm is pre-trained 1000 times in a simulated environment to ensure convergence.

[0036] Next is visualization and user interaction. Risk areas and flight paths are displayed as 3D heatmaps using the Unity engine, with green indicating low risk and red indicating high risk. The display resolution is adjusted according to a formula: ; Users can zoom in or out on the screen and tap the grid to view specific risk values.

[0037] Furthermore, the risk warning and system self-checking functions further enhance security, when a certain grid point When the risk area is detected, the system will trigger an audible alarm and highlight the warning display, then calculate the risk area. ; .

[0038] For unmanned vessels, the system also includes a self-test function, monitoring the operating status of five components: GPS, AIS, sonar, processor, and power supply, and calculating reliability. ; Based on signal threshold judgment (e.g., 1 for GPS signal below -130dBm), if Switch to standby mode.

[0039] The system also includes a communication encryption and data security module, which uses a combination of symmetric and asymmetric encryption algorithms to encrypt navigation data transmitted between the ship and the shore-based control center or other ships, and evaluates the security level using the following formula: In the formula, The safety factor is dimensionless. This is the strength value of the symmetric encryption algorithm, which is dimensionless and ranges from [0,1]. This represents the strength value of the asymmetric encryption algorithm, which is dimensionless and ranges from [0,1]. , The weighting coefficients are dimensionless and satisfy the following conditions: ,when When this happens, the system automatically triggers a communication security warning and switches to a high-strength encryption mode to ensure the confidentiality and integrity of navigation data in a dynamic and risky environment.

[0040] Example: In this example, a cargo ship sails from Shanghai Port to Tokyo Port, a distance of approximately 2241km, and encounters the outer bands of a typhoon en route. The specific implementation process is as follows.

[0041] Step 1: Data Acquisition and Preprocessing: Time: 10:00 AM, April 7, 2025.

[0042] Meteorological data: wind speed 20 m / s, wave height 7 m, air pressure 980 hPa.

[0043] Ocean current: speed 1.5 m / s, direction .

[0044] AIS data: 4 ships within 10km ahead, at distances of 3000m, 4000m, 5000m, and 600m.

[0045] Standardized calculation: Wind speed: .

[0046] Wave height: .

[0047] Air pressure: .

[0048] .

[0049] Target grid: (31.25°N, 121.50°E) (Water depth 40m) , (LSTM predicts the approaching typhoon, based on 1-hour historical data) =0.0000038.

[0050] .

[0051] Step 2: Path planning and optimization; Initial route: 2241km straight line, speed 10m / s (Including typhoon areas, integral estimation) m J.

[0052] 0.634.

[0053] Optimized route: detour to 2445km. J.

[0054] new 0.698.

[0055] .

[0056] Step 4: Visualization and Early Warning; m.

[0057] .

[0058] Typhoon area No alarm was triggered. .

[0059] Step 5: System self-test; AIS signal is below the threshold . Continue running.

[0060] Results Analysis: This experiment successfully achieved safe navigation of a cargo ship under the influence of a typhoon using an intelligent path planning system. The initial voyage was 2241 km with an average risk index of 0.49. After optimization, the voyage increased to 2445 km (an increase of 9.1%), and the risk index significantly decreased to 0.30 (a decrease of 38.8%), while energy consumption increased by 9.1% accordingly. The optimized reward value increased from 0.634 to 0.698, indicating that the system achieved a balance between risk control and navigation efficiency.

[0061] The key achievements are reflected in three aspects: First, the risk control effect is outstanding. The optimized risk value of 0.30 is far below the alarm threshold of 0.8, and no risk alarm is triggered throughout the process. Second, the system operates stably and reliably. AIS signal detection shows that the system reliability index reaches 0.8, meeting the operational requirements. Finally, the algorithm is adaptable, with the learning rate automatically adjusted to 0.009325 based on the optimization effect. During the experiment, dynamic path planning was achieved by collecting real-time meteorological data (wind speed 20m / s, wave height 7m, air pressure 980hPa) and surrounding vessel information (4 vessels within a 10km range) and combining it with the LSTM typhoon prediction model.

[0062] Through the above embodiments of the present invention, the invention has been verified in actual shipping scenarios and is applicable to: intelligent navigation of commercial cargo ships and oil tankers; autonomous navigation of unmanned vessels; port vessel traffic management; and maritime emergency rescue. By constructing a spatiotemporal predictive risk field and deeply integrating it into path planning decisions, the present invention achieves truly intelligent ship navigation, possessing significant industrial application value and market prospects.

[0063] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A ship intelligent navigation system based on spatiotemporal prediction risk field and deep reinforcement learning, characterized in that, include: Multi-source data acquisition module: used to collect multi-dimensional data related to ship navigation in real time, including ship position, meteorological parameters, ocean current speed and surrounding ship dynamics; Dynamic risk mapping module: used to generate a predictive risk field based on the multi-source data and update the risk distribution of the navigation area in real time; the dynamic risk mapping module includes a prediction unit, which calculates the predicted environmental factors through a Long Short-Term Memory (LSTM) network. Its formula is: ； In the formula, It uses the sigmoid activation function, with an output range of [0, 1]. It is a dimensionless weight matrix. Let be the hidden state vector from the previous time step, initialized as a zero vector. The input vector consists of multi-source data at the current moment, including standardized wind speed, wave height, and ocean current velocity. The term is a dimensionless bias term; These are dimensionless values, used directly for calculations. ; Intelligent route planning module: used to generate the optimal route based on the predictive risk field using an adaptive deep reinforcement learning algorithm; Visualization and interaction module: used to present the risk field and recommended route in a three-dimensional dynamic form; The predictive risk field is generated based on the following risk value calculation formula: ； In the formula, coordinates At any moment The dimensionless risk value, As static environmental factors, To predict environmental factors, These are dynamic ship density factors, all of which are dimensionless. The weights are dimensionless and satisfy the following conditions: The static environmental factors It includes the sum of standardized meteorological factors and other static factors.

2. The ship intelligent navigation system based on spatiotemporal prediction risk field and deep reinforcement learning according to claim 1, characterized in that, The dynamic ship density factor Calculated using the following formula: ； In the formula, For the number of other ships in the area, For the first Ships and coordinates The distance, in meters. The standard deviation of the density distribution is expressed in meters. The exponential term is dimensionless. (Results) These are dimensionless values, used directly for calculations. .like 0, then .

3. The ship intelligent navigation system based on spatiotemporal prediction risk field and deep reinforcement learning according to claim 1, characterized in that, The intelligent path planning module employs an adaptive deep reinforcement learning algorithm, and its reward function is defined as: ； In the formula, The reward value is dimensionless. This represents the average risk value for the route. The length of the flight path is in meters. For reference, the maximum route length is in meters. The estimated energy consumption is expressed in joules. For reference to maximum energy consumption, the unit is joules. It is a dimensionless adjustment coefficient, and satisfies The Through formula Calculation, where Energy consumption coefficient For ship speed.

4. The ship intelligent navigation system based on spatiotemporal prediction risk field and deep reinforcement learning according to claim 3, characterized in that, The average risk value of the route The risk value is calculated by integrating the risk values ​​of each grid point along the path: ； In the formula, Path is the route output by the intelligent route planning module. Here is the arc length parameter along the path, in meters. The total length of the route. The risk value at the path point, the result It is a dimensionless value.

5. The ship intelligent navigation system based on spatiotemporal prediction risk field and deep reinforcement learning according to claim 1, characterized in that, The multi-source data acquisition module includes a satellite remote sensing unit, an AIS receiving unit, and an underwater sonar unit. The meteorological parameters are standardized to generate standardized meteorological factors using the following formula. : ； In the formula, This is the raw meteorological data. These are the minimum and maximum values ​​of the data, respectively, with units of and . Consistent The static environmental factor is a dimensionless value. ,in For other standardized static factors.

6. The ship intelligent navigation system based on spatiotemporal prediction risk field and deep reinforcement learning according to claim 1, characterized in that, The visualization interaction module supports multi-scale switching, and its risk field display resolution is adaptively adjusted according to the following formula: ； In the formula, To display the resolution, the unit is 1 / meter. These represent the maximum and minimum values ​​of the current risk field, respectively, both of which are dimensionless. This is a zoom level in meters. It is a dimensionless constant; the stated and based on The statistical value.

7. The ship intelligent navigation system based on spatiotemporal prediction risk field and deep reinforcement learning according to claim 1, characterized in that, The system also includes a risk warning module, which detects when the risk value of a certain grid point... Exceeding the preset threshold When the alarm signal is triggered, the area of ​​the risk exceeding the standard is calculated using the following formula: ； In the formula, This is an indicator function, dimensionless, with a value of 0 or 1. For dimensionless risk threshold, The area of ​​a single grid cell is expressed in square meters. The unit is square meters; the aforementioned Used to assess the scope of risks and guide route adjustments.

8. The ship intelligent navigation system based on spatiotemporal prediction risk field and deep reinforcement learning according to claim 1, characterized in that, The adaptive deep reinforcement learning algorithm introduces a dynamic learning rate during training, and its calculation formula is as follows: ； In the formula, The current dimensionless learning rate, The initial dimensionless learning rate, The dimensionless attenuation coefficient is The average risk value for the route is dimensionless; and Collaborative optimization of path planning.

9. The ship intelligent navigation system based on spatiotemporal prediction risk field and deep reinforcement learning according to claim 1, characterized in that, When the system is applied to unmanned vessels, it further includes a self-testing module for real-time monitoring of hardware status and for evaluating system reliability using the following formula: ； In the formula, It is a dimensionless reliability index. For the first Fault flags for individual hardware components, dimensionless, with values ​​of 0 (normal) or 1 (fault), determined based on a comparison of sensor signals with preset thresholds. The total number of components; When the value is below the threshold, prompt to adjust dependencies. Navigation decisions.

10. The ship intelligent navigation system based on spatiotemporal prediction risk field and deep reinforcement learning according to claim 1, characterized in that, The system also includes a communication encryption and data security module, which uses a combination of symmetric and asymmetric encryption algorithms to encrypt navigation data transmitted between the ship and the shore-based control center or other ships, and evaluates the security level using the following formula: ； In the formula, The safety factor is dimensionless. This is the strength value of the symmetric encryption algorithm, which is dimensionless and ranges from [0,1]. This represents the strength value of the asymmetric encryption algorithm, which is dimensionless and ranges from [0,1]. , The weighting coefficients are dimensionless and satisfy the following conditions: ,when When this happens, the system automatically triggers a communication security warning and switches to a high-strength encryption mode.

Images