Basic emergency evidence closed-loop generation method and system based on instruction action mapping

Abstract

The invention discloses a basic-level emergency evidence closed-loop generation method and system based on instruction action mapping, and relates to the field of basic-level emergency informatization management, and the method comprises the steps: carrying out the standardized structured convergence of a command side instruction, on-site action monitoring and evidence data; performing semantic analysis and task modeling on the instruction, generating a feature vector containing an instruction element and a weight, and endowing the feature vector with a unique number to realize instruction materialization; performing behavior identification and space-time standardization on the action monitoring data, generating an action feature vector, endowing the action feature vector with a unique number, establishing an index relationship between actions and evidences, and realizing traceable association of the evidences; constructing an instruction-action mapping matrix and calculating an association score, and determining a target action set corresponding to the instruction; and performing fusion evaluation on the multi-source evidence associated with the target action, and generating an instruction closed-loop index and a closed-loop record. Therefore, quantitative closed loop and verifiable trace leaving of instruction execution are realized.

Description

Technical Field

[0001] This application relates to the field of grassroots emergency information management technology, and in particular to a method and system for generating closed-loop grassroots emergency evidence based on command action mapping. Background Technology

[0002] Grassroots emergency management faces highly uncertain scenarios such as disasters and public health emergencies. It usually requires the rapid formation of consistent actions along the "monitoring and early warning - command and dispatch - on-site execution - feedback and review" chain. However, grassroots units generally face objective constraints such as limited resources, insufficient professional personnel, and weak information infrastructure. As a result, any delay in any link of the response chain will amplify the loss of rescue efficiency and management risks.

[0003] In the current construction of grassroots emergency information systems, the acquisition and aggregation of multi-source data is often still fragmented: on the one hand, the sources of on-site data are diverse (command text / voice / video, positioning, sensors, monitoring and drones, etc.); on the other hand, the data interface standards across devices and departments are not unified and the spatiotemporal benchmarks are difficult to align, which can easily lead to the information chain being broken at the "last mile," making it difficult for the command side to make judgments and dispatch based on unified facts.

[0004] At the level of information sharing and process supervision, the common practice at the grassroots level still tends to be "manual confirmation + telephone dispatch + manual registration". Data barriers and scattered records make it difficult to grasp the progress of execution in a timely manner after the instructions are issued, and the on-site feedback also lacks objective evidence that can be verified, making it difficult to quantify the execution process and results of instructions and to trace responsibility. Summary of the Invention

[0005] This application provides a method, system, storage medium, computer program product, and electronic device for generating closed-loop emergency evidence at the grassroots level based on command action mapping. This is intended to at least solve the problem in current related technologies where the command side has difficulty obtaining consistent and verifiable factual evidence, resulting in the inability to quantify the progress of command execution and the lack of objective support for retrospective analysis.

[0006] In a first aspect, embodiments of this application provide a method for generating closed-loop emergency evidence at the grassroots level based on command-action mapping. The method includes: acquiring multi-source emergency data corresponding to an emergency event, and performing standardization processing on the multi-source emergency data to obtain a corresponding structured data set; the multi-source emergency data includes command data from the command side, action monitoring data from the field side, and evidence data related to field execution; performing semantic parsing and task modeling on the command data in the structured data set, extracting command elements related to emergency response, generating a unique command number for each command, and determining a command weight for each command to characterize the urgency and importance of the command, thereby generating a command feature vector containing the command elements and the command weight to obtain a command entity set; performing behavior recognition and standardization processing on the action monitoring data in the structured data set, extracting action type and action spatiotemporal attributes, and generating a unique action number for each action. The system generates an action feature vector containing the action type and the spatiotemporal attributes of the action to obtain an action entity set, and establishes an index relationship between each action entity and the corresponding evidence data. A mapping matrix is ​​constructed based on the instruction entity set and the action entity set. For each pair of instruction entities and action entities in the mapping matrix, a corresponding association score is calculated, and a mapping determination is made based on the association score to determine the target action set corresponding to each instruction, and an instruction-action mapping relationship is established. For each instruction, based on the target action set corresponding to the instruction and the evidence data located through the index relationship with the target action set, a multi-source evidence fusion evaluation is performed on the located evidence data to determine an instruction closed-loop index used to characterize the completeness and sufficiency of the instruction's execution, and a closed-loop record is generated. The closed-loop record includes the instruction number, the action number in the target action set, the corresponding evidence data index, and the instruction closed-loop index.

[0007] Secondly, embodiments of this application provide a closed-loop generation system for grassroots emergency evidence based on command-action mapping. The system includes: a multi-source data structuring unit, used to acquire multi-source emergency data corresponding to an emergency event and perform standardization processing on the multi-source emergency data to obtain a corresponding structured data set; the multi-source emergency data includes command data from the command side, action monitoring data from the field side, and evidence data related to field execution; a command entity generation unit, used to perform semantic parsing and task modeling on the command data in the structured data set, extract command elements related to emergency response, generate a unique command number for each command, and determine a command weight for each command to characterize the urgency and importance of the command, thereby generating a command feature vector containing the command elements and the command weight to obtain a command entity set; and an action entity and evidence index generation unit, used to perform behavior recognition and standardization processing on the action monitoring data in the structured data set, extract action types and action spatiotemporal attributes, and generate an index for each action. A unique action number is used to generate an action feature vector containing the action type and the spatiotemporal attributes of the action, thereby obtaining a set of action entities and establishing an index relationship between each action entity and the corresponding evidence data; a mapping matrix construction and judgment unit is used to construct a mapping matrix based on the set of instruction entities and the set of action entities, calculate the corresponding association score for each pair of instruction entities and action entities in the mapping matrix, and perform mapping judgment based on the association score to determine the target action set corresponding to each instruction and establish an instruction-action mapping relationship; a closed-loop index evaluation and recording unit is used to perform multi-source evidence fusion evaluation on the located evidence data based on the target action set corresponding to the instruction and the evidence data located through the index relationship with the target action set for each instruction, to determine the instruction closed-loop index used to characterize the execution completeness and evidence sufficiency of the instruction, and generate a closed-loop record; the closed-loop record includes the instruction number, the action number in the target action set, the corresponding evidence data index, and the instruction closed-loop index.

[0008] Thirdly, an electronic device is provided, comprising: at least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the steps of the grassroots emergency evidence closed-loop generation method based on instruction-action mapping according to any embodiment of the present application.

[0009] Fourthly, embodiments of this application provide a storage medium storing a computer program thereon, characterized in that, when the program is executed by a processor, it implements the steps of the grassroots emergency evidence closed-loop generation method based on instruction action mapping of any embodiment of this application.

[0010] Fifthly, embodiments of this application provide a computer program product, including a computer program / instruction, which, when executed by a processor, implements the steps of the grassroots emergency evidence closed-loop generation method based on instruction-action mapping according to any embodiment of this application.

[0011] The method and system for generating closed-loop emergency evidence at the grassroots level based on command and action mapping provided in this application can achieve at least the following technical effects: (1) Elevate the “instruction-action-evidence” chain from loose records to a computable, consistent, and related object. By standardizing and structuring multi-source emergency data, and abstracting command-side instructions into instruction entities with unique numbers and weights, and on-site execution into action entities with type and spatiotemporal attributes, the previously difficult-to-align instruction descriptions, on-site behaviors, and evidence materials are given a unified semantic and spatiotemporal anchor. On this basis, by using a mapping matrix to make a global association determination between instruction entities and action entities, the set of target actions corresponding to an instruction can be made explicit and definitive, avoiding the vague judgment of “who is doing it, what they did, and whether it belongs to the instruction” based solely on human experience, thereby significantly improving the visibility of the execution chain and the consistency of scheduling on the command side.

[0012] (2) Using action index evidence and conducting multi-source fusion evaluation to form quantifiable instruction closed-loop measurement and auditable records. By establishing an index relationship between action entities and evidence data, each action is naturally bound to verifiable evidence; then, around the target action set corresponding to the instruction, a fusion evaluation is performed on the located multi-source evidence to generate an instruction closed-loop index that simultaneously represents the completeness of execution and the sufficiency of evidence. Thus, the system can not only provide a unified quantitative result on whether it is completed, to what extent it is completed, and whether the evidence is sufficient, but also solidify the instruction number, action number, evidence index, and closed-loop index into a closed-loop record, realizing traceable, verifiable, and reviewable output from process to result, thereby upgrading execution supervision from post-event spot checks to process-oriented management based on objective evidence and indicators.

[0013] This technical solution utilizes command-action mapping to establish a calculable, deterministic correspondence between command intentions and on-site behaviors. Furthermore, it couples this correspondence with multi-source evidence fusion and evaluation to automatically generate a closed-loop evidence record with quantified indices. This transforms command execution from a weakly correlated "issue-wait-feedback" model into a mappable, verifiable, measurable, and traceable closed-loop generation model, thereby stably improving coordination consistency and full-process supervision capabilities in highly uncertain scenarios. Attached Figure Description

[0014] To more clearly illustrate the technical solutions in the embodiments of this application 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 some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0015] Figure 1 A flowchart illustrating an example of a method for generating closed-loop grassroots emergency evidence based on command action mapping according to an embodiment of this application is shown. Figure 2 This paper illustrates an example of the operational mechanism of a grassroots emergency evidence closed-loop generation method based on command action mapping according to an embodiment of this application; Figure 3 The graph shows a comparison of instruction-action mapping accuracy under different concurrent instruction scales; Figure 4 The following is a heatmap showing the closed-loop exponential distribution of the method in the embodiments of this application under different instruction complexity and resource availability conditions; Figure 5 A structural block diagram of an example of a grassroots emergency evidence closed-loop generation system based on instruction action mapping according to an embodiment of this application is shown. Detailed Implementation

[0016] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0017] To address the pain points of grassroots emergency management, some information-based auxiliary technologies have emerged, but these mainly focus on macro-situational awareness, communication channel construction, or resource computing in single links, making it difficult to fully adapt to the needs of closed-loop management across the entire grassroots chain.

[0018] In the area of ​​intelligent emergency command, some current technologies propose solutions based on scenario twins. These solutions utilize sensor networks to collect environmental and status information, construct three-dimensional virtual scenarios through augmented reality and digital twin technologies, and combine graph neural networks and other algorithms for situational analysis and strategy generation. While these technologies can achieve real-time rendering and risk prediction of complex scenarios, improving the intelligence level of decision-making, they are typically designed for city-level command centers, requiring extremely high computing resources and modeling processes, resulting in high deployment costs that are difficult for grassroots units to afford. More importantly, these solutions focus on the visualization of virtual scenarios and macro-level situational simulations, lacking a mechanism for accurately linking downlink commands with actual grassroots execution, and failing to solve the micro-level verification problem of "whether commands have been implemented."

[0019] Regarding command and dispatch systems, many current technologies employ multimedia convergence dispatch solutions, integrating video surveillance, satellite positioning, GIS geographic information, and voice conferencing to establish communication channels between the front and rear command centers. While these systems, to some extent, ensure real-time communication capabilities during rescue operations and solve the "seeing and hearing" problem, they essentially still focus on ensuring communication links and one-way information transmission. In the command execution phase, these systems still rely on manual reporting and confirmation, lacking automatic detection algorithms for command execution and failing to establish a standardized closed-loop of execution evidence based on objective data. This makes it difficult for the command level to quantitatively evaluate the quality of execution.

[0020] Furthermore, in terms of resource scheduling and support in specific scenarios, some current technologies include resource matching methods for specific areas such as chemical industrial parks, or drone communication command methods based on AI perception. The former typically calculates material needs and matches inventory based on accident type and diffusion model, while the latter uses intelligent perception systems to dynamically adjust drones to repair communication interruptions. However, these technical solutions often focus on static material allocation calculations or communication restoration at the infrastructure level, mainly addressing the questions of "what is needed" or "is the signal working?", without involving the dynamic tracking of specific law enforcement and rescue operations. Existing solutions generally lack the technical means to logically map discrete on-site actions to superior instructions, making it difficult to form a complete traceability chain of "instruction issuance - action execution - evidence feedback".

[0021] However, none of the aforementioned technologies can support the automatic establishment of a traceable mapping between instructions and actions under conditions of limited resources at the grassroots level, nor can they support an information processing solution for generating objective closed-loop evidence.

[0022] It should be understood that the above description of the relevant technologies is intended only to help the public better understand the inventive spirit and motivation of this application, and is not intended to limit this application. Furthermore, the technical solutions described in the above-mentioned relevant technologies are not prior art, and may also be undisclosed technical solutions, such as those under research or in the laboratory stage.

[0023] The technical solutions in this application, including the collection, storage, use, processing, transmission, provision, and disclosure of users' personal information, comply with relevant laws and regulations and do not violate public order and good morals.

[0024] Figure 1 A flowchart illustrating an example of a method for generating closed-loop grassroots emergency evidence based on instruction action mapping according to an embodiment of this application is shown.

[0025] Regarding the executing entity of the method in this application, it can be any controller or processor with computing or processing capabilities, such as a grassroots emergency management platform controller. By using the "command-action" mapping as a hub, multi-source on-site evidence is aggregated to the command level through action indexes, and a closed-loop index and structured closed-loop records are output through fusion evaluation, thereby forming a traceable, quantifiable, and verifiable evidence closed-loop generation mechanism. This not only achieves the structured solidification of the command execution chain but also provides a unified and verifiable benchmark for subsequent process supervision, debriefing analysis, and responsibility delineation.

[0026] In some examples, it may be integrated into an electronic device or terminal through software, hardware, or a combination of both, and the type of terminal or electronic device may be diverse.

[0027] like Figure 1 As shown, in step S110, multi-source emergency data corresponding to the emergency event is acquired, and the multi-source emergency data is standardized to obtain a corresponding structured data set; the multi-source emergency data includes command data from the command side, action monitoring data from the field side, and evidence data related to field execution.

[0028] Here, a unified data collection portal is established for the same emergency event to obtain multi-source emergency data corresponding to that event. More specifically, instruction data may include instruction text, speech-to-text transcription, instruction issuance time, issuer identification, and receiving unit identification; action monitoring data may include the location trajectory of on-site personnel / vehicles, attendance / check-in records, intercom or terminal reports, sensor status changes, equipment on / off status, and drone patrol flight records; evidence data may include images, video clips, audio, raw sensor curves, law enforcement recorder / surveillance footage, drone images and their metadata.

[0029] To ensure that data from different sources can be compared and used in subsequent calculations in a unified manner, standardized processing is performed on multi-source emergency data. On the one hand, the original fields of different protocols or interfaces are mapped to a unified set of fields (e.g., unified as "timestamp, spatial location, source type, acquisition device identifier, responsible entity identifier, content payload / media pointer", etc.), and missing fields are marked with a consistent completion strategy (e.g., set to empty and record the reason for the missing field). On the other hand, time and spatial references are aligned. For example, the local time of each terminal is uniformly converted to the same standard time zone timestamp and the estimated time deviation value is recorded. Latitude and longitude / projected coordinates are uniformly converted to a preset coordinate system and precision is marked, thereby reducing the spatiotemporal incomparability caused by cross-departmental and cross-device issues.

[0030] Through the above processing, a structured data set is obtained, which allows instructions, actions and evidence to be expressed under the same data standard. This enables subsequent operations to be directly based on structured fields for retrieval, association and scoring calculation without relying on manual interpretation and manual splicing.

[0031] In step S120, semantic parsing and task modeling are performed on the instruction data in the structured data set to extract instruction elements related to emergency response. A unique instruction number is generated for each instruction, and an instruction weight is determined for each instruction to characterize the urgency and importance of the instruction. Then, an instruction feature vector containing instruction elements and instruction weights is generated to obtain an instruction entity set.

[0032] In some implementations, the instruction text / transcribed text can be segmented and entity extracted to parse out instruction elements that include at least: the object of the action (e.g., "flooding at a certain location / suspected cases at a certain point / landslide risk at a certain road section"), the action to be taken (e.g., "lockdown, transfer, investigation, emergency repair, disinfection, traffic control"), the scope of the action (administrative division / road section / grid number / geofence), time constraints (standardized expressions of start time, end time, duration or constraints such as "immediately / as soon as possible"), the responsible entity (department / team / position / personnel), and resource constraints (number of vehicles, materials, personnel, equipment type).

[0033] To ensure that each instruction can be uniquely referenced and traced throughout the entire process, a unique instruction number can be generated for each instruction. For example, a deterministic number can be generated based on "event identifier + instruction issuance timestamp + issuer identifier + instruction content summary". The key fields used to generate the number are saved in the structured record for auditing and review.

[0034] Meanwhile, to reflect the differences in urgency and importance of different instructions in emergency response, this step determines the instruction weight for each instruction. The instruction weight can be generated using a method of "rule element normalization + optional historical statistical calibration": for example, first, multiple sub-scores are generated based on factors such as keywords, object type, scope of impact, time constraint strength, and whether personnel safety is involved in the instruction, and then these are merged into the instruction weight according to preset weights. The source fields and calculation paths of each sub-score are recorded to ensure that the weights are interpretable and verifiable.

[0035] When generating instruction entities, the "instruction elements + instruction weights" are jointly encoded into an instruction feature vector (e.g., formed by concatenating action category codes, scope codes, time constraint codes, responsible entity codes, and instruction weights), ultimately resulting in a set of instruction entities. Thus, after being element-based and vectorized, instructions can directly participate in subsequent correlation scoring and mapping determinations, and differentiated evaluation and ranking of high-priority tasks can be achieved based on instruction weights, thereby improving the interpretability and consistency of command-side supervision of task organization and execution.

[0036] In step S130, the action monitoring data in the structured data set is subjected to behavior recognition and normalization processing, action type and action spatiotemporal attributes are extracted, and a unique action number is generated for each action. Then, an action feature vector containing action type and action spatiotemporal attributes is generated to obtain a set of action entities, and an index relationship between each action entity and the corresponding evidence data is established.

[0037] In some implementations, action monitoring data can be aggregated by responsible entity or execution unit, and action segments can be identified based on temporal continuity, spatial continuity, and state event trigger points. For example, continuous trajectory segments of the same person / vehicle in a certain area and on-site check-in events can together constitute an "on-site handling action"; changes in equipment switching quantities and continuous operating intervals can constitute an "equipment start-up / shutdown / drainage operation action"; and flight paths and shooting time windows recorded by drones can constitute an "inspection and evidence collection action".

[0038] After identifying action segments, the actions are standardized and labeled. For example, each action is assigned an action type (a predefined set of action types, such as "arrival confirmation, on-site investigation, lockdown setup, personnel transfer, equipment repair, disinfection operations, traffic control, and inspection and evidence collection") and spatiotemporal attributes (start / end time, key locations, trajectory lines, coverage area, multi-point operation sequence, etc.). The accuracy or reliability source of the action's spatiotemporal attributes is recorded (e.g., GPS accuracy, base station positioning accuracy, equipment time deviation estimation, etc.). To ensure the actions are traceable throughout the entire process, a unique action number can be generated for each action, and an action feature vector can be formed based on "action type + action spatiotemporal attributes" to obtain a set of action entities.

[0039] Furthermore, to ensure "verifiable evidence" rather than "scattered evidence," an index relationship between action entities and evidence data has been established. For example, evidence data can be automatically linked to the corresponding action number based on rules such as "evidence generation time falling within the action time window," "evidence spatial location falling within the action coverage area," and "evidence source equipment / responsible entity being consistent with or verifiablely related to the action execution entity." For evidence that cannot be automatically linked, semi-automatic or manual confirmation and linking can be achieved through metadata tagging of the evidence data (such as selecting the action number on-site, QR code / tag binding, task sheet backfilling), and the confirmation method and confirmer are recorded. Therefore, on-site execution is standardized into enumerable and comparable action entities, and evidence becomes traceable supporting material for the action through the index relationship.

[0040] In step S140, a mapping matrix is ​​constructed based on the set of instruction entities and the set of action entities. For each pair of instruction entities and action entities in the mapping matrix, the corresponding association score is calculated, and a mapping determination is made based on the association score to determine the target action set corresponding to each instruction and establish the instruction-action mapping relationship.

[0041] Here, a mapping matrix is ​​constructed based on the set of instruction entities and the set of action entities, so that each pair of "instruction entity-action entity" corresponds to a computable correlation score, thereby realizing the automated attribution determination between instructions and on-site execution.

[0042] In some implementations, rows of the mapping matrix can correspond to instruction numbers, and columns can correspond to action numbers; for any pair (instruction) in the matrix... ,action When calculating association scores, multiple interpretable matching dimensions can be combined. Examples of matching dimensions are as follows:

[0043] (1) Consistency score between action type and instruction handling action (whether the action type meets the action category required by the instruction). (2) Spatial matching score (the degree of overlap between the action coverage area and the instruction range, or the distance between the key location of the action and the location specified by the instruction); (3) Time matching score (whether the action time window falls within the instruction time constraint range, and whether it meets the timeliness requirements of "immediate / deadline" type constraints); (4) Responsibility subject matching score (whether the action execution subject and the instruction responsibility subject are consistent or have an authorization / coordination relationship); (5) Evidence support score (including the number of evidences in the action index, the coverage of evidence types, and the completeness of evidence time).

[0044] The above sub-scores are merged according to preset weights to obtain the correlation score, and the sub-score details are saved in the structured record for subsequent interpretation of the correlation results.

[0045] Furthermore, when determining the mapping based on association scores, a strategy combining threshold determination and Top-K selection can be adopted. For example, candidate actions below the threshold are first filtered out, and then a preset number of actions are selected from the remaining candidates according to their association scores as the target action set. For cases where the same action may be associated with multiple instructions, mutually exclusive or weakly mutually exclusive constraints can be introduced (e.g., prioritizing assignment to the instruction with the highest association score, or allowing one action to serve multiple instructions but requiring separate references and annotations of the sharing relationship in the closed-loop record), thereby ensuring the stability and interpretability of the mapping results. Thus, through the fusion of matrix-based full-pair calculation and interpretable scoring, the correspondence between instructions and actions is explicitly and repeatably determined, avoiding mismatches or omissions caused by relying solely on manual reconciliation, and providing a clear matching boundary relationship of "which actions correspond to each instruction."

[0046] In step S150, for each instruction, based on the target action set corresponding to the instruction and the evidence data located through the index relationship with the target action set, a multi-source evidence fusion evaluation is performed on the located evidence data to determine the instruction closed-loop index used to characterize the completeness of the instruction execution and the sufficiency of evidence, and a closed-loop record is generated.

[0047] In some implementations, the "expected action coverage requirements" of the instruction can be determined first based on the instruction elements. For example, the required actions may correspond to one or more necessary action types and their minimum coverage conditions (such as requiring "on-site confirmation + handling operations + verification and evidence collection" or requiring coverage of key points within a specified area). Then, a coverage assessment is performed on the target action set to obtain the execution integrity evaluation result. Subsequently, the sufficiency and consistency of the evidence data are assessed: for example, the credibility of the evidence source (source type of the collection device / system), temporal integrity (whether the evidence time continuously covers the key stages of the action), spatial consistency (whether the evidence location is consistent with the action location), and content consistency (whether the evidence content can support the authenticity of the action type) are fused and scored. When there is mutual corroboration from multiple sources of evidence, the sufficiency evaluation is increased; when evidence is missing or there are temporal and spatial contradictions, the sufficiency evaluation is decreased. Finally, the execution integrity evaluation and the evidence sufficiency evaluation are merged to obtain the instruction closed-loop index, and the output of the index can be differentiated by combining the instruction weight (for example, the same evidence gap is reflected as a stricter closed-loop judgment on high-weight instructions), thereby ensuring that the index can reflect both the execution facts and the task importance constraints.

[0048] Closed-loop records are used to solidify the closed-loop results into auditable objects. These records can include: instruction number, action number in the target action set, corresponding evidence data index, and instruction closed-loop index. In some optional implementations, key components of the closed-loop index (such as completeness sub-scores, sufficiency sub-scores, and missing evidence types triggering gaps) can also be recorded to enhance review capabilities. Thus, the execution status of an instruction is no longer limited to "whether feedback was given / whether verbal confirmation was given," but is quantified into a comparable closed-loop index. Furthermore, each closed-loop conclusion can be traced back to specific evidence materials and related actions through the evidence index in the closed-loop record, thereby providing a locatable, interpretable, and verifiable data analysis foundation for subsequent process supervision, debriefing assessment, and accountability verification.

[0049] Regarding the implementation details of the standardization process in step S110, in some examples of the embodiments of this application, a distributed clock synchronization protocol is used to align the instruction data, action monitoring data and evidence data with a time reference, so as to generate a unified timestamp for each data record and map the multi-source emergency data to a unified time coordinate axis.

[0050] Here, after receiving multi-source emergency data, the system first performs time base alignment processing using a distributed clock synchronization protocol. By establishing a unified clock reference between the command center, the field center, and the evidence collection center, and combining link round-trip delay measurement and clock offset estimation, a unified timestamp is generated for each data record. The original time fields of data from different sources are then mapped to the same time axis according to clock offset and delay compensation rules, so that the time of instruction issuance, the time of action, and the time of evidence collection can be directly aligned on the same time series.

[0051] Then, noise filtering is performed on the sensor time-series data in the multi-source emergency data, and piecewise linear interpolation is used to repair the missing sampling points caused by communication interruption to form continuous time-series data segments; background subtraction is used to extract effective segments containing dynamic behavior from the video data in the multi-source emergency data.

[0052] Here, after completing the unified timeline mapping, the system performs differentiated preprocessing for quality defects in different data types of multi-source emergency data to ensure the input stability of subsequent behavior recognition and evidence assessment. Specifically, for sensor time-series data, noise filtering is first performed on the sampled sequence to suppress the interference of jitter, spikes, and abnormal abrupt changes on state judgment. Then, when missing sampling points caused by communication interruption or packet loss are detected, piecewise linear interpolation is used to repair the continuity of the missing intervals, and the repaired sequence is organized into continuous time-series data segments. By denoising and repairing the continuity of the time-series data, the availability of spatiotemporal trajectory and state features required for behavior recognition is improved.

[0053] For video data, the system focuses on dynamic behavior information, uses background subtraction to separate the moving foreground region from the video stream, and extracts valid segments containing dynamic behavior. At the same time, it records the start and end times and keyframe positions of the valid segments so as to establish a stable association with the action number and evidence index. By extracting dynamic segments from the video, redundant calculations and evidence noise caused by a large number of static images are reduced, thereby improving the consistency between evidence quality scoring and action credibility assessment from the source.

[0054] Furthermore, in accordance with a unified data interface standard, the multi-source emergency data that has undergone time base alignment and corresponding preprocessing is encapsulated into structured data units, and the structured data units are aggregated to form a structured data set; the structured data unit includes a unified timestamp, geographic coordinates, data source identifier, and content carrier.

[0055] Here, after completing time alignment and type preprocessing, the system performs structured encapsulation of multi-source emergency data according to a unified data interface standard to eliminate field omissions and differences in caliber caused by inconsistent interfaces across departments and devices.

[0056] Specifically, the content carrier is used to carry the corresponding instruction text, sensor segment references, or valid video segment references, and may be accompanied by data integrity tags or quality tags to support subsequent weight reduction processing; then the system aggregates the structured data units according to event identifiers and time windows to form a structured data set, enabling subsequent steps to access the data with a unified data object, improving the manageability and computability of multi-source data within the same event scope, thereby reducing the data docking cost in engineering implementation and improving the stability of the overall closed-loop generation link.

[0057] Regarding the implementation details of determining the instruction weight in step S120, in some examples of embodiments of this application, the instruction weight is quantitatively determined based on the instruction elements obtained from semantic parsing. urgency level of the instruction With priority coefficient And based on the unified timestamp in the structured dataset, calculate the first... The time delay of the instruction issuance time relative to the current time. .

[0058] Specifically, the system uses a semantic parsing module to extract and quantize the received unstructured instruction data. More specifically, the system identifies keywords (such as "fire," "collapse," and "patrol") in the instructions based on a pre-built emergency domain ontology library, and matches these keywords with the corresponding urgency level. (For example, quantifying "fire" as 0.9 and "routine patrol" as 0.2); simultaneously, the sender's rank or mission security level in the instruction metadata is read to determine the priority coefficient. To introduce the dynamic decay characteristic over time, the system uses the unified timestamps in the structured data set generated in the preceding steps to calculate the time difference between the instruction issuance time and the current system processing time, thus obtaining the time delay. Accordingly, abstract business instructions are transformed into multidimensional numerical features that can be computed by computers.

[0059] Then, an initial weight calculation model is constructed, incorporating positive gain and negative decay, based on the urgency level. Priority coefficient With time delay Calculate the first Initial weight of the instruction : Equation (1) In the formula, The positive influence coefficient is... It is a negative attenuation coefficient. This is the Sigmoid normalization function, used to map the results of linear combinations to a preset numerical range.

[0060] In the model of equation (1), the linear combination part embodies a multi-factor weighted game. Specifically, As a positive impact coefficient, it ensures that instructions with high urgency or high priority receive higher initial scores; and As a negative attenuation coefficient, it reflects the time-sensitive nature of emergency tasks, that is, with delay... As the value of an instruction increases, its value decreases linearly. (Sigmoid normalization function) This is used to nonlinearly map the output of the above linear combination to the (0, 1) interval, which not only eliminates the influence of outliers, but also ensures the numerical stability after the fusion of different dimensional features, and realizes the accurate quantification of the "absolute value" of emergency instructions.

[0061] Furthermore, considering that multiple instructions are often executed concurrently and resources are limited in grassroots emergency scenarios, Softmax normalization is performed on all initial weights in the instruction entity set to obtain a result reflecting the first... Instruction weights, relative importance of instructions : Equation (2) In the formula, This represents the total number of instructions currently pending processing in the instruction entity set. This is used to iterate over variables for instructions.

[0062] In equation (2), the exponential operation widens the gap between high-scoring and low-scoring instructions, making important tasks more prominent; at the same time, the denominator is used to calculate the current task to be processed. The instructions perform summation and normalization, converting isolated initial weights. Relative weights converted into probability distribution form This not only solves the problem of resource preemption that cannot be addressed by simply relying on absolute scores, but also provides a statistically significant benchmark for subsequent priority queue sorting, ensuring that the system can prioritize responding to the emergency command with the highest global value at the current moment under limited computing power and resources.

[0063] Regarding the implementation details of step S130, in some examples of embodiments of this application, a multi-source consistency check is performed based on video observation information, on-site reporting information and operation log information in the action monitoring data, and the target action type is determined when the multi-source consistency check meets the preset consistency constraints.

[0064] Specifically, the system implements a cross-validation mechanism based on multimodal data to address the unreliability of single-source data. In some implementations, the system can read video streams, on-site speech-transcribed text, and equipment operation logs from the action monitoring data in parallel. For example, when the video analysis algorithm detects the feature of "personnel running," while the operation log shows "sudden increase in the displacement rate of the inspection equipment," but the on-site reported text is "fixed-point duty," the system will determine a data conflict. Only when the visual features of the video, the semantic features of the text, and the behavioral features of the logs meet the logical consistency constraint within a preset time window (e.g., all three point to "patrolling" or "handling") will the system output a definite target action type. This effectively filters out noisy data caused by sensor false alarms or human recording errors, significantly improving the accuracy and robustness of behavior recognition in complex grassroots environments.

[0065] Then, the pre-set emergency action template library is called to standardize the encoding of the target action type, so as to map the non-standard expressions on the scene into a unified set of standard action types. Based on the pre-set regional dialect keyword table, a semantic mapping relationship between regional expressions and standard semantics is established to convert text descriptions containing regional expressions into standard semantics.

[0066] Specifically, addressing the pain point of inconsistent data standards at the grassroots level, the system utilizes a pre-built emergency action template library and a dialect mapping engine for standardization. In some implementations, the system first matches and encodes the determined target action type (such as "firefighting" or "water monitoring") with standard terms in the template library (such as "firefighting operations" or "flood monitoring"). In particular, considering that grassroots personnel often use regional expressions, the system loads a pre-defined list of regional dialect keywords and uses natural language processing technology to construct semantic mapping relationships, automatically converting non-standard text containing "dialectal expressions" (such as words specifically referring to landslides in a certain local dialect) into unified standard semantic descriptions. This effectively eliminates the ambiguity and regional differences in natural language expressions, ensuring that machine algorithms can accurately understand grassroots reported data from different regions and cultural backgrounds.

[0067] Next, an action feature vector containing action type components and action spatiotemporal attribute components is constructed, and a unique action number is generated for each action to form a set of action entities; the action spatiotemporal attribute components include the action's execution time, geographical coordinates, duration, and resource consumption indicators.

[0068] Specifically, the system transforms standardized action data into computer-computable mathematical representations, i.e., constructing action feature vectors and generating unique identifiers. In some implementations, the system can use one-hot encoding or embedded vector techniques to quantify action types, while normalizing the action's execution time (timestamp), geographical coordinates (latitude and longitude), duration, and resource consumption indicators (such as fuel consumption and material consumption), combining them to form a multi-dimensional action feature vector. Furthermore, the system can also use hash algorithms or auto-incrementing sequence generators to assign a unique action ID to each physical action, thereby forming a structured set of action entities. Thus, discrete business attributes are transformed into points in a unified vector space.

[0069] Furthermore, based on the spatiotemporal attribute components of the action, the evidence retrieval range corresponding to each action entity is determined, and evidence data records within the evidence retrieval range are selected from the evidence data to generate an evidence data catalog. A traceable index relationship is established between the action number and the evidence data catalog. The evidence data catalog includes at least two of the following types of index identifiers: video keyframe index identifier, positioning trajectory segment index identifier, sensor time sequence segment index identifier, and log entry index identifier.

[0070] Specifically, the system establishes a logical connection between actions and evidence based on the spatiotemporal attributes of the physical world, generating a traceable evidence data catalog. In some implementations, the system sets a time buffer window (e.g., 5 minutes before and after) centered on the execution time of the action entity, and sets a spatial retrieval radius centered on geographical coordinates to determine the evidence retrieval range. It then traverses the storage system, filtering out video keyframes, location trajectory segments, sensor time-series data, and log entries falling within this spatiotemporal range. Thus, the system does not directly copy this massive amount of raw data, but records its storage address or file pointer, establishing an index relationship between the action number and these addresses to generate an evidence data catalog containing multiple types of index identifiers. Therefore, without increasing storage redundancy, it achieves millisecond-level location tracking from "abstract actions" to "concrete evidence," providing solid data support for subsequent end-to-end review and accountability.

[0071] Regarding the implementation details of calculating the correlation score in step S140, in some examples of embodiments of this application, the analysis of the first step... The instruction feature vector of the instruction and the first instruction The action feature vectors of each action are used to extract instruction semantic elements. Action type characteristics Expected completion time of the instruction Action execution time Responsible entities Executor of the action and instruction weight Among them, the expected completion time of the instruction. With the time of action execution Determined based on a unified timestamp in a structured dataset.

[0072] Here, the system evaluates each pair of instruction entities to be evaluated in the mapping matrix (the first pair). Article ) and Action Entities (Article 1) (Number of elements), performing deep feature parsing and alignment operations. In some implementations, the system deconstructs instruction semantic elements from the instruction feature vector. (e.g., "search and rescue" vector), expected completion time Responsible entities (e.g., "First Squad") and the instruction weights calculated from the preceding steps. Simultaneously, action type features are extracted from the action feature vector. Actual execution time and the executor Furthermore, to eliminate the interference of time discrepancies between distributed devices on association determination, the system can enforce the determination based on a "uniform timestamp" in the structured data set. and This ensures a consistent comparison benchmark across the time dimension. Consequently, heterogeneous instruction and action data are unified under the same feature space and time coordinate system.

[0073] Then, the weighted association scoring model is invoked to comprehensively consider the impact of semantic similarity, temporal relevance, subject matching, and instruction weight on association, and to calculate the first... The instruction entity corresponding to the instruction is the same as the first instruction. The correlation score between the action entities corresponding to each action : Equation (3) In the formula, Indicates instruction semantic elements Action type characteristics semantic similarity; For the subject matching indicator function, when Belonging to The value is 1 if the condition is met, and 0 otherwise. These are the adjustment coefficients for the corresponding feature dimensions; It is the Sigmoid normalization function; This is a time-dependent term.

[0074] In equation (3), the system invokes a weighted correlation scoring model to perform multidimensional fusion calculations on the extracted features. This model is not a simple linear superposition, but rather adjusts the coefficients... The contributions of the four dimensions are dynamically weighted: The semantic similarity between the instruction intent and the action content was quantified (e.g., using cosine similarity calculation). As a permission verification item, via an indicator function Determining whether the executor is a liable party not only affects the score, but also logically excludes interference from unauthorized personnel; This introduces an instruction importance bias to ensure that high-weight instructions receive a higher base score in competitive matching. Finally, the system utilizes the Sigmoid function. By applying nonlinear activation to the weighted sum, the values ​​are forcibly mapped to the probability interval of (0, 1). This constructs a comprehensive evaluation system capable of assessing the causal probability of "instruction-action" from multiple perspectives, including semantics, authority, and importance, thus avoiding the limitations of single-dimensional matching.

[0075] Equation (4) Equation (5) in, For preset time scale parameters, This is the time decay coefficient; This is a time difference function used to output the action execution time. With the expected completion time of the instruction nonnegative time interval between .

[0076] As shown in formulas (4) and (5), the system employs a time correlation calculation mechanism based on exponential decay, targeting the most sensitive time element in emergency scenarios. More specifically, the system first uses formula (5) to calculate the action execution time. With the expected time of the instruction nonnegative time interval between (i.e., response delay). Then, this interval is substituted into the exponential model of formula (4), where... As a preset time scale parameter (e.g., set to 30 minutes), it is used to normalize the time difference; and The key time decay coefficient determines the rate at which the correlation decreases over time. The physical implication of this mathematical design is that the smaller the response delay, the lower the correlation coefficient. The closer the result is to 1 (strong correlation), the higher the score becomes; as the delay increases, the score drops exponentially. Therefore, it can effectively eliminate interference from actions that, although semantically matched, occur long after the instruction was issued (likely irrelevant events), accurately capturing the "immediate" causal features in emergency response.

[0077] Regarding the implementation details of establishing the instruction action mapping relationship in step S140, in some examples of the embodiments of this application, a dynamic resource spatiotemporal occupancy status table is constructed. The dynamic resource spatiotemporal occupancy status table is used to record and update the occupancy status of personnel resources, equipment resources and material resources in different time windows and the corresponding resource identifiers.

[0078] Specifically, the system initializes a dynamic resource spatiotemporal occupancy status table in an in-memory database (such as Redis) or a high-performance cache to maintain the real-time availability of all key elements in grassroots emergency scenarios. In some implementations, the system assigns a unique resource identifier to each registered personnel (such as grid workers, volunteers), equipment (such as drones, law enforcement recorders), and supplies (such as sandbags, first aid kits), and records their status along a timeline. The status table's data structure adopts a "resource ID-time window linked list" format, with each time window precisely marking the start and end times of resource occupancy. This maps the discrete and dynamically changing resource states in the physical world into a computer-readable digital twin model, providing a globally unified resource view for subsequent multi-task concurrent scheduling, effectively avoiding resource "double booking" or scheduling conflicts caused by information lag.

[0079] Then, based on instruction weight The instruction entity set is sorted in descending order to construct a priority processing queue, and for each instruction in the priority processing queue, a correlation score is applied. Select action entities that meet the preset mapping threshold conditions from the action entity set to construct a candidate action set.

[0080] Specifically, the system introduces a priority-driven greedy matching strategy to ensure that critical instructions receive priority responses. In some implementations, the system reads the instruction weights calculated in previous steps. The system sorts the current set of pending instructions in descending order to generate a priority processing queue. Then, it retrieves the highest-priority instructions from the queue sequentially, iterates through all action entities, and assigns scores based on their associations. The algorithm filters actions, retaining only those with scores higher than a preset mapping threshold (e.g., 0.7), and constructs a dedicated set of candidate actions for each instruction. This establishes a "priority tasks" scheduling principle, preventing low-value tasks from monopolizing valuable computing or physical resources. Furthermore, the initial threshold screening significantly reduces the search space for subsequent resource conflict detection, improving the algorithm's operational efficiency.

[0081] Subsequently, resource constraint checks are performed on the action entities in the candidate action set according to the priority processing queue. The resource constraint checks include querying the dynamic resource spatiotemporal occupancy status table to determine whether there are time window overlaps or conflicts in the personnel resources, equipment resources, and material resources required by the candidate action entity during the execution time period of the candidate action entity.

[0082] Specifically, the system enters the core resource constraint detection phase, aiming to verify the feasibility of "logical matching" in "physical reality." In some implementations, for each action entity in the candidate set, the system analyzes its associated executing entity and the equipment used, extracting the actual execution time period. Then, using this time period as the query condition, the dynamic resource spatiotemporal occupancy status table is retrieved to determine whether the corresponding resource identifier already has a locked record within this time window (i.e., checking for interval overlap on the time axis). Thus, by utilizing the mutual exclusion principle of time windows for collision detection, physical paradoxes such as "the same person performing different tasks at the same time" can be accurately identified, ensuring that the mapping relationship generated by the system is logically consistent and executable in the real physical world.

[0083] Furthermore, if there is no conflict in the resource constraint detection, the corresponding resource identifier's occupancy time window is locked for the candidate action entity in the dynamic resource spatiotemporal occupancy status table, and the candidate action entity is included in the target action set to establish the instruction-action mapping relationship between the instruction and the target action set.

[0084] Specifically, under the ideal condition that no conflicts are detected in resource constraint detection, the system performs atomic resource locking and mapping binding operations. In some implementations, the system immediately writes a new occupancy time window for all resource identifiers involved in the action entity in the dynamic resource spatiotemporal occupancy status table, marking their status as "locked" to prevent subsequent low-priority instructions from calling the same resource again; at the same time, the system formally establishes an association pointer between the instruction number and the target action number in the database, including it in the target action set. Thus, not only is the process from algorithmic judgment to data solidification completed, but the "locking" mechanism also achieves exclusive resource management, ensuring the integrity and exclusivity of the evidence chain corresponding to high-priority instructions.

[0085] In the event of a conflict in resource constraint detection, an alternative selection strategy is triggered to select the candidate action entity that meets the resource constraints and has the highest association score from the candidate action set, and establish an instruction-action mapping relationship between the instruction and the candidate action entity. When there is no candidate action entity that meets the resource constraints, the instruction is marked as resource blocked and suspended. When the occupancy status of the corresponding resource identifier in the dynamic resource spatiotemporal occupancy status table changes, the instruction is re-evaluated to re-determine the target action set.

[0086] Specifically, for resource conflict scenarios, the system is designed with a flexible suboptimal substitution and dynamic suspension mechanism. In some implementations, when the optimal action (highest score) encounters a resource conflict, the system does not directly discard the instruction. Instead, it automatically backtracks through the candidate action set to find a replacement action with the second-highest related score and no resource conflict for "downgrade matching." If all candidate actions conflict, the system marks the instruction as "resource blocked" and places it in a suspension queue, while simultaneously registering a listener on the status table. Once the relevant resource's occupancy status is released or changes, the listener triggers a re-evaluation process for the suspended instruction. This endows the system with strong robustness and adaptability, minimizing instruction execution "deadlock" caused by resource shortages and ensuring the continuity of the emergency response process.

[0087] In some examples of embodiments of this application, after establishing the instruction action mapping relationship, based on the instruction elements extracted from the instruction entity, a preset processing flow template library is invoked to process the instruction action. Each instruction is broken down into a sequence of instruction subtasks arranged in the expected order of processing. ,in This represents the total number of subtasks in the instruction subtask sequence.

[0088] Here, the system performs a logical mapping from abstract instructions to specific work processes. In some implementations, the system, based on semantic elements extracted from the instruction entity (such as "flood control and drainage"), calls a pre-defined handling process template library to retrieve the corresponding standard operating procedure (SOP). The system then decomposes the instruction into a sequence of instruction subtasks arranged in a pre-defined logical order. For example, the "drainage command" can be broken down into "s1: set up warning tape", "s2: connect water pump", and "s3: start pumping". This established a "standard logical benchmark" for emergency response, transforming general task instructions into measurable, discretized operational nodes.

[0089] Then, based on the spatiotemporal attributes of each action entity in the target action set, the target action set is sorted according to the action execution time to obtain the target action sequence. ,in This represents the total number of actions in the target action sequence. A sequence alignment model is constructed based on the instruction subtask sequence and the target action sequence. This model is used to align each instruction subtask with the target action sequence, while satisfying preset semantic matching and temporal order constraints. Determine its alignment position in the target action sequence. ;in, Representation and Instruction Subtasks Alignment Actions The sequence number in the target action sequence.

[0090] Here, the system constructs and logically aligns the physical action sequence at the actual execution level. In some implementations, the system extracts the execution timestamps of all action entities in the target action set, rearranges the actions in chronological order, and generates a sequence of length [length missing]. target action sequence Subsequently, the system constructs a sequence alignment model (such as one based on dynamic programming or semantic matching algorithms) to align each expected instruction subtask with the given semantic similarity (e.g., the content of "action a3" matches "subtask s1"). Find its corresponding position index in the actual action sequence For example, if the logically first subtask is actually the third one executed, then This establishes a mapping relationship between the "logical expected order" and the "physical actual order," enabling precise detection of execution anomalies such as "acting first and reporting later" or "reversing the procedure."

[0091] Then, based on the alignment position Calculate the time series consistency index : Equation (6) Equation (7) Equation (8) in, for The order deviation index of instructions. The order decay coefficient is... Iterate through the variables for subtask indices; This is a sequence difference function used to output sequence numbers. and Non-negative order intervals between ; Based on the total number of subtasks The maximum order interval is determined and used to normalize the non-negative order intervals.

[0092] Here, the order deviation index of the command is calculated based on equations (8) and (7). Specifically, the system uses equation (8) to calculate To express the "displacement" of each subtask, i.e., the expected sequence number. Alignment number with actual The absolute difference between them directly reflects the degree of disorder in a single-step operation. Furthermore, the system aggregates the displacements of all subtasks using equation (7). In equation (7), the denominator... It played a crucial role in normalization because, within a length of In a sequence, the maximum possible misalignment distance is limited by the sequence length; by dividing by The deviation value is constrained within the range of [0, 1], and then the average value is calculated. Furthermore, when the total number of instruction subtasks... At that time, there was no order deviation, and the system directly... Set it to 0. This eliminates the influence of task length on error calculation, generating a dimensionless statistical indicator that reflects the overall disorder of the execution process. .

[0093] Based on the above deviation index, the system uses equation (6) to finally calculate the time series consistency index. In this mathematical model, Defined as a sequential decay coefficient (e.g., a value of -2.0), it employs exponential decay logic. When it approaches 0 (i.e., executed completely in sequence with no deviation), Approaching 1 (full marks); and once Increase (increase in disorder) It will decrease dramatically at an exponential rate. Through this non-linear design, it imposes "severe penalties" on procedural violations in emergency response. Compared to linear deduction, exponential decay can more sensitively identify and amplify serious process reversal risks, forcing the execution end to strictly adhere to standard operating procedures.

[0094] Furthermore, in the time series consistency index If the value is below the preset consistency threshold, the closed-loop record corresponding to the instruction is marked as a sequential risk closed-loop record, and supplementary execution steps or supplementary evidence prompts are output based on the alignment results of the instruction subtask sequence and the target action sequence.

[0095] Here, the system will calculate the result. Compare with a preset consistency threshold (e.g., 0.8). If If the value falls below a threshold, the system immediately marks the closed-loop record of the instruction as "sequence risk" and triggers an intelligent feedback mechanism. As a result, the system can analyze specific misalignment nodes based on the alignment results and generate refined prompts, such as "The 'on-site containment' step is detected to be lagging behind the 'cleanup operation'; please provide supplementary explanations or correct evidence." This not only identifies problems but also provides specific rectification directions, achieving a management upgrade from "post-event accountability" to "process correction," ensuring the rigor and compliance of the emergency evidence chain in terms of logical sequence.

[0096] Regarding the implementation details of step S150, in some examples of embodiments of this application, the system delves into the micro-level, performing quality analysis on various types of evidence data associated with each action entity in the target action set. For each action entity in the target action set, the system parses the... The first action entity associated with the Multidimensional feature vectors of evidence-like data (such as on-site law enforcement camera videos) (e.g., video bitrate, resolution, illumination intensity, and jitter amplitude), and calculate the corresponding raw quality metric using a preset modal feature fusion function. :

[0097] Equation (9) In the formula, This includes the sharpness and illumination components of the video modality, the accuracy factor and drift component of the positioning information modality, or the signal-to-noise ratio and sampling rate component of the sensor information modality. For the purpose of addressing the type of evidence Feature dimensionality reduction and fusion operators.

[0098] Here, by calling a preset modal feature fusion function, the high-dimensional features are subjected to dimensionality reduction and fusion processing (e.g., using principal component analysis (PCA) or weighted summation) to generate a scalar form of the original quality metric. .

[0099] Using a preset quality normalization function Mapping raw quality metrics to evidence quality scores : Equation (10) In the formula, Used to eliminate dimensional differences between different types of evidence and map scores to the [0,1] interval.

[0100] To address the issue of inconsistent data dimensions across different sensors (e.g., GPS error is in meters, signal-to-noise ratio is in decibels), the mass normalization function in equation (10) is further utilized. (Such as the Sigmoid or Min-Max scaling function), mapping the original metric to a standardized score within the interval [0, 1]. Therefore, a unified quality evaluation benchmark across modalities was constructed, making the evidence quality of heterogeneous devices such as cameras, locators, and sensors mathematically comparable.

[0101] Then, based on the preset evidence weights Weighted fusion is performed on the evidence quality scores of the same action entity to calculate the action credibility of that entity. : Equation (11) In the formula, The total number of evidence types, and satisfying the normalization constraint. .

[0102] Here, the system calculates the overall credibility of the action entity based on the differences in the authority of different types of evidence. Specifically, the system loads preset evidence weights. (For example, in evidence collection scenarios, video evidence has a greater weight than log evidence), and the quality of all evidence under the action entity is scored using Equation (11). Perform a linear weighted summation to obtain the action credibility. This formula ensures the stability of the results through normalization constraints. Therefore, the system can dynamically adjust weights according to the business scenario (such as reducing the reliance on visual weights in night mode), comprehensively utilize multi-source information for verification, avoid misjudgments caused by single device failures, and generate a comprehensive indicator reflecting the "degree of conclusive evidence" of the specific action.

[0103] Then, extract the instructions. The semantic elements constitute the demand set Extracting action entities Action attributes constitute an attribute set Calculate action entities based on set operations For instructions Element coverage weight : Equation (12) In the formula, The cardinality of a set. Used to quantify the completeness of how well an action entity fulfills the requirements of an instruction at the semantic level; when At that time, Set to the default value.

[0104] Here, the system shifts from analyzing the quality of evidence at the physical level to analyzing task completion at the semantic level. In some implementations, the system extracts instructions separately. The semantic elements (such as "location: area A", "action: search and rescue", "object: trapped personnel") constitute the demand set. and action entities The attribute features constitute the attribute set Based on equation (12), the system calculates the ratio of the cardinality of the intersection of the two sets to the cardinality of the demand set, thereby obtaining the element coverage weight. To quantify "how much of the task required by the instruction was accomplished," if the instruction required three elements, and the action only implemented one, then... This accurately identifies the depth of content matching between actions and instructions, preventing invalid high scores where "the evidence is of high quality (the video is very clear), but the content captured is irrelevant to the instructions."

[0105] Subsequently, the time series consistency index was fused. Credibility of actions and element coverage weight Calculate the closed-loop exponent of the instruction. : Equation (13) In the formula, Indication of instructions The corresponding set of target actions.

[0106] Here, the system aggregates the evaluation results from three dimensions: timing, quality, and content, and uses equation (13) to calculate the final instruction closed-loop index. Specifically, a fusion logic of weighted averaging and global correction was adopted, using element coverage as the basis. As a weight, the credibility of the action A weighted average was applied to ensure that only actions covering the core elements of the instruction had a significant contribution to the total score based on the quality of their evidence. Secondly, the result was multiplied by a temporal consistency index. As a penalty; this means that even if the evidence is clear and comprehensive, if the execution order is seriously violated ( (If the index is low), the final closed-loop index will still be significantly lowered, thereby achieving a rigorous three-in-one assessment of "timeliness, completeness, and compliance".

[0107] In some cases, when When this occurs, it indicates that the target action set does not semantically cover any instruction elements, and the system will... Set it directly to 0.

[0108] Furthermore, the instruction closed-loop index Write the closed-loop record; if If the evidence quality score is below the preset closed-loop threshold, then the evidence quality score of each action entity in the target action set is traversed. The system filters out low-quality evidence types that are below a preset evidence quality threshold, and generates supplementary evidence requirements that include the action number to be supplemented and the low-quality evidence type based on the filtering results.

[0109] Here, the system triggers an adaptive closed-loop feedback mechanism based on the calculation results. In some implementations, when When the score falls below a preset closed-loop threshold (e.g., 0.85), the system automatically initiates a diagnostic procedure, iterating through the evidence quality scores of all action entities in the target action set. The system compares these scores one by one with a preset evidence quality threshold (e.g., 0.6) to accurately pinpoint the specific evidence dimensions that are "dragging down" the process (e.g., "the location data for Operation No. 3 is severely drifting"). Based on the screening results, the system generates structured supplementary evidence requirements, clearly indicating the operation number and evidence type to be supplemented. The technical effect of this step is to achieve a closed loop from "scoring" to "correction," guiding frontline personnel to carry out targeted remediation and ensuring that the final archived emergency evidence chain is complete, credible, and compliant.

[0110] Figure 2 The diagram illustrates an example of the operational mechanism of a grassroots emergency evidence closed-loop generation method based on command action mapping according to an embodiment of this application. The diagram shows how the system achieves the transformation process from multi-source heterogeneous data to standardized closed-loop evidence through modular collaboration.

[0111] like Figure 2As shown, at the data access layer, the system receives command inputs, operation logs, sensor data, and historical knowledge in parallel. The command parsing module and action extraction module then extract structured features from the intent of the superior command and the on-site execution behavior. These feature data are then routed to the core computing engine: on one hand, the priority and weighting module establishes a logical mapping based on the urgency of the command and the spatiotemporal attributes of the action and calculates the weights; on the other hand, the evidence fusion and evaluation module combines sensor readings with historical prior models to measure the quality of the multimodal data.

[0112] The above calculation results are aggregated in the compliance index calculation module. Based on this, the system comprehensively calculates the instruction closed-loop index (or compliance index), and generates closed-loop evidence and a quantitative performance report that includes both temporal logic and evidence quality. At the same time, when the compliance index or evidence quality triggers a risk threshold, the system automatically generates an alarm notification and feeds it back to the command end, thus forming a complete closed loop of "instruction issuance - action execution - evidence anchoring - feedback review", ensuring the visualization and traceability of the entire emergency response process.

[0113] To verify the effectiveness and superiority of the proposed solution, two sets of simulation experiments were designed and compared with baseline methods. The experiments used a simulation platform to generate multi-scenario emergency task data, including instruction texts, action sequences, sensor records, and evidence quality scores. The performance differences between the baseline method (based solely on time matching and manual verification) and the proposed method were compared based on common workflows in grassroots emergency management.

[0114] Experiment 1: Comparison of Command-Action Mapping Accuracy Experimental Setup: To verify the algorithm's performance under high concurrency and complex interference scenarios, this experiment set up 10 sets of instructions with different concurrency scales, with the number of instructions in each set increasing from 10 to 100 in a gradient manner. Simultaneously, a simulation engine was used to generate a hybrid action dataset for each set of instructions, containing both "real matching actions" and "noise interference actions." A random perturbation function was used to parameterize the "execution time delay," "semantic description features," and "attribution of responsibility" for the actions, simulating spatiotemporal misalignment and semantic ambiguity in real-world scenarios. Based on this data, the proposed Weighted Causal Inference Mapping (WCIM) algorithm was compared with the baseline fixed-time-window matching method. The instruction-action mapping accuracy of the two methods under different data scales (i.e., the proportion of correctly matched pairs output by the algorithm to the number of manually labeled real matches) was statistically calculated to evaluate the effectiveness of the proposed solution.

[0115] Figure 3The graph shows the comparison of command-action mapping accuracy under different concurrent command scales. The horizontal axis represents the number of concurrent commands, which is used to simulate the complexity of different emergency scenarios from low load to high load. The vertical axis represents the accuracy of command-action matching.

[0116] like Figure 3 As shown, the accuracy of the baseline method mainly fluctuates between 0.60 and 0.70, and with the increase in the number of concurrent instructions (especially in the range of 50 to 80), the accuracy declines significantly because simple time matching is insufficient to distinguish high-density overlapping actions. In contrast, the accuracy curve of the method in this application is consistently significantly higher than that of the traditional method, maintaining a high level between 0.78 and 0.88, and does not decrease with the increase of concurrency. This indicates that, thanks to the joint constraints of multi-dimensional features such as semantics, time, subject, and weight, the solution of this invention can effectively eliminate interference noise in high-concurrency scenarios. Compared with the baseline method, its mapping accuracy is improved by an average of about 15% to 20%, demonstrating superior scalability and robustness.

[0117] Experiment 2: Performance of the closed-loop exponent under different complexity and resource conditions Experimental setup: To evaluate the instruction closed-loop exponent proposed in this invention (… To assess sensitivity and robustness in varying environments, this experiment designed a two-dimensional simulation test matrix containing 25 scenario combinations. "Instruction complexity level" is defined as independent variable one (values ​​1 to 5), used to map the length of instruction subtask sequences and the complexity of logical dependencies; "resource availability ratio" is defined as independent variable two (values ​​0.2 to 1.0), used to simulate the proportion of idle time windows in a dynamic resource spatiotemporal occupancy status table. In each scenario combination, 1000 sets of instruction and action sequences are randomly generated, and corresponding multimodal evidence data are generated according to a preset probability distribution. Subsequently, the average instruction closure index (...) under each scenario is calculated using the evaluation model constructed in this invention. The focus is on examining situations where resource constraints cause actions to be suspended (triggering a blocking state) or excessive complexity leads to timing misalignments (impacting...). When this happens, can the index objectively reflect changes in execution quality?

[0118] Figure 4 The diagram shows a heatmap of the closed-loop index distribution of the method in the embodiments of this application under different instruction complexity and resource availability conditions. The horizontal axis represents the instruction complexity level, the vertical axis represents the resource availability ratio, and the color bars on the right represent the normalized instruction closed-loop index values ​​(the brighter the color, the higher the index).

[0119] like Figure 4The data distribution reveals that the closed-loop index exhibits significant characteristics of "resource sensitivity" and "complexity extreme constraint": In cases of extreme resource scarcity (ratio of 0.2, highest level), many key actions are suspended or rendered unenforceable due to resource conflicts, leading to a significant decrease in the weight of element coverage. Consequently, the closed-loop index generally remains in the low range of 0 to 0.12, objectively reflecting the physical constraints under resource scarcity. Conversely, in the "sweet spot" of sufficient resources (ratio > 0.8, next two levels) and moderate complexity (level 3), the system can establish the most complete mapping and evidence chain, resulting in a peak closed-loop index (1.0). Furthermore, even under moderately resource-constrained conditions (e.g., ratio of 0.6), this method can still calculate a closed-loop index with discriminative power (0.2 to 0.56) based on existing partial evidence. This indicates that compared to the traditional binary evaluation model that often directly judges a task as "failure" (0 points) when resources are insufficient, the method of this invention can provide more refined quantitative evaluation results, effectively characterizing the true completeness of instruction execution in complex environments.

[0120] The proposed method for generating closed-loop emergency evidence at the grassroots level based on command-action mapping is significantly superior to traditional scheduling methods that rely solely on time window matching or manual confirmation. Its core advantage lies in constructing a weighted causal reasoning mapping mechanism (WCIM) and a dynamic resource spatiotemporal occupancy model for limited resources at the grassroots level. Unlike digital twin systems that require extremely high computing power, this solution effectively solves mapping ambiguities and resource conflicts in high-concurrency scenarios by multi-dimensionally fusing command semantics, time delays, responsible parties, and dynamic weights. This enables lightweight deployment and accurate assessment even in environments with weak grassroots information infrastructure.

[0121] Furthermore, this invention innovatively establishes a closed-loop evaluation system that includes a time series consistency index and multi-source evidence fusion assessment. This is achieved through the quantitative calculation of the instruction closed-loop index (…). The system can transform discrete field data into objective indicators reflecting execution quality, achieving dual verification of "execution sequence compliance" and "evidence chain integrity". In particular, the automatic feedback and supplementation mechanism based on threshold judgment can generate supplementary requirements in real time when evidence gaps are found, completely solving the pain point of "managing the sending but not the verification" in traditional communication and dispatch systems, and greatly improving the transparency, standardization and full-chain traceability of grassroots emergency law enforcement.

[0122] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of combined actions. However, those skilled in the art should understand that this application is not limited to the described order of actions, as some steps may be performed in other orders or simultaneously according to this application. Secondly, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions and modules involved are not necessarily essential to this application. In the above embodiments, the descriptions of each embodiment have their own emphasis; for parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0123] Figure 5 A structural block diagram of an example of a grassroots emergency evidence closed-loop generation system based on instruction action mapping according to an embodiment of this application is shown.

[0124] like Figure 5 As shown, the grassroots emergency evidence closed-loop generation system 500 based on command action mapping includes a multi-source data structuring unit 510, a command entity generation unit 520, an action entity and evidence index generation unit 530, a mapping matrix construction and judgment unit 540, and a closed-loop index evaluation and recording unit 550.

[0125] The multi-source data structuring unit 510 is used to acquire multi-source emergency data corresponding to emergency events and perform standardization processing on the multi-source emergency data to obtain a corresponding structured data set; the multi-source emergency data includes command data from the command side, action monitoring data from the field side, and evidence data related to field execution.

[0126] The instruction entity generation unit 520 is used to perform semantic parsing and task modeling on the instruction data in the structured data set, extract instruction elements related to emergency response, generate a unique instruction number for each instruction, and determine the instruction weight for each instruction to characterize the urgency and importance of the instruction, thereby generating an instruction feature vector containing the instruction elements and the instruction weight to obtain an instruction entity set.

[0127] The action entity and evidence index generation unit 530 is used to perform behavior recognition and standardization processing on the action monitoring data in the structured data set, extract action type and action spatiotemporal attributes, generate a unique action number for each action, and then generate an action feature vector containing the action type and the action spatiotemporal attributes to obtain an action entity set, and establish an index relationship between each action entity and the corresponding evidence data.

[0128] The mapping matrix construction and determination unit 540 is used to construct a mapping matrix based on the instruction entity set and the action entity set, calculate the corresponding association score for each pair of instruction entities and action entities in the mapping matrix, and perform mapping determination based on the association score to determine the target action set corresponding to each instruction and establish the instruction-action mapping relationship.

[0129] The closed-loop index evaluation and recording unit 550 is used to perform multi-source evidence fusion evaluation on the located evidence data based on the target action set corresponding to the instruction and the evidence data located through the index relationship with the target action set for each instruction, so as to determine the instruction closed-loop index used to characterize the completeness and sufficiency of the instruction execution, and generate a closed-loop record; the closed-loop record includes the instruction number, the action number in the target action set, the corresponding evidence data index, and the instruction closed-loop index.

[0130] In some embodiments, this application provides a non-volatile computer-readable storage medium storing one or more programs including execution instructions. The execution instructions can be read and executed by electronic devices (including but not limited to computers, servers, or network devices) to perform the steps of any of the above-described methods for generating closed-loop emergency evidence based on instruction action mapping in this application.

[0131] In some embodiments, this application also provides a computer program product, the computer program product including a computer program stored on a non-volatile computer-readable storage medium, the computer program including program instructions, which, when executed by a computer, cause the computer to perform the steps of any of the above-described methods for generating closed-loop grassroots emergency evidence based on instruction-action mapping.

[0132] In some embodiments, this application also provides an electronic device comprising: at least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the steps of a method for generating closed-loop emergency evidence based on instruction action mapping.

[0133] The above-described product can perform the methods provided in the embodiments of this application, and has the corresponding functional modules and beneficial effects for performing the methods. Technical details not described in detail in this embodiment can be found in the methods provided in the embodiments of this application.

[0134] The electronic devices in this application can exist in various forms, including but not limited to: mobile communication devices, ultra-mobile personal computer devices, portable entertainment devices, or other airborne electronic devices with data interaction functions.

[0135] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0136] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented using software plus a general-purpose hardware platform, or of course, using hardware. Based on this understanding, the above technical solutions, in essence or the parts that contribute to the related technology, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0137] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A method for generating closed-loop emergency evidence at the grassroots level based on command and action mapping, characterized in that, The method includes: Acquire multi-source emergency data corresponding to the emergency event, and perform standardization processing on the multi-source emergency data to obtain a corresponding structured data set; the multi-source emergency data includes command data from the command side, action monitoring data from the field side, and evidence data related to field execution; Semantic parsing and task modeling are performed on the instruction data in the structured data set to extract instruction elements related to emergency response. A unique instruction number is generated for each instruction, and an instruction weight is determined for each instruction to characterize the urgency and importance of the instruction. Then, an instruction feature vector containing the instruction elements and the instruction weight is generated to obtain an instruction entity set. The action monitoring data in the structured data set is subjected to behavior recognition and normalization processing to extract action type and action spatiotemporal attributes, and a unique action number is generated for each action. Then, an action feature vector containing the action type and the action spatiotemporal attributes is generated to obtain a set of action entities, and an index relationship is established between each action entity and the corresponding evidence data. A mapping matrix is ​​constructed based on the set of instruction entities and the set of action entities. For each pair of instruction entities and action entities in the mapping matrix, a corresponding association score is calculated. A mapping determination is made based on the association score to determine the target action set corresponding to each instruction and to establish an instruction-action mapping relationship. For each instruction, based on the target action set corresponding to the instruction and the evidence data located through the index relationship with the target action set, a multi-source evidence fusion evaluation is performed on the located evidence data to determine the instruction closed-loop index used to characterize the completeness and sufficiency of the instruction execution, and a closed-loop record is generated; the closed-loop record includes the instruction number, the action number in the target action set, the corresponding evidence data index, and the instruction closed-loop index.

2. The method according to claim 1, characterized in that, The standardized processing of the multi-source emergency data to obtain the corresponding structured data set includes: A distributed clock synchronization protocol is used to align the instruction data, the action monitoring data, and the evidence data with a time reference, so as to generate a unified timestamp for each data record and map the multi-source emergency data to a unified time coordinate axis. Noise filtering is performed on the sensor time-series data in the multi-source emergency data, and piecewise linear interpolation is used to repair the missing sampling points caused by communication interruption to form continuous time-series data segments; background subtraction is used to extract effective segments containing dynamic behavior from the video data in the multi-source emergency data. In accordance with a unified data interface standard, the multi-source emergency data that has undergone time base alignment and corresponding preprocessing is encapsulated into structured data units, and the structured data units are aggregated to form the structured data set; the structured data unit includes a unified timestamp, geographic coordinates, data source identifier, and content carrier.

3. The method according to claim 2, characterized in that, The process of determining instruction weights for each instruction to characterize its urgency and importance includes: Based on the instruction elements obtained from the semantic parsing, the first... urgency level of the instruction With priority coefficient And based on the unified timestamp in the structured data set, calculate the first... The time delay of the instruction issuance time relative to the current time. ; Construct an initial weight calculation model that includes positive gain and negative attenuation, based on the aforementioned urgency level. The priority coefficient With the time delay Calculate the first Initial weight of the instruction : ， In the formula, The positive influence coefficient is... It is a negative attenuation coefficient. This is the Sigmoid normalization function, used to map the results of linear combinations to a preset numerical range; Softmax normalization is performed on all initial weights in the instruction entity set to obtain a result reflecting the first... Instruction weights, relative importance of instructions : ， In the formula, The total number of instructions currently pending processing in the instruction entity set. This is used to iterate over variables for instructions.

4. The method according to claim 1, characterized in that, The process of performing behavior recognition and normalization on the action monitoring data in the structured dataset, extracting action types and spatiotemporal attributes, generating a unique action number for each action, and then generating an action feature vector containing the action type and spatiotemporal attributes to obtain an action entity set, and establishing an index relationship between each action entity and the corresponding evidence data, includes: Based on the video observation information, on-site reporting information and operation log information in the action monitoring data, a multi-source consistency check is performed, and the target action type is determined when the multi-source consistency check meets the preset consistency constraints. The pre-set emergency action template library is invoked to standardize the encoding of the target action type, so as to map the non-standard expressions on the scene into a unified set of standard action types, and establish a semantic mapping relationship between regional expressions and standard semantics based on a pre-set regional dialect keyword table, so as to convert the text description containing regional expressions into standard semantics. Construct an action feature vector containing action type components and action spatiotemporal attribute components, and generate a unique action number for each action to form an action entity set; the action spatiotemporal attribute components include the action's execution time, geographical coordinates, duration, and resource consumption indicators; Based on the spatiotemporal attribute components of the action, the evidence retrieval range corresponding to each action entity is determined, and evidence data records within the evidence retrieval range are filtered from the evidence data to generate an evidence data catalog. A traceable index relationship is established between the action number and the evidence data catalog. The evidence data catalog includes at least two of the following types of index identifiers: video keyframe index identifier, positioning trajectory segment index identifier, sensor time sequence segment index identifier, and log entry index identifier.

5. The method according to claim 2, characterized in that, The calculation of the corresponding association score for each pair of instruction entities and action entities in the mapping matrix includes: Analysis of the first The instruction feature vector of the instruction and the first instruction The action feature vectors of each action are used to extract instruction semantic elements. Action type characteristics Expected completion time of the instruction Action execution time Responsible entities Executor of the action and instruction weight The expected completion time of the instruction is specified in the provided text. With the execution time of the action Determined based on the unified timestamp in the structured data set; The weighted association scoring model is invoked to comprehensively consider the impact of semantic similarity, temporal relevance, subject matching, and instruction weight on association, and to calculate the first... The instruction entity corresponding to the instruction is the same as the first instruction. The correlation score between the action entities corresponding to each action : ， In the formula, Indicates instruction semantic elements With the action type characteristics semantic similarity; For the subject matching indicator function, when Belonging to The value is 1 if the condition is met, and 0 otherwise. These are the adjustment coefficients for the corresponding feature dimensions; It is the Sigmoid normalization function; This is a time-related term; ， ， in, For preset time scale parameters, This is the time decay coefficient; This is a time difference function used to output the execution time of the action. With respect to the expected completion time of the instruction nonnegative time interval between .

6. The method according to claim 5, characterized in that, The mapping determination based on the associated score to identify the target action set corresponding to each instruction and establish an instruction-action mapping relationship includes: A dynamic resource spatiotemporal occupancy status table is constructed, which is used to record and update the occupancy status and corresponding resource identifiers of personnel resources, equipment resources and material resources in different time windows. Based on the instruction weight The instruction entity set is sorted in descending order to construct a priority processing queue, and for each instruction in the priority processing queue, an association score is applied. Select action entities that meet the preset mapping threshold conditions from the action entity set to construct a candidate action set; According to the order of the priority processing queue, resource constraint detection is performed on the action entities in the candidate action set. The resource constraint detection includes querying the dynamic resource spatiotemporal occupancy status table to determine whether there is a time window overlap conflict in the personnel resources, equipment resources and material resources required by the candidate action entity within the execution time period of the candidate action entity. If there is no conflict in the resource constraint detection, the occupancy time window of the corresponding resource identifier is locked for the candidate action entity in the dynamic resource spatiotemporal occupancy status table, and the candidate action entity is included in the target action set to establish the instruction-action mapping relationship between the instruction and the target action set. In the event of a conflict in resource constraint detection, an alternative selection strategy is triggered to filter out the candidate action entities that meet the resource constraints and have the highest association score from the candidate action set, and establish an instruction-action mapping relationship between the instruction and the candidate action entities; when there are no candidate action entities that meet the resource constraints, the instruction is marked as resource blocked and suspended, and when the occupancy status of the resource identifier corresponding to the dynamic resource spatiotemporal occupancy status table changes, a re-evaluation of the instruction is triggered to re-determine the target action set.

7. The method according to claim 6, characterized in that, After establishing the command-action mapping relationship, the method further includes: Based on the instruction elements extracted from the instruction entity, a preset processing flow template library is invoked to process the instruction. Each instruction is broken down into a sequence of instruction subtasks arranged in the expected order of processing. ,in The total number of subtasks in the instruction subtask sequence; Based on the spatiotemporal attributes of each action entity in the target action set, the target action set is sorted according to the action execution time to obtain the target action sequence. ,in The total number of actions in the target action sequence; A sequence alignment model is constructed based on the instruction subtask sequence and the target action sequence. This sequence alignment model is used to align each instruction subtask with the target action sequence, provided that preset semantic matching and temporal order constraints are met. Determine its alignment position in the target action sequence. ;in, Indicates the subtask of the instruction. Alignment Actions The sequence number in the target action sequence; Based on the alignment position Calculate the time series consistency index : ， ， ， in, for The order deviation index of instructions. The sequential decay coefficient is... Iterate through the variables for subtask sequence numbers; This is a sequence difference function used to output sequence numbers. and Non-negative order intervals between ; Based on the total number of subtasks The determined maximum order interval is used to normalize the non-negative order interval; The time series consistency index If the value is below a preset consistency threshold, the closed-loop record corresponding to the instruction is marked as a sequential risk closed-loop record, and supplementary execution step prompts or supplementary evidence prompts are output based on the alignment result of the instruction subtask sequence and the target action sequence.

8. The method according to claim 7, characterized in that, For each instruction, based on the target action set corresponding to the instruction and the evidence data located through an index relationship with the target action set, a multi-source evidence fusion evaluation is performed on the located evidence data to determine the instruction closed-loop index used to characterize the completeness and sufficiency of the instruction's execution, and a closed-loop record is generated, including: For each action entity in the target action set, parse the first... The first action entity associated with the Multidimensional feature vectors of evidence data And calculate the corresponding original quality metric value using a preset modal feature fusion function. : ， In the formula, This includes the sharpness and illumination components of the video modality, the accuracy factor and drift component of the positioning information modality, or the signal-to-noise ratio and sampling rate component of the sensor information modality. For the purpose of addressing the type of evidence Feature reduction and fusion operators; Using a preset quality normalization function Map the original quality metric values ​​to evidence quality scores. : ， In the formula, Used to eliminate the dimensional differences between different types of evidence and map the scores to the [0,1] interval; Based on preset evidence weights Weighted fusion is performed on the evidence quality scores of the same action entity to calculate the action credibility of that entity. : ， In the formula, The total number of evidence types, and satisfying the normalization constraint. ; Extraction instructions The semantic elements constitute the demand set Extracting action entities Action attributes constitute an attribute set Calculate action entities based on set operations For instructions Element coverage weight : ， In the formula, The cardinality of a set. Used to quantify the completeness of how well an action entity fulfills the requirements of an instruction at the semantic level; when At that time, Set to the default value; Integrating the time series consistency index Credibility of the actions stated and the feature coverage weight Calculate the closed-loop exponent of the instruction. : ， In the formula, Indication of instructions The corresponding set of target actions; The instruction closed-loop index Write the closed-loop record; if If the evidence quality score is lower than the preset closed-loop threshold, then the evidence quality score of each action entity in the target action set is traversed. The system filters out low-quality evidence types that are below a preset evidence quality threshold, and generates supplementary evidence requirements that include the action number to be supplemented and the low-quality evidence type based on the filtering results.

9. A grassroots emergency evidence closed-loop generation system based on command action mapping, characterized in that, The system includes: A multi-source data structuring unit is used to acquire multi-source emergency data corresponding to an emergency event and perform standardization processing on the multi-source emergency data to obtain a corresponding structured data set; the multi-source emergency data includes command data from the command side, action monitoring data from the field side, and evidence data related to field execution; The instruction entity generation unit is used to perform semantic parsing and task modeling on the instruction data in the structured data set, extract instruction elements related to emergency response, generate a unique instruction number for each instruction, and determine the instruction weight for each instruction to characterize the urgency and importance of the instruction, thereby generating an instruction feature vector containing the instruction elements and the instruction weight to obtain an instruction entity set. The action entity and evidence index generation unit is used to perform behavior recognition and standardization processing on the action monitoring data in the structured data set, extract action type and action spatiotemporal attributes, generate a unique action number for each action, and then generate an action feature vector containing the action type and the action spatiotemporal attributes to obtain an action entity set, and establish an index relationship between each action entity and the corresponding evidence data. The mapping matrix construction and determination unit is used to construct a mapping matrix based on the instruction entity set and the action entity set, calculate the corresponding association score for each pair of instruction entities and action entities in the mapping matrix, and perform mapping determination based on the association score to determine the target action set corresponding to each instruction and establish the instruction-action mapping relationship. The closed-loop index evaluation and recording unit is used to perform multi-source evidence fusion evaluation on the located evidence data based on the target action set corresponding to the instruction and the evidence data located through the index relationship with the target action set for each instruction, so as to determine the instruction closed-loop index used to characterize the completeness and sufficiency of the instruction execution, and generate a closed-loop record; the closed-loop record includes the instruction number, the action number in the target action set, the corresponding evidence data index, and the instruction closed-loop index.

Images