An emergency event state classification method and system

Abstract

The application belongs to the technical field of data processing, and discloses an emergency event state classification method and system; the method comprises the following steps: obtaining a time sequence state graph, a plan target graph, a disposal operation sequence and a complete reference frame library according to multi-source states, emergency plans and historical disposal records; forming a complete reference frame record or performing a prediction determination according to the graph similarity of a current state graph, a plan target graph and a reference state graph; generating a candidate prediction graph through each operation prefix of the disposal operation sequence, forming a prediction frame record containing an operation cut-off serial number and a prediction residual; after forming a backward reference frame, generating an interpolation graph according to a forward reference frame and the backward reference frame, forming an interpolation frame record or a complete reference frame record; and establishing an inter-frame dependence graph and restoring a state graph of a target sampling time according to the three types of frame records; the application is used for reducing the repeated complete storage of the state graph and maintaining the state restoration relationship.

Description

Technical Field

[0001] This application relates to the field of data processing technology, and more specifically, to a method and system for classifying the status of emergency events. Background Technology

[0002] During emergency response, sensors, video analytics systems, operational systems, and manual reporting channels continuously generate multi-source status data, including equipment status, personnel distribution, resource allocation, impact scope, and response progress. Furthermore, relationships such as attribution, connectivity, impact propagation, and resource accessibility exist between different status objects. Therefore, the emergency state at a single sampling moment typically needs to be represented by a state graph containing nodes, edges, and corresponding attributes, forming a continuously evolving sequence of state graphs according to the sampling time. To meet the requirements of event review, accountability, contingency plan assessment, and historical status query, the emergency management system needs to permanently store the state graph sequence and be able to restore the event state at any target sampling moment.

[0003] Existing emergency event state graph management methods typically save nodes, edges, and attributes completely at each sampling moment, or retain a portion of the complete state graph at fixed time intervals, while saving incremental changes for adjacent state graphs. The complete saving method allows direct reading of historical states, but as the sampling frequency, the number of state objects, and the event duration increase, duplicate nodes, duplicate relationships, and attributes with small changes will continuously occupy storage space. Fixed-interval sampling or adjacent difference storage methods primarily determine the storage content based on the sampling time or the amount of local data change, without distinguishing whether the state change is a predictable change caused by the emergency response process or a continuous transition between two known states.

[0004] Therefore, existing technologies struggle to determine whether the current state diagram must be completely preserved, or whether it can be accurately reconstructed based on existing state information, according to the actual evolution of the emergency event's state. For example, in a smoke incident at a rail transit station, equipment alarms, personnel evacuation, passage closures, and the arrival of rescue resources can generate a large number of continuous state diagrams in a short period. If all of them are preserved completely, it would result in a large amount of duplicate diagram structures and duplicate attribute data. If compression is only performed according to fixed intervals or adjacent differences, the transition points of the response phase may be treated the same as ordinary transition states, leading to the restored historical state not accurately reflecting key nodes, object relationships, or attribute changes. Therefore, existing technologies face the problem of not being able to implement differentiated storage of continuous emergency event state diagrams while ensuring that the state diagram can be accurately reconstructed at any sampling moment.

[0005] In view of this, this application proposes an emergency event status classification method and system to solve the above problems. Summary of the Invention

[0006] To address the shortcomings of existing technologies, the purpose of this application is to provide an emergency event state classification method and system. By sequentially performing complete reference frame determination, prediction frame determination, and interpolation frame determination on the current state graph, complete reference frame records, prediction frame records, or interpolation frame records are formed. Based on the three types of frame records, an inter-frame dependency graph is established and the complete reference frame library is updated. This achieves the goals of reducing the repeated and complete storage of nodes, edges, and attributes, restoring the state graph at the target sampling time, and supporting subsequent state graph classification.

[0007] This application provides the following technical solution: an emergency event status classification method, including:

[0008] Acquire multi-source status data, emergency plans, and historical response records for the same emergency event; construct a time-series status diagram based on the multi-source status data; obtain the plan target map, plan identifier, and response operation sequence from the emergency plan; and establish a complete reference frame library based on the historical response records.

[0009] Read the current state diagram from the time sequence state diagram and generate the current frame identifier. Calculate the graph similarity between the current state diagram and the target diagram of the plan and the complete state diagram in the complete reference frame library. Based on the graph similarity judgment result, generate a complete reference frame record for the current state diagram or perform prediction frame judgment.

[0010] Select a complete state map from the complete reference frame library, process the complete state map according to the disposal operation sequence to obtain a prediction map, compare the prediction map with the current state map to obtain prediction similarity and prediction residual, generate a prediction frame record containing reference frame identifier, contingency plan identifier and prediction residual according to the prediction similarity determination result, or retain the current state map and perform interpolation frame determination.

[0011] Obtain reference frame records before and after the current state map, recover the forward and backward reference state maps based on the reference frame records, obtain an interpolation map based on the two reference state maps and the sampling time, compare the interpolation map with the current state map to obtain interpolation similarity and interpolation residual, and generate an interpolation frame record containing two reference frame identifiers, interpolation parameters and interpolation residual based on the interpolation similarity determination result, or generate a complete reference frame record based on the current state map;

[0012] Write the current frame identifier and sampling time into the corresponding frame record. Establish an inter-frame dependency graph based on the reference relationships in the complete reference frame record, the predicted frame record, and the interpolated frame record. Restore the state graph of the target sampling time based on the frame record corresponding to the target sampling time and the inter-frame dependency graph.

[0013] An emergency event status classification system, implementing the aforementioned emergency event status classification method, includes:

[0014] The data preparation module is used to acquire multi-source status data, emergency plans and historical handling records for the same emergency event, construct a time-series status diagram based on the multi-source status data, obtain the plan target diagram, plan identifier and handling operation sequence from the emergency plan, and establish a complete reference frame library based on the historical handling records.

[0015] The complete reference frame determination module is used to read the current state diagram from the time-series state diagram and generate the current frame identifier, calculate the graph similarity between the current state diagram and the target diagram of the plan and the complete state diagram in the complete reference frame library, and generate a complete reference frame record or perform prediction frame determination based on the graph similarity determination result.

[0016] The prediction frame determination module is used to select a complete state map from the complete reference frame library, process the complete state map according to the disposal operation sequence to obtain a prediction map, compare the prediction map with the current state map to obtain the prediction similarity and prediction residual, generate a prediction frame record containing the reference frame identifier, the contingency plan identifier and the prediction residual according to the prediction similarity determination result, or retain the current state map and perform interpolation frame determination.

[0017] The interpolation frame determination module is used to obtain reference frame records before and after the current state map, recover the forward reference state map and the backward reference state map based on the reference frame records, obtain the interpolation map based on the two reference state maps and the sampling time, compare the interpolation map with the current state map to obtain the interpolation similarity and interpolation residual, and generate an interpolation frame record containing the two reference frame identifiers, interpolation parameters and interpolation residual based on the interpolation similarity determination result, or generate a complete reference frame record based on the current state map;

[0018] The state recovery module is used to write the current frame identifier and sampling time into the corresponding frame record, establish an inter-frame dependency graph based on the reference relationship in the complete reference frame record, the prediction frame record and the interpolation frame record, and restore the state graph of the target sampling time based on the frame record and the inter-frame dependency graph corresponding to the target sampling time.

[0019] The technical effects and advantages of the emergency event status classification method and system proposed in this application are as follows:

[0020] This application uses the graph similarity between the current state graph, the target plan graph, and the reference state graphs in the complete reference frame library to form complete reference frame records for state graphs that need to be stored independently. By applying different operation prefixes of the disposal operation sequence to the target reference state graph, the prediction graph and operation deadline number corresponding to the current state graph are determined, so that the prediction frame can record the disposal progress corresponding to the current state, and the state graph is restored based on the target complete reference frame, the target plan identifier, the operation deadline number, and the prediction residual. For state graphs that fail the prediction judgment, interpolation judgment is performed after the backward reference frame is formed, so that the forward reference frame and the backward reference frame have a clear temporal relationship. By establishing an inter-frame dependency graph through three types of frame records, the state graph at the target sampling time can be restored along the reference relationship, thereby reducing the storage of repeated nodes, edges, and attributes in the continuous state graph. Attached Figure Description

[0021] Figure 1 This is a schematic diagram of an emergency event status classification method according to this application;

[0022] Figure 2 A schematic diagram of a scenario constructed for the emergency event sequence state diagram in this application;

[0023] Figure 3 This is an example diagram showing the frame classification results and inter-frame dependencies of the emergency event state diagram in this application.

[0024] Figure 4 This is a schematic diagram of the module connections of the emergency event status diagram classification management system in this application. Detailed Implementation

[0025] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0026] Example 1:

[0027] Please see Figure 1-4 As shown, for ease of use Figure 3 This indicates that the complete reference frame, prediction frame, and interpolation frame are in Figure 3In this embodiment, I, P, and B are used as graphic symbols, representing only three types of state graph frame records and not video encoded frames. This embodiment provides an emergency event state classification method applicable to emergency state recording, compressed storage, historical backtracking, and contingency plan selection in urban rail transit, chemical industrial parks, natural disaster response, and security command for large-scale events. The execution unit includes a state acquisition unit, a state graph construction unit, a contingency plan parsing unit, a complete reference frame library initialization unit, a frame classification unit, a prediction unit, an interpolation unit, an inter-frame dependency graph unit, a state recovery unit, and a complete reference frame library management unit. The complete reference frame records, predicted frame records, and interpolated frame records are stored in corresponding... The storage area is maintained by a complete reference frame library management unit and an inter-frame dependency graph unit; the state recovery module includes a frame record writing unit, an inter-frame dependency graph unit, and a state recovery unit; the emergency command system allows the write time granularity, replenishment period, maintenance period, reference statistics window, and state graph retention period to be read from the storage policy configuration record; the replenishment period and maintenance period are both represented as positive integer multiples of the unified sampling period, and the reference statistics window and state graph retention period are both greater than zero; the starting point of the observation time window is the emergency event creation time, and the ending point is the event closure time or operation recovery time recorded by the business system; when the emergency event has not yet been closed, the current sampling time is used as the temporary end point of the observation time window. When the reference statistics window is not configured separately, the validity period of the complete reference frame is taken; the method includes:

[0028] Acquire multi-source status data, emergency plans, and historical response records for the same emergency event; construct a time-series status diagram based on the multi-source status data; obtain the plan target map, plan identifier, and response operation sequence from the emergency plan; and establish a complete reference frame library based on the historical response records.

[0029] In specific implementations, the methods for obtaining the timing state diagram, the contingency target diagram, the contingency identifier, the handling operation sequence, and the complete reference frame library include:

[0030] The status acquisition unit collects sensor status, video analysis results, manual reporting records, and business system status under the same emergency event according to a unified sampling period. The unified sampling period is determined by reading the sensor acquisition period, video analysis output period, manual reporting refresh period, and business system refresh period. The maximum value among the refresh periods is taken as the base period, and the base period is rounded up to the time granularity allowed to be written by the emergency command system. The status acquisition unit writes the acquisition records within the same sampling time into the sampling record set according to the event number, sampling time, and source type. The sampling record set enters the status diagram construction unit.

[0031] The event type and current handling stage of the current emergency event are read from the event base record and handling stage change record in the business system. When there are multiple event types in the event base record, the last event type that has been reviewed and confirmed is used. The current handling stage is the handling stage change record with the latest timestamp before the sampling time. If there is no handling stage change record before the sampling time, the current handling stage is recorded as the initial abnormal stage. When the event type cannot be uniquely determined, the contingency plan target map matching is not automatically performed, and the current status map is written to the pending review record.

[0032] The state diagram construction unit reads the site code, equipment code, personnel area code, emergency resource point code, and command node code from the sampling record set; the object identifier is formed by concatenating the object source system number, object type, and object code; the object category, object name, line, area, and source type are written into the node basic fields; and the site node, equipment node, personnel area node, emergency resource point node, and command node are written into the node set.

[0033] The state diagram construction unit reads line connection relationships, equipment ownership relationships, impact propagation relationships, resource reachability relationships, and command and transmission relationships from the sampling record set and service configuration records; the edge identifier is formed by concatenating the starting object identifier, the ending object identifier, the relationship type, and the relationship source number; the direction, distance, capacity, communication path, impact direction, and source record are written into the edge basic fields to obtain the edge set.

[0034] The state graph construction unit writes equipment status, congestion level, personnel density, temperature, smoke concentration, resource quantity, and disposal progress into node attributes according to object identifiers; and writes traffic capacity, communication delay, physical distance, resource scheduling time, and impact intensity into edge attributes according to edge identifiers; when multiple sources of state parameters exist for the same object at the same sampling time, the source priority is determined according to the order of sensor status, video analysis results, business system status, and manual reporting records, and the state parameter with the highest source priority is written; when multiple state parameters exist for the same source, the state parameter with the timestamp closest to the sampling time is read.

[0035] The state graph construction unit combines the node set, edge set, node attributes, and edge attributes into a state graph corresponding to the sampling time, and arranges them in ascending order of sampling time to form a time-series state graph sequence; the emergency state graph is represented as follows: ;in, Represents an emergency status diagram. This represents a set of nodes, such as sites, devices, and people. Represents a set of edges, such as those with connectivity and influence relationships. This represents a matrix of node attributes, such as device status and pedestrian traffic. This represents the edge attribute matrix, such as flow and latency. Represents the dimension of node attributes. The edge attribute dimension is represented; the temporal state diagram sequence is represented as: ;in, Indicates the sampling time The corresponding emergency status diagram, This indicates the number of state diagrams within the observation time window.

[0036] The contingency plan parsing unit prioritizes reading the contingency plan identifier, event type, response stage, affected object code, response object code, resource scheduling object code, target state constraint, and response action fields from the structured form of the emergency plan. It then performs a consistency comparison between the affected object code, response object code, and resource scheduling object code and the object identifier of the state node. If the object code corresponds to a state node, the corresponding object is written into the contingency plan target node set; if the object code does not correspond to a state node, the corresponding field is written into the record to be reviewed. The contingency plan identifier is the unique identifier of the emergency plan. The target contingency plan identifier is the contingency plan identifier of the contingency plan to which the target response operation sequence belongs, determined by comparing the graph similarity between the candidate prediction graph and the current state graph. The matching contingency plan identifier is the contingency plan identifier of the target graph to which the contingency plan with the highest graph similarity to the current state graph belongs during the complete reference frame determination process.

[0037] The contingency plan parsing unit breaks down the target state constraints into target object code, target attribute name, target attribute value, and target effective stage; it writes the target attribute value into the target node attribute or target edge attribute according to the target object code, and writes the target effective stage into the stage field of the contingency plan target graph; when performing contingency plan target graph matching in the future, only contingency plan target graphs whose event type and stage field are consistent with the current event type and current handling stage are used as contingency plan target graphs to be compared.

[0038] When the contingency plan text includes the starting object code, the ending object code, and the relationship type, the contingency plan target edge is established based on the three fields; when the contingency plan text only provides the target edge attributes but not the relationship type, the existing relationship type between the starting object and the ending object is read from the business configuration record or the current state diagram, and the corresponding relationship is written into the contingency plan target diagram; when there are multiple relationship types or the starting object and the ending object cannot be determined, the target edge is not automatically formed, and the corresponding target state constraint is written into the record to be reviewed.

[0039] The contingency plan analysis unit breaks down the action field into execution order, action object code, action type, and action parameters, and arranges them in the execution order to form a sequence of actions; the action types include node addition, node deletion, node attribute modification, adding, deleting, modifying attributes, and resource scheduling.

[0040] In one optional implementation, when the emergency plan does not provide an object code but only records the object name, the plan parsing unit establishes an object name mapping table; it reads the object source system number, object type, object code, and standard object name from the object basic record of the business system, and reads the historical name, abbreviation, and alias corresponding to the standard object name from the reviewed emergency plan, historical handling records, and manual correction records; it establishes a mapping between the standard object name, historical name, abbreviation, and alias and the same object code, deletes duplicate records with the same name and object code, and obtains the object name mapping table; when the same name corresponds to multiple object codes, the corresponding name is marked as an ambiguous name, the object code is not automatically determined based on the ambiguous name, and the text unit containing the ambiguous name is written into the record to be reviewed.

[0041] The method for constructing the action description mapping table is as follows: Seven action types are set: node addition, node deletion, node attribute modification, edge addition, edge deletion, edge attribute modification, and resource scheduling. Action descriptions and state changes before and after action execution are read from reviewed emergency plans and historical execution records. When the number of objects increases, the action description is mapped to node addition; when the number of objects decreases, it is mapped to node deletion; when object attribute values ​​change, it is mapped to node attribute modification; when a new relationship is added between objects, it is mapped to edge addition; when a relationship is dissolved between objects, it is mapped to edge deletion; when relationship attribute values ​​change, it is mapped to edge attribute modification; and when the number of resources, resource location, or the object to which a resource points changes, it is mapped to resource scheduling. The reviewed and confirmed action descriptions and corresponding action types are written into the action description mapping table. When the same action description corresponds to multiple action types, the corresponding action description is marked as an ambiguous action description, and no disposal operation is automatically generated based on the ambiguous action description.

[0042] The contingency plan analysis unit divides the unstructured contingency plan text into text units according to periods, semicolons, line breaks, and table rows. Based on character length from longest to shortest, consecutive characters that exactly match the names in the object name mapping table are identified as object name candidates, and their corresponding object codes are read. Similarly, based on character length from longest to shortest, consecutive characters that exactly match the action descriptions in the action description mapping table are identified as action description candidates, and their corresponding action types are read. When a text unit contains only one object name candidate and one action description candidate, the corresponding object code is used as the action object code. When a text unit contains multiple object name candidates, the object name candidate with the shortest character distance after the action description candidate is selected first; if no object name candidate follows the action description candidate, the object name candidate with the shortest character distance before the action description candidate is selected. If two object name candidates have the same character distance as the action description candidate, or if no object name candidate or action description candidate is identified, no action is automatically generated, and the corresponding text unit is written to the pending review record.

[0043] When step numbers or sequence numbers exist in the unstructured contingency plan text, the execution order is determined by ascending order of step numbers or sequence numbers; when step numbers or sequence numbers do not exist, the execution order is determined by the order in which the text units appear in the contingency plan text. Action parameters are recorded according to action type: Node addition record: new node identifier, node type, and initial node attributes; Node deletion record: target node identifier; Node attribute modification record: target node identifier, attribute name, and target attribute value; Edge addition record: starting node identifier, ending node identifier, edge type, and initial edge attribute; Edge deletion record: edge identifier, or record starting node identifier, ending node identifier, and edge type; Edge attribute modification record: edge identifier, attribute name, and target attribute value; Resource scheduling record: resource object identifier, target object identifier, resource quantity, target location, and scheduling time. When the required fields for a corresponding action type are missing, a disposal operation is not automatically generated, and the corresponding text unit is written to the pending review record.

[0044] The object name mapping table and action description mapping table are updated based on the review results in the records to be reviewed. When the review results confirm that the new object name corresponds to the existing object code, the new object name and the corresponding object code are written into the object name mapping table. When the review results confirm that the new action description corresponds to the existing action type, the new action description and the corresponding action type are written into the action description mapping table. The new mapping is used starting from the contingency plan analysis task after the update time, and does not retroactively modify the contingency plan target map and disposal operation sequence formed before the update time.

[0045] The contingency plan database is represented as follows: ;in, Indicates the contingency plan database. Indicates the first Each contingency plan identifier, Indicates the first Each contingency plan corresponds to a target map. Indicates the first The sequence of procedures for handling each contingency plan. Indicates the first The priority of each contingency plan This indicates the total number of contingency plans; the contingency plan database is indexed by event type, response stage, and scope of application; the database is set up as follows: read approved emergency plans, historical response review records, and expert-confirmed exercise records; write the plan number, event type, response stage, target status diagram, response operation sequence, plan priority, and source record into the database; when multiple plans exist for the same event type, indexes are created separately according to the plan's scope of application, response stage, and priority.

[0046] The complete reference frame library management unit establishes an initial complete reference frame library based on historical handling records. The contingency plan target graph is used for contingency plan matching similarity calculation in the complete reference frame determination and is not directly written into the complete reference frame library. The unit reads the state graphs in the historical handling records that are identified as initial anomaly, response escalation, critical turning point, resumption of operation, or baseline operation, and writes the complete node set, complete edge set, node attributes, and edge attributes of the corresponding state graphs into the complete reference frame record. The contingency plan target graph is only used for contingency plan matching similarity calculation in the complete reference frame determination and is not directly written into the complete reference frame library as a complete reference frame. The unit assigns a frame identifier to each complete reference frame record according to the event number, sampling time, and intra-event frame sequence number, and writes the frame identifier, event type, handling stage, complete graph structure, complete attributes, source record, matching contingency plan identifier, writing time, last reference time, and reference count into the initial complete reference frame library.

[0047] The upper limit of the complete reference frame library size is determined based on the storage capacity allocated to the online complete reference frame library and the average storage size of a single complete reference frame, calculated as follows: ;in, This indicates the upper limit of the complete reference frame library size. This indicates the storage capacity allocated to the online complete reference frame library; This indicates the average amount of storage that has been written to a complete reference frame; This indicates rounding down; when a complete reference frame has not yet been formed, the upper limit of the complete reference frame library size is calculated after the first complete reference frame is written.

[0048] Read the current state diagram from the time-series state diagram and generate the current frame identifier. Calculate the graph similarity between the current state diagram and the target diagram of the plan and the complete state diagram in the complete reference frame library. Based on the graph similarity judgment result, generate a complete reference frame record from the current state diagram or perform prediction frame judgment.

[0049] In specific implementations, methods for generating complete reference frame records or performing prediction frame determination include:

[0050] The frame classification unit reads the current state diagram from the temporal state diagram sequence according to the sampling time. Generate the current frame identifier according to the event number, current sampling time, and event intraframe sequence number; and convert the current state diagram. Compare with each contingency target map in the contingency plan library. and each complete reference frame in the complete reference frame library To form a pair of state diagrams to be compared, for any pair of state diagrams to be compared, let the first state diagram be... The second state diagram is , First state diagram The set of nodes, First state diagram The set of edges, This is the node attribute matrix of the first state graph. This is the edge attribute matrix of the first state graph. For the second state diagram The set of nodes, For the second state diagram The set of edges, This is the node attribute matrix of the second state graph. This is the edge attribute matrix of the second state graph.

[0051] Frame classification unit reading and Find pairs of nodes with the same object identifier and write them into the directly aligned node set. For nodes that are not included in the direct alignment node set, calculate the candidate similarity values ​​for supplementary nodes, and establish a supplementary alignment node set according to the candidate similarity values ​​for supplementary nodes. The directly aligned node set and the supplementary aligned node set are merged to form the node alignment set. .

[0052] For nodes that are not directly aligned in the first state graph Nodes not directly aligned in the second state graph Node type consistency value Determined based on whether the node types are the same; when the node types are the same... When the node types are different Node name similarity value according to Calculate, where, Represents a node The set of node name word segments, Represents a node The set of node name word segments; when and When at least one node name is empty, Take 0.

[0053] The node name segmentation set is obtained by performing the longest match segmentation through the object name mapping table; consecutive characters that do not match the object name mapping table are segmented according to a single Chinese character, consecutive numbers, or consecutive letters; repeated words are only retained once.

[0054] Node attribute similarity value according to Calculate; where, This represents the normalized distance value of a node attribute; normalized distance value of a node attribute according to Calculation; where the set of node attributes involved in the comparison is the union of the names of the two node attribute fields. This indicates the number of node attributes involved in the comparison. Indicates the first Normalized distance of each node attribute; when the set of node attributes is empty. When the first When the first attribute exists in only one node to be compared, the second attribute... The normalized distance for each attribute is set to 1; attributes where neither node exists are not included in the attribute set for comparison; when the maximum level value of a rank attribute equals the minimum level value, the normalized distance is 0 if the two attribute values ​​are the same, and 1 if the two attribute values ​​are different; the handling of missing edge attributes and zero range is the same as for node attributes; when both node sets of the two state graphs are empty, the node coverage similarity is set to 1; when only one node set is empty, the node coverage similarity is set to 0; when both edge sets of the two state graphs are empty, the edge coverage similarity is set to 1; when only one edge set is empty, the edge coverage similarity is set to 0.

[0055] When the node attribute is a numeric attribute ;in, express The attribute value of the j-th node in the middle. express The Middle Each node attribute value, Indicates the first Normalized range of each node attribute; normalized range Prioritize using the difference between the allowed maximum and minimum values ​​of the corresponding attribute in the business configuration table. If a corresponding allowed range does not exist in the business configuration record, ... Take the difference between the 95th percentile and the 5th percentile of the corresponding attribute in the historical samples; when equal to 0 and equal hour, Take 0; when equal to 0 and Not equal to hour, Set to 1; when the node attribute is a hierarchical attribute, ;in, Indicates the first The maximum level value of a level-type attribute. Indicates the first The minimum level value for a level-based attribute; when the node attribute is an enumeration attribute or a text label attribute. and When the same Take 0, and Not at the same time Take 1.

[0056] Supplementing node candidate similarity values according to Calculate; where, Indicates the candidate weights for node type consistency values. This indicates the candidate weights for node name similarity values. This represents the candidate weights for node attribute similarity values, and When setting weights, select node pairs from the historical state graph that are manually confirmed to belong to the same object but whose object identifiers are missing or changed as positive samples, and select node pairs that are manually confirmed not to belong to the same object as negative samples; enumerate candidate node weight combinations that satisfy the condition that the sum of their weights is one; for each candidate node weight combination, calculate the supplementary candidate similarity values ​​for all positive and negative samples, and form a candidate node alignment threshold set by combining zero, one, and non-repeating supplementary candidate similarity values; for each candidate node weight combination and each candidate node alignment threshold, perform one-to-one node alignment according to the candidate similarity values ​​from largest to smallest, and calculate the correct alignment ratio of positive samples and the exclusion ratio of negative samples; select the candidate node weight combination and candidate node alignment threshold with the largest average ratio; when multiple combinations correspond to the same average value, first select the combination with the larger weight for the same node type, and then select the candidate node alignment threshold with the larger value; when there are no historical samples, , , .

[0057] The frame classification unit sorts the candidate pairs of supplementary nodes in descending order of their candidate similarity values; reads the candidate pairs of nodes sequentially; when the candidate similarity value of the supplementary nodes is not less than the node supplementary alignment threshold and neither node is occupied by other candidate pairs, the candidate pairs of nodes are written into the supplementary alignment node set; when multiple candidate thresholds correspond to the same average value, the candidate threshold with the larger value is selected; when there are no historical samples, the node supplementary alignment threshold is set to 0.60.

[0058] The edge alignment process is as follows: Frame classification unit reads... and Find edge pairs with the same edge identifier and add them to the directly aligned edge set. For edges not included in the directly aligned edge set For each edge, calculate the candidate similarity value of the supplementary edge, and establish a set of supplementary aligned edges based on the candidate similarity values. Directly aligned edge set and supplementary alignment edge set Merge to form an edge-aligned set .

[0059] The calculation process for supplementary edge candidate similarity values ​​is as follows: For the first state graph Any side that is not directly aligned Second state diagram Any side that is not directly aligned When the starting nodes of two edges form an aligned node pair in the node alignment set, the starting node alignment value is... ,otherwise When the endpoints of two edges form an aligned node pair in the node alignment set, the endpoint alignment value is... ,otherwise When the edge types are the same, the edges When the edge types are different .

[0060] Edge attribute similarity value according to Calculate; where, Represents the normalized distance value of the edge attribute; normalized distance value of the edge attribute according to Calculate; where, This indicates the number of edge attributes involved in the comparison. Indicates the first The normalized distance of edge attributes; the handling method for numerical attributes, hierarchical attributes, enumerated attributes and missing attributes of edge attribute normalized distance is the same as the handling method for node attribute normalized distance.

[0061] Supplement edge candidate similarity values according to Calculate; where, Indicates the alignment value weight of the starting node. Indicates the alignment value weight of the endpoint node. Indicates the edge type consistency value weight. This represents the weight of the edge attribute similarity value. The method for setting the edge candidate weight and the edge supplement alignment threshold is the same as the method for setting the node candidate weight and the node supplement alignment threshold. Positive samples are edge pairs that are manually confirmed to belong to the same relation, and negative samples are edge pairs that are manually confirmed not to belong to the same relation. When there are no historical samples, all four weights are set to 0.25, and the edge supplement alignment threshold is set to 0.60.

[0062] After completing node alignment and edge alignment, the frame classification unit calculates structural similarity, node attribute similarity, and edge attribute similarity; structural similarity... according to Calculate; where, Indicates node coverage similarity. Indicates edge cover similarity. Indicates the node coverage weight. Indicates the edge cover weight. Node coverage similarity according to Calculate edge cover similarity. according to Calculate; where, This represents the set of edges that are successfully matched between two state graphs. Number of matching edges; node coverage weight and edge coverage weight The setting method is as follows: Statistically calculate node coverage similarity and edge coverage similarity from historical plan matching samples, and calculate the plan matching accuracy corresponding to different combinations of node coverage weights and edge coverage weights based on manually confirmed matching results; and The value range is set to 0 to 1, and is limited. and The sum is 1; the weight search step size is determined based on the number of historical plan matching samples. When the number of historical plan matching samples is not less than 30, the weight search step size is set to 0.01; when the number of historical plan matching samples is less than 30, both the node coverage weight and the edge coverage weight are set to 0.5; the node coverage weight and edge coverage weight combinations are traversed according to the weight search step size, and the weight combination with the highest plan matching accuracy is selected as the node coverage weight. and edge coverage weight When the number of historical matching samples is less than 30, Take 0.5, Take 0.5.

[0063] Node attribute similarity according to Calculate; where, Represents the set of node matches between two state graphs. The number of matching nodes; Indicates alignment of node pairs Similarity values ​​of node attributes; when the node alignment set is empty. Edge attribute similarity according to Calculate; where, Indicates alignment of edges Similarity values ​​of edge attributes; when the edge alignment set is empty. .

[0064] Image similarity according to ;in, and This represents two state diagrams. Indicates structural similarity. Indicates the similarity of node attributes. Indicates the similarity of edge attributes. , and The weights are defined as follows: manually confirmed matching state graph pairs are selected as positive samples, and manually confirmed non-matching state graph pairs are selected as negative samples; enumerations satisfying... Candidate weight combinations are calculated; the graph similarity of all samples is calculated using each candidate weight combination, and the corresponding judgment threshold is determined according to the method for setting the complete reference frame judgment threshold; the pass rate of positive samples and the exclusion rate of negative samples are statistically analyzed, and the candidate weight combination with the largest average of the two is selected as the graph similarity weight; when there are no labeled samples, .

[0065] Within the same classification cycle, the frame classification unit selects one graph similarity calculation method from weighted graph similarity, graph edit distance similarity, and graph kernel similarity. The sample calculation and operation judgment of the complete reference frame judgment threshold, the predicted frame judgment threshold, and the interpolated frame judgment threshold all adopt the selected graph similarity calculation method, and different graph similarity calculation methods are not mixed within the same classification cycle. When changing the graph similarity calculation method, the three judgment thresholds are recalculated according to the corresponding threshold setting method.

[0066] In one alternative implementation, graph similarity is calculated using graph edit distance. The formula is ;in, This indicates the state diagram. Convert to state diagram Minimum number of editing operations required and These represent the node sets of the two state diagrams, respectively. and Let each represent the edge set of the two state graphs; the graph similarity is 1 when both state graphs are empty.

[0067] In one alternative implementation, a graph kernel function is used to calculate graph similarity. The formula is ;in, The graph kernel function is denoted by a random walk kernel or a shortest path kernel. The graph similarity is 1 when the denominator is zero and both state graphs are empty, and 0 when the denominator is zero and at least one state graph is not empty.

[0068] The frame classification unit calculates the graph similarity between the current state graph and each plan target graph, as well as the graph similarity between the current state graph and each complete reference frame, and takes the maximum graph similarity between the plan target graph and the plan matching similarity as the plan matching similarity. The maximum similarity of the corresponding graphs of the complete reference frame is taken as the reference frame coverage similarity. When the contingency plan database is empty, Set to 0 when the complete reference frame library is empty. Take 0; when When the current state graph is determined to be a complete reference frame for matching the contingency plan, the contingency plan corresponding to the target graph with the highest graph similarity is selected as the matching contingency plan; when and When the current state graph is determined to be a new type of complete reference frame; for both types of complete reference frames, the current frame identifier is read according to the event number, current sampling time, and event intra-frame sequence number. The current frame identifier, complete nodes, complete edges, node attributes, edge attributes, sampling time, event type, handling stage, and matching plan identifier of the current state graph are written into the complete reference frame storage area to form the current complete reference frame record. Among them, the matching plan identifier of the new type of complete reference frame is empty; when and When the current state diagram and sampling time are transferred to the prediction and judgment process, the process is initiated.

[0069] For a complete reference frame that matches the pre-plan, when At that time, the frame classification unit according to Determine the best matching plan ;in, This represents the set of emergency response plan identifiers obtained from the analysis of the emergency response plan. This indicates the first in the set of emergency response plans. Individual candidate plan identifiers; Indication and contingency plan identification Corresponding contingency plan target map; Indicates the current state diagram and the first Graph similarity between target maps of each contingency plan; The identifier of the plan with the highest image similarity; read the identifier of the current frame. Mark the current frame Current state diagram Current sampling time Complete reference frame type identifier, best matching contingency plan identifier The event type, handling stage, and source record are written into the complete reference frame storage area to form the current complete reference frame record.

[0070] In one alternative implementation, the pre-plan matching complete reference frame record is represented as follows: Write to the complete reference frame storage area and add to the complete reference frame library; among which, Indicates the current frame identifier. This represents the complete nodes, edges, and attributes of the current state graph. Indicates the sampling time. Indicates the best matching plan identifier. Indicates the complete reference frame type identifier, This indicates the event type, stage of handling, and source record.

[0071] For the new type of complete reference frame, when and At that time, read the current frame identifier. Mark the current frame Current state diagram Current sampling time The complete reference frame type identifier, empty contingency plan identifier, event type, handling stage, and source record are written into the complete reference frame storage area to form the current complete reference frame record.

[0072] In one alternative implementation, the new type of complete reference frame record is represented as: Write to the complete reference frame storage area and add to the complete reference frame library; among which, This indicates that there is no matching plan.

[0073] When generating a complete reference frame record based on the current state diagram, the current frame identifier, complete state diagram, current sampling time, complete reference frame type identifier, matching plan identifier, event type, handling stage, source record, writing time, last reference time, and reference count are written in a unified complete reference frame record format. If there is no matching plan, the matching plan identifier is empty. When writing a complete reference frame record for the first time, the writing time is taken as the system time when the corresponding record is written to the complete reference frame storage area, the last reference time is empty, and the reference count is recorded as 0.

[0074] When the plan matches the similarity and When the current state diagram does not meet the conditions for determining a complete reference frame matching the pre-set plan, nor does it meet the conditions for determining a new type of complete reference frame, the frame classification unit will transfer the current state diagram and the current sampling time into the prediction and determination process.

[0075] Complete reference frame determination threshold The setting method is as follows: Positive samples are obtained by reading state diagram and plan target diagram pairs, and state diagram and complete reference frame pairs, from historical handling records that are manually confirmed to belong to the same event type and have the same handling stage. Negative samples are obtained by reading state diagram and plan target diagram pairs, and state diagram and complete reference frame pairs, from historical handling records that have different event types or inconsistent handling stages. The graph similarity of each positive and negative sample is calculated using the graph similarity calculation method selected during classification runtime. The graph similarity scores of 0, 1, and all samples that are unique are arranged in ascending order to form a candidate threshold set. For each candidate threshold, the ratio of the number of positive samples with graph similarity not less than the candidate threshold to the total number of positive samples is used as the positive sample pass rate. The ratio of the number of negative samples with graph similarity less than the candidate threshold to the total number of negative samples is used as the negative sample exclusion rate. The average of the positive sample pass rate and the negative sample exclusion rate is calculated, and the candidate threshold with the largest average is taken as the complete reference frame determination threshold. When multiple candidate thresholds have the same average value, the candidate threshold with the largest value is selected; when no historical samples exist, [the threshold is...]. The initial value is set to 0.82, and it is updated according to the above method after historical samples are formed.

[0076] Select a complete state map from the complete reference frame library, process the complete state map according to the disposal operation sequence to obtain a prediction map, compare the prediction map with the current state map to obtain prediction similarity and prediction residual, generate a prediction frame record containing reference frame identifier, contingency plan identifier and prediction residual according to the prediction similarity determination result, or retain the current state map and perform interpolation frame determination.

[0077] In a specific implementation, the method for generating the predicted frame record includes:

[0078] The prediction unit selects complete reference frame records from the complete reference frame library whose sampling time is earlier than the sampling time of the current state diagram. If the current state diagram has already recorded an event type, it retains complete reference frame records whose event type matches the event type of the current state diagram as candidate complete reference frame records. If the current state diagram has not recorded an event type, it selects complete reference frame records from the complete reference frame library that have the same node identifier or the same edge endpoint identifier as the current state diagram as candidate complete reference frame records. If the current state diagram has already recorded an event type, the prediction unit selects a record from the contingency plan library whose event type matches the event type of the current state diagram, and whose handling object code matches the candidate event type. The sequence of handling operations that has the same node identifier or edge endpoint identifier in the complete reference frame record is selected as the candidate handling operation sequence. When the current state diagram does not record the event type, the corresponding event type is read for each candidate complete reference frame record, and the sequence of handling operations that has the same event type as the corresponding candidate complete reference frame record and whose handling object code has the same identifier as the node identifier or edge endpoint identifier in the corresponding candidate complete reference frame record is selected from the contingency plan library as the corresponding candidate handling operation sequence. When the candidate complete reference frame set is empty or the candidate handling operation sequence set is empty, the prediction graph calculation is not performed, and the current state diagram and sampling time are transferred to the interpolation judgment process.

[0079] For each candidate prediction combination consisting of a candidate complete reference frame record and a candidate disposal operation sequence, the prediction unit reads the reference state diagram in the candidate complete reference frame record and represents the candidate disposal operation sequence as the first disposal operation to the Kth disposal operation according to the execution order.

[0080] The prediction unit copies the reference state diagram K+1 times. For the kth copy of the reference state diagram, the first k disposal operations in the disposal operation sequence are executed in the execution order to obtain the kth candidate prediction diagram. k is recorded as the corresponding candidate operation cutoff number. The value of k is from 0 to K. When k is 0, no disposal operation is executed and the reference state diagram copy is directly used as the candidate prediction diagram.

[0081] The prediction unit calculates the graph similarity between each candidate prediction graph and the current state graph, and determines the candidate prediction graph with the highest graph similarity as the prediction graph. The candidate complete reference frame record that generates the prediction graph is determined as the target complete reference frame record. The plan identifier of the plan to which the corresponding disposal operation sequence belongs is determined as the target plan identifier, and the corresponding candidate operation deadline number is determined as the operation deadline number. When multiple candidate prediction graphs have the same graph similarity, the target complete reference frame record with the sampling time closest to the current state graph is selected first. If the sampling times are still the same, the disposal operation sequence with the highest plan priority is selected. If the plan priorities are still the same, the candidate prediction graph with the smaller operation deadline number is selected.

[0082] When a node is added, the node identifier, node type, and node attributes are written to the candidate complete reference frame copy. When a node is deleted, the corresponding node and edges with the corresponding node as the start or end point are deleted. When a node attribute is modified, the target attribute value is written to the corresponding node attribute field. When an edge is added, the edge identifier, start node identifier, end node identifier, edge type, and edge attribute are written. When an edge is deleted, the corresponding edge is deleted. When an edge attribute is modified, the target attribute value is written to the corresponding edge attribute field. When resource scheduling is performed, the resource quantity of resource point nodes, the scheduling time of resource reachable edges, and the processing progress of the target node are updated. After all processing operations in the corresponding operation prefix are completed, a candidate prediction graph is obtained.

[0083] In one alternative implementation, the sequence of processing operations is represented as follows: At that time, for the first Each operation prefix corresponds to a candidate prediction graph represented as follows: ;when hour, ;in, Indicates compound operation symbols; This indicates the execution order of the disposal operation sequence, using the candidate complete reference state diagram. As the initial state diagram, execute item 1 through item 2 in sequence. The term graph transformation operation takes the state graph obtained from the previous term graph transformation operation as the input of the next term graph transformation operation. This represents the complete state diagram recorded in the candidate complete reference frame. Indicates the first Item diagram transformation operations, It also serves as the cutoff number for the corresponding candidate operation.

[0084] The prediction unit calculates the current state diagram respectively. With each candidate prediction plot The graph similarity between the graphs is calculated; the candidate prediction graph with the highest graph similarity is selected as the target prediction graph; the candidate complete reference frame corresponding to the target prediction graph is selected as the target reference frame, and the candidate disposal operation sequence corresponding to the target prediction graph is selected as the target disposal operation sequence; when the graph similarity of multiple candidate prediction graphs is the same, the candidate prediction combination with the smallest difference between the sampling time of the target reference frame and the sampling time of the current state graph is selected; when the sampling time difference is still the same, the candidate prediction combination with the highest priority of the plan is selected; the frame identifier in the target reference frame record is read as the reference frame identifier, and the plan identifier of the plan to which the target disposal operation sequence belongs is read as the target plan identifier.

[0085] The prediction unit aligns the current state graph with the target prediction graph using node alignment and edge alignment methods. Nodes and edges present in the current state graph but not in the target prediction graph are recorded as node addition residuals and edge addition residuals, respectively. Nodes and edges present in the target prediction graph but not in the current state graph are recorded as node deletion residuals and edge deletion residuals, respectively. Attribute fields with different attribute values ​​in aligned nodes or edges, along with the target attribute value in the current state graph, are recorded as attribute modification residuals. The prediction residuals are obtained by aggregating these node addition residuals, node deletion residuals, edge addition residuals, edge deletion residuals, and attribute modification residuals. , ;in, This indicates the graph transformation operations required to restore the target prediction graph to the current state graph.

[0086] When the prediction similarity between the current state graph and the prediction graph, i.e., the graph similarity, is not less than the prediction frame determination threshold. When the current frame identifier, target complete reference frame identifier, target contingency plan identifier, operation deadline number, prediction residual, sampling time, prediction similarity, and prediction frame type identifier are written into the prediction frame record; when the prediction similarity is less than the prediction frame judgment threshold, the current state diagram, current frame identifier, and sampling time are retained until the same emergency event forms a complete reference frame record or prediction frame record with a sampling time later than the current state diagram.

[0087] In one alternative implementation, predictive frame recording Represented as:

[0088] ;in, Indicates the target's complete reference frame identifier. Indicates the target contingency plan identifier. Indicates the operation deadline number. Indicates predicted similarity. Indicates the predicted residual. This indicates the predicted frame type identifier.

[0089] Predicted frame determination threshold The setting method is as follows: Read the complete reference frame, the actual executed contingency plan operation, and the stage result status diagram under the same event number from the historical handling records. Form a positive sample by combining the corresponding complete reference frame, contingency plan operation, and stage result status diagram. Form a negative sample by combining complete reference frames, contingency plan operations, and status diagrams of different event types, handling objects that do not overlap, or that do not belong to the same handling chain. For each predicted positive and negative sample, read the handling operation sequence in the sample according to the execution order. Apply the first zero to the first K handling operations of the sequence to the complete reference frame of the sample, obtaining K+1 candidate prediction maps. Calculate the graph similarity between each candidate prediction map and the sample status diagram. Use the maximum graph similarity as the predicted similarity of the corresponding sample and record the operation cutoff number that generates the maximum graph similarity. Form a candidate threshold set by combining 0, 1, and all sample prediction similarities. Select the candidate threshold with the largest average positive sample pass rate and negative sample exclusion rate as the prediction frame judgment threshold. When the average values ​​of multiple candidate thresholds are the same, the candidate threshold with the largest value is selected; when the number of positive samples is zero or the number of negative samples is zero, The initial value is 0.75, and it is reset after both positive and negative samples are generated simultaneously.

[0090] Obtain reference frame records before and after the current state map, recover the forward and backward reference state maps based on the reference frame records, obtain an interpolation map based on the two reference state maps and the sampling time, compare the interpolation map with the current state map to obtain interpolation similarity and interpolation residual, generate an interpolation frame record containing two reference frame identifiers, interpolation parameters and interpolation residual based on the interpolation similarity determination result, or generate a complete reference frame record based on the current state map.

[0091] In specific implementations, methods for generating interpolated frame records or complete reference frame records include:

[0092] When the prediction similarity of the current state graph is less than the prediction frame determination threshold, the current state graph, the current frame identifier, and the sampling time are written into the set of states to be determined.

[0093] For each complete reference frame record or prediction frame record formed, the interpolation unit reads the corresponding event number and sampling time, and selects a state map from the set of states to be determined that has the same event number and a sampling time earlier than the newly formed frame record. For each selected state map, the frame record with a sampling time earlier than the corresponding state map is selected from the already formed complete reference frame record and prediction frame record as a forward reference candidate frame, and the frame record with a sampling time later than the corresponding state map is selected as a backward reference candidate frame.

[0094] When no forward reference candidate frame exists, a complete reference frame record is generated for the corresponding state diagram according to the unified complete reference frame record format described above, where the matching plan identifier is empty; when no backward reference candidate frame exists, the corresponding state diagram is retained; when no backward reference candidate frame exists after the observation time window for the same emergency event ends, a complete reference frame record is generated for the corresponding state diagram according to the unified complete reference frame record format described above, where the matching plan identifier is empty.

[0095] When the forward or backward reference frame record is a complete reference frame record, the reference state diagram in the complete reference frame record is read. When the forward or backward reference frame record is a prediction frame record, the target reference state diagram is read according to the target complete reference frame identifier in the prediction frame record, the disposal operation sequence is read according to the target contingency plan identifier, the disposal operations up to the operation deadline number in the disposal operation sequence are executed in the execution order, and then node addition, node deletion, edge addition, edge deletion and attribute modification are executed according to the prediction residual to restore the corresponding state diagram.

[0096] The interpolation unit calculates the interpolation position parameters for each set of forward and backward reference candidate frames. ; ;in, , ; This indicates the sampling time of the current state diagram. Indicates the sampling time of the forward reference frame. Indicates the sampling time of the backward reference frame;

[0097] The interpolation unit directly aligns nodes in two reference frames according to the object identifier and directly aligns edges in two reference frames according to the edge identifier. For nodes and edges that are not directly aligned, a supplementary alignment relationship is established according to the supplementary node candidate similarity value and the supplementary edge candidate similarity value. The union of nodes in two reference frames is taken as the candidate node set, and the union of edges in two reference frames is taken as the candidate edge set.

[0098] For numerical or hierarchical node attributes and edge attributes that exist and are aligned in both reference frames, the bidirectional interpolation model is used. Calculate the interpolated attribute values; where, This represents the attribute values ​​in the candidate interpolation plot. This represents the attribute value in the forward reference frame. This represents the attribute value in the backward reference frame. For enumerated attributes or text tag attributes, the attribute value is taken if the attribute values ​​in the current and backward reference frames are the same; otherwise, the attribute value is taken if the attribute values ​​in the current and backward reference frames are different. The attribute value of the preceding reference frame is taken at the same time. If the attribute values ​​of the preceding and following reference frames are inconsistent, and When the attribute value of the back reference frame is used, if the attribute involved in the interpolation is a level attribute and the interpolated attribute value is not a valid level value of the level attribute, the absolute difference between the interpolated attribute value and each valid level value is calculated, and the valid level value with the smallest absolute difference is written into the candidate interpolation map. If there are two valid level values ​​with the same absolute difference, the level value in the forward reference state map is used when the interpolation parameter is less than 0.5, and the level value in the back reference state map is used when the interpolation parameter is not less than 0.5.

[0099] For nodes or edges that exist only in the forward reference frame, the object's preserved value is... For nodes or edges that exist only in the backward reference frame, the object's preserved value is... When the object retention value is not less than the object retention threshold When the object retention value is less than the object retention threshold, the corresponding node or edge is retained in the candidate node set or candidate edge set. When an object is not retained, the corresponding node or edge is removed from the candidate node set or candidate edge set. After the object retention judgment is completed, the start node and end node of each edge in the candidate edge set are checked. If the start node or end node is not retained in the candidate node set, the corresponding edge is removed from the candidate edge set. The candidate node set, candidate edge set, interpolated node attributes, and interpolated edge attributes together form the candidate interpolation graph. .

[0100] The interpolation unit calculates the graph similarity between the current state graph and each candidate interpolation graph, and selects the candidate interpolation graph with the highest graph similarity as the target interpolation graph. When multiple candidate interpolation graphs have the same graph similarity, the candidate interpolation graph with the smallest sampling time interval between the forward reference frame and the backward reference frame is selected. The frame identifiers of the forward reference frame and the backward reference frame corresponding to the target interpolation graph are read.

[0101] The interpolation unit aligns the current state graph with the target interpolation graph using node and edge alignment methods; it then generates the interpolation residuals according to the recording methods for node additions, node deletions, edge additions, edge deletions, and attribute differences in the prediction residuals. , .

[0102] When the interpolation similarity between the current state map and the target interpolation map Not less than the interpolation frame determination threshold At that time, the current frame identifier is read according to the event number, the current sampling time, and the sequence number of the event frame. Write the current frame identifier, the forward reference frame identifier, the backward reference frame identifier, the interpolation position parameters, the interpolation residual, the interpolation similarity, the current sampling time, the interpolation frame type identifier, and the source record into the interpolation frame storage area to form the current interpolation frame record.

[0103] When the interpolation similarity between the current state map and the target interpolation map is less than the interpolation frame determination threshold At that time, the current complete reference frame record is generated based on the current state diagram according to the unified complete reference frame record format mentioned above, wherein the matching plan identifier is empty, and the current complete reference frame record is written into the complete reference frame storage area and the complete reference frame library.

[0104] In one alternative implementation, the interpolated frame record is represented as follows:

[0105] ;in, Indicates the forward reference frame identifier. Indicates the backward reference frame identifier. Indicates the interpolation residual. Indicates the interpolation frame type identifier. This indicates the event type, stage of handling, and source record.

[0106] Object retention threshold Threshold determination in interpolation frames Previously set; read nodes and edges that exist only in the forward reference frame or only in the backward reference frame from historical transition samples, and check whether the corresponding nodes and edges exist in the transition state graph; for nodes or edges that exist only in the forward reference frame, according to Calculate the retained values ​​of the objects; for nodes or edges that exist only in the backward reference frame, according to Calculate the object retention value; take the nodes or edges that exist in the transition state graph as positive samples and the nodes or edges that do not exist in the transition state graph as negative samples; form a candidate threshold set with 0, 1 and all object retention values; select the candidate threshold with the largest average of the positive sample retention ratio and the negative sample deletion ratio as the object retention threshold; when multiple candidate thresholds correspond to the same average value, select the candidate threshold with the largest value; when there are no positive or negative samples, the object retention threshold is initially set to 0.50.

[0107] Interpolation frame determination threshold The setting method is as follows: Generate sample interpolation maps using the determined object retention threshold; read the transition state map between two complete reference frames or prediction frames under the same event number from the historical handling records, and combine the forward reference frame, backward reference frame, and transition state map into positive samples; combine state maps of different event types, discontinuous handling stages, or two reference frames not belonging to the same handling link into negative samples; calculate the interpolation position parameters according to the sampling time and obtain the sample interpolation map, and calculate the graph similarity between the sample interpolation map and the sample state map; form a candidate threshold set using 0, 1, and all sample graph similarities; select the candidate threshold with the largest average ratio of positive samples and negative samples as the interpolation frame determination threshold; when multiple candidate thresholds correspond to the same average value, select the candidate threshold with the largest value; when positive or negative samples are empty, the initial interpolation frame determination threshold is 0.70.

[0108] Write the current frame identifier and sampling time into the corresponding frame record. Establish an inter-frame dependency graph based on the reference relationships in the complete reference frame record, the predicted frame record, and the interpolated frame record. Restore the state graph of the target sampling time based on the frame record corresponding to the target sampling time and the inter-frame dependency graph.

[0109] In a specific implementation, the method for restoring the state diagram at the target sampling time includes:

[0110] The inter-frame dependency graph unit reads the current frame identifier from the current frame record and writes the current frame identifier into the frame node set. When the current frame record is a complete reference frame record, only the current frame node is written. When the current frame record is a prediction frame record, the target complete reference frame identifier is read, a prediction dependency edge is established from the reference frame node to the current prediction frame node, and the reference frame identifier, the current prediction frame identifier, the target prediction identifier, the prediction similarity, and the establishment time are written into the prediction dependency edge. When the current frame record is an interpolation frame record, the forward reference frame identifier and the backward reference frame identifier are read, interpolation dependency edges are established from the two reference frame nodes to the current interpolation frame node, and the corresponding reference frame identifier, the current interpolation frame identifier, the interpolation position parameter, the interpolation similarity, and the establishment time are written into the interpolation dependency edge.

[0111] In one alternative implementation, the inter-frame dependency graph is represented as: ;in, Represents the set of frame nodes. Represents the complete set of reference frames. Represents the set of predicted frames. Represents the set of interpolated frames. Represents the set of dependent edges.

[0112] Before writing predicted or interpolated dependency edges, the inter-frame dependency graph unit performs dependency cycle detection. For a dependency edge to be written that points from a reference frame node to the current frame node, a depth-first traversal is performed from the current frame node along the existing directed edges in the inter-frame dependency graph. If the traversal can reach the reference frame node to which the dependency edge to be written is intended, it is determined that the dependency edge to be written will form a dependency cycle, and the writing of the corresponding dependency edge is stopped. If the traversal cannot reach the reference frame node to which the dependency edge to be written is intended, the corresponding dependency edge is written. If the dependency edge to be written will form a dependency cycle, the dependency edge to be written is not written, the predicted frame record or interpolated frame record corresponding to the invalid current frame identifier is canceled or marked, and the current complete reference frame record is generated using the current state graph retained during classification, ensuring that the same current frame identifier corresponds to only one valid frame record.

[0113] After receiving a target sampling time query request, the state recovery module reads the target frame record based on the target sampling time. If the target frame record is a complete reference frame record, it reads the complete nodes, complete edges, node attributes, and edge attributes to obtain the target state graph. If the target frame record is a prediction frame record, it reads the corresponding reference state graph based on the target complete reference frame identifier, reads the disposal operation sequence based on the target contingency plan identifier, and executes the disposal operations up to the operation deadline in the disposal operation sequence according to the execution order to obtain the basic prediction graph. It then executes node addition, node deletion, edge addition, edge deletion, and attribute modification according to the prediction residual to obtain the target state graph. If the target frame record is an interpolation frame record, it recovers two reference frames based on the forward reference frame identifier and the backward reference frame identifier, obtains the basic interpolation graph according to the node alignment, edge alignment, attribute interpolation, and object preservation methods used when the interpolation frame was formed, and then executes node addition, node deletion, edge addition, edge deletion, and attribute modification according to the interpolation residual to obtain the target state graph.

[0114] In one alternative implementation, the formula for recovering the complete reference frame is: The formula for predicting frame recovery is: The formula for interpolating frame recovery is:

[0115] ;in, This represents the complete reference frame state diagram after recovery. This represents the state diagram of the recovered predicted frame. This represents the state diagram of the recovered interpolated frame. This represents the complete reference frame in the complete reference frame storage area. This indicates the first to the second items in the sequence of actions corresponding to the target contingency plan identifier. Item handling operation, This represents the target reference state diagram. Indicates the predicted residual; This represents the state diagram of the recovered interpolated frame. This represents the recovered forward reference frame state diagram. This represents the restored backward reference frame state diagram. Indicates the interpolation position parameter. Indicates the interpolation residual. This indicates that a graph transformation operation is performed based on the prediction residual or interpolation residual in the corresponding frame record.

[0116] When the current frame record is a complete reference frame record and there is no identical current frame identifier in the complete reference frame library, the current complete reference frame record is added to the complete reference frame library. When a complete reference frame is selected as a prediction frame reference frame, a forward or backward reference frame for interpolation frames, or is directly read in a status query, the reference count of the corresponding complete reference frame is incremented by one, and the current time is written to the last reference time. When the same complete reference frame is used multiple times in the same processing procedure, the reference count is incremented only once.

[0117] The number of contingency plans that can be associated with a complete reference frame is obtained by comparing the target map of the contingency plan that is consistent with the response stage of the complete reference frame. The event type and response stage of the complete reference frame are read, and the target map of the contingency plan that is consistent with the event type and response stage is selected from the contingency plan library. The graph similarity between the complete reference frame and each target map of the contingency plan is calculated, and the number of contingency plans with a graph similarity of not less than the judgment threshold of the complete reference frame is taken as the number of contingency plans that can be associated with the complete reference frame.

[0118] If the event type to which the restored target state graph belongs already exists, only complete reference frame records whose event types are consistent with the event type of the emergency event to which the target state graph belongs are selected from the complete reference frame library, and the maximum graph similarity between the target state graph and the selected complete state graph is calculated; when the maximum graph similarity is less than the subtype threshold, a new variant complete reference frame record is generated for the target state graph; if no complete reference frame record with the same event type exists, it is processed as a new type complete reference frame.

[0119] Subtype threshold The setting method is as follows: Select state graph pairs that have been reviewed and confirmed to belong to different handling stages or different impact ranges from the historical state graphs of the same event type as variant positive samples, and select repeated state graph pairs that have been reviewed and confirmed to belong to the same handling stage and have the same impact range as variant negative samples; calculate the graph similarity of each sample state graph pair, and form a candidate threshold set by zero, one, and all sample graph similarities; for each candidate threshold, calculate the proportion of variant positive samples with graph similarity less than the candidate threshold as the variant recognition rate, and calculate the proportion of variant negative samples with graph similarity not less than the candidate threshold as the repeated state exclusion rate; select the candidate threshold with the largest average of variant recognition rate and repeated state exclusion rate as the subtype threshold; when multiple candidate thresholds correspond to the same average value, select the candidate threshold with the smaller value to reduce the number of repeated states written into the complete reference frame library; when variant positive samples or variant negative samples are empty, do not perform new variant complete reference frame determination, only use reference frame coverage similarity to perform existing complete reference frame determination; set the subtype threshold after both types of samples have been formed.

[0120] Every time the complete reference frame library management unit completes a supplementary cycle, it reads candidate frame records from the prediction frame storage area and the interpolation frame storage area; it reconstructs the complete state diagram corresponding to each candidate frame according to the inter-frame dependency graph, and calculates the candidate supplementary score based on the reconstructed complete state diagram; it selects the candidate frame with the largest candidate supplementary score, assigns a supplementary complete reference frame identifier to the reconstructed complete state diagram, and writes the supplementary complete reference frame identifier, the reconstructed complete state diagram, the original candidate frame identifier, the sampling time, and the source record into the complete reference frame storage area and the complete reference frame library; when the complete reference frame library reaches its size limit, it first performs complete reference frame elimination, and then writes the supplementary reference frames. The complete reference frame is supplemented; the original predicted frame record, the original interpolated frame record, and the original dependencies remain unchanged; the supplemented complete reference frame record does not replace the corresponding original predicted frame record or the original interpolated frame record; when the supplemented complete reference frame record and the corresponding original predicted frame record or the original interpolated frame record have the same sampling time, when the state recovery module determines the target frame record based on the sampling time, the corresponding original predicted frame record or the original interpolated frame record is used as the target frame record; the supplemented complete reference frame record is used for subsequent graph similarity calculation between the current state graph and the complete reference frame, selection of candidate complete reference frames, and selection of reference frames for the predicted frames or interpolated frames formed later.

[0121] The supplementary score for candidates is calculated using the following formula: ;in, This represents the candidate supplementary score of the candidate state diagram. This represents the storage ratio. , This represents the storage amount of the candidate state graph residuals. This indicates the total storage size of the candidate state graph. hour, , This represents the ratio of the number of queries to the candidate state graph within the current supplementary period to the maximum number of queries to all candidate state graphs. The maximum number of queries to all candidate state graphs is zero. Take zero, This represents the maximum graph similarity between the candidate state graph and all complete reference frames in the complete reference frame library; when the complete reference frame library is empty. Set the value to 0; when candidate supplementary scores are the same, prioritize the candidate frame with more query counts, and if the number of query counts are still the same, select the candidate frame with larger residual storage; when the complete reference frame library has not reached its size limit, write the candidate state diagram with the largest candidate supplementary score into the complete reference frame library; when the complete reference frame library reaches its size limit, first perform complete reference frame elimination, and then write it into the candidate state diagram.

[0122] When the current number of records in the complete reference frame library is not less than the upper limit and a new complete reference frame record is to be written, the complete reference frame library management unit calculates the elimination score of each complete reference frame based on the expiration value, the duration of non-reference, and the insufficient coverage value of the contingency plan. The complete reference frame with the highest elimination score that exceeds the elimination threshold is removed from the online complete reference frame library. Complete reference frames are removed sequentially from high to low elimination scores until the number of records in the online complete reference frame library is no greater than the upper limit minus 1, at which point a new complete reference frame record is written. The removed complete reference frames are written to the archive storage area, retaining the frame identifier, sampling time, and historical recovery index. The archive storage area retains the complete node set, complete edge set, node attributes, and edge attributes of the complete reference frame. The historical recovery index records at least the mapping relationship between the frame identifier and the corresponding archive storage location. When the state recovery module reads the reference state diagram based on the reference frame identifier, it first searches in the online complete reference frame library. If the corresponding frame identifier is not found, the corresponding complete reference frame is read from the archive storage area based on the historical recovery index. The elimination of an online complete reference frame does not delete its complete state diagram data used for historical state recovery.

[0123] The supplementary complete reference frame record is used as a reference record for subsequent frame classification and prediction reference frame selection. It does not replace the original prediction frame record or interpolation frame record, and is not written to the target state query index with the sampling time as the key. When performing state recovery for the target sampling time, the original valid frame record of that sampling time is still read. The supplementary complete reference frame identifier is only used for subsequent reference frame selection and inter-frame dependency establishment.

[0124] Complete reference frame elimination score for: ;in, Represents a complete reference frame Elimination score , and Indicates the elimination weight. , This indicates the expiration value, which occurs when the difference between the current time and the time the complete reference frame was written is greater than the validity period of the complete reference frame. hour Select 1 if the value is 1, otherwise select 0. This indicates a duration value that is not referenced; the complete reference frame has never been referenced. If the value is 1, and the complete reference frame has already been referenced, it is calculated according to the following formula: , This indicates a reference to the statistics window. Indicates the last time it was cited. This indicates insufficient coverage of contingency plans. When the contingency plan database is not empty, it is calculated using the following formula: The number of contingency plans that can be associated with a complete reference frame is obtained by comparing the complete reference frame with the target maps of contingency plans that are consistent with the handling stage. The complete reference frame library management unit reads the event type and handling stage of the complete reference frame, selects the target maps of contingency plans with consistent event type and handling stage from the contingency plan library, calculates the graph similarity between the complete reference frame and each target map of the contingency plan, and takes the number of contingency plans with a graph similarity of not less than the judgment threshold of the complete reference frame as the number of contingency plans that can be associated with the complete reference frame. This represents a set of candidate contingency plans whose event type and handling phase are consistent with the complete reference frame. This refers to the set of candidate plans whose graph similarity to the target graph of a plan is not lower than the threshold for judging the complete reference frame; when the plan library is empty. Set to 0;

[0125] The method for setting elimination weights and elimination thresholds is as follows: Read complete reference frames that should be manually removed from the online complete reference frame library from historical maintenance records as positive samples, and read complete reference frames that should be manually retained in the online complete reference frame library as negative samples; enumerate candidate weight combinations that satisfy the condition that the sum of weights is one, and for each candidate weight combination, form a candidate elimination threshold set by eliminating scores of 0, 1, and all samples; calculate the proportion of positive samples with elimination scores not less than the candidate elimination thresholds and the proportion of negative samples with elimination scores less than the candidate elimination thresholds; select the candidate weight combination and candidate elimination threshold with the largest average of the two proportions; when there are no historical maintenance samples, all three weights are taken as one-third, and elimination is performed from high to low elimination scores when the complete reference frame library reaches its size limit.

[0126] The validity period of a complete reference frame is determined based on the time interval between the time of writing a historical complete reference frame and the time of the last successful reference. The time interval corresponding to all complete reference frames with successful reference records is calculated, and the 90th percentile value of all time intervals is taken as the validity period of the complete reference frame. When there is no successful reference record, the longest review period of the historical processing record is taken as the validity period of the complete reference frame. After each maintenance cycle, the validity period of the complete reference frame is recalculated using the newly added reference record.

[0127] Let the average storage size of a single complete reference frame be... The average storage size of a single prediction frame is The average storage size of a single interpolated frame is ;but ; ;in, Indicates the storage amount of the current frame identifier. Indicates the storage size of a reference frame identifier. This indicates the storage volume of the target contingency plan identifier or handling operation. This indicates the amount of predicted residual storage. This indicates the storage size of the interpolation position parameters. This indicates the storage size of the interpolation residual. This indicates the sampling time, frame type, and storage size of the source record.

[0128] Predicted frame relative storage relative storage size of interpolated frames They are respectively: , For those containing A complete reference frame One predicted frame and A sequence of state diagrams for each interpolated frame, relative storage requirements for Storage compression ratio for Storage compression ratio for .

[0129] When the state graph enters the frame classification process, the complete nodes, complete edges, node attributes, edge attributes, event numbers, and sampling times of the current state graph are written into the quality verification cache. After each supplementary cycle, the state graph corresponding to each sampling time in the quality verification cache is restored according to the inter-frame dependency graph, and the graph similarity between the restored state graph and the original state graph in the quality verification cache is calculated. After the calculation is completed, the original state graph that has been verified is deleted from the quality verification cache.

[0130] When the average restored similarity is lower than the lower limit of restored similarity, the prediction frame determination threshold and the interpolation frame determination threshold are adjusted to the smallest candidate threshold in the corresponding candidate threshold set that is greater than the current threshold. When the average restored similarity is not lower than the lower limit of restored similarity and the storage compression ratio is lower than the compression ratio target value, the prediction frame determination threshold and the interpolation frame determination threshold are adjusted to the largest candidate threshold in the corresponding candidate threshold set that is less than the current threshold. When there is no corresponding larger or smaller value in the candidate threshold set, the current threshold remains unchanged. When the average restored similarity is lower than the lower limit of restored similarity in the next supplementary period after the threshold is lowered, the adjustment of the most recent threshold reduction is withdrawn.

[0131] The method for setting the lower limit of recovery similarity is as follows: Read the manually confirmed acceptable recovery state graph and the original state graph from the historical state recovery record; calculate the graph similarity of each group of state graphs; and take the fifth percentile value of all graph similarities as the lower limit of recovery similarity; compression ratio target value. The number of state diagrams is calculated based on the available capacity allocated to state diagram storage, the average storage size of complete reference frames, and the expected number of state diagrams within the retention period. The expected number of state diagrams within the retention period is the maximum value of the actual number of state diagrams generated in the most recent three complete retention periods; if there are fewer than three complete retention periods, the maximum value among the existing complete retention periods is used; if there are no historical complete retention periods, the number is obtained by dividing the state diagram retention period by the uniform sampling period and rounding up. ;in, Indicates available storage capacity. This indicates the number of state diagrams expected to be generated within the retention period. This represents the average storage amount of a single complete reference frame. When the number of expected state maps within the retention period is 0 or the average storage amount of a single complete reference frame is 0, the compression ratio target value is 0. When there are no historical state recovery records for which the recovery results are manually confirmed to be acceptable, the automatic adjustment of the predicted frame judgment threshold and the interpolated frame judgment threshold is not performed, and the current threshold remains unchanged.

[0132] After the complete reference frame library is added, supplemented, and eliminated, the complete reference frame library management unit outputs the updated complete reference frame library; the frame classification unit reads the updated complete reference frame library at the next sampling time, performs graph similarity calculation between the current state graph and the complete reference frames, and predicts reference frame selection, thereby forming a closed loop of classification processing between consecutive sampling times.

[0133] In a specific application example, a smoke emergency event at a rail transit station is used as an example to illustrate the process of recording complete reference frames, predictive frames, interpolated frames, and the recovery of the target state diagram. The values ​​in this application example are used to illustrate the data processing process of the aforementioned method and do not limit the actual values ​​of the complete reference frame determination threshold, the predictive frame determination threshold, and the interpolated frame determination threshold. In actual operation, each threshold is determined and updated according to the aforementioned corresponding threshold setting method.

[0134] Assuming the current classification cycle uses the aforementioned weighted graph similarity calculation method, the complete reference frame determination threshold is... Set the threshold to 0.82 for predicting frames. Set the threshold to 0.75 for interpolation frame determination. The value is 0.70. The state diagram in this application example includes a site area node N1, a smoke detection device node N2, a personnel area node N3, an affected passage node N4, and a rescue resource node N5. It also includes the attribution edges between the device and the site area, the regional association edges between the personnel area and the site area, the connection edges between the affected passage and the site area, and the reachability edges between the rescue resource and the site area. Node N2 has a smoke concentration attribute, node N3 has a personnel evacuation completion rate attribute, node N4 has a passage status attribute, and node N5 has a rescue resource status attribute. The sequence of actions corresponding to plan P01 includes the first action. Second treatment operation and third-party processing ,in Used to change the status of the affected channel to closed. Used to change the evacuation status to "activated". Used to change the status of rescue resources to "dispatched".

[0135] At the first sampling time of 10:00 The state diagram construction unit obtains the state diagram. The smoke concentration was 100, the evacuation completion rate was 0.10, the affected passages were open, and the rescue resources were not dispatched; the frame classification unit was obtained according to the aforementioned graph similarity calculation method. The maximum map similarity with each contingency plan target map is 0.85; since 0.85 is not lower than the complete reference frame determination threshold of 0.82, therefore... Generate a complete reference frame record And record the corresponding current frame identifier and complete state diagram. Sampling time Complete reference frame type identifier, matching contingency plan identifier P01, event type, handling stage, and source record.

[0136] At the second sampling time of 10:01 Obtain the state diagram The smoke concentration was 84, the evacuation completion rate was 0.32, the affected passage was closed, the evacuation status was initiated, and the rescue resources status was not dispatched. The similarity between the contingency plan and the target map is 0.70, which is lower than the threshold of 0.82 for determining the complete reference frame. With complete reference frame Complete state diagram in The similarity of the reference frame coverage between them is 0.90, which is not lower than the complete reference frame determination threshold of 0.82. Therefore, for Perform prediction frame determination.

[0137] The prediction unit applies the first three actions of plan P01 to the complete reference frame I0. Four candidate prediction maps were obtained; the first two processing operations and Corresponding candidate prediction map and The graphs with the highest similarity are identified, therefore the operation cutoff number is determined to be 2. Following the aforementioned graph similarity calculation method, the structural similarity of the candidate prediction graph is 1.00, the node attribute similarity is 0.91, and the edge attribute similarity is 0.88. With each of the three graph similarity weights set to one-third, the predicted similarity is (1.00 + 0.91 + 0.88) / 3 = 0.93. Since 0.93 is not lower than the prediction frame determination threshold of 0.75, the candidate prediction graph is determined as the prediction graph, and the prediction graph is converted... The required attribute modification records form the predicted residuals. For example, the smoke concentration in the predicted graph is 82. When the smoke concentration is 84, the smoke concentration is changed from 82 to 84 and written into the prediction residual; the evacuation completion rate in the prediction graph is 0.30. When the evacuation completion rate is 0.32, the evacuation completion rate is changed from 0.30 to 0.32 and written into the prediction residual; this generates the prediction frame record. , At least associate a complete reference frame Target plan P01, Operation deadline number 2, Predicted residual and sampling time .

[0138] At the third sampling time of 10:02 Obtain the state diagram The smoke concentration was 67, the evacuation completion rate was 0.57, the affected passage was closed, the evacuation status was initiated, and rescue resources were being dispatched. The similarity between the proposed prediction and the reference frame coverage is less than 0.82, and therefore the prediction frame is determined. The prediction unit compares each candidate prediction image with the reference frame coverage. The maximum predicted similarity was 0.71; since 0.71 is lower than the prediction frame determination threshold of 0.75, no prediction frame record was generated, and the original frame was retained. The corresponding current frame identifier and sampling time Waiting for the sampling time to be formed later than A complete reference frame record or a predicted frame record.

[0139] At the fourth sampling time of 10:03 Obtain the state diagram The smoke concentration was 46, the evacuation completion rate was 0.78, the affected passage was closed, the evacuation status was initiated, and the rescue resources status was dispatched. The maximum map similarity between the target map and the corresponding plan map in the recovery and disposal phase is 0.87; since 0.87 is not lower than the complete reference frame determination threshold of 0.82, it will be... Generate a complete reference frame record ; After formation, sampling time It already has a sampling time earlier than Predicted frame record And sampling time later than Complete reference frame record The interpolation unit will Recorded as a forward reference frame, Recorded as a backward reference frame.

[0140] The interpolation unit first recovers The corresponding forward reference state diagram; according to Reading the target complete reference frame identifier Complete state diagram in Read the response sequence based on the target contingency plan identifier P01 and execute it sequentially. and Then, based on the predicted residuals Modify the attributes to restore the sampling time. State diagram Backward reference frame For complete reference frame recording, therefore read directly. Complete state diagram in .

[0141] according to =10:01 =10:02 and Interpolation parameters were calculated at 10:03. ;for and Linear interpolation is performed on all numerical attributes that exist and are already aligned. The interpolation result for smoke concentration is 84×(1-0.5)+46×0.5=65, and the interpolation result for the personnel evacuation completion rate is 0.32×(1-0.5)+0.78×0.5=0.55. The remaining nodes, edges, and attributes are processed according to the aforementioned hierarchical attributes, enumerated attributes, and object preservation rules to form a candidate interpolation graph. .

[0142] Compare candidate interpolation images according to the aforementioned graph similarity calculation method. and actual state diagram The structural similarity is 1.00, the node attribute similarity is 0.95, and the edge attribute similarity is 0.98. When the weights of the three graph similarities are all one-third, the interpolation similarity is (1.00 + 0.95 + 0.98) / 3 ≈ 0.98. Since 0.98 is not lower than the interpolation frame determination threshold of 0.70, it will be... The target interpolation map is identified; the smoke concentration in the target interpolation map is 65, while... The smoke concentration in the target plot is 67, thus creating an attribute modification residual that changes the smoke concentration from 65 to 67; the evacuation completion rate in the target interpolation plot is 0.55, while... The evacuation completion rate is 0.57, thus creating an attribute modification residual that changes the evacuation completion rate from 0.55 to 0.57; the corresponding graph transformation operations are then aggregated to form interpolation residuals. And generate sampling time Corresponding interpolation frame record , Associated forward reference frame Backward reference frame Interpolation parameter 0.5 and interpolation residual .

[0143] The resulting inter-frame dependencies include those formed by… point to The predictive dependency, and the relationship between the predicted dependency and the predicted dependency. and Pointing to respectively Interpolation dependency; when the target sampling time of the state recovery task is At that time, the state recovery module directly reads Complete state diagram in When the target sampling time is At that time, according to Read And contingency plan P01, execution ends at operation deadline number 2. and and perform prediction residuals Restore the smoke concentration to 84, the personnel evacuation completion rate to 0.32, and the corresponding nodes, edges, and other attributes to obtain the original state diagram. Consistent recovery state diagram.

[0144] When the target sampling time is At that time, the state recovery module according to Forward reference frame identifier recovery corresponding Read directly based on the backward reference frame identifier corresponding The interpolation process was re-executed with an interpolation parameter of 0.5 to obtain a basic interpolation map showing a smoke concentration of 65 and a personnel evacuation completion rate of 0.55. Then, the interpolation residuals were used to... Modify the smoke concentration to 67, the personnel evacuation completion rate to 0.57, and perform modifications to other nodes, edges, or attributes recorded in the interpolation residuals to obtain the restored state graph. ;Will Sampling time in quality verification cache Corresponding original state diagram The comparison is performed using the aforementioned graph similarity calculation method, assuming all nodes, edges, and attributes have been restored to their original recorded values. and The image similarity is 1; therefore, a complete reference frame Can be used as a prediction frame The recovery basis, predicted frame and subsequent complete reference frames It can also be used as an interpolation frame. The forward and backward recovery bases are used to recover the original state diagram at the corresponding sampling time by predicting the residuals and interpolating the residuals.

[0145] Example 2:

[0146] Please see Figure 4 As shown, this embodiment provides an emergency event status classification system, including: a data preparation module, a complete reference frame determination module, a prediction frame determination module, an interpolation frame determination module, and a status recovery module. The data preparation module includes a status acquisition unit, a status diagram construction unit, a contingency plan analysis unit, and a complete reference frame library initialization unit; the complete reference frame determination module includes a frame classification unit and a complete reference frame library management unit; the prediction frame determination module includes a prediction unit and a prediction frame record generation unit; the interpolation frame determination module includes an interpolation unit and an interpolation frame record generation unit; and the status recovery module includes an inter-frame dependency graph unit and a status recovery unit.

[0147] The data preparation module is used to acquire multi-source status data, emergency plans, and historical handling records for the same emergency event. It constructs a time-series status diagram based on the multi-source status data, obtains the plan target map, plan identifier, and handling operation sequence from the emergency plan, and establishes a complete reference frame library based on the historical handling records.

[0148] The complete reference frame determination module is used to read the current state diagram from the time-series state diagram and generate the current frame identifier, calculate the graph similarity between the current state diagram and the target diagram of the plan and the complete state diagram in the complete reference frame library, and generate a complete reference frame record or perform prediction frame determination based on the graph similarity determination result.

[0149] The prediction frame determination module is used to select a complete state map from the complete reference frame library, process the complete state map according to the disposal operation sequence to obtain a prediction map, compare the prediction map with the current state map to obtain prediction similarity and prediction residual, generate a prediction frame record containing reference frame identifier, contingency plan identifier and prediction residual according to the prediction similarity determination result, or retain the current state map and perform interpolation frame determination.

[0150] The interpolation frame determination module is used to obtain reference frame records before and after the current state map, recover the forward and backward reference state maps based on the reference frame records, obtain an interpolation map based on the two reference state maps and the sampling time, compare the interpolation map with the current state map to obtain the interpolation similarity and interpolation residual, generate an interpolation frame record containing the two reference frame identifiers, interpolation parameters and interpolation residual based on the interpolation similarity determination result, or generate a complete reference frame record based on the current state map.

[0151] The state recovery module is used to write the current frame identifier and sampling time into the corresponding frame record, establish an inter-frame dependency graph based on the reference relationship in the complete reference frame record, the prediction frame record and the interpolation frame record, and restore the state graph of the target sampling time based on the frame record and the inter-frame dependency graph corresponding to the target sampling time.

[0152] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for classifying the status of emergency events, characterized in that, include: Acquire multi-source status data, emergency plans, and historical response records for the same emergency event; construct a time-series status diagram based on the multi-source status data; obtain the plan target map, plan identifier, and response operation sequence from the emergency plan; and establish a complete reference frame library based on the historical response records. Read the current state diagram from the time sequence state diagram and generate the current frame identifier. Calculate the graph similarity between the current state diagram and the target diagram of the plan and the complete state diagram in the complete reference frame library. Based on the graph similarity judgment result, generate a complete reference frame record for the current state diagram or perform prediction frame judgment. Select a complete state map from the complete reference frame library, process the complete state map according to the disposal operation sequence to obtain a prediction map, compare the prediction map with the current state map to obtain prediction similarity and prediction residual, generate a prediction frame record containing reference frame identifier, contingency plan identifier and prediction residual according to the prediction similarity determination result, or retain the current state map and perform interpolation frame determination. Obtain reference frame records before and after the current state map, recover the forward and backward reference state maps based on the reference frame records, obtain an interpolation map based on the two reference state maps and the sampling time, compare the interpolation map with the current state map to obtain interpolation similarity and interpolation residual, and generate an interpolation frame record containing two reference frame identifiers, interpolation parameters and interpolation residual based on the interpolation similarity determination result, or generate a complete reference frame record based on the current state map; Write the current frame identifier and sampling time into the corresponding frame record. Establish an inter-frame dependency graph based on the reference relationships in the complete reference frame record, the predicted frame record, and the interpolated frame record. Restore the state graph of the target sampling time based on the frame record corresponding to the target sampling time and the inter-frame dependency graph.

2. The emergency event status classification method according to claim 1, characterized in that, The multi-source status data includes sensor status, video analysis results, business system status, and manual reporting records; Methods for constructing timing state diagrams include: Synchronize and associate multiple source status data in time to obtain a set of status data under the same emergency event; State nodes are established based on object identifiers in the state data set, state edges are established based on the relationships between objects, the state time corresponding to the state node is obtained based on the collection time, and the object state, object type, state time and object relationship are written into the attributes of the corresponding node or edge to generate a time-series state diagram.

3. The emergency event status classification method according to claim 1, characterized in that, Methods for obtaining the contingency plan target map, contingency plan identifiers, and response operation sequences include: Obtain the emergency plan text corresponding to the current emergency event and read the plan identifier in the emergency plan text; construct the plan target map based on the target objects, the relationships between target objects and the target status information in the emergency plan text; extract the handling operations based on the handling steps and execution order in the emergency plan text, and arrange them into a handling operation sequence according to the execution order.

4. The emergency event status classification method according to claim 1, characterized in that, Methods for establishing a complete reference frame library include: Obtain the historical state graph, historical event type, and historical handling stage from the historical handling records; determine the historical state graph containing the complete set of nodes, the complete set of edges, node attributes, and edge attributes as the complete state graph that can be independently recovered, and write the complete state graph, the corresponding sampling time, the event type, and the handling stage into the complete reference frame record; associate and store multiple complete reference frame records according to the event type and handling stage to form a complete reference frame library.

5. The emergency event status classification method according to claim 1, characterized in that, The method for calculating graph similarity includes: Obtain the nodes, edges, node attributes, and edge attributes of the two state graphs to be compared; establish node alignment relationships and edge alignment relationships based on node identifier, node type, node attributes, edge connection relationships, and edge attributes; calculate the graph similarity between the two state graphs based on the number of aligned nodes, the number of aligned edges, node attribute similarity, and edge attribute similarity.

6. The emergency event status classification method according to claim 1, characterized in that, Methods for generating a complete reference frame record include: Calculate the graph similarity between the current state graph and each contingency target graph, and use the maximum graph similarity as the contingency matching similarity; calculate the graph similarity between the current state graph and each complete state graph in the complete reference frame library, and use the maximum graph similarity as the reference frame coverage similarity; when the contingency matching similarity is not lower than the complete reference frame judgment threshold, generate a contingency matching type complete reference frame record for the current state graph; when the contingency matching similarity is lower than the complete reference frame judgment threshold and the reference frame coverage similarity is lower than the complete reference frame judgment threshold, generate a new type of complete reference frame record for the current state graph; when the contingency matching similarity is lower than the complete reference frame judgment threshold and the reference frame coverage similarity is not lower than the complete reference frame judgment threshold, perform prediction frame judgment on the current state graph.

7. The emergency event status classification method according to claim 1, characterized in that, Methods for obtaining candidate prediction maps include: Candidate complete reference frame records are selected from the complete reference frame library whose sampling time is earlier than the current state graph sampling time and whose event type is consistent with the current emergency event. Based on the event type of the current emergency event and the node or edge endpoint object identifiers in the candidate complete reference frame records, candidate disposal operation sequences are selected from the disposal operation sequences corresponding to the emergency plan. Each candidate disposal operation sequence is divided into multiple operation prefixes from the first zero operations to the first K operations. Each operation prefix is ​​applied to the complete state graph in the corresponding candidate complete reference frame record to obtain a candidate prediction graph, where K is the number of disposal operations contained in the candidate disposal operation sequence, and the number of disposal operations contained in each operation prefix is ​​used as the candidate operation cutoff number.

8. The emergency event status classification method according to claim 7, characterized in that, Methods for obtaining predicted frame records include: Calculate the graph similarity between each candidate prediction graph and the current state graph. Use the maximum graph similarity as the prediction similarity. Determine the candidate prediction graph corresponding to the maximum graph similarity as the prediction graph. Determine the corresponding candidate complete reference frame record as the target complete reference frame record. Determine the plan identifier of the plan to which the corresponding candidate disposal operation sequence belongs as the target plan identifier. Determine the corresponding candidate operation deadline number as the operation deadline number. Determine the node addition, node deletion, edge addition, edge deletion, node attribute modification, and edge attribute modification records required to convert the prediction graph into the current state graph as the prediction residual. When the prediction similarity between the prediction graph and the current state graph is not lower than the prediction frame judgment threshold, write the current frame identifier, target complete reference frame identifier, target plan identifier, operation deadline number, prediction residual, and current sampling time into the prediction frame record.

9. The emergency event status classification method according to claim 8, characterized in that, The method for determining the prediction frame determination threshold includes: Predicted positive and negative samples are obtained from historical treatment records. Each predicted sample includes a complete reference state map, a treatment operation sequence, and a sample state map. For each predicted positive and negative sample, the first zero to the first K treatment operations of the treatment operation sequence are applied to the complete reference state map to obtain multiple candidate prediction maps. The graph similarity between each candidate prediction map and the sample state map is calculated, and the maximum graph similarity is taken as the prediction similarity of the corresponding predicted sample. The prediction similarities of 0, 1, and all predicted samples are used to form a candidate prediction threshold set. The candidate prediction threshold with the largest average positive sample pass rate and negative sample rejection rate is selected as the prediction frame determination threshold.

10. The emergency event status classification method according to claim 1, characterized in that, Methods for obtaining forward reference frame records and backward reference frame records include: When no prediction frame record is generated in the current state diagram, the current state diagram, the current frame identifier, and the sampling time are retained; forward reference candidate frame records are selected from the complete reference frame records and prediction frame records with sampling times earlier than the current state diagram, and backward reference candidate frame records are selected from the complete reference frame records and prediction frame records with sampling times later than the current state diagram. Forward reference candidate frame records and backward reference candidate frame records belonging to the same emergency event are combined into a candidate reference frame combination.

11. The emergency event status classification method according to claim 10, characterized in that, Methods for generating candidate interpolation maps include: The forward reference state map and backward reference state map in each candidate reference frame combination are recovered respectively. The interpolation parameters are calculated according to the sampling time corresponding to the forward reference state map, backward reference state map and current state map respectively. The nodes and edges in the forward reference state map and backward reference state map are aligned, and the nodes, edges, node attributes and edge attributes are interpolated according to the interpolation parameters to obtain the candidate interpolation map.

12. The emergency event status classification method according to claim 11, characterized in that, The interpolation process includes: For numerical attributes that exist and are aligned in both reference state diagrams, linear interpolation is performed based on the interpolation parameters; for graded attributes, the interpolation result is mapped to the effective grade value with the smallest absolute difference; for enumerated or text label attributes, the attribute value is taken when the attribute values ​​are the same, and the forward or backward attribute value is selected based on the interpolation parameters when the attribute values ​​are different. For a node or edge that exists only in one reference state graph, when it is in the forward reference state graph, the object retention value is a subtracted interpolation parameter, and when it is in the backward reference state graph, the object retention value is an interpolation parameter; retain nodes or edges whose object retention value is not less than the object retention threshold, and delete edges that are not retained at the endpoint nodes; The object retention threshold is determined based on the presence of single-sided nodes or edges in the transition state graph in the historical transition samples.

13. The emergency event status classification method according to claim 11, characterized in that, Methods for generating interpolated frame records include: Calculate the graph similarity between each candidate interpolation graph and the current state graph. Use the maximum graph similarity as the interpolation similarity. Determine the candidate interpolation graph corresponding to the maximum graph similarity as the interpolation graph, and determine the corresponding forward reference frame identifier, backward reference frame identifier, and interpolation parameters. Determine the records of node addition, node deletion, edge addition, edge deletion, node attribute modification, and edge attribute modification required to convert the interpolation graph into the current state graph as the interpolation residual. When the interpolation similarity is not less than the interpolation frame determination threshold, write the current frame identifier, forward reference frame identifier, backward reference frame identifier, interpolation parameters, interpolation residual, and current sampling time into the interpolation frame record.

14. The emergency event status classification method according to claim 13, characterized in that, The method for determining the interpolation frame determination threshold includes: Obtain historical interpolation positive samples and historical interpolation negative samples. Each historical interpolation sample includes a forward reference state map, a backward reference state map, a sample state map, and the sampling time of the three. Generate a sample interpolation map based on the forward reference state map, the backward reference state map, and the sampling time of the three. Calculate the interpolation similarity between the sample interpolation map and the sample state map. Form a candidate interpolation threshold set by the interpolation similarities corresponding to zero, one, and all historical interpolation samples. Select the candidate interpolation threshold with the largest average positive sample pass rate and negative sample rejection rate as the interpolation frame determination threshold.

15. The emergency event status classification method according to claim 1, characterized in that, Methods for establishing inter-frame dependency graphs include: The current frame identifier corresponding to the complete reference frame record, the predicted frame record, and the interpolated frame record is used as the frame node; the prediction dependency edge between the complete reference frame node and the predicted frame node is established based on the target complete reference frame identifier in the predicted frame record; the interpolation dependency edge between the forward reference frame node, the backward reference frame node, and the interpolated frame node is established based on the forward reference frame identifier and the backward reference frame identifier in the interpolated frame record.

16. The emergency event status classification method according to claim 15, characterized in that, Methods for restoring the state diagram at the target sampling time include: Read the corresponding target frame record based on the target sampling time; When the target frame record is a complete reference frame record, the corresponding complete state diagram is read; When the target frame is recorded as a predicted frame, the corresponding complete state diagram is read according to the target complete reference frame identifier, the handling operation sequence is read according to the target plan identifier, and the handling operations with operation sequence numbers not greater than the operation deadline sequence number are executed in the execution order. When the operation deadline sequence number is 0, no handling operation is executed. Then, the state diagram at the target sampling time is restored according to the predicted residual. When the target frame is recorded as an interpolated frame, the forward reference state diagram and the backward reference state diagram are recovered according to the forward reference frame identifier and the backward reference frame identifier, respectively. The basic interpolation diagram is generated according to the forward reference state diagram, the backward reference state diagram and the interpolation parameters. Then, the state diagram at the target sampling time is recovered according to the interpolation residual.

17. The emergency event status classification method according to claim 16, characterized in that, Methods for updating the complete reference frame library include: Obtain the recovered target state map; select complete reference frame records from the complete reference frame library whose event type matches the event type of the emergency event to which the target state map belongs, and calculate the maximum graph similarity between the target state map and the complete state map in the selected complete reference frame records; when there is no complete reference frame record with the same event type, or when the maximum graph similarity is lower than the subtype threshold, generate a new complete reference frame record for the target state map and write it into the complete reference frame library; wherein, the subtype threshold is determined based on the historical state map of the same event type.

18. An emergency event status classification system, implementing the emergency event status classification method according to any one of claims 1-17, characterized in that, include: The data preparation module is used to acquire multi-source status data, emergency plans and historical handling records for the same emergency event, construct a time-series status diagram based on the multi-source status data, obtain the plan target diagram, plan identifier and handling operation sequence from the emergency plan, and establish a complete reference frame library based on the historical handling records. The complete reference frame determination module is used to read the current state diagram from the time-series state diagram and generate the current frame identifier, calculate the graph similarity between the current state diagram and the target diagram of the plan and the complete state diagram in the complete reference frame library, and generate a complete reference frame record or perform prediction frame determination based on the graph similarity determination result. The prediction frame determination module is used to select a complete state map from the complete reference frame library, process the complete state map according to the disposal operation sequence to obtain a prediction map, compare the prediction map with the current state map to obtain the prediction similarity and prediction residual, generate a prediction frame record containing the reference frame identifier, the contingency plan identifier and the prediction residual according to the prediction similarity determination result, or retain the current state map and perform interpolation frame determination. The interpolation frame determination module is used to obtain reference frame records before and after the current state map, recover the forward reference state map and the backward reference state map based on the reference frame records, obtain the interpolation map based on the two reference state maps and the sampling time, compare the interpolation map with the current state map to obtain the interpolation similarity and interpolation residual, and generate an interpolation frame record containing the two reference frame identifiers, interpolation parameters and interpolation residual based on the interpolation similarity determination result, or generate a complete reference frame record based on the current state map; The state recovery module is used to write the current frame identifier and sampling time into the corresponding frame record, establish an inter-frame dependency graph based on the reference relationship in the complete reference frame record, the prediction frame record and the interpolation frame record, and restore the state graph of the target sampling time based on the frame record and the inter-frame dependency graph corresponding to the target sampling time.

Images