LLM-based ship assisted driving method and system

Abstract

This invention relates to the field of intelligent ship technology and discloses a method and system for ship assisted driving based on LLM (Large Language Model). The method includes: collecting first state data of the target ship, second state data of the target object, environmental state data, and navigation rule data corresponding to the target ship's target journey; inputting the first state data, second state data, environmental state data, and navigation rule data into a pre-trained large language model to generate dynamic risk factors for the target ship. These dynamic risk factors indicate navigation risk objects, navigation risk factors, and corresponding risk levels for the target ship; and generating a set of control instructions for the target ship based on the dynamic risk factors. These control instructions are used to instruct the target ship on hazard avoidance guidance information and / or regulate the target ship's first state data and / or execute drive-away operations against navigation risk objects. Therefore, implementing this invention can improve the decision-making accuracy and flexibility of ship assisted driving.

Description

Technical Field

[0001] This invention relates to the field of intelligent ship technology, and in particular to a method and system for implementing ship assisted driving based on LLM. Background Technology

[0002] The navigation environment for ships is complex and ever-changing. To ensure navigation safety, modern ships are usually equipped with a variety of navigation aids, such as automatic identification systems, radar, electronic chart displays and information systems, to perceive the surrounding environment and obtain target information.

[0003] Traditional ship navigation assistance systems are mostly based on rule engines or traditional mathematical models, such as calculating the nearest encounter distance and shortest encounter time, to construct collision risk models. However, in practice, it has been found that such methods often provide static and isolated risk assessments for ship navigation, making it difficult to conduct dynamic and global risk extrapolation and coupled analysis for complex interactive scenarios involving multiple objectives and spanning time and space (such as multi-ship encounters and continuous avoidance). As a result, the decision-making accuracy and flexibility of ship navigation assistance are relatively low.

[0004] Therefore, it is particularly important to propose a technical solution to improve the accuracy and flexibility of ship assisted driving decisions. Summary of the Invention

[0005] This invention provides a method and system for ship assisted driving based on LLM, which can improve the accuracy and flexibility of ship assisted driving decisions.

[0006] To address the aforementioned technical problems, the first aspect of this invention discloses a method for implementing ship assisted driving based on LLM, the method comprising:

[0007] Collect first state data of the target vessel, second state data of the target object, environmental state data, and navigation rule data corresponding to the target vessel's target journey. The target object and the environmental state data respectively correspond to a preset range corresponding to the target vessel.

[0008] The first state data, the second state data, the environmental state data, and the navigation rule data are input into a pre-trained large language model to generate a dynamic risk factor for the target vessel. The dynamic risk factor is used to indicate the navigation risk object, navigation risk factor, and corresponding risk level for the target vessel.

[0009] Based on the dynamic risk factors, a set of control instructions for the target vessel is generated. The set of control instructions is used to instruct the target vessel on risk avoidance guidance information and / or to regulate the first state data of the target vessel and / or to perform a drive-away operation on the navigation risk object.

[0010] As an optional implementation, in a first aspect of the present invention, the step of inputting the first state data, the second state data, the environmental state data, and the navigation rule data into a pre-trained large language model to generate the dynamic risk factor of the target vessel includes:

[0011] Based on the first state data and the second state data, analyze the relative motion trajectory data between the target vessel and the target object;

[0012] Based on the relative motion trajectory data, a temporal collision risk coefficient between the target vessel and the target object is calculated. The temporal collision risk coefficient is used to represent the real-time collision risk level between the target vessel and the target object.

[0013] The environmental state data is analyzed, and the fluid disturbance data within the preset range is analyzed.

[0014] Based on the fluid disturbance data, attitude offset corrections are generated for the target vessel and the target object, respectively.

[0015] Based on the navigation rule data, the current trajectory avoidance weight of the target vessel is generated, and the trajectory avoidance weight is used to quantify the current degree of active avoidance responsibility of the target vessel.

[0016] The timing collision risk coefficient, the attitude offset correction amount, and the trajectory avoidance weight are input into a pre-trained large language model to generate the dynamic risk factor of the target ship.

[0017] As an optional implementation, in the first aspect of the present invention, generating a set of control instructions for the target vessel based on the dynamic risk factor includes:

[0018] Based on the dynamic risk factors, a risk propagation network is constructed for the target vessel, and the risk propagation network is used to indicate the risk propagation path and correlation of the dynamic risk factors to the target vessel.

[0019] Based on the risk propagation network, a set of control instructions for the target vessel is generated.

[0020] As an optional implementation, in the first aspect of the present invention, constructing a risk propagation network for the target vessel based on the dynamic risk factor includes:

[0021] The target vessel is identified as the target node for risk transmission, and the navigation risk object in the dynamic risk factor is identified as the source node for risk transmission.

[0022] Based on the navigation risk factors and their corresponding risk levels in the dynamic risk factors, the spatial gradient distribution of the connection edges from the source node to the target node is quantified to construct a risk propagation network for the target vessel.

[0023] As an optional implementation, in the first aspect of the invention, before generating the set of control instructions for the target vessel based on the risk propagation network, the method further includes:

[0024] Identify the associated nodes of the dynamic risk factors;

[0025] Construct the associated risk propagation path between the dynamic risk factors and the associated nodes;

[0026] Based on the associated risk propagation path and the risk propagation network, a multi-level transmission strength coefficient is calculated between the target vessel and the associated node via the dynamic risk factor. The multi-level transmission strength coefficient is used to represent the degree of risk transmission to the associated node during the process of the target vessel avoiding the dynamic risk factor.

[0027] The risk propagation network is updated based on the multi-level transmission strength coefficient.

[0028] As an optional implementation, in the first aspect of the invention, generating the set of control instructions for the target vessel based on the risk propagation network includes:

[0029] Based on the risk propagation network, the risk change rate of the navigation risk object is calculated, and the risk change rate is used to represent the change of the risk level of the navigation risk object over time.

[0030] Based on the risk change rate and the risk level, the navigation risk objects are prioritized to determine the processing order;

[0031] Based on the processing order and the navigation rule data, a corresponding risk avoidance strategy is retrieved from a preset rule base. The risk avoidance strategy is used to indicate the response method for a specific navigation risk object.

[0032] Based on the navigation rules data, the allocation of responsibility between the target vessel and the navigation risk object is analyzed, and the allocation of responsibility is used to indicate the degree of responsibility in a potential collision accident;

[0033] Based on the allocation of responsibilities, the risk avoidance strategy is adjusted to generate preliminary control instructions;

[0034] The conflict items between the initial control commands are analyzed, and the command execution sequence is planned based on the risk propagation network to generate a set of control commands for the target vessel.

[0035] As an optional implementation, in the first aspect of the present invention, the method further includes:

[0036] After generating the set of control instructions, the changing parameters of the dynamic risk factors and the risk propagation network are monitored in real time.

[0037] Based on the monitored changes in parameters, the effectiveness value of the set of control instructions is evaluated, and the effectiveness value is used to represent the degree of control exerted by the set of control instructions on the dynamic risk factor and the risk propagation network.

[0038] When the validity value is less than a preset threshold, the target adjustment item in the control instruction set is determined according to the change parameter; the control instruction set is updated according to the target adjustment item and the change parameter.

[0039] A second aspect of this invention discloses a ship-assisted driving system based on LLM, the system comprising:

[0040] The data acquisition module is used to acquire first state data of the target vessel, second state data of the target object, environmental state data, and navigation rule data corresponding to the target vessel's target journey. The target object and the environmental state data respectively correspond to a preset range corresponding to the target vessel.

[0041] The generation module is used to input the first state data, the second state data, the environmental state data, and the navigation rule data into a pre-trained large language model to generate the dynamic risk factor of the target ship. The dynamic risk factor is used to indicate the navigation risk object, navigation risk factor, and corresponding risk level for the target ship.

[0042] The generation module is further configured to generate a set of control instructions for the target vessel based on the dynamic risk factor. The set of control instructions is used to instruct the target vessel on risk avoidance guidance information and / or to regulate the first state data of the target vessel and / or to perform a drive-away operation on the navigation risk object.

[0043] As an optional implementation, in a second aspect of the present invention, the generation module inputs the first state data, the second state data, the environmental state data, and the navigation rule data into a pre-trained large language model to generate the dynamic risk factors of the target vessel. The specific method for generating these factors includes:

[0044] Based on the first state data and the second state data, analyze the relative motion trajectory data between the target vessel and the target object;

[0045] Based on the relative motion trajectory data, a temporal collision risk coefficient between the target vessel and the target object is calculated. The temporal collision risk coefficient is used to represent the real-time collision risk level between the target vessel and the target object.

[0046] The environmental state data is analyzed, and the fluid disturbance data within the preset range is analyzed.

[0047] Based on the fluid disturbance data, attitude offset corrections are generated for the target vessel and the target object, respectively.

[0048] Based on the navigation rule data, the current trajectory avoidance weight of the target vessel is generated, and the trajectory avoidance weight is used to quantify the current degree of active avoidance responsibility of the target vessel.

[0049] The timing collision risk coefficient, the attitude offset correction amount, and the trajectory avoidance weight are input into a pre-trained large language model to generate the dynamic risk factor of the target ship.

[0050] As an optional implementation, in the second aspect of the present invention, the specific method by which the generation module generates the set of control instructions for the target vessel based on the dynamic risk factor includes:

[0051] Based on the dynamic risk factors, a risk propagation network is constructed for the target vessel, and the risk propagation network is used to indicate the risk propagation path and correlation of the dynamic risk factors to the target vessel.

[0052] Based on the risk propagation network, a set of control instructions for the target vessel is generated.

[0053] As an optional implementation, in a second aspect of the invention, the specific method by which the generation module constructs a risk propagation network for the target vessel based on the dynamic risk factor includes:

[0054] The target vessel is identified as the target node for risk transmission, and the navigation risk object in the dynamic risk factor is identified as the source node for risk transmission.

[0055] Based on the navigation risk factors and their corresponding risk levels in the dynamic risk factors, the spatial gradient distribution of the connection edges from the source node to the target node is quantified to construct a risk propagation network for the target vessel.

[0056] As an optional implementation, in a second aspect of the invention, the system further includes:

[0057] The first determining module is used to determine the associated nodes of the dynamic risk factor before the generating module generates the set of control instructions for the target vessel based on the risk propagation network;

[0058] The construction module is used to construct the associated risk propagation path between the dynamic risk factor and the associated node;

[0059] The calculation module is used to calculate the multi-level transmission strength coefficient between the target vessel and the associated node via the dynamic risk factor based on the associated risk propagation path and the risk propagation network. The multi-level transmission strength coefficient is used to represent the degree of risk transmission to the associated node during the process of the target vessel avoiding the dynamic risk factor.

[0060] The first update module is used to update the risk propagation network based on the multi-level transmission strength coefficient.

[0061] As an optional implementation, in the second aspect of the present invention, the specific method by which the generation module generates the set of control instructions for the target vessel based on the risk propagation network includes:

[0062] Based on the risk propagation network, the risk change rate of the navigation risk object is calculated, and the risk change rate is used to represent the change of the risk level of the navigation risk object over time.

[0063] Based on the risk change rate and the risk level, the navigation risk objects are prioritized to determine the processing order;

[0064] Based on the processing order and the navigation rule data, a corresponding risk avoidance strategy is retrieved from a preset rule base. The risk avoidance strategy is used to indicate the response method for a specific navigation risk object.

[0065] Based on the navigation rules data, the allocation of responsibility between the target vessel and the navigation risk object is analyzed, and the allocation of responsibility is used to indicate the degree of responsibility in a potential collision accident;

[0066] Based on the allocation of responsibilities, the risk avoidance strategy is adjusted to generate preliminary control instructions;

[0067] The conflict items between the initial control commands are analyzed, and the command execution sequence is planned based on the risk propagation network to generate a set of control commands for the target vessel.

[0068] As an optional implementation, in a second aspect of the invention, the system further includes:

[0069] The monitoring module is used to monitor the changes in the dynamic risk factors and the risk propagation network parameters in real time after the generation module generates the set of control instructions.

[0070] An evaluation module is used to evaluate the effectiveness value of the set of control instructions based on the monitored changes in parameters. The effectiveness value is used to represent the degree of control exerted by the set of control instructions on the dynamic risk factor and the risk propagation network.

[0071] The second determining module is used to determine the target adjustment item in the set of control instructions based on the changing parameters when the effectiveness value is less than a preset threshold.

[0072] The second update module is used to update the set of control instructions based on the target adjustment item and the changed parameters.

[0073] A third aspect of the present invention discloses another LLM-based ship assisted driving system, the system comprising:

[0074] Memory containing executable program code;

[0075] A processor coupled to the memory;

[0076] The processor calls the executable program code stored in the memory to execute the LLM-based ship assisted driving implementation method disclosed in the first aspect of the present invention.

[0077] The fourth aspect of the present invention discloses a computer storage medium storing computer instructions, which, when invoked, are used to execute the LLM-based ship assisted driving implementation method disclosed in the first aspect of the present invention.

[0078] Compared with the prior art, the embodiments of the present invention have the following beneficial effects:

[0079] In this embodiment of the invention, first state data of the target vessel, second state data of the target object, environmental state data, and navigation rule data corresponding to the target vessel's target journey are collected. The target object and environmental state data correspond to preset ranges corresponding to the target vessel. The first state data, second state data, environmental state data, and navigation rule data are input into a pre-trained large language model to generate dynamic risk factors for the target vessel. The dynamic risk factors are used to indicate the navigation risk objects, navigation risk factors, and corresponding risk levels for the target vessel. Based on the dynamic risk factors, a set of control instructions for the target vessel is generated. The set of control instructions is used to instruct the target vessel on risk avoidance guidance information and / or control the first state data of the target vessel and / or perform driving-away operations on navigation risk objects. It is evident that implementing this invention enables multi-dimensional and comprehensive perception of the target vessel's navigation environment by collecting first-state data, second-state data, environmental state data, and navigation rule data. This improves the comprehensiveness and accuracy of environmental situational awareness, thus providing a solid data foundation for subsequent risk assessment. By inputting multi-source data into a pre-trained large language model to generate dynamic risk factors, the invention allows for natural language understanding, multimodal information fusion, and complex logical reasoning based on the large language model. This enhances the ability to analyze and quantify complex, fuzzy, and unstructured risk factors (such as rule clauses and environmental interference), enabling the transformation from raw data to structured data. The intelligent leap in risk knowledge generates dynamic risk factors that systematically analyze multiple factors, including comprehensive hydrological and meteorological conditions, waterway conditions, traffic density, and international and local navigation rules. This improves the comprehensiveness, accuracy, and flexibility of dynamic risk assessment for target vessels. Based on the dynamic risk factors, a set of control instructions is generated, including hazard avoidance guidance, state control, and expulsion operations. This provides diversified and hierarchical response strategies, ranging from information warnings to direct control and from internal adjustments to external interventions. This improves the accuracy, flexibility, and operability of ship assisted driving decisions, thereby facilitating the coverage of scenario requirements with different risk levels and types, enhancing the safety of ship assisted driving, and achieving closed-loop intelligent protection for ship safety. Attached Figure Description

[0080] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0081] Figure 1 This is a flowchart illustrating a method for implementing ship assisted driving based on LLM, as disclosed in an embodiment of the present invention.

[0082] Figure 2This is a schematic diagram of a ship assisted driving system based on LLM disclosed in an embodiment of the present invention;

[0083] Figure 3 This is a schematic diagram of another LLM-based ship assisted driving system disclosed in an embodiment of the present invention;

[0084] Figure 4 This is a schematic diagram of another LLM-based ship assisted driving system disclosed in an embodiment of the present invention. Detailed Implementation

[0085] To enable those skilled in the art to better understand the present invention, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0086] The terms "first," "second," etc., used in the specification, claims, and accompanying drawings of this invention are used to distinguish different objects, not to describe a specific order. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or end that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, products, or ends.

[0087] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of the invention. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0088] This invention discloses a method and system for ship assisted navigation based on LLM (Low-Low Motion Model). By collecting first-state data of the target ship, second-state data of the target object, environmental state data, and navigation rule data, it can achieve multi-dimensional and comprehensive perception of the target ship's navigation environment, improving the comprehensiveness and accuracy of environmental situational awareness, thus providing a solid data foundation for subsequent risk assessment. By inputting multi-source data into a pre-trained large language model to generate dynamic risk factors, it can perform natural language understanding, multimodal information fusion, and complex logical reasoning based on the large language model, improving the ability to analyze and quantify complex, fuzzy, and unstructured risk factors (such as rule clauses and environmental interference). The current intelligent leap from raw data to structured risk knowledge generates dynamic risk factors that systematically analyze multiple factors, including hydrological and meteorological conditions, waterway conditions, traffic density, and international and local navigation rules. This improves the comprehensiveness, accuracy, and flexibility of dynamic risk assessment for target vessels. Based on these dynamic risk factors, a set of control commands is generated, encompassing hazard avoidance guidance, state control, and disengagement operations. This provides diversified and hierarchical response strategies, ranging from information warnings to direct control, and from internal adjustments to external intervention. This enhances the accuracy, flexibility, and operability of ship assisted navigation decisions, thereby facilitating the coverage of scenarios with different risk levels and types, improving the safety of ship assisted navigation, and achieving closed-loop intelligent protection for ship safety. These will be explained in detail below.

[0089] Example 1

[0090] Please see Figure 1 , Figure 1 This is a flowchart illustrating a method for implementing ship assisted driving based on LLM, as disclosed in an embodiment of the present invention. Wherein, Figure 1 The described LLM-based ship assisted driving implementation method can be applied to ship assisted driving equipment, and also to intelligent devices associated with ship assisted driving equipment. These intelligent devices include, but are not limited to, one or more of cloud devices, edge computing devices, relay devices, base station devices, urban management devices, and intelligent connected devices; the embodiments of this invention do not limit the application to these devices. Figure 1 As shown, the LLM-based ship assisted driving implementation method may include the following operations:

[0091] 101. Collect the first state data of the target vessel, the second state data of the target object, the environmental state data, and the navigation rule data corresponding to the target vessel's target journey. The target object and environmental state data are respectively within the preset range corresponding to the target vessel.

[0092] In this embodiment of the invention, optionally, the first state data can be collected by the target ship's own navigation system (such as GPS, GNSS), inertial measurement unit, compass and log, etc.; optionally, it specifically includes the ship's real-time geographical location, heading, speed, heading, turning rate, etc., to accurately describe the target ship's own motion state.

[0093] The second state data may include dynamic and static information of target objects such as other ships, offshore fixed facilities (such as drilling platforms), floating objects, and flying objects within a preset range of the target ship, obtained through sensing devices such as automatic identification systems, radar, and optical / infrared camera systems configured in the target ship and / or its associated equipment. This information may include the ID, location, course, speed, length, width, and type of other ships, in order to construct a traffic situation map around the target ship and identify potential threats.

[0094] Environmental status data can be collected through equipment such as meteorological sensors, current meters, wave radars, and depth sensors (depth sounders) configured on and / or in the associated equipment of the target vessel. This information corresponds to the environmental information within the preset range mentioned above or differs from the environmental information collected in the second state data mentioned above. Specifically, it can include wind speed, wind direction, wave height, wave direction, current speed, current direction, visibility, water temperature, and water depth topography data to quantify the impact of the external environment on the vessel's maneuverability and collision risk. In special circumstances, the dynamic risk factors in the subsequent assessment may be more inclined towards extreme weather in the environment, specifically such as wind and waves.

[0095] Navigation rules data can refer to all rule texts related to the target itinerary (i.e., the planned route from the port of departure to the port of destination). This data is extracted and matched from electronic chart systems, wayfinding guides, and stored rule bases (such as the International Regulations for Preventing Collisions at Sea), navigation rules of the home country, navigation rules of the host country, and navigation rules of the destination country. The system dynamically loads and parses the corresponding rule clauses based on the vessel's current position (e.g., whether within a traffic separation scheme or in port waters) and the encounter situation (e.g., head-on collisions, crossings, overtaking). Its purpose is to provide a basis for risk assessment and decision-making, including legal rules, navigation rules, liability attribution, and customary practices.

[0096] 102. Input the first state data, second state data, environmental state data and navigation rule data into the pre-trained large language model to generate the dynamic risk factor of the target ship. The dynamic risk factor is used to indicate the navigation risk object, navigation risk factor and corresponding risk level for the target ship.

[0097] In this embodiment of the invention, optionally, the above four types of data (usually preprocessed and vectorized numerical or textual data) can be used as input prompts and fed into a pre-trained large language model (such as an LLM fine-tuned with maritime domain text and data). The LLM uses its powerful multimodal information fusion, natural language understanding and logical reasoning capabilities to comprehensively evaluate the input information and obtain the above dynamic risk factors.

[0098] Furthermore, optional, dynamic risk factors are the structured risk assessment results output by LLM. Their function is to condense the complex navigation environment into several key risk dimensions:

[0099] Navigation risk targets: clearly identify the most significant sources of threat, such as "bulk carrier A at 30 degrees to port and 3 nautical miles away" and "fishing ground floating facility B in the starboard direction."

[0100] Navigation risk factors: Analyze the specific causes of the risks, such as "crossing encounter with bulk carrier A", "affected by strong southwest current pressure, there is a tendency to press towards dangerous objects", and "poor visibility and limited visual lookout".

[0101] Risk level: Each risk object and risk factor is quantitatively or graded, such as "collision risk: high" or "stranding risk: medium". This is usually a confidence score or risk level based on model inference.

[0102] In this embodiment of the invention, as an optional implementation, the above-mentioned inputting the first state data, second state data, environmental state data, and navigation rule data into a pre-trained large language model to generate dynamic risk factors for the target vessel includes:

[0103] Based on the first state data and the second state data, analyze the relative motion trajectory data between the target vessel and the target object;

[0104] Based on the relative motion trajectory data, the temporal collision risk coefficient between the target vessel and the target object is calculated. The temporal collision risk coefficient is used to represent the real-time collision risk level between the target vessel and the target object.

[0105] Analyze environmental status data and analyze fluid disturbance data within a preset range;

[0106] Based on fluid disturbance data, generate attitude offset corrections for the target vessel and the target object respectively;

[0107] Based on navigation rules data, the current trajectory avoidance weight of the target vessel is generated. The trajectory avoidance weight is used to quantify the current degree of active avoidance responsibility of the target vessel.

[0108] The temporal collision risk coefficient, attitude deviation correction amount, and trajectory avoidance weight are input into the pre-trained large language model to generate the dynamic risk factors of the target ship.

[0109] In this embodiment of the invention, optionally, during the process of generating dynamic risk factors, the first state data, the second state data, the environmental state data, and the navigation rule data can be further processed, and the processed data such as the time-series collision risk coefficient, attitude offset correction amount, and trajectory avoidance weight are then input into the pre-trained large language model to improve the accuracy and comprehensiveness of the generation of dynamic risk factors for the target ship.

[0110] Further, optionally, for relative motion trajectory data analysis: based on the position and velocity vectors in the first and second state data, through coordinate transformation and kinematic calculations, the relative velocity vector and relative motion trajectory line between the target vessel and each target object are obtained. Its function is to predict the shortest encounter distance (DCPA) and shortest encounter time (TCPA) between the two parties within a future period, which are traditional core indicators of collision risk.

[0111] For calculating the temporal collision risk coefficient: based on the relative motion trajectory, not only is the current DCPA / TCPA calculated, but its trend over time (derivative) is also analyzed. Factors such as the type of target object (e.g., high risk for oil tankers, potential for sudden turning by fishing vessels) and size are comprehensively considered to calculate a comprehensive temporal collision risk coefficient that varies with the time series (e.g., a continuous value between 0 and 1). Its purpose is to quantify real-time collision risk more dynamically and comprehensively, rather than simply providing a static snapshot.

[0112] For fluid disturbance data analysis: Wave and current information in environmental state data is analyzed, and the disturbance forces and moments they generate on the ship are calculated using fluid dynamics models or empirical models. Its purpose is to assess the impact of external environmental forces on the ship's heading and course-keeping capabilities.

[0113] For attitude deviation correction generation: Based on the calculated fluid disturbance forces, the expected leeway and yaw of the vessel (and target object) are predicted. The attitude deviation correction is an estimate of this predicted deviation (such as yaw angle correction or position drift). Its function is to quantify the impact of environmental factors in advance as a correction to the actual trajectory of the vessel, making subsequent risk assessments closer to reality.

[0114] For trajectory avoidance weight generation: Navigation rule data is input into a rule parser. This parser determines the obligations the target vessel should fulfill according to the rules (e.g., whether to proceed directly or give way) based on the current encounter situation between the target vessel and the target object (determined by their relative motion trajectories). The trajectory avoidance weight is a quantified value (e.g., a higher weight for giving way and a lower weight for proceeding directly) used to represent the degree of the target vessel's responsibility to actively avoid the obstacle. Its purpose is to ensure that the system considers not only physical risks but also legal responsibilities when making decisions, prioritizing the fulfillment of avoidance obligations.

[0115] For the final risk factor generation: the calculated quantitative data of the three dimensions—temporal collision risk coefficient (physical risk), attitude deviation correction (environmental impact), and trajectory avoidance weight (rule responsibility)—are used as a set of refined, highly correlated feature vectors and input into the LLM. Based on this, the LLM, combined with its built-in maritime knowledge, performs final synthesis and decision-making to generate the aforementioned complete dynamic risk factors.

[0116] As can be seen, implementing this optional embodiment can extend from static spatial relationships to dynamic time series analysis by analyzing relative motion trajectories and calculating time-series collision risk coefficients, improving the foresight and real-time nature of collision risk prediction, thus facilitating earlier identification of potential threats. By parsing environmental data to generate fluid disturbance data and attitude offset corrections, abstract environmental impacts can be quantified into predictions and corrections of the ship's specific motion attitude, improving the accuracy of ship motion modeling and its robustness to environmental disturbances, thus enabling subsequent risk assessments to more closely reflect the ship's actual physical response. By generating trajectory avoidance weights based on navigation rule data, legal provisions and navigation practices can be transformed into quantifiable liability indicators, improving the compliance and rationality of system decisions, thus ensuring that generated instructions comply with maritime collision avoidance rules and avoiding new legal risks caused by improper decisions. Finally, inputting the above-mentioned refined physical risk, environmental impact, and rule liability feature vectors into the LLM provides the LLM with high-value, highly relevant decision-making basis, reducing the computational complexity and inference uncertainty of the LLM in processing raw data, thereby improving the efficiency and accuracy of the final dynamic risk factor generation.

[0117] 103. Based on dynamic risk factors, generate a set of control instructions for the target vessel. The set of control instructions is used to instruct the target vessel on risk avoidance guidance information and / or control the target vessel's first state data and / or execute drive-away operations against navigation risk objects.

[0118] In this embodiment of the invention, optionally, for risk avoidance guidance information, advisory information can be provided to crew members and the ship's associated responsible parties / units, such as highlighting risk targets on the display screen, generating and broadcasting voice warnings such as "Beware of the crossing vessel to the right front", and marking recommended safe routes on the electronic chart. Its function is to assist the driver in making decisions.

[0119] For regulating the first state data, the target vessel can be automatically intervened in driving after further authorization or by direct control, with the aim of automatically changing the vessel's own state to avoid risks;

[0120] For driving away operations: For non-ship-related risk objects (such as small boats, floating objects, or flying objects that are too close) or vessels that do not act in accordance with the rules, alarms or driving away signals are issued through communication systems (such as VHF broadcasts, VHF digital selective calling (DSC)) or specific equipment (such as sonar pulse transmitters, strong light, lasers, and explosive water cannons). The purpose is to actively eliminate or mitigate external threats.

[0121] As can be seen, implementing the embodiments of the present invention enables multi-dimensional and comprehensive perception of the navigation environment of the target vessel by collecting first-state data, second-state data, environmental state data, and navigation rule data. This improves the comprehensiveness and accuracy of environmental situational awareness, thus providing a solid data foundation for subsequent risk assessment. By inputting multi-source data into a pre-trained large language model to generate dynamic risk factors, the invention enables natural language understanding, multimodal information fusion, and complex logical reasoning based on the large language model. This improves the ability to analyze and quantify complex, fuzzy, and unstructured risk factors (such as rule clauses and environmental interference), and achieves the transformation from raw data to structured data. By intelligently transforming risk knowledge, dynamic risk factors are generated through a comprehensive analysis of multiple factors, including hydrological and meteorological conditions, waterway conditions, traffic density, and international and local navigation rules. This improves the comprehensiveness, accuracy, and flexibility of dynamic risk assessment for target vessels. Based on these dynamic risk factors, a set of control instructions is generated, encompassing hazard avoidance guidance, state control, and dispersal operations. This provides diversified and hierarchical response strategies, ranging from information warnings to direct control and from internal adjustments to external intervention. This enhances the accuracy, flexibility, and operability of ship assisted driving decisions, thereby facilitating the coverage of scenario requirements with different risk levels and types, improving the safety of ship assisted driving, and achieving closed-loop intelligent protection for ship safety.

[0122] In this embodiment of the invention, as another optional implementation, the above-mentioned generation of a set of control instructions for the target vessel based on dynamic risk factors includes:

[0123] Based on dynamic risk factors, a risk propagation network is constructed for the target vessel. The risk propagation network is used to indicate the risk propagation path and correlation of dynamic risk factors to the target vessel.

[0124] Based on the risk propagation network, a set of control instructions for the target vessel is generated.

[0125] In this embodiment of the invention, optionally, for constructing a risk propagation network: this step treats dynamic risk factors as multiple risk sources in a complex system. The system models the relationship between these risk sources and the target vessel in the form of a risk propagation network. This network can be a graph structure, which visualizes and quantifies the causal relationships, correlations, and risk transmission paths between different risk factors. For example, severe weather (node ​​A) leads to difficulty in vessel maneuvering (node ​​B), thereby increasing the risk of collision with another vessel (node ​​C). Constructing this network is the foundation for systematic and global decision-making.

[0126] Regarding the generation of control instructions: Based on the constructed risk propagation network, the system can analyze which node is the key risk point, block which propagation path is most effective to form a chain effect, or form an avoidance sequence, etc. By evaluating the impact of different control measures on each node and edge in the network, a set of control instructions that can optimize the reduction of global risk is generated, rather than just targeting a single risk.

[0127] As can be seen, implementing this optional embodiment, by constructing a risk propagation network based on dynamic risk factors, can organize discrete risk points into a related and hierarchical risk topology, improving the systematic analysis capability of the coupling, transmission, and superposition relationships between risks. This facilitates understanding the overall risk situation from a global rather than isolated perspective. Generating a set of control instructions based on the risk propagation network enables decisions to be based on a deep understanding of the risk system structure, improving the systematicness and overall optimization of control strategies. This facilitates the generation of optimal decisions that not only address current direct risks but also consider the interrelationships between risks, achieving a combination of "treating the symptoms" and "addressing the root cause."

[0128] In this optional embodiment, as an optional implementation method, the above-mentioned construction of a risk propagation network for the target vessel based on dynamic risk factors includes:

[0129] The target vessel is identified as the target node for risk transmission, and the navigation risk object in the dynamic risk factors is identified as the source node for risk transmission.

[0130] Based on the navigation risk factors and their corresponding risk levels in the dynamic risk factors, the spatial gradient distribution of the connection edges from the source node to the target node is quantified to construct a risk propagation network for the target vessel.

[0131] In this embodiment of the invention, optionally, the target node can always be the "target vessel" itself, because the ultimate goal of all risk analysis is to protect the safety of the vessel itself.

[0132] For a source node, it can be any navigation risk object identified in the dynamic risk factors (such as ship A, ship B, obstacle C, wave risk range D, wave height E, typhoon risk range F, etc.). It should be noted that for natural conditions such as waves, wave height, and typhoons, the risk needs to be assessed in conjunction with the ship's own load capacity and resistance capabilities, based on its risk range. For example, not the entire coverage area of ​​a typhoon poses a risk to the target ship (e.g., capsizing, breakage), but only a portion of the typhoon's wind circle will bring these risks. Each source node represents an independent external risk source; quantification of the spatial gradient distribution of the connection edges:

[0133] For each "connection edge" (source node -> target node), it is necessary to quantify the strength and direction of its risk transmission, which is achieved by analyzing navigation risk factors and their corresponding risk levels;

[0134] Spatial gradient distribution is a key concept, describing the rate of change of risk in space. For example, the closer the source node (other ships) is to the target node (this ship), the more the risk level increases, not linearly, but sharply (the gradient becomes larger); a risk factor with "high relative speed" will produce a larger risk gradient than one with "low relative speed".

[0135] By calculating the rate of change of risk level relative to spatial location (distance, orientation) and combining this with the type of risk factor (collision, stranding, etc.), the spatial gradient distribution of each connecting edge is quantified. The result defines the "weight" or "strength" of each edge in the network, thus completing the construction of the entire risk propagation network. A connecting edge with a steep gradient distribution signifies high risk and urgency.

[0136] As can be seen, by determining the target node and the source node, this optional embodiment can clearly define the starting point and the ending point of risk transmission, improving the standardization and clarity of the risk propagation model construction. By quantifying the spatial gradient distribution of the connection edges from the source node to the target node, it is possible to use mathematical methods to accurately describe the degree of change of risk with spatial distance, orientation and other factors, improving the scientificity and accuracy of the connection edge weights in the risk propagation network. This is beneficial for distinguishing the urgency and severity of different risks, and thus provides the possibility for subsequent quantitative decision-making based on network structure (such as finding the most critical path).

[0137] In this optional embodiment, as another optional implementation, the method further includes, before generating the set of control instructions for the target vessel based on the risk propagation network:

[0138] Identify the relevant nodes of dynamic risk factors;

[0139] Construct a dynamic risk propagation path between risk factors and related nodes;

[0140] Based on the associated risk propagation path and risk propagation network, the multi-level transmission strength coefficient between the target vessel and the associated nodes via the dynamic risk factors is calculated. The multi-level transmission strength coefficient is used to represent the degree of risk transmission to the associated nodes during the process of the target vessel avoiding the dynamic risk factors.

[0141] The risk propagation network is updated based on the multi-level transmission strength coefficient.

[0142] In this embodiment of the invention, optionally, for determining associated nodes: after identifying the direct risk source node (source node), the system will further reason. Associated nodes refer to those targets that currently pose a low risk or do not constitute a direct threat, but may become high-risk targets due to evasive maneuvers taken by the vessel. For example, to avoid "source node A" (a fishing boat) on its starboard side, the vessel might turn left, thus heading towards another originally safe merchant ship "B". In this case, ship B is an associated node.

[0143] To construct a related risk propagation path: The system simulates the aforementioned avoidance maneuver (such as "hard to port"), predicts the ship's new trajectory, and calculates whether this new trajectory will create a new risk path between the ship and the associated node (ship B). This path from "source node A" to "target node (the ship)" and then to "associated node B" is the related risk propagation path;

[0144] For calculating the multi-level transmission strength coefficient: This coefficient quantifies the magnitude of the indirect risk (to B) caused by the ship's avoidance maneuver while eliminating the original risk (to A). It is a numerical value based on simulation calculations, taking into account the DCPA / TCPA of the new trajectory, the attributes of associated nodes, etc. Its role is to provide an assessment of "side effects" for decision-making;

[0145] For updating the risk propagation network: newly identified associated nodes and associated risk propagation paths, along with the calculated multi-level transmission strength coefficients, are added as new edge weights to the existing risk propagation network. Thus, the network is upgraded from a "first-level network" that only describes direct risks to a "multi-level network" that can describe risk chains, providing support for generating more comprehensive regulatory instructions.

[0146] As can be seen, by identifying associated nodes and constructing associated risk propagation paths, this optional embodiment can proactively predict secondary and derivative risks that may be caused by actions taken to avoid current risks, thereby improving the system's ability to anticipate the "side effects" of decisions and thus helping to avoid the decision-making trap of "robbing Peter to pay Paul." By calculating multi-level transmission strength coefficients and updating the risk propagation network, the quantitative assessment results of indirect risks can be fed back to the global risk model, improving the completeness and dynamism of the risk propagation network. This helps to ensure that the generated control instructions are solutions after a global risk and benefit trade-off, achieving true safety and risk avoidance.

[0147] In this optional embodiment, as yet another optional implementation, the above-mentioned generation of a set of control instructions for the target vessel based on the risk propagation network includes:

[0148] Based on the risk propagation network, the risk change rate of the navigation risk object is calculated. The risk change rate is used to represent the change of the risk level of the navigation risk object over time.

[0149] Based on the rate of change of risk and the degree of risk, the navigation risk objects are prioritized to determine the order of treatment;

[0150] Based on the processing order and navigation rule data, the corresponding risk avoidance strategy is retrieved from the preset rule base. The risk avoidance strategy is used to indicate the response to a specific navigation risk object.

[0151] Based on navigation rules data, the allocation of responsibility between the target vessel and the navigation risk object is analyzed. The allocation of responsibility is used to indicate the degree of responsibility in a potential collision accident.

[0152] Based on the allocation of responsibilities, adjust the risk avoidance strategy to generate initial control instructions;

[0153] Analyze the conflicts between initial control commands and, based on the risk propagation network, plan the command execution sequence to generate a set of control commands for the target vessel.

[0154] In this embodiment of the invention, optionally, for calculating the risk change rate: the system continuously monitors the risk level value of each source node in the risk propagation network. The risk change rate can be the first derivative of this risk value with time (calculated by differentiating risk values ​​over consecutive time steps). Its function is to identify which threats are "accelerating" to become dangerous, and addressing these imminent threats should have the highest priority.

[0155] For priority ranking: all risk objects are comprehensively ranked based on two dimensions: risk level (how dangerous is it currently) and risk change rate (how quickly it changes), generating a processing order list. For example, objects with "high risk level and high risk change rate" are ranked first;

[0156] For retrieving risk avoidance strategies: The system maintains a preset rule base, which stores various standardized risk avoidance strategies. These strategies are a set of "IF (situation / risk type) THEN (suggested action)". The system matches strategies from the rule base to each high-risk object sequentially according to the processing order. For example, the strategy retrieved for "giving way vessel in a starboard crossing encounter" might be "turn to starboard and pass behind the other vessel" or "slow down and let the other vessel pass first";

[0157] For analyzing liability allocation: a rule parser can be invoked to conduct in-depth analysis of the encounter situation between the vessel and each risky object, based on rules such as the International Regulations for Preventing Collisions at Sea (ICP-5), determining who bears primary responsibility and who bears secondary responsibility. The liability allocation result will influence strategy selection; the system will tend to generate instructions that require the vessel to fulfill its avoidance obligations, thus avoiding liability for accidents due to improper actions.

[0158] For generating initial control instructions and resolving conflicts: After adjusting the retrieved strategies according to the allocation of responsibilities, initial control instructions are generated. However, these instructions may conflict (e.g., one strategy requires "acceleration," while another requires "deceleration"). The system will resolve these conflicts and, based on the risk propagation network, assess the consequences of executing different instructions, making trade-offs and choices.

[0159] Regarding the timing of command execution: Ultimately, the system doesn't simply output a set of commands, but rather plans a timeline-based action plan, i.e., the command execution sequence. For example: "At time T0: sound a short blast, begin turning 5 degrees to the right; at T0+30 seconds: increase the turn to 15 degrees; at T0+60 seconds: confirm the avoidance effect, and if good, maintain direction and speed." This ensures the orderliness and safety of the action, generating the final set of control commands.

[0160] As can be seen, implementing this optional embodiment, by calculating the rate of change of risk and prioritizing it, can identify which risks are rapidly deteriorating, improving the system's responsiveness to the dynamic evolution of risks. This helps ensure that limited resources are prioritized for dealing with the most urgent and dangerous threats, achieving optimal allocation of decision-making resources. By retrieving risk avoidance strategies from a preset rule base and adjusting them in conjunction with responsibility allocation, it ensures that the generated preliminary instructions conform to both best practice paradigms and legal division of rights and responsibilities, improving the standardization and compliance of decision-making, thus facilitating the generation of safe and reasonable instructions. By parsing instruction conflict items and planning the instruction execution sequence, it can coordinate contradictions between different instructions at the operational level, improving the internal consistency and timing coordination of the final instruction set. This helps ensure that a series of operations can be executed smoothly and orderly, reducing the risk of operational chaos caused by instruction conflicts, and achieving precise and controllable ship control.

[0161] In an optional embodiment, the method further includes:

[0162] After generating the set of control instructions, the changing parameters of dynamic risk factors and risk propagation networks are monitored in real time.

[0163] Based on the monitored changes in parameters, the effectiveness value of the set of control instructions is evaluated. The effectiveness value is used to indicate the degree to which the set of control instructions regulates dynamic risk factors and risk propagation networks.

[0164] When the effectiveness value is less than the preset threshold, the target adjustment item in the control instruction set is determined based on the changed parameters; the control instruction set is updated based on the target adjustment item and the changed parameters.

[0165] In this embodiment of the invention, optionally, for real-time monitoring of changing parameters: while executing the set of control instructions, the system continuously monitors the changing parameters of dynamic risk factors and risk propagation networks, which represent the system state. These parameters include, but are not limited to: risk level value, risk change rate, number of network nodes, and connection edge weights;

[0166] For evaluating effectiveness: The system defines an effectiveness metric to measure whether the currently executed set of instructions has achieved the expected effect (i.e., reduced risk). This value can be calculated by comparing the decrease in risk parameters after instruction execution with the expected decrease, or simply by monitoring whether the risk level continues to rise;

[0167] Regarding judgment and command updates: When the effectiveness value falls below a preset threshold (meaning the command is ineffective or the situation is worsening), a decision loop is triggered. The system analyzes which link has gone wrong based on the latest changing parameters (identifying the target adjustment item, such as "insufficient steering angle" or "the target vessel not acting as expected"), and then quickly recalculates or fine-tunes the original set of control commands to generate an updated set of commands that is more adapted to the current rapidly changing environment, thereby achieving continuous and adaptive assisted driving.

[0168] As can be seen, by implementing this optional embodiment, a real-time feedback loop for the effectiveness of decision execution can be constructed by monitoring changing parameters in real time and evaluating the effectiveness value of the control command set. This improves the system's adaptability to dynamic environments and its closed-loop control capability for decision-making, thereby facilitating the timely detection of decision failures or sudden changes in circumstances. When the effectiveness value is insufficient, the target adjustment item can be determined and the control command set can be updated, enabling feedback-based online strategy optimization and rapid iteration. This improves the robustness and reliability of the entire assisted driving system, thereby helping to cope with sudden uncertainties in complex sea areas and achieving continuous and stable intelligent assisted driving.

[0169] Example 2

[0170] Please see Figure 2 , Figure 2 This is a schematic diagram of a ship assisted driving system based on LLM, as disclosed in an embodiment of the present invention. This LLM-based ship assisted driving system can be applied to ship assisted driving equipment, and also to intelligent devices associated with the ship assisted driving equipment. These intelligent devices include, but are not limited to, one or more of cloud devices, edge computing devices, relay devices, base station devices, urban management devices, and intelligent connected devices; the embodiments of the present invention do not impose limitations on this. Figure 2 As shown, the LLM-based ship assisted driving system may include:

[0171] The acquisition module 201 is used to acquire the first state data of the target vessel, the second state data of the target object, the environmental state data, and the navigation rule data corresponding to the target vessel's target journey. The target object and the environmental state data are respectively within the preset range corresponding to the target vessel.

[0172] The generation module 202 is used to input the first state data, the second state data, the environmental state data and the navigation rule data into the pre-trained large language model to generate the dynamic risk factor of the target ship. The dynamic risk factor is used to indicate the navigation risk object, navigation risk factor and corresponding risk level for the target ship.

[0173] The generation module 202 is also used to generate a set of control instructions for the target vessel based on dynamic risk factors. The set of control instructions is used to instruct the target vessel on risk avoidance guidance information and / or control the target vessel's first state data and / or perform driving-away operations on navigation risk objects.

[0174] As can be seen, implementing the embodiments of the present invention enables multi-dimensional and comprehensive perception of the navigation environment of the target vessel by collecting first-state data, second-state data, environmental state data, and navigation rule data. This improves the comprehensiveness and accuracy of environmental situational awareness, thus providing a solid data foundation for subsequent risk assessment. By inputting multi-source data into a pre-trained large language model to generate dynamic risk factors, the invention enables natural language understanding, multimodal information fusion, and complex logical reasoning based on the large language model. This improves the ability to analyze and quantify complex, fuzzy, and unstructured risk factors (such as rule clauses and environmental interference), and achieves the transformation from raw data to structured data. By intelligently transforming risk knowledge, dynamic risk factors are generated through a comprehensive analysis of multiple factors, including hydrological and meteorological conditions, waterway conditions, traffic density, and international and local navigation rules. This improves the comprehensiveness, accuracy, and flexibility of dynamic risk assessment for target vessels. Based on these dynamic risk factors, a set of control instructions is generated, encompassing hazard avoidance guidance, state control, and dispersal operations. This provides diversified and hierarchical response strategies, ranging from information warnings to direct control and from internal adjustments to external intervention. This enhances the accuracy, flexibility, and operability of ship assisted driving decisions, thereby facilitating the coverage of scenario requirements with different risk levels and types, improving the safety of ship assisted driving, and achieving closed-loop intelligent protection for ship safety.

[0175] In this embodiment of the invention, as an optional implementation, the generation module 202 inputs the first state data, the second state data, the environmental state data, and the navigation rules data into a pre-trained large language model to generate the dynamic risk factors of the target ship. The specific methods for generating these factors include:

[0176] Based on the first state data and the second state data, analyze the relative motion trajectory data between the target vessel and the target object;

[0177] Based on the relative motion trajectory data, the temporal collision risk coefficient between the target vessel and the target object is calculated. The temporal collision risk coefficient is used to represent the real-time collision risk level between the target vessel and the target object.

[0178] Analyze environmental status data and analyze fluid disturbance data within a preset range;

[0179] Based on fluid disturbance data, generate attitude offset corrections for the target vessel and the target object respectively;

[0180] Based on navigation rules data, the current trajectory avoidance weight of the target vessel is generated. The trajectory avoidance weight is used to quantify the current degree of active avoidance responsibility of the target vessel.

[0181] The temporal collision risk coefficient, attitude deviation correction amount, and trajectory avoidance weight are input into the pre-trained large language model to generate the dynamic risk factors of the target ship.

[0182] As can be seen, implementing this optional embodiment can extend from static spatial relationships to dynamic time series analysis by analyzing relative motion trajectories and calculating time-series collision risk coefficients, improving the foresight and real-time nature of collision risk prediction, thus facilitating earlier identification of potential threats. By parsing environmental data to generate fluid disturbance data and attitude offset corrections, abstract environmental impacts can be quantified into predictions and corrections of the ship's specific motion attitude, improving the accuracy of ship motion modeling and its robustness to environmental disturbances, thus enabling subsequent risk assessments to more closely reflect the ship's actual physical response. By generating trajectory avoidance weights based on navigation rule data, legal provisions and navigation practices can be transformed into quantifiable liability indicators, improving the compliance and rationality of system decisions, thus ensuring that generated instructions comply with maritime collision avoidance rules and avoiding new legal risks caused by improper decisions. Finally, inputting the above-mentioned refined physical risk, environmental impact, and rule liability feature vectors into the LLM provides the LLM with high-value, highly relevant decision-making basis, reducing the computational complexity and inference uncertainty of the LLM in processing raw data, thereby improving the efficiency and accuracy of the final dynamic risk factor generation.

[0183] In this embodiment of the invention, as another optional implementation, the specific method by which the generation module 202 generates the set of control instructions for the target vessel based on dynamic risk factors includes:

[0184] Based on dynamic risk factors, a risk propagation network is constructed for the target vessel. The risk propagation network is used to indicate the risk propagation path and correlation of dynamic risk factors to the target vessel.

[0185] Based on the risk propagation network, a set of control instructions for the target vessel is generated.

[0186] As can be seen, implementing this optional embodiment, by constructing a risk propagation network based on dynamic risk factors, can organize discrete risk points into a related and hierarchical risk topology, improving the systematic analysis capability of the coupling, transmission, and superposition relationships between risks. This facilitates understanding the overall risk situation from a global rather than isolated perspective. Generating a set of control instructions based on the risk propagation network enables decisions to be based on a deep understanding of the risk system structure, improving the systematicness and overall optimization of control strategies. This facilitates the generation of optimal decisions that not only address current direct risks but also consider the interrelationships between risks, achieving a combination of "treating the symptoms" and "addressing the root cause."

[0187] In this optional embodiment, as an optional implementation method, the specific way in which the generation module 202 constructs a risk propagation network for the target vessel based on dynamic risk factors includes:

[0188] The target vessel is identified as the target node for risk transmission, and the navigation risk object in the dynamic risk factors is identified as the source node for risk transmission.

[0189] Based on the navigation risk factors and their corresponding risk levels in the dynamic risk factors, the spatial gradient distribution of the connection edges from the source node to the target node is quantified to construct a risk propagation network for the target vessel.

[0190] As can be seen, by determining the target node and the source node, this optional embodiment can clearly define the starting point and the ending point of risk transmission, improving the standardization and clarity of the risk propagation model construction. By quantifying the spatial gradient distribution of the connection edges from the source node to the target node, it is possible to use mathematical methods to accurately describe the degree of change of risk with spatial distance, orientation and other factors, improving the scientificity and accuracy of the connection edge weights in the risk propagation network. This is beneficial for distinguishing the urgency and severity of different risks, and thus provides the possibility for subsequent quantitative decision-making based on network structure (such as finding the most critical path).

[0191] In this optional embodiment, as another alternative implementation, such as Figure 3 As shown, the system also includes:

[0192] The first determining module 203 is used to determine the associated nodes of dynamic risk factors before the generating module 202 generates the set of control instructions for the target vessel based on the risk propagation network.

[0193] Module 204 is used to construct the associated risk propagation path between dynamic risk factors and related nodes;

[0194] The calculation module 205 is used to calculate the multi-level transmission strength coefficient between the target vessel and the associated nodes via the dynamic risk factor based on the associated risk transmission path and risk transmission network. The multi-level transmission strength coefficient is used to represent the degree of risk transmission to the associated nodes during the process of the target vessel avoiding the dynamic risk factor.

[0195] The first update module 206 is used to update the risk propagation network based on the multi-level transmission strength coefficient.

[0196] As can be seen, by identifying associated nodes and constructing associated risk propagation paths, this optional embodiment can proactively predict secondary and derivative risks that may be caused by actions taken to avoid current risks, thereby improving the system's ability to anticipate the "side effects" of decisions and thus helping to avoid the decision-making trap of "robbing Peter to pay Paul." By calculating multi-level transmission strength coefficients and updating the risk propagation network, the quantitative assessment results of indirect risks can be fed back to the global risk model, improving the completeness and dynamism of the risk propagation network. This helps to ensure that the generated control instructions are solutions after a global risk and benefit trade-off, achieving true safety and risk avoidance.

[0197] In this optional embodiment, as yet another optional implementation, the specific method by which the generation module 202 generates the set of control instructions for the target vessel based on the risk propagation network includes:

[0198] Based on the risk propagation network, the risk change rate of the navigation risk object is calculated. The risk change rate is used to represent the change of the risk level of the navigation risk object over time.

[0199] Based on the rate of change of risk and the degree of risk, the navigation risk objects are prioritized to determine the order of treatment;

[0200] Based on the processing order and navigation rule data, the corresponding risk avoidance strategy is retrieved from the preset rule base. The risk avoidance strategy is used to indicate the response to a specific navigation risk object.

[0201] Based on navigation rules data, the allocation of responsibility between the target vessel and the navigation risk object is analyzed. The allocation of responsibility is used to indicate the degree of responsibility in a potential collision accident.

[0202] Based on the allocation of responsibilities, adjust the risk avoidance strategy to generate initial control instructions;

[0203] Analyze the conflicts between initial control commands and, based on the risk propagation network, plan the command execution sequence to generate a set of control commands for the target vessel.

[0204] As can be seen, implementing this optional embodiment, by calculating the rate of change of risk and prioritizing it, can identify which risks are rapidly deteriorating, improving the system's responsiveness to the dynamic evolution of risks. This helps ensure that limited resources are prioritized for dealing with the most urgent and dangerous threats, achieving optimal allocation of decision-making resources. By retrieving risk avoidance strategies from a preset rule base and adjusting them in conjunction with responsibility allocation, it ensures that the generated preliminary instructions conform to both best practice paradigms and legal division of rights and responsibilities, improving the standardization and compliance of decision-making, thus facilitating the generation of safe and reasonable instructions. By parsing instruction conflict items and planning the instruction execution sequence, it can coordinate contradictions between different instructions at the operational level, improving the internal consistency and timing coordination of the final instruction set. This helps ensure that a series of operations can be executed smoothly and orderly, reducing the risk of operational chaos caused by instruction conflicts, and achieving precise and controllable ship control.

[0205] In an optional embodiment, such as Figure 3 As shown, the system also includes:

[0206] The monitoring module 207 is used to monitor the changing parameters of dynamic risk factors and risk propagation networks in real time after the generation module 202 generates the set of control instructions.

[0207] The evaluation module 208 is used to evaluate the effectiveness value of the set of control instructions based on the monitored changes in parameters. The effectiveness value is used to indicate the degree of control of the set of control instructions on dynamic risk factors and risk propagation networks.

[0208] The second determining module 209 is used to determine the target adjustment item in the control instruction set based on the changing parameters when the effectiveness value is less than the preset threshold.

[0209] The second update module 210 is used to update the control instruction set based on the target adjustment items and changed parameters.

[0210] As can be seen, by implementing this optional embodiment, a real-time feedback loop for the effectiveness of decision execution can be constructed by monitoring changing parameters in real time and evaluating the effectiveness value of the control command set. This improves the system's adaptability to dynamic environments and its closed-loop control capability for decision-making, thereby facilitating the timely detection of decision failures or sudden changes in circumstances. When the effectiveness value is insufficient, the target adjustment item can be determined and the control command set can be updated, enabling feedback-based online strategy optimization and rapid iteration. This improves the robustness and reliability of the entire assisted driving system, thereby helping to cope with sudden uncertainties in complex sea areas and achieving continuous and stable intelligent assisted driving.

[0211] Example 3

[0212] Please see Figure 4 , Figure 4This is a schematic diagram of another LLM-based ship assisted driving system disclosed in this invention. This LLM-based ship assisted driving system can be applied to ship assisted driving equipment, and also to intelligent devices associated with the ship assisted driving equipment. These intelligent devices include, but are not limited to, one or more of cloud devices, edge computing devices, relay devices, base station devices, urban management devices, and intelligent connected devices; this invention does not limit the scope of these devices. Figure 4 As shown, the LLM-based ship assisted driving system may include:

[0213] Memory 301 that stores executable program code.

[0214] Processor 302 coupled to memory 301.

[0215] The processor 302 calls the executable program code stored in the memory 301 to execute the steps in the LLM-based ship assisted driving implementation method described in Embodiment 1 of the present invention.

[0216] Example 4

[0217] This invention discloses a computer storage medium storing computer instructions. When these computer instructions are invoked, they are used to execute the steps in the LLM-based ship assisted driving implementation method described in Embodiment 1 of this invention.

[0218] Example 5

[0219] This invention discloses a computer program product, which includes a non-transitory computer storage medium storing a computer program, and the computer program is operable to cause a computer to perform the steps in the LLM-based ship assisted driving implementation method described in Embodiment 1.

[0220] The system embodiments described above are merely illustrative. The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical modules; that is, they may be located in one place or distributed across multiple network modules. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0221] Through the detailed description of the above embodiments, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, including read-only memory (ROM), random access memory (RAM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), one-time programmable read-only memory (OTPROM), electrically-Erasable Programmable Read-Only Memory (EEPROM), compact disc read-only memory (CD-ROM) or other optical disc storage, disk storage, magnetic tape storage, or any other computer-readable medium that can be used to carry or store data.

[0222] Finally, it should be noted that the LLM-based ship assisted driving implementation method and system disclosed in the embodiments of the present invention are merely preferred embodiments of the present invention, and are only used to illustrate the technical solutions of the present invention, not to limit it; 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. An LLM-based ship auxiliary driving implementation method, characterized in that, The method comprises: Collecting first state data of a target ship, second state data of a target object, environmental state data, and navigation rule data corresponding to a target trip of the target ship, the target object and the environmental state data corresponding to a preset range corresponding to the target ship respectively; Inputting the first state data, the second state data, the environmental state data and the navigation rule data into a pre-trained large language model to generate a dynamic risk factor of the target ship, the dynamic risk factor being used to indicate a navigation risk object, a navigation risk factor and a corresponding risk degree of the target ship; According to the dynamic risk factor, a risk propagation network for the target ship is constructed, which is used to indicate the risk propagation path and the correlation of the dynamic risk factor to the target ship; According to the risk propagation network, a set of control instructions of the target ship is generated, which is used to indicate the risk avoidance guidance information of the target ship and / or to control the first state data of the target ship and / or to perform a repelling operation on the navigation risk object.

2. The LLM-based ship's assisted piloting implementation method according to claim 1, characterized in that, The first state data, the second state data, the environmental state data and the navigation rule data are input into the pre-trained large language model to generate the dynamic risk factor of the target ship, comprising: According to the first state data and the second state data, the relative motion trajectory data between the target ship and the target object is analyzed; According to the relative motion trajectory data, a time sequence collision risk coefficient between the target ship and the target object is calculated, which is used to represent the real-time collision risk degree between the target ship and the target object; The environmental state data is analyzed to analyze the fluid disturbance data within the preset range; According to the fluid disturbance data, the attitude offset correction amount corresponding to each of the target ship and the target object is generated; According to the navigation rule data, the current trajectory avoidance weight of the target ship is generated, which is used to quantify the current active avoidance responsibility degree of the target ship; The time sequence collision risk coefficient, the attitude offset correction amount and the trajectory avoidance weight are input into the pre-trained large language model to generate the dynamic risk factor of the target ship.

3. The LLM-based ship's assisted piloting implementation method according to claim 1, characterized in that, According to the dynamic risk factor, a risk propagation network for the target ship is constructed, comprising: The target ship is determined as a target node of risk transmission, and the navigation risk object in the dynamic risk factor is determined as a source node of risk transmission; According to the navigation risk factor and the corresponding risk degree in the dynamic risk factor, the connection edge space gradient distribution of the source node to the target node is quantified to construct a risk propagation network for the target ship.

4. The LLM-based ship's assisted piloting implementation method according to claim 1, characterized in that, Before the set of control instructions of the target ship is generated according to the risk propagation network, the method further comprises: Determine the associated node of the dynamic risk factor; Construct the associated risk propagation path between the dynamic risk factor and the associated node; According to the associated risk propagation path and the risk propagation network, a multi-level conduction intensity coefficient between the target ship and the associated node via the dynamic risk factor is calculated, the multi-level conduction intensity coefficient being used to represent a degree of risk conduction of the target ship to the associated node in the process of avoiding the dynamic risk factor; According to the multi-level conduction intensity coefficient, the risk propagation network is updated.

5. The LLM-based ship's assisted piloting implementation method according to claim 1, characterized in that, The generating, according to the risk propagation network, of the set of control instructions of the target ship comprises: According to the risk propagation network, a risk change rate of the navigation risk object is calculated, the risk change rate being used to represent a change situation of the navigation risk object corresponding to the risk degree over time; According to the risk change rate and the risk degree, the navigation risk object is prioritized to determine a processing order; According to the processing order and the navigation rule data, a corresponding risk avoidance strategy is retrieved from a preset rule library, the risk avoidance strategy being used to indicate a coping manner for a specific navigation risk object; According to the navigation rule data, a responsibility allocation between the target ship and the navigation risk object is analyzed, the responsibility allocation being used to represent a degree of responsibility in a potential collision accident; According to the responsibility allocation, the risk avoidance strategy is adjusted to generate preliminary control instructions; Conflicting items between the preliminary control instructions are resolved, and an instruction execution time sequence is planned based on the risk propagation network to generate a set of control instructions of the target ship.

6. The LLM-based ship's assisted piloting implementation method according to claim 5, characterized in that, The method further comprises: After the set of control instructions is generated, a change parameter of the dynamic risk factor and the risk propagation network is monitored in real time; According to the monitored change parameter, an effectiveness value of the set of control instructions is evaluated, the effectiveness value being used to represent a degree of control of the set of control instructions on the dynamic risk factor and the risk propagation network; When the effectiveness value is less than a preset threshold value, a target adjustment item in the set of control instructions is determined according to the change parameter; and the set of control instructions is updated according to the target adjustment item and the change parameter.

7. An LLM-based ship auxiliary piloting implementation system, characterized by, The system is used to implement the LLM-based ship auxiliary driving implementation method according to any one of claims 1-6, and comprises: An acquisition module is configured to acquire first state data of a target ship, second state data of a target object, environmental state data, and navigation rule data corresponding to a target trip of the target ship, the target object and the environmental state data corresponding to a preset range of the target ship, respectively; A generation module is configured to input the first state data, the second state data, the environmental state data, and the navigation rule data into a pre-trained large language model to generate a dynamic risk factor of the target ship, the dynamic risk factor being used to indicate a navigation risk object, a navigation risk factor, and a corresponding risk degree of the target ship. The generation module is further configured to generate a set of regulation instructions of the target ship according to the dynamic risk factor, the set of regulation instructions being used to instruct the target ship to execute a risk avoidance guiding operation and / or to regulate the first state data of the target ship and / or to execute a driving-off operation on the navigation risk object.

8. An LLM-based ship auxiliary piloting implementation system, characterized by, The system comprises: a memory storing executable program codes; a processor coupled with the memory; the processor invokes the executable program codes stored in the memory to execute the LLM-based ship auxiliary driving implementation method according to any one of claims 1-6.

9. A computer storage medium, characterized in that The computer storage medium stores computer instructions, which are invoked to execute the LLM-based ship auxiliary driving implementation method according to any one of claims 1-6.

Images