Post-earthquake multi-task cooperative optimization method and system

Abstract

This invention relates to the field of data processing technology for management or prediction, and particularly to a post-earthquake multi-task collaborative optimization method and system. The method includes: constructing optimization tasks under multi-layered information feedback and initial optimization functions for each task; constructing a unified search space and evaluating initial candidate solutions to determine the initial set of hazardous areas, the initial set of communication blind spots, and the initial set of inspection routes; constructing spatial geometric constraint inequalities and penalty terms for each task to obtain new optimization functions for the corresponding tasks; constructing learning mode determination rules to obtain updated candidate solutions for each task; calculating the current optimal solution for each task in the new population, predicting the optimal solution, and obtaining the predicted optimal solution for each task; synchronously updating the set of hazardous areas, the set of communication blind spots, and the set of inspection routes until the iteration is complete, and then outputting the optimal solution for each task. This invention effectively improves the accuracy of finding the optimal solution for each task.

Description

Technical Field

[0001] This invention relates to the field of data management or prediction technology, and in particular to a post-earthquake multi-task collaborative optimization method and system. Background Technology

[0002] Aftershock source location, emergency communication base station deployment, and drone patrols are interconnected modules in the rapid restoration of communication and disaster assessment after an earthquake. Intelligent optimization algorithms are used to solve these problems due to their simplicity and adaptability. However, current optimization approaches for these tasks have significant limitations: most studies employ single-task optimization strategies to solve the three tasks independently or sequentially, resulting in high computational redundancy and high computational resource consumption.

[0003] While there is considerable mature research on existing multi-task optimization algorithms, most are general optimization frameworks that can only complete shallow information interaction based on candidate solutions during the solution evolution process. They do not establish strong correlation mechanisms for the three types of tasks, and cannot coordinate the mutual constraints between the three types of tasks, nor can they dynamically modify the planning scheme based on the spatial range of each task's output. Ultimately, this leads to conflicts between different task output schemes, which are out of touch with the actual working conditions in the earthquake zone, making it difficult to form a well-implemented, integrated decision-making process.

[0004] At the same time, the complex on-site environment in the earthquake zone will bring multiple interferences to the aftershock source inversion. Traditional solution mechanisms are difficult to accurately restore the true spatial coordinates of the aftershock source and cannot completely and clearly delineate the high-risk geological restricted areas after the earthquake. When planning the deployment of emergency communication base stations, due to the objective constraints of the limited total amount of emergency communication base station equipment resources, the existing solutions are difficult to maximize the communication coverage of the disaster area and cannot meet the actual needs of uninterrupted communication transmission for the entire area of ​​emergency rescue operations. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to provide a post-earthquake multi-task collaborative optimization method and system, which enables multi-task synchronous parallel iteration and global joint optimization, effectively improving the accuracy of optimal solution for each task, and can output an integrated planning scheme that meets the on-site space access prohibition rules and has stronger landing reliability.

[0006] This invention is achieved through the following technical solution: A post-earthquake multi-task collaborative optimization method includes the following steps: S1: After the earthquake, carry out the task of locating the aftershock source, delineate the set of dangerous areas in the whole region, deploy emergency communication base stations in the area of ​​dangerous areas, deploy drone inspection tasks in the airspace above the dangerous areas and communication blind spots, and construct the initial optimization function for each task. S2: Construct a unified search space for aftershock source location tasks, emergency communication base station deployment tasks, and UAV inspection tasks. In the unified search space, construct an independent population for each task. Each population contains multiple candidate solutions. Evaluate the initial candidate solutions and determine the initial dangerous area set, the initial communication blind zone set, and the initial inspection route set. S3: Based on the initial dangerous area set, initial communication blind zone set, and initial inspection route set, all spatial constraints are uniformly and equivalently transformed into spatial geometric constraint inequalities for emergency communication base station deployment tasks and UAV inspection tasks. Penalty terms are constructed for the spatial geometric constraint inequalities of emergency communication base station deployment tasks and UAV inspection tasks. The penalty terms are substituted into the initial optimization function of the corresponding task to obtain the new optimization function of the corresponding task. S4: Construct learning mode determination rules for each candidate solution, and obtain updated candidate solutions for each task based on the learning mode determination rules and the corresponding new optimization functions; S5: Merge the updated candidate solutions of each task to form a new population, and calculate the current optimal solution of each task in the new population. When the number of iterations reaches the preset number, predict the subsequent optimal solution based on the historical optimal solution sequence within the sliding window of each task to obtain the predicted optimal solution of each task. S6: If the predicted optimal solution is better, then update the current optimal solution of each task to the predicted optimal solution, and simultaneously update the dangerous area set, communication blind spot set, and inspection route set based on the current optimal solution of each task, until the optimal solution of each task is output after the iteration is completed.

[0007] Furthermore, the initial optimization function for the aftershock source location task in step S1 is Equation (1), the initial optimization function for the emergency communication base station deployment task is Equation (2), and the initial optimization function for the UAV inspection task is Equation (3): (1); (2); (3); in: Indicates the first The sensor and the first The sensor received the first The theoretical time difference of each aftershock source signal. Indicates the first The sensor and the first The sensor received the first The actual time difference of each aftershock source signal This represents the Euclidean distance between two points. Indicates the first Spatial coordinates of the aftershock source Indicates the first Sensor coordinates, Indicates the first Sensor coordinates, This indicates the speed at which seismic waves propagate through the Earth's strata. Indicates the first The sensor received the first The timing of each aftershock source signal Indicates the first The sensor received the first The timing of each aftershock source signal This represents the initial optimization function for the aftershock source location task. This prevents positive numbers with a denominator of 0. Indicates the number of aftershock source signals. Indicates the number of sensors. This represents the initial optimization function for the emergency communication base station deployment task. Represents the set consisting of all base station locations. Indicates the first Coverage target point locations Indicates the probability of joint coverage by multiple base stations. This represents the number of target points covered by the discretization of the reconstructed area. Indicates constraints. This represents the decision variables for the deployment of emergency communication base stations. The x-coordinate range represents the area where communications were damaged in the disaster. Indicates the first The x-coordinate of each emergency base station Indicates the first The vertical coordinate of each emergency base station, This represents the vertical coordinate range of the area where communications were destroyed after the disaster. This indicates the danger zone predicted by the aftershock source location mission. Indicates the area where communications were destroyed after the disaster. Locations of each disaster area This indicates the number of disaster-stricken points within the area where communications were destroyed after the disaster. This represents the set of points within the disaster area whose communications were destroyed after the disaster. The first deployment Location of an emergency base station Indicates any, This indicates that it exists. This indicates the maximum distance between disaster-affected locations and emergency base stations within the area where communications are disrupted after a disaster. This indicates a point on the inspection route. This indicates the optimal inspection route. Indicates a communication blind spot. This represents the decision variables for drone inspection tasks. This indicates the total flight distance of the drone inspection. This represents the weighted sum of the urgency levels of the disaster in the affected areas. This is the initial optimization function for the UAV inspection mission. This indicates the total number of disaster areas to be inspected. Indicates the first Each disaster-stricken area was inspected. Indicates the first The urgency of the disaster situation at each inspected disaster area point The weighting coefficient represents the total flight distance. The weighting coefficients represent the total weighted sum of the urgency levels of disaster areas. Indicates the drone inspection route. This represents the set of communication blind spots under the optimal deployment scheme. Indicates the number of emergency base stations. This represents the horizontal plane projection operator.

[0008] In the optimized version, when deploying the drone inspection task in step S1, a random key encoding method is used, and the dimension of the decision variable is equal to the total number of disaster areas to be surveyed. The dimension component is the corresponding first dimension component. The random coded keys of each disaster area point are mapped one-to-one with the disaster areas to be inspected. The order of drone inspection visits to disaster areas is generated by encoding and decoding as follows: First, all random key values ​​in the decision variables are sorted in descending order. Then, the original dimension index corresponding to each key value after sorting is extracted. The final output index sequence is used as the order of drone inspections of disaster areas.

[0009] Furthermore, in step S2, the initial candidate solutions are evaluated using the following method: S211: Decode the data in the unified search space back into the original space according to equation (4): (4); in: Indicates the decoding function. For the task The The data of candidate solutions in the unified search space Indicates task No. The candidate solution of the nth Dimensional data, express Data after decoding back to the original space. Indicates task No. The lower bound of the original variable's value. Indicates task No. The upper bound of the values ​​of the original variable. Indicates task The total dimension of its own decision variables. This indicates sequentially applying the first dimension to the second dimension. Each dimension of the data is calculated according to the formula in parentheses, and then the calculation results of each dimension are concatenated in dimensional order to generate a complete original vector. S212: Substitute each decoded candidate solution into the corresponding initial optimization function to calculate the value of the corresponding initial optimization function. Based on the value of the corresponding initial optimization function, evaluate and select the decoded candidate solution corresponding to the maximum value as the optimal solution for each task. Based on the optimal solution for each task, determine the initial dangerous area set, the initial communication blind zone set, and the initial inspection route set.

[0010] Furthermore, the spatial geometric constraint inequalities transformed in step S3, the emergency communication base station deployment task, are as follows: S311: Regarding the requirement that emergency base stations should be deployed in areas where communication is damaged, this is transformed into an inequality judgment rule (5): (5); in: This indicates the first emergency communication base station deployment task under the condition of base station deployment in areas where communication is destroyed. Inequalities This represents the decision variables for the deployment of emergency communication base stations. Indicates the first The x-coordinate of each emergency base station Indicates the first The vertical coordinate of each emergency base station, The x-coordinate range represents the area where communications were damaged in the disaster. This represents the vertical coordinate range of the area where communications were destroyed after the disaster. Indicates the number of emergency base stations; S312: Emergency communication base station deployment work ensures the restoration of communication services to disaster-stricken areas where communication has been interrupted. This requires that the surrounding areas of each disaster-stricken communication point be protected. At least one base station shall be deployed within a range of meters. Based on this, the inequality is constructed as equation (6): (6); in: This indicates that the emergency communication base station deployment task is aimed at the first Inequality regarding base station coverage limitations at a single point of communication disruption. This represents the set of points within the disaster area whose communications were destroyed after the disaster. This indicates the maximum distance between disaster-affected locations and emergency base stations within the area where communications are disrupted after a disaster. Indicates the area where communications were destroyed after the disaster. Coordinates of the disaster area locations This indicates the number of disaster-stricken points within the area where communications were disrupted after the disaster. S313: For the deployment of emergency communication base stations and the set of prohibited dangerous areas, the criterion is that the distance from the deployment location of each emergency base station to the center of the dangerous area is greater than or equal to the preset radius of the dangerous area. The inequality is constructed as Equation (7): (7); in: This represents the Lth inequality for the deployment of emergency communication base stations under the restricted conditions of no-entry dangerous areas. Indicates the preset number Danger radius of each aftershock source Indicates the predicted first The epicenter of each aftershock Coordinates along the axial direction, Indicates the predicted first The epicenter of each aftershock Coordinates along the axial direction, This indicates the number of aftershock source signals. Indicates the number of emergency base stations. This indicates the danger zone predicted by the aftershock source location mission. S314: For the inspection route set, the entire flight area is projected onto a two-dimensional horizontal plane to carry out distance geometric calculations. Based on the idea that the shortest distance from a point to a finite line segment is greater than or equal to the blind zone radius, an inequality is established as equation (8): (8); in: Indicates the first From the communication blind spot to the optimal inspection route The disaster-stricken areas that were inspected and The projection coefficient of the line segment connecting the inspected disaster area points, where: This indicates that the deployment of emergency communication base stations is carried out under the condition that the drone inspection route does not pass through communication blind spots. Inequalities This indicates the first in the optimal inspection route. Two-dimensional plane coordinates of the disaster area points being inspected. Indicates the first Coordinates of a communication blind spot Indicates the radius of the blind zone. This indicates the total number of disaster areas to be inspected. Indicates the number of communication blind spots. This indicates the optimal inspection route for a drone inspection mission.

[0011] Furthermore, the spatial geometric constraint inequalities transformed in step S3 of the UAV inspection task are as follows: S321: Combining the set of hazardous areas output by aftershock location, the entire flight area is projected onto a two-dimensional horizontal plane, and a geometric judgment criterion is set: the shortest distance from the center of each hazardous area to the inspection route under the two-dimensional projection is not less than the preset hazardous radius. The resulting inequality is equation (9): (9); in: Indicates the first From the epicenter of each aftershock to the inspection route The disaster-stricken areas that were inspected and The projection coefficient of the line segment connecting the inspected disaster area points. This indicates the first time that a drone inspection mission is conducted under restricted airspace conditions in a prohibited danger zone. Inequalities, Indicates the first The disaster-stricken areas that were inspected Axial coordinates, Indicates the first The disaster-stricken areas that were inspected Coordinates along the axial direction, Indicates the predicted first The epicenter of each aftershock Coordinates along the axial direction, Indicates the predicted first The epicenter of each aftershock Coordinates along the axial direction, This indicates the number of aftershock source signals. This indicates the total number of disaster areas to be inspected. This indicates the danger zone predicted by the aftershock source location mission. Indicates the preset number Danger radius of each aftershock source; S322: For the communication blind zone given in the base station deployment work, the entire flight area is projected onto a two-dimensional horizontal plane. Based on the idea that the shortest distance from the center of the communication blind zone to each line segment is greater than or equal to the radius of the blind zone, it is transformed into the following inequality as equation (10): (10); in: This represents the optimal base station deployment scheme. From the first communication blind spot to the inspection route The disaster-stricken areas that were inspected and The projection coefficient of the line segment connecting the inspected disaster area points. This indicates the first time that a drone inspection mission is conducted under restricted airspace conditions in a no-entry communication blind zone. Inequalities, The preset communication blind zone radius, The set of communication blind spots under the optimal base station deployment scheme. blind spots Axial coordinates, The set of communication blind spots under the optimal base station deployment scheme. blind spots Coordinates along the axial direction, This represents the total number of communication blind spots in the optimal base station deployment scheme. This represents the set of communication blind spots under the optimal deployment scheme. This represents the decision variables for drone inspection tasks.

[0012] Furthermore, in step S3, a penalty term is constructed for the spatial geometric constraint inequalities of the emergency communication base station deployment task and the UAV inspection task. The penalty term is then substituted into the initial optimization function of the corresponding task to obtain the new optimization function of the corresponding task, as follows: S331: Application of spatial geometric constraint inequalities The penalty technique is used to optimize the initial function of the emergency communication base station deployment task. Combining the spatial geometric constraint inequality, we transform it according to equation (11) to obtain the emergency base station deployment optimization function with a penalty term. The initial optimization function for the drone inspection task Combining the spatial geometric constraint inequality, we transform it according to equation (12) to obtain the UAV inspection task optimization function with the introduction of a penalty term. : (11); (12); in: This represents the decision variables for the deployment of emergency communication base stations. This represents the penalty parameter that increases with the number of iterations. This indicates the danger zone predicted by the aftershock source location mission. This indicates the optimal inspection route for a drone inspection mission. This indicates that base stations should be deployed in areas where communication is disrupted, under the emergency communication base station deployment task. Inequalities This indicates that the emergency communication base station deployment task is aimed at the first Inequality regarding base station coverage limitations at a single point of communication disruption. This represents the Lth inequality for the deployment of emergency communication base stations under the restricted conditions of no-entry dangerous areas. This indicates that the deployment of emergency communication base stations is carried out under the condition that the drone inspection route does not pass through communication blind spots. Inequalities Indicates the number of communication blind spots. This indicates the first time that a drone inspection mission is conducted under restricted airspace conditions in a prohibited danger zone. Inequalities, Indicates the number of emergency base stations. This indicates the number of aftershock source signals. This indicates the total number of disaster areas to be inspected. This represents the total number of communication blind spots in the optimal base station deployment scheme. This represents the decision variables for drone inspection tasks. This represents the set of communication blind spots under the optimal deployment scheme. This refers to the first drone inspection mission conducted under restricted airspace conditions in a no-entry communication blind zone. Inequalities, This indicates the number of disaster-stricken points within the area where communications were disrupted after the disaster. S332: By using the piecewise smooth function (13) and By combining penalty techniques, a new emergency base station deployment optimization function with the introduction of a smooth penalty term is obtained according to equation (14). According to equation (15), the new UAV inspection optimization function after introducing a smoothing penalty term is obtained. ; (13); (14); (15); in: Represents a smooth function. This represents the value that the function takes on the left side of any inequality. Represents the smoothness parameter. Represents the natural constant.

[0013] Furthermore, the method for obtaining the updated candidate solutions for each task in step S4 is as follows: S411: For each candidate solution in each task, construct the learning mode determination rule according to equation (16): (16); in: Indicates task No. The second iteration The learning mode selected from the candidate solutions Indicates task The optimal solution continuously stagnates during the update algebra. This indicates the preset critical threshold for the optimal solution to stagnate and update. Represents logical OR, This indicates logical AND. Indicates the current iteration number. Indicates task No. The second iteration Spatial characteristic coefficients of candidate solutions This represents the critical coefficient for partitioning the neighborhood of the candidate solution; S412: When a candidate solution is determined to enter the multi-task collaborative learning mode, firstly, the adaptability is evaluated according to Equation (17) to obtain the adaptability of the source task to the target task, and the adaptability is normalized. Then, the distance proximity between the target task and the source task is calculated according to Equation (18). Then, the normalized adaptability and the distance proximity are weighted and fused to obtain a comprehensive evaluation value. Then, the probability of collaborative learning between the target task and the source task is calculated according to the comprehensive evaluation value. Based on the probability of collaborative learning between the target task and the source task, the selected source task is determined. A reference solution is selected from the selected source tasks using a hierarchical learning method. Based on the selected reference solution, multi-task collaborative learning is performed on the candidate solution according to Equation (19) to obtain the updated candidate solutions for each task. (17); (18); (19); in: Indicates the first Source task to target task Adaptability This represents the objective task of introducing a smoothing penalty term. The optimization function, Indicates the decoding function. Indicates the first Individual source task The second iteration There are 10 candidate solutions. Indicates task The total dimension of its own decision variables. For the target task With the The decision space distance of each source task Indicates the number of candidate solutions. Indicate the target task No. The iteration of the ... There are 10 candidate solutions. Indicate the target task With the Proximity between source tasks Represents the natural constant. Indicates task No. The iteration of the ... There are 10 candidate solutions. Indicates task No. The iteration of the ... There are 10 candidate solutions. This represents the preset internal learning baseline coefficient. This represents the preset baseline coefficient for heterogeneous task learning. The stochastic adjustment factor represents the update magnitude of the guiding term for the optimal candidate solution within the dynamic control task. This represents a stochastic adjustment factor that dynamically regulates the update magnitude of the heterogeneous task reference solution guidance term. Indicates task No. The optimal candidate solution for the next iteration. Indicates from the source task The selected reference solution; S413: When a candidate solution is determined to enter the internal collaborative learning mode, follow-up learning is first performed to guide the candidate solution to learn and update towards the optimal candidate solution according to equation (20), and the target task after follow-up learning is obtained. The candidate solutions are then used to follow up on the target task using a consultation and learning mechanism. The candidate solutions are updated according to equation (21) to obtain the target task after consultation and learning. The candidate solutions are then reviewed and learned through positive reinforcement or negative correction of the current state, and the target task after consultation learning is determined according to equation (22). The candidate solutions are updated to obtain the updated candidate solutions for each task: (20); (twenty one); (twenty two); in: Indicates the goals and tasks following the learning process. No. The iteration of the ... There are 10 candidate solutions. For the task No. Sub-iteration dependency factor This represents the preset maximum value of the dependency factor. This indicates the maximum preset number of iterations. Indicates the goals and tasks after consultation and learning. No. The iteration of the ... There are 10 candidate solutions. This indicates that teachers are learning stochastic adjustment factors. This indicates a peer-learning random adjustment factor. Indicates the current task after follow-up learning. The optimal candidate solution; Indicates from the task The average of two randomly selected top students from the set of top students. Indicates the random adjustment factor for determining the direction of the review. This represents the preset baseline coefficient for retaining the original state. This represents the preset neighborhood self-introspection fine-tuning baseline coefficient.

[0014] Furthermore, the method for obtaining the optimal solution for each task prediction in step S5 is as follows: S511: Merge the updated candidate solutions of each task to form a new population, and calculate the current optimal solution of each task according to equation (23): (twenty three); in: Indicate the target task No. The optimal solution in the next iteration. This represents the independent variable that corresponds to the maximum value of the function. This represents the objective task of introducing a smoothing penalty term. The optimization function, Indicate the target task No. The iteration of the ... One decoding candidate solution, Indicates the decoding function. Indicates task No. The iteration of the ... There are 10 candidate solutions. Indicates task The total dimension of its own decision variables. Indicates the number of candidate solutions; S512: Targeting the The optimization task uses a sliding window to collect the most recent consecutive... The optimal solution of the generation is constructed into a time series set of the optimal solution. The operator is combined with a dual-encoder and a single-decoder to construct a three-stage architecture of dual encoder-linear evolution-single decoder. The two encoders perform nonlinear transformations on the current optimal solution of each task, and the feature vectors output by the two encoders are concatenated to form a joint hidden state vector. The decoder takes the joint hidden state vector as input and reconstructs the original optimal solution through nonlinear transformation according to equation (24): (twenty four); in: Indicates task No. The optimal solution in the next iteration. express The joint hidden state vector obtained after mapping by the dual encoder, This indicates two independent nonlinear encoders. and Together they form a dual-channel parallel encoder. express Linear evolution operator, express The hidden state vector obtained after linear deduction Indicates a single-channel decoder. express The original optimal solution after reconstruction. This indicates the preset sliding window length parameter used to store historical optimal solutions; S513: Both the encoder and decoder use fully connected neural networks for network training. After the network training is completed, Based on equation (25), the optimal solution is predicted to obtain the predicted optimal solution for each task: (25); in: Indicate the target task No. The predicted optimal solution in the next iteration.

[0015] A post-earthquake multi-task collaborative optimization system is used to execute a post-earthquake multi-task collaborative optimization method as described in any one of the above, which includes an initial optimization function construction module for each task, an initial candidate solution evaluation module, a new optimization function acquisition module, an updated candidate solution acquisition module for each task, a predicted optimal solution acquisition module for each task, and an optimal solution output module for each task. The initial optimization function construction module for each task is used to carry out aftershock source location tasks after an earthquake, delineate the set of dangerous areas in the whole region, deploy emergency communication base stations in the area of ​​dangerous areas, deploy drone inspection tasks in the airspace above dangerous areas and communication blind spots, and construct the initial optimization function for each task. The initial candidate solution evaluation module is used to construct a unified search space for aftershock source location tasks, emergency communication base station deployment tasks, and UAV inspection tasks. In the unified search space, an independent population is constructed for each task, and each population contains multiple candidate solutions. The initial candidate solutions are evaluated to determine the initial dangerous area set, the initial communication blind zone set, and the initial inspection route set. The new optimization function acquisition module is used to uniformly and equivalently transform all spatial constraints into spatial geometric constraint inequalities for emergency communication base station deployment tasks and UAV inspection tasks based on the initial dangerous area set, initial communication blind zone set, and initial inspection route set. It also constructs penalty terms for the spatial geometric constraint inequalities of emergency communication base station deployment tasks and UAV inspection tasks, and substitutes the penalty terms into the initial optimization function of the corresponding task to obtain the new optimization function of the corresponding task. The updated candidate solution acquisition module for each task is used to construct a learning mode determination rule for each candidate solution, and obtain the updated candidate solution for each task based on the learning mode determination rule and the new optimization function of the corresponding task. The module for obtaining the predicted optimal solution for each task is used to merge the updated candidate solutions of each task to form a new population, and calculate the current optimal solution for each task in the new population. When the number of iterations reaches a preset number, the module predicts the subsequent optimal solution based on the historical optimal solution sequence within the sliding window of each task, and obtains the predicted optimal solution for each task. The optimal solution output module for each task synchronously updates the dangerous area set, communication blind spot set, and inspection route set based on the current optimal solution of each task until the iteration is completed and the optimal solution of each task is output.

[0016] Beneficial effects of the invention: This invention constructs an integrated joint solution framework adapted to post-earthquake multi-tasks, transforming the output results of three sub-tasks—aftershock source location, emergency communication base station deployment, and UAV inspection—into spatial constraints of the seismic field. It also completes the unified standardization of inequalities. Then, through learning mode determination, hierarchical learning, phased learning, and time-series optimal solution prediction, it performs synchronous parallel iteration and global joint optimization of the three sub-tasks, effectively improving the accuracy of optimal solution solving for each task. It can meet the on-site spatial no-entry rules and has stronger reliability in implementation. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of the overall process of the present invention. Detailed Implementation

[0018] A post-earthquake multi-task collaborative optimization method, the overall flowchart of which is shown in Figure 1, includes the following steps: S1: After the earthquake, carry out the task of locating the aftershock source, delineate the set of dangerous areas in the whole region, deploy emergency communication base stations in the area of ​​dangerous areas, deploy drone inspection tasks in the airspace above the dangerous areas and communication blind spots, and construct the initial optimization function for each task. After an earthquake, sensing equipment is rapidly deployed in the affected area to support the monitoring of subsequent aftershocks. Based on the theory of seismic wave propagation, the theoretical and actual time differences between signals received by any two sensors can be obtained. Under ideal conditions where the source coordinates are unbiased, the observed time difference of seismic waves arriving at different sensors perfectly matches the theoretical time difference, and the corresponding time difference residual is 0. With minimizing the sum of squared global residuals as the optimization objective, a maximization form is constructed through reciprocal transformation, thereby constructing the initial optimization function for the aftershock source location task as Equation (1).

[0019] After each iteration, the predicted [number]th ... The coordinates of the aftershock source are as follows: , This indicates the number of aftershock source signals, based on the set danger radius. Using the two-dimensional plane coordinates of the earthquake source Center of the circle A two-dimensional planar danger zone is generated with radius [missing information]; the danger zones corresponding to all aftershocks are merged to form a global danger zone. ,Will Feedback is sent to emergency communication base station deployment tasks and drone inspection tasks.

[0020] Emergency communication base station deployment tasks can use mobile vehicle-mounted base stations to maximize communication coverage in communication-damaged areas as the optimization goal. At the same time, it is required that base stations must be deployed in communication-damaged areas, prohibited from being deployed in dangerous areas, and at least one base station must be deployed around the disaster-stricken area where communication is interrupted. The drone flight path must achieve continuous communication coverage throughout. In each round of solution, a set of communication blind spots is output and fed back to the drone inspection task to guide the drone to avoid the no-signal blind areas throughout the entire process.

[0021] After an earthquake, disaster-stricken areas for communication repair can be designated based on the actual situation, and communication repair teams can be deployed in the areas where communication is damaged. There are 10 emergency base stations, and the coverage radius of each emergency base station is 1000. Discretize the reconstructed region into A grid of equal area is used, and the center point of each grid is the target point for communication service coverage. The total number of target points covered is 1. If the target point for coverage is located within a circle with the base station as the center and a radius of... Within the circle, it indicates that the grid point can achieve communication restoration. Coverage target points by The probability of coverage by an emergency base station for: , Indicates the first Coverage target points With the The distance between emergency base stations This indicates the coverage radius of the base station.

[0022] Each coverage target point can be covered by multiple base station nodes, and the probability of joint coverage by multiple base stations. for The coverage rate is the ratio of the number of covered target points to the total number of covered target points. , .

[0023] In areas where native communication is lost, after removing the effective coverage area of ​​newly built base stations based on the base station deployment plan, the remaining uncovered grid areas constitute the set of communication blind spots. Emergency communication base stations should be located within the communication disruption area, i.e., [areas with...]. , , The x-coordinate range represents the area where communications were damaged in the disaster. Indicates the first The x-coordinate of each emergency base station Indicates the first The vertical coordinate of each emergency base station, This represents the vertical coordinate range of the area where communication was destroyed after the disaster. Furthermore, at least one emergency communication base station must be deployed at the point of communication disruption in the disaster area. Surrounding area Within a 1.5-meter radius, emergency communication base stations must not be deployed in dangerous areas. Internally, and ensure the optimal inspection trunk line. any point on Not in communication blind spots. Based on the above conditions, the initial optimization function for the emergency communication base station deployment task is obtained as equation (2).

[0024] After obtaining the optimal deployment scheme for emergency communication base stations in each round of solution, the set of communication blind spots corresponding to the optimal base station deployment scheme is fed back to the UAV inspection mission to ensure that the disaster inspection route avoids communication blind spots.

[0025] The drone inspection mission can adopt a fixed-altitude low-altitude flight mode. An optimization function is constructed by comprehensively considering the total flight range and the urgency of the disaster area. The order of inspection visits to the disaster area can be generated by random key encoding. The inspection route must avoid dangerous areas and communication blind spots to ensure that the drone is not damaged by landslides or falling rocks and to avoid losing contact. The optimal inspection route in each round is fed back to the base station deployment task to iteratively correct the base station layout and ensure that the drone inspection route achieves uninterrupted communication coverage and transmits data in real time.

[0026] The low-altitude reconnaissance drone departs from its starting point and sequentially traverses all disaster areas, completing only one low-altitude aerial survey of each area. The inspection process simultaneously achieves two optimization objectives: minimizing the total flight distance of the drone from its starting point to the final surveyed disaster area, thus improving the efficiency of disaster assessment; and prioritizing aerial surveys of disaster areas with higher risks of damaged buildings and trapped people, ensuring priority surveying of key disaster areas, and ensuring that any coordinate point along the drone's inspection route is covered. Landing over dangerous areas and areas with no communication coverage is prohibited.

[0027] In summary, the initial optimization function for the UAV inspection task can be constructed as Equation (3).

[0028] (1); (2); (3); in: Indicates the first The sensor and the first The sensor received the first The theoretical time difference of each aftershock source signal. Indicates the first The sensor and the first The sensor received the first The actual time difference of each aftershock source signal This represents the Euclidean distance between two points. Indicates the first Spatial coordinates of the aftershock source Indicates the first Sensor coordinates, Indicates the first Sensor coordinates, This indicates the speed at which seismic waves propagate through the Earth's strata. Indicates the first The sensor received the first The timing of each aftershock source signal Indicates the first The sensor received the first The timing of each aftershock source signal This represents the initial optimization function for the aftershock source location task. This prevents positive numbers with a denominator of 0. Indicates the number of aftershock source signals. Indicates the number of sensors. This represents the initial optimization function for the emergency communication base station deployment task. Represents the set consisting of all base station locations. Indicates the first Coverage target point locations Indicates the probability of joint coverage by multiple base stations. This represents the number of target points covered by the discretization of the reconstructed area. Indicates constraints. This represents the decision variables for the deployment of emergency communication base stations. The x-coordinate range represents the area where communications were damaged in the disaster. Indicates the first The x-coordinate of each emergency base station Indicates the first The vertical coordinate of each emergency base station, This represents the vertical coordinate range of the area where communications were destroyed after the disaster. This indicates the danger zone predicted by the aftershock source location mission. Indicates the area where communications were destroyed after the disaster. Locations of each disaster area This indicates the number of disaster-stricken points within the area where communications were destroyed after the disaster. This represents the set of points within the disaster area whose communications were destroyed after the disaster. The first deployment Location of an emergency base station Indicates any, This indicates that it exists. This indicates the maximum distance between disaster-affected locations and emergency base stations within the area where communications are disrupted after a disaster. Indicates any point on the inspection route. This indicates the optimal inspection route. Indicates a communication blind spot. This represents the decision variables for drone inspection tasks. This indicates the total flight distance of the drone inspection. This represents the weighted sum of the urgency levels of the disaster in the affected areas. This is the initial optimization function for the UAV inspection mission. This indicates the total number of disaster areas to be inspected. Indicates the first Each disaster-stricken area was inspected. Indicates the first The urgency of the disaster situation at each inspected disaster area point The weighting coefficient represents the total flight distance. The weighting coefficients represent the total weighted sum of the urgency levels of disaster areas. Indicates the drone inspection route. This represents the set of communication blind spots under the optimal deployment scheme. Indicates the number of emergency base stations. This represents the horizontal plane projection operator.

[0029] In an optimized deployment of drone inspection missions, a random key encoding method can be used, where the dimension of the decision variable equals the total number of disaster areas to be surveyed, and the decision variable of dimension i is... The dimension component is the corresponding first dimension component. The random coded keys of each disaster area point are mapped one-to-one with the disaster areas to be inspected. The order of drone inspection visits to disaster areas is generated by encoding and decoding as follows: First, all random key values ​​in the decision variables are sorted in descending order. Then, the original dimension index corresponding to each key value after sorting is extracted. The final output index sequence is used as the order of drone inspections of disaster areas.

[0030] The optimal inspection route obtained from each round of calculation Feedback is sent to the emergency communication base station deployment task, and the deployment plan of the base station is used to meet the full communication coverage requirements of the inspection route.

[0031] S2: Construct a unified search space for aftershock source location tasks, emergency communication base station deployment tasks, and UAV inspection tasks. In the unified search space, construct an independent population for each task. Each population contains multiple candidate solutions. Evaluate the initial candidate solutions and determine the initial dangerous area set, the initial communication blind zone set, and the initial inspection route set. The task of locating aftershock sources is designated as Task 1, the deployment of emergency communication base stations as Task 2, and drone inspection as Task 3. To achieve coordinated optimization of aftershock location, emergency base station deployment, and drone inspection, a unified search space can be constructed for these three tasks. The dimension of the unified search space is determined by the largest dimension of the decision variables for all tasks, i.e. After determining the unified search space, for each task to be optimized... Initialize a containing A population of candidate solutions, each candidate solution having a dimension equal to The values ​​of each component of the candidate solution are constrained by The interval is then used to evaluate the initial candidate solutions for each task.

[0032] Specifically, the following methods can be used to evaluate the initial candidate solutions: S211: Decode the data in the unified search space back into the original space according to equation (4): (4); in: Indicates the decoding function. For the task The The data of candidate solutions in the unified search space Indicates task No. The candidate solution of the nth Dimensional data, express Data after decoding back to the original space. Indicates task No. The lower bound of the original variable's value. Indicates task No. The upper bound of the values ​​of the original variable. Indicates task The total dimension of its own decision variables. This indicates sequentially applying the first dimension to the second dimension. Each dimension of the data is calculated according to the formula in parentheses, and then the calculation results of each dimension are concatenated in dimensional order to generate a complete original vector. S212: Substitute each decoded candidate solution into the corresponding initial optimization function to calculate the value of the corresponding initial optimization function. Based on the value of the corresponding initial optimization function, evaluate and select the decoded candidate solution corresponding to the maximum value as the optimal solution for each task. Based on the optimal solution for each task, determine the initial dangerous area set, the initial communication blind zone set, and the initial inspection route set.

[0033] S3: Based on the initial set of hazardous areas, initial set of communication blind spots, and initial set of inspection routes, all spatial constraints are uniformly and equivalently transformed into spatial geometric constraint inequalities for the emergency communication base station deployment task and the UAV inspection task. Penalty terms are then constructed for these spatial geometric constraint inequalities, and these terms are substituted into the initial optimization function of the corresponding task to obtain a new optimization function. Specifically, the transformed spatial geometric constraint inequalities in the emergency communication base station deployment task are as follows: S311: Regarding the requirement that emergency base stations should be deployed in areas where communication is damaged, this is transformed into an inequality judgment rule (5): (5); in: This indicates that base stations should be deployed in areas where communication is disrupted, under the emergency communication base station deployment task. Inequalities This represents the decision variables for the deployment of emergency communication base stations. Indicates the first The x-coordinate of each emergency base station Indicates the first The vertical coordinate of each emergency base station, The x-coordinate range represents the area where communications were damaged in the disaster. This represents the vertical coordinate range of the area where communications were destroyed after the disaster. Indicates the number of emergency base stations; S312: Emergency communication base station deployment work ensures the restoration of communication services to disaster-stricken areas where communication has been interrupted. This requires that the surrounding areas of each disaster-stricken communication point be protected. At least one base station shall be deployed within a range of meters. Based on this, the inequality is constructed as equation (6): (6); in: This indicates that the emergency communication base station deployment task is aimed at the first Inequality regarding base station coverage limitations at a single point of communication disruption. This represents the set of points within the disaster area whose communications were destroyed after the disaster. This indicates the maximum distance between disaster-affected locations and emergency base stations within the area where communications are disrupted after a disaster. Indicates the area where communications were destroyed after the disaster. Coordinates of the disaster area locations This indicates the number of disaster-stricken points within the area where communications were disrupted after the disaster. S313: For the deployment of emergency communication base stations and the set of prohibited dangerous areas, the criterion is that the distance from the deployment location of each emergency base station to the center of the dangerous area is greater than or equal to the preset radius of the dangerous area. The inequality is constructed as Equation (7): (7); in: This represents the Lth inequality for the deployment of emergency communication base stations under the restricted conditions of no-entry dangerous areas. Indicates the preset number Danger radius of each aftershock source Indicates the predicted first The epicenter of each aftershock Coordinates along the axial direction, Indicates the predicted first The epicenter of each aftershock Coordinates along the axial direction, This indicates the number of aftershock source signals. Indicates the number of emergency base stations. This indicates the danger zone predicted by the aftershock source location mission. S314: For the inspection route set, the entire flight area is projected onto a two-dimensional horizontal plane to carry out distance geometric calculations. Based on the idea that the shortest distance from a point to a finite line segment is greater than or equal to the blind zone radius, an inequality is established as equation (8): (8); in: Indicates the first From the communication blind spot to the optimal inspection route The disaster-stricken areas that were inspected and The projection coefficient of the line segment connecting the inspected disaster area points. This indicates that the deployment of emergency communication base stations is carried out under the condition that the drone inspection route does not pass through communication blind spots. Inequalities This indicates the first in the optimal inspection route. Two-dimensional plane coordinates of the disaster area points being inspected. Indicates the first Coordinates of a communication blind spot Indicates the radius of the blind zone. This indicates the total number of disaster areas to be inspected. Indicates the number of communication blind spots. This indicates the optimal inspection route for a drone inspection mission.

[0034] Specifically, the spatial geometric constraint inequalities transformed in the UAV inspection mission are as follows: S321: Combining the set of hazardous areas output by aftershock location, the entire flight area is projected onto a two-dimensional horizontal plane, and a geometric judgment criterion is set: the shortest distance from the center of each hazardous area to the inspection route under the two-dimensional projection is not less than the preset hazardous radius. The resulting inequality is equation (9): (9); in: Indicates the first From the epicenter of each aftershock to the inspection route The disaster-stricken areas that were inspected and The projection coefficient of the line segment connecting the inspected disaster area points. This indicates the first time that a drone inspection mission is conducted under restricted airspace conditions in a prohibited danger zone. Inequalities, Indicates the first The disaster-stricken areas that were inspected Axial coordinates, Indicates the first The disaster-stricken areas that were inspected Coordinates along the axial direction, Indicates the predicted first The epicenter of each aftershock Coordinates along the axial direction, Indicates the predicted first The epicenter of each aftershock Coordinates along the axial direction, This indicates the number of aftershock source signals. This indicates the total number of disaster areas to be inspected. This indicates the danger zone predicted by the aftershock source location mission. Indicates the preset number Danger radius of each aftershock source; S322: For the communication blind zone given in the base station deployment work, the entire flight area is projected onto a two-dimensional horizontal plane. Based on the idea that the shortest distance from the center of the communication blind zone to each line segment is greater than or equal to the radius of the blind zone, it is transformed into the following inequality as equation (10): (10); in: This represents the optimal base station deployment scheme. From the first communication blind spot to the inspection route The disaster-stricken areas that were inspected and The projection coefficient of the line segment connecting the inspected disaster area points. This indicates the first time that a drone inspection mission is conducted under restricted airspace conditions in a no-entry communication blind zone. Inequalities, The preset communication blind zone radius, The set of communication blind spots under the optimal base station deployment scheme. blind spots Axial coordinates, The set of communication blind spots under the optimal base station deployment scheme. blind spots Coordinates along the axial direction, This represents the total number of communication blind spots in the optimal base station deployment scheme. This represents the set of communication blind spots under the optimal deployment scheme. This represents the decision variables for drone inspection tasks.

[0035] For each spatial geometric constraint inequality, a piecewise smooth function can be used for approximation. The smoothing parameter decreases monotonically with the number of iterations, and then a penalty term is constructed based on the violation quantity after smoothing.

[0036] Specifically, the method for constructing penalty terms for the spatial geometric constraint inequalities of emergency communication base station deployment tasks and UAV inspection tasks, and substituting these penalty terms into the initial optimization function of the corresponding task to obtain the new optimization function for the corresponding task is as follows: S331: Application of spatial geometric constraint inequalities The penalty technique is used to optimize the initial function of the emergency communication base station deployment task. Combining the spatial geometric constraint inequality, we transform it according to equation (11) to obtain the emergency base station deployment optimization function with a penalty term. The initial optimization function for the drone inspection task Combining the spatial geometric constraint inequality, we transform it according to equation (12) to obtain the UAV inspection task optimization function with the introduction of a penalty term. : (11); (12); in: This represents the decision variables for the deployment of emergency communication base stations. This represents the penalty parameter that increases with the number of iterations. This indicates the danger zone predicted by the aftershock source location mission. This indicates the optimal inspection route for a drone inspection mission. This indicates that base stations should be deployed in areas where communication is disrupted, under the emergency communication base station deployment task. Inequalities This indicates that the emergency communication base station deployment task is aimed at the first Inequality regarding base station coverage limitations at a single point of communication disruption. This represents the Lth inequality for the deployment of emergency communication base stations under the restricted conditions of no-entry dangerous areas. This indicates that the deployment of emergency communication base stations is carried out under the condition that the drone inspection route does not pass through communication blind spots. Inequalities Indicates the number of communication blind spots. This indicates the first time that a drone inspection mission is conducted under restricted airspace conditions in a prohibited danger zone. Inequalities, Indicates the number of emergency base stations. This indicates the number of aftershock source signals. This indicates the total number of disaster areas to be inspected. This represents the total number of communication blind spots in the optimal base station deployment scheme. This represents the decision variables for drone inspection tasks. This represents the set of communication blind spots under the optimal deployment scheme. This refers to the first drone inspection mission conducted under restricted airspace conditions in a no-entry communication blind zone. Inequalities, This indicates the number of disaster-stricken points within the area where communications were disrupted after the disaster. S332: By using the piecewise smooth function (13) and By combining penalty techniques, a new emergency base station deployment optimization function with the introduction of a smooth penalty term is obtained according to equation (14). According to equation (15), the new UAV inspection optimization function after introducing a smoothing penalty term is obtained. ; (13); (14); (15); in: Represents a smooth function. This represents the value that the function takes on the left side of any inequality. Represents the smoothness parameter. It is a positive number that gradually decreases with each iteration. Represents the natural constant.

[0037] By using piecewise smooth functions and Combining punishment techniques can overcome Under punishment The operator suffers from the defects of being non-smooth and non-differentiable.

[0038] S4: Construct learning mode determination rules for each candidate solution, and obtain updated candidate solutions for each task based on the learning mode determination rules and the corresponding new optimization functions; For each candidate solution in each task, a learning mode determination rule can be constructed based on the number of consecutive stagnant update generations of the optimal solution in each task and the spatial feature coefficients of the candidate solutions. Specifically: A preset critical threshold for optimal solution stagnation is established. When the number of consecutive stagnant update generations of the optimal solution exceeds this threshold, the local diversity of the solutions is measured based on the spatial distance between the candidate solutions and the optimal solution. When candidate solutions are distributed within the neighborhood of the optimal solution, to avoid getting trapped in local optima, multi-task collaborative learning is enabled to expand the optimization range and escape local optima. When candidate solutions are far from the optimal region, the candidate solutions adopt the internal collaborative learning mode for this task.

[0039] Specifically, the method for obtaining the updated candidate solutions for each task is as follows: S411: For each candidate solution in each task, construct the learning mode determination rule according to equation (16): (16); in: Indicates task No. The second iteration The learning mode selected from the candidate solutions Indicates task The optimal solution continuously stagnates during the update algebra. This indicates the preset critical threshold for the optimal solution to stagnate and update. Represents logical OR, This indicates logical AND. Indicates the current iteration number. Indicates task No. The second iteration Spatial characteristic coefficients of candidate solutions This represents the critical coefficient for partitioning the neighborhood of the candidate solution; When satisfied When, the spatial characteristic coefficients of the candidate solution The calculation formula is as follows: ; in: It is a Euclidean norm; Indicates task No. In the nth iteration There are 10 candidate solutions. For the task No. The optimal candidate solution in the next iteration; Indicates candidate solutions With the optimal candidate solution Euclidean distance; Indicates task No. In the nth iteration There are 10 candidate solutions; This indicates the number of candidate solutions for each task; For the task No. The optimal candidate solution is obtained from all candidate solutions in the next iteration. The average Euclidean distance; To represent a very small positive number, to prevent the denominator from being 0. For candidate solutions The relative distance coefficient; Represents the hyperbolic tangent function. This is the hyperbolic tangent scaling control coefficient.

[0040] S412: When a candidate solution is determined to enter the multi-task collaborative learning mode, firstly, the adaptability is evaluated according to Equation (17) to obtain the adaptability of the source task to the target task, and the adaptability is normalized. Then, the distance proximity between the target task and the source task is calculated according to Equation (18). Then, the normalized adaptability and the distance proximity are weighted and fused to obtain a comprehensive evaluation value. Then, the probability of collaborative learning between the target task and the source task is calculated according to the comprehensive evaluation value. Based on the probability of collaborative learning between the target task and the source task, the selected source task is determined. A reference solution is selected from the selected source tasks. Based on the selected reference solution, multi-task collaborative learning is performed on the candidate solution according to Equation (19) to obtain the updated candidate solutions for each task. (17); (18); (19); in: Indicates the first Source task to target task Adaptability This represents the objective task of introducing a smoothing penalty term. The optimization function, when When equal to 1, for , Indicates the decoding function. Indicates the first Individual source task The second iteration There are 10 candidate solutions. Indicates task The total dimension of its own decision variables. For the target task With the The decision space distance of each source task Indicates the number of candidate solutions. Indicate the target task No. The iteration of the ... There are 10 candidate solutions. Indicate the target task With the Proximity between source tasks Represents the natural constant. Indicates task No. The iteration of the ... There are 10 candidate solutions. Indicates task No. The iteration of the ... There are 10 candidate solutions. This represents the preset internal learning baseline coefficient. This represents the preset baseline coefficient for heterogeneous task learning. The stochastic adjustment factor represents the update magnitude of the guiding term for the optimal candidate solution within the dynamic control task. This represents a stochastic adjustment factor that dynamically regulates the update magnitude of the heterogeneous task reference solution guidance term. Indicates task No. The optimal candidate solution for the next iteration. Indicates from the source task The selected reference solution; The fitness evaluation is used to quantify the actual performance of the other two source task candidate solutions under the target task optimization objective, and directly reflects the usability of the source tasks. The greater the fitness, the more useful the source tasks are to the target tasks.

[0041] The fit is normalized as follows: ; Indicates the first Source task to target task Normalized fit Indicates the source task traversal index.

[0042] Decision space distance is used to measure the proximity of two task populations in the solution space, judging task relevance from a structural perspective. Using the average Euclidean distance between candidate solutions can more realistically reflect the differences in population distribution; the smaller the distance, the closer the spatial distribution of the tasks, and the higher the value of collaborative learning.

[0043] The expression for the weighted fusion of normalized fitness and proximity is: ; Indicate the target task With the The comprehensive evaluation value of each source task This represents the preset comprehensive evaluation balance coefficient.

[0044] The target task can then be calculated using the following formula. With the The probability of collaborative learning of individual source tasks : ; in: Indicate the target task With the The comprehensive evaluation value of each source task.

[0045] Then, based on the probability of collaborative learning between the target task and the source task, the selected source task is determined according to the following formula: ;in: This indicates the final source task number determined after screening. Indicate the target task The task number of the first source task. Indicate the target task The task number of the second source task. Indicate the target task The probability of collaborative learning with the first source task. This represents a uniformly random number selected from the source task.

[0046] Furthermore, the method for selecting a reference solution from the selected source tasks is as follows: [Select the source tasks...] All candidate solutions are sorted from highest to lowest based on the value of their new optimization function. The top-ranked candidate solutions are designated as superior solutions, the bottom-ranked as inferior solutions, and the remaining candidate solutions are designated as intermediate solutions, forming a hierarchical set. In the multi-task collaborative learning phase, optimization information from superior solutions is extracted with high probability, optimization experience from intermediate solutions is reused with moderate probability, and features of inferior solutions are referenced with low probability. The specific rules are as follows: ;in, Indicates from the source task The selected reference solution; Indicates from the source task A randomly selected superior solution; Indicates from the source task A randomly selected intermediate solution; From the source task A randomly selected inferior solution; This represents the pre-defined sampling probability threshold between superior and intermediate solutions. This represents the pre-defined sampling probability threshold between intermediate and inferior solutions. This represents the probability sampling interval for determining random numbers in the stratified solution set.

[0047] Selecting with probability Then, we can perform multi-task collaborative learning on the candidate solutions according to equation (19) to obtain the updated candidate solutions for each task.

[0048] Equation 19 introduces two random numbers, which can enhance the ability to explore in the solution space.

[0049] After completing the multi-task collaborative learning, the candidate solutions before and after the update are decoded according to equation (4), and then input into the new optimization function for calculation. In the solution update stage, a greedy strategy is adopted to retain high-quality solutions. The specific formula of the greedy strategy is as follows: This strategy filters the two candidate solutions before and after the update. If the updated candidate solution has a higher value of the new optimization function, the new solution is adopted; otherwise, the original solution from the previous generation is retained.

[0050] In the internal collaborative learning model, after each stage of follow-up learning, consultation learning, and debriefing learning, the same greedy strategy is used to retain the high-quality solution by applying the above method to the two solutions before and after the update.

[0051] S413: When a candidate solution is determined to enter the internal collaborative learning mode, follow-up learning is first performed to guide the candidate solution to learn and update towards the optimal candidate solution according to equation (20), and the target task after follow-up learning is obtained. The candidate solutions are then used to follow up on the target task using a consultation and learning mechanism. The candidate solutions are updated according to equation (21) to obtain the target task after consultation and learning. The candidate solutions are then reviewed and learned through positive reinforcement or negative correction of the current state, and the target task after consultation learning is determined according to equation (22). The candidate solutions are updated to obtain the updated candidate solutions for each task: (20); (twenty one); (twenty two); in: Indicates the goals and tasks following the learning process. No. The iteration of the ... There are 10 candidate solutions. For the task No. Sub-iteration dependency factor The value of will be dynamically and adaptively adjusted according to the current iteration number, thereby achieving reasonable control of the learning intensity at different stages. This represents the preset maximum value of the dependency factor. This indicates the maximum preset number of iterations. Indicates the goals and tasks after consultation and learning. No. The iteration of the ... There are 10 candidate solutions. This indicates that teachers are learning stochastic adjustment factors. This indicates a peer-learning random adjustment factor. Indicates the current task after follow-up learning. The optimal candidate solution; Indicates from the task The average of two randomly selected top students from the set of top students. The knowledge of the teacher and the knowledge of the student work synergistically to promote the evolution of candidate solutions. Indicates the random adjustment factor for determining the direction of the review. This represents the preset baseline coefficient for retaining the original state. This represents the preset neighborhood self-introspection fine-tuning baseline coefficient. and They respectively play a regulatory role in self-consolidation and self-reflection, when When the conditions are favorable, positive adjustments are made to push the candidate solution closer to a better state; conversely, negative adjustments are made to guide the candidate solution away from its unfavorable state, thus achieving fine-tuning.

[0052] The internal collaborative learning model imitates the learning habits of top students and is designed with three stages of learning, which include follow-up learning, consultation learning, and debriefing learning.

[0053] The follow-up learning phase draws on the habit of high-achieving students who actively follow and deeply participate in the teacher's teaching pace in class, rather than passively receiving information. By guiding candidate solutions to learn from the optimal solution, the quality of candidate solutions can be improved.

[0054] In real-world learning scenarios, becoming an excellent student often involves more than just classroom learning; students proactively seek help from teachers and high-achieving classmates during breaks. Inspired by this behavior, this invention employs a consultation-based learning mechanism. It constructs a set of high-achieving students based on the values ​​of a new optimization function, allowing students to learn from both teachers and other high-achieving peers.

[0055] After completing their school studies, high-achieving students often engage in self-directed review and reflection at home. The update mechanism simulates this habit of high-achieving students, enabling a refined exploration of the solution space through positive reinforcement or negative correction of their current state.

[0056] S5: Merge the updated candidate solutions of each task to form a new population, and calculate the current optimal solution of each task in the new population. When the number of iterations reaches the preset number, predict the subsequent optimal solution based on the historical optimal solution sequence within the sliding window of each task to obtain the predicted optimal solution of each task. Specifically, the method for obtaining the optimal solution for each task prediction is as follows: S511: Merge the updated candidate solutions of each task to form a new population, and calculate the current optimal solution of each task according to equation (23): (twenty three); in: Indicate the target task No. The optimal solution in the next iteration. This represents the independent variable that corresponds to the maximum value of the function. This represents the objective task of introducing a smoothing penalty term. The optimization function, when When equal to 1, for , Indicate the target task No. The iteration of the ... One decoding candidate solution, Indicates the decoding function. Indicates task No. The iteration of the ... There are 10 candidate solutions. Indicates task The total dimension of its own decision variables. Indicates the number of candidate solutions; S512: Targeting the The optimization task uses a sliding window to collect the most recent consecutive... The optimal solution of the generation is constructed into a time series set of the optimal solution. The operator is combined with a dual-encoder and a single-decoder to construct a three-stage architecture of dual encoder-linear evolution-single decoder. The two encoders perform nonlinear transformations on the optimal solutions of each task, and the feature vectors output by the two encoders are concatenated to form a joint hidden state vector. The decoder takes the joint hidden state vector as input and reconstructs the original optimal solution through nonlinear transformation according to equation (24): (twenty four); in: Indicates task No. The optimal solution in the next iteration. express The joint hidden state vector obtained after mapping by the dual encoder, This indicates two independent nonlinear encoders. and Together they form a dual-channel parallel encoder. express The linear evolution operator is responsible for performing linear derivations within the latent space. express The hidden state vector obtained after linear deduction This indicates a single-channel decoder used to reconstruct the joint hidden state into the original optimal solution. express The original optimal solution after reconstruction. This indicates the preset sliding window length parameter used to store historical optimal solutions; S513: Both the encoder and decoder use fully connected neural networks for network training. After the network training is completed, Based on equation (25), the optimal solution is predicted to obtain the predicted optimal solution for each task. If the predicted optimal solution for each task is better than the current optimal solution for each task, the optimal solution for each task is updated based on the predicted optimal solution for each task. (25); in: Indicate the target task No. The predicted optimal solution in the next iteration.

[0057] When both the encoder and decoder are trained using fully connected neural networks, the following loss function is constructed by combining the state reconstruction loss of the dual encoder-single decoder and the latent space linear evolution loss: ; in, For the loss function, the first term The state reconstruction loss is used to ensure that the decoder can accurately reconstruct the original optimal solution from the latent features; the second term... The latent space linear evolution loss is used to constrain the latent feature sequences to follow... The linear dynamics of the operator definition.

[0058] S6: If the predicted optimal solution is better, then update the current optimal solution of each task to the predicted optimal solution, and simultaneously update the dangerous area set, communication blind spot set, and inspection route set based on the current optimal solution of each task, until the optimal solution of each task is output after the iteration is completed.

[0059] This invention constructs an integrated joint solution framework adapted to post-earthquake multi-tasks. In each round of optimization, a three-layer information feedback link is built-in to transform the output results of three sub-tasks—aftershock source location, emergency communication base station deployment, and UAV inspection—into seismic field spatial constraints. Inequality unification and standardization are completed for subsequent guidance of other tasks. Through learning mode determination, hierarchical learning, phased learning, and time-series optimal solution prediction, the three sub-tasks are synchronously and iteratively combined globally for optimization, effectively improving the accuracy of optimal solution for each task. This results in an integrated planning scheme that meets on-site spatial prohibition rules and has stronger implementation reliability.

[0060] A post-earthquake multi-task collaborative optimization system is used to execute a post-earthquake multi-task collaborative optimization method as described in any one of the above, which includes an initial optimization function construction module for each task, an initial candidate solution evaluation module, a new optimization function acquisition module, an updated candidate solution acquisition module for each task, a predicted optimal solution acquisition module for each task, and an optimal solution output module for each task. The initial optimization function construction module for each task is used to carry out aftershock source location tasks after an earthquake, delineate the set of dangerous areas in the whole region, deploy emergency communication base stations in the area of ​​dangerous areas, deploy drone inspection tasks in the airspace above dangerous areas and communication blind spots, and construct the initial optimization function for each task. The initial candidate solution evaluation module is used to construct a unified search space for aftershock source location tasks, emergency communication base station deployment tasks, and UAV inspection tasks. In the unified search space, an independent population is constructed for each task, and each population contains multiple candidate solutions. The initial candidate solutions are evaluated to determine the initial dangerous area set, the initial communication blind zone set, and the initial inspection route set. The new optimization function acquisition module is used to uniformly and equivalently transform all spatial constraints into spatial geometric constraint inequalities for emergency communication base station deployment tasks and UAV inspection tasks based on the initial dangerous area set, initial communication blind zone set, and initial inspection route set. It also constructs penalty terms for the spatial geometric constraint inequalities of emergency communication base station deployment tasks and UAV inspection tasks, and substitutes the penalty terms into the initial optimization function of the corresponding task to obtain the new optimization function of the corresponding task. The updated candidate solution acquisition module for each task is used to construct a learning mode determination rule for each candidate solution, and obtain the updated candidate solution for each task based on the learning mode determination rule and the new optimization function of the corresponding task. The module for obtaining the predicted optimal solution for each task is used to merge the updated candidate solutions of each task to form a new population, and calculate the current optimal solution for each task in the new population. When the number of iterations reaches a preset number, the module predicts the subsequent optimal solution based on the historical optimal solution sequence within the sliding window of each task, and obtains the predicted optimal solution for each task. The optimal solution output module for each task synchronously updates the dangerous area set, communication blind spot set, and inspection route set based on the current optimal solution of each task until the iteration is completed and the optimal solution of each task is output.

[0061] In summary, the post-earthquake multi-task collaborative optimization method and system provided by this invention effectively improves the accuracy of finding the optimal solution for each task and meets the requirements of on-site space access prohibition rules and landing reliability.

[0062] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A post-earthquake multi-task collaborative optimization method, characterized in that: Includes the following steps: S1: After the earthquake, carry out the task of locating the aftershock source, delineate the set of dangerous areas in the whole region, deploy emergency communication base stations in the area of ​​dangerous areas, deploy drone inspection tasks in the airspace above the dangerous areas and communication blind spots, and construct the initial optimization function for each task. S2: Construct a unified search space for aftershock source location tasks, emergency communication base station deployment tasks, and UAV inspection tasks. In the unified search space, construct an independent population for each task. Each population contains multiple candidate solutions. Evaluate the initial candidate solutions and determine the initial dangerous area set, the initial communication blind zone set, and the initial inspection route set. S3: Based on the initial dangerous area set, initial communication blind zone set, and initial inspection route set, all spatial constraints are uniformly and equivalently transformed into spatial geometric constraint inequalities for emergency communication base station deployment tasks and UAV inspection tasks. Penalty terms are constructed for the spatial geometric constraint inequalities of emergency communication base station deployment tasks and UAV inspection tasks. The penalty terms are substituted into the initial optimization function of the corresponding task to obtain the new optimization function of the corresponding task. S4: Construct learning mode determination rules for each candidate solution, and obtain updated candidate solutions for each task based on the learning mode determination rules and the corresponding new optimization functions; S5: Merge the updated candidate solutions of each task to form a new population, and calculate the current optimal solution of each task in the new population. When the number of iterations reaches the preset number, predict the subsequent optimal solution based on the historical optimal solution sequence within the sliding window of each task to obtain the predicted optimal solution of each task. S6: If the predicted optimal solution is better, then update the current optimal solution of each task to the predicted optimal solution, and simultaneously update the dangerous area set, communication blind spot set, and inspection route set based on the current optimal solution of each task, until the optimal solution of each task is output after the iteration is completed.

2. The post-earthquake multi-task collaborative optimization method according to claim 1, characterized in that: The initial optimization function for the aftershock source location task in step S1 is Equation (1), the initial optimization function for the emergency communication base station deployment task is Equation (2), and the initial optimization function for the UAV inspection task is Equation (3). （1）； （2）； （3）； in: Indicates the first The sensor and the first The sensor received the first The theoretical time difference of each aftershock source signal. Indicates the first The sensor and the first The sensor received the first The actual time difference of each aftershock source signal This represents the Euclidean distance between two points. Indicates the first Spatial coordinates of the aftershock source Indicates the first Sensor coordinates, Indicates the first Sensor coordinates, This indicates the speed at which seismic waves propagate through the Earth's strata. Indicates the first The sensor received the first The timing of each aftershock source signal Indicates the first The sensor received the first The timing of each aftershock source signal This represents the initial optimization function for the aftershock source location task. This prevents positive numbers with a denominator of 0. Indicates the number of aftershock source signals. Indicates the number of sensors. This represents the initial optimization function for the emergency communication base station deployment task. Represents the set consisting of all base station locations. Indicates the first Coverage target point locations Indicates the probability of joint coverage by multiple base stations. This represents the number of target points covered by the discretization of the reconstructed area. Indicates constraints. This represents the decision variables for the deployment of emergency communication base stations. The x-coordinate range represents the area where communications were damaged in the disaster. Indicates the first The x-coordinate of each emergency base station Indicates the first The vertical coordinate of each emergency base station, This represents the vertical coordinate range of the area where communications were destroyed after the disaster. This indicates the danger zone predicted by the aftershock source location mission. Indicates the area where communications were destroyed after the disaster. Locations of each disaster area This indicates the number of disaster-stricken points within the area where communications were destroyed after the disaster. This represents the set of points within the disaster area whose communications were destroyed after the disaster. The first deployment Location of an emergency base station Indicates any, This indicates that it exists. This indicates the maximum distance between disaster-affected locations and emergency base stations within the area where communications are disrupted after a disaster. This indicates a point on the inspection route. This indicates the optimal inspection route. Indicates a communication blind spot. This represents the decision variables for drone inspection tasks. This indicates the total flight distance of the drone inspection. This represents the weighted sum of the urgency levels of the disaster in the affected areas. This is the initial optimization function for the UAV inspection mission. This indicates the total number of disaster areas to be inspected. Indicates the first Each disaster-stricken area was inspected. Indicates the first The urgency of the disaster situation at each inspected disaster area point The weighting coefficient represents the total flight distance. The weighting coefficients represent the total weighted sum of the urgency levels of disaster areas. Indicates the drone inspection route. This represents the set of communication blind spots under the optimal deployment scheme. Indicates the number of emergency base stations. This represents the horizontal plane projection operator.

3. The post-earthquake multi-task collaborative optimization method according to claim 2, characterized in that: In step S1, when deploying the drone inspection task, a random key encoding method is used, and the dimension of the decision variable is equal to the total number of disaster areas to be surveyed. The dimension component is the corresponding first dimension component. The random coded keys of each disaster area point are mapped one-to-one with the disaster areas to be inspected. The order of drone inspection visits to disaster areas is generated by encoding and decoding as follows: First, all random key values ​​in the decision variables are sorted in descending order. Then, the original dimension index corresponding to each key value after sorting is extracted. The final output index sequence is used as the order of drone inspections of disaster areas.

4. The post-earthquake multi-task collaborative optimization method according to claim 1, characterized in that: In step S2, the initial candidate solutions are evaluated using the following method: S211: Decode the data in the unified search space back into the original space according to equation (4): （4）； in: Indicates the decoding function. For the task The The data of candidate solutions in the unified search space Indicates task No. The candidate solution of the nth Dimensional data, express Data after decoding back to the original space. Indicates task No. The lower bound of the original variable's value. Indicates task No. The upper bound of the values ​​of the original variable. Indicates task The total dimension of its own decision variables. This indicates sequentially applying the first dimension to the second dimension. Each dimension of the data is calculated according to the formula in parentheses, and then the calculation results of each dimension are concatenated in dimensional order to generate a complete original vector. S212: Substitute each decoded candidate solution into the corresponding initial optimization function to calculate the value of the corresponding initial optimization function. Based on the value of the corresponding initial optimization function, evaluate and select the decoded candidate solution corresponding to the maximum value as the optimal solution for each task. Based on the optimal solution for each task, determine the initial dangerous area set, the initial communication blind zone set, and the initial inspection route set.

5. The post-earthquake multi-task collaborative optimization method according to claim 1, characterized in that: The spatial geometric constraint inequalities transformed in step S3, the emergency communication base station deployment task, are as follows: S311: Regarding the requirement that emergency base stations should be deployed in areas where communication is damaged, this is transformed into an inequality judgment rule (5): （5）； in: This indicates the first emergency communication base station deployment task under the condition of base station deployment in areas where communication is destroyed. Inequalities This represents the decision variables for the deployment of emergency communication base stations. Indicates the first The x-coordinate of each emergency base station Indicates the first The vertical coordinate of each emergency base station, The x-coordinate range represents the area where communications were damaged in the disaster. This represents the vertical coordinate range of the area where communications were destroyed after the disaster. Indicates the number of emergency base stations; S312: Emergency communication base station deployment work ensures the restoration of communication services to disaster-stricken areas where communication has been interrupted. This requires that the surrounding areas of each disaster-stricken communication point be protected. At least one base station shall be deployed within a range of meters. Based on this, the inequality is constructed as equation (6): (6); in: This indicates that the emergency communication base station deployment task is aimed at the first Inequality regarding base station coverage limitations at a single point of communication disruption. This represents the set of points within the disaster area whose communications were destroyed after the disaster. This indicates the maximum distance between disaster-affected locations and emergency base stations within the area where communications are disrupted after a disaster. Indicates the area where communications were destroyed after the disaster. Coordinates of the disaster area locations This indicates the number of disaster-stricken points within the area where communications were disrupted after the disaster. S313: For the deployment of emergency communication base stations and the set of prohibited dangerous areas, the criterion is that the distance from the deployment location of each emergency base station to the center of the dangerous area is greater than or equal to the preset radius of the dangerous area. The inequality is constructed as Equation (7): （7）； in: This represents the Lth inequality for the deployment of emergency communication base stations under the restricted conditions of no-entry dangerous areas. Indicates the preset number Danger radius of each aftershock source Indicates the predicted first The epicenter of each aftershock Coordinates along the axial direction, Indicates the predicted first The epicenter of each aftershock Coordinates along the axial direction, This indicates the number of aftershock source signals. Indicates the number of emergency base stations. This indicates the danger zone predicted by the aftershock source location mission. S314: For the inspection route set, the entire flight area is projected onto a two-dimensional horizontal plane to carry out distance geometric calculations. Based on the idea that the shortest distance from a point to a finite line segment is greater than or equal to the blind zone radius, an inequality is established as equation (8): （8）； in: Indicates the first From the communication blind spot to the optimal inspection route The disaster-stricken areas that were inspected and The projection coefficient of the line segment connecting the inspected disaster area points. This indicates that the deployment of emergency communication base stations is carried out under the condition that the drone inspection route does not pass through communication blind spots. Inequalities This indicates the first in the optimal inspection route. Two-dimensional plane coordinates of the disaster area points being inspected. Indicates the first Coordinates of a communication blind spot Indicates the radius of the blind zone. This indicates the total number of disaster areas to be inspected. Indicates the number of communication blind spots. This indicates the optimal inspection route for a drone inspection mission.

6. The post-earthquake multi-task collaborative optimization method according to claim 1, characterized in that: The spatial geometric constraint inequalities transformed in step S3 of the UAV inspection task are as follows: S321: Combining the set of hazardous areas output by aftershock location, the entire flight area is projected onto a two-dimensional horizontal plane, and a geometric judgment criterion is set: the shortest distance from the center of each hazardous area to the inspection route under the two-dimensional projection is not less than the preset hazardous radius. The resulting inequality is equation (9): （9）； in: Indicates the first From the epicenter of each aftershock to the inspection route The disaster-stricken areas that were inspected and The projection coefficient of the line segment connecting the inspected disaster area points. This indicates the first time that a drone inspection mission is conducted under restricted airspace conditions in a prohibited or dangerous area. Inequalities, Indicates the first The disaster-stricken areas that were inspected Axial coordinates, Indicates the first The disaster-stricken areas that were inspected Coordinates along the axial direction, Indicates the first The disaster-stricken areas that were inspected Axial coordinates, Indicates the first The disaster-stricken areas that were inspected Coordinates along the axial direction, Indicates the predicted first The epicenter of each aftershock Coordinates along the axial direction, Indicates the predicted first The epicenter of each aftershock Coordinates along the axial direction, This indicates the number of aftershock source signals. This indicates the total number of disaster areas to be inspected. This indicates the danger zone predicted by the aftershock source location mission. Indicates the preset number Danger radius of each aftershock source; S322: For the communication blind zone given in the base station deployment work, the entire flight area is projected onto a two-dimensional horizontal plane. Based on the idea that the shortest distance from the center of the communication blind zone to each line segment is greater than or equal to the radius of the blind zone, it is transformed into the following inequality as equation (10): (10)； in: This represents the optimal base station deployment scheme. From a communication blind spot to the inspection route The disaster-stricken areas that were inspected and The projection coefficient of the line segment connecting the inspected disaster area points. This indicates the first time that a drone inspection mission is conducted under restricted airspace conditions in a no-entry communication blind zone. Inequalities, The preset communication blind zone radius, The set of communication blind spots under the optimal base station deployment scheme. blind spots Axial coordinates, The set of communication blind spots under the optimal base station deployment scheme. blind spots Coordinates along the axial direction, This represents the total number of communication blind spots in the optimal base station deployment scheme. This represents the set of communication blind spots under the optimal deployment scheme. This represents the decision variables for drone inspection tasks.

7. The post-earthquake multi-task collaborative optimization method according to claim 1, characterized in that: In step S3, the spatial geometric constraint inequalities for the emergency communication base station deployment task and the UAV inspection task are used to construct penalty terms. Substituting these penalty terms into the initial optimization function of the corresponding task yields the new optimization function for that task, as follows: S331: Application of spatial geometric constraint inequalities The penalty technique is used to optimize the initial function of the emergency communication base station deployment task. Combining the spatial geometric constraint inequality, we transform it according to equation (11) to obtain the emergency base station deployment optimization function with a penalty term. The initial optimization function for the drone inspection task Combining the spatial geometric constraint inequality, we transform it according to equation (12) to obtain the UAV inspection task optimization function with the introduction of a penalty term. : （11）； （12）； in: This represents the decision variables for the deployment of emergency communication base stations. This represents the penalty parameter that increases with the number of iterations. This indicates the danger zone predicted by the aftershock source location mission. This indicates the optimal inspection route for a drone inspection mission. This indicates that base stations should be deployed in areas where communication is disrupted, under the emergency communication base station deployment task. Inequalities This indicates that the emergency communication base station deployment task is aimed at the first Inequality regarding base station coverage limitations at a single point of communication disruption. This represents the Lth inequality for the deployment of emergency communication base stations under the restricted conditions of no-entry dangerous areas. This indicates that the deployment of emergency communication base stations is carried out under the condition that the drone inspection route does not pass through communication blind spots. Inequalities Indicates the number of communication blind spots. This indicates the first time that a drone inspection mission is conducted under restricted airspace conditions in a prohibited or dangerous area. Inequalities, Indicates the number of emergency base stations. This indicates the number of aftershock source signals. This indicates the total number of disaster areas to be inspected. This represents the total number of communication blind spots in the optimal base station deployment scheme. This represents the decision variables for drone inspection tasks. This represents the set of communication blind spots under the optimal deployment scheme. This refers to the first drone inspection mission conducted under restricted airspace conditions in a no-entry communication blind zone. Inequalities, This indicates the number of disaster-stricken points within the area where communications were disrupted after the disaster. S332: By using the piecewise smooth function (13) and By combining penalty techniques, a new emergency base station deployment optimization function with the introduction of a smooth penalty term is obtained according to equation (14). According to equation (15), the new UAV inspection optimization function after introducing a smoothing penalty term is obtained. ; （13）； （14）； （15）； in: Represents a smooth function. This represents the value that the function takes on the left side of any inequality. Represents the smoothness parameter. Represents the natural constant.

8. The post-earthquake multi-task collaborative optimization method according to claim 1, characterized in that: The method for obtaining the updated candidate solutions for each task in step S4 is as follows: S411: For each candidate solution in each task, construct the learning mode determination rule according to equation (16): （16）； in: Indicates task No. The second iteration The learning mode selected from the candidate solutions Indicates task The optimal solution continuously stagnates during the update algebra. This indicates the preset critical threshold for the optimal solution to stagnate and update. Represents logical OR, This indicates logical AND. Indicates the current iteration number. Indicates task No. The second iteration Spatial characteristic coefficients of candidate solutions This represents the critical coefficient for partitioning the neighborhood of the candidate solution; S412: When a candidate solution is determined to enter the multi-task collaborative learning mode, firstly, the adaptability is evaluated according to Equation (17) to obtain the adaptability of the source task to the target task, and the adaptability is normalized. Then, the distance proximity between the target task and the source task is calculated according to Equation (18). Then, the normalized adaptability and the distance proximity are weighted and fused to obtain a comprehensive evaluation value. Then, the probability of collaborative learning between the target task and the source task is calculated according to the comprehensive evaluation value. Based on the probability of collaborative learning between the target task and the source task, the selected source task is determined. A reference solution is selected from the selected source tasks using a hierarchical learning method. Based on the selected reference solution, multi-task collaborative learning is performed on the candidate solution according to Equation (19) to obtain the updated candidate solutions for each task. （17）； （18）； （19）； in: Indicates the first Source task to target task Adaptability This represents the objective task of introducing a smoothing penalty term. The optimization function, Indicates the decoding function. Indicates the first Individual source task The second iteration There are 10 candidate solutions. Indicates task The total dimension of its own decision variables. For the target task With the The decision space distance of each source task Indicates the number of candidate solutions. Indicate the target task No. The iteration of the ... There are 10 candidate solutions. Indicate the target task With the Proximity between source tasks Represents the natural constant. Indicates task No. The iteration of the ... There are 10 candidate solutions. Indicates task No. The iteration of the ... There are 10 candidate solutions. This represents the preset internal learning baseline coefficient. This represents the preset baseline coefficient for heterogeneous task learning. The stochastic adjustment factor represents the update magnitude of the guiding term for the optimal candidate solution within the dynamic control task. This represents a stochastic adjustment factor that dynamically regulates the update magnitude of the heterogeneous task reference solution guidance term. Indicates task No. The optimal candidate solution for the next iteration. Indicates from the source task The selected reference solution; S413: When a candidate solution is determined to enter the internal collaborative learning mode, follow-up learning is first performed to guide the candidate solution to learn and update towards the optimal candidate solution according to equation (20), and the target task after follow-up learning is obtained. The candidate solutions are then used to follow up on the target task using a consultation and learning mechanism. The candidate solutions are updated according to equation (21) to obtain the target task after consultation and learning. The candidate solutions are then reviewed and learned through positive reinforcement or negative correction of the current state, and the target task after consultation learning is determined according to equation (22). The candidate solutions are updated to obtain the updated candidate solutions for each task: （20）； （21）； （22）； in: Indicates the goals and tasks following up on the learning process. No. The iteration of the ... There are 10 candidate solutions. For the task No. Sub-iteration dependency factor This represents the preset maximum value of the dependency factor. This indicates the maximum preset number of iterations. Indicates the goals and tasks after consultation and learning. No. The iteration of the ... There are 10 candidate solutions. This indicates that teachers are learning stochastic adjustment factors. This indicates a peer-learning random adjustment factor. Indicates the current task after follow-up learning. The optimal candidate solution; Indicates from the task The average of two randomly selected top students from the set of top students. Indicates the random adjustment factor for determining the direction of the review. This represents the preset baseline coefficient for retaining the original state. This represents the preset neighborhood self-introspection fine-tuning baseline coefficient.

9. The post-earthquake multi-task collaborative optimization method according to claim 1, characterized in that: The method for obtaining the optimal solution for each task prediction in step S5 is as follows: S511: Merge the updated candidate solutions of each task to form a new population, and calculate the current optimal solution of each task according to equation (23): (23)； in: Indicate the target task No. The optimal solution in the next iteration. This represents the independent variable that corresponds to the maximum value of the function. This represents the objective task of introducing a smoothing penalty term. The optimization function, Indicate the target task No. The iteration of the ... One decoding candidate solution, Indicates the decoding function. Indicates task No. The iteration of the ... There are 10 candidate solutions. Indicates task The total dimension of its own decision variables. Indicates the number of candidate solutions; S512: Targeting the The optimization task uses a sliding window to collect the most recent consecutive... The optimal solution of the generation is constructed into a time series set of the optimal solution. The operator is combined with a dual-encoder and a single-decoder to construct a three-stage architecture of dual encoder-linear evolution-single decoder. The two encoders perform nonlinear transformations on the optimal solutions of each task, and the feature vectors output by the two encoders are concatenated to form a joint hidden state vector. The decoder takes the joint hidden state vector as input and reconstructs the original optimal solution through nonlinear transformation according to equation (24): （24）； in: Indicates task No. The optimal solution in the next iteration. express The joint hidden state vector obtained after mapping by the dual encoder, This indicates two independent nonlinear encoders. and Together they form a dual-channel parallel encoder. express Linear evolution operator, express The hidden state vector obtained after linear deduction Indicates a single-channel decoder. express The original optimal solution after reconstruction. This indicates the preset sliding window length parameter used to store historical optimal solutions; S513: Both the encoder and decoder use fully connected neural networks for network training. After the network training is completed, Based on equation (25), the optimal solution is predicted to obtain the predicted optimal solution for each task: （25）； in: Indicate the target task No. The predicted optimal solution in the next iteration.

10. A post-earthquake multi-task collaborative optimization system, characterized in that: The method is used to execute a post-earthquake multi-task collaborative optimization method as described in any one of claims 1 to 9, which includes an initial optimization function construction module for each task, an initial candidate solution evaluation module, a new optimization function acquisition module, an updated candidate solution acquisition module for each task, a predicted optimal solution acquisition module for each task, and an optimal solution output module for each task. The initial optimization function construction module for each task is used to carry out aftershock source location tasks after an earthquake, delineate the set of dangerous areas in the whole region, deploy emergency communication base stations in the area of ​​dangerous areas, deploy drone inspection tasks in the airspace above dangerous areas and communication blind spots, and construct the initial optimization function for each task. The initial candidate solution evaluation module is used to construct a unified search space for aftershock source location tasks, emergency communication base station deployment tasks, and UAV inspection tasks. In the unified search space, an independent population is constructed for each task, and each population contains multiple candidate solutions. The initial candidate solutions are evaluated to determine the initial dangerous area set, the initial communication blind zone set, and the initial inspection route set. The new optimization function acquisition module is used to uniformly and equivalently transform all spatial constraints into spatial geometric constraint inequalities for emergency communication base station deployment tasks and UAV inspection tasks based on the initial dangerous area set, initial communication blind zone set, and initial inspection route set. It also constructs penalty terms for the spatial geometric constraint inequalities of emergency communication base station deployment tasks and UAV inspection tasks, and substitutes the penalty terms into the initial optimization function of the corresponding task to obtain the new optimization function of the corresponding task. The updated candidate solution acquisition module for each task is used to construct a learning mode determination rule for each candidate solution, and obtain the updated candidate solution for each task based on the learning mode determination rule and the new optimization function of the corresponding task. The module for obtaining the predicted optimal solution for each task is used to merge the updated candidate solutions of each task to form a new population, and calculate the current optimal solution for each task in the new population. When the number of iterations reaches a preset number, the module predicts the subsequent optimal solution based on the historical optimal solution sequence within the sliding window of each task, and obtains the predicted optimal solution for each task. The optimal solution output module for each task synchronously updates the dangerous area set, communication blind spot set, and inspection route set based on the current optimal solution of each task until the iteration is completed and the optimal solution of each task is output.

Images