A spatial resilience evaluation method and system for public health risk prevention and control

Abstract

This invention discloses a spatial resilience assessment method and system for public health risk prevention and control. An automated data acquisition and fusion platform is created to obtain sets of building information, environmental information, and spatial operation and maintenance resilience indicators, generating a unified spatial resilience feature vector. This feature vector is then input into a constructed spatiotemporal attention-based multi-task network model, which outputs a comprehensive resilience index that is adaptively dual-state ("normal-emergency"), thus improving the accuracy of spatial resilience assessment.

Description

Technical Field

[0001] This invention relates to the field of spatial resilience assessment technology, and in particular to a spatial resilience assessment method and system for public health risk prevention and control. Background Technology

[0002] Current campus space resilience assessments largely rely on expert experience and lack dynamic, quantitative resilience evaluation, especially when switching between "normal" and "emergency" states, where evaluation criteria become vague. Therefore, a scientific, intelligent, and quantifiable method for assessing campus space resilience is urgently needed to address these issues. Summary of the Invention

[0003] This invention provides a spatial resilience evaluation method and system for public health risk prevention and control, which solves the technical problems existing in the prior art and realizes a scientific, intelligent and quantifiable spatial resilience evaluation method.

[0004] This invention provides a spatial resilience assessment method for public health risk prevention and control, comprising: Obtain building information set Environmental information collection Space operation and maintenance resilience index set .

[0005] The building information set Environmental information collection Space operation and maintenance resilience index set Integrate into spatial resilience feature vector .

[0006] The spatial resilience feature vector Input the spatiotemporal attention multi-task network model, through the expression Obtain the dimensionality-upgraded feature vector ;in, For activation function, This is the weight matrix. This is the bias vector.

[0007] Through expression Get the query matrix Key matrix Value matrix ;in, These are the learnable query, key, and value projection matrices, respectively.

[0008] Through expressions Obtain the spatial attention weight matrix ;in, is the dimension of the key vector.

[0009] Through expression Features that enhance spatial interaction .

[0010] Through expression Obtain the gate vector in the feature dimension ;in, It is the sigmoid activation function. For gating layer parameters, It is a binary state encoding vector.

[0011] Through expression Obtain the final features of state adaptation. ;in, This indicates element-wise multiplication.

[0012] Through expression Obtain the comprehensive resilience index ;in, This is the weight matrix. This is the bias vector.

[0013] Specifically, it also includes: The building information set Environmental information collection Space operation and maintenance resilience index set Transformed into a refined simulation environment: Spatial nodes in the simulation environment represent a discrete functional area within the building and are associated with its geometric boundaries, functional types, and personnel capacity. Each spatial node is bound to a corresponding spatial operation and maintenance resilience index, and each connecting edge is associated with access attributes.

[0014] Each agent encapsulates an independent health state machine and a bimodal behavior policy library: In daily scenarios, the agent triggers behavior based on preset routines, employing a destination-attractive-based behavior model, combined with spatial functional semantic weights and real-time pedestrian density, through... The algorithm dynamically plans the optimal path; in emergency scenarios, the agent's behavior pattern is reconstructed: the overtly disturbed agent moves to the designated area according to the preset emergency transfer path, and the shortest emergency transfer route is calculated using Dijkstra's algorithm; the stable agent activates the evacuation procedure and calculates the individual's movement trajectory through the social force model.

[0015] Through formula Calculate the perturbation probability of the agent ;in, This represents the basic diffusion rate of the disturbance source. The current personnel density of the space unit. This is a correction factor for space environment risk. This is the simulation time step.

[0016] Through formula The computational agent is in a potentially disturbed state. Transition to a dominant disturbed state probability ;in, It is the reciprocal of the average duration of the potential disturbance phase.

[0017] Through formula The computational agent changes from a dominant perturbed state Switch to recovery status probability ;in, This represents the corresponding recovery rate.

[0018] Disturbance probability probability Sum of probabilities At each simulation time step The system updates the health status of all agents in parallel, simulating the evolution of the population's health status under public health risk disturbances. After the simulation, the state trajectories of each agent are statistically analyzed, and four core prevention and control label data are aggregated: global disturbance impact rate. Defined as the ratio of the peak cumulative number of affected individuals to the total number of affected individuals; effective reproduction number. The actual number of affected individuals per unit of explicitly disturbed agent within the diffusion period is calculated using a sliding window; evacuation completion time. Record the time taken from the triggering of the operational disturbance to the arrival of the last agent at the safety exit; emergency response capacity compliance rate. Calculate the ratio of available emergency response units to the peak number of affected people at the end of the emergency.

[0019] Through weighted synthesis function Generate ground truth labels for the prevention and control effectiveness of the spatiotemporal attention multi-task network model.

[0020] Specifically, it also includes: Through formula The space environment risk correction coefficient was calculated. ;in, For the first A space operation and maintenance resilience index value. These are the weighting coefficients for the corresponding indicators. This represents the total number of space operation and maintenance resilience indicators used in the calculation. This is the index for the corresponding space operation and maintenance resilience indicators.

[0021] Specifically, the loss function of the spatiotemporal attention multi-task network model is: ;in, The loss weighting coefficients predicted by the comprehensive resilience index. The loss weighting coefficient is used for the joint prediction of prevention and control effectiveness indicators. The loss weight coefficients for the L2 regularization term. The loss weight coefficients for the L1 regularization term are... Global disturbance impact rate The sub-weight coefficients, Effective reproduction number The sub-weight coefficients, Evacuation completion time The sub-weight coefficients, To ensure the emergency response capacity meets the standards The sub-weight coefficients, This represents the mean squared error loss in the comprehensive resilience index prediction. Global disturbance impact rate The predicted mean squared error loss, Effective reproduction number The predicted mean squared error loss, Evacuation completion time The predicted mean squared error loss, To ensure the emergency response capacity meets the standards The predicted mean squared error loss, The L2 regularization term represents the model parameters. Feature importance vector The L1 regularization term.

[0022] Specifically, it also includes: Through expression Get the first The feature importance weight of each indicator ;in, The spatial resilience feature vector The first in Individual indicator values, The spatial resilience feature vector The first in Individual indicator values, For summation index.

[0023] like The preset value will then be the corresponding indicator. Recorded as a key indicator.

[0024] Through expression Key indicators were calculated The toughness gain per unit increase ;in, The number of sampling points. This represents the comprehensive resilience prediction result of the spatiotemporal attention multi-task network model. Including key indicators Other indicators and characteristics, This represents the amplitude of a one-sided disturbance. This is the index of the number of sampling points.

[0025] Through expression The change in evacuation completion time was calculated. ;in, The evacuation completion time of the spatiotemporal attention multi-task network model is... Prediction function.

[0026] Through expression The change in the global disturbance impact rate was calculated. ;in, The global perturbation impact rate of the spatiotemporal attention multi-task network model. Prediction function.

[0027] Through expression The change in effective reproduction number was calculated. ;in, The effective regeneration number of the spatiotemporal attention multi-task network model. Prediction function.

[0028] Through expression The change in the emergency response capacity compliance rate was calculated. ;in, The emergency response capacity compliance rate of the spatiotemporal attention multi-task network model. Prediction function.

[0029] Through expression Calculate the overall benefit score ;in, The penalty coefficient is... Indicates taking 0 and The larger value between Indicates taking 0 and The larger value between Indicates taking 0 and The larger value between Indicates taking 0 and The larger of the two values.

[0030] recommend The preset modification method corresponding to the highest indicator.

[0031] This invention also provides a spatial resilience assessment system for public health risk prevention and control, comprising: The raw data acquisition module is used to obtain a set of building information. Environmental information collection Space operation and maintenance resilience index set .

[0032] The spatial resilience feature vector acquisition module is used to obtain the building information set. Environmental information collection Space operation and maintenance resilience index set Integrate into spatial resilience feature vector .

[0033] The feature vector dimensionality upscaling module is used to upscale the spatial resilience feature vector. Input the spatiotemporal attention multi-task network model, through the expression Obtain the dimensionality-upgraded feature vector ;in, For activation function, This is the weight matrix. This is the bias vector.

[0034] The matrix calculation module is used to perform calculations using expressions. Get the query matrix Key matrix Value matrix ;in, These are the learnable query, key, and value projection matrices, respectively.

[0035] The spatial attention weight matrix acquisition module is used to obtain the spatial attention weight matrix through an expression. Obtain the spatial attention weight matrix ;in, is the dimension of the key vector.

[0036] Spatial interaction enhancement feature acquisition module, used to obtain features through expressions Features that enhance spatial interaction .

[0037] The gated vector acquisition module is used to obtain vectors through expressions. Obtain the gate vector in the feature dimension ;in, It is the sigmoid activation function. For gating layer parameters, It is a binary state encoding vector.

[0038] The state-adaptive feature acquisition module is used to obtain features through expressions. Obtain the final features of state adaptation. ;in, This indicates element-wise multiplication.

[0039] The module for obtaining the comprehensive resilience index is used to obtain the index through an expression. Obtain the comprehensive resilience index ;in, This is the weight matrix. This is the bias vector.

[0040] Specifically, it also includes: The refined environment simulation module is used to collect the building information. Environmental information collection Space operation and maintenance resilience index set Transformed into a refined simulation environment: Spatial nodes in the simulation environment represent discrete functional areas within a building, associated with their geometric boundaries, functional types, and personnel capacity. Each spatial node is bound to a corresponding spatial operation and maintenance resilience index, and each connecting edge is associated with access attributes. Each intelligent agent encapsulates an independent health state machine and a bimodal behavior strategy library: In daily scenarios, the agent triggers behavior according to preset work-rest rules, adopting a destination-attractive-based behavior model, combining spatial functional semantic weights and real-time pedestrian density, through... The algorithm dynamically plans the optimal path; in emergency scenarios, the agent's behavior pattern is reconstructed: the overtly disturbed agent moves to the designated area according to the preset emergency transfer path, and the shortest emergency transfer route is calculated using Dijkstra's algorithm; the stable agent activates the evacuation procedure and calculates the individual's movement trajectory through the social force model.

[0041] The disturbance probability calculation module is used to calculate the probability of disturbance using the formula. Calculate the perturbation probability of the agent ;in, This represents the basic diffusion rate of the disturbance source. The current personnel density of the space unit. This is a correction factor for space environment risk. This is the simulation time step.

[0042] The first state transition probability calculation module is used to calculate the probability using the formula. The computational agent is in a potentially disturbed state. Transition to a dominant disturbed state probability ;in, It is the reciprocal of the average duration of the potential disturbance phase.

[0043] The second state transition probability calculation module is used to calculate the probability using the formula. The computational agent changes from a dominant perturbed state Switch to recovery status probability ;in, This represents the corresponding recovery rate.

[0044] The prevention and control tag data acquisition module is used for the probability of disturbance. probability Sum of probabilities At each simulation time step The system updates the health status of all agents in parallel, simulating the evolution of the population's health status under public health risk disturbances. After the simulation, the state trajectories of each agent are statistically analyzed, and four core prevention and control label data are aggregated: global disturbance impact rate. Defined as the ratio of the peak cumulative number of affected individuals to the total number of affected individuals; effective reproduction number. The actual number of affected individuals per unit of explicitly disturbed agent within the diffusion period is calculated using a sliding window; evacuation completion time. Record the time taken from the triggering of the operational disturbance to the arrival of the last agent at the safety exit; emergency response capacity compliance rate. Calculate the ratio of available emergency response units to the peak number of affected people at the end of the emergency.

[0045] The module for obtaining the true value label of prevention and control effectiveness is used to obtain the weighted comprehensive function. Generate ground truth labels for the prevention and control effectiveness of the spatiotemporal attention multi-task network model.

[0046] Specifically, it also includes: The space environment risk correction coefficient calculation module is used to calculate the risk correction coefficient using the formula. The space environment risk correction coefficient was calculated. ;in, For the first A space operation and maintenance resilience index value. These are the weighting coefficients for the corresponding indicators. This represents the total number of space operation and maintenance resilience indicators used in the calculation. This is the index for the corresponding space operation and maintenance resilience indicators.

[0047] Specifically, the loss function of the spatiotemporal attention multi-task network model is: ;in, The loss weighting coefficients predicted by the comprehensive resilience index. The loss weighting coefficient is used for the joint prediction of prevention and control effectiveness indicators. The loss weight coefficients for the L2 regularization term. The loss weight coefficients for the L1 regularization term are... Global disturbance impact rate The sub-weight coefficients, Effective reproduction number The sub-weight coefficients, Evacuation completion time The sub-weight coefficients, To ensure the emergency response capacity meets the standards The sub-weight coefficients, This represents the mean squared error loss in the comprehensive resilience index prediction. Global disturbance impact rate The predicted mean squared error loss, Effective reproduction number The predicted mean squared error loss, Evacuation completion time The predicted mean squared error loss, To ensure the emergency response capacity meets the standards The predicted mean squared error loss, The L2 regularization term represents the model parameters. Feature importance vector The L1 regularization term.

[0048] Specifically, it also includes: The feature importance weight calculation module is used to calculate the feature importance weight using an expression. Get the first The feature importance weight of each indicator ;in, The spatial resilience feature vector The first in Individual indicator values, The spatial resilience feature vector The first in Individual indicator values, For summation index.

[0049] Key indicator tagging module, used for... The preset value will then be the corresponding indicator. Recorded as a key indicator.

[0050] The toughness gain calculation module is used to calculate the toughness gain through the expression Key indicators were calculated The toughness gain per unit increase ;in, The number of sampling points. This represents the comprehensive resilience prediction result of the spatiotemporal attention multi-task network model. Including key indicators Other indicators and characteristics, This represents the amplitude of a one-sided disturbance. This is the index of the number of sampling points.

[0051] The evacuation completion time change calculation module is used to calculate the change through an expression. The change in evacuation completion time was calculated. ;in, The evacuation completion time of the spatiotemporal attention multi-task network model is... Prediction function.

[0052] The global disturbance impact rate change calculation module is used to calculate the change through an expression. The change in the global disturbance impact rate was calculated. ;in, The global perturbation impact rate of the spatiotemporal attention multi-task network model. Prediction function.

[0053] The effective reproduction number change calculation module is used to calculate the change through an expression. The change in effective reproduction number was calculated. ;in, The effective regeneration number of the spatiotemporal attention multi-task network model. Prediction function.

[0054] The emergency response capacity compliance rate change calculation module is used to calculate the change through an expression. The change in the emergency response capacity compliance rate was calculated. ;in, The emergency response capacity compliance rate of the spatiotemporal attention multi-task network model. Prediction function.

[0055] The comprehensive benefit score calculation module is used to calculate the comprehensive benefit score using an expression. Calculate the overall benefit score ;in, The penalty coefficient is... Indicates taking 0 and The larger value between Indicates taking 0 and The larger value between Indicates taking 0 and The larger value between Indicates taking 0 and The larger of the two values.

[0056] The modification method recommendation module is used to recommend modifications. The preset modification method corresponding to the highest indicator.

[0057] One or more technical solutions provided in this invention have at least the following technical effects or advantages: 1. Create an automated airborne system for acquiring and fusing multi-source heterogeneous data to obtain building information sets. Environmental information collection Space operation and maintenance resilience index set A unified spatial resilience feature vector is generated; then the spatial resilience feature vector is input into the constructed spatiotemporal attention multi-task network model, which outputs a comprehensive resilience index that is adaptive in both normal and emergency situations, thus improving the accuracy of spatial resilience evaluation.

[0058] 2. This invention achieves automated collection and high-precision fusion of multi-dimensional spatial resilience indicators, significantly improving the integrity and timeliness of the data.

[0059] 3. In terms of assessment model construction, an innovative simulation mechanism is proposed that couples building features with operational disturbance diffusion and evacuation dynamics. Spatial indicators are quantified as environmental risk factors and embedded into the disturbance probability calculation. The impact of spatial features on the infection probability is quantified in real time through environmental risk correction coefficients, generating highly credible prevention and control effectiveness labels, which further improves the accuracy of spatial resilience assessment.

[0060] 4. A three-stage model optimization strategy (virtual data pre-training → real data fine-tuning → state switching enhancement) is proposed, which combines multi-task regularization constraints to solve the problems of few-sample learning and overfitting.

[0061] 5. A decision-making closed loop driven by sensitivity analysis was established, which dynamically generates quantitative transformation suggestions based on feature importance, forming an intelligent decision-making chain from evaluation to optimization. Attached Figure Description

[0062] Figure 1 A flowchart of a spatial resilience evaluation method for public health risk prevention and control provided in an embodiment of the present invention. Detailed Implementation

[0063] To address the technical problems existing in the prior art, this invention provides an intelligent evaluation method for the resilience of building space operations based on multi-source data fusion and artificial intelligence. It achieves closed-loop evaluation through a technical chain of "data acquisition → simulation label generation → intelligent model training → decision output". This method first collects static geometric attributes, dynamic environmental data, and operation and maintenance indicators of building space. After spatiotemporal alignment, cleaning, and standardization, a unified spatial resilience feature vector and spatial topology network structure are generated to construct a refined simulation environment. Then, building features are embedded into an improved social force evacuation model and a four-state group operation evolution dynamics framework. The probability of disturbance is calculated through the interaction of personnel density and environmental risk correction coefficients, outputting a ground truth label for prevention and control effectiveness. Based on this, a spatiotemporal attention multi-task network model is constructed. Using the spatial resilience feature vector as input and the ground truth label for prevention and control effectiveness as the training objective, a spatial attention mechanism is used to analyze the correlation between indicators. Combined with state gating units, a dual-state adaptive feature recombination of "normal times" and "emergency times" is achieved, simultaneously outputting the corresponding comprehensive resilience index, effectiveness prediction, and feature importance weights for gradient analysis. Model training employs virtual-real data transfer learning, state switching enhancement, and multi-task regularization strategies. Finally, sensitivity analysis is triggered by user-specified states to generate a quantitative transformation suggestion report. The spatiotemporal attention multi-task network model, as the core of intelligent evaluation, is a deep learning architecture specifically designed for the quantitative analysis of building space resilience proposed in this invention. This model, through a multi-task learning framework, achieves deep analysis of spatial features and cross-state generalization capabilities, and innovatively constructs a feature interaction mechanism under a dual spatiotemporal vision. The model input is a normalized spatial resilience feature vector. The output simultaneously generates three objectives: a comprehensive resilience index, a prediction of prevention and control effectiveness, and a feature importance weight. The model achieves end-to-end mapping through cascaded embedding layers, a spatiotemporal attention module, and a multi-task prediction head.

[0064] The input embedding layer first takes the original feature vector Perform high-dimensional semantic upsizing. This is achieved through fully connected transformation. ,Will Dimensional input mapping to Hidden space, in which This is the weight matrix. This is the bias vector. This operation not only enhances the feature representation capability, but also unifies different types of indicators into a comparable vector space, laying the foundation for subsequent attention mechanisms.

[0065] The spatiotemporal attention module comprises two parallel interactive modeling units: spatial and state. Spatial attention simulates the interrelationships between architectural indicators, and its calculation process employs a scaled dot product attention mechanism. ; in, These are the learnable query, key, and value projection matrices, respectively. This is the spatial attention weight matrix, whose elements are... The physical meaning is the index For indicators The intensity of the impact. The feature representation enhances spatial interaction while preserving the original feature dimensional structure.

[0066] State attention focuses on the differences between "normal" and "emergency" scenarios. It introduces a binary state encoding vector. Indicates the normal state. (indicating an emergency situation), and combine it with After splicing, state adaptation is achieved through a gating mechanism: ; in, For gating layer parameters, It is the sigmoid activation function. The gate vector is the feature dimension. This represents element-wise multiplication. This is the final feature representation for state adaptation.

[0067] Multi-task prediction heads sharing features Based on this, three independent subnets branch out. The main task outputs a comprehensive resilience index through a fully connected layer and sigmoid activation: Its range It directly reflects the overall resilience level of the space. Task 2 uses linear layers. Simultaneously predict four prevention and control effectiveness indicators. ,in An exponential transformation is required to ensure positive values. Task 3 calculates the comprehensive resilience output based on the gradient-weighted activation mapping principle. For input features Gradient sensitivity: ; in, For the first The feature importance weights of the indicators satisfy .gradient Obtained through automatic differentiation via backpropagation, its absolute value reflects the input features. The marginal impact of minor perturbations on the overall resilience score is investigated. This design enables the spatiotemporal attention multi-task network model to combine predictive accuracy with decision interpretability, outputting both quantitative scores and analyzing key influencing factors, thus providing precise direction for campus renovation.

[0068] Therefore, the three-layer technical architecture of "multi-source heterogeneous data fusion - multi-agent simulation verification - AI model-driven evaluation" proposed in this embodiment of the invention realizes a complete closed loop from objective data collection to comprehensive intelligent evaluation.

[0069] To better understand the above technical solutions, the following will provide a detailed explanation of the technical solutions in conjunction with the accompanying drawings and specific implementation methods.

[0070] like Figure 1 As shown, the spatial resilience assessment method for public health risk prevention and control provided in this embodiment of the invention includes: Step S110: Obtain the building information set Environmental information collection Space operation and maintenance resilience index set .

[0071] This step will be explained in detail: By parsing semantically defined component entities (such as IfcDoor, IfcStair, IfcSpace) and attribute sets, key parameters such as the net width of door and window openings, the width of stair flights, room area, and functional classification are directly extracted. Furthermore, layer and primitive recognition are performed, using a pre-trained image recognition model to distinguish wall lines, door and window symbols, and room labels, and then reconstructing the topological relationship network based on the drawing scale. The Floyd-Warshall algorithm is applied to count the number of disjoint shortest paths between any two key nodes. Its core analytical process can be formally represented as follows: For a given building model M, the analytical function concatenates the aforementioned feature vectors according to predefined index calculation rules, performs normalization processing, and finally outputs an index matrix. .

[0072] By connecting to an existing network of environmental sensors using standard protocols, data on carbon dioxide concentration, temperature, humidity, and population density in various areas are acquired. This time-series data is then compiled into a dataset. It is used to verify and calibrate static design parameters. It also provides a structured mobile application through a manual data verification interface to guide assessors in on-site input or file retrieval of space operation and maintenance resilience indicators. ; This interface transforms unstructured information into structured fields and binds them to specific spatial location codes, integrating them into an attribute set. This ensures that the data has spatial attributes.

[0073] Step S120: Collect building information Environmental information collection Space operation and maintenance resilience index set Integrate into spatial resilience feature vector .

[0074] This step will be explained in detail: Receive building information set Environmental information collection Space operation and maintenance resilience index set First, spatiotemporal alignment is performed. The three types of data are uniformly mapped to the building space coordinate system via the spatiotemporal alignment engine: dynamic data is associated with the nearest BIM unit through spatial indexing, while static data is treated as a time-independent variable. For spatial alignment, the location information of manual verification and IoT sensing points is associated with the nearest BIM spatial unit through spatial indexing. For temporal alignment, dynamic data is aligned to the evaluation time. Slicing is performed based on a baseline, and static data is considered time-independent.

[0075] Subsequently, data cleaning and standardization were performed. Numerical indicators were dimensionless to eliminate the influence of dimensions. For the first... Individual indicators The range standardization method is used: ; in, and To find the theoretical or empirical minimum and maximum values ​​of this indicator, thus mapping the indicator value to... The interval is converted from a Boolean index to a binary value of 0 / 1. Finally, a unified and normalized spatial resilience feature vector is output. This feature vector, along with the accompanying spatial-timestamp context information. It serves directly as the input to downstream evaluation models and simulations, forming the cornerstone of the entire system's intelligent evaluation.

[0076] Step S130: Transfer the spatial resilience feature vector Input the spatiotemporal attention multi-task network model, through the expression Obtain the dimensionality-upgraded feature vector ;in, For activation function, This is the weight matrix. This is the bias vector.

[0077] Step S140: Through expression Get the query matrix Key matrix Value matrix ;in, These are the learnable query, key, and value projection matrices, respectively.

[0078] Step S150: Through the expression Obtain the spatial attention weight matrix ;in, is the dimension of the key vector.

[0079] Step S160: Through expression Features that enhance spatial interaction .

[0080] Step S170: By expression Obtain the gate vector in the feature dimension ;in, It is the sigmoid activation function. For gating layer parameters, It is a binary state encoding vector.

[0081] Step S180: By expression Obtain the final features of state adaptation. ;in, This indicates element-wise multiplication.

[0082] Step S190: Through the expression Obtain the comprehensive resilience index ;in, This is the weight matrix. This is the bias vector.

[0083] To achieve deep integration of building space physical characteristics, personnel evacuation behavior, and operational disturbance diffusion dynamics across the spatiotemporal dimensions, generate ground truth labels for prevention and control effectiveness, and improve the accuracy of evaluation, the following measures are also included: Collect building information Environmental information collection Space operation and maintenance resilience index set Transformed into a refined simulation environment: Spatial nodes in the simulation environment represent discrete functional areas within a building, associated with their geometric boundaries, functional types, and personnel capacity. Each spatial node is bound to a corresponding spatial operation and maintenance resilience index, and each connecting edge is associated with access attributes, including passage direction constraints, width attenuation factors, and congestion tolerance values. The geometric features of spatial units are discretized through meshing, forming the basic units for calculating personnel movement trajectories and detecting close-range contact. Among them, the spatial unit is a finer-grained discrete mesh in the simulation environment, used for personnel movement and contact detection. Each spatial node is further meshed to obtain several spatial units.

[0084] The resulting topological network directly serves as the underlying map structure for agent path planning and behavior simulation. During the simulation, the agent's position always falls within the spatial cell corresponding to a certain spatial node. Movement across nodes must traverse edges and is constrained by edge attributes. The agents generated in the simulation represent individual students and teachers, each encapsulating an independent health state machine (following the four-state group operation model with four stages: stable). → Potential disturbance → Dominant disturbance →Restore (and a bimodal behavior strategy library): In daily scenarios, the agent triggers behavior based on preset routines, employing a destination-attractive-based behavior model, combined with spatial functional semantic weights and real-time pedestrian density, through... The algorithm dynamically plans the optimal path. When a disturbance event is detected (such as the density of obviously disturbed agents in a certain area exceeding a threshold) and the system switches to an emergency scenario, the agent behavior pattern is reconstructed in the emergency scenario: obviously disturbed agents move to the designated area according to the preset emergency transfer path, and the shortest emergency transfer route is calculated using the Dijkstra algorithm; agents in a stable state activate the evacuation procedure and calculate the individual movement trajectory through a social force model. This model simultaneously solves for physical forces and psychological driving forces, dynamically simulates the formation of congestion, the extension of queues, and the dissipation of evacuation flows, and accurately captures the dynamic characteristics of the crowd in an emergency.

[0085] Through formula Calculate the perturbation probability of the agent ;in, This represents the basic diffusion rate of the disturbance source, which is determined by the characteristics of the disturbance source; The personnel density of the current spatial unit is obtained through statistical analysis of the spatial distribution of intelligent agents; This is a correction factor for space environment risk. This is the simulation time step.

[0086] Specifically, through the formula The space environment risk correction coefficient was calculated. ;in, For the first The larger the value of the space operation and maintenance resilience index, the better the environmental resilience. The weighting coefficients for the corresponding indicators are determined through regression analysis of historical public health event data or the expert Delphi method. This represents the total number of space operation and maintenance resilience indicators used in the calculation. This is an index for the corresponding spatial operational resilience index. This formula quantifies the amplification / suppression effect of the built environment on the diffusion of risk. When a certain index... When approaching the minimum value (poor environmental resilience), Increase the probability of being disturbed; conversely, when , Decreasing the value inhibits diffusion. Weight This reflects the differences in the contribution of different indicators to the risk of spread.

[0087] Through formula The computational agent is in a potentially disturbed state. Transition to a dominant disturbed state probability ;in, It is the reciprocal of the average duration of the potential disturbance phase.

[0088] Through formula The computational agent changes from a dominant perturbed state Switch to recovery status probability ;in, This represents the corresponding recovery rate.

[0089] Disturbance probability probability Sum of probabilities At each simulation time step The system updates the health status of all agents in parallel, simulating the evolution of the population's health status under public health risk disturbances. After the simulation, the state trajectories of each agent are statistically analyzed, and four core prevention and control label data are aggregated: global disturbance impact rate. Defined as the ratio of the peak cumulative number of affected individuals to the total number of affected individuals; effective reproduction number. The actual number of affected individuals per unit of explicitly disturbed agent within the diffusion period is calculated using a sliding window; evacuation completion time. Record the time elapsed from the triggering of the operational disturbance to the arrival of the last agent at the safety exit to assess emergency evacuation efficiency; emergency response capacity compliance rate. At the end of the emergency, the ratio of available emergency response units to the peak number of affected people is calculated to verify resource matching.

[0090] Through weighted synthesis function Generate ground truth labels for the prevention and control effectiveness of a spatiotemporal attention multi-task network model. By finely coupling architectural spatial features and diffusion dynamics parameters, generate high-fidelity training labels with spatiotemporal correlation, providing a data foundation for model evaluation.

[0091] The training and optimization process of the spatiotemporal attention multi-task network model is explained in detail below: First, based on the generated training dataset, which covers samples with different building layouts, population densities, and status scenarios, each sample... Includes spatial feature vectors The corresponding true value of prevention and control effectiveness and status labels , For the sample The dataset consists of binary state-encoded vectors. The dataset is divided into training, validation, and test sets in a 7:2:1 ratio, with the test set strictly separated and used only for final model evaluation. To improve data utilization efficiency, a stratified sampling strategy is employed to ensure a balanced distribution of various campus building types across the subsets.

[0092] The loss function design adopts a multi-task weighted mechanism. The loss function of the spatiotemporal attention multi-task network model is: ;in, The loss weighting coefficients predicted by the comprehensive resilience index. The loss weighting coefficient is used for the joint prediction of prevention and control effectiveness indicators. The loss weight coefficients for the L2 regularization term. The loss weight coefficients for the L1 regularization term are... Global disturbance impact rate The sub-weight coefficients, Effective reproduction number The sub-weight coefficients, Evacuation completion time The sub-weight coefficients, To ensure the emergency response capacity meets the standards The sub-weight coefficients, This represents the mean squared error loss in the comprehensive resilience index prediction. Global disturbance impact rate The predicted mean squared error loss, Effective reproduction number The predicted mean squared error loss, Evacuation completion time The predicted mean squared error loss, To ensure the emergency response capacity meets the standards The predicted mean squared error loss, The L2 regularization term represents the model parameters, preventing overfitting (the loss weight coefficient of the L2 regularization term). ), Feature importance vector The L1 regularization term (the loss weight coefficient of the L1 regularization term) Interpretability is improved by enforcing sparsity.

[0093] The optimization process employs a three-stage progressive strategy. The first stage involves working on a virtual campus dataset. The Adam optimizer is used for pre-training, with a learning rate set to [value missing]. , The second phase involves generating a synthetic dataset through parametric modeling, covering diverse building layout topologies, population density distributions, and equipment configuration schemes; this is done using real campus data. Fine-tuning, the learning rate decreased And freeze the embedding layer parameters, Physical data collected from the actual campus environment, including building information sets. Environmental information collection Space operation and maintenance resilience index set The third stage involves state-sensitive enhancement training for each sample. respectively =Normal and The model is input into the emergency state, and the bi-state prediction loss is calculated and backpropagated to improve the model's responsiveness to state transitions. An early stopping mechanism is used throughout training; training terminates when the validation set loss does not decrease for 10 consecutive epochs.

[0094] To achieve the linkage between feature importance weights obtained through gradient analysis and sensitivity analysis, generating quantifiable resilience modification suggestions, and forming a closed-loop decision-making process of "assessment-attribution-optimization," the following measures are also included: Through expression Get the first The feature importance weight of each indicator ;in, Spatial resilience feature vector The first in A normalized index value, Spatial resilience feature vector The first in A normalized index value, For summation index.

[0095] like The preset value will then be the corresponding indicator. Recorded as a key indicator, within the range Perform equidistant sampling.

[0096] Through expression Key indicators were calculated The toughness gain per unit increase ;in, The number of sampling points. The comprehensive resilience prediction results for the spatiotemporal attention multi-task network model are as follows. Including key indicators Other indicators and characteristics, This represents the amplitude of a one-sided disturbance. This is the index of the number of sampling points.

[0097] In this embodiment, .

[0098] Through expressions The change in evacuation completion time was calculated. ;in, Evacuation completion time for spatiotemporal attention multi-task network models Prediction function.

[0099] Through expressions The change in the global disturbance impact rate was calculated. ;in, Global perturbation impact rate of spatiotemporal attention multi-task network model Prediction function.

[0100] Through expressions The change in effective reproduction number was calculated. ;in, Effective regeneration number for spatiotemporal attention multi-task network models Prediction function.

[0101] Through expressions The change in the emergency response capacity compliance rate was calculated. ;in, To achieve the target emergency response capacity of the spatiotemporal attention multi-task network model Prediction function.

[0102] All changes are based on the same perturbation sampling point. Calculations are performed to ensure the comparability of results. This is done through expressions. Calculate the overall benefit score ;in, The penalty coefficient is... Indicates taking 0 and The larger value between Indicates taking 0 and The larger value between Indicates taking 0 and The larger value between Indicates taking 0 and The larger of the two values.

[0103] recommend The preset modification method corresponding to the highest indicator.

[0104] This invention also provides a spatial resilience assessment system for public health risk prevention and control, comprising: The raw data acquisition module is used to obtain a set of building information. Environmental information collection and Space Operation and Maintenance Resilience Index Set .

[0105] The spatial resilience feature vector acquisition module is used to collect building information. Environmental information collection and Space Operation and Maintenance Resilience Index Set Integrate into spatial resilience feature vector .

[0106] The feature vector dimensionality enhancement module is used to enhance the spatial resilience of feature vectors. Input the spatiotemporal attention multi-task network model, through the expression Obtain the upgraded feature vector ;in, For activation function, This is the weight matrix. This is the bias vector.

[0107] The matrix calculation module is used to perform calculations using expressions. Get the query matrix Key matrix Value matrix ;in, These are the learnable query, key, and value projection matrices, respectively.

[0108] The spatial attention weight matrix acquisition module is used to obtain the spatial attention weight matrix through an expression. Obtain the spatial attention weight matrix ;in, is the dimension of the key vector.

[0109] Spatial interaction enhancement feature acquisition module, used to obtain features through expressions Features that enhance spatial interaction .

[0110] The gated vector acquisition module is used to obtain vectors through expressions. Obtain the gate vector in the feature dimension ;in, It is the sigmoid activation function. For gating layer parameters, It is a binary state encoding vector.

[0111] The state-adaptive feature acquisition module is used to obtain features through expressions. Obtain the final features of state adaptation. ;in, This indicates element-wise multiplication.

[0112] The comprehensive resilience index acquisition module is used to obtain the index through an expression. Obtain the comprehensive resilience index ;in, This is the weight matrix. This is the bias vector.

[0113] To achieve deep integration of building space physical characteristics, personnel evacuation behavior, and operational disturbance diffusion dynamics across the spatiotemporal dimensions, generate ground truth labels for prevention and control effectiveness, and improve the accuracy of evaluation, the following measures are also included: The refined environment simulation module is used to aggregate building information. Environmental information collection and Space Operation and Maintenance Resilience Index Set Transformed into a refined simulation environment: Spatial nodes in the simulation environment represent discrete functional areas within a building, associated with their geometric boundaries, functional types, and personnel capacity. Each spatial node is bound to a corresponding spatial operation and maintenance resilience index, and each connecting edge is associated with access attributes, including passage direction constraints, width attenuation factors, and congestion tolerance values. The geometric features of spatial units are discretized through meshing, forming the basic units for calculating personnel movement trajectories and detecting close-range contact. The spatial unit is a finer-grained discrete mesh in the simulation environment, used for personnel movement and contact detection. Each spatial node is further meshed to obtain several spatial units. The resulting topological network directly serves as the underlying map structure for agent path planning and behavior simulation. During the simulation, the agent's position always falls within the spatial unit corresponding to a certain spatial node. When moving across nodes, it must traverse edges and is constrained by edge attributes. The agents generated in the simulation represent individual students and teachers. Each agent encapsulates an independent health state machine (following the four-state group operation model with four stages: stable). Potential disturbance Dominant disturbance recover (and a bimodal behavior strategy library): In daily scenarios, the agent triggers behavior based on preset routines, employing a destination-attractive-based behavior model, combined with spatial functional semantic weights and real-time pedestrian density, through... The algorithm dynamically plans the optimal path. When a disturbance event is detected (such as the density of obviously disturbed agents in a certain area exceeding a threshold) and the system switches to an emergency scenario, the agent behavior pattern is reconstructed in the emergency scenario: obviously disturbed agents move to the designated area according to the preset emergency transfer path, and the shortest emergency transfer route is calculated using the Dijkstra algorithm; agents in a stable state activate the evacuation procedure and calculate the individual movement trajectory through a social force model. This model simultaneously solves for physical forces and psychological driving forces, dynamically simulates the formation of congestion, the extension of queues, and the dissipation of evacuation flows, and accurately captures the dynamic characteristics of the crowd in an emergency.

[0114] The disturbance probability calculation module is used to calculate the probability of disturbance using the formula. Calculate the perturbation probability of the agent ;in, This represents the basic diffusion rate of the disturbance source, which is determined by the characteristics of the disturbance source; The personnel density of the current spatial unit is obtained through statistical analysis of the spatial distribution of intelligent agents; This is a correction factor for space environment risk. This is the simulation time step.

[0115] Specifically, it also includes: The space environment risk correction coefficient calculation module is used to calculate the risk correction coefficient using the formula. The space environment risk correction coefficient was calculated. ;in, For the first The larger the value of the space operation and maintenance resilience index, the better the environmental resilience. The weighting coefficients for the corresponding indicators are determined through regression analysis of historical operational event data or the expert Delphi method. This represents the total number of space operation and maintenance resilience indicators used in the calculation. This is an index for the corresponding spatial operational resilience index. This formula quantifies the amplification / suppression effect of the built environment on the diffusion of risk. When a certain index... When approaching the minimum value (poor environmental resilience), Increase the probability of being disturbed; conversely, when , Decreasing the value inhibits diffusion. Weight This reflects the differences in the contribution of different indicators to the risk of spread.

[0116] The first state transition probability calculation module is used to calculate the probability using the formula... The computational agent is composed of potentially disturbed states. Transition to a dominant disturbed state probability ;in, It is the reciprocal of the average duration of the potential disturbance phase.

[0117] The second state transition probability calculation module is used to calculate the probability using the formula. The computational agent changes from a dominant perturbed state Switch to recovery status probability ;in, This represents the corresponding recovery rate.

[0118] The prevention and control tag data acquisition module is used for the probability of disturbance. probability Sum of probabilities At each simulation time step The system updates the health status of all agents in parallel, simulating the evolution of the population's health status under public health risk disturbances. After the simulation, the state trajectories of each agent are statistically analyzed, and four core prevention and control label data are aggregated: global disturbance impact rate. Defined as the ratio of the peak cumulative number of affected individuals to the total number of affected individuals; effective reproduction number. The actual number of affected individuals per unit of explicitly disturbed agent within the diffusion period is calculated using a sliding window; evacuation completion time. Record the time elapsed from the triggering of the operational disturbance to the arrival of the last agent at the safety exit to assess emergency evacuation efficiency; emergency response capacity compliance rate. At the end of the emergency, the ratio of available emergency response units to the peak number of affected people is calculated to verify resource matching.

[0119] The module for obtaining the true value label of prevention and control effectiveness is used to obtain the weighted comprehensive function. Generate ground truth labels for the prevention and control effectiveness of a spatiotemporal attention multi-task network model. By finely coupling architectural spatial features and diffusion dynamics parameters, generate high-fidelity training labels with spatiotemporal correlation, providing a data foundation for model evaluation.

[0120] The training and optimization process of the spatiotemporal attention multi-task network model is explained in detail below: First, based on the generated training dataset, which covers samples with different building layouts, population densities, and status scenarios, each sample... Includes spatial feature vectors The corresponding true value of prevention and control effectiveness and status labels , For the sample The dataset consists of binary state-encoded vectors. The dataset is divided into training, validation, and test sets in a 7:2:1 ratio, with the test set strictly separated and used only for final model evaluation. To improve data utilization efficiency, a stratified sampling strategy is employed to ensure a balanced distribution of various campus building types across the subsets.

[0121] The loss function design adopts a multi-task weighted mechanism. The loss function of the spatiotemporal attention multi-task network model is: ;in, The loss weighting coefficients predicted by the comprehensive resilience index. The loss weighting coefficient is used for the joint prediction of prevention and control effectiveness indicators. The loss weight coefficients for the L2 regularization term. The loss weight coefficients for the L1 regularization term. Global disturbance impact rate The sub-weight coefficients, Effective reproduction number The sub-weight coefficients, Evacuation completion time The sub-weight coefficients, To ensure the emergency response capacity meets the standards The sub-weight coefficients, This represents the mean squared error loss in the comprehensive resilience index prediction. Global disturbance impact rate The predicted mean squared error loss, Effective reproduction number The predicted mean squared error loss, Evacuation completion time The predicted mean squared error loss, To ensure the emergency response capacity meets the standards The predicted mean squared error loss, The L2 regularization term represents the model parameters, preventing overfitting (the loss weight coefficient of the L2 regularization term). ), Feature importance vector The L1 regularization term (the loss weight coefficient of the L1 regularization term) Interpretability is improved by enforcing sparsity.

[0122] The optimization process employs a three-stage progressive strategy. The first stage involves working on a virtual campus dataset. The Adam optimizer is used for pre-training, with a learning rate set to 1. , The second phase involves generating a synthetic dataset through parametric modeling, covering diverse building layout topologies, population density distributions, and equipment configuration schemes; this is done using real campus data. Fine-tuning, the learning rate decreased And freeze the embedding layer parameters, Physical data collected from the actual campus environment, including building information sets. Environmental information collection and Space Operation and Maintenance Resilience Index Set The third stage involves state-sensitive enhancement training for each sample. respectively =Normal and =In case of emergency, the model is input, the bi-state prediction loss is calculated and backpropagated to improve the model's responsiveness to state transitions. An early stopping mechanism is used throughout the training process, and training is terminated when the validation set loss does not decrease for 10 consecutive epochs.

[0123] To achieve the linkage between feature importance weights obtained through gradient analysis and sensitivity analysis, generating quantifiable resilience modification suggestions, and forming a closed-loop decision-making process of "assessment-attribution-optimization," the following measures are also included: The feature importance weight calculation module is used to calculate the feature importance weight using an expression. Get the first The feature importance weight of each indicator ;in, Spatial resilience feature vector The first in A normalized index value, Spatial resilience feature vector The first in A normalized index value, For summation index.

[0124] Key indicator tagging module, used for... The preset value will then be the corresponding indicator. Recorded as a key indicator, within the range Perform equidistant sampling.

[0125] The toughness gain calculation module is used to calculate the toughness gain through the expression Key indicators were calculated The toughness gain per unit increase ;in, The number of sampling points. The comprehensive resilience prediction results for the spatiotemporal attention multi-task network model are as follows. Including key indicators Other indicators and characteristics, This represents the amplitude of a one-sided disturbance. This is the index of the number of sampling points.

[0126] In this embodiment, .

[0127] The evacuation completion time change calculation module is used to calculate the change through an expression. The change in evacuation completion time was calculated. ;in, Evacuation completion time for spatiotemporal attention multi-task network models Prediction function.

[0128] The global disturbance impact rate change calculation module is used to calculate the change through an expression. The change in the global disturbance impact rate was calculated. ;in, Global perturbation impact rate of spatiotemporal attention multi-task network model Prediction function.

[0129] The effective reproduction number change calculation module is used to calculate the change through an expression. The change in effective reproduction number was calculated. ;in, Effective regeneration number for spatiotemporal attention multi-task network models Prediction function.

[0130] The emergency response capacity compliance rate change calculation module is used to calculate the change through an expression. The change in the emergency response capacity compliance rate was calculated. ;in, To achieve the target emergency response capacity of the spatiotemporal attention multi-task network model Prediction function.

[0131] The comprehensive benefit score calculation module is used when all changes are based on the same disturbance sampling point. Calculations are performed to ensure the comparability of results. This is done through an expression. Calculate the overall benefit score ;in, The penalty coefficient is... Indicates taking 0 and The larger value between Indicates taking 0 and The larger value between Indicates taking 0 and The larger value between Indicates taking 0 and The larger of the two values.

[0132] The modification method recommendation module is used to recommend modifications. The preset modification method corresponding to the highest indicator.

[0133] In summary, this invention provides an intelligent quantitative evaluation method for campus space resilience that integrates architecture, artificial intelligence, and public safety science. This invention develops a spatial resilience evaluation model based on attention mechanisms and multi-task learning. It utilizes spatial attention mechanisms to analyze the nonlinear correlations between indicators, combines state gating units to achieve dual-state adaptive feature recombination, outputs a comprehensive resilience index, analyzes the impact weights of each spatial element on prevention and control effectiveness, and generates quantitative optimization suggestions. This method is particularly suitable for the planning, design, renovation, and emergency response capability assessment of primary and secondary school campuses under the background of public health risk prevention and control.

[0134] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0135] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0136] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0137] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0138] Any aspects of this invention not described in detail in the embodiments are well-known techniques to those skilled in the art. Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this invention and not to limit it. Although this invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of this invention without departing from the spirit and scope of this invention, and all such modifications and substitutions should be covered within the scope of the claims of this invention.

Claims

1. A spatial resilience assessment method for public health risk prevention and control, characterized in that, include: Obtain building information set Environmental information collection and Space Operation and Maintenance Resilience Index Set ; The building information set Environmental information collection and Space Operation and Maintenance Resilience Index Set Integrate into spatial resilience feature vector ; The spatial resilience feature vector Input the spatiotemporal attention multi-task network model, through the expression Obtain the upgraded feature vector ;in, For activation function, This is the weight matrix. It is the bias vector; Through expressions Get the query matrix Key matrix Value matrix ;in, These are the learnable query, key, and value projection matrices, respectively. Through expressions Obtain the spatial attention weight matrix ;in, The dimension of the key vector; Through expressions Features that enhance spatial interaction ; Through expressions Obtain the gate vector in the feature dimension ;in, It is the sigmoid activation function. For gating layer parameters, It is a binary state encoding vector; Through expressions Obtain the final features of state adaptation. ;in, This represents element-wise multiplication; Through expressions Obtain the comprehensive resilience index ;in, This is the weight matrix. This is the bias vector.

2. The spatial resilience assessment method for public health risk prevention and control as described in claim 1, characterized in that, Also includes: The building information set Environmental information collection and Space Operation and Maintenance Resilience Index Set Transformed into a refined simulation environment: Spatial nodes in the simulation environment represent a discrete functional area within the building and are associated with its geometric boundaries, functional types, and personnel capacity. Each spatial node is bound to a corresponding spatial operation and maintenance resilience index, and each connecting edge is associated with access attributes. Each agent encapsulates an independent health state machine and a bimodal behavior policy library: In daily scenarios, the agent triggers behavior based on preset routines, employing a destination-attractive-based behavior model, combined with spatial functional semantic weights and real-time pedestrian density, through... The algorithm dynamically plans the optimal path; in emergency scenarios, the agent's behavior pattern is reconstructed: the overtly disturbed agent moves to the designated area according to the preset emergency transfer path, and the shortest emergency transfer route is calculated using the Dijkstra algorithm; the stable agent activates the evacuation procedure and calculates the individual's movement trajectory through the social force model. Through formula Calculate the perturbation probability of the agent ;in, This represents the basic diffusion rate of the disturbance source. The current personnel density of the space unit. This is a correction factor for space environment risk. This is the simulation time step; Through formula The computational agent is composed of potentially disturbed states. Transition to a dominant disturbed state probability ;in, It is the reciprocal of the average duration of the potential disturbance phase; Through formula The computational agent changes from a dominant perturbed state Switch to recovery status probability ;in, This corresponds to the recovery rate; Disturbance probability probability Sum of probabilities At each simulation time step The system updates the health status of all agents in parallel, simulating the evolution of the population's health status under public health risk disturbances. After the simulation, the state trajectories of each agent are statistically analyzed, and four core prevention and control label data are aggregated: global disturbance impact rate. Defined as the ratio of the peak cumulative number of affected individuals to the total number of affected individuals; effective reproduction number. The actual number of affected individuals per unit of explicitly disturbed agent within the diffusion period is calculated using a sliding window; evacuation completion time. Record the time taken from the triggering of the operational disturbance to the arrival of the last agent at the safety exit; emergency response capacity compliance rate. Calculate the ratio of available emergency response units to the peak number of affected people at the end of the emergency; Through weighted synthesis function Generate ground truth labels for the prevention and control effectiveness of the spatiotemporal attention multi-task network model.

3. The spatial resilience assessment method for public health risk prevention and control as described in claim 2, characterized in that, Also includes: Through formula The space environment risk correction coefficient was calculated. ;in, For the first A space operation and maintenance resilience index value, These are the weighting coefficients for the corresponding indicators. This represents the total number of space operation and maintenance resilience indicators used in the calculation. This is the corresponding index of space operation and maintenance resilience indicators.

4. The spatial resilience assessment method for public health risk prevention and control as described in claim 2, characterized in that, The loss function of the spatiotemporal attention multi-task network model is: ;in, The loss weighting coefficients predicted by the comprehensive resilience index. The loss weighting coefficient is used for the joint prediction of prevention and control effectiveness indicators. The loss weight coefficients for the L2 regularization term. The loss weight coefficients for the L1 regularization term. Global disturbance impact rate The sub-weight coefficients, Effective reproduction number The sub-weight coefficients, Evacuation completion time The sub-weight coefficients, To ensure the emergency response capacity meets the standards The sub-weight coefficients, This represents the mean squared error loss in the comprehensive resilience index prediction. Global disturbance impact rate The predicted mean squared error loss, Effective reproduction number The predicted mean squared error loss, Evacuation completion time The predicted mean squared error loss, To ensure the emergency response capacity meets the standards The predicted mean squared error loss, The L2 regularization term represents the model parameters. Feature importance vector The L1 regularization term.

5. The spatial resilience assessment method for public health risk prevention and control as described in any one of claims 1-4, characterized in that, Also includes: Through expressions Get the first The feature importance weight of each indicator ;in, The spatial resilience feature vector The first in Individual indicator values, The spatial resilience feature vector The first in Individual indicator values, For summation index; like The preset value will then be the corresponding indicator. Recorded as a key indicator; Through expressions Key indicators were calculated The toughness gain per unit increase ;in, The number of sampling points. This represents the comprehensive resilience prediction result of the spatiotemporal attention multi-task network model. Including key indicators Other indicators and characteristics, This represents the amplitude of a one-sided disturbance. The index for the number of sampling points; Through expressions The change in evacuation completion time was calculated. ;in, The evacuation completion time of the spatiotemporal attention multi-task network model is... Prediction function; Through expressions The change in the global disturbance impact rate was calculated. ;in, The global perturbation impact rate of the spatiotemporal attention multi-task network model. Prediction function; Through expressions The change in effective reproduction number was calculated. ;in, The effective regeneration number of the spatiotemporal attention multi-task network model. Prediction function; Through expressions The change in the emergency response capacity compliance rate was calculated. ;in, The emergency response capacity compliance rate of the spatiotemporal attention multi-task network model. Prediction function; Through expressions Calculate the overall benefit score ;in, The penalty coefficient is... Indicates taking 0 and The larger value between Indicates taking 0 and The larger value between Indicates taking 0 and The larger value between Indicates taking 0 and The larger of the two values; recommend The preset modification method corresponding to the highest indicator.

6. A spatial resilience assessment system for public health risk prevention and control, characterized in that, include: The raw data acquisition module is used to obtain a set of building information. Environmental information collection and Space Operation and Maintenance Resilience Index Set ; The spatial resilience feature vector acquisition module is used to obtain the building information set. Environmental information collection and Space Operation and Maintenance Resilience Index Set Integrate into spatial resilience feature vector ; The feature vector dimensionality upscaling module is used to upscale the spatial resilience feature vector. Input the spatiotemporal attention multi-task network model, through the expression Obtain the upgraded feature vector ;in, For activation function, This is the weight matrix. It is the bias vector; The matrix calculation module is used to perform calculations using expressions. Get the query matrix Key matrix Value matrix ;in, These are the learnable query, key, and value projection matrices, respectively. The spatial attention weight matrix acquisition module is used to obtain the spatial attention weight matrix through an expression. Obtain the spatial attention weight matrix ;in, The dimension of the key vector; Spatial interaction enhancement feature acquisition module, used to obtain features through expressions Features that enhance spatial interaction ; The gated vector acquisition module is used to obtain vectors through expressions. Obtain the gate vector in the feature dimension ;in, It is the sigmoid activation function. For gating layer parameters, It is a binary state encoding vector; The state-adaptive feature acquisition module is used to obtain features through expressions. Obtain the final features of state adaptation. ;in, This represents element-wise multiplication; The comprehensive resilience index acquisition module is used to obtain the index through an expression. Obtain the comprehensive resilience index ;in, This is the weight matrix. This is the bias vector.

7. The spatial resilience assessment system for public health risk prevention and control as described in claim 6, characterized in that, Also includes: The refined environment simulation module is used to collect the building information. Environmental information collection and Space Operation and Maintenance Resilience Index Set Transformed into a refined simulation environment: Spatial nodes in the simulation environment represent discrete functional areas within a building, associated with their geometric boundaries, functional types, and personnel capacity. Each spatial node is bound to a corresponding spatial operation and maintenance resilience index, and each connecting edge is associated with access attributes. Each intelligent agent encapsulates an independent health state machine and a bimodal behavior strategy library: In daily scenarios, the agent triggers behavior according to preset work-rest rules, adopting a destination-attractive-based behavior model, combining spatial functional semantic weights and real-time pedestrian density, through... The algorithm dynamically plans the optimal path; in emergency scenarios, the agent's behavior pattern is reconstructed: the overtly disturbed agent moves to the designated area according to the preset emergency transfer path, and the shortest emergency transfer route is calculated using the Dijkstra algorithm; the stable agent activates the evacuation procedure and calculates the individual's movement trajectory through the social force model. The disturbance probability calculation module is used to calculate the probability of disturbance using the formula. Calculate the perturbation probability of the agent ;in, This represents the basic diffusion rate of the disturbance source. The current personnel density of the space unit. This is a correction factor for space environment risk. This is the simulation time step; The first state transition probability calculation module is used to calculate the probability using the formula... The computational agent is composed of potentially disturbed states. Transition to a dominant disturbed state probability ;in, It is the reciprocal of the average duration of the potential disturbance phase; The second state transition probability calculation module is used to calculate the probability using the formula. The computational agent changes from a dominant perturbed state Switch to recovery status probability ;in, This corresponds to the recovery rate; The prevention and control tag data acquisition module is used for the probability of disturbance. probability Sum of probabilities At each simulation time step The system updates the health status of all agents in parallel, simulating the evolution of the population's health status under public health risk disturbances. After the simulation, the state trajectories of each agent are statistically analyzed, and four core prevention and control label data are aggregated: global disturbance impact rate. Defined as the ratio of the peak cumulative number of affected individuals to the total number of affected individuals; effective reproduction number. The actual number of affected individuals per unit of explicitly disturbed agent within the diffusion period is calculated using a sliding window; evacuation completion time. Record the time taken from the triggering of the operational disturbance to the arrival of the last agent at the safety exit; emergency response capacity compliance rate. Calculate the ratio of available emergency response units to the peak number of affected people at the end of the emergency; The module for obtaining the true value label of prevention and control effectiveness is used to obtain the weighted comprehensive function. Generate ground truth labels for the prevention and control effectiveness of the spatiotemporal attention multi-task network model.

8. The spatial resilience assessment system for public health risk prevention and control as described in claim 7, characterized in that, Also includes: The space environment risk correction coefficient calculation module is used to calculate the risk correction coefficient using the formula. The space environment risk correction coefficient was calculated. ;in, For the first A space operation and maintenance resilience index value, These are the weighting coefficients for the corresponding indicators. This represents the total number of space operation and maintenance resilience indicators used in the calculation. This is the corresponding index of space operation and maintenance resilience indicators.

9. The spatial resilience assessment system for public health risk prevention and control as described in claim 7, characterized in that, The loss function of the spatiotemporal attention multi-task network model is: ;in, The loss weighting coefficients predicted by the comprehensive resilience index. The loss weighting coefficient is used for the joint prediction of prevention and control effectiveness indicators. The loss weight coefficients for the L2 regularization term. The loss weight coefficients for the L1 regularization term. Global disturbance impact rate The sub-weight coefficients, Effective reproduction number The sub-weight coefficients, Evacuation completion time The sub-weight coefficients, To ensure the emergency response capacity meets the standards The sub-weight coefficients, This represents the mean squared error loss in the comprehensive resilience index prediction. Global disturbance impact rate The predicted mean squared error loss, Effective reproduction number The predicted mean squared error loss, Evacuation completion time The predicted mean squared error loss, To ensure the emergency response capacity meets the standards The predicted mean squared error loss, The L2 regularization term represents the model parameters. Feature importance vector The L1 regularization term.

10. The spatial resilience assessment system for public health risk prevention and control as described in any one of claims 6-9, characterized in that, Also includes: The feature importance weight calculation module is used to calculate the feature importance weight using an expression. Get the first The feature importance weight of each indicator ;in, The spatial resilience feature vector The first in Individual indicator values, The spatial resilience feature vector The first in Individual indicator values, For summation index; Key indicator tagging module, used for... The preset value will then be the corresponding indicator. Recorded as a key indicator; The toughness gain calculation module is used to calculate the toughness gain through the expression Key indicators were calculated The toughness gain per unit increase ;in, The number of sampling points. This represents the comprehensive resilience prediction result of the spatiotemporal attention multi-task network model. Including key indicators Other indicators and characteristics, This represents the amplitude of a one-sided disturbance. The index for the number of sampling points; The evacuation completion time change calculation module is used to calculate the change through an expression. The change in evacuation completion time was calculated. ;in, The evacuation completion time of the spatiotemporal attention multi-task network model is... Prediction function; The global disturbance impact rate change calculation module is used to calculate the change through an expression. The change in the global disturbance impact rate was calculated. ;in, The global perturbation impact rate of the spatiotemporal attention multi-task network model. Prediction function; The effective reproduction number change calculation module is used to calculate the change through an expression. The change in effective reproduction number was calculated. ;in, The effective regeneration number of the spatiotemporal attention multi-task network model. Prediction function; The emergency response capacity compliance rate change calculation module is used to calculate the change through an expression. The change in the emergency response capacity compliance rate was calculated. ;in, The emergency response capacity compliance rate of the spatiotemporal attention multi-task network model. Prediction function; The comprehensive benefit score calculation module is used to calculate the comprehensive benefit score using an expression. Calculate the overall benefit score ;in, The penalty coefficient is... Indicates taking 0 and The larger value between Indicates taking 0 and The larger value between Indicates taking 0 and The larger value between Indicates taking 0 and The larger of the two values; The modification method recommendation module is used to recommend modifications. The preset modification method corresponding to the highest indicator.

Images