Search and rescue boat area coverage cruising method and system

Abstract

This invention relates to the technical field of vessels, specifically to a method and system for area coverage cruise of search and rescue vessels. A data fusion module acquires and fuses real-time status data from heterogeneous search and rescue platforms and multi-source environmental data from the search and rescue area. A drift prediction and information gain evaluation module generates and dynamically updates a target presence probability distribution map based on the fused data and calculates the comprehensive search value of the heterogeneous search and rescue platforms performing candidate search actions. A cross-domain collaborative path planning module receives the comprehensive search value and plans a collaborative cruise path for the heterogeneous search and rescue platforms based on the principle of maximizing the comprehensive search value. An adaptive control execution module controls the heterogeneous search and rescue platforms to execute the collaborative cruise path and adaptively adjusts the formation configuration or task allocation based on platform status data and communication link quality data. This invention solves engineering problems such as communication interruption and energy depletion, ensuring system operation in harsh environments.

Description

Technical Field

[0001] This invention relates to the technical field of boats, and specifically to a method and system for area coverage patrol of a search and rescue boat. Background Technology

[0002] With the acceleration of urbanization and the impact of climate change, the search and rescue needs of rivers, as an important part of inland water systems, are becoming increasingly prominent. River environments are typically narrow, winding, and have complex currents such as rapids, floods, and shallow waters. They are also easily disturbed by bank vegetation, bridges, and human activities, leading to inefficiencies and slow responses in traditional search and rescue methods. Existing technologies mostly rely on single platforms or manual operation, failing to achieve comprehensive coverage and real-time adaptation. There is an urgent need for multi-platform collaborative systems to improve search accuracy, shorten rescue time, and reduce risks. However, existing river search and rescue systems still have many limitations in areas such as data fusion, drift prediction, path planning, and target confirmation.

[0003] In the existing field of river search and rescue technology, multi-platform collaborative systems have become a key method to improve the efficiency of inland waterway or river basin searches. For example, announcement number CN106772515A discloses a rapid judgment and determination system and method for inland waterway vessel accidents, as well as a search and rescue system. This system uses AIS base station reliability area analysis to classify areas into four categories: most reliable, reliable, relatively reliable, and unreliable. Combined with environmental information such as weather and sea conditions, it quickly judges the vessel accident and simulates the route of the wrecked vessel and the signal collection of surrounding vessels to determine the accident area. Finally, based on the size of the area: less than 2 nautical miles, a single-ship fan-shaped search is used; 2-10 nautical miles, a single-ship extended square search is used; and more than 10 nautical miles, a parallel search or ship-machine collaborative search is used to generate a search and rescue plan, supporting rescue operations in river waterways. The core of this technology lies in accident identification and area division, but the problem is that the data fusion module lacks real-time adaptive adjustment to dynamic water flow in the river, such as rapids and floods. This leads to positioning errors or mission interruptions in winding river sections or shallow water areas, and it cannot effectively handle the uncertainty of drift prediction.

[0004] CN115027627A discloses an intelligent unmanned surface vessel (USV) system for watershed safety inspection and rescue. This system utilizes onboard robots such as security inspection robots, surveying robots, and rescue and salvage robots, along with image recognition, to perform inspections and rescues along the watershed and at key river points. This includes video acquisition, detection of potential hazards such as surface pollutants, emergency response (e.g., drowning victim rescue and material retrieval), and a visual servo control method using a fuzzy cerebellum model neural network algorithm to control the robotic arm. It combines electronic tag positioning and visual navigation for USV navigation, positioning, and trajectory tracking analysis. While the system emphasizes autonomous cruising and target detection, it suffers from insufficient integration with aerial platforms, such as UAVs or underwater equipment. This results in limited search coverage due to narrow river channels and riverbank obstacles. Furthermore, the lack of advanced prediction mechanisms, such as neural network-enhanced particle filters (NN-PF), makes it unable to accurately predict river drift trajectories, and it is prone to high-entropy uncertainty in fast-flowing environments.

[0005] Publication No. CN110104139B discloses a maritime patrol device and its usage method for an unmanned surface vessel (USV) carrying a drone. The device includes a hull made of solar panels, a drone take-off and landing platform system, and an automatic drone battery replacement and charging system. The usage method involves the USV navigating along a planned trajectory, the drone taking off for patrol, transmitting real-time images / data, marking a location when the battery is low, returning to the marked point, automatically replacing the battery, and then returning to continue operation. This technology emphasizes heterogeneous collaboration, but a problem exists: it does not link the confidence distribution with the probability distribution map (PDM), resulting in insufficient adaptability on devices at the riverbank and an inability to dynamically adjust the search boundary to cope with low-confidence, high-probability areas, such as shallow-water floating objects.

[0006] While the aforementioned existing technologies have made some progress in river platform collaboration, drift prediction, and path planning, they generally suffer from problems such as loose coupling between modules, weak adaptability to complex river conditions (i.e., weak capacity in rapids, shallow water, and bends), low computational efficiency, and insufficient target confirmation accuracy. These problems lead to low search and rescue efficiency and an inability to effectively cope with dynamic uncertainties. Summary of the Invention

[0007] Therefore, the purpose of this invention is to provide a method and system for area coverage patrol of search and rescue boats, which solves the technical problems existing in the search and rescue of accidents involving new energy vessels in inland waterways, such as inaccurate drift prediction, blind path planning, low efficiency of multi-platform collaboration, poor system robustness, and high false alarm rate of target confirmation.

[0008] To achieve the above objectives, the present invention provides the following technical solution:

[0009] A search and rescue boat area coverage cruise system includes:

[0010] The data fusion module is used to acquire and fuse real-time status data of heterogeneous search and rescue platforms and multi-source environmental data of the search and rescue area. The heterogeneous search and rescue platforms include at least two of the following: surface, underwater and aerial platforms.

[0011] The drift prediction and information gain assessment module, which is connected to the data fusion module, is used to: generate and dynamically update the target existence probability distribution map based on the fused data, and calculate the comprehensive search value of the heterogeneous search and rescue platform to perform candidate search actions;

[0012] The cross-domain collaborative path planning module, which is connected to the drift prediction and information gain evaluation module, is used to receive the comprehensive search value and plan a collaborative cruise path for heterogeneous search and rescue platforms based on the principle of maximizing the comprehensive search value.

[0013] The adaptive control execution module, connected to the cross-domain collaborative path planning module, is used to control the heterogeneous search and rescue platforms to execute collaborative cruise paths. Based on platform status data and communication link quality data, it adaptively adjusts the formation configuration or task allocation to optimize the overall search and rescue coverage and operational robustness of the system.

[0014] The present invention is further configured such that: the data fusion module includes:

[0015] A multi-source sensor interface for acquiring at least one of wind field data, water flow data, image data, and underwater flow field data;

[0016] The platform status monitoring submodule is used to obtain data on the remaining battery life, communication link quality, and positioning accuracy of heterogeneous search and rescue platforms.

[0017] The data preprocessing unit is used to synchronize environmental data and platform status data to a unified spatiotemporal reference.

[0018] The present invention is further configured such that the drift prediction and information gain evaluation module includes:

[0019] The drift prediction submodule uses a particle filter based on neural network enhancement to dynamically estimate the target position and drift trajectory. The neural network is used to optimize the particle weight update and resampling process.

[0020] The present invention is further configured such that the drift prediction and information gain evaluation module includes:

[0021] The information gain evaluation submodule is used to calculate the overall search value, which is weighted based on the following parameters:

[0022] Information gain entropy reduction based on probability distribution plot;

[0023] Environmental perception effectiveness factors related to environmental data in the search and rescue area;

[0024] Heterogeneous collaborative gain factor characterizing the contribution of current actions to the future search performance of other heterogeneous platforms;

[0025] Energy and time cost factors required to perform candidate search actions;

[0026] It also outputs the comprehensive search value to the cross-domain collaborative path planning module as a control input for search path priority ranking and decision-making.

[0027] The present invention is further configured such that: the cross-domain collaborative path planning module adopts a distributed model predictive control algorithm, and solves the collaborative cruise path based on the platform's dynamic model, energy consumption constraints and minimum safe distance constraints.

[0028] The present invention is further configured such that: the adaptive control execution module includes:

[0029] The communication quality assessment submodule is used to trigger the switching of the preset communication protocol mode or adjust the data compression rate when the communication link quality is lower than the preset communication threshold, so as to maintain the continuity of task data transmission.

[0030] The energy balance controller is used to plan a return path for any search and rescue platform when its remaining endurance is lower than a preset return threshold.

[0031] The formation reconfiguration trigger is used to reassign tasks to the remaining platforms when any search and rescue platform fails or leaves the formation for a preset time.

[0032] The present invention is further configured to include a federated learning target recognition module, which enables heterogeneous search and rescue platforms to train target recognition models using local sensor data on their respective local computing units, upload model parameters or gradients to a central server to aggregate and generate a global model, and distribute it to each platform to improve recognition accuracy and environmental adaptability.

[0033] The present invention is further configured to include a two-stage target confirmation mechanism of focusing and verification, used in the first stage to control the aerial search and rescue platform to perform hovering observation and motion vector analysis after detecting a suspected target;

[0034] In Phase Two, the surface or underwater search and rescue platform is controlled to perform convergence verification based on the results of motion vector analysis; when at least two different types of sensors confirm the target and meet the preset spatiotemporal consistency conditions, it is determined as a confirmed discovery;

[0035] The mechanism achieves high-confidence confirmation of suspected targets through multimodal verification across air, surface, and underwater platforms, thereby reducing the false detection rate.

[0036] A method for area coverage patrol of a search and rescue boat includes the following steps:

[0037] S1. Real-time acquisition and fusion of status data of heterogeneous search and rescue platforms and multi-source environmental data of the search and rescue area. The heterogeneous search and rescue platforms include at least two of the following: surface, underwater and air platforms.

[0038] S2. Generate and dynamically update the target existence probability distribution map, and calculate the comprehensive search value of candidate search actions;

[0039] S3. Based on the principle of maximizing the comprehensive search value, plan collaborative patrol routes for heterogeneous search and rescue platforms;

[0040] S4. Control the execution path of heterogeneous search and rescue platforms, and adaptively adjust the path or formation configuration based on platform status data and communication link quality data;

[0041] S5. Repeat steps S2-S4.

[0042] The present invention is further configured such that the prediction and evaluation steps include: generating and updating the probability distribution map using a neural network-enhanced particle filter, and calculating the comprehensive search value.

[0043] The planning step employs a distributed model predictive control algorithm to solve for the cooperative cruise path;

[0044] The execution and adjustment steps also include real-time monitoring of the information coverage of the searched high-probability areas. When the coverage reaches the preset coverage threshold and no target is found, the search area boundary is automatically expanded and the probability threshold of the high-probability areas is adjusted.

[0045] Compared with the shortcomings of the prior art, the beneficial effects of the present invention are as follows:

[0046] By establishing a comprehensive search value model, which deeply couples physical environment, platform cost, and multi-platform collaboration into the calculation of information gain, the decision-making basis for path planning shifts from theoretical optimality to engineering optimality.

[0047] It solves engineering problems such as communication interruption and power depletion, ensuring the system can operate in harsh environments; while the focus-verification module fully leverages the unique advantages of the air-water-submarine heterogeneous platform, combining the wide-area reconnaissance capabilities of UAVs with the close-range verification capabilities of USVs / AUVs, greatly reducing the false alarm rate. Attached Figure Description

[0048] Figure 1 This is a system framework diagram of the present invention;

[0049] Figure 2 A graph showing the change in entropy of the search area over time;

[0050] Figure 3 A graph showing the cumulative probability of discovery (POD) changing over search and rescue time;

[0051] Figure 4 A comparative curve of the system's cumulative energy consumption over search and rescue time;

[0052] Figure 5 This is a comparison chart of the comprehensive performance indicators (TTP, ECA, FPR) of ablation experiments. Detailed Implementation

[0053] Reference Figure 1 Further explanation of embodiments of the present invention is provided, which offer a search and rescue boat area coverage patrol system for inland waterway new energy vessel accidents. This system, through the tight coupling of four core modules and the integration of advanced perception and recognition algorithms, achieves a complete closed loop from data fusion to autonomous decision-making. The overall workflow of the system is as follows:

[0054] The data fusion module collects environmental data and platform status data in real time from sensors deployed in the inland waterway environment and on various heterogeneous platforms (UAV, USV, AUV). This data is sent to the system's drift prediction and information gain evaluation module. On the one hand, this module uses a neural network-enhanced particle filter to predict the drift trajectory of accident targets with high precision and generate a dynamic target existence probability distribution map. On the other hand, it innovatively calculates a comprehensive search value V, which is a complex indicator that integrates information gain, environmental effectiveness, synergistic effect and platform cost.

[0055] The cross-domain collaborative path planning module acquires the V value and, with the sole objective of maximizing the comprehensive search value, uses a distributed model predictive control algorithm to calculate the optimal collaborative cruise path for all platforms. The adaptive control execution module is responsible for controlling the platforms to strictly execute this path and proactively handling unexpected situations such as communication interruptions and energy consumption alarms during execution to ensure mission robustness. Meanwhile, the federated learning target recognition module and the focus-verification two-stage target confirmation mechanism, as high-level perception units of the system, work in parallel, greatly improving the accuracy and anti-interference capability of target recognition.

[0056] The data fusion module forms the perception foundation of the entire system. Its multi-source sensor interfaces are configured to access specific data streams specific to inland waterway environments, including infrared thermal imaging and lidar from UAVs, multi-beam sonar from USVs, and side-scan sonar and ADCP from AUVs. Its platform status monitoring submodule acquires real-time alarms for each platform when remaining endurance is below 25%, communication link quality data below -100dBm, and positioning accuracy data with a PDOP value less than 2.0 via the MQTT protocol or CAN bus. All these heterogeneous data are processed by the data preprocessing unit, which preferably uses extended Kalman filtering for smoothing and employs PTP to synchronize all data streams to a unified spatiotemporal reference with an error no greater than 50ms.

[0057] The drift prediction and information gain evaluation modules are fully implemented in one paragraph as follows:

[0058] To address the pain point of inaccurate prediction of complex inland waterway hydrology, the drift prediction submodule employs a neural network-enhanced particle filter.

[0059] The neural network is a Long Short-Term Memory (LSTM) network. It is trained using historical inland waterway hydrological data and real trajectory data of known floating objects, with the previous time step as its input. and current environmental data (e.g., wind field, upstream reservoir flow), the output is the predicted state for the next time step. In implementation, the prediction step of the particle filter is replaced by calling this pre-trained LSTM model to generate new particle positions: ,in This is process noise. This method allows for a better fit of the nonlinear dynamics of the inland river to the dynamic estimation of the target position and drift trajectory. After the NN-PF generates a high-precision target existence probability distribution map, the information gain evaluation submodule begins to work. A weighted calculation model for the comprehensive search value V is established, and its mathematical model is as follows:

[0060]

[0061] in, It is the final score of platform k executing candidate action j. It is the basic information gain, which is calculated using the standard Shannon entropy formula:

[0062] This is used to answer how much uncertainty can be theoretically reduced for action j. It is an environmental perception effectiveness factor. UAV visible light camera at night A value close to 0 corrects for the "actual availability" of information gain. It is a heterogeneous synergistic gain factor used to quantify the synergistic value brought by UAV's wide-area scanning to guide the accurate detection and strike of AUV. Its calculation method can be the contribution to the expected future information gain of other platforms, i.e. . and This refers to the energy consumption and time cost factors required to perform the action. These are weighting coefficients that can be adjusted according to the task mode. For example, in a time-priority implementation, the weights... This will be set to the maximum value; in the energy-optimized implementation, the weight... This will be set to the maximum value. Finally, this submodule outputs the comprehensive search value V to the cross-domain collaborative path planning module as the control input for search path priority ranking and decision-making.

[0063] The cross-domain collaborative path planning module receives a list of comprehensive search values ​​V from the evaluation module and uses this as its sole objective function. In this embodiment, this module preferably employs a distributed model predictive control algorithm. The objective function of this algorithm is to maximize the cumulative comprehensive search value V of all platforms k over the next N time steps:

[0064]

[0065] When solving this objective function, the algorithm strictly adheres to the constraints:

[0066] Platform dynamics model: That is, the kinematic and dynamic equations of the platform;

[0067] Energy consumption constraints: That is, after completing the planned route, the remaining battery power must be greater than the return threshold;

[0068] Minimum safe distance (obstacle avoidance): This refers to platform collision prevention;

[0069] Geofencing constraints: This means that flying over bridges or restricted airspace is prohibited. In this way, the calculated cooperative cruise path is physically accessible, safe, and has the highest overall value.

[0070] The adaptive control execution module is an autonomous safeguard unit that ensures the system's robustness in harsh inland waterway environments. Its three sub-functions work in concert: the communication quality assessment sub-module continuously monitors link quality; if a packet loss rate exceeding 15% is detected under bridges, it immediately triggers a switch to a preset communication protocol mode, such as automatically downgrading from high-bandwidth 5G to low-bandwidth LoRa, and adjusts the data compression rate to maintain the continuity of mission data transmission; the energy consumption balancing controller, upon detecting that the AUV's remaining endurance is below 25%, immediately plans a return path for it, causing it to leave the formation; at this point, the formation reconfiguration trigger detects that the AUV has been out of formation for 60 seconds and immediately triggers a task redistribution to the remaining platforms, ensuring that the search and rescue mission is not interrupted due to a single platform's departure.

[0071] The federated learning target recognition module provides AI recognition capabilities for this system, particularly suitable for scenarios with limited inland waterway communication. Its working principle is as follows: heterogeneous search and rescue platforms use local sensor data to infer and fine-tune a pre-set target recognition model on their respective local computing units. Instead of uploading raw, massive image data, each platform uploads model parameters or gradients to a central server. The server uses an algorithm to aggregate the gradients from all platforms, generating a global model that incorporates the "experience" of all platforms, and then distributes it to each platform in a continuous cycle to improve recognition accuracy and environmental adaptability.

[0072] "Focus-Validation" is used to reduce false positive rates, and its workflow is as follows:

[0073] In Phase One, the aerial search and rescue platform detected a suspected thermal anomaly at an altitude of 50 meters using thermal imaging. It immediately locked onto the target and conducted hovering observations and motion vector analysis to estimate its drift trajectory.

[0074] In Phase Two, the system automatically dispatches surface or underwater search and rescue platforms to perform convergence verification based on motion vector analysis results. For example, a USV travels downstream of the target to verify the presence of a leak using gas sensors, while an AUV dives to verify whether it is a sunken heavy object using sonar. The system only determines a confirmed discovery when at least two different types of sensors confirm the target and the preset spatiotemporal consistency conditions are met.

[0075] Corresponding to the above system, the search and rescue boat area coverage patrol method of the present invention has also been fully implemented. The method includes:

[0076] First, real-time acquisition and fusion of inland waterway environmental and status data from all platforms.

[0077] Next, in the prediction and evaluation step, a neural network-enhanced particle filter is used to generate and update the target presence probability distribution map. Following this, the comprehensive search value V for all candidate actions is calculated. This value is calculated at least in part based on the information gain entropy reduction calculated using the Shannon entropy correlation formula, and incorporates environmental, cooperative, and cost factors. The subsequent planning step, based on the principle of maximizing the comprehensive search value V, uses a distributed model predictive control algorithm to solve the cooperative cruising path for all heterogeneous platforms. In the execution and adjustment step, the platform executes the path and adaptively adjusts the path or formation based on the platform's state. This step also includes real-time monitoring of the information coverage of high-probability searched areas. When the coverage reaches a preset coverage threshold and no target is found, the system automatically expands the search area boundary and adjusts the probability threshold of the high-probability area. Finally, the system iteratively executes the prediction and evaluation, planning, and execution and adjustment steps, forming a dynamic, closed-loop, and highly intelligent search and rescue process.

[0078] Comparative Experiment: The experiment was conducted in a high-fidelity simulated inland waterway environment (winding river section, water flow velocity 1-3 m / s), simulating the drifting event of a new energy ship's battery pack (continuously releasing heat) after falling into the water. All experiments were run 1000 times using the Monte Carlo method to obtain statistically significant results.

[0079] The experimental and control groups were set up as follows:

[0080] The embodiments of this invention adopt the complete technical solution of this invention, namely NN-PF prediction + DMPC (based on V model) planning + focusing - verification and confirmation.

[0081] Comparative Example 1 (Prediction Module Ablation): This group is used to verify the necessity of the NN-PF of the present invention. This group uses a standard particle filter (PF) (without LSTM neural network enhancement) to replace the NN-PF, while other modules (V model, DMPC, focus-verification) are consistent with the embodiments of the present invention.

[0082] The study aims to verify whether the PF standard's inability to accurately fit complex (nonlinear) inland river flows will cause search and rescue operations to fail at their root.

[0083] Comparative Example 2 (Decision Module Ablation): This group was used to verify the necessity of the comprehensive search value V model of the present invention. This group used standard information gain ΔH to replace the comprehensive search value V as the sole decision criterion for DMPC, i.e., ignoring environmental effectiveness (…). ), Synergistic gain ( ) and cost (C). Verify theoretical optimality (only) Is the decision made by the engineer inferior to the optimal decision-making process?

[0084] Comparative Example 3 (Verification Module Ablation): This group was used to verify the focusing-verification mechanism of the present invention. This group used a single sensor threshold (e.g., reporting if UAV infrared > threshold) to replace the two-stage multimodal confirmation mechanism of focusing-verification. The verification aimed to determine whether the lack of multimodal confirmation would lead to a sharp increase in the false alarm rate (FPR), thereby wasting valuable platform resources and ultimately delaying search and rescue time (TTP).

[0085] The experimental results were analyzed after 1000 Monte Carlo simulations, as shown in the attached figure. Figure 1 The comparison curves showing the change in search area entropy over search and rescue time clearly demonstrate the efficiency of different schemes in reducing system uncertainty (entropy). The entropy value decreases most rapidly in this embodiment of the invention, thanks to its V-model consistently guiding the platform to perform the highest-value actions. Comparative Example 2, due to the use of only... The decision-making process is suboptimal, and the entropy decreases at a significantly slower rate than in this invention.

[0086] However, the most crucial comparison is with scale 1, where the entropy value decreases extremely slowly. This decisively proves our prediction: because the standard PF cannot accurately predict the complex hydrology of inland rivers, the PDM it generates is flawed from the outset. An incorrect PDM renders all subsequent information gain calculations and path planning meaningless, even the most advanced V-model is powerless. This confirms the absolute necessity of using NN-PF as the basis for high-precision prediction in this invention.

[0087] As attached Figure 2 The cumulative probability of discovery (POD) curves shown in the comparison curves of search and rescue time show that the POD curve of the embodiment of the present invention rises the fastest, reaching 100% discovery on average at around 28.3 minutes, which is significantly faster than all the comparative examples.

[0088] This advantage is in the appendix Figure 3 The comparison curves showing the changes in cumulative system energy consumption over search and rescue time further confirm this. The energy consumption curve of this embodiment of the invention reaches the plateau period earliest (i.e., the task is completed and energy consumption no longer increases), and the total energy consumption (ECA) is the lowest (4520kJ).

[0089] Most noteworthy is the comparison between this invention and Comparative Example 2 (standard). Comparison: The decision model in Comparative Example 2 does not consider cost (C) and environmental effectiveness (C). This led to many foolish decisions that appeared to have high information gain but were extremely costly in engineering (e.g., repeatedly sending UAVs to conduct visual searches in dense fog, or sending low-power platforms to perform long-distance missions). As a result, although it eventually completed the mission, its time-to-output ratio (TTP) (42.1 minutes) and energy consumption rate (ECA) were far higher than those of this invention. This demonstrates that the integrated search value (V) model of this invention is the core of achieving "engineering optimality" rather than "theoretical optimality."

[0090] As attached Figure 4 The chart showing the comparison of comprehensive performance indicators (TTP, ECA, FPR) in the ablation experiment visually summarizes all indicators. This embodiment of the invention achieves the lowest possible TTP (search and rescue time) and ECA (total energy consumption).

[0091] In particular, regarding the false positive rate (FPR), Comparative Example 3 (single verification) has an FPR as high as 35.0%, while the embodiment of this invention has an FPR of only 8.5%. This indicates that the system of Comparative Example 3 wastes a significant amount of valuable UAV and USV resources chasing and verifying "false targets" (such as water surface reflections and driftwood). The focus-verification mechanism of this invention, through multimodal, two-stage cross-verification using UAVs and USVs / AUVs, filters out the vast majority of false alarms early on, ensuring that search and rescue resources are always used for the highest-value missions, thereby guaranteeing the overall optimization of TTP and ECA.

[0092] The ablation experimental data of this invention strongly demonstrate that its superior performance does not stem from a single technology, but rather from the synergistic gains of its core components. NN-PF provides an accurate predictive foundation, the Integrated Search Value (V) model provides the most efficient decision-making engine, and the focus-verification mechanism provides the most reliable execution guarantee. All three are indispensable, together constructing an advanced system that offers significant advantages over existing technologies in search and rescue efficiency, energy consumption control, and robustness.

[0093] 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 search and rescue boat area coverage cruise system, characterized in that, include: The data fusion module is used to acquire and fuse real-time status data of heterogeneous search and rescue platforms and multi-source environmental data of the search and rescue area. The heterogeneous search and rescue platforms include at least two of the following: surface, underwater and aerial platforms. The drift prediction and information gain assessment module, which is connected to the data fusion module, is used to: generate and dynamically update the target existence probability distribution map based on the fused data, and calculate the comprehensive search value of the heterogeneous search and rescue platform to perform candidate search actions; The cross-domain collaborative path planning module, which is connected to the drift prediction and information gain evaluation module, is used to receive the comprehensive search value and plan a collaborative cruise path for heterogeneous search and rescue platforms based on the principle of maximizing the comprehensive search value. The adaptive control execution module, connected to the cross-domain collaborative path planning module, is used to control the heterogeneous search and rescue platforms to execute collaborative cruise paths and adaptively adjust the formation configuration or task allocation based on platform status data and communication link quality data.

2. The search and rescue boat area coverage cruise system according to claim 1, characterized in that, The data fusion module includes: A multi-source sensor interface for acquiring at least one of wind field data, water flow data, image data, and underwater flow field data; The platform status monitoring submodule is used to obtain data on the remaining battery life, communication link quality, and positioning accuracy of heterogeneous search and rescue platforms. The data preprocessing unit is used to synchronize environmental data and platform status data to a unified spatiotemporal reference.

3. The search and rescue boat area coverage cruise system according to claim 2, characterized in that, The drift prediction and information gain evaluation module includes: The drift prediction submodule uses a particle filter based on neural network enhancement to dynamically estimate the target position and drift trajectory. The neural network is used to optimize the particle weight update and resampling process.

4. A search and rescue boat area coverage cruise system according to claim 3, characterized in that, The drift prediction and information gain evaluation module includes: The information gain evaluation submodule is used to calculate the overall search value, which is weighted based on the following parameters: Information gain entropy reduction based on probability distribution plot; Environmental perception effectiveness factors related to environmental data in the search and rescue area; Heterogeneous collaborative gain factor characterizing the contribution of current actions to the future search performance of other heterogeneous platforms; Energy and time cost factors required to perform candidate search actions; It also outputs the comprehensive search value to the cross-domain collaborative path planning module as a control input for search path priority ranking and decision-making.

5. A search and rescue boat area coverage cruise system according to claim 4, characterized in that, The cross-domain collaborative path planning module adopts a distributed model predictive control algorithm and solves the collaborative cruise path based on the platform's dynamic model, energy consumption constraints, and minimum safe distance constraints.

6. A search and rescue boat area coverage cruise system according to claim 5, characterized in that, The adaptive control execution module includes: The communication quality assessment submodule is used to trigger the switching of the preset communication protocol mode or adjust the data compression rate when the communication link quality is lower than the preset communication threshold, so as to maintain the continuity of task data transmission. The energy balance controller is used to plan a return path for any search and rescue platform when its remaining endurance is lower than a preset return threshold. The formation reconfiguration trigger is used to reassign tasks to the remaining platforms when any search and rescue platform fails or leaves the formation for a preset time.

7. A search and rescue boat area coverage cruise system according to claim 6, characterized in that, This includes a federated learning target recognition module: which enables heterogeneous search and rescue platforms to train target recognition models using local sensor data on their respective local computing units, upload model parameters or gradients to a central server to aggregate and generate a global model, and then distribute it to each platform to improve recognition accuracy and environmental adaptability.

8. A search and rescue boat area coverage cruise system according to claim 7, characterized in that, This includes a two-stage target confirmation mechanism of focusing and verification, which is used in the first stage to control the aerial search and rescue platform to perform hovering observation and motion vector analysis after detecting a suspected target; In Phase Two, the surface or underwater search and rescue platform is controlled to perform convergence verification based on the results of motion vector analysis; when at least two different types of sensors confirm the target and meet the preset spatiotemporal consistency conditions, it is determined as a confirmed discovery; The mechanism achieves high-confidence confirmation of suspected targets through multimodal verification across air, surface, and underwater platforms, thereby reducing the false detection rate.

9. A method for area coverage patrol of a search and rescue boat, characterized in that, Includes the following steps: S1. Real-time acquisition and fusion of status data of heterogeneous search and rescue platforms and multi-source environmental data of the search and rescue area. The heterogeneous search and rescue platforms include at least two of the following: surface, underwater and air platforms. S2. Generate and dynamically update the target existence probability distribution map, and calculate the comprehensive search value of candidate search actions; S3. Based on the principle of maximizing the comprehensive search value, plan collaborative patrol routes for heterogeneous search and rescue platforms; S4. Control the execution path of heterogeneous search and rescue platforms, and adaptively adjust the path or formation configuration based on platform status data and communication link quality data; S5. Repeat steps S2-S4.

10. A method for area coverage patrol of a search and rescue boat according to claim 9, characterized in that: The prediction and evaluation steps include: generating and updating the probability distribution map using a neural network-enhanced particle filter, and calculating the comprehensive search value. The planning step employs a distributed model predictive control algorithm to solve for the cooperative cruise path; The execution and adjustment steps also include real-time monitoring of the information coverage of the searched high-probability areas. When the coverage reaches the preset coverage threshold and no target is found, the search area boundary is automatically expanded and the probability threshold of the high-probability areas is adjusted.

Images