A sea-air cooperative search and rescue smooth scheduling method based on anti-disturbance rolling horizon

Abstract

The application discloses a sea-air cooperative search and rescue smooth scheduling method based on anti-disturbance rolling time domain, and comprises the following steps: according to a target drift space probability atlas, extracting a task point set and a real-time state of sea-air heterogeneous search and rescue resources, and constructing a dynamic topology correlation matrix representing a reachable relationship between the resources and the task points and a scheduling cost; solving a multi-objective function model to obtain a candidate scheme set in which expected search and rescue efficiency, scheduling cost and stability are not mutually dominated; respectively calculating a Hamming distance between a candidate scheme in the candidate scheme set and a last execution scheme, comparing the Hamming distance with a shock threshold to obtain a remaining candidate scheme, and sorting the remaining candidate scheme based on a final comprehensive weight, and selecting a candidate scheme with the closest degree in the candidate scheme set as an optimal smooth scheduling scheme of a current rolling window of an accident sea area. The method can dynamically respond to sea state changes while inhibiting dramatic changes in resource assignment caused by small disturbances.

Description

Technical Field

[0001] This invention belongs to the field of rescue technology and relates to a smooth scheduling method for sea-air collaborative search and rescue based on anti-disturbance rolling time domain. Background Technology

[0002] Maritime search and rescue is characterized by its dynamic, uncertain, and multi-objective complexity. After accidents such as ship collisions, fires, capsizing, and people falling overboard, distressed targets continue to move due to the influence of wind fields, current fields, wave directions, and their own drift characteristics. The accessibility and operational conditions of search and rescue resources such as rescue vessels, helicopters, and drones also change over time. Existing research on maritime search and rescue scheduling mainly focuses on issues such as search and rescue area determination, target drift prediction, resource allocation, path planning, and multi-objective optimization. However, many methods still primarily rely on situational data at a certain moment to form scheduling plans, lacking the ability to continuously absorb feedback from subsequent environmental changes, target probability distribution updates, and execution deviations, resulting in insufficient dynamic adaptability.

[0003] Rolling time-domain control, through a closed-loop "prediction-execution-feedback-correction" approach, optimizes scheduling within the current window for a future period, executing only instructions for shorter timeframes and correcting the plan in the next window based on the latest information. This concept aligns well with the characteristics of maritime search and rescue missions, such as continuous target drift, constant environmental disturbances, and the inability to lock in long-term plans all at once. However, existing maritime search and rescue methods still lack a systematic design for rolling closed-loop scheduling mechanisms for the coordinated execution of heterogeneous sea and air resources.

[0004] Furthermore, if rolling optimization is applied directly to maritime search and rescue without considering the continuity of instructions between adjacent rolling windows, the system may still frequently adjust the mission area, direction of action, or resupply arrangements when there are slight changes in wind direction, current speed, target drift probability, or resource status. This can cause the scheduling plan to jump around, affecting on-site collaborative execution. Therefore, rolling scheduling in maritime search and rescue not only needs to have dynamic update capabilities but also needs to consider the smoothness of cross-window plans.

[0005] Furthermore, joint sea-air search and rescue involves multiple types of resources, multiple mission areas, and multiple spatiotemporal constraints, resulting in a complex scheduling space. The scheme generation process is prone to insufficient searching or local convergence. The final scheme determination must also consider both algorithm calculation results and on-site risk assessment; otherwise, it may affect decision-making efficiency, objectivity, and feasibility. Therefore, existing technologies still require a joint sea-air search and rescue scheduling method that can integrate rolling time-domain dynamic updates, smooth scheme control, complex candidate search, and on-site executability coordination into a unified framework. Based on this, this invention proposes a smooth scheduling method for joint sea-air search and rescue based on an anti-disturbance rolling time domain. Summary of the Invention

[0006] To address the significant impact of environmental disturbances such as wind, currents, and waves on maritime search and rescue missions, the location of distressed targets, the status of search and rescue resources, and on-site execution conditions all continuously change over time. Existing maritime search and rescue scheduling methods are mostly based on information generated from the initial situation of the accident or information from a local stage, lacking a rolling closed-loop mechanism that can continuously correct decision results by incorporating real-time environmental changes, target drift, and execution feedback. This makes them unsuitable for the demands of long-duration, dynamic search and rescue missions. While rolling time-domain control is highly adaptable to such dynamic decision-making problems, there is still a lack of effective methods for applying its system to sea-air collaborative search and rescue scheduling, and further suppressing frequent task switching, scheme jumps, and on-site execution instability caused by small disturbances during the rolling update process. Furthermore, sea-air heterogeneous resource collaborative scheduling also faces problems such as high difficulty in candidate scheme search, insufficient integration of automatic optimization results and on-site risk assessment, and insufficient connection between rolling updates and execution feedback, insufficient cross-window instruction continuity control, and inadequate search for complex candidate schemes. The technical solution adopted in this invention is: a smooth scheduling method for sea-air collaborative search and rescue based on an anti-disturbance rolling time domain, comprising the following steps:

[0007] S1: Obtain real-time multi-source environmental information streams of the accident sea area. S2: Divide the accident sea area into several spatial grids, and form a target drift spatial probability map based on the real-time multi-source environmental information flow of the accident sea area; S3: Obtain the real-time status of heterogeneous sea and air search and rescue resources; S4: Extract the real-time status of the task point set and heterogeneous sea and air search and rescue resources based on the target drift space probability map, and construct a dynamic topological correlation matrix that represents the reachability relationship and scheduling cost between resources and task points; S5: Within the current scrolling window, based on the dynamic topological correlation matrix, construct a multi-objective function model that includes expected search and rescue effectiveness indicators, scheduling cost indicators, and Hamming distance, to generate candidate scheduling schemes for the accident sea area; S6: Based on the non-dominated sorting genetic algorithm, the multi-objective function model is solved to obtain a set of candidate solutions that are not mutually dominant among the expected search and rescue effectiveness, scheduling cost and solution stability represented by Hamming distance; S7: From the candidate solution set, select the candidate solutions with a preset proportion before ranking based on the non-dominated ranking level and the crowding distance to form an elite set, and perform embedded simulated annealing local search on the elite solutions in the elite set to obtain the expanded candidate solution set. S8: Calculate the Hamming distance between each candidate scheme in the current window and the reference task assignment sequence in the overlapping prediction time domain of the previous window, and compare the Hamming distance with the oscillation threshold. When the Hamming distance of all candidate schemes in the candidate scheme set is greater than the oscillation threshold, return to S6 and regenerate candidate schemes. When there is a candidate scheme whose Hamming distance is less than or equal to the oscillation threshold, take the candidate scheme whose Hamming distance is less than or equal to the oscillation threshold as the remaining candidate scheme. S9: Based on the entropy weight calculated from the decision matrix of the remaining candidate schemes and the benchmark weight generated from the sea state parameters, the machine weights of each evaluation index of the candidate schemes are obtained. Then, the weights of each index are corrected based on the machine weights to obtain the final comprehensive weights. S10: Based on the final comprehensive weight, select the candidate scheme with the highest proximity as the optimal smooth scheduling scheme for the current rolling window of the accident sea area; S11: Extract the scheduling instructions within the current execution time domain from the candidate solutions with the highest proximity and issue them for execution; after the execution time domain ends, collect the actual state of resources, new observation information of the target, and environmental change information, correct the initial state of the next rolling window, and enter the next round of rolling optimization.

[0008] Furthermore, the process of selecting the candidate scheme with the highest proximity based on the final comprehensive weight as the optimal smooth scheduling scheme for the current rolling window of the accident sea area is as follows: Based on the final comprehensive weight, the expected search and rescue effectiveness index, comprehensive scheduling cost index, and Hamming distance index of the remaining candidate schemes are uniformly standardized. Based on the standardized index, a weighted standardized decision matrix is ​​constructed. Among them, the expected search and rescue effectiveness index is a positive index, and the comprehensive scheduling cost index and Hamming distance index are negative indexes. The optimal values ​​of each indicator in the weighted standardized decision matrix constitute the positive ideal solution, and the worst values ​​of each indicator constitute the negative ideal solution. In the positive ideal solution, the expected search and rescue effectiveness indicator is set to the maximum value, and the comprehensive scheduling cost indicator and Hamming distance indicator are set to the minimum value. In the negative ideal solution, the expected search and rescue effectiveness indicator is set to the minimum value, and the comprehensive scheduling cost indicator and Hamming distance indicator are set to the maximum value. The distances between each candidate scheme and the positive and negative ideal solutions are calculated respectively. Based on the distances between each candidate scheme and the positive and negative ideal solutions, the proximity of each candidate scheme is calculated. The candidate scheme with the highest proximity is selected as the optimal smooth scheduling scheme for the current rolling window of the accident sea area.

[0009] Furthermore, the expression for the target drift spatial probability map is as follows:

[0010] in, Indicates the first The target location at the start time of the scrolling window is... The probability of each spatial grid is plotted, and this probability map is used for subsequent calculations of task point priority, expected search and rescue effectiveness, and resource allocation benefits. Initial location of the maritime accident. C: Wind field vector, C: Flow field vector High waves, The first [section / delineation] obtained from the sea area of ​​the accident A spatial grid, : Spatial grid number; The total number of spatial grids obtained from the division of the accident sea area.

[0011] Furthermore, the expression for the dynamic topological correlation matrix is ​​as follows:

[0012] in, Indicates the first The first scrolling window The resource and the first Dynamic topological association information between task points Indicates the search and rescue resource number. Indicates the task point number. Indicates the scrolling window number; Indicates the first Is the resource executable? One task point, Indicates the first The resource and the first The distance between each task point Indicates the first The resource reached the first Estimated arrival time of each task point Indicates the first The resource executes the first The flight distance and fuel cost corresponding to each mission point Indicates the first The resource executes the first The executability risk at each task point; if If the value is 0, then the resource-task combination will not be included in the candidate search space.

[0013] Furthermore, the expression for the multi-objective function model, which includes the expected search and rescue effectiveness index, the scheduling cost index, and the Hamming distance, is as follows:

[0014] in, This indicates the expected search and rescue effectiveness index of the candidate solutions. This represents the overall scheduling cost indicator. This represents the Hamming distance between the current candidate solution and the previous execution solution; Expected search and rescue effectiveness indicators of candidate solutions The probability of target appearance, resource arrival time, and target survival probability are determined, and the specific expression is as follows:

[0015] in, Indicates that the resource has reached the first The probability that the target is still alive at each task point. Indicates the candidate solution for the first Coverage or search intensity of each task point; The comprehensive scheduling cost index is a weighted average of range cost, time cost, fuel cost, resupply waiting cost, and mission switching cost, and the specific expression is as follows:

[0016] in, This indicates the total flight distance or total distance traveled by the candidate option. This represents the total time required for resources to arrive and for tasks to be executed. Indicates the cost of fuel or energy consumption. This indicates the time cost incurred during resupply, waiting, or mission coordination. This represents the execution cost incurred when resources switch tasks between adjacent time periods; , , , , These are the weighting coefficients for the corresponding cost items.

[0017] Further: From the candidate solution set, based on the non-dominated ranking level and combined with the crowding distance, a preset proportion of candidate solutions are selected to form an elite set, and embedded simulated annealing local search is performed on the elite solutions in the elite set to obtain the expanded candidate solution set as follows: Select a predetermined proportion of candidate solutions from the candidate solution set to form an elite set, and perform embedded simulated annealing local search on the elite solutions. For elite programs that concentrate elites Generate Gaussian perturbation And map the perturbation results back to the set of discrete tasks:

[0018] in, This represents a valid task mapping function, used to correct the perturbed value to the task number that can be executed by search and rescue resources. When the perturbation causes the task number to exceed the task range, it is truncated or mapped to the nearest valid task point. When the perturbation causes the resource execution capability to fail to meet the task requirements, the perturbation scheme is repaired or discarded. Calculate the energy difference between the perturbation solution and the original elite solution. :

[0019]

[0020] Among them, the variable marked with "ˆ" represents the index after normalization; like This indicates that the new solution is no worse than the original solution, so the new solution is accepted; if According to the Metropolis criterion, in terms of probability Accept this inferior solution:

[0021] in, Given the current annealing temperature; generate a random number between 0 and 1. ,like If the solution is good, accept the inferior solution; otherwise, reject the solution. Temperature is determined according to... Update, in which .

[0022] Furthermore: the formula for calculating the Hamming distance between the candidate solutions in the current window and the reference task assignment sequence in the overlapping prediction time domain of the previous window is as follows:

[0023] in, To the extent of search and rescue resources, The number of comparable time steps. This is an indicator function that indicates when the current candidate solution differs from the task assignment at the same resource and comparable time step in the previous execution solution. Select 1; otherwise select 0.

[0024] Furthermore, the expression for the machine weights is as follows:

[0025] in: Indicates the first Machine weights for each evaluation metric It is the decision matrix of candidate solutions. These are the baseline weights for generating sea state parameters; Indicates the fusion coefficient; Correction coefficients for each indicator input And obtain the final comprehensive weight:

[0026] in, The number of evaluation indicators; the correction coefficient is set to... ;when When, it means that the first one is not changed. Machine weights for each evaluation metric; when When, it indicates the importance of improving this indicator; when When the time is right, it indicates that the importance of the indicator has been reduced.

[0027] Furthermore: Based on the final comprehensive weight, the evaluation function for the comprehensive effect of multiple indicators is constructed as follows:

[0028] Among them: Among them, This represents the candidate solutions within the k-th scrolling window. The multi-index comprehensive effect evaluation function value, This represents the candidate scheduling scheme generated within the k-th rolling window; These represent the final comprehensive weights corresponding to the expected search and rescue effectiveness index, the comprehensive dispatch cost index, and the Hamming distance index, respectively. Indicate candidate solutions Normalized expected search and rescue effectiveness index value; Indicate candidate solutions Normalized comprehensive scheduling cost index value; Indicate candidate solutions Compared with the previous implementation plan The normalized Hamming distance index value between them; This represents the scheduling scheme that has been executed in the (k-1)th rolling window and the reference task assignment sequence formed in the overlapping prediction time domain.

[0029] The present invention provides a smooth scheduling method for sea-air collaborative search and rescue based on anti-disturbance rolling time domain. Unlike the macro decision-making method based on stage division, this method combines RHC dynamic update, multi-target candidate generation, local escape search, cross-window stability constraints and expert weight correction. This method enables the system to dynamically respond to changes in sea state while suppressing drastic changes in resource allocation caused by small disturbances. Unlike macro-level decision-making methods based on search and rescue phase divisions, this invention focuses on the micro-level continuity control of the scheduling instruction sequence during continuous rolling optimization. By introducing cross-window stability constraints, local optimal escape mechanisms, human-machine collaboration weight correction, and truncation execution feedback mechanisms, it achieves dynamic updates and smooth execution of the sea-air collaborative search and rescue scheduling scheme.

[0030] To address the issue of insufficient coordination between rolling updates and execution feedback, this invention employs a rolling time-domain closed-loop scheduling mechanism. Within the current rolling window, the system formulates a complete scheduling plan in the prediction time domain, but only truncates the instructions issued in the current execution time domain. After the execution time domain ends, feedback information such as actual resource locations, speed deviations, new target observations, and environmental changes is collected to correct the initial state of the next rolling window, forming a recursive closed loop of "prediction optimization—truncation execution—feedback correction—window rolling update".

[0031] To address the problem of insufficient control over cross-window instruction continuity, this invention introduces Hamming distance as a cross-window stability index and simultaneously applies it to the stability objective in a multi-objective model, the oscillation threshold constraint in the candidate scheme selection stage, and the negative index in the final TOPSIS ranking, thereby suppressing drastic changes in scheduling schemes between adjacent rolling windows.

[0032] To address the problem of insufficient search for complex candidate solutions, this invention uses NSGA-II to generate a Pareto candidate solution set and further performs embedded simulated annealing local search on elite solutions. Local escape is achieved through Gaussian perturbation and Metropolis acceptance criterion, thereby expanding the search range of candidate solutions and reducing the risk of premature convergence.

[0033] To address the issue of insufficient consideration of on-site executability, this invention constructs an HMC-TOPSIS subjective and objective fusion evaluation mechanism. It integrates the entropy weight calculated from the decision matrix of candidate schemes, the system benchmark weight generated from sea state parameters, and the expert correction coefficient into a final comprehensive weight. Then, the remaining candidate schemes are ranked using TOPSIS, thereby enabling the final scheme to take into account search and rescue effectiveness, scheduling cost, command continuity, and on-site execution feasibility.

[0034] This invention constructs a progressive closed-loop scheduling process centered on "continuous control of cross-window scheduling instructions." First, sea state, target drift probability, and the status of heterogeneous sea and air resources are acquired, and a multi-objective scheduling model is built within the current rolling window. Then, a Pareto candidate scheme set is generated using NSGA-II, and Gaussian perturbation and Metropolis local escape are applied to elite schemes using embedded simulated annealing (SA). Subsequently, the Hamming distance between each candidate scheme and the previous executed scheme is calculated, and non-smooth schemes are eliminated using an oscillation threshold λ. Finally, the entropy weight, the system baseline weight generated from sea state parameters, and the expert correction coefficient are integrated into the final comprehensive weight. TOPSIS is used to rank the remaining schemes, extracting the scheme with the highest comprehensive score, and only issuing scheduling instructions within the current execution time domain.

[0035] This invention works step by step around the same technical objective: Rolling Time Control (RHC) is used for dynamic updates, NSGA-II is used to generate multi-objective candidate solutions, SA is used to avoid local convergence after adding stability constraints, Hamming distance is used to quantify and eliminate cross-window instruction oscillations, and the Human-Machine Collaborative-Approximation Ideal Solution Ranking Method (HMC-TOPSIS) is used to integrate on-site expert experience and determine the final execution solution from the remaining smooth solutions.

[0036] The present invention proposes a smooth scheduling method for sea-air collaborative search and rescue based on disturbance-resistant rolling time domain, which has at least the following advantages compared with the prior art: (1) Transform the continuity of cross-window scheduling instructions into a computable and constrainable stability index.

[0037] This invention introduces Hamming distance The system quantifies the task assignment differences between the current candidate scheduling scheme and the previous execution scheme, and uses these differences for stability guidance in the candidate scheme generation stage, threshold elimination in the candidate scheme screening stage, and negative evaluation in the final ranking stage. As a result, the system can proactively suppress frequent resource reallocation caused by minor sea state disturbances such as wind, current, and waves during the rolling optimization process, reducing the risk of high-frequency oscillations in scheduling commands.

[0038] (2) Improve the on-site feasibility of the sea-air collaborative search and rescue dispatch plan.

[0039] Unlike dynamic scheduling methods that only pursue mathematical optimization within a single window, this invention uses an oscillation threshold. By eliminating non-smooth candidate solutions with excessively large task switching, the final dispatch instructions maintain good continuity between adjacent rolling windows. This mechanism can reduce the additional fuel consumption, communication burden, and execution waiting time caused by frequent course changes, rerouting, reassignment, or resupply plan adjustments for search and rescue resources such as rescue ships, helicopters, and drones.

[0040] (3) Enhance global search capability in high-dimensional discrete scheduling space.

[0041] This invention, based on the Pareto candidate solution set generated by NSGA-II, embeds a simulated annealing local search mechanism into elite solutions. This mechanism generates new neighborhood solutions through Gaussian perturbation and accepts poor short-term solutions with a certain probability according to the Metropolis criterion. This mechanism helps the algorithm cross local optima, expands the search range of candidate solutions, and obtains a more comprehensive candidate solution set within a limited rolling computation time, providing a more reliable basis for subsequent subjective and objective fusion ranking.

[0042] (4) Achieve computable fusion of expert experience and machine evaluation results.

[0043] This invention does not rely on expert judgment as a passive manual selection process after the solution is generated, but rather uses an expert correction coefficient. The system transforms the assessments of maritime emergency command personnel regarding wind force, wave height, visibility, communication conditions, and on-site execution risks into calculable weighted adjustment quantities. The system integrates entropy weight, system baseline weight, and expert correction coefficients into the final comprehensive weight. Then, TOPSIS is used to rank the candidate schemes that meet the Hamming distance threshold constraint, so that the final scheme can simultaneously take into account search and rescue effectiveness, scheduling cost and command continuity.

[0044] (5) Form a rolling closed-loop scheduling mechanism for strong disturbance sea conditions.

[0045] This invention in the prediction time domain Internally, forward optimization is performed, but only the current execution time domain is truncated. The system executes scheduling instructions within the time domain; after execution, it updates the initial state of the next rolling window based on the actual location of resources, speed deviations, environmental changes, and new target observation information. This mechanism avoids locking long-term scheduling plans at once and can continuously correct scheduling results based on sea state disturbances and execution deviations, thereby improving the dynamic adaptability and anti-disturbance capability of sea-air collaborative search and rescue scheduling. Attached Figure Description

[0046] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0047] Figure 1 It is a flowchart of the overall smooth scheduling of sea-air collaborative search and rescue based on the anti-disturbance rolling time domain; Figure 2 This is a schematic diagram of the perturbation-resistant rolling time domain (RHC) spatiotemporal sliding mechanism; Figure 3 It is an elite individual local search logic graph with embedded SA operator; Figure 4 This is a schematic diagram of scheme stability discrimination based on Hamming distance, where (a) is the scheme matrix at time t-1, (b) is the scheme matrix at time t, and (c) is the difference matrix; Figure 5 This is the architecture diagram of the HMC subjective and objective integrated evaluation model; Figure 6 This is a schematic diagram of the target drift space probability map and the task area extraction. Detailed Implementation

[0048] It should be noted that, unless otherwise specified, the embodiments and features in the embodiments of the present invention can be combined with each other. The present invention will be described in detail below with reference to the accompanying drawings and embodiments.

[0049] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The following description of at least one exemplary embodiment is merely illustrative and is in no way intended to limit the present invention or its application or use. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0050] A smooth scheduling method for sea-air collaborative search and rescue based on disturbance-resistant rolling time domain includes the following steps: S1: Obtain real-time multi-source environmental information streams of the accident sea area. S2: Divide the accident sea area into several spatial grids, and form a target drift spatial probability map based on the real-time multi-source environmental information flow of the accident sea area; S3: Obtain the real-time status of heterogeneous sea and air search and rescue resources; S4: Extract the real-time status of the task point set and heterogeneous sea and air search and rescue resources based on the target drift space probability map, and construct a dynamic topological correlation matrix that represents the reachability relationship and scheduling cost between resources and task points; S5: Within the current scrolling window, based on the dynamic topological correlation matrix, construct a multi-objective function model that includes expected search and rescue effectiveness indicators, scheduling cost indicators, and Hamming distance, to generate candidate scheduling schemes for the accident sea area; S6: Based on the non-dominated sorting genetic algorithm, the multi-objective function model is solved to obtain a set of candidate solutions that are not mutually dominant among the expected search and rescue effectiveness, scheduling cost and solution stability represented by Hamming distance; S7: Select the top-ranked elite solutions from the candidate solution set to form an elite set, and perform embedded simulated annealing local search on the elite solutions to obtain the expanded candidate solution set. S8: Calculate the Hamming distance between each candidate scheme in the current window and the reference task assignment sequence in the overlapping prediction time domain of the previous window, and compare the Hamming distance with the oscillation threshold. When the Hamming distance of all candidate schemes in the candidate scheme set is greater than the oscillation threshold, return to S6 and regenerate candidate schemes. When there is a candidate scheme whose Hamming distance is less than or equal to the oscillation threshold, take the candidate scheme whose Hamming distance is less than or equal to the oscillation threshold as the remaining candidate scheme. S9: Based on the entropy weight calculated from the decision matrix of the remaining candidate schemes and the benchmark weight generated from the sea state parameters, the machine weights of each evaluation index of the candidate schemes are obtained. Then, the weights of each index are corrected based on the machine weights to obtain the final comprehensive weights. S10: Based on the final comprehensive weight, select the candidate scheme with the highest proximity as the optimal smooth scheduling scheme for the current rolling window of the accident sea area; S11: Extract the scheduling instructions within the current execution time domain from the candidate solutions with the highest proximity and issue them for execution; after the execution time domain ends, collect the actual state of resources, new observation information of the target, and environmental change information, correct the initial state of the next rolling window, and enter the next round of rolling optimization.

[0051] In the NSGA-II stage of this invention, a multi-objective function set is used to generate candidate solutions; in the SA stage, an energy function is used to determine whether a new solution with perturbation is acceptable; and in the TOPSIS stage, the final comprehensive weight is used to rank the solutions that have passed the smoothness screening.

[0052] The process of selecting the candidate scheme with the highest proximity as the optimal smooth scheduling scheme for the current rolling window of the accident sea area based on the final comprehensive weight is as follows: Based on the final comprehensive weight, the expected search and rescue effectiveness index, comprehensive scheduling cost index, and Hamming distance index of the remaining candidate schemes are uniformly standardized, and a weighted standardized decision matrix is ​​constructed. The expected search and rescue effectiveness index is a positive index, while the comprehensive scheduling cost index and Hamming distance index are negative indices. The optimal values ​​of each index in the weighted standardized decision matrix constitute the positive ideal solution, and the worst values ​​constitute the negative ideal solution. In the positive ideal solution, the expected search and rescue effectiveness index is at its maximum value, and the comprehensive scheduling cost index and Hamming distance index are at their minimum values. In the negative ideal solution, the expected search and rescue effectiveness index is at its minimum value, and the comprehensive scheduling cost index and Hamming distance index are at their maximum values. The distances between each candidate scheme and the positive and negative ideal solutions are calculated respectively. Based on the distances between each candidate scheme and the positive and negative ideal solutions, the proximity degree of each candidate scheme is calculated. The candidate scheme with the highest proximity degree is selected as the optimal smooth scheduling scheme for the current rolling window of the accident sea area.

[0053] The role of standardization is to convert the expected search and rescue effectiveness index, comprehensive dispatch cost index, and Hamming distance index in this invention into a unified dimensionless scale for comparison.

[0054] The expected search and rescue effectiveness indicators are determined by the probability of target appearance, the time of resource arrival, and the probability of target survival; The comprehensive scheduling cost index is a weighted average of flight cost, time cost, fuel cost, resupply waiting cost, and mission switching cost; The Hamming distance metric is determined by the number of differences in task assignment between the current candidate scheme and the previous execution scheme at the same resources and the same comparable time steps.

[0055] The three types of indicators have different calculation sources, physical meanings, and numerical scales. Therefore, before using the final comprehensive weight for TOPSIS ranking, they need to be standardized so that the three types of indicators can be included in the same weighted decision matrix for comprehensive evaluation.

[0056] Specifically, let the standardized decision matrix be... The final overall weight is ,in, , , The weighted standardized decision matrix V corresponds to the expected search and rescue effectiveness index, the comprehensive dispatch cost index, and the Hamming distance index, respectively. according to The matrix is ​​constructed by standardizing the evaluation indicators to eliminate differences in the dimensions and numerical scales of different evaluation indicators, and then introducing a final comprehensive weight to reflect the relative importance of different evaluation indicators in the calculation of the proximity of the remaining candidate schemes and the final scheme selection process. This process does not change the positive or negative attributes of each evaluation indicator; among them, the expected search and rescue effectiveness indicator remains a positive indicator, while the comprehensive scheduling cost indicator and the Hamming distance indicator remain negative indicators.

[0057] Furthermore, the accident sea area is divided into several spatial grids, and the process of forming a target drift spatial probability map based on the real-time multi-source environmental information flow of the accident sea area is as follows: Real-time acquisition of multi-source environmental information streams in the accident area, including wind field vectors. Flow field vector High waves ,visibility Initial accident report location And the latest observed location information of the search and rescue targets.

[0058] The accident area was divided into several spatial grids. Based on the initial reported location, wind field, wave direction, target drift model, and historical observation information, the probability of the target being located in each grid is calculated to form a target drift spatial probability map.

[0059] The expression for the target drift spatial probability map is as follows:

[0060] in, Indicates the first The target location at the start time of the scrolling window is... The probability of each spatial grid is plotted, and this probability map is used for subsequent calculations of task point priority, expected search and rescue effectiveness, and resource allocation benefits. : Initial location of the marine accident, W: Wind field vector, C: Flow field vector, H: Wave height.

[0061] The target drift spatial probability map can be represented as a probability heat map on a grid of sea areas, where each grid corresponds to a target occurrence probability value. The darker the grayscale or the higher the probability value, the higher the likelihood of the target appearing in that area. Several search and rescue mission areas can be extracted from the high-probability grids. This is used for subsequent construction of dynamic topology association matrix and rolling scheduling model, that is, for calling the multi-objective scheduling model built within the current rolling window; The search and rescue mission area It can be a single high-probability grid or a connected region formed by merging multiple adjacent high-probability grids; the base or supply node is denoted as... Therefore, the target drift spatial probability map realizes the connection from environmental and target situational awareness to the generation of subsequent scheduling task areas.

[0062] Real-time data collection and quantitative characterization of the status of heterogeneous sea and air search and rescue resources, the resource set denoted as... Each resource The status includes the current position Maximum speed or flight speed Remaining battery life Load capacity Current task saturation Executable task types and communication status.

[0063] Among them, task saturation Indicates the first The search and rescue resources were in the first The task occupancy level at the start of each scrolling window can be calculated using the following formula:

[0064] in, Representing resources The number of tasks that have been undertaken or assigned in the current window. Representing resources The maximum number of tasks allowed in the current window. The larger the value, the higher the current task load of the resource. The system will reduce the priority of assigning new tasks to it in subsequent task allocation to avoid excessive task concentration.

[0065] Furthermore, based on the target drift spatial probability map, a task point set V={v_1,v_2, ... v_M}, where the task point can be a high-probability drifting grid, a suspected target area, a supply point, or a base node, the expression for constructing the dynamic topological correlation matrix of the current window is as follows:

[0066] in, Indicates the first The first scrolling window The resource and the first Dynamic topological association information between task points Indicates the search and rescue resource number. Indicates the task point number. Indicates the scrolling window number; Indicates the first Is the resource executable? One task point, Indicates the first The resource and the first The distance between each task point Indicates the first The resource reached the first Estimated arrival time of each task point Indicates the first The resource executes the first The flight distance and fuel cost corresponding to each mission point Indicates the first The resource executes the first The executability risk at each task point; if If the value is 0, then the resource-task combination will not be included in the candidate search space.

[0067] Furthermore, the prediction time domain is set. and execution time domain In each scrolling window Within, for the interval Construct a multi-objective scheduling model and encode the scheduling scheme into a resource-time step task assignment matrix. The formula used is as follows:

[0068] in, Indicates the first The first window The search and rescue resources were in the first Task assignment at discrete time steps. Representing resources The set of executable and legal tasks. This represents the number of discrete time steps in the prediction time domain. Each row of the matrix corresponds to a search and rescue resource, and each column corresponds to a discrete time step in the prediction time domain. The matrix elements represent the task assignment number of the corresponding resource at the corresponding time step.

[0069] This encoding method can be used to fully represent the task allocation sequence of each search and rescue resource in the prediction time domain in matrix form, and provide a unified data structure for subsequent NSGA-II candidate scheme generation, SA local search, and Hamming distance stability calculation.

[0070] The expression for the multi-objective function model, which includes the expected search and rescue effectiveness index, the scheduling cost index, and the Hamming distance, is as follows:

[0071] in, This indicates the expected search and rescue effectiveness index of the candidate solutions. This represents the overall scheduling cost indicator. This represents the Hamming distance between the current candidate solution and the previous execution solution; In the first rolling window, since no previous execution plan has been formed, the system can directly generate initial candidate plans based on the current environmental situation and resource status, and determine the first round of execution plan without performing cross-window Hamming distance comparison. Starting from the second rolling window, the execution plan issued in the previous rolling window and its reference task assignment sequence formed within the overlapping prediction time domain are used as the previous execution plan for calculating the Hamming distance with the current candidate plan. The "reference task assignment sequence of the previous execution plan within the overlapping prediction time domain" refers to the task assignment arrangement in the complete prediction scheduling plan generated in the previous rolling window that overlaps with the prediction range of the current rolling window. For example, if the prediction time domain is 30 minutes and the execution time domain is 20 minutes, and the previous window generates a scheduling plan for 0–30 minutes but only executes the plan for 0–20 minutes, and then enters the next window after 20 minutes, generating a scheduling plan for 20–50 minutes after the current window slides, then the overlapping prediction time domain is 20–30 minutes. The search and rescue resource task arrangements already predicted in the previous window within this 20–30 minute interval are the "reference task assignment sequence of the previous execution plan within the overlapping prediction time domain." This sequence is used to compare with the new task assignments within the same time interval of the current window to calculate the Hamming distance and measure the magnitude of cross-window plan changes.

[0072] Expected search and rescue effectiveness indicators of candidate solutions The probability of target appearance, resource arrival time, and target survival probability are determined, and the specific expression is as follows:

[0073] in, Indicates that the resource has reached the first The probability that the target is still alive at each task point. Indicates the candidate solution for the first Coverage or search intensity of each task point; The comprehensive scheduling cost index is a weighted average of range cost, time cost, fuel cost, resupply waiting cost, and mission switching cost, and the specific expression is as follows:

[0074] in, This indicates the total flight distance or total distance traveled by the candidate option. This represents the total time required for resources to arrive and for tasks to be executed. Indicates the cost of fuel or energy consumption. This indicates the time cost incurred during resupply, waiting, or mission coordination. This represents the execution cost incurred when resources switch tasks between adjacent time periods; , , , , These are the weighting coefficients for the corresponding cost items. Depending on the type of search and rescue resources, sea state level, mission urgency, and available data, one or more of the above cost items can be selected for calculation, or the weighting coefficients for each cost item can be adjusted.

[0075] in, Represents the actual execution cost of task switching, Hamming distance The two represent the cross-window instruction variation of the candidate scheme relative to the previous execution scheme, and constrain the scheduling scheme from the perspectives of execution cost and instruction stability, respectively.

[0076] Scheme stability refers to the stability of scheduling schemes across rolling windows, characterized by Hamming distance, and is not an independent evaluation metric. Based on the multi-objective function model, the NSGA-II phase simultaneously considers expected search and rescue effectiveness, overall scheduling cost, and Hamming distance. The Hamming distance quantifies the task assignment difference between the current candidate scheme and the previous execution scheme, thus reflecting scheme stability.

[0077] Furthermore, NSGA-II is used to perform parallel search on the multi-objective scheduling model within the current rolling window. NSGA-II does not directly output a unique solution, but instead generates a set of Pareto candidate solutions that are independent of expected search and rescue effectiveness, scheduling cost, and solution stability. .

[0078]

[0079] During this phase, Hamming distance As one of the multiple objectives, the candidate solutions are guided by stability during generation; however, at this point, not only are some options selected. Instead of choosing the minimum acceptable solution, we retain candidate solutions that represent a compromise between different objectives.

[0080] To avoid premature convergence of NSGA-II in high-dimensional discrete spaces, the process of selecting a predetermined proportion of candidate solutions from the candidate solution set based on the non-dominated ranking level and crowding distance to form an elite set, and then performing embedded simulated annealing local search on the elite solutions in the elite set to obtain the expanded candidate solution set is as follows: From the candidate solution set The top 5% of candidate solutions are selected to form an elite set. Furthermore, embedded SA local search is performed on the elite solutions; SA refers to the existing Simulated Annealing algorithm (which is an existing technology). Embedded SA local search can be understood as embedding the Simulated Annealing local search mechanism into the optimization process of elite candidate solutions.

[0081] For the elite group Elite program Generate Gaussian perturbation And map the perturbation results back to the set of legitimate discrete tasks:

[0082] in, Indicates the original elite plan After applying Gaussian perturbation, the new perturbation solution obtained is corrected by the legal task mapping function, which is the new legal candidate scheduling scheme generated during the simulated annealing local search process; This refers to the original elite solution; This indicates that the expression follows a pattern with a mean of 0 and a variance of . Gaussian perturbation; This represents the discrete perturbation quantity obtained after rounding the Gaussian perturbation. This indicates the initial perturbation result obtained after adding discrete perturbation to the original elite scheme; This represents a valid task mapping function, used to correct the perturbed value to the task number that can be executed by the search and rescue resource. When the perturbed task number does not belong to the set of valid discrete tasks corresponding to the search and rescue resource, it is truncated or mapped to the nearest valid task point. When the perturbation causes the resource execution capability to fail to meet the task requirements, the perturbation scheme is repaired or discarded. Calculate the energy difference between the perturbation solution and the original elite solution. :

[0083]

[0084] Among them, variables marked with "ˆ" represent indicators after normalization; energy function The smaller the value, the better the overall performance of the solution. The purpose of setting the energy function is to provide a unified criterion for judging whether a new solution is acceptable during the local search phase of SA.

[0085] like This indicates that the new solution is no worse than the original solution, so the new solution is accepted; if According to the Metropolis criterion, with probability Accept the inferior solution and combine it with the annealing temperature update rule to achieve local escape search. Expand the candidate solution set through the above process to enhance the ability of candidate solutions to escape local optima.

[0086]

[0087] in, Given the current annealing temperature; generate a random number between 0 and 1. ,like If the solution is good, accept the inferior solution; otherwise, reject the solution. Temperature is determined according to... Update, in which .

[0088] For the candidate solution set after SA expansion Calculate each candidate solution one by one. Compared with the previous implementation plan Hamming distance between them:

[0089] in, To the extent of search and rescue resources, The number of comparable time steps. This is an indicator function that indicates when the current candidate solution differs from the task assignment at the same resource and comparable time step in the previous execution solution. Select 1; otherwise select 0.

[0090] therefore, The larger the value, the more task switching occurs in the current plan compared to the previous execution plan, and the more obvious the scheduling oscillations.

[0091] In this embodiment, the comparable time step range is the overlapping prediction time domain of the current scrolling window and the previous scrolling window on the time axis, and this range corresponds to the previous execution scheme. Reference task assignment sequence and current candidate scheme The system compares the task assignment sequences at the same time location within the overlapping prediction time domain. For tasks with the same search and rescue resources and the same discrete time step, the system compares them item by item. If the task assignment at the corresponding location differs between the current candidate scheme and the previous execution scheme, it is recorded as a task switch and the Hamming distance is included. Therefore, the Hamming distance is used to characterize the temporal variation of the scheduling scheme between adjacent rolling windows.

[0092] Non-smooth candidate solutions are eliminated, and an oscillation threshold is set. And determine whether the candidate solution satisfies:

[0093] If the above conditions are met, it means that the task variation of the candidate solution relative to the previous execution solution is within the allowable range, and the solution will proceed to the subsequent HMC-TOPSIS ranking; if If the candidate solution is deemed to have a significant risk of oscillation, the system will directly eliminate that solution. If all candidate solutions are eliminated, the system returns to step S6 to regenerate candidate solutions. In other implementations, the oscillation threshold can be reset based on the urgency of the task, the amount of resources, and the sea state risk level.

[0094] It should be noted that the Hamming distance has three functions in this invention: First, it serves as a stability optimization objective in the NSGA-II stage, guiding the smooth generation of candidate solutions; second, it serves as a hard constraint in the S302 stage, eliminating non-smooth solutions; and third, it serves as a negative indicator in the TOPSIS ranking stage, participating in the final comprehensive evaluation.

[0095] Perform human-machine collaborative HMC subjective and objective weight fusion and TOPSIS ranking; In response to satisfy The candidate solutions are ranked using a human-machine collaborative (HMC) method that combines subjective and objective metrics. In HMC, experts do not directly reject machine results; instead, they influence the final ranking by adjusting the weights of the indicators. The final solution is still calculated by the system based on the objective metrics of the candidate solutions.

[0096] The criteria for inclusion in the TOPSIS ranking in this invention include: expected search and rescue effectiveness. (Positive indicators), overall scheduling costs (Negative indicator) and Hamming distance (Negative indicator). First, calculate the entropy weight based on the decision matrix of the candidate solutions. Simultaneously, system baseline weights are generated based on sea state parameters such as real-time wind speed, wave height, and visibility. .

[0097] The expression for the machine weight is as follows:

[0098] in: It is the decision matrix of candidate solutions. These are the baseline weights for generating sea state parameters; Correction coefficients for each indicator input And obtain the final comprehensive weight:

[0099] in, To evaluate the number of indicators; and to avoid excessive deviation from objective calculation results due to human intervention, in this implementation, the expert correction coefficient is set to... .when At that time, it was stated that the experts would not change the first... Machine weights for each evaluation metric; when When, it indicates the importance of improving this indicator; when When the value is reduced, it indicates a decrease in the importance of the indicator. The expert correction coefficient is only used to adjust the indicator weight; it does not directly replace the objective indicator value of the candidate solution, nor does it directly specify the final solution. The final solution is still determined by the TOPSIS closeness calculation result. In other embodiments, the range of the expert correction coefficient can be adjusted according to the mission level, sea state risk level, and system access control requirements.

[0100] Machine weight refers to the machine weight corresponding to the three types of evaluation indicators: expected search and rescue effectiveness, comprehensive scheduling cost, and Hamming distance.

[0101] Furthermore: Based on the final comprehensive weights, the evaluation function for the combined effect of multiple indicators is:

[0102] in, This represents the candidate solutions within the k-th scrolling window. The multi-index comprehensive effect evaluation function value, This represents the candidate scheduling scheme generated within the k-th rolling window; These represent the final comprehensive weights corresponding to the expected search and rescue effectiveness index, the comprehensive dispatch cost index, and the Hamming distance index, respectively. Indicate candidate solutions Normalized expected search and rescue effectiveness index value; Indicate candidate solutions Normalized comprehensive scheduling cost index value; Indicate candidate solutions Compared with the previous implementation plan The normalized Hamming distance index value between them; This represents the scheduling schemes already executed in the (k-1)th rolling window and their reference task assignment sequence formed within the overlapping prediction time domain. The expected search and rescue effectiveness index is a positive indicator, while the comprehensive scheduling cost index and Hamming distance index are negative indicators. Therefore, the evaluation function takes a positive sign for the expected search and rescue effectiveness index and a negative sign for the comprehensive scheduling cost index and Hamming distance index. The larger the value, the better the overall evaluation effect of the candidate solution.

[0103] In this invention, NSGA-II is responsible for candidate set generation. Mainly used for SA acceptance judgment, Used to explain the overall advantages and disadvantages of candidate solutions.

[0104] The final comprehensive weight is used in the subsequent TOPSIS ranking to weight the importance of different evaluation indicators. As a weight parameter in the multi-indicator comprehensive ranking, it affects the final closeness calculation result of each remaining candidate solution. Based on the final comprehensive weight The remaining candidate solutions are ranked using TOPSIS. Specifically, the system standardizes both positive and negative metrics to determine the ideal positive solution. and negative ideal solution Calculate the distance between each candidate solution and the two. , And obtain the degree of closeness:

[0105] The larger the value, the closer the candidate solution is to the ideal solution. The system selects the candidate solution with the highest degree of similarity. This is the optimal smooth scheduling scheme for the current rolling window.

[0106] Positive ideal solution: The positive index is maximized, and the negative index is minimized. Negative ideal solution: The positive index is minimized, and the negative index is maximized.

[0107] Perform truncation and feedback updates The system does not issue the entire prediction time domain at once. Not all instructions within, but from the final solution Extract the current execution time domain The system executes the scheduling instructions within the system.

[0108]

[0109] After the execution time domain ends, feedback information such as actual resource location, speed deviation, fuel consumption, communication delay, new target observations, and environmental changes is collected to form the execution deviation. And correct the initial state of the next window:

[0110] The window then scrolls forward. Then, S1 to S6 are re-executed to form a recursive dynamic closed loop.

[0111] Example 1 Figure 1 The overall operational logic of this invention is illustrated, corresponding to steps S101 to S401. The operational process of this invention includes five levels: information perception, rolling optimization, smoothing determination, subjective and objective fusion ranking, truncation execution, and feedback update. S101 to S103 are used to acquire environmental information of the accident sea area, the spatial probability map of target drift, and the status of heterogeneous sea and air search and rescue resources, and to construct a dynamic topological correlation matrix; S201 to S203 are used to construct a multi-objective scheduling model in the current prediction time domain, generate a Pareto candidate scheme set through NSGA-II, and perform Gaussian perturbation and Metropolis local escape through embedded SA local search; S301 to S302 are used to calculate the Hamming distance between the current candidate scheduling scheme and the previous execution scheme, and eliminate non-smoothing schemes according to the oscillation threshold; S303 is used to perform subjective and objective fusion ranking of the remaining schemes through the HMC-TOPSIS mechanism, and output the smoothing scheme with the highest comprehensive score. The following example, a merchant ship collision in a certain sea area, illustrates the implementation of this invention. The initial reported location of the accident is... =(122.5°E, 30.2°N). The system acquires the wind field. Flow field High waves and visibility And generate a target drift spatial probability map based on the target drift model.

[0112] Set up search and rescue resource collection ,in As a professional rescue vessel, For search and rescue helicopters, For drones or patrol boats, the system constructs a dynamic topological association matrix based on the location, speed, endurance, and mission capabilities of each resource.

[0113] Table 1. Logic Table for Assigning and Coding Heterogeneous Sea and Air Resources

[0114] The resource-time step codes in Table 1 above are used as inputs to the discrete task assignment sequence for Hamming distance calculation. Indicates the first Each mission search and rescue area, Indicates a base or supply node.

[0115] Table 2 Rolling Optimization Parameter Setting Table

[0116] The parameters described above are exemplary values ​​used in this embodiment and are not intended to limit the scope of protection of this invention. In different implementations, the prediction time domain, execution time domain, population size, annealing parameters, and oscillation threshold can be adjusted according to the accident level, the quantity of search and rescue resources, the sea state level, communication delay, and the urgency of the mission.

[0117] The following is an example of a numerical closed-loop method for candidate selection and ranking: Several candidate solutions are generated within the current scrolling window. After local expansion using NSGA-II and SA, solutions A, B, and C are obtained, and the following indices are calculated.

[0118] Table 3. Candidate Solution Screening and TOPSIS Ranking Results

[0119] In this example, scheme B has the highest expected search and rescue efficiency, but its Hamming distance of 3 exceeds the oscillation threshold of 2, indicating that this scheme involves too many task switching compared to the previous execution scheme, and therefore it is eliminated. Schemes A and C satisfy the smoothness constraint and are entered into the HMC-TOPSIS sorting. Scheme A has the highest overall score, so the system selects scheme A as the optimal smooth scheduling scheme for the current window.

[0120] The following is an example of HMC subjective and objective weight fusion and TOPSIS ranking: Let the evaluation index be , , The entropy weights are calculated based on the candidate solution decision matrix, and the system baseline weights are generated based on the current sea state. Assume the fused machine weights are:

[0121] If experts believe that the current wave height is high and visibility is low, and that frequent task switching will increase the execution risk, then enter the expert correction factor:

[0122] The unnormalized weights are then:

[0123] The final comprehensive weights are obtained after normalization:

[0124] This shows that the experts did not directly reject the machine's results, but rather increased the importance of the Hamming distance index, making the system more inclined to choose the option with less change in task and more continuous execution under harsh sea conditions.

[0125] Table 4. Subjective and objective evaluation index system based on stability regularization constraints

[0126] The execution and next window update are truncated. After selecting Option A, instead of issuing all instructions within the full 4-hour forecast time domain, only the scheduling instructions within the current 30-minute execution time domain of Option A are issued. For example: Continue to Area search, implement Rapid arrival mission in the region implement Area reconnaissance mission. The above instructions only apply to the current execution time domain. This does not mean locking the entire prediction time domain. The subsequent arrangements within the window. After 30 minutes, the system collects the actual location of the resource, trajectory deviation, wind flow field update, and new target observation information, corrects the initial state of the next window, and re-executes the above process.

[0127] Figure 2 This paper demonstrates the spatiotemporal sliding logic of the anti-disturbance rolling time-domain control in this invention, corresponding to steps S201 and S401. The horizontal axis represents the time axis t. The current rolling window k covers a prediction time domain T_p, but the system only truncates the scheduling instructions issued within the current execution time domain Δt. After the execution time domain ends, the system collects feedback information such as sea state changes, resource trajectory deviations, speed deviations, and new target observations, corrects the initial state of the next rolling window k+1, and shifts the window forward by one execution time domain Δt. This mechanism enables the system to perform look-ahead optimization within a longer prediction time domain while avoiding locking the long-term scheduling scheme at once, thus ensuring that the scheduling instructions can be continuously updated with sea state disturbances. The scheduling scheme; S401 is used to issue and execute the truncated instructions within the current execution time domain and feed back the execution deviation to the next rolling window, forming an anti-disturbance dynamic closed loop.

[0128] Figure 3This illustrates the mechanism by which the elite solution escapes local optima in this invention, corresponding to step S203. Background contour lines represent the energy topography of complex, non-convex scheduling schemes, and local valleys represent local optima where the algorithm might get stuck. The system selects elite solutions from the Pareto candidate solution set generated by NSGA-II, and generates new neighborhood solutions centered on these elite solutions through Gaussian perturbation. Then, it determines whether to accept the perturbated new solution based on the Metropolis criterion. If the new solution is better than the original solution, it is accepted directly; if the new solution is poor in the short term, it can also be accepted with a certain probability, thereby overcoming local energy barriers and expanding the search range of candidate solutions. This figure illustrates that the embedded SA local search is not an independent replacement of NSGA-II, but rather a local expansion and global escape enhancement of the Pareto elite solution.

[0129] Figure 4 This is a schematic diagram of scheme stability discrimination based on Hamming distance, where (a) is the scheme matrix at time t-1, (b) is the scheme matrix at time t, and (c) is the difference matrix; Figure 4 The method for calculating the Hamming distance stability index in this invention is illustrated, corresponding to steps S301 and S302. The left-hand matrix represents the reference task assignment state of the previous execution scheme X_(k-1)^exec within the comparable time step range, the middle matrix represents the task assignment state of the current candidate scheduling scheme X_k within the same comparable time step range, and the right-hand matrix is ​​the difference matrix between the two. The difference matrix can be obtained through a logical XOR operation, where a value of 1 indicates that the current candidate scheme has undergone a change in task assignment relative to the previous execution scheme, and a value of 0 indicates that the task assignment remains unchanged. The number of elements with a value of 1 in the difference matrix is ​​the Hamming distance D_H. This index is used to quantify the task switching amplitude between adjacent rolling windows and serves as a stability target in the candidate scheme generation stage, a hard constraint in the candidate scheme screening stage, and a negative evaluation index in the final TOPSIS ranking stage.

[0130] Figure 5This invention demonstrates the subjective and objective fusion evaluation architecture of Human-Machine Collaboration (HMC), corresponding to step S303. The model input includes three types of candidate scheme evaluation indicators: expected search and rescue effectiveness Z_sar, comprehensive scheduling cost Z_cost, and Hamming distance D_H. Z_sar is a positive indicator, while Z_cost and D_H are negative indicators. The system first calculates entropy weights based on the candidate scheme decision matrix and generates system baseline weights based on real-time sea state parameters such as wind speed, wave height, and visibility. Maritime emergency command experts input expert correction coefficients based on on-site risk assessments. The system integrates the entropy weights, system baseline weights, and expert correction coefficients into a final comprehensive weight and uses the TOPSIS method to rank the remaining candidate schemes that meet the Hamming distance threshold constraint. This mechanism does not allow experts to directly reject machine results; instead, it transforms expert experience into computable weight corrections. Candidate schemes first satisfy the Hamming distance threshold constraint D_H≤λ before entering the HMC-TOPSIS ranking, ensuring that the final output scheduling scheme simultaneously considers search and rescue effectiveness, scheduling cost, and command continuity.

[0131] Figure 6 The process of constructing the target drift spatial probability map and extracting the mission region is demonstrated. The system is based on the initial accident report location. Wind field vector Flow field vector High waves Based on historical observation information, the accident area was discretized into several spatial grids, and the probability of the target appearing in each grid was calculated. In the image, the darker the grayscale or the higher the probability value, the higher the likelihood that the target will appear in that area. The system extracts several search and rescue mission areas from the high-probability grid. , , ... and use them as task point inputs for subsequent dynamic topology association matrix and rolling scheduling model; the figure also shows base or supply nodes. .

[0132] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A smooth scheduling method for sea-air collaborative search and rescue based on disturbance-resistant rolling time domain, characterized in that: Includes the following steps: S1: Obtain real-time multi-source environmental information streams of the accident sea area. S2: Divide the accident sea area into several spatial grids, and form a target drift spatial probability map based on the real-time multi-source environmental information flow of the accident sea area; S3: Obtain the real-time status of heterogeneous sea and air search and rescue resources; S4: Extract the real-time status of the task point set and heterogeneous sea and air search and rescue resources based on the target drift space probability map, and construct a dynamic topological correlation matrix that represents the reachability relationship and scheduling cost between resources and task points; S5: Within the current scrolling window, based on the dynamic topological correlation matrix, construct a multi-objective function model that includes expected search and rescue effectiveness indicators, scheduling cost indicators, and Hamming distance, to generate candidate scheduling schemes for the accident sea area; S6: Based on the non-dominated sorting genetic algorithm, the multi-objective function model is solved to obtain a set of candidate solutions that are not mutually dominant among the expected search and rescue effectiveness, scheduling cost and solution stability represented by Hamming distance; S7: From the candidate solution set, select the candidate solutions with a preset proportion before ranking based on the non-dominated ranking level and the crowding distance to form an elite set, and perform embedded simulated annealing local search on the elite solutions in the elite set to obtain the expanded candidate solution set. S8: Calculate the Hamming distance between each candidate scheme in the current window and the reference task assignment sequence in the overlapping prediction time domain of the previous window, and compare the Hamming distance with the oscillation threshold. When the Hamming distance of all candidate schemes in the candidate scheme set is greater than the oscillation threshold, return to S6 and regenerate candidate schemes. When there is a candidate scheme whose Hamming distance is less than or equal to the oscillation threshold, take the candidate scheme whose Hamming distance is less than or equal to the oscillation threshold as the remaining candidate scheme. S9: Based on the entropy weight calculated from the decision matrix of the remaining candidate schemes and the benchmark weight generated from the sea state parameters, the machine weights of each evaluation index of the candidate schemes are obtained. Then, the weights of each index are corrected based on the machine weights to obtain the final comprehensive weights. S10: Based on the final comprehensive weight, select the candidate scheme with the highest proximity as the optimal smooth scheduling scheme for the current rolling window of the accident sea area; S11: Extract the scheduling instructions within the current execution time domain from the candidate solutions with the highest proximity and issue them for execution; after the execution time domain ends, collect the actual state of resources, new observation information of the target, and environmental change information, correct the initial state of the next rolling window, and enter the next round of rolling optimization.

2. The smooth scheduling method for sea-air collaborative search and rescue based on anti-disturbance rolling time domain as described in claim 1, characterized in that: The process of selecting the candidate scheme with the highest proximity as the optimal smooth scheduling scheme for the current rolling window of the accident sea area based on the final comprehensive weight is as follows: Based on the final comprehensive weight, the expected search and rescue effectiveness index, comprehensive scheduling cost index, and Hamming distance index of the remaining candidate schemes are uniformly standardized. Based on the standardized index, a weighted standardized decision matrix is ​​constructed. Among them, the expected search and rescue effectiveness index is a positive index, and the comprehensive scheduling cost index and Hamming distance index are negative indexes. The optimal values ​​of each indicator in the weighted standardized decision matrix constitute the positive ideal solution, and the worst values ​​of each indicator constitute the negative ideal solution. In the positive ideal solution, the expected search and rescue effectiveness indicator is set to the maximum value, and the comprehensive scheduling cost indicator and Hamming distance indicator are set to the minimum value. In the negative ideal solution, the expected search and rescue effectiveness indicator is set to the minimum value, and the comprehensive scheduling cost indicator and Hamming distance indicator are set to the maximum value. The distances between each candidate scheme and the positive and negative ideal solutions are calculated respectively. Based on the distances between each candidate scheme and the positive and negative ideal solutions, the proximity of each candidate scheme is calculated. The candidate scheme with the highest proximity is selected as the optimal smooth scheduling scheme for the current rolling window of the accident sea area.

3. The smooth scheduling method for sea-air collaborative search and rescue based on anti-disturbance rolling time domain as described in claim 1, characterized in that: The expression for the target drift spatial probability map is as follows: in, Indicates the first The target location at the start time of the scrolling window is... The probability of each spatial grid is plotted, and this probability map is used for subsequent calculations of task point priority, expected search and rescue effectiveness, and resource allocation benefits. Initial location of the maritime accident. C: Wind field vector, C: Flow field vector High waves, The first [section / delineation] obtained from the sea area of ​​the accident A spatial grid, : Spatial grid number; The total number of spatial grids obtained from the division of the accident sea area.

4. The smooth scheduling method for sea-air collaborative search and rescue based on anti-disturbance rolling time domain as described in claim 1, characterized in that: The expression for the dynamic topological association matrix is ​​as follows: in, Indicates the first The first scrolling window The resource and the first Dynamic topological association information between task points Indicates the search and rescue resource number. Indicates the task point number. Indicates the scrolling window number; Indicates the first Is the resource executable? One task point, Indicates the first The resource and the first The distance between each task point Indicates the first The resource reached the first Estimated arrival time of each task point Indicates the first The resource executes the first The flight distance and fuel cost corresponding to each mission point Indicates the first The resource executes the first The executability risk at each task point; if If the value is 0, then the resource-task combination will not be included in the candidate search space.

5. The smooth scheduling method for sea-air collaborative search and rescue based on anti-disturbance rolling time domain as described in claim 1, characterized in that: The expression for the multi-objective function model, which includes the expected search and rescue effectiveness index, the scheduling cost index, and the Hamming distance, is as follows: in, This indicates the expected search and rescue effectiveness index of the candidate solutions. This represents the overall scheduling cost indicator. This represents the Hamming distance between the current candidate solution and the previous execution solution; Expected search and rescue effectiveness indicators of candidate solutions The probability of target appearance, resource arrival time, and target survival probability are determined, and the specific expression is as follows: in, Indicates that the resource has reached the first The probability that the target is still alive at each task point. Indicates the candidate solution for the first Coverage or search intensity of each task point; The comprehensive scheduling cost index is a weighted average of range cost, time cost, fuel cost, resupply waiting cost, and mission switching cost, and the specific expression is as follows: in, This indicates the total flight distance or total distance traveled by the candidate option. This represents the total time required for resources to arrive and for tasks to be executed. Indicates the cost of fuel or energy consumption. This indicates the time cost incurred during resupply, waiting, or mission coordination. This represents the execution cost incurred when resources switch tasks between adjacent time periods; , , , , These are the weighting coefficients for the corresponding cost items.

6. The smooth scheduling method for sea-air collaborative search and rescue based on anti-disturbance rolling time domain as described in claim 1, characterized in that: From the candidate solution set, based on the non-dominated ranking level and combined with the crowding distance, a predetermined proportion of candidate solutions are selected to form an elite set. Then, an embedded simulated annealing local search is performed on the elite solutions in the elite set to obtain the expanded candidate solution set as follows: Select a predetermined proportion of candidate solutions from the candidate solution set to form an elite set, and perform embedded simulated annealing local search on the elite solutions. For elite programs that concentrate elites Generate Gaussian perturbation And map the perturbation results back to the set of discrete tasks: in, This represents a valid task mapping function, used to correct the perturbed value to the task number that can be executed by search and rescue resources. When the perturbation causes the task number to exceed the task range, it is truncated or mapped to the nearest valid task point. When the perturbation causes the resource execution capability to fail to meet the task requirements, the perturbation scheme is repaired or discarded. Calculate the energy difference between the perturbation solution and the original elite solution. : Among them, the variable marked with "ˆ" represents the index after normalization; like This indicates that the new solution is no worse than the original solution, so the new solution is accepted; if According to the Metropolis criterion, in terms of probability Accept this inferior solution: in, Given the current annealing temperature; generate a random number between 0 and 1. ,like If the solution is good, accept the inferior solution; otherwise, reject the solution. Temperature is determined according to... Update, in which .

7. The smooth scheduling method for sea-air collaborative search and rescue based on anti-disturbance rolling time domain as described in claim 1, characterized in that: The formula for calculating the Hamming distance between the candidate solutions in the current window and the reference task assignment sequence in the overlapping prediction time domain of the previous window is as follows: in, To determine the quantity of search and rescue resources, The number of comparable time steps. This is an indicator function that indicates when the current candidate solution differs from the task assignment at the same resource and comparable time step in the previous execution solution. Select 1; otherwise select 0.

8. The smooth scheduling method for sea-air collaborative search and rescue based on anti-disturbance rolling time domain as described in claim 1, characterized in that: The expression for the machine weight is as follows: in: Indicates the first Machine weights for each evaluation metric It is the decision matrix of candidate solutions. These are the baseline weights for generating sea state parameters; Indicates the fusion coefficient; Correction coefficients for each indicator input And obtain the final comprehensive weight: in, The number of evaluation indicators; the correction coefficient is set to... ;when When, it means that the first one is not changed. Machine weights for each evaluation metric; when When, it indicates the importance of improving this indicator; when When the time is right, it indicates that the importance of the indicator has been reduced.

9. The smooth scheduling method for sea-air collaborative search and rescue based on anti-disturbance rolling time domain as described in claim 1, characterized in that: Based on the final comprehensive weight, the evaluation function for the comprehensive effect of multiple indicators is constructed as follows: Among them: Among them, This represents the candidate solutions within the k-th scrolling window. The multi-index comprehensive effect evaluation function value, This represents the candidate scheduling scheme generated within the k-th rolling window; These represent the final comprehensive weights corresponding to the expected search and rescue effectiveness index, the comprehensive dispatch cost index, and the Hamming distance index, respectively. Indicate candidate solutions Normalized expected search and rescue effectiveness index value; Indicate candidate solutions Normalized comprehensive scheduling cost index value; Indicate candidate solutions Compared with the previous implementation plan The normalized Hamming distance index value between them; This represents the scheduling scheme that has been executed in the (k-1)th rolling window and the reference task assignment sequence formed in the overlapping prediction time domain.

Images