Adaptive risk assessment method and system for disaster prevention and mitigation emergency drill

Abstract

The disaster prevention and reduction emergency drill-oriented adaptive risk assessment method and system of the present application belongs to the cross technical field of information technology and emergency management, and comprises the following steps: S1, constructing a plan knowledge graph and generating an actual behavior graph; the construction of the plan knowledge graph comprises: analyzing the emergency drill plan, extracting the response units and task nodes defined in the plan as plan entity nodes, extracting the command chain and task dependency relationship between entities as the plan relationship edge, and generating the plan knowledge graph as a static assessment benchmark; the generation of the actual behavior graph comprises: real-time fusion and analysis of the multi-source heterogeneous data in the drill process to form a structured state flow unified in time and space, and constructing an actual behavior graph dynamically evolving with time according to the structured state flow; the present application provides a dynamic and comparable data basis for risk assessment by constructing double graphs, i.e., the plan knowledge graph constructed offline and the actual behavior graph generated in real time.

Description

Technical Field

[0001] This invention relates to the intersection of information technology and emergency management, specifically to an adaptive risk assessment method and system for disaster prevention and mitigation emergency drills. Background Technology

[0002] In the risk assessment of disaster prevention and mitigation emergency drills, existing technical methods suffer from static nature and a single assessment dimension. These methods mainly rely on comparing the actual execution process of the drill with the fixed emergency plan to perform a static compliance audit.

[0003] The current state and shortcomings of this model are mainly reflected in:

[0004] Limited assessment dimensions and inability to conduct in-depth cause analysis: The fundamental limitation of existing technologies lies in their limited assessment dimensions. They can only determine whether a behavior conforms to the plan, but cannot conduct in-depth cause analysis of behaviors that deviate from the plan.

[0005] Unable to effectively distinguish the nature of deviation behaviors: Due to the inability to attribute them at a deep level, existing technologies cannot effectively distinguish between two fundamentally different types of deviation behaviors. One is the disorderly failure caused by command errors or insufficient training, which is harmful. The other is the more efficient organizational resilience adaptation behavior proactively adopted by the exercise unit to cope with unexpected situations on the ground, which is beneficial.

[0006] As a result, traditional risk assessment methods cannot accurately perceive the true state of the exercise. They may mistakenly label efficient organizational resilience behaviors taken to deal with emergencies as risks, and at the same time, they cannot accurately identify disorderly failures caused by command or training problems. This greatly affects the accuracy and depth of risk assessment and is not conducive to the scientific evaluation of the exercise and the improvement of emergency response capabilities.

[0007] The information disclosed in the background section above is only intended to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0008] The purpose of this invention is to provide an adaptive risk assessment method and system for disaster prevention and mitigation emergency drills, so as to solve the problems mentioned in the background art.

[0009] The technical solution of the present invention includes the following steps:

[0010] S1. Constructing a contingency plan knowledge graph and generating an actual behavior graph; The construction of the contingency plan knowledge graph includes: parsing the emergency drill contingency plan, extracting the response units and task nodes defined in the plan as contingency plan entity nodes, extracting the command chain and task dependency relationship between entities as contingency plan relationship edges, and generating the contingency plan knowledge graph as a static evaluation benchmark; The generation of the actual behavior graph includes: real-time fusion and parsing of multi-source heterogeneous data during the drill process, forming a structured state flow with a unified spatiotemporal benchmark, and constructing an actual behavior graph that dynamically evolves over time based on the structured state flow;

[0011] S2. Conduct adaptive deviation quantification assessment; Adaptive deviation quantification assessment includes: receiving the contingency plan knowledge graph and the actual behavior graph as input, calculating the collaborative deviation index to quantify the correlation strength of the deviation behavior of multiple response units from the contingency plan, and when the collaborative deviation index is higher than the preset collaborative threshold, starting the calculation of the response consistency score to assess whether the deviation behavior has obtained logical support from other units in the system.

[0012] S3. Determine the risk level and issue an early warning; the determination of the risk level includes: comprehensive coordination deviation index and response consistency score. Based on the preset determination rule model, determine the risk level of the current exercise status and issue an early warning.

[0013] Preferably, the calculation of the co-deviation index in S2 specifically includes:

[0014] Determine the coordination coefficient of the contingency plan; the coordination coefficient of the contingency plan is determined based on the contingency plan knowledge graph. If any two response units in the contingency plan are specified to have a direct coordination relationship, the coordination coefficient of the contingency plan is 1, otherwise it is 0.

[0015] Calculate the dynamic interaction weights; the dynamic interaction weights are obtained by normalizing the actual interaction intensity between units in the actual behavior graph.

[0016] Obtain the multidimensional deviation vector; the multidimensional deviation vector is calculated by comparing the attribute differences of the corresponding units in the actual behavior graph and the contingency plan knowledge graph.

[0017] Based on the vector similarity between the pre-plan coordination coefficient, dynamic interaction weight, and multi-dimensional deviation vector, the coordination deviation index is calculated and output.

[0018] Preferably, the calculation of the response consistency score in S2 specifically includes:

[0019] The actual behavior graph is input into the graph neural network model to generate state embedding vectors for each response unit in the graph.

[0020] For response units that deviate, extract the set of neighboring nodes in the actual behavior graph;

[0021] Based on the state embedding vector of the off-unit, the state embedding vector of the neighboring node, and the learnable weight matrix in the graph neural network model, calculate the behavior alignment inner product value;

[0022] The behavior-aligned inner product value is normalized using an activation function to generate a response consistency score.

[0023] Preferably, the learnable weight matrix in the graph neural network model is obtained through offline supervised learning using graph snapshots of historical training data and corresponding expert annotations.

[0024] Preferably, the risk level determination in S3 specifically includes:

[0025] If the coordination deviation index is lower than or equal to the preset coordination deviation threshold, the system risk level is determined to be normal.

[0026] If the coordination deviation index is higher than the coordination deviation threshold, and the system's average response consistency score is higher than or equal to the preset response consistency threshold, then the system risk status is determined to be organizational resilience.

[0027] If the coordination deviation index is higher than the coordination deviation threshold and the system's average response consistency score is lower than the response consistency threshold, the system risk level is determined to be disorder failure, and an early warning is triggered.

[0028] Preferably, the following steps are also included:

[0029] After injecting a pre-set disturbance event during the exercise, the organizational resilience recovery time is calculated. The calculation of the organizational resilience recovery time includes: recording the occurrence time of the disturbance event and continuously monitoring the time change rate of the system-level coordination deviation index and response consistency score; when the absolute value of the time change rate falls below the pre-set stability threshold, it is determined as the time when the system returns to stability; the time span between the occurrence time of the disturbance event and the time when the system returns to stability is the organizational resilience recovery time.

[0030] Preferably, the preset stability threshold is determined by statistical analysis of the normal volatility of system-level indicators in historical stability training data.

[0031] An adaptive risk assessment system for disaster prevention and mitigation emergency drills includes the following modules:

[0032] The data processing module is used to parse emergency drill plans to build a plan knowledge graph and to integrate multi-source heterogeneous data in real time to generate a structured state flow.

[0033] The dynamic modeling module is used to construct a dynamic behavioral graph that evolves over time based on the structured state flow.

[0034] The deviation calculation module is used to compare the contingency plan knowledge graph with the actual behavior graph to calculate the collaborative deviation index;

[0035] The consistency assessment module is used to calculate the response consistency score based on a graph neural network model when the coordination deviation index is higher than the preset coordination threshold.

[0036] The risk decision-making module is used to comprehensively analyze the collaborative deviation index and response consistency score, and determine the risk level and output early warning based on preset judgment rules.

[0037] This invention provides an improved adaptive risk assessment method and system for disaster prevention and mitigation emergency drills, which, compared with existing technologies, has the following improvements and advantages:

[0038] 1. By constructing a dual graph, namely an offline-built contingency plan knowledge graph and a real-time generated actual behavior graph, a dynamic and comparable data foundation is provided for risk assessment. This step unifies static, unstructured contingency plans and dynamic, multi-source heterogeneous field data under a structured graph data model, making it possible to quantitatively analyze the differences between the two.

[0039] 2. The final conclusion is output through a pre-defined risk assessment rule model. The working principle of this model is based on a series of IF-THEN logical rules. The pre-defined collaborative deviation threshold and pre-defined response consistency threshold in these rules are set by statistical analysis of historical exercise data and combined with expert experience. For example, the quantile covering 95% of the fluctuations in normal exercise data is taken as the threshold to ensure that only statistically significant changes in indicators trigger the transition of risk levels. This mechanism deterministically maps complex continuous values ​​of dual indicators to discrete risk levels such as normal, organizational resilience, and disordered failure, thereby achieving accurate risk assessment and early warning. Attached Figure Description

[0040] The present invention will be further explained below with reference to the accompanying drawings and embodiments:

[0041] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0042] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.

[0043] Example 1

[0044] Please see Figure 1 This invention provides an adaptive risk assessment method for disaster prevention and mitigation emergency drills, comprising the following steps:

[0045] S1. Constructing a contingency plan knowledge graph and generating an actual behavior graph; The construction of the contingency plan knowledge graph includes: parsing the emergency drill contingency plan, extracting the response units and task nodes defined in the plan as contingency plan entity nodes, extracting the command chain and task dependency relationship between entities as contingency plan relationship edges, and generating the contingency plan knowledge graph as a static evaluation benchmark; The generation of the actual behavior graph includes: real-time fusion and parsing of multi-source heterogeneous data during the drill process, forming a structured state flow with a unified spatiotemporal benchmark, and constructing an actual behavior graph that dynamically evolves over time based on the structured state flow;

[0046] S2. Conduct adaptive deviation quantification assessment; Adaptive deviation quantification assessment includes: receiving the contingency plan knowledge graph and the actual behavior graph as input, calculating the collaborative deviation index to quantify the correlation strength of the deviation behavior of multiple response units from the contingency plan, and when the collaborative deviation index is higher than the preset collaborative threshold, starting the calculation of the response consistency score to assess whether the deviation behavior has obtained logical support from other units in the system.

[0047] S3. Determine the risk level and issue an early warning; The determination of the risk level includes: a comprehensive coordination deviation index and a response consistency score. Based on a preset determination rule model, the risk level of the current exercise status is determined and an early warning message is issued.

[0048] This invention provides an adaptive risk assessment method for disaster prevention and mitigation emergency drills. The method aims to solve the technical problem that existing risk assessment methods are static and cannot effectively distinguish whether deviations from the plan during drills are beneficial organizational resilience or harmful disorderly failures.

[0049] Compared with existing technologies that rely on comparing the actual execution process with a fixed plan to perform static compliance audits, this invention represents a substantial technological advancement. The fundamental limitation of existing technologies lies in their singular evaluation dimensions, which fail to provide in-depth attribution of deviations from the plan. Consequently, they cannot effectively distinguish between disorderly failures caused by command errors or insufficient training and the more efficient organizational resilience and adaptive behaviors proactively adopted by exercise units to cope with unexpected situations on the ground. This invention, by introducing a series of interconnected quantitative evaluation mechanisms, achieves a leap from compliance audits to adaptive insights.

[0050] The implementation of the method begins with S1, constructing a contingency plan knowledge graph and generating an actual behavior graph. This step establishes a benchmark and provides real-time input for subsequent risk assessment. The construction of the contingency plan knowledge graph is an offline preparation before the exercise, aiming to transform unstructured emergency plans into machine-readable structured data as a static benchmark for evaluation. In this embodiment, the text file of the emergency exercise plan is input into a parsing engine based on natural language processing technology. This engine is responsible for parsing the emergency exercise plan, extracting response units defined in the plan, such as fire brigades and medical teams, and task nodes, such as evacuating the public and delivering supplies, through entity recognition and relation extraction technologies, as entity nodes in the plan. Simultaneously, the engine further extracts the command chain between entities, such as the command relationship from the command center to the fire brigade and task dependency relationships, such as task A must start after task B is completed, as relation edges in the plan. These nodes and edges together generate the contingency plan knowledge graph.

[0051] ;

[0052] in, : A knowledge graph of the contingency plan, serving as a static evaluation benchmark; It is a collection of entity nodes in the contingency plan; It is a set of pre-planned relationship edges; this graph is stored in a fixed manner, providing an invariant reference system for all subsequent deviation calculations;

[0053] During the exercise, the generation of the actual behavior map was executed to create a real-time, high-fidelity digital model of the exercise's actual state. This step involved real-time fusion and analysis of multi-source heterogeneous data from the exercise process. This data could include GPS positioning data from emergency unit handheld terminals, radio voice communication records, and on-site video surveillance footage. A centralized data processing platform received this data, added a unified server timestamp, and projected it onto a unified geographic coordinate system to form a structured state flow with a unified spatiotemporal reference. Based on this structured state flow, a dynamic behavior map that evolves over time was constructed.

[0054] ;

[0055] in, : A dynamic graph of actual behavior that evolves over time; Actual Behavior Mapping in The set of entity nodes at any given moment; Actual Behavior Mapping in The set of edges relating to time; : Indicates the time spent during the drill;

[0056] For example, when a communication record appears in the state stream where fire brigade A requests support from medical team B, the system will create an edge carrying time and content attributes between the node representing fire brigade A and the node representing medical team B in the graph, thereby dynamically reflecting the actual interaction on site.

[0057] The method proceeds to S2 for adaptive deviation quantification assessment; the purpose of this step is to accurately quantify the difference between the practiced behavior and the contingency plan through mathematical models, and to explore the intrinsic nature of the difference; this step receives the contingency plan knowledge graph generated in the previous steps. With actual behavior map As input, a collaborative deviation index is calculated, which quantifies the correlation strength of multiple response units' deviations from the planned behavior. The underlying logic is that organized adaptive behavior is usually reflected in the consistency of multiple related units in the timing and direction of deviation, while disordered failure is manifested as isolated and random deviations. When the collaborative deviation index is higher than a preset collaborative threshold, the calculation of the response consistency score is initiated. The role of the response consistency score is to further analyze in depth to assess whether the collaborative deviation behavior is logically supported by other units within the system, thereby providing a key basis for distinguishing between organizational resilience and disordered failure.

[0058] The method executes step S3, determines the risk level, and outputs an early warning. This step is the decision-making output link of the entire method. Its purpose is to transform the complex quantitative indicators calculated in the previous steps into risk status conclusions that are easy for commanders to understand. This step integrates two core indicators, the coordination deviation index and the response consistency score, and determines the risk level of the current exercise state based on a preset judgment rule model. For example, it is classified as normal, organizational resilience, or disorderly failure. The corresponding early warning information is then output to the exercise command center.

[0059] This embodiment successfully transforms static compliance auditing into dynamic, in-depth risk insight by constructing a dual map of contingency plans and reality, and introducing a mechanism for quantitatively assessing the synergy and consistency of deviation behaviors. It can accurately distinguish between efficient organizational resilience behaviors proactively taken to cope with emergencies and disorderly failures caused by insufficient command or training, thereby greatly improving the accuracy and depth of risk assessment and providing a reliable basis for the scientific evaluation and capability enhancement of exercises.

[0060] The calculation of the co-deviation index in S2 specifically includes:

[0061] Determine the coordination coefficient of the contingency plan; the coordination coefficient of the contingency plan is determined based on the contingency plan knowledge graph. If any two response units in the contingency plan are specified to have a direct coordination relationship, the coordination coefficient of the contingency plan is 1, otherwise it is 0.

[0062] Calculate the dynamic interaction weights; the dynamic interaction weights are obtained by normalizing the actual interaction intensity between units in the actual behavior graph.

[0063] Obtain the multidimensional deviation vector; the multidimensional deviation vector is calculated by comparing the attribute differences of the corresponding units in the actual behavior graph and the contingency plan knowledge graph.

[0064] Based on the vector similarity between the pre-plan coordination coefficient, dynamic interaction weight, and multi-dimensional deviation vector, the coordination deviation index is calculated and output.

[0065] In this embodiment, the calculation method of the collaborative deviation index in S2 is specified; the purpose of the collaborative deviation index is to quantify the degree of interrelation between the deviation behaviors of multiple response units from the plan; the calculation formula is defined as:

[0066] ;

[0067] in, :time The co-deviation index is a dimensionless value; : Coordination coefficient of contingency plan; Dynamic interaction weights; :unit The multidimensional deviation vector; : Represents different response units; Vector similarity function; :time;

[0068] The calculation logic for this index is as follows:

[0069] To ensure that calculations are performed only between units that are related in the plan, the coordination coefficient of the plan needs to be determined. The contingency coordination coefficient is a binary representation of the collaborative relationships between units in the contingency plan, derived from the contingency plan knowledge graph. In this embodiment, by traversing... If any two response units in the contingency plan are specified to have a direct collaborative relationship, then this coefficient... Set to 1 if specified, otherwise 0;

[0070] To ensure that units with more frequent actual interactions have a higher weighting in the index calculation, dynamic interaction weights need to be calculated. Dynamic interaction weights are a normalized measure of the actual interaction intensity between units during the exercise, derived from the actual behavior graph. The calculation method is to statistically analyze the data within a time window. Frequency of interaction between the two Then, it is obtained by normalizing the actual interaction intensity between units in the actual behavior graph;

[0071] To quantify the magnitude and direction of the deviation, a multidimensional deviation vector needs to be obtained. The multidimensional deviation vector represents the difference between the actual state of a single response unit at a specific moment and the state specified in the plan; it is derived from comparison. and The result; in this embodiment, the vector is calculated by comparing the attributes of the corresponding units in the actual behavior graph and the pre-plan knowledge graph, such as position, state, and task execution order difference;

[0072] Based on the above-mentioned coordination coefficient of the contingency plan Dynamic interaction weights With multidimensional deviation vector Vector similarity between In this embodiment, a cosine similarity function is used to calculate and output the cooperative deviation index. ;

[0073] This embodiment clarifies and specifies the calculation method of the collaborative deviation index, making the quantification process of the correlation strength of deviation behaviors of multiple units operable and reproducible. It not only considers whether the unit deviates, but also integrates the collaborative planning of the plan, the actual interaction intensity on site, and the directionality of the deviation behavior. As a result, the calculated index can more accurately reflect the systematicness and organization of the deviation behavior, providing a more solid and reliable quantitative basis for distinguishing between organizational resilience and disorderly failure.

[0074] The technical superiority of this solution lies in the construction of a dual graph, namely an offline-built contingency plan knowledge graph and a real-time generated actual behavior graph, which provides a dynamic and comparable data foundation for risk assessment. This step unifies static, unstructured contingency plans and dynamic, multi-source heterogeneous field data under a structured graph data model, making it possible to quantitatively analyze the differences between the two.

[0075] To explore the intrinsic nature of this difference, the scheme introduces a cooperative deviation index; its physical meaning is not simply to measure the magnitude of individual deviations, but rather to quantify the cooperative strength of a group of interconnected units in their deviation behaviors; the index is calculated as follows:

[0076] ;

[0077] in, Co-deviation index : Contingency plan coordination coefficient Dynamic interaction weights : Multidimensional deviation vector Vector similarity function;

[0078] This formula can be fully calculated as described in the instruction manual; among which, the contingency coordination coefficient... With multidimensional deviation vector Obtained through analysis of the contingency plan knowledge graph and comparison of dual graphs; dynamic interactive weights. This is derived from the statistics and normalization of the interaction frequency between units in the actual behavior map; the formula uses the pre-plan coordination coefficient. The calculations are limited to units with collaborative requirements in the plan, and dynamic interaction weights are used. Amplify the impact of units with close actual interactions through vector similarity functions. Assess the consistency of these units' deviations from the correct direction; the values ​​remain positive and are trending upwards. The value represents the physical fact that a systematic and coordinated deviation has occurred in the system, rather than an isolated and random error.

[0079] The calculation of the response consistency score in S2 specifically includes:

[0080] The actual behavior graph is input into the graph neural network model to generate state embedding vectors for each response unit in the graph.

[0081] For response units that deviate, extract the set of neighboring nodes in the actual behavior graph;

[0082] Based on the state embedding vector of the off-unit, the state embedding vector of the neighboring node, and the learnable weight matrix in the graph neural network model, calculate the behavior alignment inner product value;

[0083] The behavior-aligned inner product value is normalized using an activation function to generate a response consistency score;

[0084] The learnable weight matrix in the graph neural network model is obtained through offline supervised learning using graph snapshots of historical training data and corresponding expert annotations.

[0085] In this embodiment, the calculation method of the response consistency score in S2 is specified; when the co-deviation index When a significant increase occurs, this calculation needs to be initiated. Its purpose is to determine, from the perspective of system response, whether the cooperative deviation is a benign adaptation with internal support or a malignant failure due to lack of cooperation; for any unit exhibiting a significant deviation... The formula for calculating the response consistency score is defined as follows:

[0086] ;

[0087] in, :unit The response consistency score is a dimensionless value. : Response units that have deviated significantly; :time; : Sigmoid activation function; :unit The set of neighboring nodes; :unit The state embedding vector; Learnable weight matrix; :unit The set of neighboring nodes One of the neighboring nodes; : Represents the transpose of a vector; : Represents the set of neighboring nodes The number of nodes in the middle;

[0088] The calculation process for this score is as follows:

[0089] actual behavior mapping The input is fed into a pre-trained graph neural network model to generate state embedding vectors for each response unit in the graph. The state embedding vector is a low-dimensional, dense real-number vector that encodes the unit's own attributes and its complex structural information in higher-order neighborhoods. It originates from the output of the graph neural network model.

[0090] In this embodiment, the graph neural network model can preferably be a graph attention network (GAT). The advantage of using GAT is that its core self-attention mechanism can dynamically assign different importance weights to neighboring nodes, which aligns perfectly with the present invention's aim to evaluate the interaction strength between different units. GAT effectively assigns different importance weights to neighboring nodes. Aggregate its neighbors Information to generate new state embedding vectors At that time, an attention coefficient will be calculated. :

[0091] ;

[0092] in, It is a learnable linear transformation matrix; Attention coefficient, representing the neighboring nodes For nodes Importance weights; : Response unit, and All The neighboring nodes; Exponential function; LeakyReLU activation function; A learnable weight vector; : indicates vector transpose; A learnable linear transformation matrix; : Represents vector concatenation operation; :unit The set of neighboring nodes; It is a learnable weight vector. Indicates vector concatenation; unit Updated state embedding vector It is a weighted sum of the features of all its neighboring nodes, and the weights are the learned attention coefficients. By stacking multiple GAT layers, the model can capture complex, high-order neighborhood structure information, thereby generating high-quality state embedding vectors. and This provides a solid foundation for the accurate calculation of subsequent response consistency scores;

[0093] For the response unit that deviates Extract its actual behavior map The set of neighbor nodes in The set of neighboring nodes is at time 1000. With nodes The set of all directly connected nodes serves to define the scope of direct associations for consistency assessment; its source is... Graph structure;

[0094] Based on deviation unit State embedding vector The state embedding vector of neighboring nodes and the learnable weight matrix in graph neural network models Calculate the behavior-aligned inner product; learnable weight matrix. This is a core parameter of the graph neural network model, defining which combinations of states are consistent in the feature space. As mentioned above, to ensure the accuracy of this matrix's judgment, it is obtained through offline supervised learning using graph snapshots of historical training data and corresponding expert annotations. Expert annotations are provided by domain experts who label behavioral segments in historical training as either consistent collaboration or inconsistent chaos. The model learns from these annotations and adjusts the matrix accordingly. The parameters are set such that when the behavior of two nodes is considered consistent by the expert, the embedding vector is... The transformed inner product is larger;

[0095] For example, the expert annotation and supervised learning process can be summarized in the following steps:

[0096] Constructing labeled samples: From snapshots of historical exercise data, experts select response units with direct interaction. As labeled samples; for each sample, experts assign a label based on the context of the exercise, the communication content, and the final result. ;in, Representative Unit and The interaction at this moment is an effective collaboration or a consistent response; This indicates that the interaction is invalid or chaotic;

[0097] Define the loss function: The goal of supervised learning is to make the alignment score output by the model fit the expert annotations for a pair of units. The behavior-aligned inner product value is ; You can use the Sigmoid function The binary cross-entropy loss function is used as the learning target; for a labeled sample The loss is:

[0098] ;

[0099] in, For a labeled sample Binary cross-entropy loss; There are direct interactive response unit pairs; The label is marked by experts. It is 1 when the interaction is effective collaboration or consistent response, and 0 otherwise. Logarithmic function; : Sigmoid function; Unit pair The behavior-aligned inner product value is denoted by . ;

[0100] Model training: by minimizing the total loss of all labeled samples Optimization algorithms such as gradient descent are used to iteratively update the weight matrix. The parameters; after sufficient training, the matrix This encodes the criteria for judging what constitutes consistent behavior in expert knowledge;

[0101] The calculated average behavior-aligned inner product value is then passed through an activation function, such as the Sigmoid function. Normalization is performed to generate a response consistency score with a value range between (0, 1). ;

[0102] This embodiment introduces a graph neural network-based computational method to deepen the evaluation of logical support from macroscopic phenomena to the microscopic feature space alignment level. By supervising the training of the graph neural network model, the consistency judgment criteria are derived from real data and expert experience, rather than subjective settings, which greatly enhances the objectivity and accuracy of the evaluation. This approach can capture nonlinear and high-order internal interaction patterns of the system, thus enabling a more reliable distinction between organizational resilience that has been widely recognized and supported by the system and disordered failure that is disconnected from other parts when cooperative deviation occurs.

[0103] Once a systematic coordinated deviation is identified, a response consistency score is introduced to further determine its nature. The score aims to assess whether the deviation behavior is logically supported by other units within the system; the calculation formula is:

[0104] ;

[0105] in, Response consistency score Activation function : Neighbor node set, : State embedding vector Learnable weight matrix;

[0106] The calculation of the formula relies on a pre-defined graph neural network model; the model works by encoding consistent or inconsistent chaotic behavior patterns labeled by experts into a learnable weight matrix through offline supervised learning of historical training data. During real-time evaluation, the model generates a state embedding vector for each unit that includes information about its own structure and that of its neighborhood. The formula calculates the embedding vectors of the offset cell and its neighboring cells via a matrix. The transformed inner product quantifies the degree of alignment of its behavior within the learned collaborative patterns; a high The value, in physical terms, indicates that the deviation behavior of a unit is highly consistent with the behavior of its surrounding collaborators in the deep feature space. This consistency aligns with effective collaborative patterns recognized by experts in historical data.

[0107] Example 2

[0108] The risk level determination in S3 specifically includes:

[0109] If the coordination deviation index is lower than or equal to the preset coordination deviation threshold, the system risk level is determined to be normal.

[0110] If the coordination deviation index is higher than the coordination deviation threshold, and the system's average response consistency score is higher than or equal to the preset response consistency threshold, then the system risk status is determined to be organizational resilience.

[0111] If the coordination deviation index is higher than the coordination deviation threshold and the system's average response consistency score is lower than the response consistency threshold, the system risk level is determined to be disorder failure, and an early warning is triggered.

[0112] This embodiment, based on Embodiment 1, specifies the risk level determination logic in S3. The core step is to establish a pre-defined determination rule model. The model's function is to map the continuous quantitative indicators output by the upstream module into discrete risk levels with clear guiding significance. The model is based on a series of IF-THEN logical rules. To ensure the effectiveness of the rules, threshold parameters included, such as the collaborative deviation threshold, are used. and response consistency threshold The source is the calibration and optimization achieved through statistical analysis of massive amounts of historical exercise data and combined with expert experience;

[0113] The logic for determining the risk level is as follows:

[0114] If the co-deviation index Lower than or equal to the preset collaborative deviation threshold If the system's behavior is basically in line with the plan or the degree of deviation is within an acceptable range, the system risk level is determined to be normal.

[0115] If the co-deviation index Higher than the collaborative deviation threshold When this indicates a significant and systematic deviation in the system's behavior, a response consistency score needs to be introduced for further analysis; and the system's average response consistency score... Higher than or equal to the preset response consistency threshold This indicates that the deviation behavior has broad logical support and effective coordination within the system, thus the system risk status is determined to be organizational resilience; this situation usually corresponds to more effective on-site decisions made by the training unit in response to emergencies.

[0116] If the co-deviation index Higher than the collaborative deviation threshold And the system's average response consistency score Below the response consistency threshold This indicates that the systemic deviation behavior lacks internal coordination and support, manifesting as chaos and individual actions. The system risk level is then determined to be disorderly failure, triggering an early warning and pushing high-level risk information to the command center.

[0117] This embodiment establishes a clear and automatically executable risk assessment logic through explicit and hierarchical IF-THEN rules. This approach transforms the complex dual-indicator assessment process into a simple and intuitive risk level output. It avoids the subjectivity and ambiguity that may occur when interpreting data manually, ensuring the timeliness and consistency of risk assessment results. This enables commanders to quickly and accurately grasp the core risks of the exercise and make corresponding decisions.

[0118] The model outputs a final conclusion through a pre-defined risk assessment rule model. The working principle of the model is based on a series of IF-THEN logical rules. The pre-defined collaborative deviation threshold and the pre-defined response consistency threshold in the rules are set by statistical analysis of historical exercise data and combined with expert experience. For example, the quantile covering 95% of the fluctuations in normal exercise data is taken as the threshold to ensure that only statistically significant changes in indicators trigger the transition of risk levels. This mechanism deterministically maps complex continuous values ​​of dual indicators to discrete risk levels such as normal, organizational resilience, and disordered failure, thereby achieving accurate risk assessment and early warning.

[0119] Example 3

[0120] It also includes the following steps:

[0121] After injecting a pre-set disturbance event during the exercise, the organizational resilience recovery time is calculated. The calculation of the organizational resilience recovery time includes: recording the occurrence time of the disturbance event and continuously monitoring the time change rate of the system-level coordination deviation index and response consistency score; when the absolute value of the time change rate falls back to the pre-set stability threshold, it is determined as the system recovery stability time; the time span between the occurrence time of the disturbance event and the system recovery stability time is the organizational resilience recovery time.

[0122] The preset stability threshold is determined by statistical analysis of the normal volatility of system-level indicators in historical stability exercise data;

[0123] This embodiment, based on Embodiment 1, adds a step for directly measuring the system's dynamic recovery capability: after injecting a preset disturbance event during the exercise, the tissue resilience recovery time is calculated; tissue resilience recovery time is a macroscopic evaluation indicator, aiming to quantify the time required for the system to recover to a stable state after suffering an external shock. The shorter the time, the stronger the system resilience; the calculation formula is defined as:

[0124] ;

[0125] in, : Tissue resilience recovery time, its dimension is time; The time when the disturbance event occurred; The system returned to stable condition at that moment;

[0126] The time calculation process is as follows:

[0127] Record the time of occurrence of the disturbance event. Disturbance events are pre-set contingency scenarios in drills designed to test the system's response capabilities. Examples include simulating the failure of critical communication facilities or adding a secondary disaster point to the drill scenario. Determined by specific event markers in the real-time state stream;

[0128] After event injection, the system continuously monitors the system-level coordination deviation index. The system average of the response consistency score And calculate the absolute value of their rate of change over time, i.e. and ;

[0129] The system will determine whether the system has returned to stability; system stability is defined as the absolute values ​​of the time-varying rates of two core indicators simultaneously falling back to preset stability thresholds over a certain period of time. ; at the absolute value of the rate of change over time, i.e. and When the system returns to a preset stable threshold, the start time of the stable state is determined as the time when the system recovers stability. ;

[0130] To ensure the objectivity of the judgment, a preset stability threshold is used. The determination of the threshold is based on clear criteria; its function is to distinguish between normal, minor fluctuations in the system and drastic changes after experiencing disturbances; the threshold is determined by analyzing system-level indicators in historical stability training data. and The threshold is determined through statistical analysis of the normal volatility; for example, the 95th percentile of the absolute value of historical normal volatility can be used as a threshold to ensure that stability is only determined when the system truly stabilizes. Since this threshold is used for comparison with the rate of change over time, its dimension is the reciprocal of time. ;

[0131] When the disturbance event occurs When the system returns to stability The time span is the tissue resilience recovery time. ;

[0132] This embodiment introduces the calculation of tissue resilience recovery time, providing a new, quantifiable, macro-level dimension for assessing the dynamic resilience of emergency systems; it goes beyond assessing risk status at a specific point in time, focusing instead on the entire dynamic recovery process of the system after experiencing an impact; and further, it utilizes a stability threshold... The definition of the source ensures the scientific nature and objectivity of the macro-assessment, making the resilience assessment results comparable under different exercise scenarios, and providing direct data support for the revision of emergency plans and the adjustment of personnel training priorities.

[0133] Example 4

[0134] An adaptive risk assessment system for disaster prevention and mitigation emergency drills includes the following modules:

[0135] The data processing module is used to parse emergency drill plans to build a plan knowledge graph and to integrate multi-source heterogeneous data in real time to generate a structured state flow.

[0136] The dynamic modeling module is used to construct a dynamic behavioral graph that evolves over time based on the structured state flow.

[0137] The deviation calculation module is used to compare the contingency plan knowledge graph with the actual behavior graph to calculate the collaborative deviation index;

[0138] The consistency assessment module is used to calculate the response consistency score based on a graph neural network model when the coordination deviation index is higher than the preset coordination threshold.

[0139] The risk decision-making module is used to comprehensively analyze the collaborative deviation index and response consistency score, and determine the risk level and output early warning based on preset judgment rules;

[0140] This invention provides an adaptive risk assessment system for disaster prevention and mitigation emergency drills. The system is configured to execute any of the aforementioned methods; physically, the system can be deployed on a server in an emergency command center or a cloud computing platform; logically, it includes mutually cooperating functional modules.

[0141] The data processing module, serving as the system's data entry point and preprocessing center, is configured to perform two functions: in offline mode, it is used to parse emergency drill plans to construct a plan knowledge graph; during online drills, it is used to fuse multi-source heterogeneous data in real time to generate a structured state stream.

[0142] The dynamic modeling module receives a structured state stream generated by the data processing module; it is configured to construct a dynamic behavioral graph that evolves over time based on the structured state stream. This enables high-fidelity modeling of the actual exercise process;

[0143] The deviation calculation module, serving as the starting point for quantitative evaluation, receives a contingency plan knowledge graph and an actual behavior graph. It is configured to compare the contingency plan knowledge graph with the actual behavior graph and, based on a given method, calculates a collaborative deviation index. ;

[0144] The consistency assessment module is triggered after the deviation calculation module. It is configured to initiate the assessment when the consistency deviation index exceeds a preset consistency threshold. The module integrates a pre-trained graph neural network model and calculates a response consistency score based on the graph neural network model according to the methodology. ;

[0145] The risk decision-making module, as the system's decision output unit, receives quantitative indicators generated by the deviation calculation module and the consistency assessment module. It is configured to comprehensively analyze the collaborative deviation index and the response consistency score, and, based on preset judgment rules, determine the risk level and output an early warning.

[0146] In addition, the system may also include a resilience measurement module, which, upon receiving a signal from a disturbance event injection, calls the outputs of the deviation calculation module and the consistency assessment module to calculate the tissue resilience recovery time.

[0147] This embodiment provides a complete and engineerable system architecture through a clear division of functional modules. Each module has a clear responsibility, a well-defined data flow, and works collaboratively, transforming the abstract assessment method into a concrete and operable system. The system can automatically complete the entire process from data access, dynamic modeling, quantitative assessment to decision-making and early warning, significantly improving the efficiency and intelligence level of emergency drill risk assessment, and providing powerful technical tools and quantitative basis for achieving scientific, accurate, and real-time monitoring and assessment of the drill process.

[0148] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. 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 be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. An adaptive risk assessment method for disaster prevention and mitigation emergency drill, characterized in that, The method comprises the following steps: S1, constructing a pre-plan knowledge graph and generating an actual behavior graph; the construction of the pre-plan knowledge graph comprises: analyzing the emergency drill pre-plan, extracting the response units and task nodes defined in the pre-plan as pre-plan entity nodes, extracting the command chain and task dependency relationship between entities as pre-plan relationship edges, and generating a pre-plan knowledge graph as a static evaluation benchmark; the generation of the actual behavior graph comprises: real-time fusion and analysis of multi-source heterogeneous data in the drill process to form a structured state flow with unified time and space benchmarks, and constructing an actual behavior graph that dynamically evolves over time based on the structured state flow; S2, adaptive deviation quantitative evaluation; adaptive deviation quantitative evaluation comprises: receiving the pre-plan knowledge graph and the actual behavior graph as input, calculating the collaborative deviation index to quantify the correlation strength of multiple response units deviating from the pre-plan behavior, and starting the calculation of the response consistency score when the collaborative deviation index is higher than the preset collaborative threshold to evaluate whether the deviating behavior obtains logical support from other units in the system; S3, determining the risk level and outputting the warning; the determination of the risk level comprises: comprehensively considering the collaborative deviation index and the response consistency score, determining the risk level of the current drill state according to the preset determination rule model, and outputting the warning information; The calculation of the collaborative deviation index in S2 specifically comprises: determining the pre-plan collaborative coefficient; the pre-plan collaborative coefficient is determined according to the pre-plan knowledge graph; if any two response units in the pre-plan are defined to have a direct collaborative relationship, the pre-plan collaborative coefficient is 1, otherwise it is 0; calculating the dynamic interaction weight; the dynamic interaction weight is obtained by normalizing the actual interaction strength between units in the actual behavior graph; obtaining a multi-dimensional deviation vector; the multi-dimensional deviation vector is calculated by comparing the attribute difference of the corresponding units in the actual behavior graph and the pre-plan knowledge graph; based on the vector similarity between the pre-plan collaborative coefficient, the dynamic interaction weight and the multi-dimensional deviation vector, the collaborative deviation index is calculated and outputted; The calculation of the response consistency score in S2 specifically comprises: inputting the actual behavior graph into the graph neural network model to generate state embedding vectors for each response unit in the graph; for the response unit that deviates, extract the neighbor node set in the actual behavior graph; based on the state embedding vector of the deviating unit, the state embedding vector of the neighbor node and the learnable weight matrix in the graph neural network model, calculate the behavior alignment inner product value; normalizing the behavior alignment inner product value through the activation function to generate the response consistency score. 2.The disaster prevention and mitigation emergency drill oriented adaptive risk assessment method according to claim 1, characterized in that, The learnable weight matrix in the graph neural network model is obtained by using the graph snapshots of the historical drill data and the corresponding expert annotations for offline supervised learning. 3.The disaster prevention and mitigation emergency drill oriented adaptive risk assessment method according to claim 1, characterized in that, The determination of the risk level in S3 specifically comprises: if the collaborative deviation index is lower than or equal to the preset collaborative deviation threshold, the system risk level is determined to be normal; if the collaborative deviation index is higher than the collaborative deviation threshold, and the average response consistency score of the system is higher than or equal to the preset response consistency threshold, the system risk level is determined to be organizational resilience. If the synergy deviation index is higher than the synergy deviation threshold value, and the response consistency score of the system is lower than the response consistency threshold value, it is determined that the system risk level is disordered failure, and a warning is triggered. 4.The disaster prevention and mitigation emergency drill oriented adaptive risk assessment method according to claim 3, characterized in that, Further comprising the following steps: After injecting a preset disturbance event in the drill, calculate the organizational resilience recovery time; The calculation of the organizational resilience recovery time includes: recording the time of the occurrence of the disturbance event, and continuously monitoring the time rate of change of the synergy deviation index and the response consistency score at the system level; When the absolute value of the time rate of change falls below the preset stability threshold, the system recovery stability time is determined; the time span between the time of the occurrence of the disturbance event and the system recovery stability time is the organizational resilience recovery time. 5.The disaster prevention and mitigation emergency drill oriented adaptive risk assessment method according to claim 4, characterized in that, The preset stability threshold is determined by statistical analysis of the normal fluctuation rate of system-level indicators in historical stable drill data.

6. An adaptive risk assessment system for disaster prevention and mitigation emergency drill, applied to the adaptive risk assessment method for disaster prevention and mitigation emergency drill in any one of claims 1 to 5, characterized in that, Comprising the following modules: A data processing module for parsing the emergency drill plan to build a plan knowledge graph, and fusing real-time multi-source heterogeneous data to generate a structured state flow; A dynamic modeling module for constructing an actual behavior graph that evolves dynamically over time based on the structured state flow; A deviation calculation module for comparing the plan knowledge graph with the actual behavior graph to calculate the synergy deviation index; A consistency evaluation module for calculating the response consistency score based on a graph neural network model when the synergy deviation index is higher than the preset synergy threshold; A risk decision module for integrating the synergy deviation index and the response consistency score, and determining the risk level and outputting a warning according to the preset determination rule.

Images