A kind of patrol boat autonomous cooperative perception and dynamic planning system and method

Abstract

This application relates to the field of intelligent navigation and path planning technology for maritime vessels, and discloses an autonomous collaborative perception and dynamic planning system and method for patrol boats. The system includes a base station and a patrol boat end: the base station consists of main modules for marine environment modeling, group federated learning aggregation, task management, and communication, responsible for generating and distributing lightweight knowledge packages; the patrol boat end comprises four main modules: multi-sensor perception, local lightweight intelligence, online planning, and propulsion and steering control, and is equipped with a dual-mode communication module supporting satellite links and inter-boat micro-communication. This method, through an architecture of "knowledge package infusion - local perception - intelligent prediction - rolling planning - dual-mode collaboration," realizes the transformation of patrol boat groups from centralized command to distributed edge intelligence in communication-constrained open-sea environments, significantly improving the system's autonomy, adaptability, and mission reliability in dynamic marine environments.

Description

Technical Field

[0001] This invention relates to the field of intelligent navigation and path planning technology on water, and more specifically, to an autonomous collaborative perception and dynamic planning system and method for patrol boats. Background Technology

[0002] The existing surface patrol boat system is a technical architecture used for multi-target patrol missions in complex sea areas (such as the open sea and island / reef areas), mainly adopting a "centralized computing at base stations - passive execution at the vessel" model. This architecture centralizes marine environmental data (such as wind, waves, and current fields) and patrol mission instructions at the base station for processing, and then transmits them to the patrol boat in real time via satellite or 5G long-distance communication links, thereby realizing path planning and task allocation.

[0003] Existing technologies are mainly used to improve patrol efficiency, such as in marine surveillance, environmental monitoring, or security patrols, by using multi-vessel collaboration to cover all sea areas and achieve data collection and real-time response.

[0004] For example, Chinese Patent Publication No. CN120506959A discloses a "Cooperative Path Planning Method, System, and Device for Marine Cross-Domain Unmanned Systems." This prior art provides a cooperative framework that includes multiple types of unmanned vehicles (such as unmanned surface vessels, USVs), with path optimization and task allocation performed by a coordinator on the base station side. Although this scheme can improve planning efficiency to some extent when communication is unobstructed, its overall cooperative logic heavily relies on the central computing and real-time command issuance of the base station. This means that once the remote communication link is interrupted due to severe weather (causing satellite signal attenuation) or human interference, the cooperative mechanism of the entire system will fail: the base station cannot update path commands according to the changing environment, and the patrol boat cannot report on-site emergencies (such as the appearance of dynamic targets or local sea condition deterioration), causing the pre-planned path to deviate from reality and become unable to be autonomously adjusted, thus leading to mission failure or even safety accidents.

[0005] Therefore, the shortcomings of existing technologies lie in their over-reliance on the stability of the communication link between the central base station and terminal equipment. This leads to system paralysis during communication interruptions, resulting in the loss of mission continuity and autonomous survivability in complex and unreliable communication environments. This severely limits the reliable application of existing systems in real-world scenarios such as remote seas and island reefs where communication support capabilities are limited. In view of this, this invention proposes a novel water patrol boat system architecture and method. The aim is to fundamentally solve the system failure problem caused by communication interruptions by decentralizing intelligent decision-making capabilities to the boat's deck and establishing a distributed collaborative mechanism, thereby achieving more reliable and autonomous patrol missions. Summary of the Invention

[0006] In view of the shortcomings of the existing technology, the purpose of this application is to provide an autonomous collaborative perception and dynamic planning system and method for patrol boats.

[0007] To achieve the above objectives, this application provides the following technical solution:

[0008] An autonomous collaborative perception and dynamic planning system for patrol boats, characterized in that it includes a base station and a patrol boat, wherein:

[0009] The base station includes:

[0010] The marine environment modeling module is used to receive multi-source marine data such as satellite remote sensing and ocean current forecasts, generate a high-precision four-dimensional marine environment model, and output spatiotemporal gridded marine prediction parameters.

[0011] The group federated learning aggregation module is used to receive local parameters and the latest spatiotemporal gridded ocean prediction parameters uploaded by each patrol boat, update and output the global model parameters;

[0012] The task management module is used to generate lightweight knowledge packages and distribute them to the patrol boat as the basis for local intelligent operation on the patrol boat.

[0013] The network communication module is used for data interaction between the base station and the patrol boat, transmitting lightweight knowledge packages via satellite links, etc.

[0014] The patrol boat end includes a multi-sensor sensing module, used to receive real-time marine data from optical, radar, AIS, and meteorological instruments, as well as local boat status data, and output fused local structured environmental data.

[0015] The local lightweight intelligent module is used to receive lightweight knowledge packages and local structured environmental data sent by the base station, and to predict and output data on future navigation costs and risks in local sea areas.

[0016] Online planning module: used to calculate and output optimal control sequence data based on navigation costs and risks, local vessel status and reference patrol routes;

[0017] Propulsion and steering control module: used to receive optimal control sequence data, execute actual navigation commands, and feed back the boat status data to the local multi-sensor perception module;

[0018] Network communication module: responsible for communicating with the base station, receiving and sending local parameters;

[0019] Inter-vessel micro-communication module: used to receive local navigation and risk data from other patrol vessels, and to send its own navigation and risk data.

[0020] In a preferred embodiment, the spatiotemporal gridded ocean prediction parameters are used in the population federated learning aggregation module and the task management module, including wind and wave thresholds, safety distances, drag coefficients, environmental heterogeneous parameters, and global environmental baseline distribution data.

[0021] In a preferred embodiment, the group federated learning aggregation module constructs a group federated learning aggregation model;

[0022] The group federated learning aggregation model analyzes the spatiotemporal gridded ocean prediction parameters and the local parameters uploaded by each patrol boat, and outputs the latest global model parameters.

[0023] The global model parameters include the neural network weight matrix, bias vector, and model hyperparameters of the global model.

[0024] In a preferred embodiment, the lightweight knowledge package generated by the base station task management module includes:

[0025] The spatiotemporal gridded ocean prediction parameters are derived from the ocean environment modeling module;

[0026] The global model parameters are derived from the population federated learning aggregation module.

[0027] Task execution metadata, which includes the patrol area boundary coordinate sequence, reference path points, time constraints, task priority, target weight allocation, and coverage threshold;

[0028] Control and optimization parameters: used in the patrol boat's online planning module, including state weights, control weights, and risk weights.

[0029] In a preferred embodiment, the multi-sensor sensing module employs spatiotemporal registration and Kalman filtering to analyze multi-source real-time marine data collected by local patrol boat's optical, radar, AIS, and meteorological instruments, along with local boat status data, and outputs fused local structured environmental data.

[0030] The local structured environment data, as one of the local parameters, is fed back to the base station via network communication.

[0031] In a preferred embodiment, the local lightweight intelligent module employs a neural network-based regression prediction architecture to construct a lightweight intelligent model;

[0032] Based on the lightweight intelligent model, analyze the lightweight knowledge package data and local structured environment data to predict the future navigation costs and risks, and output the navigation cost and risk results data.

[0033] Based on the updated model weights after the lightweight intelligent model runs, the parameters are output to the local parameters and fed back to the base station via backbone network communication.

[0034] In a preferred embodiment, the online planning module employs a planning framework that integrates long-term and short-term optimization to construct an online planning model;

[0035] Based on the online planning model, analyze lightweight knowledge package data, navigation cost and risk outcome data, and local vessel status data to predict and output the optimal control sequence;

[0036] The optimal control sequence includes speed commands and rudder angle commands.

[0037] In a preferred embodiment, the propulsion and rudder control module executes speed commands and rudder angle commands based on the received optimal control sequence data;

[0038] The propulsion and steering control module collects and outputs current submarine status data, which is then transmitted to the local multi-sensor perception module, online planning module, and inter-submarine micro-communication module.

[0039] In a preferred embodiment, the inter-vessel micro-communication module receives local structured environmental data, vessel status data, and patrol task data sent by each patrol vessel, and sends the local structured environmental data, vessel status data, and patrol task data back to each patrol vessel.

[0040] A method for autonomous collaborative perception and dynamic planning of patrol boats, characterized by the following steps: S1: The base station initializes the system, completes global marine environment modeling and task integration, generates a lightweight knowledge package and distributes it to each patrol boat.

[0041] S2: After receiving the lightweight knowledge package, the patrol boat starts the patrol mission and activates the local multi-sensor perception module to collect real-time marine environmental data around the boat.

[0042] S3: The patrol boat continuously monitors the communication link status. When the communication is good, it enters the cloud-edge collaboration mode. When the communication is interrupted, it switches to the autonomous collaboration mode.

[0043] S4: When the patrol boat is in cloud-edge collaboration mode, each patrol boat will upload the latest local parameters to the base station. The base station will update the lightweight knowledge package and distribute it to each patrol boat to achieve cloud-edge collaboration optimization.

[0044] S5: In the autonomous collaborative mode, the patrol boat relies on local and surrounding marine environment perception, navigation cost and risk prediction, and information sharing among neighboring boats to autonomously plan its route and complete its mission.

[0045] S6: If the patrol boat detects that all the predetermined patrol targets have been achieved and the mission is completed, the patrol boat will stop operating and send a confirmation message of mission completion to the base station.

[0046] If the patrol boat detects extreme weather conditions through its local multi-sensor sensing module, it will immediately initiate a safe return to port and transmit the mission interruption command and surrounding environmental data back to the base station.

[0047] By adopting the above technical solution, the beneficial effects of the present invention are as follows:

[0048] This invention fundamentally reduces reliance on remote communication, significantly improving the system's survivability and mission robustness in intermittent connectivity environments. Existing technologies become paralyzed when communication is interrupted, while this invention pre-configures global intelligence on the vessel's end using a "knowledge package" mechanism. Combined with a lightweight, locally-based intelligent model and a predictive control online planner, this allows the patrol vessel to continue making autonomous decisions and operating continuously based on prior knowledge and real-time perception even after a complete interruption of remote communication links such as satellite communication. The system's primary communication requirements have shifted from high-frequency, high-bandwidth real-time command / video stream transmission in traditional architectures to low-frequency, small-data-volume model parameter updates, effectively overcoming communication bottlenecks at sea and ensuring the uninterrupted nature of core missions.

[0049] This invention achieves highly adaptive decision-making to dynamic marine environments, significantly improving the accuracy of path planning and navigation efficiency. Addressing the shortcomings of existing methods that rely on static environmental models and lead to inaccurate planning, this invention uses a multi-sensor sensing module to acquire real-time data on wind, waves, and currents around the vessel. A local lightweight intelligent model then predicts navigation costs and risks for future periods, and finally, a predictive control online planner performs rolling optimization. This closed-loop mechanism of "sensing-prediction-rolling optimization" allows the planned path to dynamically adapt to the spatiotemporal changes in actual sea conditions, effectively avoiding high-energy-consuming and high-risk areas, thereby significantly reducing navigation time and energy consumption while ensuring safety.

[0050] This invention enhances the intelligence and autonomous collaboration capabilities of multi-vessel clusters, ensuring high system efficiency without centralized command. It achieves distributed information exchange between patrol vessels through inter-vessel micro-communication modules, sharing key local situational information such as "patrolled areas." Combined with a predictive control online planner integrating neighboring vessel information, it realizes self-organizing collaboration without a base station center for scheduling. This approach overcomes the low collaboration efficiency of existing static area division methods, autonomously avoiding task duplication or coverage blind spots, and maintaining efficient and orderly collaborative patrol operations even during base station communication interruptions.

[0051] Through multi-layered technological integration, a qualitative leap in the overall system performance has been achieved, moving from "centralized remote control" to "distributed swarm intelligence." This invention is not a partial improvement on existing technologies, but rather a transformation of the system's technological paradigm through the deep and organic integration of technologies such as a "cloud-edge-device" layered architecture, swarm federated learning, and local parameter prediction control. The base station focuses on global knowledge aggregation and evolution, while the submarine focuses on local autonomous decision-making and collaboration. The two form a bidirectional intelligent iteration through federated learning. This enables the system not only to survive communication interruptions but also to acquire advanced intelligent features such as continuous learning, autonomous adaptation to complex environments, and collaborative evolution, resulting in a qualitative leap in overall performance. Attached Figure Description

[0052] Figure 1 This is a schematic diagram of an autonomous collaborative perception and dynamic planning system architecture for patrol boats according to the first embodiment of this application;

[0053] Figure 2 This is a schematic diagram of a patrol boat autonomous collaborative perception and dynamic planning method according to the second embodiment of this application. Detailed Implementation

[0054] In the following description, many technical details are presented to help the reader better understand this application. However, those skilled in the art will understand that the technical solutions claimed in this application can be implemented even without these technical details and various variations and modifications based on the following embodiments.

[0055] To make the objectives, technical solutions, and advantages of this application clearer, the embodiments of this application will be described in further detail below with reference to the accompanying drawings.

[0056] The first embodiment of this application discloses an autonomous collaborative perception and dynamic planning system for patrol boats. The system includes a base station and a patrol boat terminal, wherein:

[0057] The base station includes: a marine environment modeling module, used to receive multi-source marine data such as satellite remote sensing and ocean current forecasts, generate a high-precision four-dimensional marine environment model, and output spatiotemporal gridded marine prediction parameters; a group federated learning aggregation module, used to receive local parameters and the latest spatiotemporal gridded marine prediction parameters uploaded by each patrol vessel, update and output global model parameters; a task management module, used to generate lightweight knowledge packages and distribute them to the patrol vessel, serving as the basis for local intelligent operation on the patrol vessel; and a network communication module, used for data interaction between the base station and the patrol vessel, transmitting lightweight knowledge packages via satellite links, etc.

[0058] The patrol boat end includes: a multi-sensor perception module for receiving real-time oceanographic data from optical, radar, AIS, and meteorological instruments, as well as local boat status data, and outputting fused local structured environmental data; a local lightweight intelligent module for receiving lightweight knowledge packages and local structured environmental data from the base station, predicting and outputting future navigation costs and risks in the local sea area; an online planning module for calculating and outputting optimal control sequence data based on navigation costs and risks, local boat status, and reference patrol paths; a propulsion and steering control module for receiving optimal control sequence data, executing actual navigation commands, and feeding back boat status data to the local multi-sensor perception module; a network communication module for communicating with the base station, receiving and sending local parameters; and an inter-boat micro-communication module for receiving local navigation area and risk data from other patrol boats and sending its own navigation area and risk data.

[0059] A schematic diagram of a specific patrol boat autonomous collaborative perception and dynamic planning system architecture is shown below. Figure 1 As shown.

[0060] The implementation methods of each module on the base station side are as follows:

[0061] The marine environment modeling module receives multi-source marine data, including satellite remote sensing and ocean current forecasts, and generates a 1km×1km spatiotemporally gridded four-dimensional marine environment model. It outputs spatiotemporally gridded marine prediction parameters (including wind and wave thresholds). safe distance drag coefficient , , , Heterogeneous environmental parameters, global environmental baseline distribution Q( (etc.), used by the federated learning aggregation module and task management module of the supply group.

[0062] The group federated learning aggregation module constructs a group federated learning aggregation model, taking into account spatiotemporal gridded ocean prediction parameters and local parameters uploaded by each patrol vessel (including local structured environmental data and model weights updated after running the local lightweight intelligent model), and aggregates and outputs updated global model parameters. This module is associated with the task management module and encapsulates the aggregation results into a lightweight knowledge package to adapt to the knowledge evolution needs after communication interruptions.

[0063] The specific aggregation formula is as follows:

[0064]

[0065] in: The global model parameter vector updated in round t+1 (directly loaded into the local lightweight intelligent model on the patrol boat as the prediction function) parameters This provides lightweight knowledge packages to the task management module. The total number of patrol boats participating in the training; For the first The aggregate weighting coefficient of the patrol boats; No. patrol boats in The local model parameters after the round of updates; For the first The local loss function of a patrol boat; This is the consistency weighting coefficient (hyperparameter); These are the parameters of the previous round of global model; For environmental adaptation weighting coefficients (hyperparameters); M is the number of environmental factor types; Environmental heterogeneous parameters (source: marine environment modeling module); for Divergence measures the degree of difference between two probability distributions. For the first The local sea area where the vessel is located Data distribution of environmental factors (source: local structured environmental data fed back from patrol boats); For the entire sea area The data distribution of environmental factors comes from the base (source: marine environment modeling module).

[0066] This module introduces an environmental distribution difference metric based on KL divergence, enabling the federated aggregation process to not only learn the common data characteristics of each vessel but also explicitly compensate for the environmental heterogeneity of different patrol areas. This mechanism effectively alleviates the performance degradation problem of traditional methods in scenarios where data is not independent and identically distributed, thereby improving the adaptability and generalization ability of the global model distributed to the vessel for different local environments, and providing a more accurate prior knowledge foundation for intelligent decision-making on the vessel.

[0067] The task management module generates a lightweight knowledge package based on data such as patrol tasks and global model parameters. The specific content includes:

[0068] Global model parameters: Neural network weight matrix ( , ,..., ), corresponding to the local lightweight intelligent model on the patrol boat. Layer network; bias vector ( , ,..., Model hyperparameters (learning rate, batch size, activation function type, etc.).

[0069] Spatiotemporal gridded ocean prediction parameters: wind and wave threshold safe distance drag coefficient , , , Heterogeneous environmental parameters, global environmental baseline distribution Q( )wait.

[0070] Task execution metadata: patrol area boundary coordinate sequence, reference path points Including time constraints, task priority and target weight allocation, and coverage thresholds. (Used for determining the termination of patrol boat missions).

[0071] Control and optimization parameters: used in the patrol boat's online planning module, including state weights. Control weights Risk weights .

[0072] The network communication module is responsible for data interaction between the base station and the patrol boat, transmitting knowledge packets via satellite links and other means.

[0073] The implementation methods of each module on the patrol boat are as follows:

[0074] The multi-sensor sensing module receives real-time ocean data from multiple sources, including optical, radar, AIS, and meteorological instruments, as well as local vessel status data. Through spatiotemporal registration and Kalman filtering techniques, it outputs fused structured environmental data, specifically including:

[0075] a. Fusion environment vector X: A feature vector of dimension F, containing... , , , , , , , , Real-time sensing elements.

[0076] b. Perturbation sequence : Future time-period disturbance prediction sequence based on wind, wave and flow field estimation.

[0077] c. Environment Matrix : The spatiotemporal gridded environmental state matrix around the submarine.

[0078] Furthermore, it also includes: using local structured environmental data as one of the local parameter data and feeding it back to the base station via network communication.

[0079] The local lightweight intelligent module adopts a neural network-based regression prediction architecture to build a lightweight intelligent model. The lightweight knowledge package issued by the base station and the data output by the local multi-sensor perception module are used as the input data for the model to predict the future. The cost and risk of a flight over a specific time period (e.g., 30 minutes). The specific prediction formula model is as follows:

[0080]

[0081] in, To fuse environment vectors; Model parameter set ( , , ... , ); For the future The predicted cost vector for a given period, the first component of the vector The energy consumption prediction value (in kWh) is obtained by approximate integration through network calculation. , , , , Source: Lightweight knowledge package, boats Wind speed ,wind direction High waves Flow rate Flow direction (Source: Local multi-sensor perception module); second component of the vector. The comprehensive risk prediction value is obtained through a weighted summation operation of the network. , For the first A risk indicator function, (As risk weights).

[0082] Furthermore, it also includes: using the updated model weights after the lightweight intelligent model is run as one of the local parameter data, and feeding them back to the base station through the network communication module.

[0083] This module transforms complex energy consumption and risk integration calculations into an efficient forward propagation process using a neural network model, significantly reducing computational complexity and meeting the real-time requirements under limited onboard computing resources. By integrating global model priors and local real-time perception data, the module's prediction output simultaneously achieves consistency with the global strategy and high fidelity to the local dynamic environment, providing subsequent planning modules with accurate and forward-looking cost and risk assessments. The online planning module adopts a planning framework that integrates long-term and short-term optimizations, constructing an online planning model. Lightweight knowledge package data from the base station, locally predicted navigation cost and risk results, and local boat state data are used as model output data to predict and output the optimal control sequence. The specific prediction formula model is as follows:

[0084]

[0085] Constrained by:

[0086]

[0087] in, It refers to the optimal control sequence, including speed commands and rudder angle commands; It refers to the time from the present moment to the future. The control input sequence for each step; It is a path tracking item. It refers to patrol boats in The state vector at time t, This is the state weight matrix (source: Lightweight Knowledge Package). This refers to the control smoothing term. In the finger The control input vector applied to the patrol boat at all times To control the weights (source: lightweight knowledge package); This refers to risk aversion measures. For the risk function ( The data originates from the environmental matrix in the multi-sensor perception module. ), Risk weighting coefficient (source: lightweight knowledge package); This refers to long-term forecasting. The predicted cost vector (source: local lightweight intelligent module). For long-term prediction weighting coefficients (dynamically adjusted based on communication status); current state Real-time feedback from the propulsion and steering control systems; reference path Task execution metadata from a lightweight knowledge package; dynamic constraints Based on the Nomoto ship model For environmental disturbance.

[0088] The rolling optimization framework employed in this module offers a technological advantage in achieving multi-objective collaborative optimization. In the objective function, the "path tracking term" and "control smoothing term" ensure the baseline performance of the navigation; the "risk avoidance term" achieves local dynamic obstacle avoidance and disturbance resistance control based on real-time environmental data; and the "long-term prediction term" maps forward-looking cost risks into the optimization process, guiding the system to avoid potentially unfavorable areas ahead. This design ensures that the planning results meet both immediate safety requirements and long-term economic and safety requirements.

[0089] The propulsion and steering module receives the optimal control sequence u∗ output from the online planning module and converts it into specific thruster thrust and rudder steering signals, driving the patrol boat to perform precise navigation maneuvers. This module also collects real-time data on the boat's motion (position, velocity, and heading angle) as feedback signals. The data is output to the multi-sensor perception module for correction and fed back to the online planning module as the initial state for a new round of optimization. Simultaneously, gridded status information of the patrolled sea area is generated based on the actual navigation trajectory. The output is sent to the inter-submarine micro-communication module to achieve distributed collaboration.

[0090] Further advancements include the acquisition and output of current boat status data (boat position, speed, heading, attitude angles and their rates of change, etc.) by the steering and control module, which is then transmitted to the local multi-sensor perception module and the inter-boat micro-communication module.

[0091] This module forms a closed-loop control system of "perception-decision-execution-feedback" by feeding back the hull status to the perception and planning module in real time. This mechanism enables the system to continuously correct itself based on the actual execution effect of commands, improving the system's robustness and control accuracy in the face of model uncertainties and external disturbances. Simultaneously, the precise information on the patrolled areas it provides offers crucial decision-making support for inter-vessel coordination, effectively integrating the local actions of individual vessels into the overall collaborative efficiency of the cluster.

[0092] The inter-vessel micro-communication module, serving as a key hub for achieving distributed autonomous collaboration, integrates an IEEE 802.11s self-organizing wireless network card and a VHF data exchange communication link (VDES) to form a primary and backup dual-link system, ensuring the robustness and self-healing capability of the communication link under complex sea conditions. Its core data includes: a matrix of the surrounding environment generated by the multi-sensor perception module. Real-time boat status and grid map information of the patrolled area generated by the propulsion and steering control module , and the navigation cost and risk data C(t+Δt) predicted by the local lightweight intelligent module.

[0093] This module enhances the reliability of the collaborative communication link under harsh conditions through a redundant design with primary and backup dual links. The exchange of key situational information, such as patrolled areas, enables the fleet to achieve self-organized coverage and task coordination of patrol areas without central dispatch, effectively reducing blind spots and redundant patrols. This distributed information exchange mechanism is the core technical guarantee for maintaining efficient collaborative operation of the system after communication interruptions.

[0094] The network communication module is responsible for data interaction with the base station, including receiving lightweight knowledge packages sent by the base station and uploading local parameters when communication is good, to ensure cloud-edge connectivity.

[0095] The second embodiment of this application relates to a method for autonomous cooperative perception and dynamic planning of patrol boats, the method flow of which is as follows: Figure 2 As shown, the method includes the following steps S1 to S6:

[0096] In step S1: The base station initializes the system, completes global marine environment modeling and task integration, generates a lightweight knowledge package, and distributes it to each patrol boat.

[0097] Specifically, upon startup at the base station, the marine environment modeling module integrates multi-source marine data to construct a spatiotemporally gridded four-dimensional environmental model, generating spatiotemporally gridded marine prediction parameters including parameters such as wind and wave thresholds and safe distances. Simultaneously, the task management module, in conjunction with patrol task planning (such as area boundaries and reference path points), encapsulates the initial global model parameters generated by the group federated learning aggregation module into a lightweight knowledge package, which is then distributed to all patrol vessels via the network communication module.

[0098] In step S2: After receiving the lightweight knowledge package, the patrol boat initiates its patrol mission and simultaneously activates its local multi-sensor perception module to collect real-time marine environmental data around the boat. Specifically, each patrol boat receives and loads the lightweight knowledge package to initialize its local lightweight intelligent module and online planning module. It then initiates the patrol mission and activates its multi-sensor perception module to continuously collect real-time data from optical, radar, AIS, and meteorological instruments. After fusing this data with the boat's status data, it outputs structured local environmental data. In step S3: The patrol boat continuously monitors the communication link status. When communication is good, it enters the cloud-edge collaborative mode; when communication is interrupted, it switches to the autonomous collaborative mode. Specifically, the patrol boat monitors the communication link status (latency, packet loss rate) in real time through the network communication module and makes a mode judgment based on preset thresholds (latency < 500ms and packet loss rate < 5%). When communication is good, it outputs a mode signal mode=1, triggering the cloud-edge collaborative mode and executing step S4; when communication is interrupted, it outputs mode=0, switches to the autonomous collaborative mode, and executes step S5. This judgment result serves as the benchmark for the system's global mode switching, while historical communication data is used for adjusting heterogeneous environmental parameters in group federated learning.

[0099] In step S4: When the patrol boat is in cloud-edge collaborative mode, each patrol boat uploads the latest integrated local parameters to the base station. The base station updates the lightweight knowledge package and distributes it to each patrol boat to achieve cloud-edge collaborative optimization.

[0100] Specifically, in a good communication mode, the patrol boats upload their local parameters to the base station. The base station's federated learning aggregation module receives the local parameters from each patrol boat, updates the global model parameters through the federated learning aggregation model, and the task management module encapsulates the updated global model parameters, spatiotemporal gridded ocean prediction parameters, task execution metadata, and control and optimization parameters into a new lightweight knowledge package, which is then distributed to each patrol boat. Upon receiving the latest lightweight knowledge package, the patrol boats update their local execution tasks, achieving cloud-edge collaborative optimization. In this mode, the system uses the base station as the central hub for global optimization, with each patrol boat executing the base station's optimization commands.

[0101] In step S5: When the patrol boat is in autonomous collaborative mode, it relies on local and surrounding marine environment perception, navigation cost and risk prediction, and information sharing among neighboring boats to autonomously plan its route and complete the mission.

[0102] Specifically, in communication interruption mode, the patrol boat switches to autonomous collaborative operation. The multi-sensor perception module continuously provides environmental data vectors, disturbance sequences, and environmental matrices. The local lightweight intelligent model, based on model parameters from a lightweight knowledge package, outputs predicted costs and risks to the online planning module. The online planning module simultaneously receives shared information from neighboring boats (including patrolled areas and risk alarms) from the inter-boat micro-communication module, and, combining this with a reference path and the current boat status, generates the optimal control sequence through optimization calculations. The propulsion and steering control systems then execute... It also provides real-time vessel status feedback, forming a local closed loop. The inter-vessel micro-communication module enables distributed collaboration among patrol vessels, sharing surrounding environmental data and current patrol status.

[0103] In step S6: If the patrol boat detects that all the predetermined patrol targets have been achieved and the mission is completed, the patrol boat stops operating and sends a confirmation message of mission completion to the base station.

[0104] If the patrol boat detects extreme weather conditions through its local multi-sensor sensing module, it will immediately initiate a safe return to port and transmit the mission interruption command and surrounding environmental data back to the base station.

[0105] The patrol boat continuously monitors the mission's execution status and uses a dual mechanism to determine termination: when the patrolled area coverage rate reaches a certain threshold... Exceeding the threshold set in the knowledge package When J represents the total number of grid cells in the task area, the task is considered to have been completed normally, and confirmation information is sent back to the base station; when the multi-sensor sensing module detects extreme weather indicators in the environmental data (such as wind speed exceeding a threshold), the task is considered to have been completed normally. Upon such an event, the system immediately triggers a safe return-to-base procedure and transmits the interruption alarm and environmental data back to the base station. The determination result serves as the endpoint of the mission loop and provides data support for the optimization of the knowledge package for subsequent missions, ensuring the integrity and safety of the system under complex sea conditions.

[0106] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any ordinary changes and substitutions made by those skilled in the art within the scope of the technical solution of the present invention should be included within the protection scope of the present invention.

Claims

1. A patrol boat autonomous cooperative perception and dynamic planning system, characterized in that, Including base station terminals and patrol boat terminals, among which: The base station includes: The marine environment modeling module is used to receive multi-source marine data such as satellite remote sensing and ocean current forecasts, generate a high-precision four-dimensional marine environment model, and output spatiotemporal gridded marine prediction parameters. The group federated learning aggregation module is used to receive local parameters and the latest spatiotemporal gridded ocean prediction parameters uploaded by each patrol boat, update and output the global model parameters; The task management module is used to generate lightweight knowledge packages and distribute them to the patrol boat as the basis for local intelligent operation on the patrol boat. The network communication module is used for data interaction between the base station and the patrol boat, transmitting lightweight knowledge packages via satellite links, etc. The patrol boat end includes: The multi-sensor sensing module is used to receive real-time ocean data from optics, radar, AIS, and meteorological instruments, as well as local vessel status data, and output fused local structured environmental data. The local lightweight intelligent module is used to receive lightweight knowledge packages and local structured environmental data sent by the base station, and to predict and output data on future navigation costs and risks in local sea areas. Online planning module: used to calculate and output optimal control sequence data based on navigation costs and risks, local vessel status and reference patrol routes; Propulsion and steering control module: used to receive optimal control sequence data, execute actual navigation commands, and feed back the boat status data to the local multi-sensor perception module; Network communication module: responsible for communicating with the base station, receiving and sending local parameters; Inter-vessel micro-communication module: used to receive local navigation and risk data from other patrol vessels, and to send its own navigation and risk data.

2. The patrol boat autonomous collaborative perception and dynamic planning system according to claim 1, characterized in that, The spatiotemporal gridded ocean prediction parameters are used in the group federated learning aggregation module and the task management module, including wind and wave thresholds, safe distances, drag coefficients, environmental heterogeneous parameters, and global environmental baseline distribution data.

3. The patrol boat autonomous collaborative perception and dynamic planning system according to claim 2, characterized in that, The group federated learning aggregation module constructs a group federated learning aggregation model; The group federated learning aggregation model analyzes the spatiotemporal gridded ocean prediction parameters and the local parameters uploaded by each patrol boat, and outputs the latest global model parameters. The global model parameters include the neural network weight matrix, bias vector, and model hyperparameters of the global model.

4. The autonomous co-perception and dynamic planning system for a patrol boat of claim 3, wherein, The lightweight knowledge package generated by the base station task management module includes: The spatiotemporal gridded ocean prediction parameters are derived from the ocean environment modeling module; The global model parameters are derived from the population federated learning aggregation module. Task execution metadata includes the patrol area boundary coordinate sequence, reference path points, time constraints, task priority, target weight allocation, and coverage threshold; Control and optimization parameters: used in the patrol boat's online planning module, including state weights, control weights, and risk weights.

5. The autonomous co-perception and dynamic planning system for a patrol boat according to claim 4, wherein, The multi-sensor sensing module uses spatiotemporal registration and Kalman filtering to analyze multi-source real-time marine data collected by local patrol boat optical, radar, AIS and meteorological equipment and local boat status data, and outputs fused local structured environmental data. The local structured environment data, as one of the local parameters, is fed back to the base station via network communication.

6. The patrol boat autonomous collaborative perception and dynamic planning system according to claim 5, characterized in that, The local lightweight intelligent module adopts a neural network-based regression prediction architecture to construct a lightweight intelligent model; Based on the lightweight intelligent model, analyze the lightweight knowledge package data and local structured environment data to predict the future navigation costs and risks, and output the navigation cost and risk results data. Based on the updated model weights after the lightweight intelligent model runs, the parameters are output to the local parameters and fed back to the base station via backbone network communication.

7. The autonomous co-perception and dynamic planning system for a patrol boat of claim 6, wherein, The online planning module adopts a planning framework that integrates long-term and short-term optimization to construct an online planning model; Based on the online planning model, analyze lightweight knowledge package data, navigation cost and risk outcome data, and local vessel status data to predict and output the optimal control sequence; The optimal control sequence includes speed commands and rudder angle commands.

8. The patrol boat autonomous collaborative perception and dynamic planning system according to claim 7, characterized in that, The propulsion and rudder control module executes speed and rudder angle commands based on the received optimal control sequence data; The propulsion and steering control module collects and outputs the current submarine status data, which is then transmitted to the local multi-sensor fusion perception module, online planning module, and inter-submarine micro-communication module.

9. The autonomous co-perception and dynamic planning system for a patrol boat of claim 8, wherein, The inter-vessel micro-communication module receives local structured environmental data, vessel status data, and patrol task data sent by other patrol vessels, and sends its own local structured environmental data, vessel status data, and patrol task data to each patrol vessel.

10. A method for autonomous collaborative perception and dynamic planning of patrol boats, characterized in that, Includes the following steps: Step 1: The base station initializes the system, completes global marine environment modeling and task integration, generates a lightweight knowledge package, and distributes it to each patrol boat. Step 2: After receiving the lightweight knowledge package, the patrol boat starts the patrol mission and activates the local multi-sensor perception module to collect real-time marine environmental data around the boat. Step 3: The patrol boat continuously monitors the communication link status. When communication is good, it enters the cloud-edge collaboration mode; when communication is interrupted, it switches to the autonomous collaboration mode. Step 4: When the patrol boat is in cloud-edge collaborative mode, each patrol boat will upload the latest integrated local parameters to the base station. The base station will update the global model parameters and send out the optimized lightweight knowledge package to achieve cloud-edge collaborative optimization. Step 5: In the autonomous collaborative mode, the patrol boat relies on local and surrounding marine environment perception, navigation cost and risk prediction, and information sharing among neighboring boats to autonomously plan its route and complete the mission. Step Six: If the patrol boat detects that all the predetermined patrol targets have been achieved and the mission is completed, the patrol boat will stop operating and send a confirmation message of mission completion to the base station. If the patrol boat detects extreme weather conditions through its local multi-sensor sensing module, it will immediately initiate a safe return to port and transmit the mission interruption command and surrounding environmental data back to the base station.

Images