A hub capability platform data collaborative processing method for urban emergency

Abstract

The application discloses a kind of city emergency-oriented hub ability platform data collaborative processing method, this method includes: based on adaptive space-time slice and semantic knowledge graph dynamic data access and alignment;Based on CP-ABE and quantitative correlation analysis task-driven data desensitization and routing;Based on the cloud edge network self-organizing collaboration of improved Raft election algorithm and hash chain check;Based on the digital twin deduction and backtracking of LSTM prediction model and multi-agent simulation.This application realizes the real-time standardized processing of multi-source heterogeneous emergency data, reduces the transmission delay of key instructions, makes the system have the ability of network survival, reduces the response delay caused by manual intervention.

Description

Technical Field

[0001] This invention belongs to the field of data processing technology, and in particular relates to a data collaborative processing method for a central capability platform for urban emergency response. Background Technology

[0002] With the deepening of smart city construction, urban emergency management, as a key area for ensuring public safety, directly impacts a city's resilience and sustainable development through its level of informatization and intelligence. Currently, urban emergency management is characterized by diversified types of emergencies, expanded impact scope, and increasingly demanding timeliness and precision in response. Traditional models relying on manual dispatch and single-point responses are no longer sufficient to meet the needs of modern urban emergency management. Especially when facing major emergencies such as floods, fires, and hazardous chemical leaks, it is necessary to efficiently integrate sensor data, decision-making instructions, and dispatch resources scattered across different departments. This places higher demands on data processing capabilities, cross-departmental collaboration efficiency, and real-time response speed.

[0003] However, existing technological solutions still have significant shortcomings in practical applications. Existing technologies largely focus on the perception and post-disaster analysis of specific disasters, making it difficult to support real-time emergency command and resource scheduling during an event. Furthermore, they lack unified data semantics and access control mechanisms for business collaboration, leading to "data silos" in cross-departmental data interaction. While some solutions utilize large models for multimodal management, their core lies in the allocation of computing resources and load prediction, failing to address the information silo problem at the emergency business logic level. In addition, existing technologies lack the self-organizing capabilities of edge nodes under extreme communication conditions. The traditional tree-structured hierarchical reporting model will lead to a break in the collaborative link if the front-end node loses connection, and the response speed remains insufficient for extremely rapid disasters such as explosions and toxic gas leaks. Simultaneously, existing research largely remains at the theoretical framework level, failing to propose specific data structures and algorithms to automatically resolve cross-organizational data conflicts and priority adjudication issues, resulting in the persistent contradiction of departmental silos and fragmented information exchange.

[0004] Therefore, addressing the shortcomings of the existing technologies has become the focus of efforts for those skilled in the field. Summary of the Invention

[0005] The purpose of this invention is to provide a data collaborative processing method for a central capability platform for urban emergency response, which can completely solve the shortcomings of the existing technology.

[0006] The objective of this invention is achieved through the following technical solution:

[0007] A data collaborative processing method for a central capability platform for urban emergency response includes the following steps:

[0008] Step 1: Based on adaptive spatiotemporal slicing and semantic alignment, a dynamic data access gateway is constructed to parse various types of emergency data and encapsulate them into unified data objects. Real-time semantic alignment and conflict resolution of multi-source heterogeneous data are achieved by constructing an emergency semantic knowledge graph containing an entity mapping relationship library and an attribute association matrix. The size of the spatiotemporal slice window is dynamically determined using an adaptive algorithm based on disaster evolution rate assessment.

[0009] Step 2: Task-driven dynamic data desensitization and routing. Using a task-driven data extrapolation mechanism, after the emergency command center generates an emergency task, the task is automatically decomposed by a reinforcement learning model based on a hierarchical task network. The minimum dataset required to execute the task is determined based on the correlation quantification calculation method. The encrypted policy attribute base encryption algorithm is used to automatically remove classified information and push the processed data to specific departments.

[0010] Furthermore, this method also defines a smooth switching mechanism between normal mode and offline mode:

[0011] When the system is in a normal cloud connection state, the complete task-driven data deduction mechanism described in step 2 is executed, that is, the emergency command center performs task decomposition, minimum dataset calculation and attribute-based encrypted routing.

[0012] When an edge node detects a connection interruption with the cloud, the system automatically switches from "centralized mode" to "edge autonomous mode." In edge autonomous mode, the edge self-organizing network no longer relies on a cloud-based central hub for task decomposition. Instead, it performs simplified task breakdown and data routing based on a locally pre-built lightweight task template library.

[0013] Task decomposition: The edge temporary master node selects a matching template from commonly used task templates cached locally (such as "gas leak evacuation" and "fire rescue"), and no longer runs the reinforcement learning model;

[0014] Data requirements analysis: A fixed relevance threshold and a preset minimum dataset list are used, without dynamic quantitative calculations;

[0015] Data routing: CP-ABE encryption is not performed; plaintext or simple encryption is used for transmission based only on the department identifier pre-configured on the node.

[0016] Once the network is restored, the system automatically switches back to the central drive mode. Edge nodes upload operation logs and hash chains from the period of network outage to the cloud for integrity verification and post-event archiving. The switching latency between the two modes is controlled within 100 milliseconds to ensure uninterrupted emergency response.

[0017] Step 3: Cloud-edge self-organizing data collaboration method when the edge node detects an interruption in connection with the cloud. When the edge node detects an interruption in connection with the cloud, it triggers the edge self-organizing network mode. Through an election algorithm based on the improved Raft protocol, a temporary master node is selected by comprehensively considering the remaining computing resources, communication bandwidth and geographical advantages. The nodes share disaster data using short-distance communication. After the network is restored, the data integrity is verified by using a hash chain based on SHA-256 and incrementally synchronized to the cloud hub.

[0018] Step 4: Data extrapolation and retrospection based on digital twins. Based on a unified data mirror, an extrapolation model is built in virtual space. Decision-makers use a simulation engine based on the multi-agent model (ABM) to conduct sand table extrapolation to predict the development of the situation, and use a time series prediction model based on long short-term memory network (LSTM) to predict the potential results of different decision options. All data operations have timestamps synchronized with a global time server to support post-event auditing and accountability.

[0019] Furthermore, the adaptive algorithm described in step 1 obtains the average rate of change by calculating the first derivative of the key indicators. and according to the function Dynamically adjust the window size, where As the reference window, This serves as a reference rate of change threshold.

[0020] Furthermore, the emergency semantic knowledge graph mentioned in step 1 is constructed using a hybrid method combining rules and statistics; the conflict resolution is based on the authority priority of the data source, timestamps, and consistency rules based on RDF triples for arbitration.

[0021] Furthermore, in the relevance quantification calculation method described in step 2, for data object D and subtask T, the relevance score Rel(D,T) is composed of a weighted sum of semantic similarity, spatial similarity and timeliness matching degree, and only data objects with scores greater than a preset threshold are retained to form the minimum dataset.

[0022] Furthermore, the comprehensive scoring formula for the election algorithm described in step 3 is as follows: ,in, This represents the current CPU load rate. This represents the currently available communication bandwidth. This is the theoretical maximum bandwidth; The distance between the candidate node and the center of the disaster area is the Euclidean distance. The priority level is preset for the node; α, β, γ, and δ are weight coefficients.

[0023] Furthermore, the structure of each data operation block in the hash chain verification described in step 3 is as follows:<Index,Timestamp, DataHash, PrevBlockHash, Nonce, CurrentBlockHash> Integrity verification is achieved by recursively verifying the continuity of hash values.

[0024] Furthermore, the LSTM-based time-series prediction model described in step 4 takes historical situation feature sequences and decision scheme vectors as inputs and outputs future predicted situation sequences and final result vectors. The simulation engine based on the multi-agent model includes vehicle agents, personnel agents, and resource agents, and adopts discrete event simulation and time step advancement mechanisms.

[0025] A data collaborative processing system for a central capability platform for urban emergency response, used to execute the method described in any of the above, includes a global perception layer, an intelligent central capability layer, and an edge collaboration layer;

[0026] The global perception layer is equipped with a dynamic data access gateway, which is used to receive multi-source heterogeneous data streams from IoT devices, video surveillance systems, voice communication systems and geographic information systems, and send the parsed and encapsulated unified data object to the intelligent central capability layer.

[0027] The intelligent central capability layer is communicatively connected to the global perception layer. The intelligent central capability layer includes a task decomposition unit, a data requirement analysis unit, a dynamic routing unit, and a digital twin module. The task decomposition unit is used to break down emergency tasks into several sub-tasks and determine the dependencies between them. The data requirement analysis unit, connected to the task decomposition unit, is used to calculate the minimum dataset required to execute each sub-task based on the dependencies. The dynamic routing unit, connected to the data requirement analysis unit, is used to determine the data push path based on the minimum dataset. The digital twin module is used to construct a digital twin model based on a unified data object for sand table simulation.

[0028] The edge collaboration layer is communicatively connected to the intelligent central capability layer. The edge collaboration layer includes multiple edge nodes, each of which has a status monitoring module, a self-organizing network communication module, and an incremental synchronization module. The edge nodes have bidirectional communication links with the cloud central hub, and the edge nodes have short-distance communication links with each other.

[0029] Furthermore, it also includes a data quality monitoring module and a cross-departmental data conflict resolution module; the data quality monitoring module is used to monitor the integrity, accuracy and timeliness of the accessed data; the cross-departmental data conflict resolution module is used to handle semantic conflicts and permission conflicts between multiple departments based on priority rules, timestamp rules and permission level rules.

[0030] Furthermore, the preset thresholds involved in the system, including the communication delay threshold and packet loss rate threshold, are set and dynamically updated using an adaptive threshold determination method based on historical statistical distribution. The adaptive threshold determination method involves: establishing a Gaussian distribution model based on performance data collected during normal system operation, calculating the mean μ and standard deviation σ of the performance data; and then applying the formula... Calculate the dynamic threshold, where Th represents the dynamic threshold at the current moment, μ represents the mean of the performance data, σ represents the standard deviation of the performance data, and k is the alarm coefficient, with a value of 3.

[0031] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0032] 1. Through the adaptive spatiotemporal slicing and semantic alignment technology in step 1, real-time standardized processing of multi-source heterogeneous emergency data was achieved, effectively solving the problems of data heterogeneity and semantic conflict;

[0033] 2. By using the task-driven data dynamic desensitization and routing technology in step 2, the transmission delay of critical instructions is significantly reduced, effectively solving the problem of data flooding bandwidth.

[0034] 3. By using the cloud-edge self-organizing data collaboration method in step 3, the system is made capable of surviving network outages, effectively solving the collaboration problem under extreme communication conditions.

[0035] 4. Through the digital twin technology based on LSTM and multi-agent simulation in step 4, the fully automated data flow from perception to scheduling is realized, effectively reducing the response delay caused by manual intervention;

[0036] This invention achieves a balance between high robustness, high timeliness, and data security and compliance through a layered and decoupled architecture design. Attached Figure Description

[0037] Figure 1 This is a schematic diagram of the overall technical solution architecture of the present invention;

[0038] Figure 2 This is a schematic diagram of the core principle framework of the task-driven data dynamic desensitization and routing mechanism in this invention;

[0039] Figure 3 This is a logical flowchart of the dynamic data access and semantic alignment based on adaptive spatiotemporal slicing in this invention.

[0040] Figure 4 This is a schematic diagram of the multi-level interaction relationship and data flow of the cloud-edge disconnection self-organizing data collaboration method in this invention;

[0041] Figure 5This is a logical flowchart of the data extrapolation and backtracking based on digital twins in this 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] See Figures 1 to 5 This embodiment provides a complete technical implementation scheme for a data collaborative processing method for a central capability platform in urban emergency scenarios. The method adopts a layered and decoupled system architecture, divided into a global perception layer, an intelligent central capability layer, and an edge collaboration layer. The global perception layer is responsible for the real-time access of multi-source heterogeneous data; the intelligent central capability layer undertakes the core functions of data processing, task decomposition, and resource scheduling; and the edge collaboration layer is deployed at various emergency site nodes, undertaking the key tasks of local data processing and self-organizing collaboration during network outages. The entire system achieves fully automated data flow from data perception to scheduling execution through a four-step core processing flow.

[0045] Step 1: Dynamic Data Access and Semantic Alignment Based on Adaptive Spatiotemporal Slicing

[0046] The implementation of Step 1 first requires the construction of a dynamic data access gateway. This gateway, as a core component of the global perception layer, is deployed at the data entry point and is responsible for receiving heterogeneous data streams from multiple sources, including IoT devices, video surveillance systems, voice communication systems, and geographic information systems. The dynamic data access gateway integrates multiple data parsing adapters, each performing standardized parsing processing for a specific type of data source. Specifically, the IoT device data adapter parses numerical data reported by various sensors, the video surveillance system adapter extracts keyframes and metadata from the video stream, the voice communication system adapter converts voice signals into structured text, and the geographic information system adapter processes vector and raster data. All adapters output intermediate data objects in a unified format to ensure consistency in subsequent processing.

[0047] After data parsing, the system introduces an emergency semantic knowledge graph to achieve real-time semantic alignment and conflict resolution of multi-source heterogeneous data. The core data structure of the emergency semantic knowledge graph includes two parts: an entity mapping relation library and an attribute association matrix.

[0048] The entity mapping database is constructed using a hybrid approach combining rules and statistics. The rule layer predefines metadata pattern matching rules for each data source system, implemented through a pre-defined configuration table [Source System Identifier, Source Field Name, Target Entity Type, Target Entity ID, Mapping Confidence]. For new data sources without pre-defined rules, the statistics layer employs entity alignment algorithms based on string similarity (e.g., edit distance, Jaro-Winkler distance) and spatial distance (latitude and longitude Euclidean distance). If the similarity between the source identifier description text and the target entity name is greater than 0.85, or the geographical distance between them is less than 50 meters, a candidate mapping relationship is established and added to the mapping database after confirmation by the system administrator. An automatic update mechanism is implemented: every 1000 new data entries received by the data access gateway, incremental learning is triggered, using the DBSCAN clustering algorithm to cluster identifiers in the new data source, automatically discovering new candidate mapping pairs.

[0049] The attribute association matrix A is an n×n matrix, where n is the number of data attribute types. Matrix elements This represents the semantic association strength between attribute i and attribute j, and its value range is... The initial values ​​are set based on an expert rule base. For example, the correlation strength between "rainfall" and "flood risk level" is preset to 0.9, and the correlation strength between "gas concentration" and "explosion risk level" is preset to 0.95.

[0050] Dynamic learning: The system records the co-occurrence frequency of data in actual emergency tasks. If attributes i and j are simultaneously accessed or queried frequently within the same spatiotemporal slice, dynamic enhancement is performed according to the following formula:

[0051] ;

[0052] in:

[0053] The updated association strength is automatically truncated to the [0,1] interval after calculation;

[0054] The correlation strength before the update;

[0055] Λ=0.01 is the learning rate, which controls the step size of each update;

[0056] For co-occurrence frequencies, This represents the number of times attributes i and j are accessed simultaneously within the most recent K=1000 spacetime slices. This represents the total number of queries for all attributes during the same period.

[0057] The unified data object is the key output of step 1, and its data structure design includes six core fields. The first field is the data identifier, using a globally unique UUID encoding to uniquely identify each data object within the system. The second field is the timestamp, recording the precise time of data generation or access, with millisecond-level precision. The third field is the spatial coordinates, expressed using a unified geographic coordinate system (WGS84), including longitude, latitude, and altitude information. The fourth field is the data type, identifying the category to which the data belongs, such as video data, IoT sensor data, voice data, geographic information data, etc. The fifth field is the semantic tag, generated in real-time based on the emergency semantic knowledge graph, with tag content including the entity type corresponding to the data, the associated disaster type, and the corresponding emergency response level. The sixth field is the priority field, dynamically calculated and determined based on the severity and scope of the disaster, with a value ranging from 0 to 100, where a higher value indicates a higher priority.

[0058] The spatiotemporal slicing mechanism uses an adaptive algorithm to dynamically determine the window size. The specific steps are as follows:

[0059] Step 1.1: Extract features from the real-time data stream and calculate the time-series change rate of key indicators. For the gas leak scenario, the key indicator is the first derivative of the gas concentration value. For flood scenarios, the key indicator is the rate of change of water level over time. For earthquake scenarios, the key indicator is peak ground acceleration. .

[0060] Step 1.2: Initial Window Size Set to the system default of 10 seconds. In the next time window... Within this window, calculate the average rate of change of key indicators. ,in Let N be the index value of the i-th sampling point, and N be the number of sampling points within the window.

[0061] Step 1.3: Based on the average rate of change The size of the next time window is dynamically adjusted. The specific adjustment rules are as follows: > When a scenario is identified as rapidly evolving, the time window should be narrowed to capture rapid data changes. The next window size should be adjusted accordingly. ;when ≦ When the scenario is determined to be a slow-evolutionary scenario, the time window should be expanded to reduce computational overhead. The next window size... .in Used as the base time window (default 10 seconds). This serves as a reference rate of change threshold. To prevent the window from being too small or too large, a minimum window size is set. Seconds, maximum window Seconds. For gas leak scenarios, Set to 0.5; for flood scenarios, Set to 0.1. The system calculates in real time. and The ratio automatically adjusts the window size to achieve adaptive partitioning.

[0062] Conflict resolution algorithm: When two data sources provide conflicting values ​​for the same property of the same entity, the following arbitration process is executed:

[0063] First, compare the predefined authority levels of the data sources (e.g., "Meteorological Bureau Professional Stations" is higher than "Community Volunteer Reports").

[0064] Second, if the priorities are the same, the data updated with the timestamp will be adopted;

[0065] Third, conflict detection based on RDF triples is introduced. If new data contradicts more than 50% of the related facts in the knowledge graph, manual review is triggered.

[0066] Step 2: Task-driven dynamic data desensitization and routing

[0067] Step 2 changes the traditional broadcast-style distribution model, adopting a task-driven data extrapolation mechanism to achieve precise data delivery. The core logic of this mechanism is that after the emergency command center generates an emergency task, the system automatically breaks down the task and calculates the minimum dataset required to execute it, pushing only this minimum dataset to the relevant implementing departments, rather than broadcasting all data to all departments. This mechanism contains three core functional units: a task decomposition unit, a data requirement analysis unit, and a dynamic routing unit.

[0068] (a) Task decomposition unit

[0069] The task decomposition unit receives emergency task instructions from the emergency command center, parses the instructions into several sub-tasks, and determines the execution order and dependencies between the sub-tasks. For example, an earthquake emergency rescue task can be decomposed into sub-tasks of rubble search, medical rescue, traffic control, and material allocation, with each sub-task having a sequence and resource dependencies.

[0070] The task decomposition unit automatically decomposes tasks based on a pre-defined task template library and a dynamic learning algorithm. The task template library uses a quintuple data structure.<TaskType,Preconditions,Subtasks,Constraints,Q-value> Here, TaskType is an enumeration of task types, such as "earthquake rescue" or "gas leak response"; Preconditions are the conditions that trigger the task, such as "magnitude > 5"; Subtasks is an ordered list of subtasks, such as... Constraints are resource constraints, such as "≥ 2 excavators are required"; Q-value is the historical cumulative return value of this decomposition strategy.

[0071] The dynamic learning algorithm employs a reinforcement learning model based on a hierarchical task network (HTN). The state space is the feature vector of the current emergency task, including disaster type (one-hot encoding), impact range (numerical), available resource quantity (numerical), and response urgency (numerical), with a total dimension of 50. The action space is a decomposition template selected from a task template library, which initially contains 50 standard task decomposition templates for common emergency scenarios.

[0072] The reward function is used to evaluate the merits of task decomposition strategies and provides feedback signals to the reinforcement learning model. The formula for calculating the reward value is: ,in:

[0073] Reward is the reward value, which ranges from (−∞,0]. The larger the value (the closer to 0), the better the decomposition strategy.

[0074] The actual time to complete the task (in seconds);

[0075] The expected completion time for the task (in seconds) is set based on historical experience values ​​in the task template library.

[0076] The actual amount of resources consumed by the task (unit: standardized resource unit).

[0077] The amount of resources expected to be consumed by the task (unit: standardized resource unit).

[0078] =0.6 is the weighting coefficient for time cost;

[0079] =0.4 is the weighting coefficient for resource consumption costs, which satisfies... .

[0080] The negative sign in the formula indicates that when the actual completion time exceeds the expected time or the actual resource consumption exceeds the expected consumption, the reward value is negative, penalizing the decomposition strategy; when the actual value is better than the expected value, the reward value approaches 0. The larger the reward value (i.e., the less negative it is), the better the task decomposition strategy. The system updates the Q-network parameters of the task decomposition unit based on the reward value to optimize the decomposition strategy for subsequent tasks.

[0081] The learning algorithm uses DQN (Deep Q-Network), a three-layer fully connected network with a 50-dimensional input layer, 128 neurons in the first hidden layer, 64 neurons in the second hidden layer, and a 50-dimensional output layer. The activation function used is ReLU. Training parameters include: an experience replay pool size of M = 10000, and the target network is updated every 100 steps. Exploration Strategy (Initial) =0.1, linearly decaying to 0.01), discount factor =0.95, optimizer uses Adam, learning rate =0.001.

[0082] (II) Data Requirements Analysis Unit

[0083] The data requirements analysis unit calculates the minimum dataset required to execute each subtask based on the subtask data dependencies output by the task decomposition unit. This calculation process follows three screening strategies: data timeliness assessment, data relevance screening, and data volume optimization.

[0084] Data timeliness assessment: Based on the execution time window requirements of the sub-tasks, data within the effective time window is selected. For real-time command sub-tasks, the effective time window is set to 5 seconds; for resource scheduling sub-tasks, the effective time window is set to 30 seconds; and for disaster analysis sub-tasks, the effective time window is set to 2 minutes.

[0085] Data Relevance Screening: A data relevance quantification method based on information gain and semantic distance is introduced. For a data object D and a subtask T, its relevance score Rel(D, T) is composed of a weighted sum of three parts:

[0086] ;

[0087] in:

[0088] Rel(D, T) is the correlation score between data object D and subtask T, with a value range of [0,1].

[0089] D is the data object to be evaluated, containing location information D.loc and a timestamp. Semantic label vectors Attributes;

[0090] T represents the subtask to be executed, containing the task center location T.loc and the expected time reference point. Maximum allowable delay Attributes such as radius of influence R;

[0091] For semantic similarity, it is calculated based on the cosine similarity of semantic label vectors;

[0092] Spatial similarity is calculated based on the exponential decay function;

[0093] The timeliness matching degree is calculated based on the ratio of time difference to maximum delay;

[0094] =0.5, =0.3, =0.2 is the weighting coefficient, which satisfies .

[0095] semantic similarity ,in, The semantic label vector for data object D is one-hot encoded and has a dimension of 87. The requirement semantic vector for subtask T is generated based on the association matrix between task type and entity type.

[0096] Spatial similarity ,in, R is the Euclidean distance (in meters) between the location of the data object and the location of the task center, and R = 500 meters is the radius of influence of the task.

[0097] Timeliness matching ,in, For the timestamp of the data object; This serves as the expected data time reference point for the task. This represents the maximum allowed latency for the task. When the time deviation of the data exceeds the maximum delay, the data has expired and the match rate is 0. When the time deviation of the data does not exceed the maximum time delay, the matching degree decreases linearly as the deviation increases, and the smaller the deviation, the higher the matching degree.

[0098] Set correlation threshold =0.6, filtering out data objects with a relevance score not lower than this threshold to form a candidate dataset. ;

[0099] in, The candidate dataset is a collection of data objects that have been filtered for relevance. The data objects in this collection will enter the subsequent data volume optimization step, and the final output is the "minimum dataset". The data object to be evaluated; This is the subtask currently being processed; The correlation score between data object D and subtask T, with a value range of [0,1]; The correlation threshold is set to a value of [value to be filled in]. =0.6, when When a data object is considered to be significantly related to a subtask, it is retained; when Some were deemed to have insufficient relevance and were therefore excluded.

[0100] Then, for the candidate dataset Data volume optimization: Based on spatial grid (10m×10m) and hash value, deduplication and compression are performed, and the final output is the "minimum dataset" required for the current subtask.

[0101] (III) Attribute-based encryption and dynamic routing

[0102] The attribute-based encryption algorithm specifically employs the Ciphertext-Policy Attribute-Based Encryption (CP-ABE) scheme, whose mathematical foundation is the bilinear mapping e: ,in and It is a multiplicative cyclic group of prime order p. The specific implementation process is as follows:

[0103] System initialization (Setup): Run by the key authorization center, which inputs security parameters and generates the master key MSK and the system public key PK.

[0104] Key Generation (KeyGen): Generate a private key SK for each department (e.g., fire department, traffic police, medical). The private key is bound to the department's attribute set S, S = {"Fire", "On-site Commander", "Confidentiality Level: Confidential"}, using the following algorithm: .

[0105] Data encryption: Performed by the sender (cloud hub). First, define an access policy tree T, for example ('Fire' AND 'Field Commander') OR ('Medical' AND 'Confidentiality Level: Confidential'). Then, encrypt the data M using the public key PK and the policy tree T to generate ciphertext CT. The encryption process is as follows: .

[0106] Data Decryption: The receiver (edge ​​node or departmental system) decrypts the ciphertext CT using its own private key SK. Decryption is successful if and only if the receiver's attribute set S satisfies the access policy tree T embedded in the ciphertext. The output for a failed decryption is ⊥.

[0107] The dynamic routing unit determines the final push path for each data object based on the calculation results of the data demand analysis unit and the processing results of the attribute-based encryption algorithm. It comprehensively considers the network reachability of the target department, the real-time requirements of data transmission, and the network bandwidth load, and supports a multi-path parallel transmission mechanism.

[0108] Step 3: A self-organizing data collaboration method for cloud-edge network outages

[0109] Step 3 aims to solve the collaboration problem in extreme environments. When the edge node detects that the connection with the cloud is interrupted, it triggers the edge self-organizing network mode to achieve local collaboration.

[0110] The triggering conditions for edge self-organizing network mode include two detection dimensions. The first dimension is network connectivity status detection. The system monitors the network connectivity status between edge nodes and the cloud hub in real time. When a physical link interruption is detected, self-organizing network mode is immediately triggered. The second dimension is communication quality assessment. The system continuously evaluates the communication latency and packet loss rate of the network connection. When the communication latency exceeds a dynamic threshold or the packet loss rate exceeds a preset threshold (default 5%), the system determines that the current network quality cannot meet the real-time requirements of emergency command and control, and then triggers self-organizing network mode.

[0111] After the edge self-organizing network mode is started, the system elects a temporary master node through an election algorithm. The election algorithm is based on the improved Raft protocol. When the edge node's heartbeat times out (default 100ms without confirmation from the master node), the node's state changes from Follower to Candidate, its current term is incremented by 1, and it sends a RequestVote (RPC) to other nodes in the cluster. Nodes receiving the vote request perform a comprehensive evaluation of the candidate node, using the following formula:

[0112] ;

[0113] in, The current CPU load rate (value range 0-1) is calculated using a 5-second sampling period and a moving average of the three most recent sampled values. The currently available communication bandwidth, This is the theoretical maximum bandwidth; The Euclidean distance between the candidate node and the center of the disaster area (unit: meters); Nodes are assigned a pre-defined priority level (levels 1-5). Weighting coefficients are α=0.4, β=0.3, γ=0.2, and δ=0.1. The node priority level classification is as follows: Level 5 is for emergency command nodes (e.g., emergency communication command vehicles), Level 4 is for professional rescue nodes (e.g., fire truck onboard terminals), Level 3 is for medical rescue nodes (e.g., ambulance terminals), Level 2 is for auxiliary support nodes (e.g., material transport vehicles), and Level 1 is for ordinary terminal nodes (e.g., volunteer terminals). If a candidate node receives more than half of the total number of votes in the cluster and has the highest overall score among all candidates in this round, it is elected as the temporary leader. After being elected, the temporary leader periodically (every 50ms by default) sends heartbeat messages to other nodes to maintain its leadership position.

[0114] After the temporary master node election is completed, all edge nodes in the area enter short-range communication sharing mode. Each node uses a pre-configured short-range communication protocol, such as the IEEE 802.11p-based vehicular ad hoc network protocol or the Bluetooth Low Energy-based near-field communication protocol, to exchange data directly with each other.

[0115] Once the network is restored, the system initiates an incremental synchronization process to synchronize the data accumulated by edge nodes during the network outage to the cloud hub. The incremental synchronization strategy is based on a comparison of data version numbers and hash values: edge nodes maintain a version number and hash value for each data object, incrementing the version number and recalculating the hash value each time data is updated. During synchronization, edge nodes only report data objects whose version numbers or hash values ​​have changed to the cloud.

[0116] A hash chain verification mechanism is used to ensure data integrity and consistency during incremental synchronization. It uniformly employs the SHA-256 secure hash algorithm, outputting a fixed-length 256-bit (32-byte) hash value. Each data operation record forms a block, with the following block structure:<Index,Timestamp,DataHash,PrevBlockHash,Nonce,CurrentBlockHash> Where Index is the block's sequence number in the chain; Timestamp is the millisecond-level timestamp of the operation; DataHash is the SHA-256 hash value of the data object's content; PrevBlockHash is the CurrentBlockHash value of the previous block (the genesis block is 32 bytes all zeros); Nonce is a 32-bit random number; CurrentBlockHash is calculated as follows: After each data write or update operation is completed, the system generates a new block, sets its PrevBlockHash to the CurrentBlockHash of the previous block, calculates its own CurrentBlockHash, and appends the new block to the end of the chain. Once the network is restored, the cloud verifies the continuity of the hash chain to confirm data integrity. If a hash chain break or hash value mismatch is found, a data retransmission process is triggered.

[0117] Step 4: Data extrapolation and retrospection based on digital twins

[0118] Step 4 constructs a digital twin model based on the unified data object generated in Step 1. This model is then used to build a simulation model in virtual space, enabling data simulation and retrospective analysis. The core components of the digital twin model include three sub-models: an urban infrastructure model, an emergency resource model, and a disaster evolution model. The urban infrastructure model is built based on geographic information system (GIS) data and includes the spatial location and attribute information of key infrastructure such as urban road networks, bridges and tunnels, underground pipelines, power facilities, and communication facilities. The emergency resource model stores the location, status, and dispatch capacity information of various emergency resources, including ambulances, fire trucks, police cars, emergency material warehouses, and shelters. The disaster evolution model is built based on historical disaster data and real-time monitoring data, encompassing the evolution patterns and influencing factors of various disasters.

[0119] The dynamic adjustment mechanism of the digital twin model ensures that the simulation model in the virtual space remains synchronized with the real emergency scenario. The unified data object generated in step 1 is pushed to the digital twin module in real time, and the module updates the corresponding model parameters according to the data content. The update frequency of the model is consistent with the data push frequency; for scenarios with high real-time requirements, the update frequency can reach once per second.

[0120] The sand table simulation function is the core application scenario of digital twin models, which includes two functional components: scenario simulation unit and result prediction unit.

[0121] (I) Scenario Simulation Unit

[0122] The scenario simulation unit employs a discrete event simulation engine based on Agent-Based Modeling (ABM). The agent is defined as follows:

[0123] Vehicle agent: Attributes include location (latitude and longitude coordinates), speed, destination, and route planning. The behavioral model uses the NaSch driving model to update speed and location. Acceleration. deceleration Random slowing probability Maximum speed The following speed limits are set according to road type: 30 m / s for expressways, 15 m / s for urban arterial roads, 10 m / s for urban secondary arterial roads, and 6 m / s for local roads / residential roads.

[0124] Human agents: Attributes include location, status (safe, injured, trapped), and movement intention. Behavioral models include panic diffusion models and herd behavior models. Panic value. The evolution equation is:

[0125] ;

[0126] in:

[0127] , =20 meters is the panic perception radius;

[0128] The number of adjacent agents within the sensing range;

[0129] =0.3 is the panic transmission coefficient. =0.5 is the disaster impact coefficient. =0.05 is the panic attenuation coefficient;

[0130] For position The disaster intensity factor at the location has a value range of [0,1].

[0131] Panic Judgment Rule: When > At that moment, the agent enters a state of panic, increasing their movement speed to 1.5 times their normal speed. =0.6 is the threshold for judging panic.

[0132] Conformity model: Proxy observation radius =The prevailing movement direction of people within a 15-meter radius, with probability Following the mainstream trend, among which =0.8 represents the maximum probability of following. The number of agents within the observation range.

[0133] Resource agents: such as fire trucks and ambulances, with attributes including location, status (idle, dispatched, in operation), and task objective. The behavioral model uses the Contract Network protocol for task allocation, and the bid price calculation formula is: ;

[0134] The meanings of each parameter are as follows:

[0135] The smaller the value of the resource agent r, the more suitable the agent is to execute the task.

[0136] : Current position of resource proxy r With the mission location Euclidean distance between them (unit: meters);

[0137] Maximum movement speed of resource agents (unit: meters / second), fire truck retrieval =15 m / s, ambulance take =12 m / s;

[0138] The remaining energy percentage of resource agent r, with a value range of [0,1]. =1 indicates sufficient energy;

[0139] The resource proxy r's suitability for the task type ranges from [0,1] and is determined by the degree of matching between the proxy type and the task type (e.g., the suitability of a fire truck for a firefighting task). =1.0, adaptability to medical rescue missions =0.2);

[0140] : Weighting coefficients, corresponding to the weights of time cost, energy cost, and capability matching degree, with values ​​of respectively. =0.4, =0.3, =0.3, which satisfies .

[0141] Task allocation rules: Select bid price The smallest resource agent acts as the task executor. If the bid price differences among multiple agents are within 5%, the best fit is selected. The highest level of agent.

[0142] The environmental model includes a traffic flow model and a disaster spread model. The simulation time step is fixed. =1 second.

[0143] (ii) Result Prediction Unit

[0144] The outcome prediction unit employs a time-series prediction model based on a Long Short-Term Memory (LSTM) network. Each case in the historical case database is a time-series data record, documenting the entire process from the occurrence to the end of the event. Each record includes: event characteristics (disaster type, occurrence time, initial intensity, meteorological conditions, etc.), decision vector (a binary vector representing whether a certain measure was taken), state vector_t (affected area, number of casualties, resource consumption rate, traffic congestion index, etc., sampled every 30 seconds), and outcome vector (final number of casualties, economic losses, response time, etc.).

[0145] LSTM Model Structure: The input layer receives a sequence of historical situation features [StateVector_{t-L+1}, ..., StateVector_t] with a time window length of L=20 steps and a decision scheme DecisionVector, which are concatenated to form the input tensor. The hidden layer consists of two stacked LSTM units, each with 128 units, and a dropout rate of 0.2. The output layer is a fully connected layer with a linear activation function, outputting the predicted situation sequence [StateVector_{t+1}, ..., StateVector_{t+K}] for the next K=10 steps and the final OutcomeVector. Training uses the Adam optimizer with mean squared error (MSE) as the loss function, and the model is trained on a historical case library until the loss on the validation set no longer decreases.

[0146] Description of the auxiliary module and adaptive threshold

[0147] The invention also includes two auxiliary modules: a data quality monitoring module and a cross-departmental data conflict resolution module.

[0148] The data quality monitoring module is used to monitor the integrity, accuracy, and timeliness of accessed data in real time. Integrity monitoring checks whether data objects contain all six required fields (data identifier, timestamp, spatial coordinates, data type, semantic label, and priority). The trigger threshold is an alert when the missing field rate exceeds 5%, and data repair is triggered when it exceeds 15%. Accuracy monitoring includes numerical range verification, logical consistency verification, and cross-source cross-validation. The trigger threshold is an alert when the confidence level of abnormal data exceeds 0.8, and data repair is triggered when more than 3 out of 10 consecutive data entries are abnormal. Timeliness monitoring checks the difference between the data's timestamp and the current time. Different data types have corresponding maximum allowable delays (e.g., 1 second for gas leak concentration, 0.5 seconds for earthquake P-wave detection, 5 seconds for water level monitoring, etc.). The trigger threshold is an alert when the proportion of expired data exceeds 10%, and data repair is triggered when it exceeds 30%.

[0149] The timestamp mechanism provides accurate time records for all data operations in the digital twin model. It employs a globally unified time server to synchronize with a standard time source based on the Network Time Protocol (NTP), ensuring that the timestamps of data operations across all nodes are comparable and traceable, with millisecond-level accuracy.

[0150] The data repair process includes three processing methods: data supplementation, data correction, and data filtering. Data correction uses moving average filtering to filter outlier data points in the time series. The correction value is (Midpoint of the sequence); For the point at the beginning of the sequence (t=1), the correction value is... For the point at the end of the sequence (t=n), the correction value is... Data that fails to be corrected (data that still exceeds 1.5 times the upper limit of the normal range after correction, or deviates from adjacent points by more than 3 times the local standard deviation) is moved to the isolation area for manual review.

[0151] The cross-departmental data conflict resolution module handles semantic and permission conflicts between different departments on the same data object. The module incorporates a three-tiered arbitration logic: Level 1 is a priority rule, with departmental authority levels divided into 5 levels (Level 5 for emergency command centers, Level 4 for fire and rescue / public security command, Level 3 for traffic management / meteorological bureaus, Level 2 for material support / community grids, and Level 1 for volunteer organizations / third-party service providers); Level 2 is a timestamp rule, adopting the more recent timestamp if Level 1 cannot distinguish between them; Level 3 is a permission level rule, with data sensitivity levels divided into 5 levels (Level 1 for public data, Level 2 for internal data, Level 3 for sensitive data, Level 4 for confidential data, and Level 5 for top-secret data). If the requester's permission level equals the data sensitivity level, modification is allowed; otherwise, if it is higher, additional verification of the requester's identity (e.g., through digital certificates or biometrics) is required before modification can be performed.

[0152] For the multiple preset thresholds involved in this method (including communication delay threshold, packet loss rate threshold, data timeliness window, spatiotemporal slice reference change rate threshold, correlation threshold, etc.), an adaptive threshold determination method based on historical statistical distribution is adopted. Initial values ​​are set based on expert experience during the initial system deployment. When the system is running normally and there are no emergency events, performance data is continuously collected and fitted with a Gaussian distribution to obtain the mean μ and standard deviation σ under normal conditions. The dynamic threshold calculation formula is as follows: , where k is the alarm coefficient, k=3 (corresponding to the 99.7% confidence interval). μ and σ are calculated based on a moving window of the most recent N=60 data points, updated every minute. An alert is triggered when M=5 consecutive data points exceed the threshold. The system recalculates the baseline model and updates the threshold every 24 hours.

[0153] Taking an emergency scenario of a gas pipeline leak in a city as an example, the specific application process of the method of the present invention is explained.

[0154] Following the incident, various sensors in the comprehensive perception layer collected real-time data on gas concentration, meteorological data, and video surveillance data in the leak area. The dynamic data access gateway received this multi-source heterogeneous data and used an adaptive spatiotemporal slicing algorithm to calculate the rate of change in gas concentration. The data is semantically aligned with the emergency semantic knowledge graph and encapsulated into a unified data object. The unified data object contains key information such as the geographic coordinates of the leak location (longitude 116.397, latitude 39.908), the gas concentration monitoring value (15% LEL), and the distribution of surrounding personnel.

[0155] After receiving the unified data object, the emergency command center initiates the task-driven data derivation mechanism in step 2. The DQN model of the task decomposition unit automatically decomposes the "gas pipeline leak emergency task" into sub-tasks such as concentration monitoring, area evacuation, traffic control, and emergency repair operations. The data requirement analysis unit calculates the minimum dataset for each sub-task: for the "area evacuation" sub-task, it calculates the relevance score Rel(D, T) and filters out the personnel distribution data within a 500-meter radius of the leak point; for the "traffic control" sub-task, it filters out the real-time traffic flow data of the surrounding main roads. The CP-ABE algorithm encrypts sensitive data containing personnel identification information and sets the access policy to ('Public Security' AND 'On-site Commander'), ensuring that this data is only visible to the on-site commander of the public security department. The dynamic routing unit pushes the processed data to relevant departments such as public security, fire, medical, and transportation, with the push latency controlled within 200ms.

[0156] Edge nodes deployed at the rescue site continuously monitor the network connection status with the cloud hub. When a leak damages surrounding communication base stations, interrupting the network connection between edge nodes and the cloud, the edge ad-hoc network mode is automatically triggered. The edge nodes on site use an improved Raft election algorithm to select the node with the most remaining computing resources (CPU load rate 25%), the highest communication bandwidth (50Mbps), and the closest distance to the disaster center (50 meters) as the temporary master node. Each node shares disaster data and task instructions using a vehicle-mounted ad-hoc network based on the IEEE 802.11p protocol. Once the network is restored, a hash chain verification mechanism verifies the integrity of the data accumulated during the network outage, and incremental synchronization is performed after confirming the data is correct.

[0157] The digital twin module constructs a virtual model of the accident scene based on a unified data object. The LSTM model predicts the expected number of casualties for each scenario based on the current situation (gas concentration 15% LEL, wind speed 2 m / s, population density 500 people / km²) and multiple alternative scenarios. The ABM simulation engine simulates traffic congestion and resource allocation processes under each scenario, providing decision-makers with a reference for comparing options. All data operations are recorded with precise timestamps (milliseconds), supporting post-event review and accountability.

[0158] Example 2

[0159] This embodiment describes the specific application of the method of the present invention in an urban flood disaster emergency scenario. This scenario is characterized by a relatively slow disaster evolution but a wide impact range, requiring coordination among multiple departments for a continuous emergency response.

[0160] In the data access and semantic alignment stage of step 1, the dynamic data access gateway additionally integrates a hydrological monitoring data adapter and a drainage facility status data adapter, taking into account the characteristics of flood disaster scenarios. The adaptive spatiotemporal slicing algorithm uses the water level height change rate. As a key indicator Set to 0.1. The emergency semantic knowledge graph has added semantic mapping relationships related to flood disasters, including mappings between rainfall and urban flooding risk levels, river water levels and inundation areas, and drainage pumping station operating status and drainage capacity. The unified data object has added inundation area polygon fields and flood point level fields to accurately describe the spatial distribution and severity of flood disasters.

[0161] In the data dynamic desensitization and routing stage of step 2, considering the long duration of flood disaster emergency responses, a longer time window is adopted for data timeliness assessment. The effective time window for drainage facility status data is set to ten minutes, and the effective time window for river water level data is set to five minutes. The attribute-based encryption algorithm adds a "hydrological data confidentiality level" attribute for flood scenarios, encrypting and protecting sensitive reservoir water level data and flood control scheduling information.

[0162] In the cloud-edge network outage self-organization stage of step 3, considering the wide distribution of edge nodes in flood disaster scenarios, the election algorithm adds a geographical proximity weight, prioritizing nodes located in the center of the flooded area and with good communication conditions with other nodes as temporary master nodes. The incremental synchronization strategy, addressing the large volume of flood data, employs a differential compression algorithm to compress identical geographical coordinate data before transmission, effectively reducing network bandwidth consumption.

[0163] In the digital twin simulation phase of step 4, the urban infrastructure model focuses on loading drainage network data, the emergency resource model focuses on loading flood control material data such as drainage pump trucks and rubber boats, and the disaster evolution model constructs a rainfall-runoff evolution model based on historical flood data. The scenario simulation unit simulates the evolution of urban flooding under different rainfall intensities, and the result prediction unit predicts the implementation effects of different drainage schemes, providing decision support for flood control command.

[0164] Example 3

[0165] This embodiment describes the specific application of the method of the present invention in an urban earthquake disaster emergency response scenario. This scenario is characterized by the sudden occurrence and high destructive power of the disaster, requiring rapid response and efficient resource allocation.

[0166] In the data access and semantic alignment stage of step 1, the dynamic data access gateway integrates earthquake monitoring data adapters, building vibration monitoring data adapters, and emergency communication equipment data adapters. The adaptive spatiotemporal slicing algorithm uses peak ground acceleration. As a key indicator Setting it to 0.8 allows for a minimum window size of 1 second. The emergency semantic knowledge graph adds semantic mapping relationships related to earthquake disasters, including mappings between magnitude and damage level, building damage level and collapse risk, and the accessibility of rescue routes. The unified data object adds a building damage level field and an estimated number of trapped personnel field.

[0167] In the data dynamic desensitization and routing phase of step 2, the earthquake emergency response task was broken down into sub-tasks such as rubble search, casualty treatment, passage opening, and material transportation. When calculating the minimum dataset, the data requirements analysis unit focused on acquiring data on building distribution, road capacity, and medical resource distribution within the disaster area. Attribute-based encryption algorithms were used to apply high-level encryption to sensitive data containing the location information of trapped personnel, ensuring that rescue plan information was only visible to the rescue teams.

[0168] In the cloud-edge self-organizing phase of step 3, earthquakes often result in widespread damage to communication infrastructure, leading to more frequent triggering of edge self-organizing network modes. The election algorithm incorporates a weighted average of remaining battery power, prioritizing nodes with sufficient battery power as temporary master nodes to ensure the continuous operation of the self-organizing network system. The short-range communication protocol employs an ultra-wideband communication protocol with strong anti-interference capabilities, ensuring reliable communication in complex environments such as ruins.

[0169] In step 4, the digital twin simulation stage, the digital twin model focuses on loading building structure data and underground pipeline network data to construct a building seismic damage analysis model and a secondary disaster prediction model. The scenario simulation unit simulates the implementation effects of different rescue plans, including comparing the efficiency of different search routes and the coverage of different resource allocation plans. The results prediction unit predicts the probability of rescuing trapped personnel and the safety risks to rescue teams for each plan based on historical earthquake rescue cases.

[0170] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the specific embodiments described above, which are merely exemplary and not restrictive. The scope of protection of the present invention is defined by the claims, and any equivalent substitutions or modifications based on the technical solutions of the present invention fall within the scope of protection of the present invention.

Claims

1. A data collaborative processing method for a central capability platform for urban emergency response, characterized in that, Includes the following steps: Step 1: Based on adaptive spatiotemporal slicing and semantic alignment, a dynamic data access gateway is constructed to parse various types of emergency data and encapsulate them into unified data objects. Real-time semantic alignment and conflict resolution of multi-source heterogeneous data are achieved by constructing an emergency semantic knowledge graph containing an entity mapping relationship library and an attribute association matrix. The size of the spatiotemporal slice window is dynamically determined using an adaptive algorithm based on disaster evolution rate assessment. Step 2: Task-driven dynamic data desensitization and routing. Using a task-driven data extrapolation mechanism, after the emergency command center generates an emergency task, the task is automatically decomposed by a reinforcement learning model based on a hierarchical task network. The minimum dataset required to execute the task is determined based on the correlation quantification calculation method. The encrypted policy attribute base encryption algorithm is used to automatically remove classified information and push the processed data to specific departments. Step 3: Cloud-edge self-organizing data collaboration method when the edge node detects an interruption in connection with the cloud. When the edge node detects an interruption in connection with the cloud, it triggers the edge self-organizing network mode. Through an election algorithm based on the improved Raft protocol, a temporary master node is selected by comprehensively considering the remaining computing resources, communication bandwidth and geographical advantages. The nodes share disaster data using short-distance communication. After the network is restored, the data integrity is verified by using a hash chain based on SHA-256 and incrementally synchronized to the cloud hub. Step 4: Data extrapolation and retrospection based on digital twins. Based on a unified data mirror, an extrapolation model is built in virtual space. Decision-makers use a simulation engine based on the multi-agent model (ABM) to conduct sand table extrapolation to predict the development of the situation, and use a time series prediction model based on long short-term memory network (LSTM) to predict the potential results of different decision options. All data operations have timestamps synchronized with a global time server to support post-event auditing and accountability.

2. The method according to claim 1, characterized in that, The adaptive algorithm described in step 1 obtains the average rate of change by calculating the first derivative of the key indicators. and according to the function Dynamically adjust the window size, where As the reference window, This serves as a reference rate of change threshold.

3. The method according to claim 1, characterized in that, The emergency semantic knowledge graph mentioned in step 1 is constructed using a hybrid method that combines rules and statistics; the conflict resolution is based on the authority priority of the data source, timestamps, and consistency rules based on RDF triples for arbitration.

4. The method according to claim 1, characterized in that, The relevance quantification calculation method described in step 2, for data object D and subtask T, its relevance score Rel(D,T) is composed of a weighted sum of semantic similarity, spatial similarity and timeliness matching degree, and only data objects with scores greater than a preset threshold are retained to form the minimum dataset.

5. The method according to claim 1, characterized in that, The comprehensive scoring formula for the election algorithm described in step 3 is: ,in, This represents the current CPU load rate. This represents the currently available communication bandwidth. This is the theoretical maximum bandwidth; The distance between the candidate node and the center of the disaster area is the Euclidean distance. The priority level is preset for the node; α, β, γ, and δ are weight coefficients.

6. The method according to claim 1, characterized in that, The structure of each data operation block in the hash chain verification described in step 3 is as follows:<Index, Timestamp, DataHash, PrevBlockHash, Nonce,CurrentBlockHash> Integrity verification is achieved by recursively verifying the continuity of hash values.

7. The method according to claim 1, characterized in that, The LSTM-based time-series prediction model described in step 4 takes historical situation feature sequences and decision scheme vectors as inputs and outputs future prediction situation sequences and final result vectors. The simulation engine based on the multi-agent model includes vehicle agents, personnel agents, and resource agents, and adopts discrete event simulation and time step advancement mechanism.

8. A data collaborative processing system for a central capability platform for urban emergency response, used to execute the method according to any one of claims 1 to 7, characterized in that, It includes a global perception layer, an intelligent central capability layer, and an edge collaboration layer; The global perception layer is equipped with a dynamic data access gateway, which is used to receive multi-source heterogeneous data streams from IoT devices, video surveillance systems, voice communication systems and geographic information systems, and send the parsed and encapsulated unified data object to the intelligent central capability layer. The intelligent central capability layer is communicatively connected to the global perception layer. The intelligent central capability layer includes a task decomposition unit, a data requirement analysis unit, a dynamic routing unit, and a digital twin module. The task decomposition unit is used to break down emergency tasks into several sub-tasks and determine the dependencies between them. The data requirement analysis unit, connected to the task decomposition unit, is used to calculate the minimum dataset required to execute each sub-task based on the dependencies. The dynamic routing unit, connected to the data requirement analysis unit, is used to determine the data push path based on the minimum dataset. The digital twin module is used to construct a digital twin model based on a unified data object for sand table simulation. The edge collaboration layer is communicatively connected to the intelligent central capability layer. The edge collaboration layer includes multiple edge nodes, each of which has a status monitoring module, a self-organizing network communication module, and an incremental synchronization module. The edge nodes have bidirectional communication links with the cloud central hub, and the edge nodes have short-distance communication links with each other.

9. The system according to claim 8, characterized in that, It also includes a data quality monitoring module and a cross-departmental data conflict resolution module; the data quality monitoring module is used to monitor the integrity, accuracy and timeliness of the accessed data; the cross-departmental data conflict resolution module is used to handle semantic conflicts and permission conflicts between multiple departments based on priority rules, timestamp rules and permission level rules.

10. The system according to claim 8, characterized in that, The preset thresholds involved in the system, including communication delay threshold and packet loss rate threshold, are set and dynamically updated using an adaptive threshold determination method based on historical statistical distribution. The adaptive threshold determination method involves: establishing a Gaussian distribution model based on performance data collected during normal system operation, calculating the mean μ and standard deviation σ of the performance data; and then applying the formula... Calculate the dynamic threshold, where Th represents the dynamic threshold at the current moment, μ represents the mean of the performance data, σ represents the standard deviation of the performance data, and k is the alarm coefficient, with a value of 3.

Images