Maritime rescue method and system based on heterogeneous unmanned system

Abstract

This invention provides a maritime rescue method and system based on a heterogeneous unmanned system. The method includes: Step S1: acquiring accident information to generate a multi-dimensional decision model; the accident information includes accident type, accident location, information of personnel to be rescued, and environmental information; Step S2: automatically matching a corresponding rescue plan based on the multi-dimensional decision model; Step S3: deploying the heterogeneous unmanned system according to the rescue plan, enabling it to perform autonomous unmanned search along the rescue path and identify the searched targets; Step S4: controlling the heterogeneous unmanned system to complete the rescue based on the identification results. This invention effectively solves the main problems existing in current maritime rescue technologies and provides a complete technical solution for achieving efficient, rapid, and intelligent unmanned maritime rescue.

Description

Technical Field

[0001] This invention relates to the field of unmanned water rescue technology, specifically to a maritime rescue method and system based on heterogeneous unmanned systems. Background Technology

[0002] With the rapid development of the marine economy, maritime activities are becoming increasingly frequent, leading to a corresponding increase in the risk of maritime accidents. Maritime rescue demands extremely high timeliness, as the golden rescue window for people in the cold seawater is typically less than one hour. However, traditional maritime rescue methods mainly rely on rescue vessels and helicopters, which have inherent limitations such as slow response time, limited coverage, high susceptibility to sea conditions, and high operating costs. Furthermore, they pose a serious threat to rescue personnel in high-risk environments.

[0003] In recent years, unmanned systems technology has shown great application potential in maritime rescue due to its advantages such as eliminating the need to consider personnel risks, rapid response, and low operating costs. In particular, heterogeneous unmanned systems—integrating different platforms such as unmanned surface vessels, drones, and unmanned lifebuoys—are expected to maximize rescue effectiveness through complementary advantages. However, analysis and research of existing technologies have revealed the following significant shortcomings in the current unmanned rescue field: (1) The limited capabilities of a single platform restrict rescue effectiveness Unmanned aerial vehicles (UAVs) are limited by their endurance and payload capacity, resulting in a limited effective operating radius and making it difficult to perform long-range maritime rescue missions. Unmanned surface vessels (USVs) face a trade-off between range and speed, and their adaptability in harsh sea conditions is challenged. Unmanned lifebuoys are typically used as independent rescue units and lack effective long-range delivery methods. When operating independently, each platform struggles to overcome its inherent technological bottlenecks.

[0004] (2) Insufficient system coordination leads to low overall efficiency. Existing unmanned rescue systems mostly operate independently on a single platform or with simple command coordination, lacking deep intelligent collaboration at the mission level. Different unmanned platforms have failed to form an effective complementary mechanism, thus failing to leverage the synergistic advantages of heterogeneous systems. Particularly in cross-domain collaboration, the collaborative control technology between surface and aerial unmanned platforms is still immature, making it difficult to achieve a qualitative improvement in overall rescue efficiency.

[0005] (3) Poor environmental adaptability affects the success rate of rescue. The existing system is clearly inadequate in adapting to the dynamic marine environment. This is mainly manifested in the following ways: path planning is mostly based on static or predefined rules, failing to fully consider the dynamic impact of environmental factors such as wind and current on the location of the target in the water; the task allocation strategy lacks flexibility and is difficult to optimize in real time according to environmental changes; the system has weak fault tolerance for abnormal situations such as communication link interruptions, and its reliability decreases in environments with limited communication in the open sea.

[0006] (4) Low level of intelligent decision-making leads to disruption of rescue process. The transition from search to rescue relies heavily on manual decision-making, resulting in significant decision-making delays. The system lacks autonomous judgment and process coordination capabilities, disrupting the integrity and coherence of rescue operations. Particularly in multi-target rescue scenarios, the lack of scientific and quantifiable prioritization decision support methods means that the rescue sequence largely depends on the subjective experience of commanders, impacting the overall rescue effectiveness.

[0007] (5) Insufficient integration of the entire process Existing technological solutions often focus on specific stages of the rescue process, lacking a comprehensive, integrated design solution encompassing the entire process from response and rapid deployment to search and rescue. Insufficient coordination and information sharing between stages lead to inefficient use of rescue resources and make it difficult to complete rapid and effective rescue missions within the critical timeframe.

[0008] Patent document CN110203299B discloses a dynamic search and rescue method for a ground-air heterogeneous rescue robot, but it is mainly aimed at land rescue scenarios where drones guide land robots to carry out rescue operations. This method differs significantly from the application scenarios and external environments of this invention and is not applicable to unmanned maritime rescue.

[0009] Patent document CN116165903B discloses an optimal robust control method for a maritime unmanned rescue boat system, including the implementation of a maritime unmanned rescue boat system and an optimal robust control method during the unmanned rescue boat's navigation. The unmanned rescue boat system includes: the unmanned rescue boat body, a power unit, a navigation information input unit, an unmanned rescue boat main control unit, a personal positioning unit, and a remote control unit, realizing automatic navigation rescue, one-key return, and attitude control functions, achieving unmanned intelligent rescue at sea. The optimal robust control method adopts a PID dual closed-loop control method to realize the heading and speed control of the unmanned rescue boat, and optimizes the PID parameters through an improved slime mold optimization algorithm to achieve optimal robust control of the unmanned rescue boat.

[0010] Patent document CN115655279A discloses a path planning method for unmanned maritime rescue airships based on an improved whale algorithm. On the basis of the standard whale optimization algorithm, this method improves the shortcomings of the standard whale algorithm, such as slow convergence speed, poor global optimization ability, and low convergence accuracy, by adopting a nonlinear segmented update strategy for the convergence factor and introducing inertial weight coefficients and random mutation perturbation strategies. This enables the unmanned maritime rescue airship to reach the target position in the shortest time and successfully complete the maritime rescue mission.

[0011] However, both of the above patent documents only address the control and planning algorithms for rescue of a single unmanned system, improving the control accuracy of the single system. They do not reflect the efficiency improvement of joint search and rescue of a submarine-aircraft heterogeneous system, or how to respond quickly and increase the probability of successful rescue.

[0012] Therefore, there is an urgent need for a new maritime rescue technology solution that can overcome the above-mentioned shortcomings and achieve intelligent, adaptive, and integrated operation throughout the entire process. Summary of the Invention

[0013] To address the shortcomings of existing technologies, the purpose of this invention is to provide a maritime rescue method and system based on heterogeneous unmanned systems.

[0014] A maritime rescue method based on a heterogeneous unmanned system provided by the present invention includes: Step S1: Obtain accident information to generate a multi-dimensional decision model; The accident information includes the accident type, accident location, information on people needing rescue, and environmental information; Step S2: Based on the multi-dimensional decision-making model, automatically match the corresponding rescue plan; Step S3: Deploy a heterogeneous unmanned system according to the rescue plan, enabling it to conduct unmanned autonomous search along the rescue route and identify the searched targets; Step S4: Based on the identification results, control the heterogeneous unmanned system to complete the rescue.

[0015] Preferably, the heterogeneous unmanned system includes: The unmanned boat serves as a mother ship and command center, equipped with a pre-precision GNSS / INS integrated navigation system, radar, weather sensors, and high-power communication relay equipment; its deck is equipped with a drone take-off and landing platform and a dedicated release mechanism for unmanned lifebuoys; The drone is used as an area search and reconnaissance unit, equipped with a dual-light pod for visible light and thermal imaging, as well as an onboard AI processing unit. Unmanned lifebuoys, used as the final rescue execution unit, include vital sign detection sensors and voice communication equipment; The system is uniformly scheduled through a multi-level collaborative controller, which integrates a target drift estimation system, a dynamic path planner, and an auction-based task assigner.

[0016] Preferably, a corresponding collaborative rescue route is planned based on the type of accident.

[0017] Preferably, the multi-dimensional decision-making model includes: S total = α f 距离 ( d )+ β g 海况 (s )+ γ h 资源 ( r )+ δ k 紧急性 ( u ) The weighting coefficients satisfy the normalization condition: α+β+γ+δ=1; The rescue plan includes a long-distance rescue plan, a medium-distance rescue plan, and a short-distance rescue plan; The decision-making rules for the proposed solutions are as follows: when In such cases, choose a long-distance rescue plan; when In such cases, choose a medium-range rescue plan; when In such cases, choose a close-range rescue option; in, This indicates the total score based on the overall urgency level. Represents the weighting coefficient; f 距离 ( d ) represents the distance coefficient function. g 海况 ( s ) represents the sea state coefficient function. h 资源 ( r ) represents the resource availability coefficient function. k 紧急性 ( u ) represents the emergency urgency coefficient function.

[0018] Preferably, step S3 includes: Establish a system health assessment based on a state-space model:

[0019] in Indicates the health status of each subsystem. These are the weighting coefficients; The inspection items include: -In-depth energy system diagnostics: Battery pack state of charge assessment ; - Communication system link quality test: Signal-to-noise ratio ; - Navigation system accuracy calibration: Positioning error ; -Task payload functionality verification: Sensor response function test; - Propulsion system evaluation: Comprehensive evaluation of motor, thruster, and energy conversion efficiency; Once all checks are passed, the system automatically enters the ready state and establishes a deployment ready matrix.

[0020] Preferably, step S3 further includes: Intelligent initial path planning: The system employs a multi-objective optimization path planning algorithm to construct the cost function:

[0021] in For the cost of time, For the cost of energy, As a risk cost, the weighting coefficients satisfy... ; Environmental factors modeling includes the effects of ocean currents and wind resistance; Obstacle avoidance uses the potential field method:

[0022] in, Indicates the first Gain coefficient of each obstacle Indicates the location The value of the repulsive potential field generated by all obstacles at that location; Indicates the current position With the Obstacle locations The Euclidean distance between them; Dynamic path replanning mechanism: The system establishes a periodic path optimization loop, and the replanning trigger conditions are based on multi-factor evaluation:

[0023] in, Indicates the threshold for changes in the target location; Indicates the threshold for changes in environmental velocity; Indicates the maximum time interval threshold; This represents the Euclidean distance between the old and new target locations; and These represent the environmental velocity vectors at the current time and the previous planning time, respectively. This indicates the time interval since the last path planning was completed (in seconds or minutes).

[0024] Preferably, the search process for heterogeneous unmanned systems includes: Precisely defined search area: Based on the target motion uncertainty model, the search radius is calculated as follows:

[0025] Among them, safety factor ; The region where the probability exists is modeled using a Gaussian mixture model:

[0026] in, This represents the weight of the i-th Gaussian distribution. Let represent the mean of the i-th Gaussian distribution. Represents the covariance matrix. This represents the number of Gaussian distributions. Indicates spatial location, and N represents a Gaussian distribution; Search priorities are allocated based on the principle of maximizing information gain. Distributed intelligent task allocation: The task allocation model employs a multi-round auction mechanism. The first round of coarse-grained region allocation is based on platform capability matching.

[0027] The second round of fine-grained grid allocation employs a cost-based bidding strategy:

[0028] in ; Represents the distance cost component. Indicates the load cost component; Adaptive intelligent search execution: The drone search mode employs path planning based on information entropy:

[0029] in, This represents all the grid cells that the drone sequentially passes through in its planned search path; Indicates the information gain of the grid; Indicates the time decay factor. The constant represents the decay rate, and t represents the time variable. The unmanned surface vessel's search strategy integrates data from multiple sensors, and the detection probability model is as follows:

[0030] Among them, P 传感器i Indicates the first The independent detection probability of each sensor.

[0031] Preferably, step S4 includes: Rescue efforts initiated intelligent decision-making: Confidence fusion based on DS evidence theory:

[0032] in, and These are from two different sources of evidence. and Jiao Yuan, This is a new proposition to consider after integration; Time consistency verification uses sliding window matching:

[0033] in, Indicates the current moment. Indicates the duration of the sliding window. Used to indicate at time Does the condition for detecting a continuous positive case meet? From arrive Discrete or continuous time points; Conditions for initiating rescue operations:

[0034] in, Indicates the reliability threshold. Indicates the time consistency threshold. Indicates the spatial consistency threshold. This represents the threshold for the minimum number of independent detections. Multi-objective prioritization scientific assessment: Priority scoring based on multi-attribute utility theory:

[0035] The detailed definitions of each utility function are as follows: Life state utility:

[0036] in, This represents the weight coefficient of each sub-item. An indicator variable representing whether or not a life jacket is being worn. Indicator variable representing state of consciousness Indicator variables representing limb mobility; Environmental threat effects:

[0037] in, Indicates seawater temperature, Indicates the significant wave height. Indicates wind speed; Distance cost-utility:

[0038] in, Indicates the target's current position. Indicates the current location of the rescue platform. This represents the Euclidean distance between the two. Indicates the normalized maximum distance; The effect of time urgency: .

[0039] in, This indicates the elapsed time from the occurrence of the accident to the present moment. Indicates the critical survival time.

[0040] A maritime rescue system based on a heterogeneous unmanned system, provided by the present invention, includes: Module M1: Acquires accident information to generate a multi-dimensional decision model; The accident information includes the accident type, accident location, information on people needing rescue, and environmental information; Module M2: Based on the multi-dimensional decision-making model, automatically match the corresponding rescue plan; Module M3: Deploys a heterogeneous unmanned system according to the rescue plan, enabling it to conduct unmanned autonomous search along the rescue route and identify the searched targets; Module M4: Based on the recognition results, control the heterogeneous unmanned system to complete the rescue.

[0041] Preferably, the heterogeneous unmanned system includes: The unmanned boat serves as a mother ship and command center, equipped with a pre-precision GNSS / INS integrated navigation system, radar, weather sensors, and high-power communication relay equipment; its deck is equipped with a drone take-off and landing platform and a dedicated release mechanism for unmanned lifebuoys; The drone is used as an area search and reconnaissance unit, equipped with a dual-light pod for visible light and thermal imaging, as well as an onboard AI processing unit. Unmanned lifebuoys, used as the final rescue execution unit, include vital sign detection sensors and voice communication equipment; The system is uniformly scheduled through a multi-level collaborative controller, which integrates a target drift estimation system, a dynamic path planner, and an auction-based task assigner.

[0042] Preferably, a corresponding collaborative rescue route is planned based on the type of accident.

[0043] Preferably, the multi-dimensional decision-making model includes: S total = α f距离 ( d )+ β g 海况 ( s )+ γ h 资源 ( r )+ δ k 紧急性 ( u ) The weighting coefficients satisfy the normalization condition: α+β+γ+δ=1; The rescue plan includes a long-distance rescue plan, a medium-distance rescue plan, and a short-distance rescue plan; The decision-making rules for the proposed solutions are as follows: when In such cases, choose a long-distance rescue plan; when In such cases, choose a medium-range rescue plan; when In such cases, choose a close-range rescue option; in, This indicates the total score based on the overall urgency level. Represents the weighting coefficient; f 距离 ( d ) represents the distance coefficient function. g 海况 ( s ) represents the sea state coefficient function. h 资源 ( r ) represents the resource availability coefficient function. k 紧急性 ( u ) represents the emergency urgency coefficient function.

[0044] Preferably, the module M3 includes: Establish a system health assessment based on a state-space model:

[0045] in Indicates the health status of each subsystem. These are the weighting coefficients; The inspection items include: -In-depth energy system diagnostics: Battery pack state of charge assessment ; - Communication system link quality test: Signal-to-noise ratio ; -Navigation system accuracy calibration: Positioning error ; -Task payload functionality verification: Sensor response function test; - Propulsion system evaluation: Comprehensive evaluation of motor, thruster, and energy conversion efficiency; Once all checks are passed, the system automatically enters the ready state and establishes a deployment ready matrix.

[0046] Preferably, the module M3 further includes: Intelligent initial path planning: The system employs a multi-objective optimization path planning algorithm to construct the cost function:

[0047] in For the cost of time, For the cost of energy, As a risk cost, the weighting coefficients satisfy... ; Environmental factors modeling includes the effects of ocean currents and wind resistance; Obstacle avoidance uses the potential field method:

[0048] in, Indicates the first Gain coefficient of each obstacle Indicates the location At that point, the repulsive potential field value generated by all obstacles; Indicates the current position With the Obstacle locations The Euclidean distance between them; Dynamic path replanning mechanism: The system establishes a periodic path optimization loop, and the replanning trigger conditions are based on multi-factor evaluation: .

[0049] in, Indicates the threshold for changes in the target location; Indicates the threshold for changes in environmental velocity; Indicates the maximum time interval threshold; This represents the Euclidean distance between the old and new target locations; and These represent the environmental velocity vectors at the current time and the previous planning time, respectively. This indicates the time interval since the last path planning was completed (in seconds or minutes).

[0050] Preferably, the search process for heterogeneous unmanned systems includes: Precisely defined search area: Based on the target motion uncertainty model, the search radius is calculated as follows:

[0051] Among them, safety factor ; The region where the probability exists is modeled using a Gaussian mixture model:

[0052] in, This represents the weight of the i-th Gaussian distribution. Let represent the mean of the i-th Gaussian distribution. Represents the covariance matrix. This represents the number of Gaussian distributions. Indicates spatial location, and N represents a Gaussian distribution; Search priorities are allocated based on the principle of maximizing information gain. Distributed intelligent task allocation: The task allocation model employs a multi-round auction mechanism. The first round of coarse-grained region allocation is based on platform capability matching.

[0053] The second round of fine-grained grid allocation employs a cost-based bidding strategy:

[0054] in ; Represents the distance cost component. Indicates the load cost component; Adaptive intelligent search execution: The drone search mode employs path planning based on information entropy:

[0055] in, This represents all the grid cells that the drone sequentially passes through in its planned search path; Indicates the information gain of the grid; Indicates the time decay factor. The constant represents the decay rate, and t represents the time variable. The unmanned surface vessel's search strategy integrates data from multiple sensors, and the detection probability model is as follows:

[0056] Among them, P 传感器i Indicates the first The independent detection probability of each sensor.

[0057] Preferably, the module M4 includes: Rescue efforts initiated intelligent decision-making: Confidence fusion based on DS evidence theory:

[0058] in, and These are from two different sources of evidence. and Jiao Yuan, This is a new proposition to consider after integration; Time consistency verification uses sliding window matching:

[0059] in, Indicates the current moment. Indicates the duration of the sliding window. Used to indicate at time Does the condition for detecting a continuous positive case meet? From arrive Discrete or continuous time points; Conditions for initiating rescue operations:

[0060] in, Indicates the reliability threshold. Indicates the time consistency threshold. Indicates the spatial consistency threshold. This represents the threshold for the minimum number of independent detections. Multi-objective prioritization scientific assessment: Priority scoring based on multi-attribute utility theory:

[0061] The detailed definitions of each utility function are as follows: Life state utility:

[0062] in, This represents the weight coefficient of each sub-item. An indicator variable representing whether or not a life jacket is being worn. Indicator variables representing states of consciousness Indicator variables representing limb mobility; Environmental threat effects:

[0063] in, Indicates seawater temperature, Indicates the significant wave height. Indicates wind speed; Distance cost-utility:

[0064] in, Indicates the target's current position. Indicates the current location of the rescue platform. This represents the Euclidean distance between the two. Indicates the normalized maximum distance; The effect of time urgency: .

[0065] in, This indicates the elapsed time from the occurrence of the accident to the present moment. Indicates the critical survival time.

[0066] Compared with the prior art, the present invention has the following beneficial effects: 1. Significantly improved rescue efficiency: This invention effectively overcomes the limitations of a single platform through a heterogeneous collaborative system architecture, expanding the rescue radius several times over and increasing the response speed by more than 300%, comprehensively covering rescue scenarios at near, medium and far distances.

[0067] 2. The system's intelligence level is significantly improved; through a dynamic intelligent decision-making mechanism, the system can actively adapt to the dynamic maritime environment, achieve proactive tracking of moving targets, improve the accuracy of reaching the destination by more than 60%, and greatly reduce subsequent search time.

[0068] 3. Significantly improved collaborative efficiency: Based on the auction-based task allocation mechanism, this invention achieves true distributed, self-organizing collaborative search, improving search coverage and efficiency per unit time by approximately 40%, and enhancing system robustness.

[0069] 4. The rescue process is more coherent and efficient; through an intelligent process switching mechanism, this invention reduces the decision-making time from discovery to the start of rescue from minutes to seconds, ensuring the timeliness and coherence of rescue operations.

[0070] 5. Enhanced ability to cope with complex scenarios: The multi-target rescue priority decision-making model in this invention provides a scientific basis for complex rescue scenarios, and the overall rescue probability is expected to increase by more than 25% in scenarios where multiple people fall into the water.

[0071] 6. Integrated design throughout the entire process; This invention achieves seamless connection of the entire process from response, rush, search to rescue, with close cooperation among all links and full information sharing, which greatly improves the success rate of rescue within the golden time.

[0072] 7. Through the above-mentioned technological innovations, this invention effectively solves the main problems existing in current maritime rescue technologies and provides a complete technical solution for achieving efficient, rapid, and intelligent unmanned maritime rescue. Attached Figure Description

[0073] Other features, objects, and advantages of the present invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings: Figure 1 This is a schematic diagram of the overall architecture of the heterogeneous unmanned system in this invention.

[0074] Figure 2 This is a schematic diagram of the heterogeneous unmanned system in this invention.

[0075] Figure 3 This is a flowchart of the rescue method in this invention.

[0076] Figure 4 This is the wide-area joint search path planning diagram in this invention.

[0077] Figure 5 This is the path planning diagram for the joint search along the track in this invention.

[0078] Figure 6 This is a flowchart of the method of the present invention.

[0079] In the picture: Unmanned lifebuoy 1, drone sensing area 7 The rescue system search mission location is at coordinates 101, unmanned boat path 8. Unmanned Ark 2 drone path 9 The total area searched by three drones was 10. Unmanned vehicle search path starting point 4 search path 11 Drone search path starting point 5 drift trajectory 12 Unmanned vehicle sensing area 6 The intersection point where the unmanned boats and drones complete their respective search missions at the same time: 13 Detailed Implementation The present invention will now be described in detail with reference to specific embodiments. These embodiments will help those skilled in the art to further understand the present invention, but do not limit the invention in any way. It should be noted that those skilled in the art can make several changes and improvements without departing from the concept of the present invention. These all fall within the protection scope of the present invention.

[0080] like Figure 6 As shown, a maritime rescue method based on heterogeneous unmanned systems includes: Step S1: Obtain accident information to generate a multi-dimensional decision model; The accident information includes the accident type, accident location, information on people needing rescue, and environmental information; Step S2: Based on the multi-dimensional decision-making model, automatically match the corresponding rescue plan; Step S3: Deploy a heterogeneous unmanned system according to the rescue plan, enabling it to conduct unmanned autonomous search along the rescue route and identify the searched targets; Step S4: Based on the identification results, control the heterogeneous unmanned system to complete the rescue.

[0081] In one embodiment, the above steps specifically include: Step 1: Based on the type of accident rescue target and in conjunction with the current water flow velocity, flow direction, wind speed, and wind direction, estimate the location of the initial accident location after the rescue deployment is in place; Step 2: Generate a multi-dimensional decision model based on the location distance, current sea conditions, number of people to be rescued, and accident type. Automatically match long-range, medium-range, and short-range rescue plans based on the model results and provide prompts to staff. Step 3: Deploy the heterogeneous unmanned systems in place according to the rescue plan; Step 4: Based on the type of accident, plan the corresponding collaborative rescue route; Step 5: Conduct unmanned autonomous search based on the path; Step 6: Identify and determine suspected targets; Step 7: Use a lifebuoy to approach and provide rescue.

[0082] The multi-dimensional decision-making model in step 2 is as follows: S total = α f 距离 ( d )+ β g 海况 ( s )+ γ h 资源 ( r )+ δ k 紧急性 ( u ) The weighting coefficients satisfy the normalization condition: α + β + γ + δ = 1.

[0083] The decision-making rules for the proposed solutions are as follows: 1. When In such cases, choose a long-distance rescue option (1.5km to 5km). 2. When In such cases, choose a medium-distance rescue option (0.2km to 1.5km). 3. When In such cases, choose the nearest rescue option (<0.2km).

[0084] in: This represents the total score based on the overall urgency level, the final score calculated by the system, and its value range. Based on the total score range, the system automatically selects the corresponding rescue plan; for example: Long-distance rescue plan (>50km) Medium-distance rescue plan (20-50km) Close-range rescue plan (<20km) The weighting coefficient represents the relative importance of each influencing factor, and the constraint condition is... .

[0085] In one embodiment, its value is: Distance has the highest weight, reflecting priority given to the "golden rescue time". Sea state is the second most important factor, reflecting environmental risk. Resource availability weight Urgency weight This represents the distance coefficient function, whose input is... The output is the straight-line distance (in kilometers) between the accident site and the rescue base, and the normalized distance urgency level is given, with a value range of... .

[0086] In one embodiment, it takes the form of:

[0087] in The distance is usually set to 100 kilometers; the farther the distance, the closer the coefficient is to 1, indicating that a higher level of rescue response is needed.

[0088] The input represents the sea state coefficient function. The marine environmental status includes multi-dimensional data such as wave height, wind speed, and visibility. The output is an environmental severity score, with a value range of... .

[0089] A typical mapping relationship is as follows: Calm sea conditions (wave height < 1m, wind speed < force 5):

[0090] Moderate sea state (wave height 1-3m, wind speed 5-7):

[0091] Severe sea conditions (wave height > 3m, wind speed > force 8):

[0092] The function representing the resource availability coefficient, input... The current status of available rescue resources, including the number, location, and health of platforms, is displayed, and the output is a resource scarcity score, with a value range of [range missing]. .

[0093] The calculation logic is as follows: When sufficient resources are available. Approaching 0; When resources are scarce, and most platforms are performing tasks or requiring maintenance, Close to 1; The input represents the accident urgency coefficient function. The output is an inherent urgency score for the accident, taking into account factors such as the number of people involved, the type of accident, and the number of casualties. The score ranges from [value range missing]. .

[0094] Typical mapping relationships include: Ship sinking / capsizing:

[0095] Fire / Explosion:

[0096] Collision / Stranded:

[0097] Mechanical failure / loss of power:

[0098] Person fell into the water (location known): (Adjusted based on number of participants) Step 3, deployment and positioning, mainly involves system self-check and dynamic deployment. The system self-check is a fully automated pre-deployment inspection process that establishes a system health assessment based on a state-space model.

[0099] in Indicates the health status of each subsystem. These are the weighting coefficients.

[0100] The inspection items include: 1. In-depth diagnostics of energy systems: Battery pack state of charge assessment ; 2. Communication system link quality test: signal-to-noise ratio ; 3. Navigation system accuracy calibration: positioning error (m: meter, RMS: root mean square, also known as the quadratic mean or effective value, is the square root of the mean square (the arithmetic mean of the squares of a set of numbers). It is the expression for the generalized mean of the second power, and can also be called the second power mean.) 4. Mission payload function verification: Sensor response function test; 5. Propulsion System Evaluation: Comprehensive evaluation of the motor, thruster, and energy conversion efficiency.

[0101] Once all checks are passed, the system automatically enters the "ready state" and establishes a deployment readiness matrix.

[0102] In addition, the key steps in the dynamic sprint phase are initial path planning and dynamic path replanning: 1) Intelligent initial path planning The system employs a multi-objective optimization path planning algorithm to construct the cost function:

[0103] in For the cost of time, For the cost of energy, As a risk cost, the weighting coefficients satisfy... .

[0104] Environmental factors modeling includes: ocean current influence and wind field resistance.

[0105] Obstacle avoidance: using the potential field method

[0106] in: Indicates the location At this location, the repulsive potential field value generated by all obstacles is represented. It is a scalar; the larger the value, the stronger the repulsive force at that location, and the more likely the path planning algorithm will avoid areas with high potential field values.

[0107] Summation symbol All obstacles within the affected area are superimposed. In practical applications, only obstacles with a distance less than a certain threshold can be considered to reduce the computational load.

[0108] Indicates the first The gain coefficient (or intensity coefficient) of an obstacle is a positive real number. It reflects the degree of danger or "repulsion strength" of the obstacle. For example, for a stationary reef, It can be set relatively high; for other dynamic vessels, it can be dynamically adjusted according to their speed and size. .

[0109] Indicates the current position With the Obstacle locations The Euclidean distance between them. The squared term in the denominator reflects the characteristic that the repulsive force decreases sharply with increasing distance—the closer the distance, the larger the potential field value and the stronger the repulsive force; the farther the distance, the faster the effect weakens.

[0110] 2) Dynamic path replanning mechanism The system establishes a periodic path optimization loop, and the replanning trigger conditions are based on multi-factor evaluation:

[0111] in, and These two vectors represent the latest predicted position of the target and the target position used in the last path planning, respectively. The target continues to drift due to ocean currents and wind fields, and its predicted position is constantly updated over time. The difference between these two vectors reflects the magnitude of the change in the target's position.

[0112] Update source: Latest predicted coordinates output by the target drift estimation system every 5-10 minutes.

[0113] This represents the Euclidean distance between the old and new target positions, quantifying the displacement caused by target drift. When the target moves a large distance, the endpoint of the original path becomes meaningless, and replanning is necessary.

[0114] and These represent the environmental velocity vectors at the current time and the previous planning time, respectively. It is a vector combination that includes ocean current speed and wind speed. Ocean currents and wind fields are the main environmental factors affecting the navigation efficiency of unmanned surface vessels, and their changes will alter the optimal path.

[0115] This represents the change in the environmental velocity vector (unit: meters per second), quantifying the magnitude and direction of changes in ocean currents or wind speeds. When the environmental dynamic field changes significantly, the energy efficiency and time advantage of the original path may no longer hold true.

[0116] After entering the path planning and collaborative search phases in steps 4 and 5, the key steps are divided into the following three items: 1. Precise definition of the search area Based on the target motion uncertainty model, the search radius is calculated as follows:

[0117] Among them, safety factor .

[0118] The region where the probability exists is modeled using a Gaussian mixture model:

[0119] in, This represents the weight of the i-th Gaussian distribution. Let represent the mean of the i-th Gaussian distribution. Represents the covariance matrix. This represents the number of Gaussian distributions. Indicates spatial location, and N represents a Gaussian distribution; Search priorities are assigned based on the principle of maximizing information gain.

[0120] 2. Distributed intelligent task allocation A task allocation model employing a multi-round auction mechanism. The first round of coarse-grained region allocation is based on platform capability matching.

[0121] The second round of fine-grained grid allocation employs a cost-based bidding strategy:

[0122] in .

[0123] Let represent the distance to the cost component, and:

[0124] Indicates the current location on the platform. Go to the task grid center The Euclidean distance.

[0125] This represents the normalization factor, typically taken as the platform's maximum effective operating radius, to ensure... The greater the distance, the longer it takes for the platform to reach the mission area, the greater the energy consumption, and the higher the cost.

[0126] Represents the load cost component:

[0127] This indicates the number of tasks currently being undertaken by the platform.

[0128] This indicates the platform's maximum task capacity (e.g., a drone can execute one grid task simultaneously, while an unmanned surface vessel can execute two simultaneously).

[0129] 3. Adaptive intelligent search execution The UAV 3 search mode employs path planning based on information entropy:

[0130] Path revenue refers to the total revenue of the entire search path. It is a dimensionless scalar used to compare the merits of different candidate paths. The higher the revenue, the more valuable information the path can acquire per unit of time, so the system will choose the path with the highest revenue to execute.

[0131] Summation symbol This indicates that all grid cells traversed sequentially along the search path planned by the drone are accumulated.

[0132] The path consists of a series of discretized spatial grids, each representing a basic search unit. Summation means that the total reward of the path equals the sum of the rewards contributed by each grid on the path.

[0133] Information gain represents the information increment obtained by performing a search at a grid cell, i.e., the reduction in uncertainty about the target before and after the search.

[0134] Information gain is based on the concept of entropy in information theory. Let the probability of the target existing in the current grid be... (This can be obtained through a target drift prediction model), then the entropy of the grid is... If the search confirms that the target exists ( (become 1) or does not exist ( If the entropy decreases to 0, the information gain is the entropy value before the search. In practice, due to false alarms and missed detections in sensors, the information gain can be expressed as the expected entropy reduction.

[0135] A grid with a higher information gain indicates a higher level of uncertainty in the area or a higher probability of the target existing. Searching this grid can eliminate uncertainty to the greatest extent possible, so path planning should prioritize covering these grids.

[0136] This represents a weighting factor that decays exponentially, with a range of values. .

[0137] parameter Represents the attenuation constant ( ), to control the decay rate. The larger the value, the faster the decay. The smaller the value, the slower the decay. Its value is usually dynamically adjusted based on the urgency of the task; for example, a larger value can be set during the golden rescue time. This emphasizes searching for high-value areas as early as possible.

[0138] parameter This represents a time variable, typically indicating the time required from the current moment until the drone actually performs the grid search (e.g., flight time from the path's starting point to the grid). In some models, It can also be simplified to sequential numbering on the path, meaning that the later the grid is, the greater the attenuation.

[0139] Because targets may drift over time or their survival probability may decrease, the actual information value of a grid found later decreases due to increased uncertainty. The decay factor reflects this time urgency, prompting UAVs to prioritize searching for grids with high information gain and that can be reached as early as possible, avoiding delaying high-value grids to the end of the path.

[0140] The Unmanned Vulcan 2's search strategy integrates data from multiple sensors, and the detection probability model is as follows:

[0141] P 传感器i Indicates the first The independent detection probability of each sensor, under given conditions, for a single sensor The probability of successfully detecting the target, and its range. .

[0142] After entering the target identification and close-range rescue phases in steps 6 and 7, the key steps are divided into the following two items: 1) Intelligent decision-making for rescue activation Confidence fusion based on DS evidence theory:

[0143] Identification Frame This represents the complete set of all possible propositions. In object detection scenarios, it is typically taken as... ,in: The proposition "The target exists" The proposition "The target does not exist" This defines the domain of evidence-based reasoning, and all basic probability assignments are based on subsets (focal elements) of this framework.

[0144] Basic probability assignment function A confidence assignment function is defined for each source of evidence (such as a sensor), satisfying:

[0145] in, This indicates that the evidence supports the proposition. The precise level of trust, but does not support... Any subset of. For example: The confidence level of the sensor in believing that the target exists; The confidence level of the sensor in believing that the target does not exist; The inability of sensors to make a judgment (i.e., uncertainty) indicates the allocation of trust in the entire framework; In rescue scenarios, each sensor outputs a set of data based on its own detection results and reliability. Values. For example, if a sensor considers a target to exist with a probability of 0.7, to not exist with a probability of 0.2, and to be uncertain with a probability of 0.1, then:

[0146] Jiao Yuan This represents a subset of the identification framework, i.e., propositions. In the fusion formula: and They came from two different sources of evidence. and Jiao Yuan.

[0147] This is a new proposition to consider after integration.

[0148] This indicates that two pieces of evidence simultaneously support the proposition. The situation; This indicates a situation where two pieces of evidence contradict each other.

[0149] The molecule part of the fusion formula This indicates that all pieces of evidence in two cases support the same proposition. Summing the trust products of the focal element combinations.

[0150] This means that both Evidence 1 and Evidence 2 support the proposition. The total amount of trust. For example, if support , Also support Then they jointly support ;like support , Support for uncertainty The intersection is still . Also support .

[0151] denominator This represents the normalization factor, which is equal to 1 minus the sum of trust for all contradictory combinations.

[0152] This represents the total confidence level when two pieces of evidence contradict each other, i.e., one piece of evidence supports... And another supporter Normalization aims to eliminate inconsistencies caused by contradictions, redistribute trust, and ensure that the total trust after fusion equals 1. The denominator reflects the degree of consistency between the evidence; the greater the contradiction, the smaller the denominator, and the more the trust distribution after fusion is "diluted."

[0153] Fusion results This indicates that by integrating the two pieces of evidence, the proposition is... Overall trust level; Value range: And satisfy .

[0154] Time consistency verification uses sliding window matching:

[0155] in: : The current moment (i.e., the real-time point in time when the decision is made).

[0156] : The duration of the sliding window (unit: seconds or minutes).

[0157] Indicates the time. Does it meet the condition of "continuous positive detection"?

[0158] : is from arrive Discrete or continuous time points.

[0159] Conditions for initiating rescue operations:

[0160] This represents the reliability threshold, which is the minimum acceptable reliability value set by the system. Rescue efforts are only considered when the fusion reliability exceeds this threshold. It is typically set to [value missing]. or higher (e.g.) The threshold is dynamically adjusted based on the mission risk level. It reflects the system's tolerance for false alarms. The higher the threshold, the more cautious the system is, and the lower the false alarm rate, but it may miss real targets; the lower the threshold, the more sensitive the system is, but it is prone to initiating rescue operations due to false alarms.

[0161] This represents the time consistency threshold, which is the minimum required level of time stability for the system. It is typically set to... The threshold is determined based on the dynamic characteristics of the environment and the performance of the sensor. This threshold ensures that the detection results are not isolated transient events. For example, if a target is only detected in a single frame, the temporal consistency will be very low, and even if the confidence level of a single frame is high, rescue will not be initiated.

[0162] This represents the spatial consistency threshold, which is the maximum allowable positioning deviation tolerance of the system. It is typically set to... The corresponding allowable positioning error range (e.g.) The consistency per meter is 0.9. This threshold ensures that different observation sources have a consistent understanding of the target location, eliminating the possibility of multi-target confusion or excessive sensor bias.

[0163] This represents the minimum number of independent detections required by the system. It is typically set to [value]. or The threshold is determined based on system redundancy and task importance. This threshold mandates at least two independent sources of evidence confirming the target's existence, preventing incorrect decisions due to single points of failure or false alarms from a single sensor.

[0164] 2) Scientific evaluation of multiple objectives and priorities Priority scoring based on multi-attribute utility theory:

[0165] Detailed definitions of each utility function: a. Life state utility:

[0166] b. Environmental threat effects:

[0167] c. Distance cost-utility:

[0168] d. The effect of time urgency:

[0169] in: The indicator variable represents whether a person is wearing a life jacket, 1 (yes) or 0 (no). People wearing life jackets have a higher probability of survival and can be rescued later; those not wearing life jackets face a greater risk of drowning and should be rescued first. Therefore, this indicator contributes positively to the utility function, but its weight is relatively low (because life jackets provide buoyancy, but consciousness is more important).

[0170] This variable indicates the state of consciousness, denoted as 1 (conscious, such as waving or looking up) or 0 (unconscious, such as floating motionless). Conscious individuals can cooperate with rescue efforts and grab onto lifebuoys, increasing the chances of a successful rescue; unconscious individuals require immediate assistance, otherwise their condition may deteriorate rapidly. Therefore, this indicator usually has the highest weight.

[0171] This variable indicates limb mobility, denoted as 1 (significant limb activity, such as paddling) or 0 (no significant activity). Individuals with high mobility are able to remain afloat and actively approach rescue equipment, making rescue easier; those with low mobility or those trapped require priority rescue. This indicator is related to consciousness level but is more refined.

[0172] This represents the weight coefficient of each sub-item, satisfying... .

[0173] In one embodiment, The most important factor is the state of consciousness. The weighting can be adjusted based on environmental factors such as water temperature and sea conditions (for example, increasing the weighting of activity level at low temperatures).

[0174] This indicates seawater temperature (unit: °C). Water temperature directly affects the rate at which the human body loses heat. In cold water, survival time is drastically shortened.

[0175] Indicates the significant wave height (unit: meters). The greater the wave height, the easier it is for the target to be submerged or scattered by the waves, and the greater the difficulty of rescue.

[0176] This indicates wind speed (unit: meters per second or knots). Strong winds can increase the feeling of cold, create waves, and potentially blow objects away.

[0177] : The target's current location (latitude and longitude coordinates).

[0178] The current location of the rescue platform (such as an unmanned surface vessel).

[0179] Euclidean distance between the two (unit: meters or kilometers).

[0180] Normalized maximum distance, usually the maximum distance among all targets to be rescued, or the system's preset effective rescue radius (e.g., 5 kilometers).

[0181] : The elapsed time from the occurrence of the accident to the present moment (unit: minutes or hours).

[0182] Critical survival time is the golden rescue window (e.g., 1 hour) determined in advance based on factors such as water temperature and sea conditions.

[0183] like Figure 1 and Figure 2 As shown, this system consists of three unmanned platforms: Unmanned Boat 2 (serving as the mother ship and command center), Unmanned Aerial Vehicle 3 (serving as the area search and reconnaissance unit), and Unmanned Lifebuoy 1 (serving as the final rescue execution unit). The system is uniformly scheduled through a multi-level collaborative controller, which integrates a target drift estimation system, a dynamic path planner, and an auction-based task assigner.

[0184] Unmanned Boat 2: Equipped with a high-precision GNSS / INS integrated navigation system, radar, weather sensors, and high-power communication relay equipment. Its deck features three UAV take-off and landing platforms and a dedicated release mechanism for one unmanned lifebuoy.

[0185] Drone 3: Equipped with a dual-light pod for visible light and thermal imaging, and an onboard AI processing unit, it has the capability for vertical take-off and landing and long-endurance cruise.

[0186] Unmanned lifebuoy 1: It adopts a U-shaped rigid float and vector propulsion design, and integrates simple vital sign detection sensors and voice communication equipment.

[0187] Example 1 This invention designs a search and rescue method that enables rapid deployment and efficient rescue operations, such as... Figure 3 As shown, the process is divided into four phases: response, raid, search and rescue.

[0188] I. Response Phase Upon receiving the rescue order from the emergency rescue center (the rescue order includes the accident type, accident coordinates, type and quantity of the target to be rescued), the rescue personnel quickly complete the rescue preparation process (exterior inspection of the unmanned boat 2, drone 3 and lifebuoy → system power-on self-test → target drift estimation system calculation → selection of rescue strategy and plan → system deployment).

[0189] 1) Accident Level Assessment Decision Model This model is used to address the next question: once a real incident is confirmed, at what level and scale should the response be?

[0190] a. Core Idea of ​​the Model This is a multi-attribute decision-making model that quantifies and weights multiple key factors affecting rescue response, and finally summarizes them into a comprehensive score, which determines the response level based on the score range.

[0191] b. Mathematical Models and Symbols The formula is:

[0192] Weighting coefficients: These weights represent the relative importance of each factor in the overall decision. The sum of all weights is 1. ).

[0193] In this invention, setting the distance is the most important ( ), because it directly determines whether you can arrive within the golden time frame; secondly, sea conditions ( This affects the feasibility and risk of rescue operations; then comes the number of personnel ( ) and accident type ( ).

[0194] Distance coefficient : A question about distance The function outputs a value between 0 and 1. The farther the distance, the closer the coefficient is to 1 (indicating higher urgency, as a faster response is required). This indicates that the distance increases linearly within 5 kilometers, and anything beyond 5 kilometers is considered the highest level.

[0195] Sea state coefficient : The value is calculated based on marine meteorological data such as wave height and wind force. The more severe the sea conditions, the higher the coefficient, because the rescue is more difficult and risky, requiring a more powerful rescue platform. The coefficient is 0.3 for winds below Beaufort scale 6, 0.6 for 6-8, and 1.0 for above 8.

[0196] Personnel coefficient : A function related to the number of people in distress. The more people, the higher the coefficient. This indicates that 20 or more people constitute the highest level.

[0197] Accident type coefficient : Different basic emergency values ​​are assigned to different accident types. Ship sinking = 1.0, fire / explosion = 0.9, collision / grounding = 0.7, mechanical failure = 0.4.

[0198] c. Decision-making rules Calculated total score It is a value between 0 and 1.

[0199] Level 1 Response (Long-Distance Solution). The strongest rescue force is activated, with the Unmanned Ark 2 equipped with its full complement of gear deployed.

[0200] Level 2 response (medium-range plan). Dispatch standard rescue team.

[0201] Level 3 Response (Close Range Solution). Activate rapid response mode, potentially deploying only 3 drones and 1 unmanned lifebuoy.

[0202] 2) System health assessment based on state-space model This model is the cornerstone of the entire system's reliable and autonomous operation. It elevates the traditional "pass / fail" checklist to a quantitative and predictive health management system. The model views the entire unmanned system as a dynamic system composed of multiple interconnected subsystems. Its goal is not simply to check whether each device "can still be powered on," but rather to comprehensively assess the overall system's probability of successfully completing the predetermined rescue mission under its current state.

[0203] a) Mathematical Model The formula is:

[0204] Indicates the overall health of the system: This is a value between 0 and 1, representing the overall "health score" of the system.

[0205] It represents a perfect state.

[0206] This indicates a complete failure.

[0207] Typically, a dispatch threshold is set, for example... This means that the system has a 95% confidence level in completing the task.

[0208] - Number of subsystems: The number of independent functional units contained in the system. In your scheme, this includes: Energy systems (batteries, generators) Communication systems (satellite, 4G / 5G, radio) Navigation system (GPS / INS) Mission payload (camera, radar) Propulsion system (motor, propeller) ...etc.

[0209] -No. Health of each subsystem: It is also a value between 0 and 1, calculated from readings of multiple sensors in this subsystem. It is not a simple binary judgment, but a continuous function based on performance indicators.

[0210] System overall health integration Integrate the health status of all the above subsystems into the overall assessment:

[0211] Weighting principles: Energy System (0.30): Highest weight, fundamental guarantee Communication system (0.25): Ensures command and control Navigation system (0.20): Affects task execution accuracy Mission payload (0.15): Affects detection and identification capabilities Propulsion system (0.10): Affects maneuverability This health assessment system, based on physical models and real-time sensor data, not only provides accurate quantification of system status but also enables early identification of potential faults, providing a solid guarantee for preventive maintenance and mission reliability.

[0212] - No. Weights of each subsystem: This represents the criticality of this subsystem in completing the rescue mission. The sum of all weights is 1 ( ).

[0213] "Multiplication" model ( Compared to the additive model (weighted average), the multiplicative model has a key characteristic: it is extremely sensitive to weak links (low-health subsystems). If the health of any subsystem approaches zero, the overall health will also approach zero. This aligns with engineering reality: the failure of a single critical unit can lead to the failure of the entire task. (Weights) As an index, it serves to regulate sensitivity. The larger the weight of a subsystem, the greater its health impact on the overall result.

[0214] II. The Assault Phase The system performs task and path planning and travels at maximum speed to the search mission starting point, whose coordinates are derived from search path 11. During the journey, the target drift estimation system updates the estimated position in real time based on the real-time changes in water flow speed and direction, as well as wind speed and direction. The rescue system then... Figure 3 The method described in the text is used to adjust the path.

[0215] 1) Intelligent initial path planning a) Multi-objective cost function

[0216] This formula is a functional integral, and its goal is to find a path. This makes it possible to start from the beginning time End time The cumulative cost is minimized. This function is an optimization process that finds the optimal trajectory.

[0217] - Time cost function: Practical calculations: This is usually related to the difference between the actual speed and the maximum safe speed. If speed is reduced due to ocean currents or waves, the time cost will increase.

[0218] - Energy cost function: Actual calculations show that the energy cost is related to propulsion power and the work done to overcome environmental drag. Sailing against currents or winds significantly increases the energy cost.

[0219] - Risk cost function: Obstacle risk: The closer the distance to reefs, other vessels, or restricted areas, the higher the risk.

[0220] Environmental risks: the severity of sea conditions such as wind, waves, and currents.

[0221] Navigation risks: such as areas where GPS signals are blocked.

[0222] - Weighting coefficients: Constraints: .

[0223] Tactical choices: Gold Rescue Mode: Settings (Time weight) is the highest, and energy consumption is spared to arrive at the fastest speed.

[0224] Severe Sea State Mode: Settings (Risk weight) Medium, safety first, better to take a detour.

[0225] Persistent Patrol Mode: Settings (Energy weight) is the lowest, prioritizing ensuring battery life.

[0226] Environmental factor modeling (parts not elaborated in the formula): Ocean current influence:

[0227] This is the principle of velocity synthesis. When planning, the "going with the current" or "going against the current" effect brought by ocean currents must be taken into account; otherwise, the actual path will deviate significantly.

[0228] Obstacle avoidance:

[0229] 2) Dynamic path replanning mechanism This is a decision logic model used to determine when the path needs to be recalculated.

[0230] Triggering conditions:

[0231] - Threshold for target location change; - Environmental velocity change threshold; - Maximum time interval threshold.

[0232] III. Search Phase According to the selected rescue strategy and plan, the search will be conducted along the planned search route 11, primarily using a wide-area "bow" shaped search route and a track-following search route 11. Once a suspected target is detected and identified, the target information will be reported. Remote control personnel will then decide on a closer approach for identification and confirmation, and precise location will be achieved. For details, please refer to the rescue strategy and plan.

[0233] 1) Precise definition of the search area Given the uncertainty of target drift prediction, the scope of the sea area to be searched should be scientifically defined to avoid blind searches.

[0234] a) Initial search radius calculation model

[0235] Take the maximum value between ocean current speed and wind speed as the target's maximum possible drift speed.

[0236] The time interval from the occurrence of the accident to the present is the main factor contributing to the accumulation of uncertainty.

[0237] Safety factor (range 1.5 to 2.0), used to compensate for model errors and minor influencing factors that were not considered.

[0238] b) Modeling of regions where probability exists The system uses a Gaussian mixture model (GMM) to describe the probability distribution of the target location:

[0239] : The weight of the i-th Gaussian distribution, representing the probability of that region.

[0240] : The mean of the i-th Gaussian distribution, representing the possible center location.

[0241] : Covariance matrix, describing the elliptical shape and direction of uncertainty.

[0242] The number of Gaussian distributions is determined based on the prediction duration and environmental complexity.

[0243] 2) Distributed intelligent task allocation By using a market auction mechanism, each platform can independently bid for search tasks, thereby achieving optimal overall efficiency.

[0244] a) Design of a multi-round auction mechanism Round 1: Coarse-grained region allocation Merge multiple adjacent grids into a larger search area; The platform assigns bids based on the overall capability match:

[0245] in For platform capability vectors, This is the regional demand vector.

[0246] Second round: Fine-grained mesh allocation Within the already allocated large area, individual grids are finely allocated; The platform calculates execution costs based on real-time status:

[0247] b) Cost function Distance cost : Normalized distance from the platform's current location to the task grid; Load cost The current task load rate of the platform should be monitored to prevent overload of any single platform. Mismatch cost: The degree of mismatch between platform capabilities and task requirements.

[0248] c) Dynamic reallocation strategy Confidence-based reallocation: When the confidence level of a region changes, search resources are readjusted; State-based redistribution: When a platform has insufficient power or a device malfunctions, its tasks are transferred to other platforms; Discovery-based redistribution: When a target clue is discovered, surrounding platforms automatically converge on that area.

[0249] 3) Intelligent search execution a) Drone search strategy Parallel route search mode: Applicable scenarios: Precise search in key areas, and areas with medium confidence; Path characteristics: A series of parallel routes along a fixed direction; Parameter optimization: The spacing between flight paths is determined based on the effective detection width of the sensor, and the flight path direction is preferentially perpendicular to the dominant drift direction; Advantages: Even coverage, with no missed areas.

[0250] Track search mode: Applicable scenario: Detailed search after discovering target clues; Mechanism features: Based on Bayesian updates, search parameters are dynamically adjusted, the search gap is automatically narrowed when clues are found, and the search path is optimized in real time based on environmental feedback; Mathematical basis: Partially observable Markov decision processes (POMDPs).

[0251] b) Unmanned surface vessel search strategy Zigzag search pattern: Applicable scenarios: Continuous search of large areas of water; Path characteristics: Zigzag path, making full use of the endurance advantage of the water surface platform; Parameter adjustment: Optimize the sawtooth angle according to the ocean current direction to reduce energy consumption.

[0252] Regional focus search mode: Applicable scenario: Confirmation search after drone 3 detects a suspected target; Collaboration mechanism: Receives precise location information provided by UAV 3; performs close-range confirmation using a high-resolution optical sensor; provides a precise navigation reference for the unmanned lifebuoy 1.

[0253] 4) Adaptive optimization of the search process a) Real-time evaluation of search performance Coverage monitoring

[0254] Discovery probability modeling

[0255] b) Dynamic adjustment of search parameters Search speed optimization: Adjust platform speed based on current discovery probability; Sensor mode switching: Selecting the optimal sensor combination based on environmental conditions; Communication strategy adjustment: Optimize data transmission frequency and content based on search progress.

[0256] IV. Rescue Phase The system intelligently determines the rescue target and deploys a lifebuoy. The system is then remotely controlled by personnel to move to the vicinity of the target and provide emotional reassurance and self-rescue guidance via a loudspeaker to complete the rescue. For details, please refer to the rescue strategy and plan.

[0257] 1) Intelligent decision-making for rescue activation a) Multi-source information confidence fusion The system uses the DS evidence theory to fuse detection information from different platforms to solve the uncertainty problem of sensors.

[0258] Basic probability assignment function: For each sensor k, its basic probability assignment for the presence of the target is as follows:

[0259] Evidence fusion formula: The fusion of evidence from two independent sensors is achieved through the following formula:

[0260] The denominator is a normalization factor, which ensures that the sum of probabilities after fusion is 1.

[0261] b) Time consistency verification The system performs time series analysis on the detection results within a continuous time window:

[0262] in It is a time decay weight, with more recent detection results having a higher weight.

[0263] c) Spatial consistency check Spatial cross-validation of observation results of the same target from multiple platforms:

[0264] in These are the positioning results of the same target from different platforms.

[0265] d) Complex decision-making conditions for initiating rescue operations The following conditions must be met simultaneously to initiate a rescue operation:

[0266] 2) Scientific evaluation of multiple objectives and priorities a) Multi-attribute utility theory model Establish a rescue value assessment system based on the analytic hierarchy process (AHP). For each target k, its rescue priority score is:

[0267] Where the weight coefficients satisfy .

[0268] b) Life state utility function

[0269]

[0270] in: : Are you wearing a life jacket? (Yes = 1, No = 0); Action recognition results based on visual analysis; Assessment of limb movement frequency and amplitude; Individual weight is higher than that of individuals in a group.

[0271] c) Environmental threat utility function

[0272]

[0273] in: Water temperature threat function; : Wave height threat function; : Wind speed threat function.

[0274] d) Time-pressure utility function

[0275]

[0276] The function grows exponentially after the critical time, emphasizing extreme urgency.

[0277] 3) Standardized rescue execution procedures a) Calculation of the optimal rescue location The optimal approach position for unmanned surface vessels takes environmental compensation into account:

[0278] in: Target drift velocity estimation; : The estimated time for the lifebuoy to reach the target; : The safe distance vector in the upwind direction (usually 50 meters).

[0279] b) Rescue process status identification Computer vision-based rescue status monitoring:

[0280] state space This includes: {Approaching, Target contact, Stable connection, Rescue completed, Assistance required}.

[0281] c) Optimization of multi-target rescue sequence The multi-objective rescue problem is modeled as a dynamic traveling salesman problem with time windows: Objective function:

[0282] Constraints: Each target is accessed only once: ; Flow conservation: ; Time window constraint: .

[0283] Sub-loop constraint elimination d) Dynamic replanning mechanism During the rescue operation, the system continuously monitors changes in the target's status and triggers a replanning process when the following conditions occur: A new target has been discovered with higher priority than some targets in the current sequence; The condition of the rescued target has deteriorated, necessitating an adjustment to the subsequent rescue sequence; Sudden environmental changes rendered the original plan unfeasible; The status of platform resources has changed.

[0284] Replanning employs a rolling time-domain optimization strategy, resolving the optimization problem at each decision point but only executing the next optimal action, thus maintaining the responsiveness and adaptability of the system.

[0285] V. Rescue Strategies and Plans for Near, Medium, and Long Distance Areas a) Remote rescue strategy When the rescue distance is greater than 1.5km (the maximum rescue radius of the drone) but less than 5km (5km is the maximum rescue radius of the UAV2; if the person to be rescued drifts far away with the current, at a common current speed of 2kn, the UAV2 can ensure that the person to be rescued is still within the rescue radius when it reaches this distance), the rescue will be carried out according to the remote rescue strategy. After the system is powered on, 2 to 3 people will place the person in the water. The UAV2 will inflate and deploy upon contact with water. The UAV2 will calculate the target drift based on the initial distress coordinates and autonomously plan the optimal path. The UAV2, carrying the drone3 and the unmanned lifebuoy1, will quickly rush towards the coordinates at a speed of 10kn. During the rush, the UAV2 will monitor the changes in the target calculation data in real time. If the target is outside the detection range of the UAV2, the path will be adjusted. At the same time, the UAV2 will continuously detect the external environment and the person to be rescued through lidar and perception cameras during the rush. If a person to be rescued is found during the process, the UAV2 will approach to confirm and rescue them. If no rescued persons are found during this process, UAV 2 will proceed until it is 50m away from the estimated area (a safe buffer distance and within the detection and identification range of UAV 2), then decelerate and navigate to the planned positioning point on the joint search path 11. UAV 3 will then be launched and conduct a large-scale search of its assigned area along the planned search path, while the rescue boat searches the remaining area along the planned path. If the number of rescue targets is determined, and UAV 3 and UAV 2 have searched for and confirmed the required number of targets during their respective searches, the current search mission can be manually terminated, and the rescue mission phase can begin. In this scenario, if all targets are discovered and confirmed in advance by either UAV 3 or UAV 2, the other will terminate its search mission, and both will enter the rescue phase. If the number of rescue targets is not clearly defined in this rescue operation, UAV 3 and UAV 2 must complete the search of the entire area before jointly entering the rescue phase. If either UAV 3 discovers and confirms a rescue target and determines it to be of high risk (e.g., drowning, shipwreck), a remote terminal alarm will be triggered, and UAV 2's search mission will be manually terminated, initiating the rescue phase. UAV 3 will continue its search mission. After UAV 3 completes its own search mission, it will record the target's location. Simultaneously, a new search mission will be planned to cover the areas not yet searched by UAV 2.Once the rescue phase begins, the system categorizes and prioritizes rescue targets based on risk level: (persons in the water (guide them to self-rescue and board the unmanned vessel 2) > sunken vessel (record location coordinates and take photos and videos of the accident scene; if there are people, guide them to self-rescue and transfer them to the unmanned vessel 2) > equipment salvage (record location coordinates and take photos and videos of the scene)), (persons in the water without life-saving equipment > person in the water wearing life-saving equipment), (unconscious person in the water > conscious person in the water), (sunken vessel with people > sunken vessel without people). Rescue operations are conducted in descending order of risk level (for targets of the same risk level, rescue operations are conducted from closest to furthest distance). The drone 3 flies ahead to the rescue target for precise positioning and guides the unmanned vessel 2 to approach quickly. Once 50 meters from the rescue target, the drone slows down and stops. If the target is a person in the water, a lifebuoy is thrown to approach. Simultaneously, the rescue warning lights and walkie-talkies are activated to alert and reassure the person in the water, guiding them on self-rescue. After rescuing all persons in the water, the system returns to base. If the drone 3 experiences a low battery warning during this stage, it can return to base on its own. If the target platform is identified as a lifeboat, a lifebuoy will be deployed directly according to the rescue level described above to complete the rescue operation.

[0286] b) Mid-range rescue strategy When the rescue distance is greater than 200m (the detection range of UAV2 for people in the water throughout the day) but less than 1.5km, a medium-range rescue strategy is implemented. UAV3 takes off from the remote control center and flies to the target drift coordinate area, continuously using its sensing cameras to detect the person to be rescued. Simultaneously, UAV2 enters the water and autonomously navigates to the vicinity of the target drift coordinate area, ready to receive the final coordinates from UAV3. Once UAV3 confirms the target and completes its positioning, UAV2 approaches the target based on the final coordinates and drops a lifebuoy to begin the rescue. UAV3 then autonomously returns to the mother ship and lands.

[0287] c) Short-range rescue strategy When the rescue distance is less than 200m and external visibility is good, the search can be conducted directly by human observation. The unmanned lifebuoy 1 can be remotely controlled to the vicinity of the target for confirmation, and then the rescue can be carried out at close range. If external visibility is poor (e.g., obstruction, heavy fog, nighttime), the unmanned boat 2 carrying the unmanned lifebuoy 1 will be launched into the water and autonomously navigate to the vicinity of the initial coordinates for search and verification. If necessary, the drone 3 can be activated to assist in the search. After confirming the target, the lifebuoy is dropped, and the rescue is carried out at close range.

[0288] VI. Search Path Planning and Strategies a) Wide-area joint search path Path shape: "bow" shape, such as Figure 4As shown. The strategy is that UAV 3 and UAV 2 complete their respective search areas in the same or similar time (considering the time loss of obstacle avoidance actions during the UAV 2 search process), and the two together can cover the entire rescue area. Figure 4 In the middle, 10 represents the total area S of the joint search area.

[0289]

[0290]

[0291]

[0292] V1—UAV Search Rate (Search Area / Unit Time) V2—Search rate of the unmanned vessel 2 (search area / unit time) t — search time S—Area of ​​the entire rescue area S1—UAV Search Area S2—Search area for unmanned boat 2 Search paths 11 (UAV path 8, UAV path 9) are determined based on the search areas defined by UAV 2 and UAV 3, respectively. The distance between search path 11 and the area boundary is half the search width of each UAV 2 and UAV 3's respective perception area (UAV perception area 6, UAV search width: C, UAV perception area 7, UAV search width: D). The distance between adjacent long sides of the "bow" shape is their respective search width. The starting point of each search path 11 (UAV 2, UAV 3) is the search length extended outward from the boundary point entering their respective search areas (UAV search length: A, UAV search length: B). The coordinates 101 of the rescue system's search mission location are calculated using L=v1T+v2T. L is the straight-line distance between the starting point 4 of the UAV search path and the starting point 5 of the UAV search path, v1 is the speed of UAV 2, v2 is the speed of UAV, and T is the travel time of UAV 2 and UAV 3 from the unmanned lifebuoy 1 to their respective starting points, UAV search path starting point 4 and UAV search path starting point 5.

[0293] In a remote rescue scenario, after the unmanned boat 2 is equipped with drone 3 and unmanned lifebuoy 1, the search mission is initiated. Unmanned boat 2 and drone 3 each head to the starting point 4 of the unmanned boat search path and the starting point 5 of the drone search path, respectively. After arriving at the starting point, they each carry out the search according to the unmanned boat path 8 and the drone path 9.

[0294] b) Joint search path along the track Path morphology: The parallel path that extends to both sides of the vessel's wake during the accident, such as... Figure 5As shown. The trajectory of the accident vessel needs to be re-determined based on the target drift trajectory prediction system 12. The parallel spacing between the unmanned boat path 8 and the drone path 9 is 1 / 2 of the search length A in the unmanned boat's perception area 6. The takeoff height H of the drone 3 needs to ensure that the search width of its perception area is 2A. The starting point 4 of the unmanned boat search path is level with the starting point of the drone path 9. The starting point 5 of the drone search path is the search length B of the drone's perception area 7 extending outward from the starting point of the drone path 9. The coordinates 101 of the rescue system's search mission location are calculated based on L1=v1T1+v2T1. L1 is the straight-line distance between the starting point 4 of the unmanned boat search path and the starting point 5 of the drone search path, v1 is the speed of the unmanned boat, v2 is the speed of the drone, and T1 is the travel time of the unmanned boat 2 and the drone 3 from the unmanned lifebuoy 1 to their respective starting points, the starting points of the unmanned boat search path 4 and the drone search path 5. The total joint search area 10 is the intersection point 13 where the unmanned boat and the drone complete their respective search tasks simultaneously, calculated using L2 = v1T2 + v2T2. L2 is the total distance of the unmanned boat's path 8, v1 is the speed of the unmanned boat, v2 is the speed of the drone, and T2 is the travel time from the starting point of the unmanned boat's search path 4 and the starting point of the drone's search path 5 to the total joint search area 10.

[0295] In the remote rescue scenario, after the unmanned boat 2 is loaded with drone 3 and unmanned lifebuoy 1 and is in place, the search mission is initiated. The unmanned boat 2 and drone 3 each go to the starting point 4 of the unmanned boat search path and the starting point 5 of the drone search path, respectively. After arriving at the starting point, they search according to the unmanned boat path 8 and the drone path 9, respectively, and end at the total area 10 of the joint search.

[0296] The present invention also provides a maritime rescue system based on a heterogeneous unmanned system. The maritime rescue system based on a heterogeneous unmanned system can be implemented by executing the process steps of the maritime rescue method based on a heterogeneous unmanned system. That is, those skilled in the art can understand the maritime rescue method based on a heterogeneous unmanned system as a preferred embodiment of the maritime rescue system based on a heterogeneous unmanned system.

[0297] A maritime rescue system based on heterogeneous unmanned systems includes: Module M1: Acquires accident information to generate a multi-dimensional decision model; The accident information includes the accident type, accident location, information on people needing rescue, and environmental information; Module M2: Based on the multi-dimensional decision-making model, automatically match the corresponding rescue plan; Module M3: Deploys a heterogeneous unmanned system according to the rescue plan, enabling it to conduct unmanned autonomous search along the rescue route and identify the searched targets; Module M4: Based on the recognition results, control the heterogeneous unmanned system to complete the rescue.

[0298] The heterogeneous unmanned system includes: The unmanned boat serves as a mother ship and command center, equipped with a pre-precision GNSS / INS integrated navigation system, radar, weather sensors, and high-power communication relay equipment; its deck is equipped with a drone take-off and landing platform and a dedicated release mechanism for unmanned lifebuoys; The drone is used as an area search and reconnaissance unit, equipped with a dual-light pod for visible light and thermal imaging, as well as an onboard AI processing unit. Unmanned lifebuoys, used as the final rescue execution unit, include vital sign detection sensors and voice communication equipment; The system is uniformly scheduled through a multi-level collaborative controller, which integrates a target drift estimation system, a dynamic path planner, and an auction-based task assigner.

[0299] Based on the type of accident, plan the corresponding collaborative rescue route.

[0300] The multi-dimensional decision-making model includes: S total = α f 距离 ( d )+ β g 海况 ( s )+ γ h 资源 ( r )+ δ k 紧急性 ( u ) The weighting coefficients satisfy the normalization condition: α+β+γ+δ=1; The rescue plan includes a long-distance rescue plan, a medium-distance rescue plan, and a short-distance rescue plan; The decision-making rules for the proposed solutions are as follows: when In such cases, choose a long-distance rescue plan; when In such cases, choose a medium-range rescue plan; when In such cases, choose a close-range rescue option; in, This indicates the total score based on the overall urgency level. Represents the weighting coefficient; f 距离 ( d ) represents the distance coefficient function. g 海况 ( s ) represents the sea state coefficient function. h 资源 ( r ) represents the resource availability coefficient function. k 紧急性 ( u ) represents the emergency urgency coefficient function.

[0301] The module M3 includes: Establish a system health assessment based on a state-space model:

[0302] in Indicates the health status of each subsystem. These are the weighting coefficients; The inspection items include: -In-depth energy system diagnostics: Battery pack state of charge assessment ; - Communication system link quality test: Signal-to-noise ratio ; -Navigation system accuracy calibration: Positioning error ; -Task payload functionality verification: Sensor response function testing; - Propulsion system evaluation: Comprehensive evaluation of motor, thruster, and energy conversion efficiency; Once all checks are passed, the system automatically enters the ready state and establishes a deployment ready matrix.

[0303] The module M3 also includes: Intelligent initial path planning: The system employs a multi-objective optimization path planning algorithm to construct the cost function:

[0304] in For the cost of time, For the cost of energy, As a risk cost, the weighting coefficients satisfy... ; Environmental factors modeling includes the effects of ocean currents and wind resistance; Obstacle avoidance uses the potential field method:

[0305] in, Indicates the first Gain coefficient of each obstacle Indicates the location The value of the repulsive potential field generated by all obstacles at that location; Indicates the current position With the Obstacle locations The Euclidean distance between them; Dynamic path replanning mechanism: The system establishes a periodic path optimization loop, and the replanning trigger conditions are based on multi-factor evaluation: .

[0306] in, Indicates the threshold for changes in the target location; Indicates the threshold for changes in environmental velocity; Indicates the maximum time interval threshold; This represents the Euclidean distance between the old and new target locations; and These represent the environmental velocity vectors at the current time and the previous planning time, respectively. This indicates the time interval since the last path planning was completed (in seconds or minutes).

[0307] The search process for heterogeneous unmanned systems includes: Precisely defined search area: Based on the target motion uncertainty model, the search radius is calculated as follows:

[0308] Among them, safety factor ; The region where the probability exists is modeled using a Gaussian mixture model:

[0309] in, This represents the weight of the i-th Gaussian distribution. Let represent the mean of the i-th Gaussian distribution. Represents the covariance matrix. This represents the number of Gaussian distributions. Indicates spatial location, and N represents a Gaussian distribution; Search priorities are allocated based on the principle of maximizing information gain. Distributed intelligent task allocation: The task allocation model employs a multi-round auction mechanism. The first round of coarse-grained region allocation is based on platform capability matching.

[0310] The second round of fine-grained grid allocation employs a cost-based bidding strategy:

[0311] in ; Represents the distance cost component. Indicates the load cost component; Adaptive intelligent search execution: The drone search mode employs path planning based on information entropy:

[0312] in, This represents all the grid cells that the drone sequentially passes through in its planned search path; Indicates the information gain of the grid; Indicates the time decay factor. The constant represents the decay rate, and t represents the time variable. The unmanned surface vessel's search strategy integrates data from multiple sensors, and the detection probability model is as follows:

[0313] Among them, P 传感器i Indicates the first The independent detection probability of each sensor.

[0314] The module M4 includes: Rescue efforts initiated intelligent decision-making: Confidence fusion based on DS evidence theory:

[0315] in, and These are from two different sources of evidence. and Jiao Yuan, This is a new proposition to consider after integration; Time consistency verification uses sliding window matching:

[0316] in, Indicates the current moment. Indicates the duration of the sliding window. Used to indicate at time Does the condition for detecting a continuous positive case meet? From arrive Discrete or continuous time points; Conditions for initiating rescue operations:

[0317] in, Indicates the reliability threshold. Indicates the time consistency threshold. Indicates the spatial consistency threshold. This represents the threshold for the minimum number of independent detections. Multi-objective prioritization scientific assessment: Priority scoring based on multi-attribute utility theory:

[0318] The detailed definitions of each utility function are as follows: Life state utility:

[0319] in, This represents the weight coefficient of each sub-item. An indicator variable representing whether or not a life jacket is being worn. Indicator variables representing states of consciousness Indicator variables representing limb mobility; Environmental threat effects:

[0320] in, Indicates seawater temperature, Indicates the significant wave height. Indicates wind speed; Distance cost-utility:

[0321] in, Indicates the target's current position. Indicates the current location of the rescue platform. This represents the Euclidean distance between the two. Indicates the normalized maximum distance; The effect of time urgency: .

[0322] in, This indicates the elapsed time from the occurrence of the accident to the present moment. Indicates the critical survival time.

[0323] Those skilled in the art will understand that, besides implementing the system and its various devices, modules, and units provided by this invention in the form of purely computer-readable program code, the same functions can be achieved entirely through logical programming of the method steps, making the system and its various devices, modules, and units of this invention function in the form of logic gates, switches, application-specific integrated circuits, programmable logic controllers, and embedded microcontrollers. Therefore, the system and its various devices, modules, and units provided by this invention can be considered as a hardware component, and the devices, modules, and units included therein for implementing various functions can also be considered as structures within the hardware component; alternatively, the devices, modules, and units for implementing various functions can be considered as both software modules implementing the method and structures within the hardware component.

[0324] Specific embodiments of the present invention have been described above. It should be understood that the present invention is not limited to the specific embodiments described above, and those skilled in the art can make various changes or modifications within the scope of the claims, which do not affect the essence of the present invention. Unless otherwise specified, the embodiments and features described in this application can be arbitrarily combined with each other.

Claims

1. A maritime rescue method based on heterogeneous unmanned systems, characterized in that, include: Step S1: Obtain accident information to generate a multi-dimensional decision model; The accident information includes the accident type, accident location, information on people needing rescue, and environmental information; Step S2: Based on the multi-dimensional decision-making model, automatically match the corresponding rescue plan; Step S3: Deploy a heterogeneous unmanned system according to the rescue plan, enabling it to conduct unmanned autonomous search along the rescue route and identify the searched targets; Step S4: Based on the identification results, control the heterogeneous unmanned system to complete the rescue.

2. The maritime rescue method based on heterogeneous unmanned systems according to claim 1, characterized in that, The heterogeneous unmanned system includes: The unmanned boat (2) is used as a mother ship and command center, equipped with a preset precision GNSS / INS integrated navigation system, radar, meteorological sensors and high-power communication relay equipment; its deck is equipped with a drone (3) take-off and landing platform and a dedicated release mechanism for unmanned lifebuoys (1); The drone (3) is used as an area search and reconnaissance unit, equipped with a dual-light pod for visible light and thermal imaging and an airborne AI processing unit; Unmanned lifebuoy (1), used as the final rescue execution unit, including vital sign detection sensors and voice communication equipment; The system is uniformly scheduled through a multi-level collaborative controller, which integrates a target drift estimation system, a dynamic path planner, and an auction-based task assigner.

3. The maritime rescue method based on heterogeneous unmanned systems according to claim 1, characterized in that, Based on the type of accident, plan the corresponding collaborative rescue route.

4. The maritime rescue method based on heterogeneous unmanned systems according to claim 1, characterized in that, The multi-dimensional decision-making model includes: S total = α f 距离 ( d )+ β g 海况 ( s )+ γ h 资源 ( r )+ δ k 紧急性 ( u ) The weighting coefficients satisfy the normalization condition: α+β+γ+δ=1; The rescue plan includes a long-distance rescue plan, a medium-distance rescue plan, and a short-distance rescue plan; The decision-making rules for the proposed solutions are as follows: when In such cases, choose a long-distance rescue plan; when In such cases, choose a medium-range rescue plan; when In such cases, choose a close-range rescue option; in, This indicates the total score based on the overall urgency level. Represents the weighting coefficient; f 距离 ( d ) represents the distance coefficient function. g 海况 ( s ) represents the sea state coefficient function. h 资源 ( r ) represents the resource availability coefficient function. k 紧急性 ( u ) represents the emergency urgency coefficient function.

5. The maritime rescue method based on heterogeneous unmanned systems according to claim 1, characterized in that, Step S3 includes: Establish a system health assessment based on a state-space model: in Indicates the health status of each subsystem. These are the weighting coefficients; The inspection items include: -In-depth energy system diagnostics: Battery pack state of charge assessment ; - Communication system link quality test: Signal-to-noise ratio ; -Navigation system accuracy calibration: Positioning error ; -Task payload functionality verification: Sensor response function testing; - Propulsion system evaluation: Comprehensive evaluation of motor, thruster, and energy conversion efficiency; Once all checks are passed, the system automatically enters the ready state and establishes a deployment ready matrix.

6. The maritime rescue method based on heterogeneous unmanned systems according to claim 1, characterized in that, Step S3 further includes: Intelligent initial path planning: The system employs a multi-objective optimization path planning algorithm to construct the cost function: in For the cost of time, For the cost of energy, As a risk cost, the weighting coefficients satisfy... ; Environmental factors modeling includes the effects of ocean currents and wind resistance; Obstacle avoidance uses the potential field method: in, Indicates the first Gain coefficient of each obstacle Indicates the location The value of the repulsive potential field generated by all obstacles at that location; Indicates the current position With the Obstacle locations The Euclidean distance between them; Dynamic path replanning mechanism: The system establishes a periodic path optimization loop, and the replanning trigger conditions are based on multi-factor evaluation: in, Indicates the threshold for changes in the target location; Indicates the threshold for changes in environmental velocity; Indicates the maximum time interval threshold; This represents the Euclidean distance between the old and new target locations; and These represent the environmental velocity vectors at the current time and the previous planning time, respectively. This indicates the time interval since the last path planning was completed.

7. The maritime rescue method based on heterogeneous unmanned systems according to claim 1, characterized in that, The search process for heterogeneous unmanned systems includes: Precisely defined search area: Based on the target motion uncertainty model, the search radius is calculated as follows: in, Indicates ocean current speed. Indicates wind speed and safety factor. ; The region where the probability exists is modeled using a Gaussian mixture model: in, This represents the weight of the i-th Gaussian distribution. Let represent the mean of the i-th Gaussian distribution. Represents the covariance matrix. This represents the number of Gaussian distributions. Indicates spatial location, and N represents a Gaussian distribution; Search priorities are allocated based on the principle of maximizing information gain. Distributed intelligent task allocation: The task allocation model employs a multi-round auction mechanism. The first round of coarse-grained region allocation is based on platform capability matching. in For platform capability vectors, This represents the regional demand vector. The second round of fine-grained grid allocation employs a cost-based bidding strategy: in ; Represents the distance cost component. Indicates the load cost component; Adaptive intelligent search execution: The drone (3) search mode adopts path planning based on information entropy: in, This represents all the grid cells that the drone sequentially passes through in its planned search path; Indicates the information gain of the grid; Indicates the time decay factor. The constant represents the decay rate, and t represents the time variable. The unmanned boat (2) search strategy integrates multi-sensor data, and the detection probability model is as follows: Among them, P 传感器i Indicates the first The independent detection probability of each sensor.

8. The maritime rescue method based on heterogeneous unmanned systems according to claim 1, characterized in that, Step S4 includes: Rescue efforts initiated intelligent decision-making: Confidence fusion based on DS evidence theory: in, and These are from two different sources of evidence. and Jiao Yuan, This is a new proposition to consider after integration; Time consistency verification uses sliding window matching: in, Indicates the current moment. Indicates the duration of the sliding window. Used to indicate at time Does the condition for detecting a continuous positive case meet? From arrive Discrete or continuous time points; Conditions for initiating rescue operations: in, Indicates the reliability threshold. Indicates the time consistency threshold. Indicates the spatial consistency threshold. This represents the threshold for the minimum number of independent detections. Multi-objective prioritization scientific assessment: Priority scoring based on multi-attribute utility theory: The detailed definitions of each utility function are as follows: Life state utility: in, This represents the weight coefficient of each sub-item. An indicator variable representing whether or not a life jacket is being worn. Indicator variables representing states of consciousness Indicator variables representing limb mobility; Distance cost-utility: in, Indicates the target's current position. Indicates the current location of the rescue platform. This represents the Euclidean distance between the two. Indicates the normalized maximum distance; Environmental threat effects: in, Indicates seawater temperature, Indicates the significant wave height. Indicates wind speed; The effect of time urgency: in, This indicates the elapsed time from the occurrence of the accident to the present moment. Indicates the critical survival time.

9. A maritime rescue system based on heterogeneous unmanned systems, characterized in that, include: Module M1: Acquires accident information to generate a multi-dimensional decision model; The accident information includes the accident type, accident location, information on people needing rescue, and environmental information; Module M2: Based on the multi-dimensional decision-making model, automatically match the corresponding rescue plan; Module M3: Deploys a heterogeneous unmanned system according to the rescue plan, enabling it to conduct unmanned autonomous search along the rescue route and identify the searched targets; Module M4: Based on the recognition results, control the heterogeneous unmanned system to complete the rescue.

10. The maritime rescue system based on heterogeneous unmanned systems according to claim 9, characterized in that, The heterogeneous unmanned system includes: The unmanned boat (2) is used as a mother ship and command center, equipped with a preset precision GNSS / INS integrated navigation system, radar, meteorological sensors and high-power communication relay equipment; its deck is equipped with a drone (3) take-off and landing platform and a dedicated release mechanism for unmanned lifebuoys (1); The drone (3) is used as an area search and reconnaissance unit, equipped with a dual-light pod for visible light and thermal imaging and an airborne AI processing unit; Unmanned lifebuoy (1), used as the final rescue execution unit, including vital sign detection sensors and voice communication equipment; The heterogeneous unmanned system is uniformly scheduled through a multi-level cooperative controller, which integrates a target drift estimation system, a dynamic path planner, and an auction-based task assigner.

Images