Urban post-earthquake resilience recovery prediction and decision-making method, system, device and medium

Abstract

This invention provides a method, system, equipment, and medium for predicting and making decisions on urban post-earthquake resilience recovery, belonging to the field of urban risk management and emergency disaster prevention technology. The method includes: collecting multi-source heterogeneous post-earthquake data; constructing and updating a multi-layered coupled urban network model in real time based on a dynamic expectation-maximization algorithm; extracting key causal links based on an improved Granger causal analysis model; using an adaptive reserve pool calculation model that integrates high-order network structures to perform multi-stage prediction of resilience recovery levels; generating recovery strategies based on key causal links and prediction results through a three-level progressive decision-making framework; and outputting recovery strategies and resource scheduling schemes through a visual decision-making platform. This invention solves the problems of static network modeling, coarse characterization of coupling effects, lagging recovery decisions, and lack of coordination in existing technologies, achieving dynamic perception, accurate prediction, and collaborative decision-making of the urban post-earthquake recovery process, and improving the efficiency of earthquake-resistant resilience recovery of urban systems.

Description

Technical Field

[0001] This invention relates to the field of urban risk management and emergency disaster prevention technology, and in particular to methods, systems, equipment and media for predicting and making decisions on urban post-earthquake resilience recovery. Background Technology

[0002] With the acceleration of urbanization, urban systems are becoming increasingly complex, consisting of tightly coupled subsystems such as buildings, transportation, energy, communications, and healthcare. When sudden disasters such as earthquakes occur, urban systems are prone to cascading failures across layers, leading to large-scale functional paralysis. Therefore, how to achieve rapid, orderly, and efficient recovery of urban systems after an earthquake, i.e., improving urban "earthquake resilience," has become a core issue in the field of urban safety and emergency management.

[0003] Currently, research and practice on urban post-earthquake recovery mainly focus on the following aspects, but there are still obvious limitations: 1. Recovery strategy based on static network topology and node importance.

[0004] This type of method constructs a complex network model of the daily state of the urban system, quantifies the importance of nodes using indicators such as node degree, clustering coefficient, and betweenness centrality, and optimizes the recovery order using genetic algorithms. Its limitations are: (1) it relies on the static topology before the earthquake and cannot reflect the dynamic damage and real-time changes of the network structure after the earthquake; (2) the node importance indicator is singular and does not consider functional redundancy and alternative paths; (3) the optimization algorithm is prone to getting trapped in local optima, and the computational complexity increases sharply with the network size, making it difficult to meet the timeliness requirements of emergency response.

[0005] 2. Collaborative recovery method based on inter-system causal correlation coefficient.

[0006] This type of method attempts to quantify the correlation coefficients of the recovery impact between systems by establishing time and financial constraint equations and resilience recovery matrices of urban systems under earthquake disasters, and then generate collaborative recovery schemes. Its main problems are: (1) the correlation coefficients are mostly based on theoretical assumptions or highly simplified coupling models, which are difficult to accurately characterize the dynamic interaction effects across systems in real earthquake disaster scenarios; (2) the models are poorly adaptable to dynamic constraints such as resource scheduling delays and policy adjustments during the recovery process; (3) the weight allocation in multi-objective optimization is highly subjective and lacks objective and traceable calibration basis.

[0007] 3. Data-driven agent model and deep learning method.

[0008] This type of method uses finite element simulation or historical earthquake damage data to construct urban system models, simulate the damage to infrastructure under different earthquake scenarios, and trains agent models with deep neural networks to replace time-consuming large-scale simulations. It learns repair strategies through interaction with the simulated environment. The main challenges it faces include: (1) the model performance is heavily dependent on the quality and coverage of the training data, and its ability to predict extreme earthquake damage beyond historical experience is significantly reduced; (2) it focuses on the functional recovery indicators of physical infrastructure and does not adequately consider the needs of social systems (such as population flow and community self-rescue capabilities); (3) the agent decision-making process lacks interpretability, is difficult to integrate with the experience of emergency command personnel, and has insufficient reliability and adaptability in actual complex disaster environments.

[0009] In summary, existing technologies generally suffer from three core shortcomings when facing the highly dynamic, multi-system coupled, and highly uncertain problem of urban post-earthquake recovery: static network modeling, coarse characterization of coupling effects, and lagging recovery decisions with a lack of coordination. This leads to significant discrepancies between existing recovery predictions and actual progress, low resource allocation efficiency, and an inability to support the resilient recovery goals of "precise repair, efficient activation, and low-cost recovery." Therefore, there is an urgent need for an intelligent decision-making method that can update network status in real time, accurately quantify the dynamic coupling relationships between systems, and generate interpretable, collaborative, and optimizable recovery strategies. Summary of the Invention

[0010] The purpose of this invention is to provide a method, system, device and medium for predicting and making decisions on urban post-earthquake resilience recovery, and to solve the problems of static network modeling, rough characterization of coupling effects, and delayed recovery decisions and lack of coordination in existing urban post-earthquake recovery technologies.

[0011] To achieve the above objectives, this invention provides a method for predicting and deciding on post-earthquake resilience recovery in cities, comprising the following steps: Step S1: Collect post-earthquake multi-source heterogeneous data, and construct and update in real time a multi-layered coupled urban network model containing functional layers of transportation, energy, medical care, communication, buildings, communities and refuge sites based on the dynamic expectation-maximization algorithm; Step S2: Based on the improved Granger causal analysis model, extract key causal links from the time series data of the urban multi-layer coupled network model; Step S3: Adaptive reserve pool calculation model with integrated high-order network structure is used to predict the resilience recovery level of urban system in multiple stages; Step S4: Based on the key causal links and multi-stage prediction results, a recovery strategy is generated through a three-level progressive decision framework. The three-level progressive decision framework includes, in turn: priority repair of key links based on the proximal strategy optimization algorithm, modular collaborative recovery based on graph neural networks, and global resource balance optimization based on multi-objective evolutionary algorithm. Step S5: Output and display the recovery strategy, prediction results and resource scheduling scheme through the visualization decision-making platform.

[0012] This invention also provides an urban post-earthquake resilience recovery prediction and decision-making system for implementing the urban post-earthquake resilience recovery prediction and decision-making method described above, including: The data acquisition and network modeling module is used to collect multi-source heterogeneous data after the earthquake. Based on the dynamic expectation-maximization algorithm, it constructs and updates in real time a multi-layer coupled urban network model that includes functional layers such as transportation, energy, medical care, communication, buildings, communities and refuge sites. The causal link extraction module is used to extract key causal links from time series data of a multi-layer coupled network model of a city based on an improved Granger causal analysis model. The extraction of key causal links includes constructing a prediction model to compare error differences, introducing a distance decay factor to calculate the standardized Granger causal effect value, and screening the top-ranked causal links. The resilience recovery prediction module is used to make multi-stage predictions of the resilience recovery level of urban systems using an adaptive reservoir calculation model that integrates a high-order network structure. The prediction module includes a dynamic leakage rate adjustment unit, a reservoir state calculation unit, and a Bayesian correction unit, which are used to output the resilience prediction values ​​and their confidence intervals for multiple preset key time nodes in the future. The progressive decision generation module is used to generate recovery strategies based on key causal links and multi-stage prediction results, through a three-level progressive decision framework. The three-level progressive decision framework includes: A critical link priority repair unit based on the near-end policy optimization algorithm is used to solve the Markov decision process to generate a critical link repair sequence and maximize the cumulative reward. A modular collaborative recovery unit based on graph neural networks is used to model the synergistic effect of nodes within a module and predict the recovery benefits under different resource inputs after the initial repair of critical links, based on the division of the smallest functional units, and to formulate the modular recovery sequence. The global resource balancing optimization unit based on the multi-objective evolutionary algorithm is used to solve the Pareto optimal solution set and form a global recovery strategy with the optimization objectives of maximizing the dynamic recovery exponent, minimizing the total resource consumption and minimizing the recovery time delay. A visual decision-making platform is used to output and display recovery strategies, prediction results, and resource scheduling plans, and provides a human-computer interaction interface to support emergency commanders in adjusting decision parameters in real time.

[0013] The present invention also provides a computer device, including: a memory and a processor; the memory stores a computer program, and the processor executes the computer program to implement the steps of the above-described method for predicting and deciding on urban post-earthquake resilience recovery.

[0014] The present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the above-described method for predicting and deciding on urban post-earthquake resilience recovery.

[0015] Therefore, the present invention employs the above-mentioned method, system, equipment, and medium for predicting and deciding on urban post-earthquake resilience recovery, and the beneficial technical effects are as follows: (1) In view of the problems that existing methods rely on pre-earthquake static topology and are difficult to handle post-disaster data loss and real-time changes in network structure, this invention integrates multi-source heterogeneous real-time data and uses the dynamic expectation-maximization (EM) algorithm to iteratively infer and update the network topology, thereby realizing adaptive modeling of urban system connectivity under dynamic disaster environments such as aftershocks and continuous damage.

[0016] (2) To address the problem that existing causal analysis methods provide a coarse characterization and inaccurate quantification of complex coupling effects between urban systems, this invention improves the Granger causal analysis model by introducing a distance decay factor and a dynamic significance threshold. This method can more accurately identify high-value functional synergistic units and their influence intensity from time-series data, effectively reducing the interference of noise and spurious correlations, thereby providing an objective and reliable quantitative basis for determining priority restoration targets and accurately allocating key resources.

[0017] (3) To address the shortcomings of existing recovery strategies, which are often passive, isolated, and lack global coordination, this invention proposes a three-level progressive intelligent decision-making framework. This framework uses reinforcement learning to drive rapid repair of critical links, graph neural networks to guide modular collaborative recovery, and multi-objective optimization to achieve global resource balance, forming a closed-loop decision-making process that combines "point-line-surface-volume". This design achieves a progression and balance from rapid local response to optimal overall benefits, significantly improving the overall efficiency of emergency resource scheduling and the synergy of system function recovery. Attached Figure Description

[0018] Figure 1 This is a flowchart of the urban post-earthquake resilience recovery prediction and decision-making method of the present invention; Figure 2 A curve comparing the predicted and actual values ​​of the toughness recovery index in the test area; Figure 3 This is a comparison chart showing the resilience recovery effects under different earthquake magnitude repair strategies. Detailed Implementation

[0019] The technical solution of the present invention will be further described below with reference to the accompanying drawings and embodiments.

[0020] Unless otherwise defined, the technical or scientific terms used in this invention shall have the ordinary meaning as understood by one of ordinary skill in the art to which this invention pertains.

[0021] Example 1 This embodiment provides a method for predicting and deciding on post-earthquake resilience recovery in cities. This method integrates dynamic modeling, causal analysis, time-series forecasting, and intelligent decision-making to form a closed-loop technical process. Figure 1 As shown. Specifically, it includes the following steps: Step S1: Collect post-earthquake multi-source heterogeneous data, and construct and update a multi-layered coupled urban network model that includes functional layers of transportation, energy, medical care, communication, buildings, communities and refuge sites in real time based on the dynamic expectation-maximization algorithm.

[0022] (1) Data Acquisition and Integration: Post-earthquake urban data acquisition is easily interrupted due to damage to buildings, transportation, and communication facilities. To address the problem of missing post-earthquake data, this invention integrates pre-earthquake infrastructure department filing data with post-earthquake UAV aerial photography data, real-time ground sensor data, and on-site survey feedback to jointly construct a basic database for post-earthquake network modeling. All data is aligned using spatial coordinates and timestamps.

[0023] (2) Layered Network Construction and Mathematical Representation: The key network layers for urban earthquake response are defined as including seven functional layers: transportation, energy, medical care, communication, buildings, community, and refuge areas. Each subsystem (layer) is abstracted as an independent network, and its core elements (nodes, edges) and functional indicators are shown in Table 1. Each functional layer The network structure uses an adjacency matrix. To provide a precise description, the matrix elements are defined. for: .

[0024] Table 1 Core Elements of Functional Layers in Urban Multi-Layer Coupled Networks

[0025] (3) Cross-system aggregation network generation: The individual single-layer networks are aggregated using logical operations (such as "OR") to generate a city-wide aggregation network containing cross-system coupling relationships. The aggregation operation aims to identify key coupling nodes such as "hospital-substation" and "road-communication tower," and its mathematical expression is: ; in, The elements in the adjacency matrix of the aggregated network represent the nodes considering all seven functional layers. and nodes Does there exist at least one cross-system connection or dependency between them? Here, Boolean operations are used as the operation functions. Representing nodes respectively and Are there direct connections between the seven layers: transportation, energy, medical, communication, architecture, community, and refuge? Represents a node and There is a direct connection in at least one functional layer; if Represents a node and There are no direct connections among any of the seven functional layers. They are considered to have no direct functional coupling in the current model.

[0026] (4) Topology inference and update based on dynamic EM algorithm: When there are missing post-earthquake observation data, the system initiates an improved expectation-maximization (EM) algorithm to perform topology inference to fill in the missing connectivity information and dynamically update the network model. The algorithm process is as follows: E-step (expected step): ; in, In the first The objective function (Q function) for the E-step of the next iteration; This represents the network topology under given conditions. The probability of occurrence is in the range [0, 1]. Functional layer The complete adjacency matrix, The adjacency matrix of the aggregated network reflects the cross-layer connection information after the "OR" operation, and has a constraining effect on single-layer connections. This includes connection status data of some known nodes and edges from sources such as sensors, drone inspections, and on-site reports. For the first The current model parameter set for the next iteration ( 、 ), Functional layer Middle node Probability estimate of the existence of connections between them. This is the cross-layer coupling strength parameter.

[0027] M-step (maximization step): ; in, For the first In the next iteration, the functional layer node and The updated estimate of the probability of a connection between them, ranging from a scalar. ; In the observation data In the text, it is explicitly recorded as "there is a connection", that is... Quantity; To calculate the expected number of connections in the E-step inference, sum over all unobserved connections based on the calculated posterior distribution: ; This represents the theoretical maximum number of connected edges in the network.

[0028] Cross-layer coupling strength update: Simultaneously, update the coupling strength parameter reflecting the similarity of network structures in different functional layers. The optimization objective is: ; in, For the first The cross-layer coupling strength parameter of the next iteration. To maximize the objective function The parameters are obtained by differentiating the objective function. This represents all non-repeating functional layer pairs. Find the product. Functional layer With functional layer The probability of structural consistency, with an index range of [missing information]. ; Functional layer With functional layer The Frobenius norm characterizes the degree of difference between two adjacency matrices; by Calculation. If the two layers are completely identical, the value is 0; the greater the difference, the larger the value.

[0029] Iterative convergence: Repeat the E-step and M-step until the parameter change is less than a preset threshold. The algorithm converged.

[0030] Step S2: Based on the improved Granger causal analysis model, extract key causal links from the time series data of the urban multi-layer coupled network model.

[0031] The improved Granger causal analysis model in step S2 is specifically used for: This step aims to identify key influencing relationships and quantify the recovery posture from the dynamic network model. The specific implementation process is as follows: Urban Network High-Order Structure Representation: To capture the synergistic effects of multiple nodes such as "community-hospital-shelter," a hypergraph is used to perform high-order modeling of the system based on the multi-layered urban network constructed in step S1, the historical operation logs of each subsystem, and real-time sensor data. The node set is defined. ,in Core nodes (such as hospitals and transportation hubs). These are the functionally related nodes (such as substations, refuge areas, water supply facilities, etc.). The core of high-order structure representation lies in the indivisibility of its functions, that is: ; in, For multi-node sets (such as "substation-hospital-shelter" or "community-substation-shelter-hospital"), A function for measuring functionality, used to quantify a set. Overall performance. This refers to the functionality of a single subsystem operating independently, such as a power supply system. This refers to the functionality of another subsystem operating independently, such as a building system. The inequality indicates that the overall functionality of the system... It is not a simple linear summation of the functions of its subsets; the synergistic effect of multiple subsystems needs to be considered.

[0032] Improved Granger Causal Effect Analysis: Based on the temporal functional state data of each node in the above-mentioned high-order hypergraph network, a refined causal link extraction is performed.

[0033] Model building and comparison: for any target node Construct two autoregressive prediction models: Full model: using nodes containing suspected cause nodes Historical data predictions for all relevant nodes. The state.

[0034] Elimination Model: Exclude Nodes After receiving the data, re-predict The state.

[0035] Causality Calculation: By comparing the prediction errors of the two models and introducing a distance decay factor to eliminate the interference of spatial proximity, the standardized Granger causality value is calculated. The core calculation formula is: ; in, For nodes right The causal effect value (0-1). For inclusion Information Time The predicted value, To eliminate After the message The predicted value, For nodes Actual functional values ​​(such as the number of patients received by the hospital). The significance threshold, For nodes and Spatial distance, This is the distance attenuation coefficient.

[0036] Key causal link extraction: Calculate the causal link of all directed node pairs in the network. The values ​​(G values) are sorted. The top 10% of high-weight causal links are selected as key causal links that require priority protection and repair. The output is a structured list containing the fields: [source node, target node, G value, functional layer type].

[0037] Quantifying the Post-Earthquake Recovery Level of Urban Systems: Defining a Dynamic Recovery Index to Assess Overall and Local Recovery Status As a system-level quantitative evaluation standard.

[0038] ; in, This is a set of functional modules divided using the Infomap algorithm. For nodes exist PageRank centrality at time step Service capacity recovery rate (measured value / designed capacity). Belongs to module The set of all nodes and All are modules One of the nodes, For nodes The weighted contribution to module recovery, ranging from ; Representation module At any moment The weighted functional level; For nodes Before the earthquake ( PageRank centrality of ), with a range of ; For module The sum of the baseline importance, ranging from ; For module The set of all boundary links, For the module Summing all boundary links, including links connecting the inside and outside of the module; For boundary links Importance weights The total weight of the boundary links of a module is used as the threshold. A module whose total boundary weight exceeds 75% is considered to have strong bridging ability.

[0039] Screening of critical nodes, vulnerable links and key modules: Based on the above analysis results, network elements that need to be focused on are automatically identified, and the screening method is shown in Table 2.

[0040] In Table 2, The average of the PageRank centrality of all nodes; The standard deviation of the PageRank centrality of all nodes; For module The weighted recovery rate is a value that reflects the functional level and the baseline importance. This represents the average recovery rate across the entire network. For nodes exist The ratio of actual service capacity to designed capacity at any given time.

[0041] Table 2 Node / Link / Module Filtering Methods

[0042] Step S3: Adaptive reserve pool calculation model with integrated high-order network structure is used to predict the resilience recovery level of urban system in multiple stages.

[0043] Multi-stage prediction is achieved through an adaptive reservoir computation model, which specifically includes: Construct a prediction model based on the reserve pool calculation, whose state update follows the formula: ; in, Representing the reserve pool Moment An internal state vector can dynamically encode the historical evolution memory of a system; For the input vector, For dimension, derived from the embedding function It is composed of high-order neighbor information and real-time functional data of nodes; Given a fixed input weight matrix, It is a sparse internal connection weight matrix, whose element values ​​are usually randomly initialized in the interval [-0.1, 0.1]. The function is a computational function that introduces nonlinear fitting capabilities, which can compress the results of linear combinations to the interval [-1, 1].

[0044] Dynamic leakage rate The calculation follows a formula that is dynamically adjusted based on changes in the recovery index: ; in, yes Dynamic leakage rate at any given time, range ; It is the norm of the system recovery index change, which can quantify the range of fluctuations in the overall system recovery state. Change . for System recovery index at any given time for The system recovery index at any given time.

[0045] Historical network status time-series records and real-time sensor monitoring data are used as input data. The data is input into the prediction model for calculation. The predictive model outputs resilience prediction values ​​for multiple preset key time nodes in the future. The predicted toughness values ​​are continuously corrected using a Bayesian correction method, and the corrected prediction distribution follows the formula: ; in, Based on the original predicted value The mean is given by the historical prediction error covariance matrix. The variance is a Gaussian distribution, representing the uncertainty of the model's predictions; For beta distribution, its shape parameter The prediction error is dynamically updated based on recent data, reflecting the empirical confidence in the accuracy of the model's predictions. The result is a posterior distribution, which integrates prior information and current prediction information, and its mean is used as the corrected final prediction value.

[0046] The final output is the corrected predicted value and its confidence interval, which serves as the multi-stage prediction result.

[0047] Step S4: Based on key causal links and multi-stage prediction results, a recovery strategy is generated through a three-level progressive decision-making framework.

[0048] This framework adopts a progressive repair logic of "point-line-surface-volume", as follows: Level 1: Prioritize the repair of critical links based on the Proximity Policy Optimization (PPO) algorithm.

[0049] Problem Modeling: The repair process is modeled as a Markov Decision Process (MDP). The state is the real-time network topology and node functional state, the action is the resource allocation scheme, and the reward function is designed as the cumulative sum of the causal effects of the repaired critical links.

[0050] Objective function: Maximize cumulative reward under resource constraints. ; in, Indicates link During the period Whether it has been repaired, Total resources for a given period of time. This is a discount factor for future earnings.

[0051] Strategy generation: The PPO algorithm is used to solve the MDP and dynamically generate repair sequences for high-priority links to ensure rapid recovery of critical function flows.

[0052] Level 2: Modular collaborative recovery based on graph neural networks (GNN).

[0053] Module partitioning: After the critical links are initially repaired, the network is partitioned based on community discovery algorithms (such as Infomap) to obtain multiple Minimal Functional Units (MFUs).

[0054] Benefit prediction: Based on the internal topology (adjacency matrix) of each MFU Using node features (PageRank centrality, damage level, etc.) as input, GNN is used to predict the recovery efficiency of the unit under different resource inputs: ; in, As the smallest functional unit In the future The predicted recovery benefit is set to a value of ; For graph neural network functions; For the first The smallest functional unit. The internal topological connectivity adjacency matrix of the smallest functional unit has a dimension of . ,element That is, if the node With nodes If there is a direct connection, then Otherwise, it is 0; The node feature matrix has dimensions of . , The feature dimension of the node.

[0055] Introducing macro-level functional modules The functional module consists of multiple spatially adjacent and functionally complementary minimum units. The internal synergy index of each module is calculated. This measures the degree of collaboration among nodes within a module. A synergy threshold is set. ,satisfy The module, which contains all the smallest functional units Only those that are prioritized for inclusion in the recovery sequence will be further considered. Based on this, the unit-level benefits of GNN predictions will be further utilized. Develop a modular recovery sequence.

[0056] Level 3: Global resource balancing optimization based on the multi-objective evolutionary algorithm (NSGA-II).

[0057] A multi-objective optimization problem is established with the objectives of maximizing the dynamic recovery exponent, minimizing the total resource consumption, and minimizing the recovery time delay. The objective function is as follows: ; in, For module The weight, For time period Based on non-dominated ranking and congestion calculation, a series of optimal trade-off schemes are generated for decision-makers to choose from according to actual conditions.

[0058] The non-dominated sorting genetic algorithm-II is used to solve the multi-objective optimization problem, generate the Pareto optimal solution set, and form a global recovery strategy.

[0059] In the decision-making processes of Level 1, Level 2, and Level 3, a dynamic recovery index is used. Quantitatively assess the recovery level and determine the conditions that are met. The real-time status of key nodes and vulnerable links is fed back to the decision-making loop, forming a closed-loop feedback. This represents the link causality value. The threshold for causal effects is the top 10%. This refers to the link recovery rate.

[0060] Step S5: Output and display the recovery strategy, prediction results and resource scheduling scheme through the visualization decision-making platform.

[0061] The invention will be further illustrated below with specific examples.

[0062] The example uses street data from a city in Southwest China with a population of 444,000. The experiment integrates multi-source heterogeneous information, and after spatiotemporal alignment and standardization, constructs a city system model with a seven-layer functional network, involving 13,326 nodes and 26,450 edges. The simulation recreates the recovery process 168 hours after an earthquake, setting up three comparison strategies: a random repair strategy, a degree-centrality-first strategy, and the three-level progressive decision-making strategy proposed in this invention.

[0063] First, based on the dynamic EM algorithm, the system successfully recovered 95.7% of network connectivity information after 15 iterations, assuming a 30% data loss post-earthquake condition. (Parameter changes during algorithm convergence are also discussed.) Cross-layer coupling strength The optimization to version 1.23 indicates strong structural correlations between functional layers. In the high-order structural characterization, the sequence "Hospital A + Substation D + Main Road S1 + 5 base stations" exhibits a significant synergistic effect, with the overall functional metric value far exceeding the sum of the independent functions of each subsystem. Improved Granger causality analysis extracts the top 10% of key causal links, among which the causal effect value of substation D on hospital A is [missing value]. Main road S1 leads to shelter A This verifies the importance of cross-system dependencies.

[0064] Secondly, the Infomap algorithm is used to divide the network into 6 functional modules. The weighted recovery rate was highest in the 24 hours following the earthquake, reaching 0.609. The overall dynamic recovery index of the system... The recovery rate gradually decreased from 0.12 after the earthquake, rising to 0.467 within 24 hours and reaching 0.623 within 48 hours. The prediction error of the 72-hour recovery trend based on the adaptive reservoir calculation model was less than 3%. After Bayesian correction, the width of the prediction confidence interval narrowed from ±0.045 to ±0.021. Particularly noteworthy is that the model accurately predicted the inflection point in the recovery rate approximately 36 hours after the earthquake, at which point the system transitioned from a rapid recovery phase to a stable recovery phase.

[0065] Table 3. 24-hour recovery status of functional modules in the case study area.

[0066] Third, with a resource constraint of 500 units, the first-level PPO algorithm prioritized repairing the four critical links with the highest causal effect values, accumulating a reward of 3.48 and improving system resilience from 0.12 to 0.467 within 24 hours. In the second-level modular recovery, a collaborative threshold was set. =0.65, three highly collaborative modules (M1) were selected. =0.84, M2 =0.79, M3 =0.72), and its included MFUs, after being sorted by GNN prediction benefit, Prioritized for repair. The third-level NSGA-II optimization generated a Pareto front, selecting the scheme with high equilibrium for execution. Finally, parameter sensitivity analysis was performed to verify the robustness of the method. When the collaborative threshold... When the value varies within the range of [0.55, 0.75], the 72-hour recovery index... The variation range is ±0.04; when the critical link screening ratio varies between [5%, 15%], The variation of ±0.03 indicates that the method is insensitive to parameter selection. In the data missing test, even with a 40% missing observation rate, topology inference based on dynamic EM still maintains a connectivity recovery accuracy of over 85%. The comparative experiment included extreme scenarios: when the epicenter was directly below the case area, random repair, due to not considering higher-order synergistic effects, resulted in a recovery efficiency that dropped to 68% of the method described in this invention; while in remote earthquake scenarios with epicenter distances >100km, the module-first method, due to accurately identifying long-distance dependencies, achieved a recovery efficiency 31% higher than the traditional method.

[0067] Example 2 Urban post-earthquake resilience recovery prediction and decision-making system, including: The data acquisition and network modeling module is used to collect multi-source heterogeneous data after the earthquake. Based on the dynamic expectation-maximization algorithm, it constructs and updates in real time a multi-layer coupled urban network model that includes functional layers such as transportation, energy, medical care, communication, buildings, communities and refuge sites. The causal link extraction module is used to extract key causal links from time series data of a multi-layer coupled network model of a city based on an improved Granger causal analysis model. The extraction of key causal links includes constructing a prediction model to compare error differences, introducing a distance decay factor to calculate the standardized Granger causal effect value, and screening the top-ranked causal links. The resilience recovery prediction module is used to make multi-stage predictions of the resilience recovery level of urban systems using an adaptive reservoir calculation model that integrates a high-order network structure. The prediction module includes a dynamic leakage rate adjustment unit, a reservoir state calculation unit, and a Bayesian correction unit, which are used to output the resilience prediction values ​​and their confidence intervals for multiple preset key time nodes in the future. The progressive decision generation module is used to generate recovery strategies based on key causal links and multi-stage prediction results, through a three-level progressive decision framework. The three-level progressive decision framework includes: A critical link priority repair unit based on the near-end policy optimization algorithm is used to solve the Markov decision process to generate a critical link repair sequence and maximize the cumulative reward. A modular collaborative recovery unit based on graph neural networks is used to model the synergistic effect of nodes within a module and predict the recovery benefits under different resource inputs after the initial repair of critical links, based on the division of the smallest functional units, and to formulate the modular recovery sequence. The global resource balancing optimization unit based on the multi-objective evolutionary algorithm is used to solve the Pareto optimal solution set and form a global recovery strategy with the optimization objectives of maximizing the dynamic recovery exponent, minimizing the total resource consumption and minimizing the recovery time delay. A visual decision-making platform is used to output and display recovery strategies, prediction results, and resource scheduling plans, and provides a human-computer interaction interface to support emergency commanders in adjusting decision parameters in real time.

[0068] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, essentially, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0069] The logic and / or steps represented in the flowchart or otherwise described herein, such as a sequenced list of executable instructions that can be considered for implementing logical functions, may be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can mean any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0070] More specific examples of computer-readable media (a non-exhaustive list) include: electrical connections (electronic devices) having one or more wires, portable computer disk drives (magnetic devices), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Furthermore, computer-readable media can even be paper or other suitable media on which the program can be printed, because the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in computer memory.

[0071] It is worth noting that all contents not described in detail in this invention are existing technologies and are well known to those skilled in the art.

[0072] Therefore, this invention employs the aforementioned urban post-earthquake resilience recovery prediction and decision-making methods, systems, equipment, and media. It overcomes the limitations of static modeling by achieving adaptive updates of the post-earthquake network topology through a dynamic expectation-maximization algorithm. By introducing an improved Granger causal analysis with a distance decay factor and a dynamic significance threshold, it accurately quantifies key cross-system dependencies, resolving the problem of coarse characterization of coupling effects. Furthermore, by integrating reinforcement learning, graph neural networks, and multi-objective optimization into a three-tiered progressive decision-making framework, it systematically coordinates rapid response and global optimization, effectively addressing the fragmentation and lag issues of traditional decision-making. This invention achieves a closed-loop process from data perception, dynamic modeling, accurate analysis to intelligent decision-making, providing efficient, collaborative, and quantifiable intelligent decision support for urban post-earthquake recovery, significantly improving the overall resilience and recovery efficiency of urban systems.

[0073] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for predicting and deciding on post-earthquake resilience recovery in cities, characterized in that, Includes the following steps: Step S1: Collect post-earthquake multi-source heterogeneous data, and construct and update in real time a multi-layered coupled urban network model containing functional layers of transportation, energy, medical care, communication, buildings, communities and refuge sites based on the dynamic expectation-maximization algorithm; Step S2: Based on the improved Granger causal analysis model, extract key causal links from the time series data of the urban multi-layer coupled network model; Step S3: Adaptive reserve pool calculation model with integrated high-order network structure is used to predict the resilience recovery level of urban system in multiple stages; Step S4: Based on key causal links and multi-stage prediction results, generate a recovery strategy through a three-level progressive decision-making framework; The three-level progressive decision-making framework includes: critical link priority repair based on proximal policy optimization algorithm, modular collaborative recovery based on graph neural network, and global resource balance optimization based on multi-objective evolutionary algorithm. Step S5: Output and display the recovery strategy, prediction results and resource scheduling scheme through the visualization decision-making platform.

2. The method for predicting and deciding on urban post-earthquake resilience recovery according to claim 1, characterized in that, Step S1, which involves constructing and updating the multi-layer coupled urban network model in real time, specifically includes: Based on pre-earthquake infrastructure registration data and post-earthquake real-time observation data obtained through drone aerial photography, sensor networks, and on-site surveys, an initial adjacency matrix is ​​generated for each functional layer. When data is missing, the dynamic expectation-maximization algorithm is started to perform topology inference. The expectation step is used to calculate the connection probability of unobserved links, and the maximization step is used to update the network parameters until convergence. When maximizing the step update of network parameters, a constraint term based on the difference measure between network structures of different functional layers is introduced to dynamically correct the strength of cross-system coupling relationships.

3. The method for predicting and deciding on urban post-earthquake resilience recovery according to claim 1, characterized in that, The improved Granger causal analysis model in step S2 is specifically used for: For the target node Two prediction models are constructed: the first model uses time-series data from all relevant nodes to predict node predictions. The state; the second model excludes specific source nodes. Re-predict nodes after data The state; By comparing the difference in prediction errors between the two models, when the difference exceeds a set significance threshold... At that time, determine the node For nodes There is a causal relationship; Introducing a distance decay factor for nodes With nodes Spatial distance between After correction, the standardized Granger causality effect value was calculated. ; Based on the calculated pairs of all nodes The values ​​are sorted, and the top-ranked causal links with a set percentage are selected as key causal links.

4. The method for predicting and deciding on urban post-earthquake resilience recovery according to claim 1, characterized in that, The multi-stage prediction in step S3 is achieved through an adaptive reservoir calculation model, specifically including: Construct a prediction model based on the reserve pool calculation, whose state update follows the formula: ； in, For the reserve pool in Moment Internal state vector, For the input vector, The dimension of the input vector. Given a fixed input weight matrix, The sparse internal connection weight matrix represents the dynamic leakage rate. The calculation formula is as follows: The adjustment is made dynamically based on changes in the recovery index. ； in, The norm of the system recovery exponential change; Historical network status time-series records and real-time sensor monitoring data are used as input data. Output weight matrix Mapping the state of the reserve pool to the future Step-by-step system recovery index prediction value ,in The predicted urban resilience values ​​output by the model are input into the prediction model for calculation. The predicted toughness values ​​are continuously corrected using a Bayesian correction method, and the corrected prediction distribution is as follows: ； in, Based on the original predicted value The mean is given by the historical prediction error covariance matrix. The variance is a Gaussian distribution. It is a beta distribution. For shape parameters, It is a posterior distribution; The final output is the corrected predicted value and its confidence interval as the multi-stage prediction result.

5. The method for predicting and deciding on urban post-earthquake resilience recovery according to claim 1, characterized in that, The first level of the three-level progressive decision-making framework is the critical link priority repair based on the near-end policy optimization algorithm, which specifically includes: Based on the extracted key causal links, a proximal policy optimization algorithm is used to dynamically allocate emergency resources, and a repair sequence is generated by solving a Markov decision process; the objective function of the Markov decision process is to maximize the cumulative reward. ； in, This is the total duration of the first recovery phase. yes Discount factor for each moment It is a collection of critical links, elements Indicates from node Pointing to node Directed links, For link The corresponding Granger causality value, For binary decision variables, Deadline Total cumulative repair costs.

6. The method for predicting and deciding on urban post-earthquake resilience recovery according to claim 1, characterized in that, The second level of the three-level progressive decision-making framework is modular collaborative recovery based on graph neural networks, which specifically includes: After the initial repair of critical links is completed at the first level, the smallest functional units are determined based on the partitioning of the urban multi-layer coupled network model. By utilizing the synergistic effect of nodes within the graph neural network modeling module, the recovery benefits under different resource inputs can be predicted. Graph neural networks use the adjacency matrix of the smallest functional unit as the topological connection. The node features are used as input, including PageRank centrality and degree of corruption. The formula for predicting recovery benefits using a graph neural network is: ； in, As the smallest functional unit In the future The predicted recovery benefits For graph neural network functions, For the first The smallest functional unit This is the internal topological connection adjacency matrix of the smallest functional unit. The node feature matrix; Introducing macro-level functional modules Calculate the internal synergy index of each functional module. Set a synergistic effect threshold ,satisfy The module, which contains all the smallest functional units Prioritized inclusion in the recovery sequence, combined with the unit-level benefits predicted by GNN. Develop a modular recovery sequence.

7. The method for predicting and deciding on urban post-earthquake resilience recovery according to claim 1, characterized in that, The third level of the three-level progressive decision-making framework is global resilience equilibrium optimization based on a multi-objective evolutionary algorithm, which specifically includes: A multi-objective optimization problem is established with the objectives of maximizing the dynamic recovery exponent, minimizing the total resource consumption, and minimizing the recovery time delay. The objective function is as follows: ； in, To maximize the system's global dynamic recovery index, The total number of modules, For module The weight, To the entire planning cycle Each time step within Minimize total resources, To minimize the total recovery time; The non-dominated sorting genetic algorithm-II is used to solve the multi-objective optimization problem, generate the Pareto optimal solution set, and form a global recovery strategy.

8. A prediction and decision-making system for post-earthquake resilience recovery in cities, characterized in that: The method for predicting and deciding on urban post-earthquake resilience recovery as described in any one of claims 1-7 includes: The data acquisition and network modeling module is used to collect multi-source heterogeneous data after the earthquake. Based on the dynamic expectation-maximization algorithm, it constructs and updates in real time a multi-layer coupled urban network model that includes functional layers such as transportation, energy, medical care, communication, buildings, communities and refuge sites. The causal link extraction module is used to extract key causal links from time series data of a multi-layer coupled network model of a city based on an improved Granger causal analysis model. The extraction of key causal links includes constructing a prediction model to compare error differences, introducing a distance decay factor to calculate the standardized Granger causal effect value, and screening the top-ranked causal links. The resilience recovery prediction module is used to make multi-stage predictions of the resilience recovery level of urban systems using an adaptive reservoir calculation model that integrates a high-order network structure. The prediction module includes a dynamic leakage rate adjustment unit, a reservoir state calculation unit, and a Bayesian correction unit, which are used to output the resilience prediction values ​​and their confidence intervals for multiple preset key time nodes in the future. The progressive decision generation module is used to generate recovery strategies based on key causal links and multi-stage prediction results, through a three-level progressive decision framework. The three-level progressive decision framework includes: A critical link priority repair unit based on the near-end policy optimization algorithm is used to solve the Markov decision process to generate a critical link repair sequence and maximize the cumulative reward. A modular collaborative recovery unit based on graph neural networks is used to model the synergistic effect of nodes within a module and predict the recovery benefits under different resource inputs after the initial repair of critical links, based on the division of the smallest functional units, and to formulate the modular recovery sequence. The global resource balancing optimization unit based on the multi-objective evolutionary algorithm is used to solve the Pareto optimal solution set and form a global recovery strategy with the optimization objectives of maximizing the dynamic recovery exponent, minimizing the total resource consumption and minimizing the recovery time delay. A visual decision-making platform is used to output and display recovery strategies, prediction results, and resource scheduling plans, and provides a human-computer interaction interface to support emergency commanders in adjusting decision parameters in real time.

9. A computer device, comprising: Memory and processor; The memory stores a computer program, characterized in that when the processor executes the computer program, it implements the steps of the urban post-earthquake resilience recovery prediction and decision-making method as described in any one of claims 1-7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the urban post-earthquake resilience recovery prediction and decision-making method as described in any one of claims 1-7.

Images