Method, device and system for evaluating and allocating resources for a group of regionally damaged buildings

Abstract

This invention discloses a method, apparatus, and system for assessing and allocating resources for earthquake-damaged building clusters in a region, belonging to the field of earthquake disaster emergency management and building safety assessment technology. Addressing the problems of long assessment cycles, single standards, lack of global optimization in resource allocation, and strong network dependence in existing technologies, this invention constructs a regional feature matrix by fusing multi-source data and generates an adaptive assessment parameter set to dynamically select the optimal building earthquake damage assessment model. It employs an improved multi-objective evolutionary algorithm to generate local Pareto-optimal resource scheduling schemes at edge nodes and relies on a cloud-edge-device collaborative architecture to achieve global fusion and real-time dynamic updates of the schemes. This method can be used for rapid damage assessment and intelligent optimization scheduling of rescue resources for building clusters in earthquake-stricken areas.

Description

Technical Field

[0001] This invention relates to the field of earthquake disaster emergency management and building structural safety assessment technology. More specifically, this invention relates to a method, apparatus, and system for assessing and allocating resources for regional earthquake-damaged building complexes. Background Technology

[0002] Following an earthquake, buildings in the affected area often suffer structural damage to varying degrees. Failure to assess the extent of damage promptly and accurately can not only delay emergency response but also lead to inefficient resource allocation, thereby increasing casualties and economic losses. Current earthquake damage assessment methods primarily focus on individual buildings, relying on manual on-site inspections or data collection from local sensors. These methods are time-consuming and have limited coverage, making it difficult to meet the needs for rapid post-disaster assessment across large areas and regions.

[0003] In terms of resource allocation, most emergency management departments currently rely on pre-established distributed emergency plans or real-time manual dispatch. This approach usually fails to comprehensively consider multiple factors such as the importance of buildings, the degree of damage, the safety status of personnel, geographical location, accessibility, and real-time resource occupancy. As a result, the allocation of rescue and reinforcement resources lacks global optimality, and there is a tendency for some areas to have redundant resources while other disaster-stricken areas suffer from resource shortages.

[0004] Furthermore, existing post-earthquake emergency assessment and dispatch systems are often deployed on a single centralized server or local data center, relying on a stable network environment. Once communication is interrupted or bandwidth is limited, continuous data processing and decision support cannot be achieved within the disaster area. For areas with complex geological conditions and diverse building types, existing methods rarely make parametric adjustments to local building structural characteristics and seismic fortification requirements, resulting in a uniform assessment standard that is not adapted to the actual needs of different regions.

[0005] Therefore, there is an urgent need for a new system that can rapidly and uniformly assess earthquake-damaged building complexes at a regional scale and intelligently allocate rescue and reinforcement resources by combining multi-objective optimization methods. This system should have the ability to operate autonomously in network-limited environments and dynamically adjust assessment standards according to regional characteristics, thereby improving the efficiency and scientific nature of post-earthquake emergency response. Summary of the Invention

[0006] One object of the present invention is to solve at least the above-mentioned problems and to provide at least the advantages that will be described later.

[0007] Another objective of this invention is to provide a method, apparatus, and system for assessing and allocating resources for earthquake-damaged building clusters in a regional setting. This system addresses issues such as long assessment cycles, insufficient coverage, lack of regional adaptability of assessment standards, lack of global optimization in emergency and reinforcement resource allocation, and insufficient system stability under communication-restricted environments. The system combines speed, flexibility, and robustness, enabling it to support scientific decision-making and efficient response in post-disaster situations under varying geological conditions, building characteristics, and network environments.

[0008] To achieve these objectives and other advantages according to the present invention, a method for assessing and allocating resources for regional earthquake-damaged building complexes is provided, comprising the following steps: Acquire real-time earthquake monitoring data, aerial and satellite remote sensing image data, on-site inspection data of terminal nodes, and historical earthquake damage and regional geological database data; A regional feature matrix is ​​constructed based on historical earthquake damage and regional geological database data. An adaptive evaluation parameter set is generated based on the regional feature matrix under a preset mapping function. The optimal building seismic damage assessment model is selected based on the adaptive evaluation parameter set. Real-time earthquake monitoring data, aerial and satellite remote sensing image data, and on-site inspection data of terminal nodes are integrated according to building identification, and then input into the building seismic damage assessment model to calculate the predicted value of post-earthquake damage and the corresponding confidence level. Based on the predicted value of post-earthquake damage and the adaptive assessment parameter set, the building is classified into four safety levels, including four levels: safe and usable, restricted use, requiring reinforcement, and dangerous. A multi-dimensional optimization objective function is constructed, which includes building importance, personnel safety, traffic accessibility, and resource utilization efficiency. The weights of each optimization objective are automatically adjusted according to the post-disaster stage. An improved multi-objective evolutionary algorithm combined with constrained programming method is used to generate a local Pareto optimal resource scheduling scheme.

[0009] Preferably, it also includes: The local Pareto optimal resource scheduling scheme is uploaded to the cloud, so that the cloud can generate a global optimal resource scheduling scheme based on each local Pareto optimal resource scheduling scheme. Obtain the globally optimal resource scheduling plan from the cloud and push it to the terminal nodes.

[0010] Preferably, the region feature matrix includes the following feature data: Geological parameters: soil type, fault distribution, seismic intensity zoning, and seismic ground acceleration index; Building parameters: structural type, number of floors, age, seismic fortification level; Functional parameters: population density distribution, location of key facilities, and accessibility index of transportation network.

[0011] Preferably, the method for generating an adaptive evaluation parameter set based on the regional feature matrix under a preset mapping function, and selecting the optimal building seismic damage assessment model based on the adaptive evaluation parameter set, is as follows: The parameters in the building seismic damage assessment model are corrected by calling the preset mapping function P′=f(Tr,Bs,S1) based on the regional feature matrix. The parameters in the building seismic damage assessment model include the structural yield threshold, residual bearing capacity coefficient and failure probability curve parameters. Dynamically load and match the parameters in the building seismic damage assessment model.

[0012] Preferably, the method of generating a local Pareto optimal scheduling scheme by using an improved multi-objective evolutionary algorithm combined with a constraint programming method is as follows: the improved non-dominated sorting genetic algorithm NSGA-II is combined with a constraint programming method to dynamically adjust the Pareto optimal solution search path according to the weight of each objective, and parallel computation is performed using distributed edge nodes.

[0013] Preferably, when the resource reserve reported by the terminal node is lower than the set percentage threshold, or the accessibility index of the transportation network drops by more than the set percentage, local incremental optimization is performed, the resource scheduling plan for the earthquake-damaged area is modified, and updated to the terminal node.

[0014] Preferably, the set ratio thresholds include a resource surplus threshold of 20% and a traffic network accessibility index decline threshold of 50%.

[0015] Preferably, when communication bandwidth is insufficient, a local data caching and delay synchronization mechanism is adopted to ensure that the edge node has at least N hours of autonomous computing and scheduling execution capability in the event of network outage.

[0016] Preferably, N is 24 hours, which means ensuring that the edge node has the ability to operate autonomously for no less than 24 hours in the event of a network outage.

[0017] Preferably, it also includes: Data from the entire process of post-disaster assessment and resource allocation is uploaded to the cloud, enabling the cloud to train a building seismic damage assessment model for the next disaster using machine learning methods.

[0018] The present invention also provides a device for assessing and allocating resources for regional earthquake-damaged building complexes, comprising: The multi-source data acquisition module is used to acquire real-time earthquake monitoring data, aerial and satellite remote sensing image data, on-site inspection data of terminal nodes, and historical earthquake damage and regional geological database data. The regional feature modeling and parameter adaptation module is used to construct a regional feature matrix based on historical earthquake damage and regional geological database data, generate an adaptive evaluation parameter set based on the regional feature matrix under a preset mapping function, and load the optimal building seismic damage assessment model based on the adaptive evaluation parameter set. The building cluster seismic damage assessment module is used to integrate real-time earthquake monitoring data, aerial and satellite remote sensing image data, and on-site inspection data of terminal nodes according to building identification, and then input them into the building seismic damage assessment model to calculate the predicted value of post-earthquake damage and the corresponding confidence level of the building. Based on the predicted value of post-earthquake damage and the adaptive assessment parameter set, the building is classified into four safety levels, including four levels: safe and usable, restricted use, requiring reinforcement, and dangerous. The multi-objective optimization resource scheduling module is used to construct a multi-dimensional optimization objective function that includes building importance, personnel safety, traffic accessibility, and resource utilization efficiency. It automatically adjusts the weights of each optimization objective according to the post-disaster stage and uses an improved multi-objective evolutionary algorithm combined with constrained programming to generate a local Pareto optimal scheduling scheme.

[0019] The present invention also provides a regional earthquake-damaged building complex assessment and resource allocation system, including a cloud, an edge node, and a terminal node, wherein the edge node is used to execute the above-described regional earthquake-damaged building complex assessment and resource allocation method; The terminal node is used to collect on-site inspection data of the earthquake-damaged area and send it to the edge node, as well as to obtain and display the resource scheduling scheme of the earthquake-damaged area pushed by the edge node; The cloud is used to acquire local Pareto optimal resource scheduling schemes generated by each edge node, and then generate a global optimal resource scheduling scheme based on each local Pareto optimal resource scheduling scheme and feed it back to each edge node. It also acquires and stores real-time earthquake monitoring data, aerial and satellite remote sensing image data, on-site inspection data of terminal nodes, and historical earthquake damage and regional geological database data to train and update parameters of the building seismic damage assessment model.

[0020] The present invention includes at least the following beneficial effects: (1) Rapid global assessment: Based on multi-source data fusion and parallel computing, a batch assessment of regional building clusters can be completed in a short time, and the coverage can be extended to different cities or regions; (2) Adaptive standards: The assessment parameters are automatically adapted to different geological and architectural features, improving the accuracy and regional adaptability of the assessment results; (3) Global optimal scheduling: Multi-objective optimization combined with a dynamic weight adjustment mechanism ensures that rescue and reinforcement resources are allocated reasonably and in a timely manner throughout the entire region; (4) Anti-network interruption capability: The cloud-edge-device architecture ensures that the decision-making capability can still be sustained under network-limited conditions; (5) Dynamic iteration capability: The real-time feedback mechanism ensures that the scheduling plan evolves synchronously with the disaster situation, avoiding rescue delays or resource waste caused by information lag.

[0021] Other advantages, objectives and features of the present invention will become apparent in part from the following description, and in part from those skilled in the art through study and practice of the invention. Attached Figure Description

[0022] Figure 1 This is a flowchart of the regional earthquake-damaged building complex assessment and resource allocation method described in this invention; Figure 2 This is a flowchart of the regional feature modeling process described in this invention; Figure 3 This is a flowchart of the building seismic damage assessment and grading process described in this invention; Figure 4 This is a flowchart of the multi-objective optimization process described in this invention; Figure 5 This is an architecture diagram of the regional earthquake-damaged building complex assessment and resource allocation system described in this invention; Figure 6 This is a partial pseudocode diagram illustrating the building seismic damage assessment and grading process described in this invention; Figure 7 This is a partial pseudocode diagram of the multi-objective optimization process described in this invention. Detailed Implementation

[0023] The present invention will now be described in further detail with reference to the accompanying drawings, so that those skilled in the art can implement it based on the description.

[0024] like Figure 1 As shown, this invention provides a method for assessing and allocating resources for earthquake-damaged building complexes in a region, comprising the following steps: S1. Acquire real-time earthquake monitoring data, aerial and satellite remote sensing image data, on-site inspection data of terminal nodes, and historical earthquake damage and regional geological database data. Traditional earthquake damage assessment data collection methods often rely on a single data source or manual collection, resulting in limited coverage, untimely updates, and strong subjectivity. This invention constructs a collaborative acquisition system for multi-source heterogeneous data, achieving comprehensive data coverage from macro to micro and from historical to real-time, providing a solid data foundation for accurate assessment.

[0025] In practice, real-time earthquake monitoring data is acquired through a network of smart sensors deployed in key areas. These sensors continuously collect ground acceleration, velocity, and displacement time histories, as well as dynamic response data of building structures. The sensors have built-in computing modules that perform preliminary filtering and feature extraction on the raw signals, and then transmit the compressed feature data to edge nodes in real time via IoT communication protocols. This design ensures data timeliness while significantly reducing network transmission load.

[0026] The acquisition of aerial and satellite remote sensing imagery data forms a crucial part of macro-level situational awareness. After an earthquake, pre-programmed remote sensing data acquisition tasks are automatically triggered, deploying drones to conduct oblique photography of key areas, while simultaneously receiving synthetic aperture radar and optical imagery data from multiple satellites. After geometric correction and radiometric calibration, this remote sensing data is automatically used through deep learning algorithms to identify large-scale disaster information such as building collapses, road damage, and surface fissures, providing a direct basis for overall regional damage assessment.

[0027] The introduction of on-site inspection data from terminal nodes fills the gaps in microscopic detail information. The mobile smart terminals equipped by rescue personnel have built-in professional data acquisition applications that can systematically record structural damage details such as crack width, component misalignment, and material spalling. They also integrate a positioning module to automatically associate with building markers. The terminal application supports multiple input methods, including voice input, image annotation, and sketching, and has offline data caching capabilities to ensure continued operation even during communication interruptions. This frontline inspection data, uploaded through a secure channel, provides valuable on-site verification information for the evaluation model.

[0028] Access to historical earthquake damage and regional geological databases provides crucial background knowledge and experience for the assessment work. Standardized interfaces allow access to distributed databases, enabling the acquisition of long-term accumulated professional data such as damage distribution, geological structural characteristics, and soil response spectra from historical earthquakes in the region. When this historical data is integrated with real-time monitoring data, it effectively improves the model's understanding of specific regional earthquake response patterns, making the assessment results more consistent with local conditions.

[0029] These multi-source heterogeneous data are fused under a unified spatiotemporal benchmark. Through timestamp alignment, coordinate system unification, and data quality verification, inconsistencies between data from different sources are eliminated. The resulting multi-dimensional dataset not only provides comprehensive information support for subsequent earthquake damage assessment, but its multi-source complementary characteristics also significantly improve fault tolerance in the event of partial data loss, constituting the core innovation of this invention at the data level.

[0030] S2. Construct a regional feature matrix based on historical earthquake damage and regional geological database data. Generate an adaptive evaluation parameter set based on the regional feature matrix under a preset mapping function. Select the optimal building seismic damage assessment model based on the adaptive evaluation parameter set. Traditional building seismic damage assessment methods typically employ uniform assessment standards and parameter thresholds, neglecting significant differences in geological conditions, building types, and seismic fortification requirements across different regions. This leads to discrepancies between assessment results and actual conditions. This invention constructs a regional feature matrix and generates an adaptive assessment parameter set based on a preset mapping function, achieving precise matching between assessment standards and regional characteristics, effectively solving the adaptability problem of assessment models in different regions.

[0031] like Figure 2 As shown, in the specific implementation process, multi-dimensional feature data is first extracted from historical earthquake damage databases and regional geological databases. After standardization and preprocessing, this data is organized into a structured regional feature matrix, which contains three main categories of parameters: geological parameters such as soil type, fault distribution, seismic intensity zoning, and peak ground acceleration; architectural parameters such as structural form, number of floors, construction year, and seismic fortification level; and functional parameters such as population density distribution, location of key facilities, and transportation network accessibility index. This comprehensive feature matrix fully characterizes the region's seismic hazard characteristics, building structural vulnerability characteristics, and socio-economic importance.

[0032] Based on the constructed regional feature matrix, a preset mapping function is invoked to generate an adaptive evaluation parameter set. In the mapping function P′=f(Tr, Bs, S1), Tr represents geological feature vectors (including soil type, fault distribution, etc.), Bs represents building feature vectors (including structural form, number of floors, etc.), and S1 represents functional feature vectors (including population density distribution, transportation network accessibility index, etc.). Function f is a multivariate nonlinear function obtained through machine learning training on a large number of historical earthquake damage cases. It can accurately capture the complex relationship between regional features and evaluation parameters. Specifically, a neural network model can be used. Its training process includes: using historical earthquake data as input, using actual earthquake damage evaluation parameters as labels, and optimizing network weights through a backpropagation algorithm to minimize the error between the output parameters and the actual parameters. When a specific regional feature matrix is ​​input, the mapping function outputs a set of regionally adjusted evaluation parameters, including the structural yield threshold, residual bearing capacity coefficient, and the shape parameters of the failure probability curve. The essential function of these parameters is to recalibrate the evaluation scale according to local conditions. For example, appropriately lowering the structural yield threshold in soft soil foundation areas and adjusting the slope of the failure probability curve in high-intensity seismic fortification areas.

[0033] After obtaining the adaptive evaluation parameter set, the model selection phase begins. A model library containing various evaluation models is maintained in the cloud, such as a damage prediction model based on neural networks, a fast evaluation model based on logistic regression, and an expert system model based on fuzzy inference. Each model has its applicable scenarios and advantages. By calculating the matching degree between the adaptive evaluation parameter set and the features of each model, the evaluation model most suitable for the current regional characteristics is selected. For example, for old urban areas with dense brick-concrete structures, a model with higher sensitivity to material aging might be chosen; for new areas dominated by steel structures, an evaluation model that focuses more on node connection performance might be selected. This dynamic model selection mechanism ensures the highest degree of fit between the evaluation tool and the evaluation object, improving the reliability and practicality of the evaluation results from a methodological perspective.

[0034] The entire process, from feature matrix construction to model selection, forms a complete technical chain, realizing the transformation from static assessment to dynamic assessment. This adaptive assessment framework based on regional characteristics not only significantly improves the accuracy of seismic damage assessment but also provides a unified and flexible technical solution for disaster assessment in different geographical and architectural environments, demonstrating the substantial innovation of this invention in the field of seismic damage assessment.

[0035] S3. Integrate real-time earthquake monitoring data, aerial and satellite remote sensing image data, and on-site inspection data of terminal nodes according to building identification, and then input them into the building seismic damage assessment model to calculate the predicted value of post-earthquake damage and the corresponding confidence level. Based on the predicted value of post-earthquake damage and the adaptive assessment parameter set, classify the building into four safety levels, including four levels: safe and usable, restricted use, requiring reinforcement, and dangerous. Traditional post-earthquake building safety assessments often rely on expert on-site inspections or judgments based on a single type of data, resulting in significant drawbacks such as low efficiency, strong subjectivity, and incomplete coverage. This invention, through multi-source data fusion and an intelligent assessment model, achieves rapid, objective, and precise safety level classification of regional building clusters, completely transforming the traditional assessment approach.

[0036] like Figure 3As shown, in the specific implementation process, all data is first intelligently associated and integrated through a unified building unique identifier. This identifier system is synchronized with the basic database of the urban planning and management department to ensure the uniqueness of each building unit's identity. When vibration data from real-time monitoring sensors is received, it is automatically associated with the corresponding building identifier through the sensor's installation location information; remote sensing images, through georegistration and building outline extraction algorithms, bind identified damage features such as roof collapse and wall tilting to specific buildings; when on-site inspection personnel report data via mobile terminals, the application automatically obtains the current location and matches it with the nearest building identifier. This multi-source data fusion method based on spatial location and identity identification constructs a complete feature vector for each building, including dynamic response, appearance damage, and detailed observations.

[0037] After data integration, the fused feature vectors are input into the loaded building seismic damage assessment model. This model employs a deep neural network architecture, including convolutional layers specifically for processing time-series signals, a vision module for analyzing image features, and fully connected layers for integrating structured data. The model calculates a damage prediction value for each building through forward propagation; this value is a continuous numerical value between 0 and 1, quantitatively representing the degree of damage. Simultaneously, the model also outputs a confidence score, calculated based on the variance of the model's predictions. Specifically, this score is obtained through multiple inferences (e.g., using Dropout techniques) to obtain the distribution of predicted values; the smaller the variance, the higher the confidence score. Alternatively, it can be calculated based on a weighted average of input data quality indicators (e.g., data integrity, sensor accuracy), reflecting the reliability of the current prediction results.

[0038] After obtaining the damage prediction values, the system does not simply apply fixed thresholds for classification, but rather makes intelligent decisions based on the previously generated adaptive assessment parameter set. These region-specific parameters include adjustment factors that consider local building characteristics, seismic fortification standards, and functional importance. For example, for important public buildings such as schools and hospitals, a more conservative threshold is used, and even with the same damage prediction values, they may be classified into a higher safety level; while for newly built areas with higher seismic fortification levels, the threshold range for restricted use is appropriately relaxed. Through this dynamic threshold adjustment mechanism, continuous damage prediction values ​​are mapped to four discrete safety levels: safe and usable, restricted use, requiring reinforcement, and dangerous (partial pseudocode shown). Figure 6 (As shown).

[0039] The entire assessment process is executed in parallel within a distributed computing framework, with the system simultaneously and automatically assessing and classifying tens of thousands of buildings. The assessment results for each building include damage level, confidence score, and key evidence sources, presented visually on the command center's digital twin platform. This assessment method, based on multi-source data fusion and adaptive parameter adjustment, not only significantly improves assessment efficiency but also, through the introduction of a confidence mechanism and regional differentiation processing, significantly enhances the scientific rigor and practicality of the assessment results, providing a reliable basis for subsequent precise resource allocation.

[0040] S4. Construct a multi-dimensional optimization objective function that includes building importance, personnel safety, traffic accessibility, and resource utilization efficiency. Automatically adjust the weights of each optimization objective according to the post-disaster stage. Use an improved multi-objective evolutionary algorithm combined with constrained programming to generate a local Pareto optimal resource scheduling scheme.

[0041] Traditional emergency resource allocation methods often rely on human experience or single-objective optimization, making it difficult to achieve a balance among multiple conflicting objectives such as building importance, personnel safety, traffic conditions, and resource efficiency. This invention achieves a technological breakthrough in scientifically generating resource allocation schemes in complex post-disaster environments by constructing a multi-dimensional optimization objective function and introducing a dynamic weight adjustment mechanism, combined with an improved multi-objective evolutionary algorithm and constrained programming method.

[0042] like Figure 4 As shown, in the specific implementation process, an optimization objective function containing four core dimensions is first constructed. The objectives include maximizing building importance F1, maximizing personnel safety F2, optimizing traffic accessibility F3, and maximizing resource utilization efficiency F4. These objective functions together constitute a multi-dimensional optimization space that comprehensively reflects the needs of post-disaster relief.

[0043] Simultaneously, an automatic weight adjustment mechanism is established to dynamically optimize the allocation of importance for each objective based on the characteristics of different post-disaster stages. During the golden rescue period after the earthquake, the weight of the objective related to personnel safety is automatically increased to the highest priority, ensuring that rescue forces are prioritized for densely populated and high-risk buildings. As the post-disaster recovery phase progresses, the weight of objectives related to resource utilization efficiency and building importance is gradually increased, guiding resources towards the repair of critical infrastructure and the restoration of livelihoods. This dynamic weighting strategy, based on time stages and the evolution of the disaster situation, ensures that the generated dispatch plan remains consistent with the current disaster relief strategic priorities.

[0044] At the optimization algorithm level, the specific improvements of the improved non-dominated sorting genetic algorithm NSGA-II include: before non-dominated sorting, the objective function is scalarized based on dynamic weights, transforming the multi-objective problem into a single-objective problem and guiding the search direction; in the crowding calculation, a weighted objective spatial distance is introduced, prioritizing individuals with greater improvement potential in high-weight objective directions. During the evolutionary process, the algorithm not only sorts solutions according to their non-dominated levels but also guides the population towards the most important Pareto front region through dynamic weight information. For hard constraints such as total resource limits, task time dependencies, and path connectivity, the system strictly satisfies them through constraint programming methods to ensure the feasibility of the generated solution; while for soft constraints such as ideal resource utilization, they are transformed into additional optimization objectives for computation. This combination of rigidity and flexibility in constraint handling ensures both the feasibility of the solution's execution and the algorithm's search flexibility.

[0045] More specifically, in the complex multi-objective optimization problem of post-disaster emergency resource allocation, the improved non-dominated sorting genetic algorithm NSGA-II, based on the deep fusion constraint programming mechanism, achieves parallel solution on a distributed edge computing architecture. The innovation of this computational process lies in breaking the unbiasedness of traditional multi-objective optimization algorithms in the search direction. By introducing a dynamic weight guidance mechanism, the algorithm can intelligently focus on the most practically valuable region in the Pareto front for deep search based on the phased preferences of emergency decision-making.

[0046] The specific calculation process begins in the population initialization phase. Each edge node independently generates an initial population, where each individual represents a potential resource scheduling scheme, using integer or real-number encoding to represent the rescue team's destination, resource allocation amount, and action sequence. When generating the initial solution, the system invokes constrained programming methods to preliminarily screen the feasibility of the solution. For example, it checks whether resource allocation exceeds inventory using the resource conservation equation, or uses graph theory algorithms to verify the connectivity of scheduling paths, ensuring that the initial population has a high feasibility foundation.

[0047] After entering the evolutionary cycle, the algorithm performs crossover and mutation operations in each generation to generate offspring. Unlike the traditional NSGA-II algorithm, this improved algorithm first transforms the objective function space based on a dynamic weight vector distributed from the cloud before non-dominated sorting. This weight vector is dynamically set by the emergency command system according to the stage of the disaster, for example, it is set to [personnel safety: 0.6, building importance: 0.3, traffic efficiency: 0.1] in the early post-earthquake period and adjusted to [resource efficiency: 0.5, building importance: 0.3, personnel safety: 0.2] during the recovery period. The algorithm weights and fuses the original objective function values ​​with this weight vector to generate a set of preferred scalar fitness values, and then performs fast non-dominated sorting. This weight-based pre-sorting process allows individuals in the population that perform well in the more important objective directions at the current stage to obtain higher priority, thereby effectively guiding the population to evolve towards the Pareto front region of concern to decision-makers.

[0048] In the crowding calculation phase, the algorithm also introduces a weight-aware mechanism. Traditional crowding assessment only considers the geometric distribution density of individuals in the target space, while the improved algorithm evaluates the crowding level of individuals by calculating weighted target space distances. This improvement allows individuals with greater improvement potential along high-weight target directions to receive higher selection priority within the same non-dominated hierarchy, further strengthening the guidance of the search process. Through this design that deeply integrates decision preferences into the selection mechanism, the algorithm significantly improves the search efficiency for high-value solutions while maintaining its multi-objective optimization nature.

[0049] For handling constraints, the algorithm employs a hierarchical optimization strategy. Hard constraints, such as total resource limits and task time dependencies, are strictly protected through consistency checks in constraint programming; any entity violating a hard constraint is either directly eliminated or transformed into a feasible solution through a repair operator. For soft constraints, such as ideal travel time and optimal resource utilization, the degree of violation is transformed into an additional optimization objective, participating in non-dominated sorting along with other primary objectives. This balanced approach to constraint handling ensures the feasibility of the final solution in actual execution while retaining necessary flexibility for the algorithm's exploration near constraint boundaries.

[0050] The entire optimization process unfolds in parallel across distributed edge nodes. Each edge node independently runs the improved NSGA-II algorithm based on local data within its jurisdiction, generating a local Pareto optimal solution set for its region. These local solution sets are uploaded to the cloud via asynchronous communication. The cloud scheduling center then uses a decomposition-based multi-objective optimization framework for global coordination. Specifically, it employs a decomposition-based multi-objective evolutionary algorithm (MOEA / D) to decompose the global problem into multiple sub-problems and coordinate the solutions from each edge node. A secondary planning process using a resource contention model resolves resource conflicts. Finally, the local Pareto fronts of each edge node are effectively fused to generate a unified scheduling scheme that considers global optimality (partial pseudocode is shown below). Figure 7 (As shown). This distributed parallel architecture not only significantly improves the solution speed of large-scale optimization problems, but also ensures that regional-level emergency scheduling can continue to run autonomously based on local computing when communication between a single edge node and the cloud is interrupted, greatly enhancing the robustness and practicality of the entire system in harsh post-disaster environments.

[0051] Furthermore, a local Pareto optimal scheduling scheme is generated, and each edge node uploads the local Pareto optimal resource scheduling scheme to the cloud, so that the cloud can generate a global optimal resource scheduling scheme based on each local Pareto optimal resource scheduling scheme. The edge nodes then obtain the globally optimal resource scheduling plan from the cloud and push it to the terminal nodes.

[0052] Traditional emergency resource dispatch systems typically employ centralized optimization or fully distributed decision-making. The former struggles to function during communication disruptions, while the latter lacks global coordination, making it prone to resource allocation conflicts. This invention achieves an innovative balance between ensuring regional autonomy and maintaining global optimality by constructing a collaborative optimization mechanism of "local generation - cloud fusion - edge execution."

[0053] In the implementation process, each edge node first runs an improved multi-objective optimization algorithm in parallel based on real-time data from its jurisdiction. Each edge node independently generates a set of locally Pareto optimal scheduling schemes for its region. These schemes constitute the solution front, which includes various trade-offs among objectives such as building importance, personnel safety, traffic accessibility, and resource efficiency. To ensure transmission efficiency, the edge nodes intelligently compress these solution sets, retaining only the hypervolume contribution values ​​of the front feature points and key solutions, while also marking the regional applicability conditions and resource requirement constraints of each scheme.

[0054] After completing local optimization, each edge node uploads its local Pareto optimal solution set to the cloud fusion center via asynchronous communication channels. Upon receiving the solution sets from multiple edge nodes, the cloud initiates a global coordinated optimization algorithm. This algorithm does not simply concatenate the local solutions; instead, it constructs a global optimization model, using the local solutions from each edge node as the initial population, and performs secondary evolutionary calculations with the optimization objectives of cross-regional resource balance and overall rescue efficiency. During the fusion process, the cloud pays special attention to resource scheduling conflicts at the boundaries of adjacent edge regions, introducing boundary coordination constraints to ensure smooth coordination of rescue forces between regions. This global fusion mechanism can identify scheduling schemes that are not optimal locally but are more valuable at the global level.

[0055] After generating the globally optimal resource scheduling plan in the cloud, it is pushed back to each edge node via a publish-subscribe mechanism. Upon receiving the global plan, the edge nodes first perform a regional adaptation check, extracting the scheduling instructions relevant to their region from the global plan and making fine-tuning adjustments based on the latest local state data. This fine-tuning involves limited adjustments for specific local conditions, such as route replanning due to local road disruptions caused by sudden aftershocks, without disrupting the global optimization framework.

[0056] Ultimately, the edge nodes decompose the adapted scheduling scheme into specific executable tasks and push them to the terminal nodes via a dedicated communication protocol. The emergency application on the terminal devices presents the scheduling instructions in an intuitive and visual format, including the location of the rescue target, the optimal route, task priority, and a list of required resources. The on-site rescue team provides feedback on task execution progress through the terminals, forming a complete closed loop from decision-making to execution. This layered and collaborative optimized architecture fully leverages the real-time advantages of edge computing while maintaining a global perspective from cloud-based coordination, ensuring the scientific, efficient, and robust nature of emergency resource scheduling in complex post-disaster environments.

[0057] Furthermore, when the resource reserve reported by the terminal node is lower than the set percentage threshold, or the accessibility index of the transportation network drops by more than the set percentage, local incremental optimization is performed, the resource scheduling plan for the earthquake-damaged area is modified, and updated to the terminal node.

[0058] Traditional emergency resource dispatching systems generally lack dynamic adjustment capabilities, making it difficult to adapt to rapidly changing on-site conditions after a disaster once a plan is finalized. This invention, by establishing a triggering mechanism based on real-time feedback and a local incremental optimization algorithm, achieves synchronous adaptation of resource dispatching plans to the evolution of the disaster situation, effectively solving the technical problems of slow response and high adjustment costs in traditional systems.

[0059] By continuously monitoring key operational indicators through terminal nodes deployed at the rescue site, including real-time inventory levels of various rescue supplies, available personnel for engineering teams, and remaining medical resources, as well as road network capacity index calculated based on traffic status information transmitted from mobile terminals, this data is transmitted to edge nodes via a lightweight communication protocol and automatically compared with preset threshold conditions. When it is detected that the resource balance in any area is lower than a set percentage threshold, or that the road capacity index has decreased by more than a set percentage from the benchmark value (e.g., resources are below 20% or the road capacity index has decreased by more than 50%), a local incremental optimization process is immediately initiated. This intelligent triggering mechanism based on multi-source real-time data ensures the system's keen perception and rapid response to changes in the disaster situation.

[0060] After entering the local incremental optimization phase, an efficient optimization strategy was adopted, the core of which lies in accurately defining the optimization scope. The algorithm first identifies the subset of buildings directly affected by resource shortages or traffic disruptions, along with their associated scheduling tasks, through impact propagation analysis, strictly limiting optimization variables to these local objects. While maintaining the stability of the overall global solution structure, the optimization engine recalculates rescue routes, adjusts resource allocation order, and optimizes task scheduling for the affected areas. For example, when a main road is interrupted due to a secondary disaster, the system immediately replans alternative routes for the affected rescue tasks and allocates resources from nearby backup warehouses, rather than rescheduling all tasks in the entire area.

[0061] This local incremental optimization strategy employs a lightweight and fast optimization algorithm. By limiting the size of the optimization problem, it reduces the computational complexity from a high-order level for global problems to a linear level for local problems. During iteration, the algorithm fully utilizes the basic information of the original solution, performing targeted searches only in the affected variable space, enabling solution adjustments to be completed within tens of seconds even with the limited computing power of edge nodes. The optimization process strictly adheres to hard constraints such as resource conservation and time windows, and ensures the balance of the new solution across multiple objectives through fast non-dominated sorting.

[0062] After optimization, the modified scheduling instructions are precisely pushed to relevant terminal devices through a differential update mechanism. Only changed task information is retransmitted, and this differential transmission method greatly saves communication bandwidth. On-site rescue teams can receive updated action routes, resource allocation instructions, and schedules on their terminal devices, ensuring immediate execution of the latest plan. The entire process, from status monitoring, trigger judgment, incremental optimization to instruction update, forms a complete adaptive control closed loop, enabling the resource scheduling system to have self-correction and continuous optimization capabilities in dynamic disaster environments, significantly improving the agility of emergency response and the rationality of resource utilization.

[0063] Furthermore, when communication bandwidth is insufficient, a local data caching and delay synchronization mechanism is adopted to ensure that edge nodes have at least N hours of autonomous computing and scheduling execution capability in the event of network outage.

[0064] Traditional emergency management systems heavily rely on continuous and stable network connections; once communication is interrupted, decision-making and dispatch functions become paralyzed. This invention, through the design of an intelligent local data caching and delayed synchronization mechanism, enables edge nodes to maintain autonomous operation for extended periods even in a completely disconnected network environment. This feature has crucial practical value in scenarios where post-disaster communication infrastructure is damaged.

[0065] A multi-level caching system is deployed at edge nodes to store critical data and computing resources. When network connectivity is normal, edge nodes synchronize the latest regional building inventory, resource distribution maps, historical optimization plans, and other basic data from the cloud. Simultaneously, they pre-load necessary evaluation models and algorithm components based on predictive models. This data undergoes intelligent compression and selective storage, prioritizing high-value information such as detailed structural parameters of key buildings, real-time locations of critical rescue resources, and traffic capacity of frequently used routes. The cache management system employs a dynamic space allocation strategy, automatically adjusting storage priorities based on data type, usage frequency, and update timeliness to ensure that limited storage space accommodates the most operationally valuable information.

[0066] When communication quality degrades or is completely interrupted, edge nodes automatically switch to offline mode. In this mode, they continue to operate using locally cached data and models, still performing core functions such as building seismic damage assessment and resource scheduling optimization. To extend autonomous operation time, a resource-saving mechanism is activated, including reducing the computational precision of non-critical tasks, pausing data backup processes, and optimizing memory usage efficiency. Simultaneously, edge nodes initiate periodic self-checks, monitoring remaining storage space and computing resources. When resources fall below a safe threshold, they automatically clean up low-priority cached data to ensure the continuous operation of core functions.

[0067] After network recovery, a delayed synchronization phase begins. Edge nodes first package and upload all evaluation results, scheduling instructions, status change records, and other data generated during the offline period to the cloud. This data is precisely timestamped and version-marked, facilitating time-series consistency verification by the cloud. During synchronization, an intelligent conflict resolution algorithm handles potential data inconsistencies. For example, when the same building receives different evaluation results on different edge nodes, the optimal value is automatically selected based on the reliability of the data source, the order of collection time, and confidence scores, triggering a manual review process if necessary.

[0068] The entire caching and synchronization mechanism is designed with full consideration of the actual needs of post-disaster emergency response. Through a series of technological innovations such as data preloading, dynamic resource management, and intelligent conflict resolution, it ensures the system's continuous service capability under extreme network conditions. This resilient architecture, which can fully utilize network resources while operating independently, significantly improves the reliability and practicality of the emergency management system, providing solid technical support for life-saving and disaster response.

[0069] Furthermore, it also includes uploading the entire process data of post-disaster assessment and resource allocation to the cloud, so that the cloud can use machine learning methods to train a building seismic damage assessment model for the next disaster.

[0070] Traditional earthquake damage assessment models often rely on historical cases and laboratory data, making them ill-suited to the complex and ever-changing environmental conditions of real-world disasters. Furthermore, their long update cycles prevent them from promptly incorporating the latest disaster experience. This invention, by establishing a closed-loop learning mechanism based on post-disaster data, enables the continuous evolution and self-optimization of the assessment model, significantly improving the accuracy and efficiency of future disaster responses.

[0071] In practice, various operational data are automatically collected throughout the entire post-disaster emergency response cycle to form a complete training sample set. This data includes not only multi-source input information such as seismic characteristics, building structural responses, and remote sensing image change detection results, but also complete records of the damage prediction values, confidence scores, and final safety levels output by the assessment model. More importantly, it simultaneously collects records of actual post-disaster rescue operations, detailed reports from engineering teams' on-site investigations, and actual damage records from subsequent reinforcement and repair processes. This on-site validation data provides valuable labeled samples for model optimization.

[0072] After the data is uploaded to the cloud, it first enters a standardized preprocessing workflow. This involves timestamp alignment, coordinate unification, and data quality cleaning of the multi-source heterogeneous data to eliminate sensor errors and manual input biases. Subsequently, feature engineering is performed to extract physically meaningful features from the raw data, such as calculating the inter-story drift angle from acceleration time histories and extracting the building outline deformation rate from remote sensing imagery. These features are then correlated and integrated with static attributes such as the building's structural type and construction date. All sensitive information is anonymized during the preprocessing stage to ensure privacy and security.

[0073] Based on the processed high-quality dataset, a machine learning training process was initiated in the cloud. An incremental learning algorithm was employed, fine-tuning the parameters of the existing model rather than training from scratch. This preserved the model's existing knowledge while rapidly absorbing lessons learned from new disasters. Special attention was paid to handling imbalanced samples during training, using a weighted loss function and oversampling techniques to ensure accurate identification of a few building categories, such as those classified as "hazardous." The model structure was also specifically optimized, incorporating an attention mechanism that allows the model to adaptively focus on the features most relevant to the damage, improving interpretability and generalization capabilities.

[0074] After model training is complete, a rigorous validation process is initiated. The new model not only needs to meet accuracy requirements on the test set but also passes a physical plausibility test to ensure its predictions conform to engineering mechanics principles. Validated models are deployed to the model library of edge nodes and their applicable regional characteristics are marked. When the next disaster occurs, the most suitable model version can be automatically selected based on the actual conditions of the affected area, achieving a virtuous cycle of "using the experience of the previous disaster to guide the response to the next disaster." This continuous learning mechanism enables the system to continuously accumulate practical experience, gradually evolving into an increasingly accurate and reliable intelligent emergency decision-making platform.

[0075] Based on the same inventive concept, this invention also provides a device for assessing and allocating resources for regional earthquake-damaged building complexes. This device can be a personal computer, a server, or other apparatus that implements the aforementioned method for assessing and allocating resources for regional earthquake-damaged building complexes. The device includes: The multi-source data acquisition module is used to acquire real-time earthquake monitoring data, aerial and satellite remote sensing image data, on-site inspection data of terminal nodes, and historical earthquake damage and regional geological database data. The regional feature modeling and parameter adaptation module is used to construct a regional feature matrix based on historical earthquake damage and regional geological database data, generate an adaptive evaluation parameter set based on the regional feature matrix under a preset mapping function, and load the optimal building seismic damage assessment model based on the adaptive evaluation parameter set. The building cluster seismic damage assessment module is used to integrate real-time earthquake monitoring data, aerial and satellite remote sensing image data, and on-site inspection data of terminal nodes according to building identification, and then input them into the building seismic damage assessment model to calculate the predicted value of post-earthquake damage and the corresponding confidence level of the building. Based on the predicted value of post-earthquake damage and the adaptive assessment parameter set, the building is classified into four safety levels, including four levels: safe and usable, restricted use, requiring reinforcement, and dangerous. The multi-objective optimization resource scheduling module is used to construct a multi-dimensional optimization objective function that includes building importance, personnel safety, traffic accessibility, and resource utilization efficiency. It automatically adjusts the weights of each optimization objective according to the post-disaster stage and uses an improved multi-objective evolutionary algorithm combined with constrained programming to generate a local Pareto optimal scheduling scheme.

[0076] All relevant content of each step involved in the aforementioned embodiments of the regional earthquake-damaged building group assessment and resource allocation method can be referenced to the functional description of the corresponding functional module of the regional earthquake-damaged building group assessment and resource allocation device in the embodiments of this application, and will not be repeated here.

[0077] The module division in this embodiment is illustrative and represents only one logical functional division. In actual implementation, other division methods may be used. Furthermore, the functional modules in the various embodiments of this invention can be integrated into a single processor, exist as separate physical entities, or be integrated into a single module. The integrated modules described above can be implemented in hardware or as software functional modules.

[0078] like Figure 5 As shown, the present invention also provides a regional earthquake-damaged building complex assessment and resource allocation system, including a cloud, an edge node, and a terminal node, wherein the edge node is used to execute the above-described regional earthquake-damaged building complex assessment and resource allocation method; The terminal node is used to collect on-site inspection data of the earthquake-damaged area and send it to the edge node, as well as to obtain and display the resource scheduling scheme of the earthquake-damaged area pushed by the edge node; The cloud is used to acquire local Pareto optimal resource scheduling schemes generated by each edge node, and then generate a global optimal resource scheduling scheme based on each local Pareto optimal resource scheduling scheme and feed it back to each edge node. It also acquires and stores real-time earthquake monitoring data, aerial and satellite remote sensing image data, on-site inspection data of terminal nodes, and historical earthquake damage and regional geological database data to train and update parameters of the building seismic damage assessment model.

[0079] Following an earthquake, the regional earthquake-damaged building cluster assessment and resource allocation system provided by this invention can be rapidly deployed at the provincial emergency command center as the cloud hub, the frontline command vehicle in the disaster area as the edge node, and the handheld mobile terminals of inspection personnel as the terminal nodes. Through deep collaboration of the three-tiered architecture of cloud, edge nodes, and terminal nodes, an emergency response system with both global vision and local intelligence is constructed. This architecture innovatively solves the vulnerability of traditional centralized systems during network outages, while overcoming the lack of global coordination in fully distributed systems, achieving efficient and reliable operation in complex post-disaster environments.

[0080] In the system's implementation, terminal nodes, serving as the carriers for on-site information collection and command execution, are equipped with specialized data acquisition applications and multiple sensor interfaces. Rescue personnel use mobile terminals to systematically record damage details such as crack width and structural deformation. Simultaneously, the terminals automatically integrate location information and match it with the building's unique identifier. This on-site data is transmitted to edge nodes via secure channels, providing valuable on-site verification information for the evaluation model. The terminal devices also feature offline caching capabilities, ensuring continued operation even during communication interruptions, and automatically synchronizing data once the network is restored. Regarding command reception, the terminals present scheduling plans in a visual format, including highlighted task priorities, optimal routes, and resource allocation lists, significantly improving the efficiency of on-site rescue operations.

[0081] Edge nodes, acting as regional computing centers, are deployed in emergency command vehicles or fixed stations near disaster areas, undertaking real-time data processing and decision optimization tasks for their respective regions. Each node incorporates a multi-source data fusion engine, intelligently associating data from terminal inspections, local sensors, and remote sensing analysis according to building identifiers to construct complete building feature vectors. In terms of computing architecture, edge nodes employ a distributed parallel processing framework, capable of running multiple evaluation models and optimization algorithms simultaneously. Upon receiving a global scheduling plan from the cloud, the edge nodes adapt it regionally based on the latest local status data, making necessary fine-tuning while maintaining the global framework. This design ensures both timely decision-making and regional applicability of the solution.

[0082] As the system's central nervous system, the cloud possesses powerful data storage and computing capabilities. It maintains a comprehensive regional building information database, a historical earthquake damage case database, and a resource distribution map, providing knowledge support for global decision-making. Regarding resource scheduling, the cloud employs a decomposition-based multi-objective optimization framework, intelligently fusing local Pareto solutions uploaded from each edge node. By introducing boundary coordination constraints, it resolves resource conflicts between regions, generating a globally optimal scheduling scheme. For model training, the cloud uses machine learning algorithms to analyze the entire process data of past disasters, continuously optimizing and evaluating the model's parameters and structure, and intelligently pushing newly trained models to corresponding edge nodes. This continuous learning mechanism enables the system to continuously accumulate practical experience and gradually improve its response capabilities to future disasters.

[0083] The entire system maintains data synchronization across its three levels of nodes through an asynchronous communication mechanism and employs a publish-subscribe model for efficient command distribution. When network bandwidth is limited, the system automatically initiates data compression and differential transmission strategies to prioritize the delivery of critical commands. This resilient communication design ensures stable operation of the system under various network conditions, forming a complete closed loop from data acquisition, local assessment, and global optimization to on-site execution, providing comprehensive technical support for post-earthquake emergency response.

[0084] Although embodiments of the present invention have been disclosed above, they are not limited to the applications listed in the specification and embodiments. They can be applied to various fields suitable for the present invention. For those skilled in the art, other modifications can be easily made. Therefore, without departing from the general concept defined by the claims and their equivalents, the present invention is not limited to the specific details and illustrations shown and described herein.

Claims

1. A method for assessing and allocating resources for earthquake-damaged building complexes in a region, characterized in that, Includes the following steps: Acquire real-time earthquake monitoring data, aerial and satellite remote sensing image data, on-site inspection data of terminal nodes, and historical earthquake damage and regional geological database data; Based on historical earthquake damage and regional geological database data, a regional feature matrix is ​​constructed, comprising geological parameters, architectural parameters, and functional parameters. The geological parameters include soil type, fault distribution, seismic intensity zoning, and seismic ground acceleration index. The architectural parameters include structural form, number of floors, construction year, and seismic fortification level. The functional parameters include population density distribution, location of key facilities, and transportation network accessibility index. Based on the regional feature matrix, a preset mapping function P′=f(Tr,Bs,S1) is called to generate an adaptive evaluation parameter set containing structural yield threshold, residual bearing capacity coefficient and failure probability curve parameters. f(Tr,Bs,S1) is a multivariate nonlinear function obtained in advance through machine learning training on historical earthquake damage cases. Tr represents the feature vector of geological parameters, Bs represents the feature vector of building parameters, and S1 represents the feature vector of functional parameters. Then, the optimal building earthquake damage assessment model is selected based on the adaptive evaluation parameter set. Real-time earthquake monitoring data, aerial and satellite remote sensing image data, and on-site inspection data of terminal nodes are integrated according to building identification, and then input into the building seismic damage assessment model to calculate the predicted value of post-earthquake damage and the corresponding confidence level. Based on the predicted value of post-earthquake damage and the adaptive assessment parameter set, the building is classified into four safety levels, including four levels: safe and usable, restricted use, requiring reinforcement, and dangerous. A multi-dimensional optimization objective function is constructed, including building importance, personnel safety, traffic accessibility, and resource utilization efficiency. The weights of each optimization objective are automatically adjusted according to the post-disaster stage. A locally Pareto optimal resource scheduling scheme is generated using an improved multi-objective evolutionary algorithm combined with constrained programming. The improved multi-objective evolutionary algorithm is an improved non-dominated sorting genetic algorithm NSGA-II, whose improvements include: scalarizing the objective function based on dynamic weights before non-dominated sorting; and introducing a weighted objective spatial distance in the congestion calculation.

2. The method for assessing and allocating resources for regional earthquake-damaged building complexes as described in claim 1, characterized in that, Also includes: The local Pareto optimal resource scheduling scheme is uploaded to the cloud, so that the cloud can generate a global optimal resource scheduling scheme based on each local Pareto optimal resource scheduling scheme. Obtain the globally optimal resource scheduling plan from the cloud and push it to the terminal nodes.

3. The method for assessing and allocating resources for regional earthquake-damaged building complexes as described in claim 1, characterized in that, Methods for selecting the optimal building seismic damage assessment model based on an adaptive assessment parameter set include: The model library is maintained in the cloud and contains multiple evaluation models, each with preset applicable scenarios and model features. Calculate the matching degree between the adaptive evaluation parameter set and the model features of each evaluation model; Based on the matching degree, the evaluation model with the highest matching degree to the adaptive evaluation parameter set is selected.

4. The method for assessing and allocating resources for regional earthquake-damaged building complexes as described in claim 1, characterized in that, The method for generating local Pareto optimal scheduling schemes using an improved multi-objective evolutionary algorithm combined with constrained programming is as follows: the improved non-dominated sorting genetic algorithm NSGA-II is combined with constrained programming to dynamically adjust the Pareto optimal solution search path according to the weight of each objective, and parallel computation is performed using distributed edge nodes.

5. The method for assessing and allocating resources for regional earthquake-damaged building complexes as described in claim 1, characterized in that, When the resource reserve reported by the terminal node is lower than the set percentage threshold, or the accessibility index of the transportation network drops by more than the set percentage, local incremental optimization is performed, the resource scheduling plan for the earthquake-damaged area is modified, and updated to the terminal node.

6. The method for assessing and allocating resources for regional earthquake-damaged building complexes as described in claim 1, characterized in that, When communication bandwidth is insufficient, a local data caching and delay synchronization mechanism is adopted to ensure that the edge node has no less than N hours of autonomous computing and scheduling execution capability in the event of network outage.

7. The method for assessing and allocating resources for regional earthquake-damaged building complexes as described in claim 1, characterized in that, Also includes: Data from the entire process of post-disaster assessment and resource allocation is uploaded to the cloud, enabling the cloud to train a building seismic damage assessment model for the next disaster using machine learning methods.

8. A device for assessing and allocating resources for regional earthquake-damaged building complexes, used to perform the method as described in claim 1, characterized in that, include: The multi-source data acquisition module is used to acquire real-time earthquake monitoring data, aerial and satellite remote sensing image data, on-site inspection data of terminal nodes, and historical earthquake damage and regional geological database data. The regional feature modeling and parameter adaptation module is used to construct a regional feature matrix based on historical earthquake damage and regional geological database data, generate an adaptive evaluation parameter set based on the regional feature matrix under a preset mapping function, and load the optimal building seismic damage assessment model based on the adaptive evaluation parameter set. The building cluster seismic damage assessment module is used to integrate real-time earthquake monitoring data, aerial and satellite remote sensing image data, and on-site inspection data of terminal nodes according to building identification, and then input them into the building seismic damage assessment model to calculate the predicted value of post-earthquake damage and the corresponding confidence level of the building. Based on the predicted value of post-earthquake damage and the adaptive assessment parameter set, the building is classified into four safety levels, including four levels: safe and usable, restricted use, requiring reinforcement, and dangerous. The multi-objective optimization resource scheduling module is used to construct a multi-dimensional optimization objective function that includes building importance, personnel safety, traffic accessibility, and resource utilization efficiency. It automatically adjusts the weights of each optimization objective according to the post-disaster stage and uses an improved multi-objective evolutionary algorithm combined with constrained programming to generate a local Pareto optimal scheduling scheme.

9. A system for assessing and allocating resources for regional earthquake-damaged building complexes, characterized in that, It includes cloud, edge nodes and terminal nodes, wherein the edge nodes are used to execute the regional earthquake-damaged building complex assessment and resource allocation method as described in any one of claims 1 to 6; The terminal node is used to collect on-site inspection data of the earthquake-damaged area and send it to the edge node, as well as to obtain and display the resource scheduling scheme of the earthquake-damaged area pushed by the edge node; The cloud is used to acquire local Pareto optimal resource scheduling schemes generated by each edge node, and then generate a global optimal resource scheduling scheme based on each local Pareto optimal resource scheduling scheme and feed it back to each edge node. It also acquires and stores real-time earthquake monitoring data, aerial and satellite remote sensing image data, on-site inspection data of terminal nodes, and historical earthquake damage and regional geological database data to train and update parameters of the building seismic damage assessment model.

Images