An emergency single-graph scheduling method and system based on knowledge graph

Abstract

This invention discloses an emergency single-graph scheduling method and system based on knowledge graphs, belonging to the field of emergency management information technology. The method includes real-time acquisition of multi-source spatial data, spatiotemporal alignment and fusion processing of the multi-source spatial data to obtain spatial entity state vectors; construction of a spatiotemporal knowledge network based on the spatial entity state vectors, establishing spatial entity nodes, semantic relationship edges, taboo state nodes, and risk propagation inverse edges in the spatiotemporal knowledge network to obtain the spatiotemporal knowledge network topology. This invention solves the problem that existing forward optimization models are prone to exceeding safety limits due to relying solely on soft penalty terms, the problem that existing emergency dispatch systems cannot predict future secondary risk triggering paths after implementing a plan, and the problem that static planning schemes gradually become ineffective as uncertainty accumulates with the evolution of emergencies.

Description

Technical Field

[0001] This invention relates to the field of emergency management information technology, and in particular to an emergency single-map dispatching method and system based on knowledge graphs. Background Technology

[0002] With the continuous development of Geographic Information Systems (GIS), knowledge graph technology, and multi-objective optimization algorithms, emergency command and dispatch systems have gradually moved from the stage of single data display to the stage of intelligent decision support. Emergency single-map, as the core carrier of spatial visualization aggregation, can realize geospatial mapping of multi-source heterogeneous data; knowledge graph technology supports rapid retrieval and matching of disaster events, rescue resources, and command plans through semantic association and rule querying of emergency elements; at the dispatch optimization level, resource dispatch models based on genetic algorithms, evolutionary algorithms, or multi-agent systems use the current situation as input to solve for the optimal solution for future resource allocation. However, facing the high complexity and uncertainty of emergencies, such as chain transmission of disasters and multi-agent collaborative conflicts, as well as the limitations of the dispatch algorithms themselves, existing output results inevitably have the risk of breaching safety boundaries. These dispatch errors will propagate step by step in the cascading system. If the cascading secondary risks are not effectively modeled and handled in the system design, it may lead to unacceptable secondary disasters triggered during the execution of the dispatch plan, thus posing a potential threat to rescue safety and the lives and property of the affected people.

[0003] To improve the safety and reliability of emergency dispatch, researchers have introduced a technical approach combining knowledge graphs and multi-objective optimization. Knowledge graphs store the semantic relationships and historical patterns of emergency elements, while multi-objective optimization algorithms solve for resource allocation schemes. However, existing emergency dispatch methods still have several limitations when applied to complex disaster scenarios. For example, existing methods focus on optimal resource allocation under the current situation, lacking a mechanism to predict the transmission of secondary risks during future evolution, leading to unforeseen cascading risks after execution. Secondly, existing knowledge graphs are mainly used for historical case retrieval and rule querying, failing to deeply embed the constraint formation process of dispatch decisions, resulting in insufficient security guarantees under complex constraints. Furthermore, existing dispatch schemes are mostly one-time static planning, lacking dynamic feedback and correction mechanisms based on real-time situations during execution, making them difficult to adapt to the high uncertainty of sudden event evolution.

[0004] Existing technologies lack an emergency dispatch method that can predict cascading risk transmission in advance, deeply embed security constraints, and dynamically adapt to situational evolution. Summary of the Invention

[0005] The purpose of this invention is to propose an emergency single-graph scheduling method and system based on knowledge graphs. This invention solves the problem that existing positive optimization models are prone to exceeding safety limits due to their reliance on only soft penalty terms, the problem that existing emergency scheduling systems cannot predict the triggering paths of future secondary risks after the execution of a plan, and the problem that static planning schemes gradually become ineffective as uncertainty accumulates with the evolution of emergencies.

[0006] In a first aspect, embodiments of this application provide an emergency single-graph scheduling method based on knowledge graphs, comprising the following steps:

[0007] Real-time acquisition of multi-source spatial data, spatiotemporal alignment and fusion processing of multi-source spatial data to obtain spatial entity state vectors;

[0008] A spatiotemporal knowledge network is constructed based on the spatial entity state vector. Spatial entity nodes, semantic relation edges, taboo state nodes, and risk transmission inverse edges are established in the spatiotemporal knowledge network to obtain the spatiotemporal knowledge network topology.

[0009] Candidate scheduling schemes are generated based on the spatial entity state vectors. Using the candidate scheduling schemes and the spatiotemporal knowledge network topology as input, multi-step risk propagation simulation is performed along the semantic relationship edges to obtain the risk evolution trajectory.

[0010] Identify activated taboo state nodes from the risk evolution trajectory, and starting from the activated taboo state nodes, traverse the spatiotemporal knowledge network topology in reverse along the risk propagation reverse edge to extract the trigger threshold boundary and obtain the avoidance constraint set.

[0011] By using the avoidance constraint set as a hard constraint condition and combining it with the spatiotemporal knowledge network topology, the scheduling scheme is solved to obtain the initial scheduling scheme.

[0012] Based on the difference between the spatial entity state vector and the risk evolution trajectory, the avoidance constraint set is updated, and the initial scheduling scheme is rolled over to obtain the emergency scheduling result.

[0013] In one embodiment, multi-source spatial data is acquired in real time, and spatiotemporal alignment and fusion processing is performed on the multi-source spatial data to obtain a spatial entity state vector, including:

[0014] By integrating meteorological monitoring data, traffic flow data, video surveillance data, social media sentiment data, and data transmitted from on-site sensors, multi-source spatial data is obtained.

[0015] Spatiotemporal alignment and fusion processing of multi-source spatial data are performed to obtain spatial entity state vectors that include event point coordinates, resource point reserves, path segment accessibility, and shelter capacity.

[0016] In one embodiment, a spatiotemporal knowledge network is constructed based on spatial entity state vectors. Spatial entity nodes, semantic relation edges, taboo state nodes, and risk propagation inverse edges are established within the spatiotemporal knowledge network to obtain the spatiotemporal knowledge network topology, including:

[0017] Spatial entity nodes are generated based on spatial entity state vectors, and semantic tag attributes and spatiotemporal coordinate attributes are configured for each spatial entity node.

[0018] Establish semantic relationship edges and configure spatiotemporal triggering conditions for each semantic relationship edge. The spatiotemporal triggering conditions include time window, spatial range and threshold conditions.

[0019] Establish taboo state nodes, and build reverse risk transmission edges from the taboo state nodes to obtain the spatiotemporal knowledge network topology.

[0020] In one embodiment, candidate scheduling schemes are generated based on spatial entity state vectors. Using the candidate scheduling schemes and the spatiotemporal knowledge network topology as input, a multi-step risk propagation simulation is performed along semantic relationship edges to obtain the risk evolution trajectory, including:

[0021] Based on the resource reserves and path accessibility in the spatial entity state vector, a set of resource allocation actions is generated to obtain candidate scheduling schemes;

[0022] Starting from the resource allocation node corresponding to the candidate scheduling scheme, traverse all outgoing semantic relationship edges;

[0023] Check the spatiotemporal triggering conditions on the semantic relationship edge. When the time window includes the inference time, the spatial range includes the affected area, and the threshold condition is exceeded, activate the next level node and record the risk level increment.

[0024] The simulation process is advanced by dividing the data into preset time slices to obtain the risk evolution trajectory.

[0025] In one embodiment, activated taboo state nodes are identified from the risk evolution trajectory. Starting from the activated taboo state nodes, the spatiotemporal knowledge network topology is traversed in reverse along the risk propagation inverse edge to extract the trigger threshold boundary, thereby obtaining the avoidance constraint set, including:

[0026] Identify activated taboo state nodes from the risk evolution trajectory;

[0027] Starting from the activated taboo state node, traverse the spatiotemporal knowledge network topology in reverse along the risk propagation reverse edge, with the search depth limited to a preset number of layers;

[0028] Record the complete path from the taboo state node back to the scheduling action node, and extract the threshold boundaries of the spatiotemporal triggering conditions on the path;

[0029] Transform the threshold boundary into an avoidance constraint set.

[0030] In one embodiment, the avoidance constraint set is used as a hard constraint condition, and the scheduling scheme is solved by combining the spatiotemporal knowledge network topology to obtain an initial scheduling scheme, including:

[0031] The avoidance constraint set is used as a hard constraint condition, which together with the regular constraints constitutes a mixed constraint condition.

[0032] Under mixed constraints, a multi-objective evolutionary solution is performed to obtain an initial scheduling scheme.

[0033] In one embodiment, based on the difference between the spatial entity state vector and the risk evolution trajectory, the following is included:

[0034] Real-time acquisition of multi-source spatial data, spatiotemporal alignment and fusion processing of multi-source spatial data to obtain updated spatial entity state vectors;

[0035] Calculate the deviation between the updated spatial entity state vector and the risk evolution trajectory;

[0036] When the deviation exceeds the preset threshold, the re-analysis process is triggered.

[0037] In one embodiment, updating the evasion constraint set includes:

[0038] Monitor the tightness of each constraint in the constraint avoidance set and identify constraints that can be appropriately relaxed;

[0039] Release the optimization space to obtain the updated avoidance constraint set.

[0040] In one embodiment, rolling adjustments to the initial scheduling scheme include:

[0041] The updated avoidance constraint set is used as a hard constraint condition, and the scheduling scheme is solved by combining the spatiotemporal knowledge network topology to obtain the emergency scheduling result.

[0042] Secondly, embodiments of this application provide an emergency single-graph dispatch system based on a knowledge graph, including:

[0043] The data fusion module is used to collect multi-source spatial data in real time, perform spatiotemporal alignment and fusion processing on the multi-source spatial data, and obtain spatial entity state vectors.

[0044] The network construction module is used to construct a spatiotemporal knowledge network based on the spatial entity state vector. In the spatiotemporal knowledge network, spatial entity nodes, semantic relation edges, taboo state nodes, and risk transmission reverse edges are established to obtain the spatiotemporal knowledge network topology.

[0045] The risk simulation module is used to generate candidate scheduling schemes based on the spatial entity state vectors. Taking the candidate scheduling schemes and the spatiotemporal knowledge network topology as input, it performs multi-step risk propagation simulation along semantic relationship edges to obtain the risk evolution trajectory.

[0046] The constraint extraction module is used to identify activated taboo state nodes from the risk evolution trajectory. Starting from the activated taboo state nodes, the spatiotemporal knowledge network topology is traversed in reverse along the risk propagation reverse edge to extract the trigger threshold boundary and obtain the avoidance constraint set.

[0047] The scheme solving module is used to solve the scheduling scheme by taking the avoidance constraint set as hard constraint conditions and combining it with the spatiotemporal knowledge network topology to obtain the initial scheduling scheme.

[0048] The dynamic correction module is used to update the avoidance constraint set based on the difference between the spatial entity state vector and the risk evolution trajectory, and to perform rolling correction on the initial scheduling scheme to obtain the emergency scheduling result.

[0049] The beneficial effects of this invention are:

[0050] 1. This invention establishes a reverse risk propagation edge from the taboo state node and extracts the trigger threshold boundary by traversing along the reverse edge. It transforms the absolutely unacceptable future state into a hard constraint condition for the current scheduling action and injects it into the solution process. This solves the problem that the existing forward optimization model is prone to breaking the safety red line because it only relies on soft penalty terms. It realizes the rigid locking of the safety boundary of the scheduling scheme and the ex-ante avoidance of cascaded secondary risks.

[0051] 2. This invention establishes semantic relationship edges in a spatiotemporal knowledge network and configures spatiotemporal triggering conditions. It simulates multi-step risk propagation along the semantic relationship edges with candidate scheduling schemes as input. This solves the problem that existing emergency scheduling systems cannot predict the triggering path of future secondary risks after the execution of a scheme, and realizes the full-chain deduction from the current situation to the future risk evolution trajectory.

[0052] 3. This invention solves the problem that static planning schemes gradually fail as uncertainty accumulates with the evolution of sudden events by dynamically updating the constraint avoidance set and resolving the scheduling scheme based on the decay law of the projection confidence with the projection time and the degree of deviation between the actual situation and the projection baseline. This ensures the continuous effectiveness of the scheduling scheme during the execution process. Attached Figure Description

[0053] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0054] Figure 1 A flowchart of the steps provided for this invention;

[0055] Figure 2 This is a schematic diagram of the structure proposed in this invention. Detailed Implementation

[0056] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0057] First of all, it should be noted that after analyzing the actual application of existing emergency dispatch systems, it was found that traditional systems generally have problems such as lack of cascading risk prediction capability in positive optimization, undefined taboo states in knowledge graphs leading to optimization models exceeding safety red lines, and one-time static planning being unable to adapt to situational drift. These problems combined result in insufficient response efficiency of emergency command and dispatch, making it difficult to meet the forward-looking and adaptive dispatch requirements in complex disaster scenarios.

[0058] This invention constructs an emergency single-graph scheduling method based on knowledge graphs. This method achieves end-to-end emergency scheduling through six steps. The specific implementation of each step is described in detail below with reference to the embodiments.

[0059] refer to Figure 1 This embodiment illustrates the steps of the knowledge graph-based emergency single-graph scheduling method provided in this example. The method includes steps S1: real-time acquisition of multi-source spatial data and generation of spatial entity state vectors; S2: construction of a spatiotemporal knowledge network and generation of its topology; S3: generation of candidate scheduling schemes and cascaded risk forward inference; S4: identification of taboo states and reverse generation of avoidance constraint sets; S5: solution of scheduling schemes based on hybrid constraint conditions; and S6: rolling correction based on situational differences. Each step works together to achieve complete functions of data perception, knowledge modeling, risk inference, constraint generation, scheme solution, and dynamic correction.

[0060] S1. Real-time acquisition of multi-source spatial data, spatiotemporal alignment and fusion processing of multi-source spatial data to obtain spatial entity state vectors.

[0061] Emergency dispatch involves a wide range of multi-source data, including meteorological monitoring data, traffic flow data, video surveillance data, social media sentiment data, and data transmitted from on-site sensors. In their raw state, these data suffer from problems such as format differences, inconsistent transmission protocols, and inconsistent spatiotemporal references, which directly affect the usability of the data. Therefore, it is necessary to solve the above problems through standardization processing and spatiotemporal alignment mechanisms.

[0062] To address the diverse data sources mentioned above, a unified data format, transmission protocol, and semantic specification were established, and a standardized data access interface was constructed. Regarding the data format, numerical data retains two decimal places of precision, timestamps are standardized to year, month, day, hour, minute, and second format, and geographic coordinates adopt the World Geodetic System 1984 coordinate system. The transmission protocol employs a transport layer security (TLS) encryption mechanism, compatible with mainstream IoT transmission protocols such as message queue telemetry, Hypertext Transfer Protocol (HTTP), and Transmission Control Protocol (TCP). For semantic specifications, a unified dictionary of emergency indicators was established, standardizing the same indicator from different sources into a single standard field to eliminate semantic conflicts.

[0063] Using geospatial grid coding as the core, a cross-data source spatiotemporal alignment rule base is established. This rule base includes three categories: data source and spatial grid mapping rules, time alignment rules, and data integrity verification rules. The data source and spatial grid mapping rules clarify the mapping relationship between geographic coordinate field names and standard grid codes in each data source; the time alignment rules accurately match different types of data within the same spatial grid according to timestamps; and the data integrity verification rules set a mandatory list of core fields for each data source.

[0064] Based on a cross-data source spatiotemporal alignment rule library, multi-source monitoring data within the same spatiotemporal range are bound together to ensure a unique correspondence between all data and the spatial grid. A weighted fusion algorithm is used to perform confidence-weighted fusion of similar data from different sources. The fusion formula is as follows:

[0065] ;

[0066] In the formula, For the first Spatial entity state vector components of a spatial grid; Total number of data sources; For the first The confidence weight of each data source is dynamically calibrated based on the historical accuracy of the data source. For the first The data source in the first The original monitoring values ​​of each spatial grid.

[0067] The spatial entity state vector after fusion processing includes four dimensions: event point coordinates, resource point reserves, path segment accessibility, and shelter capacity. It is synchronously stored in the distributed time-series database of the emergency command platform for subsequent steps.

[0068] S2. Construct a spatiotemporal knowledge network based on the spatial entity state vector. Establish spatial entity nodes, semantic relation edges, taboo state nodes, and risk transmission reverse edges in the spatiotemporal knowledge network to obtain the spatiotemporal knowledge network topology.

[0069] The spatiotemporal knowledge network is constructed using a graph database, and the node types include spatial entity nodes, semantic relation nodes, and taboo state nodes.

[0070] Spatial entity nodes are generated based on spatial entity state vectors. Each spatial entity node is configured with semantic tag attributes and spatiotemporal coordinate attributes. Semantic tag attributes include node type identifier, risk level identifier, and resource type identifier. Spatiotemporal coordinate attributes include longitude, latitude, elevation, and timestamp.

[0071] Semantic relationship edges correspond to risk transmission rules, established based on a historical disaster case database and expert knowledge. These rules cover typical transmission paths such as hazardous chemical leaks to downwind residential areas, earthquakes leading to bridge damage and disruption of rescue routes, and large-scale evacuations to resettlement sites leading to infectious disease risks. Each semantic relationship edge is configured with spatiotemporal triggering conditions, which include three sub-attributes: time window, spatial range, and threshold condition. The time window defines the effective period for risk transmission, the spatial range defines the geometric boundary of the risk impact, and the threshold condition defines the critical value for risk activation.

[0072] Taboo state nodes correspond to absolutely unacceptable future states, defined according to emergency management regulations and engineering safety standards, including states such as traffic paralysis on a road section, overflowing hospital capacity, and dam exceeding pressure limits. Each taboo state node is configured with a taboo threshold attribute and a hazard level attribute.

[0073] Risk propagation reverse edges are constructed from the taboo state node, pointing to the preceding nodes that may have caused the taboo state. During construction, starting from the taboo state node, following the reverse logic of the risk propagation rules, all direct preceding nodes that could trigger the taboo state are identified, and the first layer of reverse edges is established. Then, the preceding nodes of the preceding preceding nodes are identified, and the second layer of reverse edges is established. This process iteratively continues until the complete risk propagation chain is covered. The weight of each risk propagation reverse edge is determined based on the trigger frequency of that path in historical cases.

[0074] S3. Generate candidate scheduling schemes based on spatial entity state vectors. Using the candidate scheduling schemes and spatiotemporal knowledge network topology as inputs, perform multi-step risk propagation simulation along semantic relationship edges to obtain the risk evolution trajectory.

[0075] Candidate scheduling schemes are generated based on the resource margin and path accessibility in the spatial entity state vector. The generation process is as follows: traverse all resource points and extract resource points with margins greater than the scheduling threshold as schedulable resources; calculate the reachability of each resource point to the event point based on the path segment accessibility; combine the four elements of resource type, allocation quantity, target location, and estimated arrival time to form a set of resource allocation actions, i.e., candidate scheduling schemes.

[0076] The cascaded risk forward inference engine takes candidate scheduling schemes and spatiotemporal knowledge network topology as inputs and performs multi-step risk propagation simulations along semantic relationship edges. Specifically, it starts from the resource allocation node corresponding to the candidate scheduling scheme, traverses all outgoing semantic relationship edges, checks the spatiotemporal triggering conditions on the semantic relationship edges, and activates the next-level node and records the risk level increment when the time window includes the inference time, the spatial range includes the affected area, and the threshold condition is exceeded.

[0077] The formula for calculating the risk level increment is as follows:

[0078] ;

[0079] In the formula, For the first The risk level increment of each node; To point to the first The set of all preceding semantic relation edges of each node; For the first The semantic relation edge points to the first The transmission weight of each node is determined based on historical case statistics; For the first The actual trigger value on the semantic relation edge; For the first Threshold conditions on semantic relation edges; The function ensures that risk increments are generated only when the actual trigger value exceeds the threshold.

[0080] The simulation process proceeds according to preset time slices, with a time slice interval set at 5 minutes. This value is chosen because emergency situations typically change on a minute-by-minute basis, and 5 minutes represents a balance between computational accuracy and computational load. The maximum number of simulation steps is set to 10, covering the risk evolution over the next 50 minutes, a duration that meets the requirements of the golden time window for emergency response. The engine outputs the risk evolution trajectory, including the predicted risk level values ​​for all spatial entity nodes under each time slice.

[0081] S4. Identify the activated taboo state nodes from the risk evolution trajectory. Starting from the activated taboo state nodes, traverse the spatiotemporal knowledge network topology in reverse along the risk propagation reverse edge to extract the trigger threshold boundary and obtain the avoidance constraint set.

[0082] When identifying activated taboo state nodes from the risk evolution trajectory, the risk level prediction values ​​of each time slice in the risk evolution trajectory are traversed. The prediction values ​​are compared with the taboo threshold of the taboo state node. If the prediction value is greater than or equal to the taboo threshold, the taboo state node is determined to be activated.

[0083] The reverse constraint generation process employs a depth-first search algorithm. Starting from the activated taboo state node, the spatiotemporal knowledge network topology is traversed backwards along the risk propagation inverse edge, with a search depth limit of 5 layers. This depth limit is based on the fact that cascaded risk propagation typically converges within 5 layers; beyond 5 layers, the risk propagation probability falls below the effective threshold, making further searching less cost-effective. During the traversal, the complete path backtracking from the taboo state node to the scheduling action node is recorded, and the threshold boundaries of all spatiotemporal triggering conditions along the path are extracted.

[0084] When the threshold boundary is transformed into an avoidance constraint set, the threshold conditions on each path are mapped to the numerical constraints of the corresponding scheduling action parameters. For example, if the path iterates backward from the forbidden state node (hospital overflow) to the scheduling action node (patient evacuation rate control node), and the spatiotemporal triggering condition indicates that a hospital overflow will be triggered when the patient evacuation rate exceeds 80 people per hour, then the avoidance constraint generated is that the patient evacuation rate is less than or equal to 80 people per hour. The avoidance constraint set is output to step S5 through the constraint injection interface.

[0085] S5. Using the avoidance constraint set as a hard constraint condition, and combining it with the spatiotemporal knowledge network topology, the scheduling scheme is solved to obtain the initial scheduling scheme.

[0086] The avoidance constraint set is treated as a hard constraint condition, which together with the regular constraints constitutes a hybrid constraint condition. The regular constraints include total resource constraints, path capacity constraints, and time window constraints; the hard constraints are the threshold boundaries in the avoidance constraint set, which are treated as absolutely insurmountable boundary conditions during the solution process.

[0087] The hybrid scheduling solution employs a multi-objective evolutionary algorithm framework, with the objective function comprising three dimensions: minimizing casualties, minimizing average arrival time, and maximizing resource utilization. The algorithm implementation details are as follows:

[0088] Chromosome encoding uses real numbers, with each chromosome representing a scheduling scheme, and gene positions corresponding to the allocation amount and path selection for each resource point.

[0089] In the population initialization phase, an initial population satisfying the conventional constraints is randomly generated, with a population size set to 200. This value has been experimentally verified as the optimal size balancing convergence speed and solution space coverage. Too small a size can easily lead to local optima, while too large a size results in exponentially increasing computation time. The selection operator uses tournament selection, with a tournament size of 3, balancing selection pressure and population diversity. The crossover operator uses simulated binary crossover, with a distribution index of 20 to ensure that offspring individuals are searched within the parent's neighborhood. A distribution index greater than 20 results in a narrow search range and premature convergence, while a distribution index less than 20 results in a narrow search range and premature convergence. The wide range of search terms leads to slow convergence; the mutation operator uses multinomial mutation with a distribution exponent of 20 to maintain population diversity; the elite retention strategy retains the top 10% of non-dominated solutions in each generation to prevent the loss of excellent solutions; the number of iterations is set to 100 generations, which, combined with the population size of 200, ensures that the algorithm converges to the Pareto front within a finite time; the crossover probability is set to 0.9, as a high crossover probability facilitates the rapid propagation of excellent patterns and accelerates population convergence; the mutation probability is set to 0.1, as a low mutation probability maintains population stability while preventing premature convergence of the algorithm.

[0090] The objective function formula is as follows:

[0091] ;

[0092] In the formula, For scheduling scheme The overall fitness value; The calculation scheme is a function for the number of casualties. The estimated number of casualties for all affected personnel; The calculation scheme is a function of the average arrival time. The weighted average arrival time of all available rescue resources; The calculation scheme is a resource utilization function. The ratio of the amount of resources already scheduled to the total amount of available resources; The weighting coefficients are determined using the analytic hierarchy process (AHP) and satisfy the following conditions: .

[0093] The solver outputs the initial scheduling scheme to step S6.

[0094] S6. Based on the difference between the spatial entity state vector and the risk evolution trajectory, update the avoidance constraint set, perform rolling corrections on the initial scheduling scheme, and obtain the emergency scheduling result.

[0095] The rolling correction process first re-executes step S1, collects multi-source spatial data in real time and performs spatiotemporal alignment and fusion processing to obtain the updated spatial entity state vector.

[0096] The deviation between the updated spatial entity state vector and the risk evolution trajectory is calculated using the following formula:

[0097] ;

[0098] In the formula, This refers to the degree of deviation from the situation. This represents the total number of spatial entity nodes. For the first in the risk evolution trajectory Risk level prediction value for each node; For the updated spatial entity state vector, the th The measured risk level of each node.

[0099] Simultaneously, the projection confidence level is calculated, and the projection confidence level decay formula is as follows:

[0100] ;

[0101] In the formula, For the first time after the start of the deduction Confidence level of the deduction within minutes; The initial projection confidence level is set at 0.95. Since the initial projection is based on the current complete situation data, the data completeness is high, hence the confidence level is relatively high. The attenuation coefficient is dynamically set according to the type of disaster. For fire scenarios, it is set to 0.1 per minute because the fire spread speed is relatively uniform and the situation evolution is highly predictable. For flood scenarios, it is set to 0.15 per minute because floods are affected by the uncertainty of rainfall and the situation evolution is moderately predictable. For hazardous chemical leakage scenarios, it is set to 0.2 per minute because the diffusion of hazardous chemicals is severely affected by meteorological conditions and the situation evolution is less predictable. The time span from the start time to the current time is calculated in minutes.

[0102] Trigger condition determination: when Below 0.6 or When the deviation exceeds 0.3, a re-analysis process is triggered. The threshold of 0.6 is selected because the reliability of the simulation results is insufficient when the value is below this, and the probability of misjudgment increases significantly when continuing to execute the original plan; the threshold of 0.3 is selected because a deviation of more than 30% indicates that there is a structural difference between the actual situation and the simulation baseline, and the plan needs to be revised.

[0103] The constraint relaxation identifier monitors the tightness of each constraint in the constraint avoidance set and calculates the ratio of the actual resource margin to the constraint upper limit for each constraint, as shown in the following formula:

[0104] ;

[0105] In the formula, For the first The relaxation ratio of the constraints; For the first The actual available resources or environmental capacity corresponding to each constraint; For the first The upper limit of the constraint avoidance limit. When When the value is less than 0.5, the constraint is considered to be understrength and can be relaxed appropriately to free up optimization space. The relaxation ratio threshold of 0.5 is selected based on the fact that when the actual available quantity is less than 50% of the upper limit of the constraint, it indicates that there is significant leeway in the constraint, and moderate relaxation will not jeopardize the safety boundary.

[0106] When updating the avoidance constraint set, the relaxed threshold boundary is replaced with the original constraint value to obtain the updated avoidance constraint set.

[0107] When performing rolling corrections on the initial scheduling scheme, the updated avoidance constraint set is used as a hard constraint condition. Combined with the updated spatiotemporal knowledge network topology, the scheduling scheme solution in step S5 is re-executed to obtain the emergency scheduling result.

[0108] Reference Figure 2 Based on the same inventive concept, this application also provides an emergency single-map dispatch system based on knowledge graphs, including:

[0109] The data fusion module is used to collect multi-source spatial data in real time, perform spatiotemporal alignment and fusion processing on the multi-source spatial data, and obtain spatial entity state vectors.

[0110] The network construction module is used to construct a spatiotemporal knowledge network based on the spatial entity state vector. In the spatiotemporal knowledge network, spatial entity nodes, semantic relation edges, taboo state nodes, and risk transmission reverse edges are established to obtain the spatiotemporal knowledge network topology.

[0111] The risk simulation module is used to generate candidate scheduling schemes based on the spatial entity state vectors. Taking the candidate scheduling schemes and the spatiotemporal knowledge network topology as input, it performs multi-step risk propagation simulation along semantic relationship edges to obtain the risk evolution trajectory.

[0112] The constraint extraction module is used to identify activated taboo state nodes from the risk evolution trajectory. Starting from the activated taboo state nodes, the spatiotemporal knowledge network topology is traversed in reverse along the risk propagation reverse edge to extract the trigger threshold boundary and obtain the avoidance constraint set.

[0113] The scheme solving module is used to solve the scheduling scheme by taking the avoidance constraint set as hard constraint conditions and combining it with the spatiotemporal knowledge network topology to obtain the initial scheduling scheme.

[0114] The dynamic correction module is used to update the avoidance constraint set based on the difference between the spatial entity state vector and the risk evolution trajectory, and to perform rolling correction on the initial scheduling scheme to obtain the emergency scheduling result.

[0115] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. An emergency single-graph scheduling method based on knowledge graphs, characterized in that, Includes the following steps: Real-time acquisition of multi-source spatial data, spatiotemporal alignment and fusion processing of multi-source spatial data to obtain spatial entity state vectors; A spatiotemporal knowledge network is constructed based on the spatial entity state vector. Spatial entity nodes, semantic relation edges, taboo state nodes, and risk transmission inverse edges are established in the spatiotemporal knowledge network to obtain the spatiotemporal knowledge network topology. Candidate scheduling schemes are generated based on the spatial entity state vectors. Using the candidate scheduling schemes and the spatiotemporal knowledge network topology as input, multi-step risk propagation simulation is performed along the semantic relationship edges to obtain the risk evolution trajectory. Identify activated taboo state nodes from the risk evolution trajectory, and starting from the activated taboo state nodes, traverse the spatiotemporal knowledge network topology in reverse along the risk propagation reverse edge to extract the trigger threshold boundary and obtain the avoidance constraint set. By using the avoidance constraint set as a hard constraint condition and combining it with the spatiotemporal knowledge network topology, the scheduling scheme is solved to obtain the initial scheduling scheme. Based on the difference between the spatial entity state vector and the risk evolution trajectory, the avoidance constraint set is updated, and the initial scheduling scheme is rolled over to obtain the emergency scheduling result.

2. The emergency single-graph scheduling method based on knowledge graphs according to claim 1, characterized in that, Real-time acquisition of multi-source spatial data, spatiotemporal alignment and fusion processing of the multi-source spatial data to obtain spatial entity state vectors, including: By integrating meteorological monitoring data, traffic flow data, video surveillance data, social media sentiment data, and data transmitted from on-site sensors, multi-source spatial data is obtained. Spatiotemporal alignment and fusion processing of multi-source spatial data are performed to obtain spatial entity state vectors that include event point coordinates, resource point reserves, path segment accessibility, and shelter capacity.

3. The emergency single-graph scheduling method based on knowledge graphs according to claim 1, characterized in that, A spatiotemporal knowledge network is constructed based on the spatial entity state vectors. Spatial entity nodes, semantic relation edges, taboo state nodes, and risk propagation inverse edges are established within the spatiotemporal knowledge network, resulting in the spatiotemporal knowledge network topology, including: Spatial entity nodes are generated based on spatial entity state vectors, and semantic tag attributes and spatiotemporal coordinate attributes are configured for each spatial entity node. Establish semantic relationship edges and configure spatiotemporal triggering conditions for each semantic relationship edge. The spatiotemporal triggering conditions include time window, spatial range and threshold conditions. Establish taboo state nodes, and build reverse risk transmission edges from the taboo state nodes to obtain the spatiotemporal knowledge network topology.

4. The emergency single-graph scheduling method based on knowledge graphs according to claim 1, characterized in that, Candidate scheduling schemes are generated based on the spatial entity state vectors. Using the candidate scheduling schemes and the spatiotemporal knowledge network topology as input, a multi-step risk propagation simulation is performed along semantic relation edges to obtain the risk evolution trajectory, including: Based on the resource availability and path accessibility in the spatial entity state vector, a set of resource allocation actions is generated to obtain candidate scheduling schemes; Starting from the resource allocation node corresponding to the candidate scheduling scheme, traverse all outgoing semantic relationship edges; Check the spatiotemporal triggering conditions on the semantic relationship edge. When the time window includes the inference time, the spatial range includes the affected area, and the threshold condition is exceeded, activate the next level node and record the risk level increment. The simulation process is advanced by dividing the data into preset time slices to obtain the risk evolution trajectory.

5. The emergency single-graph scheduling method based on knowledge graphs according to claim 1, characterized in that, Identify activated taboo state nodes from the risk evolution trajectory. Starting from these activated taboo state nodes, traverse the spatiotemporal knowledge network topology in reverse along the risk propagation inverse edge to extract trigger threshold boundaries and obtain the avoidance constraint set, including: Identify activated taboo state nodes from the risk evolution trajectory; Starting from the activated taboo state node, traverse the spatiotemporal knowledge network topology in reverse along the risk propagation reverse edge, with the search depth limited to a preset number of layers; Record the complete path from the taboo state node back to the scheduling action node, and extract the threshold boundaries of the spatiotemporal triggering conditions on the path; Transform the threshold boundary into an avoidance constraint set.

6. The emergency single-graph scheduling method based on knowledge graphs according to claim 1, characterized in that, By treating the avoidance constraint set as a hard constraint and combining it with the spatiotemporal knowledge network topology, the scheduling scheme is solved to obtain the initial scheduling scheme, including: The avoidance constraint set is used as a hard constraint condition, which together with the regular constraints constitutes a mixed constraint condition. Under mixed constraints, a multi-objective evolutionary solution is performed to obtain an initial scheduling scheme.

7. The emergency single-graph scheduling method based on knowledge graphs according to claim 1, characterized in that, Based on the difference between the spatial entity state vector and the risk evolution trajectory, including: Real-time acquisition of multi-source spatial data, spatiotemporal alignment and fusion processing of multi-source spatial data to obtain updated spatial entity state vectors; Calculate the deviation between the updated spatial entity state vector and the risk evolution trajectory; When the deviation exceeds the preset threshold, the re-analysis process is triggered.

8. The emergency single-graph scheduling method based on knowledge graphs according to claim 7, characterized in that, Update the avoidance constraint set, including: Monitor the tightness of each constraint in the constraint avoidance set and identify constraints that can be appropriately relaxed; Release the optimization space to obtain the updated avoidance constraint set.

9. The emergency single-graph scheduling method based on knowledge graphs according to claim 8, characterized in that, The initial scheduling scheme is subject to rolling revisions, including: The updated avoidance constraint set is used as a hard constraint condition, and the scheduling scheme is solved by combining the spatiotemporal knowledge network topology to obtain the emergency scheduling result.

10. An emergency single-map dispatch system based on knowledge graphs, characterized in that, include: The data fusion module is used to collect multi-source spatial data in real time, perform spatiotemporal alignment and fusion processing on the multi-source spatial data, and obtain spatial entity state vectors. The network construction module is used to construct a spatiotemporal knowledge network based on the spatial entity state vector. In the spatiotemporal knowledge network, spatial entity nodes, semantic relation edges, taboo state nodes, and risk transmission reverse edges are established to obtain the spatiotemporal knowledge network topology. The risk simulation module is used to generate candidate scheduling schemes based on the spatial entity state vectors. Taking the candidate scheduling schemes and the spatiotemporal knowledge network topology as input, it performs multi-step risk propagation simulation along semantic relationship edges to obtain the risk evolution trajectory. The constraint extraction module is used to identify activated taboo state nodes from the risk evolution trajectory. Starting from the activated taboo state nodes, the spatiotemporal knowledge network topology is traversed in reverse along the risk propagation reverse edge to extract the trigger threshold boundary and obtain the avoidance constraint set. The scheme solving module is used to solve the scheduling scheme by taking the avoidance constraint set as hard constraint conditions and combining it with the spatiotemporal knowledge network topology to obtain the initial scheduling scheme. The dynamic correction module is used to update the avoidance constraint set based on the difference between the spatial entity state vector and the risk evolution trajectory, and to perform rolling correction on the initial scheduling scheme to obtain the emergency scheduling result.

Images