A sudden accident visualization restoration and emergency archiving method and system based on digital twinning

Abstract

The application discloses a sudden accident visual reduction and emergency archiving method and system based on digital twinning, belongs to the technical field of digital twinning and emergency information processing, and comprises the following steps: acquiring multi-source accident data related to a sudden accident site, and forming an accident event sequence facing a twinning object based on a digital twinning object; determining state change information of the twinning object and performing source correlation identification; when the same state change information is associated with at least two types of influence sources, constructing a fact state branch and at least one contrast state branch generated by shielding the influence sources; calculating a state residual error and determining contribution information; generating an accident visual reduction result based on the state change information and the contribution information, and associating to form emergency archiving data. The application can improve the explainability of accident reduction and the traceability of emergency archiving.

Description

Technical Field

[0001] This invention relates to the field of digital twin and emergency information processing technology, and in particular to a method and system for visual reconstruction and emergency archiving of sudden accidents based on digital twins. Background Technology

[0002] With the widespread deployment of sensing devices and emergency management platforms in industrial parks, public buildings, transportation hubs, and complex infrastructure, the handling of emergencies typically generates various types of data, including video surveillance, sensor data collection, equipment operation data, access control data, personnel location data, drone inspection data, emergency commands, and manual records. Existing technologies can map some of the accident data to a digital twin scenario for process playback, or compile accident records, handling records, and evidence materials into archived files.

[0003] Existing technologies typically focus on demonstrating the accident process or preserving data source records. For changes in state from mixed sources, they often rely on time proximity, source tags, rule matching, or manual judgment for interpretation. This makes it difficult to further distinguish the actual contribution of different influencing sources to the same state change. This may lead to problems such as mistaking the effect of emergency response for the natural evolution of the accident, mistaking model completion results for actual on-site observation results, or lacking verifiable evidence for archived conclusions. Especially in scenarios such as accident debriefing, response effect evaluation, and responsibility analysis, single-line playback or static archiving is insufficient to meet the requirements of interpretability, traceability, and reproducibility.

[0004] Therefore, how to uniformly organize multi-source accident data in a digital twin environment, establish a computable attribution mechanism for changes in the state of mixed sources, and link the attribution results with visualization reconstruction and emergency archiving have become technical problems that need to be further solved in this field. Summary of the Invention

[0005] This invention overcomes the shortcomings of the prior art and provides a method and system for visual reconstruction and emergency archiving of sudden accidents based on digital twins.

[0006] To achieve the above objectives, the technical solution adopted by this invention is: a method for visual reconstruction and emergency archiving of sudden accidents based on digital twins, comprising:

[0007] S1. Acquire multi-source accident data related to the scene of the sudden accident, and perform object association and time sequence organization on the multi-source accident data based on the digital twin object to obtain an accident event sequence oriented towards the twin object;

[0008] S2. Determine the state change information of the twin object during the accident process based on the accident event sequence, and perform source association identification on the state change information. The source association identification is used to determine the association relationship between the state change information and at least one source of influence. The source of influence includes at least one of accident ontology evolution, emergency response, model completion and manual correction.

[0009] S3. When the same state change information is associated with at least two types of influence sources, a fact state branch and at least one control state branch are constructed based on the state change information. The fact state branch is used to characterize the state change result under actual data conditions, and the control state branch is generated by shielding at least one of the at least two types of influence sources.

[0010] S4. Calculate the state residual between the fact state branch and the control state branch, and determine the contribution information of the at least two types of influence sources to the state change information based on the state residual;

[0011] S5. Generate an accident visualization reconstruction result based on the state change information and the contribution information, and associate the accident visualization reconstruction result, the state residual, and the contribution information to form emergency archive data.

[0012] In a preferred embodiment of the present invention, the multi-source accident data in S1 includes digital twin basic data, accident process data, and emergency response data;

[0013] The digital twin basic data includes at least one of the following: spatial structure data of the accident site, equipment object data, sensor deployment data, traffic topology data, and initial operating status data;

[0014] The accident process data includes at least one of the following: video surveillance data, sensor data, access control status data, equipment operation data, personnel positioning data, drone data, and manually reported data during the accident.

[0015] The emergency response data includes at least one of the following: emergency command instructions, equipment operation logs, rescue personnel operation records, area lockdown records, and emergency plan execution records;

[0016] Furthermore, the multi-source accident data is standardized by performing field standardization processing based on the object identifier, spatial location, collection time, and data source in the multi-source accident data.

[0017] In a preferred embodiment of the present invention, step S1 involves object association and temporal organization of the multi-source accident data based on digital twin objects to obtain an accident event sequence oriented towards the digital twin objects, including:

[0018] A set of twin objects is constructed based on the basic data of the digital twin. The set of twin objects includes multiple twin objects with object identifiers, spatial locations, object types, and initial states.

[0019] The accident process data and emergency response data are processed into event data to obtain raw event data containing event time, event object, event location, event type and event source;

[0020] Based on the set of twin objects, the original event data is mapped to determine the target twin object corresponding to the original event data;

[0021] The original event data mapped to the same target twin object are arranged according to the event time to generate an accident event sequence oriented towards the twin object.

[0022] In a preferred embodiment of the present invention, before arranging the original event data mapped to the same target twin object according to event time, the method further includes:

[0023] Obtain the data source timestamp corresponding to the original event data;

[0024] The data source timestamp is corrected according to the preset base time and data source time offset to obtain the corrected event time;

[0025] When the timestamp of the data source is uncertain, the time confidence interval is determined based on at least one of the data source synchronization status, network latency, acquisition frequency and transmission latency.

[0026] Based on the corrected event time and the time confidence interval, the original event data are arranged into a unified accident timeline;

[0027] The time confidence interval is used to limit the branch calculation time range of the fact state branch and the control state branch.

[0028] In a preferred embodiment of the present invention, the step of performing object mapping on the original event data based on the set of twin objects to determine the target twin object corresponding to the original event data includes:

[0029] Candidate twin objects are determined from the set of twin objects based on at least one of the object identifier, spatial location, event type, and data source in the original event data;

[0030] The object mapping confidence is determined based on at least one of the following: spatial distance between the candidate twin object and the original event data, object type consistency, data field consistency, and device binding relationship;

[0031] The target twin object is determined from the candidate twin objects based on the object mapping confidence.

[0032] When the confidence level of the object mapping is lower than the preset mapping threshold, the corresponding original event data is marked as object mapping data to be confirmed, and the object mapping data to be confirmed is used as masked data or downweighted data when constructing the control state branch.

[0033] In a preferred embodiment of the present invention, step S2, determining the state change information of the twin object during the accident process based on the accident event sequence, includes:

[0034] Extract the object state values ​​of the same twin object at different time locations from the sequence of accident events;

[0035] The state change field is determined based on the differences in object state values ​​at adjacent time locations;

[0036] Based on the state change field, determine the state before the change, the state after the change, the time of the change, and the duration of the change;

[0037] The state change information is generated by combining at least one of the state change fields, the state before the change, the state after the change, the time of the change, the duration of the change, the source of the associated data, and the confidence of the object mapping.

[0038] The status change field includes at least one of the following: spatial status field, equipment operation status field, environmental status field, access status field, personnel status field, and hazardous area status field.

[0039] In a preferred embodiment of the present invention, step S2, which involves identifying the source of the state change information, includes:

[0040] The state change information is matched with the accident evolution rule base to obtain the accident evolution matching degree;

[0041] The state change information is matched with the emergency response action rule base to obtain the emergency response matching degree;

[0042] The model completion matching degree is obtained based on whether the state change information is filled in by missing data, inferred from adjacent data, or generated by a simulation model.

[0043] The manual correction matching degree is obtained based on whether the state change information is manually reported, manually reviewed, or manually overwritten.

[0044] Based on the accident evolution matching degree, emergency response matching degree, model completion matching degree, and manual correction matching degree, the source association results corresponding to the state change information are generated.

[0045] When at least two of the accident evolution matching degree, emergency response matching degree, model completion matching degree, and manual correction matching degree reach the corresponding threshold, and the corresponding data are related in time, object, or space, it is determined that the same state change information is associated with at least two types of influence sources.

[0046] In a preferred embodiment of the present invention, step S3, which involves constructing a factual state branch and at least one control state branch based on the state change information, includes:

[0047] The state before the change in the state change information is used as the common starting point for the branching of the factual state branch and the control state branch, and the branch calculation window is determined by the time of the change and the duration of the change.

[0048] Based on the effective data from the actual acquired accident process data, emergency response data, model completion data, and manually corrected data, a fact state branch is generated to characterize the state after factual changes under actual data conditions.

[0049] The control state branch includes at least one of the following:

[0050] When the state change information is simultaneously associated with the accident ontology evolution and emergency response, emergency response data that is spatiotemporally related to the state change information is shielded, and a first comparison state branch is generated based on the pre-change state, accident process data, and accident evolution rule base.

[0051] When the state change information is associated with both accident ontology evolution and model completion, the model completion data related to the state change information is masked, and a second control state branch is generated based on the state before the change, the actual observed accident process data, and the accident evolution rule base.

[0052] When the status change information is associated with both emergency response and manual correction, the manual correction data related to the status change information is masked, and a third comparison status branch is generated based on the pre-change status, accident process data, and emergency response action rule base.

[0053] In a preferred embodiment of the present invention, step S4 involves calculating the state residual between the factual state branch and the control state branch, and determining the contribution information of the at least two types of influence sources to the state change information based on the state residual, including:

[0054] When a first control state branch exists, calculate the first state residual between the state after the fact change and the first predicted state corresponding to the first control state branch, and determine the emergency response contribution based on the first state residual.

[0055] When a second control state branch exists, calculate the second state residual between the state after the fact change and the second predicted state corresponding to the second control state branch, and determine the model completion contribution based on the second state residual.

[0056] When a third control state branch exists, calculate the third state residual between the state after the fact change and the third predicted state corresponding to the third control state branch, and determine the amount of artificial correction based on the third state residual.

[0057] The contribution of accident evolution is determined based on the total state change corresponding to the state change information, as well as at least one of the determined emergency response contribution, model completion contribution, and manual correction influence.

[0058] Normalize at least two of the following: the contribution of accident evolution, the contribution of emergency response, the contribution of model completion, and the influence of manual correction, to obtain the corresponding contribution information, and generate a dominant source identifier based on the contribution information.

[0059] The state residual includes at least one of the following: numerical difference, relative rate of change, standardized distance, state level difference, state transition cost, regional coverage area difference, boundary offset, regional intersection-to-union ratio difference, trajectory distance difference, and topology access cost difference.

[0060] A digital twin-based system for visualizing and reconstructing sudden accidents and for emergency archiving includes:

[0061] The data association module is used to acquire multi-source accident data related to the scene of the sudden accident, and to perform object association and time-series organization on the multi-source accident data based on the digital twin object to obtain an accident event sequence oriented towards the twin object;

[0062] The state change determination module is used to determine the state change information of the twin object during the accident process based on the accident event sequence.

[0063] The source identification module is used to identify the source association of the state change information, determine the association between the state change information and at least one source of influence, and trigger branch attribution processing when the same state change information is associated with at least two sources of influence, wherein the sources of influence include at least one of accident ontology evolution, emergency response, model completion and manual correction;

[0064] The branch attribution module is used to construct a fact state branch and at least one control state branch based on the state change information, calculate the state residual between the fact state branch and the control state branch, and determine the contribution information of the at least two types of influence sources to the state change information based on the state residual, wherein the control state branch is generated by masking at least one of the at least two types of influence sources.

[0065] The visualization archiving module is used to generate an accident visualization reconstruction result based on the state change information and the contribution information, and to associate the accident visualization reconstruction result, the state residual, and the contribution information to form emergency archive data.

[0066] This invention addresses the shortcomings of the prior art and has the following beneficial effects:

[0067] (1) This invention acquires multi-source accident data and performs object association and temporal organization based on digital twin objects to form an accident event sequence oriented towards twin objects. Then, it performs source association identification, branch attribution and visual archiving association on the state change information. Since both accident reconstruction and emergency archiving revolve around the same state change information, it can directly achieve the two purposes of visual reconstruction of the accident process and traceability of archived evidence. Compared with the existing technology where accident playback and archive retention are separated, this invention can further improve the integrity of emergency accident review and the reliability of subsequent review.

[0068] (2) This invention converts multi-source accident data into an accident event sequence for twin objects and determines the state change information of twin objects during the accident process based on the accident event sequence. Since heterogeneous data such as video, sensors, equipment logs, personnel positioning and emergency response records are unified under the dimensions of object, time and state change, a continuous and structured accident reconstruction basis can be directly formed. Compared with the existing technology that relies on discrete data splicing or single time axis playback, this invention can further improve the spatiotemporal consistency and process interpretation ability of accident visualization reconstruction.

[0069] (3) This invention identifies the sources of state change information, constructs factual state branches and control state branches when the same state change information is associated with at least two types of influence sources, and calculates state residuals by masking at least one type of influence source. Since mixed source state changes no longer rely solely on the highest matching degree or source label for judgment, the contribution of different influence sources to state changes can be directly determined. Compared with the existing technology, which is difficult to distinguish between the natural evolution of accidents, emergency response and model completion effects, this invention can further improve the objectivity of attribution results and the accuracy of emergency response assessment.

[0070] (4) This invention associates state change information, state residuals and contribution information with the results of accident visualization and reconstruction, and forms emergency archive data. Since the visualization results not only show the state change process, but also simultaneously retain the factual state, the comparison state and the residual attribution basis, they can directly form an archive evidence chain that can be displayed, explained and verified. Compared with the existing technology where the visualization scene and archived files are separated and the archived conclusions are difficult to reproduce, this invention can further enhance the long-term utilization value of accident investigation, responsibility analysis and emergency archives. Attached Figure Description

[0071] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0072] Figure 1 This is a flowchart of a preferred embodiment of the present invention;

[0073] Figure 2 This is a system structure block diagram of the present invention. Detailed Implementation

[0074] Application Overview:

[0075] This invention can be applied to emergency response and debriefing scenarios in industrial parks, chemical plant areas, underground integrated pipe corridors, rail transit stations, urban transportation hubs, public buildings, warehousing and logistics parks, and large-scale energy infrastructure.

[0076] In the above scenarios, after an accident occurs, various types of data are typically generated, including video surveillance data, sensor data, access control data, equipment operation logs, personnel location data, drone inspection images, emergency command instructions, action records of handling equipment, operation records of rescue personnel, and manual reporting records.

[0077] The data sources, sampling frequencies, object identification rules, and timestamp precisions are different. If they are simply spliced ​​together according to a single timeline, it is easy to cause problems such as broken accident processes, incorrect object attribution, unclear sources of state changes, and lack of verification basis for archived conclusions.

[0078] This invention uses digital twin objects as the unified organizational basis for multi-source accident data, mapping objects such as on-site space, equipment, personnel, passages, hazardous areas and emergency resources into computable, displayable and archiveable twin objects, and forming an accident event sequence around the twin objects.

[0079] The system further identifies the state changes of the twin object during the accident process and determines whether the state changes are simultaneously affected by different sources such as accident ontology evolution, emergency response, model completion, and manual correction.

[0080] When the same state change has multiple sources of influence, the system constructs a factual state branch and a control state branch. By masking the state differences after a certain source of influence is eliminated, the corresponding contribution information is determined, and then a visual restoration result and emergency archive data are generated.

[0081] Example 1:

[0082] like Figure 1 As shown, this method for visualizing and reconstructing sudden accidents and archiving emergencies based on digital twins includes the following steps:

[0083] S1. Acquire multi-source accident data related to the scene of the sudden accident, and perform object association and time sequence organization on the multi-source accident data based on the digital twin object to obtain an accident event sequence oriented towards the twin object;

[0084] S2. Determine the state change information of the twin object during the accident process based on the accident event sequence, and perform source association identification on the state change information. The source association identification is used to determine the association relationship between the state change information and at least one source of influence. The source of influence includes at least one of accident ontology evolution, emergency response, model completion and manual correction.

[0085] S3. When the same state change information is associated with at least two types of influence sources, a fact state branch and at least one control state branch are constructed based on the state change information. The fact state branch is used to characterize the state change result under actual data conditions, and the control state branch is generated by shielding at least one of the at least two types of influence sources.

[0086] S4. Calculate the state residual between the fact state branch and the control state branch, and determine the contribution information of the at least two types of influence sources to the state change information based on the state residual;

[0087] S5. Generate an accident visualization reconstruction result based on the state change information and the contribution information, and associate the accident visualization reconstruction result, the state residual, and the contribution information to form emergency archive data.

[0088] The above method maps multi-source heterogeneous data generated at the scene of a sudden accident onto a digital twin object, transforming the accident data from scattered records into an object-oriented, time-oriented, and state-change-oriented sequence of accident events. The system identifies the state changes of the twin object and performs source association identification on the state changes.

[0089] For a state change that is simultaneously affected by multiple sources, the system obtains the state residual by comparing the fact state branch with the control state branch, and then determines the degree of contribution of different sources to the state change based on the state residual.

[0090] Accident reconstruction results can not only show the accident process, but also explain which data sources contributed to the changes in state, and can preserve the corresponding evidence and calculation basis in emergency archives.

[0091] In practical applications, multi-source accident data suffers from problems such as inconsistent object identification, inconsistent spatial location descriptions, offset timestamps, different collection frequencies, and some data being generated by model completion or manual correction.

[0092] If data is not correlated at the object level and corrected for time sequence, it is easy to mix up the state changes of different objects or to mistakenly include time-irrelevant data in the same incident segment during incident review.

[0093] For state changes from mixed sources, traditional single-line playback methods can usually only display the final state, making it difficult to distinguish the actual impact of natural accident evolution, emergency response actions, model completion results, and manual correction records on the state change.

[0094] This invention improves the interpretability of accident reconstruction and the traceability of emergency archiving by forming a continuous processing chain through object association, temporal organization, source identification, branch construction, and residual attribution.

[0095] Example 2:

[0096] In step S1, the system receives various types of data related to the accident from the accident site and related business platforms, converts these data into a unified format, and then assigns each piece of data to the corresponding equipment, space, personnel, passage, area or emergency resource object according to the object information in the digital twin system.

[0097] The system arranges the data under the same twin object according to a unified accident timeline to form the event sequence of that object during the accident process.

[0098] The multi-source accident data includes digital twin basic data, accident process data, and emergency response data.

[0099] Digital twin basic data includes at least one of the following: spatial structure data of the accident site, equipment object data, sensor deployment data, traffic topology data, and initial operating status data.

[0100] Accident process data includes at least one of the following: video surveillance data, sensor data, access control status data, equipment operation data, personnel location data, drone data, and manually reported data during the accident.

[0101] Emergency response data includes at least one of the following: emergency command instructions, equipment operation logs, rescue personnel operation records, area lockdown records, and emergency plan execution records.

[0102] A digital twin object refers to a data object used in a digital twin scenario to represent a physical entity, spatial unit, or logical unit in a real accident scene.

[0103] Digital twin objects can include buildings, floors, rooms, pipelines, valves, cameras, sensors, access control equipment, fire-fighting equipment, vehicles, personnel, hazardous areas, evacuation routes, and emergency resource points, etc.

[0104] Each digital twin object has at least an object identifier, object type, spatial location, and initial state.

[0105] Preferably, the digital twin object also has geometric boundaries, topological connections, bound devices, bound data sources, and a set of state fields.

[0106] Object identifiers refer to the numbers, names, equipment IDs, asset codes, GIS codes, BIM component codes, or platform internal object codes used to distinguish different twin objects.

[0107] Spatial location refers to the coordinates, floor number, area number, grid number, or topology node location of the twin object in the digital twin scene.

[0108] Event time refers to the time when incident data is collected, generated, recorded, or corrected.

[0109] The data source refers to the source of the data, which is from cameras, sensors, equipment control systems, access control systems, positioning systems, drones, manual reporting terminals, or emergency command platforms.

[0110] When acquiring multi-source accident data, the system uses a combination of data interface collection, file import, message subscription, and manual entry.

[0111] Different accident data are usually distributed across multiple systems. The system receives data output from each data source through a unified data access layer and generates an access record for each piece of access data.

[0112] Access records must include at least the data source number, access time, original timestamp, original object identifier, original spatial description, data format, integrity identifier, and verification result.

[0113] During execution, the system first reads the basic data of the digital twin to form a list of on-site objects and an object relationship diagram; then, based on the accident start time, accident end time, and a preset extended time window, it extracts relevant data from the video surveillance system, sensor system, equipment management system, access control system, positioning system, drone platform, and emergency command platform; subsequently, it converts the fields from different data sources into a unified field structure; for missing fields, abnormal formats, duplicate records, and data that clearly exceeds the reasonable range, the system does not delete them directly, but marks the data quality status so that it can be downgraded or blocked during subsequent source identification and branch attribution.

[0114] The start time of an accident can be determined by the alarm trigger time, the manual confirmation time, or the earliest abnormal sensor time.

[0115] The time for an incident to end can be determined by the time the on-site alert is lifted, the time the emergency response ends, or the time the system returns to normal.

[0116] A preset extended time window is used to retain the baseline state before the accident and the recovery information after the accident. It is preferably 5 to 60 minutes before the accident start time and 5 to 60 minutes after the accident end time.

[0117] For accidents that change rapidly, such as chemical leaks, fire spread, and personnel evacuation, a smaller extended time window can be used; for scenarios where the accident evolves slowly or the response takes a long time, a larger extended time window can be used.

[0118] For field standardization, the system converts data from different sources into unified event data that includes event number, object identifier, event location, event time, event type, data source, event value, and data quality identifier.

[0119] Event values ​​can be sensor readings, equipment open / closed status, personnel location, passageway status, hazardous area range, or manually recorded content.

[0120] Data quality identifiers are used to indicate whether the data is a true observation, model completion, manual correction, delayed reporting, low-confidence mapping, or data to be confirmed.

[0121] Through this unified structure, the system can convert data in different formats into data units that can be processed by the same sequence of accident events.

[0122] The object association processing mechanism is that after the accident data is assigned to a specific twin object, the system can further determine the state changes of that object during the accident process.

[0123] For example, if smoke sensor alarms, smoke appearing in camera footage, a certain area being sealed off, and personnel location points deviating from evacuation routes are all related to the same floor or the same room, then these data should be organized under the corresponding spatial object or related equipment object, rather than being stored as unrelated isolated data.

[0124] During execution, the system constructs a set of twin objects based on the digital twin's basic data and extracts object identifiers, spatial locations, event types, and data sources from the original event data.

[0125] If the object identifier in the original event data matches the twin object identifier, the system will prioritize using that identifier for mapping.

[0126] If the object identifier is missing or inconsistent, the system determines the candidate twin object by combining spatial distance, object type consistency, data field consistency and device binding relationship.

[0127] Candidate twin objects are twin objects that are related to the original event data in terms of space, type, or data source.

[0128] The confidence score of an object mapping can be obtained through a weighted scoring method.

[0129] The scoring factors include the degree of matching of object identifiers, the degree of proximity of spatial distance, the degree of consistency of object type, the degree of consistency of field attribution, and the degree of consistency of device binding.

[0130] For device log data, the system increases the weight of object identifiers and device binding relationships; for video recognition data, the system increases the weight of spatial location and object type; for manually reported data, the system increases the weight of spatial description and field attribution.

[0131] The confidence score for object mapping ranges from 0 to 1. A higher value indicates that the original event data is more likely to belong to the twin object.

[0132] The object mapping threshold is preferably 0.65 to 0.85, more preferably 0.75.

[0133] When the highest object mapping confidence is lower than the object mapping threshold, the system marks the original event data as object mapping data to be confirmed, and treats it as masked data or downweighted data when constructing the comparison state branch in the future.

[0134] The processing mechanism of chronological organization is that accident reconstruction requires not only clarifying which events occurred for a certain object, but also clarifying the chronological relationship of these events on a unified accident timeline.

[0135] Because different data sources may have local clock offsets, network transmission delays, batch upload delays, and manual recording lags, the system needs to correct the original timestamps and provide time confidence intervals for events whose times cannot be accurately determined.

[0136] During execution, the system reads the data source timestamp corresponding to the original event data and corrects it according to the preset base time and data source time offset.

[0137] When there is a significant transmission delay in the data source, the system further introduces a delay correction amount; the correction event time is determined according to the following formula:

[0138] ,

[0139] in, Indicates the time of the correction event; Indicates the original timestamp of the data source; This indicates the time offset of the data source relative to the reference time; This represents the correction amount for transmission delay or upload delay. It can be estimated through unified clock synchronization records before and after the incident, device heartbeat logs, or the time difference of the same event in multiple data sources. It can be obtained from network logs, message queue reception time, packet arrival time, or historical average latency.

[0140] This process allows event times recorded by different systems to be unified onto the same accident timeline, reducing misjudgments of the order of state changes caused by clock inconsistencies.

[0141] When timestamps are uncertain, the system determines the time confidence interval based on the data source synchronization status, network latency, collection frequency, and transmission latency.

[0142] The lower the sampling frequency, the greater the network latency fluctuation, the worse the data source clock synchronization status, and the wider the time confidence interval.

[0143] For high-frequency synchronous sensors, the time confidence interval can be set to milliseconds or seconds; for manually reported records, the time confidence interval can be set to minutes.

[0144] This time confidence interval is used to limit the time range for branch calculations of the fact state branch and the control state branch, so that potentially relevant data will not be excluded due to slight time deviations, and obviously irrelevant data will not be included in the same state change calculation window.

[0145] Through the above processing, S1 solves the problem of the difficulty in uniformly organizing multi-source heterogeneous accident data.

[0146] The system converts scattered accident data into accident event sequences oriented towards twin objects, enabling subsequent state change identification to be based on a unified object, a unified timeline, and a unified data structure, thereby reducing reconstruction deviations caused by misattribution of objects and out-of-order time.

[0147] In step S2, the system reads the accident event sequence of each twin object, compares the state differences of the object at different time points or different time windows, and determines whether the object has undergone a state change; then, based on the location, time, data source, direction of change, and data generation method of the state change, it determines which sources of influence the state change may be related to.

[0148] State change information refers to structured data that describes the transition of a twin object from one state to another during an accident.

[0149] Status change information may include status change fields, status before change, status after change, time of change, duration of change, associated data source, object mapping confidence, and data quality identifier.

[0150] The status change field includes at least one of the following: spatial status field, equipment operation status field, environmental status field, access status field, personnel status field, and hazardous area status field.

[0151] The spatial status field may include smoke coverage, visibility, and containment boundary; the equipment operation status field may include valve open / closed status, fan operation status, and fire pump operation status; the environmental status field may include temperature, smoke concentration, gas concentration, and pressure; the passage status field may include whether the passage is passable, closed, congested, or blocked; the personnel status field may include location, direction of movement, status of stay, and evacuation status; and the hazardous area status field may include hazard level, coverage area, and boundary offset.

[0152] The mechanism for processing state change information is that state change is not a single piece of data itself, but rather the state difference of the same twin object in the time dimension.

[0153] The system extracts the state before the change, the state after the change, the time of the change, and the duration of the change from the sequence of accident events, and then analyzes the source and contribution of the change.

[0154] During execution, the system extracts the object state value from the sequence of incident events of the same twin object.

[0155] For numerical status fields, the system compares the numerical differences between adjacent time positions or adjacent time windows; for hierarchical status fields, the system compares whether a hierarchical transition has occurred; for regional status fields, the system compares whether the regional area, regional boundary, or raster set has changed; for trajectory status fields, the system compares the direction of movement, the degree of trajectory deviation, and the dwell time; for topology accessibility fields, the system compares the channel connectivity status and access cost.

[0156] If the state difference reaches the change threshold of the corresponding field, the system marks the field as a state change field and generates state change information.

[0157] The status change threshold is determined based on the field type and on-site business requirements.

[0158] For continuous numerical fields such as temperature, concentration, and pressure, the change threshold can be determined based on the equipment range, alarm threshold, historical fluctuation range, and sensor accuracy.

[0159] For discrete fields such as hazard level and equipment open / closed status, status change recognition can be triggered as long as the level or status changes.

[0160] For the boundary of a dangerous area, whether it constitutes a state change can be determined based on the proportion of area change, boundary offset distance, or number of grid changes.

[0161] For personnel trajectories, it can be determined whether a change in status has occurred based on the distance from the preset evacuation route, the duration of stay, and the change in the direction of movement.

[0162] The mechanism for source association identification is that the same state change may not be caused by a single source.

[0163] For example, a decrease in smoke concentration in a certain area may be related to the start-up of smoke exhaust fans, changes in natural diffusion trends, and model completion; a passage being marked as impassable may be related to access control logs, abnormal personnel location, and manual correction.

[0164] The system determines the relationship between state changes and different sources of influence by using rule matching, time correlation, spatial correlation, object correlation, and data generation methods.

[0165] During execution, the system matches the state change information with the accident evolution rule base to determine whether the change conforms to the development law of the accident itself.

[0166] The accident evolution rule base can include fire spread rules, smoke diffusion rules, hazardous gas diffusion rules, equipment abnormal propagation rules, and personnel congestion diffusion rules.

[0167] The system also matches the status change information with the emergency response action rule base to determine whether the change may have been caused by an emergency response action.

[0168] The emergency response action rule library can include actions such as starting smoke exhaust fans, activating sprinkler systems, closing valves, sealing off areas, broadcasting evacuation information, guiding rescue efforts, and starting and stopping emergency equipment, as well as their expected impacts.

[0169] The system determines the degree of correlation of model completion based on whether the state change information is filled in by missing data, inferred from adjacent data, or generated by simulation model.

[0170] If the changed state is entirely derived from real observation data, the model completion correlation is low; if the changed state is mainly generated by interpolation, simulation, or prediction results, the model completion correlation is high.

[0171] The system also determines the degree of correlation between manual corrections and status change information based on whether the information was manually reported, manually reviewed, or manually overwritten.

[0172] If a manually corrected record changes a status field, hazard level, access status, or visual boundary, then that status change is related to the manual correction.

[0173] The source association threshold can be set based on data reliability and incident type.

[0174] The matching thresholds for accident evolution and emergency response are preferably between 0.60 and 0.85.

[0175] The model completion matching threshold can be set to 0.30 to 0.60, because even if the proportion of model completion is small, it may change the final visualization boundary. The manual correction matching threshold is preferably 0.50 to 0.80.

[0176] When at least two types of influence sources reach the corresponding threshold, and the corresponding data are related in time, object or space, the system determines that the same state change information is associated with at least two types of influence sources. Through the above processing, S2 solves the problem of the difficulty in distinguishing the sources of state changes.

[0177] The system can first identify which state changes belong to single-source changes and which state changes belong to mixed-source changes, so that it only performs branch attribution on state changes with multi-source associations, avoiding indiscriminate calculation of all state changes and reducing computational redundancy.

[0178] In step S3, when the system determines that a certain state change is simultaneously affected by multiple sources of influence, the system takes the state before the change as a common starting point and generates a factual state branch while retaining the actual valid data. At the same time, by shielding one of the sources of influence, the system recalculates the control state to form a control state branch that can be compared with the factual state.

[0179] The factual state branch refers to the state change result generated while retaining valid data from the actual acquired accident process data, emergency response data, model completion data, and manually corrected data.

[0180] The contrast state branch refers to the state change result generated after blocking at least one type of influence source within the same object, the same state field, the same state before the change, and the same calculation window.

[0181] Masking does not necessarily mean deleting the original data. It can also mean resetting the weight of this type of data to zero in the branch calculation, marking it as not participating in the state update, or simply retaining it as archived evidence without participating in the state calculation.

[0182] The mechanism for constructing fact state branches and control state branches is as follows: the contribution of mixed source state changes cannot be directly judged by the final state value alone. It is necessary to construct a control result that is as consistent as possible in other conditions and only lacks a certain type of influence source.

[0183] The factual state branch represents the state change under actual data conditions, while the contrast state branch represents the state change after a certain type of source is blocked. The difference between the two can reflect the impact of the blocked source on the state change.

[0184] During execution, the system uses the pre-change state in the state change information as the common starting point for the branching of the fact state branch and the control state branch, and determines the branch calculation window based on the time of change, the duration of change, and the time confidence interval.

[0185] The branch calculation window covers the valid data segments before and after the state change, ensuring that the fact state branch and the control state branch are calculated within the same time frame.

[0186] The system then extracts data related to the target twin object and its associated objects within the window, and generates the post-change state based on the actual valid data.

[0187] For state changes that are simultaneously related to the evolution of the accident itself and emergency response, the system shields emergency response data that are spatiotemporally related to the state change, and generates an emergency response comparison state branch based on the state before the change, accident process data, and accident evolution rule base.

[0188] For example, if the smoke concentration in a certain area decreases after the exhaust fan is started, the factual state branch retains the exhaust fan start-up log and smoke sensor data; the emergency response control state branch does not allow the exhaust fan start-up record to participate in the state update, but only infers the state of the area before the response action was taken based on the accident process data and accident evolution rules.

[0189] For state changes that are simultaneously associated with accident ontology evolution and model completion, the system masks the model completion data related to the state change and generates a model completion comparison state branch based on the state before the change, the actual observed accident process data, and the accident evolution rule base.

[0190] For example, some cameras may be unable to identify dangerous boundaries due to smoke obscuring them. The system uses adjacent sensors and diffusion rules to complete the boundaries of dangerous areas. The model completion control state branch only uses the actually observed boundary points and sensor data to generate a control state in order to determine the impact of model completion on the final boundary.

[0191] For status changes that are simultaneously associated with emergency response and manual correction, the system blocks the manual correction data related to the status change and generates a manual correction comparison status branch based on the pre-change status, accident process data, and emergency response action rule base.

[0192] For example, if on-site personnel manually change a passage to be impassable, but the access control log still shows it as open, the factual status branch uses the manually changed passage status; the manual correction comparison status branch does not use the manually changed record, but only calculates the passage status based on the access control status, personnel location, and handling records.

[0193] The length of the branch calculation window is determined based on the status field type.

[0194] For changes in sensor values, the window can be set to several seconds to several minutes before and after the change occurs; for the spread of dangerous areas, the window can be set to several minutes to tens of minutes; for personnel evacuation trajectories, the window can cover the entire process of personnel entering the dangerous area, receiving evacuation instructions, and leaving the dangerous area.

[0195] If the time confidence interval is wide, the branch calculation window can be appropriately expanded, but it must not exceed the maximum window preset for incident review to prevent irrelevant events from entering the branch calculation.

[0196] Through the above processing, S3 solved the problem of lacking a basis for comparative analysis of changes in the state of mixed sources.

[0197] Without altering the original accident data, the system generates a reproducible control state by masking the source, enabling comparisons between different sources of influence under the same object, time window, and state field.

[0198] In step S4, the system compares the changed state obtained from the fact state branch with the changed state obtained from each control state branch to obtain the state residual.

[0199] The larger the state residual, the more significant the impact of the blocked source on the change in the factual state.

[0200] The system then combines the confidence level of the source data to convert the state residuals into contribution values, and normalizes multiple contribution values ​​to obtain contribution information.

[0201] State residuals include at least one of the following: numerical difference, relative rate of change, standardized distance, state level difference, state transition cost, regional coverage area difference, boundary offset, regional intersection-union ratio difference, trajectory distance difference, and topology access cost difference.

[0202] Contribution information refers to data used to describe the degree of contribution of different sources of influence to the same state change, which may include contribution amount, contribution ratio, dominant source identifier, residual type, residual confidence level, and corresponding evidence index.

[0203] The calculation mechanism for state residuals is that the fact state branch and the control state branch have the same pre-change state, object scope, state field, and time window. The difference between the two mainly comes from the masked sources of influence.

[0204] By comparing the state after the fact changes with the state after the control changes, we can obtain the degree of influence of the source on the state change.

[0205] During execution, the system selects the residual measurement method based on the type of the status field.

[0206] For numerical fields such as temperature, gas concentration, smoke concentration, pressure, and personnel density, the system uses normalized numerical difference or relative rate of change to calculate residuals.

[0207] For level-based fields such as hazard level, access level, and equipment alarm level, the system uses the level difference to calculate the residual.

[0208] For spatial fields such as hazardous areas, smoke-covered areas, and sealed-off areas, the system uses the area difference, degree of overlap, and boundary offset to calculate the residual.

[0209] For trajectory fields such as personnel evacuation trajectory, vehicle movement trajectory, and drone inspection trajectory, the system calculates the residual using trajectory distance difference or path deviation degree.

[0210] For the passage topology field, the system uses the shortest passage cost difference or passage impedance difference to calculate the residual.

[0211] To facilitate unified processing across different state fields, the state residuals adopt the following general expression:

[0212] ,

[0213] in, Indicates source blocking The resulting state residuals; This indicates the state after a fact change within a fact state branch; Indicates source blocking The control group obtained after the change; This represents the difference measurement function corresponding to the status field type.

[0214] For numeric fields This represents the normalized numerical difference or relative rate of change; for rank fields, For grade differences; for regional fields, This is a combined difference resulting from regional area differences, boundary offsets, and regional intersection-union ratio differences; for trajectory-type fields, This represents the difference in trajectory distance or the degree of path deviation; for topology fields, The difference in cost of passage.

[0215] This expression only limits the core data relationship for residual calculation. The specific difference measurement method is selected according to the state field type, thereby avoiding direct mixing of data with different dimensions.

[0216] The mechanism for determining contribution information is that a certain control state branch corresponds to a certain masked source, so the residual between the control state branch and the fact state branch can be used as the basis for calculating the contribution of that source.

[0217] If the residual is large after shielding the emergency response, it indicates that the emergency response has a significant impact on the change in the actual state. If the residual is large after shielding the model completion, it indicates that the model completion has a significant impact on the final state result. If the residual is large after shielding the manual correction, it indicates that the manual correction has a significant impact on the final archived conclusion.

[0218] During execution, the system determines the contribution of emergency response based on the state residuals of the emergency response control state branch, determines the contribution of model completion based on the state residuals of the model completion control state branch, and determines the impact of manual correction based on the state residuals of the manual correction control state branch.

[0219] The above contribution figures can also be adjusted based on the confidence level of the source data.

[0220] The confidence level of source data is determined by the reliability of the data source, the confidence level of the object mapping, the time confidence interval, the amount of evidence, and the review status.

[0221] Equipment action logs, automatic sensor data collection records, and consistent data from multiple sources typically have high confidence levels; data formed solely by a single manual record and lacking supporting evidence typically have low confidence levels.

[0222] The contribution of accident evolution can be determined based on the relationship between the total state change and the contributions from other sources.

[0223] Once the contributions of emergency response, model completion, and manual correction have been determined, the system will use the remaining portion that can be explained by the accident process data and accident evolution rules as the accident evolution contribution.

[0224] If the remaining amount is less than zero due to multiple overlapping sources or data errors, the contribution of the accident evolution will be set to zero to avoid negative contribution results.

[0225] Contribution information was obtained using the following normalization method:

[0226] ,

[0227] ,

[0228] in, Indicate source The amount of contribution; Indicates source blocking The resulting state residuals; Indicate source Confidence level of the source data; Indicate source The proportion of contribution; This represents the sum of contributions from all attribution sources. It traverses the sources of the accident, including its evolution, emergency response, model completion, and manual corrections. It is a very small positive number, used to avoid the denominator being zero.

[0229] Through normalization, the contributions from different sources are converted into contribution ratios between 0 and 1, which makes it easier to display in the visualization interface and to retrieve and verify in the emergency archive.

[0230] The dominant source identifier is determined based on contribution information. If the contribution ratio of a certain source is significantly higher than that of other sources, the system marks that source as the dominant source.

[0231] If the difference between the largest contribution ratio and the second largest contribution ratio is small, the system marks the state change as being dominated by multiple sources. The threshold for the difference in dominance is preferably 0.10 to 0.20, and more preferably 0.15.

[0232] This judgment method can avoid forcibly designating a single dominant source when the contributions from multiple sources are similar, thus improving the objectivity of the attribution results.

[0233] Through the above processing, S4 solves the problem of the inability to quantify the attribution of state changes from mixed sources.

[0234] The system establishes a data processing chain that includes source masking, state recalculation, residual calculation, and contribution normalization, enabling accident reconstruction to be upgraded from single-line playback to multi-source attribution reconstruction that is calculable, interpretable, and verifiable.

[0235] In step S5, the system maps the state change information and contribution information to the digital twin scenario, generates a visual reconstruction result of the accident process, and uniformly associates and saves the visualization result, factual state branches, comparison state branches, state residuals, contribution information and original evidence index to form emergency archive data.

[0236] The results of the accident visualization reconstruction include at least one of the following: accident timeline, animation of twin object status changes, hazardous area change layer, action markers, source contribution markers, state residual comparison view, fact branch and control branch switching view, and archived evidence index.

[0237] Emergency archived data includes at least one of the following: raw data index, standardized event data, object mapping results, time correction results, status change information, source association results, branch calculation parameters, factual status branches, control status branches, status residuals, contribution information, dominant source identifiers, visual snapshots, and review records.

[0238] The mechanism for generating the visual reconstruction results of accidents is that accident reviewers not only need to read the data tables, but also need to intuitively see how the accident status changes, when the response actions are taken, and how different sources affect the status changes in the digital twin scenario.

[0239] The system maps status change information and contribution information onto twin objects, region layers, personnel trajectories, equipment status, and timelines.

[0240] During execution, the system generates a unified incident timeline based on the sequence of incident events and drives the state update of digital twin objects according to this timeline.

[0241] For each status change message, the system generates a visual node on the corresponding object, region, or path, and displays the source contribution based on the contribution information.

[0242] If the contribution ratio of emergency response is high, the system displays the correlation between response actions and state changes in the state change node; if the contribution ratio of model completion is high, the system displays that the state change includes the impact of model completion; if the impact of manual correction is high, the system displays the manual correction record and review status.

[0243] The system can also provide a switchable display of factual state branches and comparison state branches, enabling reviewers to compare the differences in state after retaining a certain type of source versus blocking a certain type of source.

[0244] The mechanism for forming emergency archived data is that accident archives not only preserve the final report, but also retain the intermediate processing link from raw data to visualized conclusions.

[0245] The system establishes an index relationship between each state change node and its corresponding data source, object mapping, time correction, source identification, branch calculation, and residual attribution results, so that any visualized conclusion can be traced back to the original evidence and calculation basis.

[0246] During execution, the system generates a unique archive number for each status change message and associates this archive number with the twin object identifier, status field, time window, and incident sequence number.

[0247] The system saves the input data index, masking source type, branch calculation window, and status update method for the factual status branch and the comparison status branch; it saves the status residual type, residual value, contribution amount, contribution ratio, and dominant source identifier; and it saves the visual snapshot, review comments, archive generation time, and archive version number.

[0248] Emergency archived data may include archive number, object identifier, time window, status field, status before change, status after change, set of status after change, set of status residuals, set of contribution information, dominant source identifier, evidence index, visualization index, and review record.

[0249] This structure ensures that the results of the accident visualization reconstruction are strongly correlated with the state residuals, contribution information, and evidence data, thus avoiding the problem of the visualization results being disconnected from the original evidence during subsequent review.

[0250] Preferably, the system can also generate integrity verification values ​​for archived records.

[0251] The integrity check value can be obtained by using a hash algorithm known in the art to calculate a summary of the key fields, evidence index, and generation time in the archived record.

[0252] This invention does not improve the internal logic of the hash algorithm, therefore no mathematical derivation of the hash algorithm is performed.

[0253] Integrity check values ​​can be used to determine whether archived records have been tampered with when reviewing or verifying archived data later, thereby improving the credibility of archived data.

[0254] Through the above processing, S5 resolved the problem of the disconnect between accident reconstruction results and archived evidence.

[0255] The same state change information can be dynamically displayed in the digital twin scenario, and can also be associated with factual state branches, comparison state branches, state residuals and contribution information, thereby realizing the visualization of the accident process, the interpretability of the attribution results, the traceability of archived data and the replayability of the review process.

[0256] Example 3:

[0257] In existing digital twin accident playback technologies, accident data is usually displayed in chronological order, or a certain state is displayed based on data source labels, indicating whether it comes from sensors, videos, manual records, or emergency systems. In sudden accident scenarios, the same state change is often caused by multiple factors.

[0258] For example, the reduction in the area of ​​danger may be related to both the stabilization of the accident itself and emergency response actions such as fire extinguishing, ventilation, and containment; the expansion of the danger boundary may include both on-site observation results and model completion results; the marking of a passage as impassable may come from both access control logs and manual corrections.

[0259] Relying solely on source labels or the highest matching degree can easily lead to misinterpreting the effectiveness of emergency response as the natural evolution of an accident, or misinterpreting model completion results as actual on-site observations.

[0260] By designing factual state branches and control state branches, mixed-source state changes have a comparable attribution basis.

[0261] The factual state branch retains the state change results under the actual data conditions, while the contrast state branch masks a certain type of influence source under the same object, same starting point, same time window, and same state field.

[0262] The state difference between the two is used to characterize the impact of the masked source on changes in the factual state.

[0263] This design differs from the traditional single-line replay approach that only displays the final state, expanding accident debriefing from displaying results to calculable source contribution analysis.

[0264] In practice, the system extracts state changes with multiple source associations from the source association results and confirms that the relevant data meet the association conditions in terms of time, object, or space.

[0265] The system then determines the state before the change, the time of the change, the duration of the change, and the time confidence interval, and uses this information to determine the conditions for branch calculation.

[0266] The system generates a factual state branch while retaining the actual valid data, and then, as needed, masks emergency response data, model completion data, or manually corrected data to generate a corresponding control state branch.

[0267] The system calculates the residual between the factual state and the control state based on the state field type, and combines the residual with the confidence level of the source data to obtain the contribution amount and contribution ratio.

[0268] In the above mechanism, object mapping confidence, time confidence interval, source association threshold, state residual, and contribution ratio are key parameters.

[0269] Object mapping confidence is used to reduce the impact of incorrect data attribution on the identification of state changes.

[0270] Time confidence intervals are used to prevent relevant data from being excluded or irrelevant data from being included due to timestamp offsets.

[0271] The source association threshold is used to filter mixed source state changes that require branch attribution.

[0272] State residuals are used to quantify the difference between the actual state and the control state.

[0273] Contribution ratios are used to transform the impact of different sources on the same state change into archived data that is visible, searchable, and verifiable.

[0274] This invention selects object identifier, spatial location, object type, field consistency, and device binding relationship as the main basis for object mapping confidence because these parameters reflect the correspondence between the original event data and the twin object from the identifier layer, spatial layer, semantic layer, field layer, and device relationship layer, respectively.

[0275] When using only object identifiers, mapping is prone to failure if the device is renamed, the code is missing, or the manual record is not standardized; when using only spatial distance, confusion is likely to occur if multiple similar devices exist in the same area.

[0276] By comprehensively judging multiple factors, the system can maintain high mapping robustness even when some fields are missing, reducing misjudgments of state changes caused by object mismatch.

[0277] This invention selects data source synchronization status, network latency, acquisition frequency, and transmission latency as the main criteria for time confidence intervals because sudden accident data not only has static time offsets during actual access, but also dynamic transmission delays and uncertainties caused by low-frequency sampling.

[0278] The system expresses time uncertainty as a confidence interval, rather than simply correcting it to a single time point. This allows it to retain potentially relevant data during branch calculations while preventing obviously irrelevant data from entering the branch window.

[0279] This invention selects different residual indices based on the type of the status field.

[0280] For continuous numerical states, using normalized numerical differences or relative rates of change can solve the problem that data from different units cannot be directly compared.

[0281] For dangerous areas, the area, degree of overlap, and boundary offset can simultaneously reflect changes in area size and spatial location.

[0282] For personnel trajectory status, the trajectory distance difference or path deviation can reflect the impact of emergency instructions, regional lockdowns, or manual corrections on personnel movement paths.

[0283] For traffic topology status, the traffic cost difference can reflect the impact of lockdowns, congestion, and detours on traffic capacity.

[0284] The above indicators all correspond to changes in technical data during digital twin incident reconstruction, rather than being a formalized packaging of purely commercial rules or administrative procedures.

[0285] Model completion can be achieved using interpolation models, rule-based models, physical diffusion models, or artificial intelligence models.

[0286] For sensor data missing for a short period of time, the system can perform time proximity interpolation based on valid observations before and after the missing time point.

[0287] This method is suitable for short-term data such as temperature, pressure, and concentration with relatively smooth trends. The maximum interpolation window is preferably 30 to 300 seconds.

[0288] If the missing time exceeds the maximum interpolation window, the system will not use simple interpolation to avoid distortion of the completion results.

[0289] For spatial completion of hazardous gas or smoke concentrations, the system can use a weighted completion method based on nearby observation points.

[0290] The closer the observation point is to the location to be completed, the greater its influence on the completion result; the farther the observation point is, the smaller its influence on the completion result. This method is suitable for scenarios where the sensor spatial distribution is uneven but the local spatial continuity is strong.

[0291] By using spatial completion, the system can form more continuous boundaries of dangerous areas when some sensors are missing, thereby improving the completeness of accident visualization and reconstruction.

[0292] When the present invention uses an artificial intelligence model for video recognition, trajectory prediction or state completion, the model may include an input preprocessing layer, a spatiotemporal feature extraction layer, an object state encoding layer, a state prediction layer and a confidence output layer.

[0293] The input preprocessing layer is used to perform time alignment, normalization, missing value imputation, and outlier removal on video frames, sensor sequences, personnel location sequences, and equipment logs.

[0294] The spatiotemporal feature extraction layer is used to extract image region features, sensor time series features, and spatial adjacency features.

[0295] The object state encoding layer is used to map multi-source features into twin object state vectors.

[0296] The state prediction layer is used to output state estimates for missing time points or missing regions.

[0297] The confidence output layer is used to output the model completion confidence score.

[0298] The input data for artificial intelligence models can include anonymized video recognition features, sensor value sequences, equipment status sequences, anonymized personnel location points, and spatial topological relationships.

[0299] The model output data may include complete state values, predicted boundaries of dangerous areas, predicted points of personnel trajectories, or confidence levels of state changes.

[0300] Training data can consist of historical accident drill data, legally authorized public emergency datasets, user-anonymized business data, and simulation-generated data.

[0301] During training, some data in a complete observation segment can be masked, and the true state before masking can be used as a supervision label, enabling the model to learn to recover the object state under missing data conditions.

[0302] Model training can employ at least one of the following: mean squared error loss, region intersection-union ratio loss, and confidence calibration loss.

[0303] Learning rate Preferred to Batch size The preferred number of training rounds is 16 to 128. The optimal value is 20 to 200, and the optimizer can be Adam or an equivalent gradient optimizer.

[0304] It should be noted that the artificial intelligence model is not an essential technical feature of this invention.

[0305] The core of this invention lies in the object-oriented organization of multi-source accident data, the identification of mixed-source state changes, the construction of factual state branches and control state branches, the calculation of state residuals, and the visualization and archiving association.

[0306] For implementation methods that only involve system architecture reconstruction, data interaction paths, timing scheduling, and multi-terminal collaboration, artificial intelligence models may not be introduced.

[0307] For hash digests, conventional communication protocol encapsulation, common database indexes, and standard encrypted transmissions that are well-known in the field, their names and technical functions should be objectively recorded without any modifications to their internal algorithms, and no mathematical derivations without technical basis should be made.

[0308] The input data and training datasets used are all from compliant sources. The data are taken from legally authorized public datasets, accident drill data, simulation-generated data, and business data that has been anonymized by users. No personal sensitive information or biometric data was collected in violation of regulations.

[0309] Personal information involved in the data has been anonymized, and unauthorized access has been restricted in system access control and archive management.

[0310] Through the above design, the present invention can clarify the role of each type of source in the formation process of the state change when the change of accident state is related to multiple sources, and bind the attribution results with the accident visualization reconstruction and emergency archives. This mechanism can reduce source misjudgment in single-line playback, improve the accuracy of emergency response effect evaluation, and make the accident archiving conclusions have a repeatable and verifiable data foundation.

[0311] Example 4:

[0312] like Figure 2 As shown, the digital twin-based emergency visualization and archiving system is used to perform the methods described in any of the above embodiments.

[0313] The system can be deployed in emergency management platforms, digital twin platforms, accident review platforms, or park integrated management platforms, or it can be implemented collaboratively by cloud servers, edge computing nodes, and local command terminals.

[0314] The system includes a data association module, a state change determination module, a source identification module, a branch attribution module, and a visualization archiving module.

[0315] The data association module is used to acquire multi-source accident data related to the scene of the sudden accident, and to perform object association and time sequence organization on the multi-source accident data based on digital twin objects to obtain an accident event sequence oriented towards the twin objects.

[0316] The data association module may include a data access unit, a field standardization unit, an object mapping unit, and a time correction unit.

[0317] The data access unit is used to receive accident data from video surveillance systems, sensor systems, equipment operation systems, access control systems, personnel positioning systems, drone platforms, and emergency command platforms.

[0318] Field normalization units are used to transform data from different sources into a unified event data structure.

[0319] The object mapping unit is used to determine the target twin object based on object identifier, spatial location, object type, field consistency, and device binding relationship.

[0320] The time correction unit is used to arrange events onto a unified incident timeline based on data source time offset, network latency, and acquisition frequency.

[0321] The state change determination module is used to determine the state change information of the twin object during the accident process based on the accident event sequence.

[0322] The state change determination module may include a state extraction unit, a state difference judgment unit, and a state change encapsulation unit.

[0323] The state extraction unit is used to extract the object state values ​​of the same twin object at different time locations from the sequence of accident events.

[0324] The status difference judgment unit is used to determine whether changes in numerical values, levels, regions, trajectories, or topological accessibility have reached the corresponding thresholds based on the field type.

[0325] The state change encapsulation unit is used to combine the state change field, the state before the change, the state after the change, the time of the change, the duration of the change, the source of the associated data, and the confidence level of the object mapping into state change information.

[0326] The source identification module is used to identify the source association of state change information, determine the association between state change information and at least one source of influence, and trigger branch attribution processing when the same state change information is associated with at least two sources of influence.

[0327] The source identification module may include an accident evolution matching unit, an emergency response matching unit, a model completion identification unit, and a manual correction identification unit.

[0328] The accident evolution matching unit is used to match state change information with the accident evolution rule base.

[0329] The emergency response matching unit is used to match status change information with the emergency response action rule base.

[0330] The model completion recognition unit is used to determine whether the state change is caused by missing data completion, adjacent data inference, or simulation model generation.

[0331] The manual correction identification unit is used to determine whether the state change was generated by manual reporting, manual review, or manual overwriting.

[0332] The branch attribution module is used to construct a fact state branch and at least one control state branch based on state change information, calculate the state residual between the fact state branch and the control state branch, and determine the contribution information of at least two types of influence sources to the state change information based on the state residual.

[0333] The branch attribution module may include a branch window determination unit, a fact branch generation unit, a comparison branch generation unit, a state residual calculation unit, and a contribution normalization unit.

[0334] The branch window determination unit is used to determine the branch calculation window based on the time of change, the duration of change, and the time confidence interval.

[0335] The fact branch generation unit is used to retain actual valid data to generate the state after fact changes.

[0336] The control branch generation unit is used to generate the control state after the change by shielding emergency response, model completion or manual correction.

[0337] The State Residual Calculation Unit is used to calculate the residual between the fact state and the control state according to the state field type.

[0338] The contribution normalization unit is used to determine the contribution ratio and dominant source identifier based on the confidence level of the residual and source data.

[0339] The visualization archiving module is used to generate accident visualization reconstruction results based on state change information and contribution information, and to associate the accident visualization reconstruction results, state residuals and contribution information to form emergency archive data.

[0340] The visualization archiving module can include a timeline rendering unit, a twin scene update unit, a source contribution display unit, a branch comparison display unit, and an archive generation unit.

[0341] The timeline rendering unit is used to display accident events according to a unified accident timeline.

[0342] The twin scene update unit is used to drive the state of the twin object to change over time.

[0343] The source contribution display unit is used to display the contribution ratio of accident ontology evolution, emergency response, model completion and manual correction.

[0344] The branch comparison display unit is used to show the differences between the fact state branch and the control state branch.

[0345] The archive generation unit is used to generate emergency archive data that includes an index of original evidence, state change information, state residuals, contribution information, and an index of visualization results.

[0346] In a preferred embodiment, the system further includes an access control module and a review control module.

[0347] The access control module is used to control the access permissions of different users to raw data, branch calculation results, manual correction records and archived results, so as to prevent unauthorized users from modifying key incident evidence.

[0348] The review management module is used to record the reviewer, review time, review comments and review results, and to save the review records with the corresponding archive number.

[0349] For manually corrected data, the system preferably records the state before correction, the state after correction, the reason for correction, the person who made the correction, and supporting evidence to improve the traceability of the source of manual correction.

[0350] In another preferred embodiment, the system may employ an edge-cloud collaborative architecture.

[0351] Edge devices are deployed at accident sites or locally within the park to quickly access logs from sensors, cameras, access control devices, and emergency equipment, and to perform initial field standardization, time correction, and object mapping.

[0352] The cloud is used for performing state change identification, source association identification, branch attribution calculation, visualization reconstruction, and emergency archiving.

[0353] Standardized event data can be transmitted between the edge and the cloud using secure communication protocols known in the field.

[0354] This invention does not improve the encapsulation method of the communication protocol; therefore, it only needs to ensure the integrity, timing, and security of data transmission.

[0355] The above-described system and the above-described method embodiments belong to the same inventive concept and can achieve the same or corresponding technical effects.

[0356] The data association module organizes multi-source accident data into twin objects and accident timelines. The state change determination module and source identification module locate mixed source state changes. The branch attribution module quantifies the contribution of different impact sources in the form of state residuals. The visualization archiving module associates the attribution results with accident reconstruction and emergency archives, ultimately improving the interpretability of sudden accident reconstruction, the accuracy of emergency response assessment, and the traceability of accident archiving.

[0357] The above embodiments illustrate the implementation of the present invention, but the present invention is not limited to the above embodiments.

[0358] For those skilled in the art, without departing from the technical concept of this invention, adaptive adjustments can be made to the multi-source accident data type, object mapping parameters, source matching rules, residual measurement method, branch masking method, visualization expression form and archived data structure.

[0359] For example, in the case of chemical leak accidents, the key status fields can be gas concentration, wind direction, valve status, and sealed-off area; in the case of rail transit accidents, the key status fields can be train operation status, platform personnel density, passageway access cost, and emergency broadcast instructions; and in the case of public building fire accidents, the key status fields can be smoke concentration, temperature, evacuation route status, fire equipment operation, and personnel evacuation trajectory.

[0360] The above adjustments are all specific application variations based on the content disclosed in this invention, and do not affect the core idea of ​​this invention to achieve technical effects through digital twin object organization, multi-source state change identification, fact and contrast branch construction, state residual attribution, and visual archiving association.

Claims

1. A method for visual reconstruction and emergency archiving of sudden accidents based on digital twins, characterized in that, include: S1. Acquire multi-source accident data related to the scene of the sudden accident, and perform object association and time sequence organization on the multi-source accident data based on the digital twin object to obtain an accident event sequence oriented towards the twin object; S2. Determine the state change information of the twin object during the accident process based on the accident event sequence, and perform source association identification on the state change information. The source association identification is used to determine the association relationship between the state change information and at least one source of influence. The source of influence includes at least one of accident ontology evolution, emergency response, model completion and manual correction. S3. When the same state change information is associated with at least two types of influence sources, a fact state branch and at least one control state branch are constructed based on the state change information. The fact state branch is used to characterize the state change result under actual data conditions, and the control state branch is generated by shielding at least one of the at least two types of influence sources. S4. Calculate the state residual between the fact state branch and the control state branch, and determine the contribution information of the at least two types of influence sources to the state change information based on the state residual; S5. Generate an accident visualization reconstruction result based on the state change information and the contribution information, and associate the accident visualization reconstruction result, the state residual, and the contribution information to form emergency archive data.

2. The method for visual reconstruction and emergency archiving of sudden accidents based on digital twins according to claim 1, characterized in that, The multi-source accident data in S1 includes digital twin basic data, accident process data, and emergency response data; The digital twin basic data includes at least one of the following: spatial structure data of the accident site, equipment object data, sensor deployment data, traffic topology data, and initial operating status data; The accident process data includes at least one of the following: video surveillance data, sensor data, access control status data, equipment operation data, personnel positioning data, drone data, and manually reported data during the accident. The emergency response data includes at least one of the following: emergency command instructions, equipment operation logs, rescue personnel operation records, area lockdown records, and emergency plan execution records; Furthermore, the multi-source accident data is standardized by performing field standardization processing based on the object identifier, spatial location, collection time, and data source in the multi-source accident data.

3. The method for visual reconstruction and emergency archiving of sudden accidents based on digital twins according to claim 1, characterized in that, In step S1, object association and temporal organization are performed on the multi-source accident data based on digital twin objects to obtain an accident event sequence oriented towards the twin objects, including: A set of twin objects is constructed based on the basic data of the digital twin. The set of twin objects includes multiple twin objects with object identifiers, spatial locations, object types, and initial states. The accident process data and emergency response data are processed into event data to obtain raw event data containing event time, event object, event location, event type and event source; Based on the set of twin objects, the original event data is mapped to determine the target twin object corresponding to the original event data; The original event data mapped to the same target twin object are arranged according to the event time to generate an accident event sequence oriented towards the twin object.

4. The method for visual reconstruction and emergency archiving of sudden accidents based on digital twins according to claim 3, characterized in that, Before arranging the original event data mapped to the same target twin object according to event time, the following is also included: Obtain the data source timestamp corresponding to the original event data; The data source timestamp is corrected according to the preset base time and the data source time offset to obtain the corrected event time; When the timestamp of the data source is uncertain, the time confidence interval is determined based on at least one of the data source synchronization status, network latency, acquisition frequency and transmission latency. Based on the corrected event time and the time confidence interval, the original event data are arranged into a unified accident timeline; The time confidence interval is used to limit the branch calculation time range of the fact state branch and the control state branch.

5. The method for visual reconstruction and emergency archiving of sudden accidents based on digital twins according to claim 3, characterized in that, The step of mapping the original event data to the twin object set to determine the target twin object corresponding to the original event data includes: Candidate twin objects are determined from the set of twin objects based on at least one of the object identifier, spatial location, event type, and data source in the original event data; The object mapping confidence is determined based on at least one of the following: spatial distance between the candidate twin object and the original event data, consistency of object type, consistency of data fields, and device binding relationship; The target twin object is determined from the candidate twin objects based on the object mapping confidence. When the confidence level of the object mapping is lower than the preset mapping threshold, the corresponding original event data is marked as object mapping data to be confirmed, and the object mapping data to be confirmed is used as masked data or downweighted data when constructing the control state branch.

6. The method for visual reconstruction and emergency archiving of sudden accidents based on digital twins according to claim 1, characterized in that, In step S2, determining the state change information of the twin object during the accident process based on the accident event sequence includes: Extract the object state values ​​of the same twin object at different time locations from the sequence of accident events; The state change field is determined based on the differences in object state values ​​at adjacent time locations; Based on the state change field, determine the state before the change, the state after the change, the time of the change, and the duration of the change; The state change information is generated by combining at least one of the state change fields, the state before the change, the state after the change, the time of the change, the duration of the change, the source of the associated data, and the confidence of the object mapping. The status change field includes at least one of the following: spatial status field, equipment operation status field, environmental status field, access status field, personnel status field, and hazardous area status field.

7. The method for visual reconstruction and emergency archiving of sudden accidents based on digital twins according to claim 1, characterized in that, The source association identification of the state change information in step S2 includes: The state change information is matched with the accident evolution rule base to obtain the accident evolution matching degree; The state change information is matched with the emergency response action rule base to obtain the emergency response matching degree; The model completion matching degree is obtained based on whether the state change information is filled in by missing data, inferred from adjacent data, or generated by a simulation model. The manual correction matching degree is obtained based on whether the state change information is manually reported, manually reviewed, or manually overwritten. Based on the accident evolution matching degree, emergency response matching degree, model completion matching degree, and manual correction matching degree, the source association results corresponding to the state change information are generated. When at least two of the accident evolution matching degree, emergency response matching degree, model completion matching degree, and manual correction matching degree reach the corresponding threshold, and the corresponding data are related in time, object, or space, it is determined that the same state change information is associated with at least two types of influence sources.

8. A method for visual reconstruction and emergency archiving of sudden accidents based on digital twins according to claim 6, characterized in that, The S3 step, which constructs a factual state branch and at least one control state branch based on the state change information, includes: The state before the change in the state change information is used as the common starting point for the branching of the factual state branch and the control state branch, and the branch calculation window is determined by the time of the change and the duration of the change. Based on the effective data from the actual acquired accident process data, emergency response data, model completion data, and manually corrected data, a fact state branch is generated to characterize the state after factual changes under actual data conditions. The control state branch includes at least one of the following: When the state change information is simultaneously associated with the accident ontology evolution and emergency response, emergency response data that is spatiotemporally related to the state change information is shielded, and a first comparison state branch is generated based on the pre-change state, accident process data, and accident evolution rule base. When the state change information is associated with both accident ontology evolution and model completion, the model completion data related to the state change information is masked, and a second control state branch is generated based on the state before the change, the actual observed accident process data, and the accident evolution rule base. When the status change information is associated with both emergency response and manual correction, the manual correction data related to the status change information is masked, and a third comparison status branch is generated based on the pre-change status, accident process data, and emergency response action rule base.

9. A method for visual reconstruction and emergency archiving of sudden accidents based on digital twins according to claim 8, characterized in that, In step S4, the state residual between the factual state branch and the control state branch is calculated, and the contribution information of the at least two types of influence sources to the state change information is determined based on the state residual, including: When a first control state branch exists, calculate the first state residual between the state after the fact change and the first predicted state corresponding to the first control state branch, and determine the emergency response contribution based on the first state residual. When a second control state branch exists, calculate the second state residual between the state after the fact change and the second predicted state corresponding to the second control state branch, and determine the model completion contribution based on the second state residual. When a third control state branch exists, calculate the third state residual between the state after the fact change and the third predicted state corresponding to the third control state branch, and determine the amount of artificial correction based on the third state residual. The contribution of accident evolution is determined based on the total state change corresponding to the state change information, as well as at least one of the determined emergency response contribution, model completion contribution, and manual correction influence. Normalize at least two of the following: the contribution of accident evolution, the contribution of emergency response, the contribution of model completion, and the influence of manual correction, to obtain the corresponding contribution information, and generate a dominant source identifier based on the contribution information. The state residual includes at least one of the following: numerical difference, relative rate of change, standardized distance, state level difference, state transition cost, regional coverage area difference, boundary offset, regional intersection-to-union ratio difference, trajectory distance difference, and topology access cost difference.

10. A system for visual reconstruction and emergency archiving of sudden accidents based on digital twins, characterized in that, include: The data association module is used to acquire multi-source accident data related to the scene of the sudden accident, and to perform object association and time-series organization on the multi-source accident data based on the digital twin object to obtain an accident event sequence oriented towards the twin object; The state change determination module is used to determine the state change information of the twin object during the accident process based on the accident event sequence. The source identification module is used to identify the source association of the state change information, determine the association between the state change information and at least one source of influence, and trigger branch attribution processing when the same state change information is associated with at least two sources of influence, wherein the sources of influence include at least one of accident ontology evolution, emergency response, model completion and manual correction; The branch attribution module is used to construct a fact state branch and at least one control state branch based on the state change information, calculate the state residual between the fact state branch and the control state branch, and determine the contribution information of the at least two types of influence sources to the state change information based on the state residual, wherein the control state branch is generated by masking at least one of the at least two types of influence sources. The visualization archiving module is used to generate an accident visualization reconstruction result based on the state change information and the contribution information, and to associate the accident visualization reconstruction result, the state residual, and the contribution information to form emergency archive data.

Images