A method for assessing the risk of an animal disease

Abstract

The application discloses an animal epidemic risk assessment method and belongs to the field of animal epidemic prevention and quarantine information processing. In order to solve the problem that animal epidemic prevention and quarantine data sources are scattered, and the transmission chain and risk nodes are difficult to be associated in time, the application realizes the technical effect of field area, batch, vehicle, route and regional risk linkage assessment through subject identification, time standardization, dynamic heterogeneous transmission graph, evidence chain gating, risk memory vector, time sequence diagram reasoning and active check closed loop.

Description

Technical Field

[0001] This invention relates to the field of animal disease prevention and quarantine information processing, and in particular to a method for assessing animal disease risk. Background Technology

[0002] Animal disease prevention and quarantine scenarios typically involve multiple types of data, including breeding records, immunization records, transportation quarantine, vehicle routes, laboratory testing, abnormal mortality reporting, and farm location. Existing information systems can record business data during the breeding, quarantine, testing, and transportation processes, but different data sources differ in subject identification, time granularity, and event description methods, making it difficult to directly form a transmission chain connecting farms, animal batches, transport vehicles, quarantine stations, and administrative regions.

[0003] In existing manual screening or single-point rule-based judgment methods, risk identification typically relies on local information such as positive test results, abnormal deaths, or single transport records. When co-transportation, cross-regional transfers, spatial proximity, and temporal proximity all play a role, manual methods struggle to promptly determine the possible paths of risk diffusion from one entity to adjacent entities, easily leading to delays in risk identification across regions and transportation links.

[0004] In addition, existing risk assessment methods are insufficient in handling expired records, underreported data, and single-point anomalies. They lack a dynamic update mechanism that can simultaneously combine the incubation period of the disease, the infectious period, the effective period of immunity, the credibility of quarantine, and testing evidence. They also lack a closed-loop mechanism to transform the uncertainty of risk scores into tasks such as laboratory sampling, on-site verification, or interception of transportation routes.

[0005] Therefore, there is a need for an animal disease risk assessment method that can address the shortcomings of existing technologies. Summary of the Invention

[0006] One objective of this invention is to propose an animal disease risk assessment method. Addressing the problem in existing technologies where data such as breeding records, immunization records, transportation quarantine, laboratory testing, and abnormal death reporting are scattered, making it difficult to timely correlate transmission chains and risk nodes, this invention proposes a technical solution that involves subject identification and time standardization of multi-source data, constructing a dynamic heterogeneous transmission graph, and combining evidence chain gating, risk memory vectors, time-series graph reasoning models, and proactive verification closed-loop for risk updates. This invention has the technical effect of enabling risk nodes across farm areas and transportation links to be updated in a linked manner with positive test results, abnormal deaths, or high-risk transportation activities.

[0007] This invention provides a method for animal disease risk assessment, comprising: S1, acquiring multi-source animal disease prevention and quarantine data and performing subject identification and time standardization to generate an event dataset with unified subject identifiers and standard timestamps; S2, generating a heterogeneous entity set with farms, animal batches, transport vehicles, quarantine stations, laboratory samples, administrative regions, and abnormal events as nodes according to the event dataset, and generating a candidate propagation edge set; S3, extracting evidence vectors for each candidate propagation edge, and gating and weighting the evidence vectors according to the incubation period, infectious period, and immunization validity period of the target disease to generate a dynamic heterogeneous propagation graph; S4; Based on the node's inventory size, immunization status, transportation frequency, test results, degree of mortality anomaly, and historical risk status, generate node temporal features and risk memory vectors; S5, input the dynamic heterogeneous propagation graph, the node temporal features, and the risk memory vectors into the trained temporal graph inference model, iteratively update the node risk status and edge propagation probability, and output risk scores, risk levels, and suspected propagation chains; S6, based on the confidence interval of the risk score and the distance-weighted local entropy of unchecked nodes in the neighborhood, generate quarantine control priorities and a check list, and write the check results back to the event dataset as input for the next round of updates.

[0008] Optionally, S1 includes: Normalize the fields of farm name, unified social credit code, farm area coordinates, animal batch ear tag or batch number, vehicle license plate, quarantine certificate number, laboratory sample number and reported event number; A subject identifier mapping table is used to merge different records of the same subject into a unified subject identifier; Convert the immunization date, transportation start time, quarantine release time, sampling time, test completion time, and abnormal death reporting time into standard timestamps; Data credibility factors are generated for missing or conflicting fields, and these data credibility factors serve as input fields for the evidence vector in step S3.

[0009] Optionally, S2 includes: Map the unified subject identifier to the node identifier, and the standard timestamp to the node attribute and event sequence; When an animal batch is transferred from the first farm to the second farm, a transfer transmission edge is generated; when two animal batches are transported in the same vehicle, a same-vehicle transport edge is generated; when a contact record appears within the same farm area time window, a same-farm contact edge is generated; when the animal is released at a quarantine station, a quarantine release edge is generated; when a laboratory sample tests positive, a test evidence edge is generated; and when the administrative region or farm area meets the preset spatial distance threshold and the event time difference does not exceed the preset time threshold, a proximity transmission edge is generated. A set of candidate propagation edges is generated from the node identifier, the node attributes, and each propagation edge.

[0010] Optionally, S3 includes: For each candidate propagation edge, the event time, spatial distance, number of animal batches, same-vehicle transportation relationship, immunization date, vaccine type, quarantine certificate status, test results, and number of abnormal deaths are read from the two nodes of the edge, and an evidence vector is generated, including contact intensity factor, time window matching factor, immune protection attenuation factor, quarantine test credibility factor, and abnormal death triggering factor. If the event time difference of a candidate propagation edge falls within the propagation time window corresponding to the incubation period and infectious period of the target disease, the candidate propagation edge is retained; otherwise, the gating value of the candidate propagation edge is set to zero. The evidence vector, the gate value, and the data credibility factor are weighted and fused to obtain dynamic edge weights, and the propagation credibility matrix and dynamic heterogeneous propagation graph are generated from the dynamic edge weights. Furthermore, the gate value is determined as follows: a source record identifier, a record timestamp, and a credibility identifier are set for each factor in the evidence vector; When the quarantine certificate status is inconsistent with the vehicle route, the immunization date is later than the transportation start time, the laboratory sample number is missing, or the abnormal death reporting time exceeds the preset reporting time limit, an evidence conflict marker is generated. The gating value is modified by combining the evidence conflict marker, the data credibility factor, and the target epidemic time window, so that the dynamic edge weight of the candidate propagation edge recording the conflict in the propagation credibility matrix is ​​reduced by a preset conflict attenuation coefficient.

[0011] Optionally, S4 includes: The node's time sequence characteristics are generated based on the corresponding stock size, immunization status, time interval from the last immunization date, number of transfers in, number of transfers out, number of quarantine releases, test result category, difference between the number of abnormal deaths and the baseline number of deaths, and the risk status of the previous round. The baseline number of deaths is the average number of deaths of the same animal category in the same field within a preset historical statistical period. Events where the difference between the number of abnormal deaths and the baseline number of deaths exceeds a preset death number threshold are identified as abnormal death cluster events. Risk memory vectors are set for farm nodes, animal batch nodes, and transport vehicle nodes respectively. The risk memory vectors include exposure risk components, infection risk components, transmission risk components, and blocking risk components. When a node receives a positive laboratory result, an abnormal death cluster event, or an incoming event from a node whose risk score exceeds a preset scoring threshold, the corresponding risk component is increased according to the event credibility factor. When a node receives a valid immunization, negative test, blockade, disinfection, or quarantine event, it reduces the corresponding risk component according to a preset attenuation coefficient.

[0012] Optionally, S5 includes: The propagation credibility matrix is ​​used as the adjacency weight of the time sequence graph reasoning model, and the node time sequence features and the risk memory vector are used as node input features. In each round of reasoning, the risk memory vectors of adjacent nodes are aggregated according to the dynamic edge weights, and the message transmission range is limited by the time mask generated by the incubation period, infectious period and immune validity period. The aggregation result is concatenated with the previous risk status of the node and then input into the gating update unit to output the current node risk status, edge propagation probability, risk score, risk level and confidence interval. Furthermore, the time-series graph reasoning model is trained in the following way: using the confirmed infection nodes, exposure nodes, blocking nodes and uninfected nodes in the historical animal disease treatment records as labeled samples, a training sample is constructed that includes a historical event dataset, a historical propagation credibility matrix, historical node time-series features and a historical risk memory vector. The training objective is a weighted sum of node risk state classification loss, edge propagation probability regression loss, propagation chain sequence consistency loss, and confidence interval calibration loss. The confidence interval calibration loss is determined by quantiles based on the inconsistency scores between the predicted risk score and the confirmation label in the calibration sample, and the quantiles are mapped to confidence interval width corrections.

[0013] Optionally, S6 includes: For each node, read the risk score and the confidence interval, and calculate the confidence interval width; When the risk score exceeds a preset score threshold and the confidence interval width exceeds a preset width threshold, the node is determined as a node to be verified. Within a preset neighborhood radius of the node to be verified, count the number of unverified nodes, the risk score distribution of unverified nodes, and the road network distance from unverified nodes to the node to be verified, and calculate the distance-weighted local entropy. Based on the distance-weighted local entropy, the edge propagation probability, and the processing capacity of the quarantine stations, a quarantine deployment priority and a list of laboratory sampling, on-site verification, or transportation route interception are generated. The verification results are converted into new event records and written into the event dataset; Furthermore, the quarantine deployment priority is generated in the following way: the distance-weighted local entropy, the edge propagation probability, the risk score, the confidence interval width, and the quarantine station processing capacity are input into the priority mapping table to obtain the node deployment score, route deployment score, and area deployment score; When the verification result is positive, an abnormal death or cluster event is found on site, or a contact event is confirmed by interception of a transportation route, the event credibility factor of the corresponding newly added event record will be set as the first credibility factor. When the verification result is negative, no abnormal death or gathering event is found on site, or no contact event is confirmed when the transportation route is intercepted, the event credibility factor of the corresponding newly added event record will be set as the second credibility factor. The first confidence factor is greater than the second confidence factor, and both are determined by a preset confidence factor table.

[0014] The beneficial effects of this invention are: 1. By unifying farms, animal batches, transport vehicles, quarantine stations, laboratory samples, administrative regions, and abnormal events into a heterogeneous entity set, and generating candidate propagation edges based on relationships such as transfer in / out, same vehicle transport, same site contact, quarantine release, positive test, abnormal death clusters, spatial proximity, and temporal proximity, it is possible to transform scattered business data into a propagation graph structure that can be reasoned about, thus providing a clear data foundation for cross-entity risk association.

[0015] 2. By extracting contact intensity factors, time window matching factors, immune protection attenuation factors, quarantine detection reliability factors, and abnormal death triggering factors from candidate transmission edges, and combining them with incubation period, infectious period, and immune validity period to generate gating values ​​and dynamic edge weights, the interference of expired records, single-point anomalies, and conflicting records on transmission judgment can be reduced, making the transmission reliability matrix match the transmission pattern of the target disease.

[0016] 3. By maintaining the risk memory vector and inputting the propagation credibility matrix, node temporal features, and risk memory vector into the trained temporal graph inference model, the node risk status and side propagation probability can be iteratively updated after the occurrence of new positive detections, abnormal death clusters, effective immunization, negative detections, or quarantine blockage events. By generating a verification list using confidence intervals and distance-weighted local entropy and writing back the verification results, a proactive verification closed loop oriented towards quarantine control can be formed. Attached Figure Description

[0017] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a flowchart of a method for assessing the risk of animal diseases.

[0018] Figure 2 This is a flowchart of step S3 of the present invention, which generates a dynamic heterogeneous propagation graph. Detailed Implementation

[0019] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0020] refer to Figures 1-2 A method for assessing animal disease risk includes: S1. Acquiring multi-source animal disease prevention and quarantine data and performing subject identification and time standardization to generate an event dataset with unified subject identifiers and standard timestamps; S2. Generating a heterogeneous entity set with farms, animal batches, transport vehicles, quarantine stations, laboratory samples, administrative regions, and abnormal events as nodes according to the event dataset, and generating a candidate propagation edge set; S3. Extracting evidence vectors from each candidate propagation edge, and gating and weighting the evidence vectors according to the incubation period, infectious period, and immunization validity period of the target disease to generate a dynamic heterogeneous propagation graph; S4. According to... S5. The node's inventory size, immunization status, transportation frequency, test results, degree of mortality anomaly, and historical risk status are used to generate node temporal features and risk memory vectors; S6. The dynamic heterogeneous propagation graph, the node temporal features, and the risk memory vectors are input into the trained temporal graph inference model to iteratively update the node risk status and edge propagation probability, and output the risk score, risk level, and suspected propagation chain; S7. Based on the confidence interval of the risk score and the distance-weighted local entropy of unchecked nodes in the neighborhood, quarantine control priorities and a check list are generated, and the check results are written back to the event dataset as the input for the next round of updates.

[0021] In this specific embodiment, S1 includes: The system reads raw records from the breeding archive, immunization registration system, transportation quarantine system, vehicle tracking system, laboratory testing system and abnormal mortality reporting system, and parses each raw record into a structured event record. The event record contains at least four types of information: subject field, time field, spatial field and business field. The main fields include farm name, unified social credit code, farm coordinates, animal batch ear tag or batch number, vehicle license plate, quarantine certificate number, laboratory sample number, and reported event number. Field normalization is performed on these fields, including unifying the character set to UTF-8, converting full-width Chinese characters to half-width characters, unifying letters to uppercase, removing leading and trailing spaces, compressing consecutive whitespace into single spaces, removing non-business punctuation marks, and retaining necessary separators. For coordinate fields, latitude and longitude are uniformly converted to the WGS-84 coordinate system and retained to 6 decimal places. Simultaneously, the administrative division code is written into the event record as a constraint field for subsequent main entity merging. Establish regular expression validation rules for numbered fields and unify the format without changing the business meaning. Validate the unified social credit code according to the 18-digit rule and remove the separator. Remove color and spaces from vehicle license plates and retain the province abbreviation and alphanumeric main body. Remove redundant characters from the prefix and suffix of quarantine certificate number, laboratory sample number and reported event number and retain the main number. Ear tags or batch numbers are written to the event log in a three-segment code of "species code-batch main number-batch serial number". When the original field does not meet the three-segment structure, a reversible standardized code is generated according to the field content and the original field is retained for traceability. Entity identification is implemented using an entity identifier mapping table, which stores the mapping relationship of "source system entity key → unified entity identifier" and is constructed according to entity type. The farm's main key is composed of the normalized unified social credit code and the farm's coordinates. When the unified social credit code exists and passes verification, it is used as the main key to complete the merging. When the unified social credit code is missing or fails verification, it is merged using the normalized farm name, administrative division code, and farm coordinates. The name similarity is calculated using edit distance similarity with a threshold of 0.92, and the coordinate distance is calculated using spherical distance with a threshold of 50 meters. Multiple records are merged into the same unified entity identifier when the name similarity is not lower than the threshold, the coordinate distance does not exceed the threshold, and the administrative division code is consistent. Animal batches are composed of normalized ear tags or batch numbers and species codes; vehicles are composed of normalized vehicle license plates; quarantine stations are composed of normalized station names and administrative division codes; laboratory samples are composed of normalized laboratory sample numbers; and abnormal events are composed of normalized reported event numbers. The unified entity identifier is generated using a fixed prefix plus a hash digest. The hash input is a concatenated string of the corresponding entity key, and the first 16 bytes are taken as the identifier entity segment using the SHA-256 algorithm. The prefix is ​​used to distinguish entity types to avoid cross-type collisions. When the entity key of the same source system already exists in the mapping table, the allocated unified entity identifier is directly reused. When it does not exist in the mapping table, a new unified entity identifier is generated according to the above rules and written into the mapping table. At the same time, the source system, source record identifier, and entity key of the original record are written into the event log to ensure traceability. Time standardization converts the immunization date, transport start time, quarantine release time, sampling time, test completion time, and abnormal death reporting time into a standard timestamp in Beijing time. The standard timestamp is represented in Unix milliseconds and written to the standard timestamp field of the event log. When the original time only contains the date and not the hours, minutes, and seconds, the hours, minutes, and seconds are set to 00:00:00 and the time granularity is marked as "day". Data reliability factors are generated for missing or conflicting fields and written to the event log, wherein the number of key fields preset for each event log is defined as follows: Define the number of fields in the key fields that are empty or have failed validation as... The number of conflicts that violate the consistency check rules is defined as The data credibility factor is defined as: ; in This represents the data reliability factor, with a value range of 0 to... This indicates the total number of key fields corresponding to this event type, and is fixedly configured by the event type table. Indicates the number of missing fields. Indicates the number of collisions; The consistency verification rules include four types of conflicts: immunization date later than the shipment departure time, quarantine release time earlier than the shipment departure time, testing completion time earlier than the sampling time, and the same vehicle license plate appearing in mutually exclusive spatial positions within the same standard timestamp window. Any conflict will be counted as a conflict item. The final output event dataset consists of the following fields: "Unified Subject Identifier, Subject Type, Event Type, Standard Timestamp, Site Coordinates, Administrative Division Code, Business Field, Source Record Identifier, and Data Trust Factor." Each event record is guaranteed to carry the unified subject identifier, standard timestamp, and data trust factor. As an input field of the evidence vector in step S3, it participates in the subsequent propagation of side evidence weighting.

[0022] In this specific embodiment, S2 includes: The system reads the event dataset and writes the unified subject identifier in each event record directly as the node identifier into the heterogeneous entity set, thereby ensuring a one-to-one correspondence between the node identifier and the unified subject identifier and making it traceable. The heterogeneous entity set is divided into farm nodes, animal batch nodes, transport vehicle nodes, quarantine station nodes, laboratory sample nodes, administrative region nodes, and abnormal event nodes according to the subject type. Each node carries node attributes, which include at least the subject type, farm coordinates, administrative division code, the set of standard timestamps associated with the node, and the event sequence number arranged in ascending order of standard timestamps. The event sequence number is generated by stably sorting all event records according to standard timestamps. The stable sorting is based on the lexicographical order of the source record identifier when the standard timestamps are the same, thus ensuring that the event sequence number is uniquely determined under the same input dataset. The candidate propagation edge set is stored with a field structure of "edge start node identifier, edge end node identifier, edge type, edge occurrence standard timestamp, edge supporting record set, and edge attribute". The edge attribute includes at least spatial distance, time difference, batch quantity, quarantine certificate number, vehicle route identifier, and source record identifier set written in step S1. The construction of the candidate propagation edge set is completed by scanning the event dataset one by one according to the event type, and missing nodes are created synchronously during the scanning process. When an event record indicates that an animal batch has been transferred from the first farm to the second farm, the system reads the batch node identifier, the first farm node identifier, the second farm node identifier, the standard timestamp corresponding to the transfer start time, and the standard timestamp corresponding to the quarantine release time from the event record. It then generates two transfer propagation edges. The first transfer propagation edge starts at the first farm node identifier and ends at the animal batch node identifier, and the standard timestamp of the edge occurrence is taken from the standard timestamp corresponding to the transfer start time. The second transfer propagation edge starts at the animal batch node identifier and ends at the second farm node identifier, and the standard timestamp of the edge occurrence is taken from the standard timestamp corresponding to the quarantine release time. At the same time, the number of animal batches is written into the edge attributes of the two transfer propagation edges, and the source record identifier of the event record is written into the edge support record set of the two transfer propagation edges. When an event record indicates that the same vehicle carries a batch of animals, the system uses "transport vehicle node identifier + quarantine certificate number" as the task key for a transport task. All animal batch node identifiers appearing under the same task key are aggregated into the same transport batch set. When the number of elements in the transport batch set is not less than 2, a same-vehicle transport edge is generated for each animal batch node in the transport batch set. The starting point of the same-vehicle transport edge is the transport vehicle node identifier and the ending point is the corresponding animal batch node identifier. The standard timestamp of the edge occurrence is the standard timestamp corresponding to the earliest transfer start time under the transport task. The quarantine certificate number and vehicle route identifier are written into the edge attributes for subsequent evidence extraction. When an event record indicates that a contact record has occurred within the same field time window, the system reads the first animal batch node identifier, the second animal batch node identifier, the field coordinates, and the standard timestamps corresponding to the contact start time and contact end time from the contact record. It generates a contact edge in the same field and writes the contact edge in the candidate propagation edge set in a bidirectional manner. That is, it writes two edges respectively: "first animal batch node identifier → second animal batch node identifier" and "second animal batch node identifier → first animal batch node identifier". The standard timestamp of the edge occurrence is taken from the standard timestamp corresponding to the contact start time, and the standard timestamp corresponding to the contact end time is written into the edge attribute as the upper bound of the contact duration. When the event log indicates that the quarantine station has released the animal, the system reads the quarantine station node identifier, the animal batch node identifier of the released animal, and the standard timestamp corresponding to the quarantine release time. It generates a quarantine release edge with the starting point of the quarantine station node identifier and the ending point of the animal batch node identifier. At the same time, the quarantine certificate number and the release result status are written into the edge attribute. When the event log indicates that the laboratory sample test is positive, the system reads the laboratory sample node identifier, the animal batch node identifier of the tested object, the standard timestamp corresponding to the test completion time, and the test result, generates a test evidence edge with the starting point of the test evidence edge being the laboratory sample node identifier and the ending point being the animal batch node identifier, and writes the test result into the edge attribute. When generating proximity propagation edges, the system uses farm nodes and administrative region nodes as proximity determination objects and calculates the spatial distance using the farm coordinates or administrative region center point coordinates in the node attributes. Simultaneously, it calculates the time difference using the latest standard timestamp in the node attributes, and follows the formula: Determine whether to write to the adjacent propagation edge; in Represents a node With nodes The trigger indicator for generating adjacent propagation edges is generated between them. This represents an indicator function, which takes the value 1 if the condition within the parentheses is true, and 0 otherwise. and These represent the node identifiers of the two nodes participating in the proximity determination. Represents a node coordinates and nodes The spherical distance between the coordinates is calculated using the great circle distance algorithm based on the WGS-84 coordinate system. Represents a node The latest standard timestamps and nodes The absolute time difference between the latest standard timestamps This indicates a preset spatial distance threshold with a value of 5000m. This represents the preset time threshold and has a value of 7d, corresponding to the number of milliseconds. when At that time, the system writes the neighbor propagation edges in a bidirectional manner and writes them into the edge attributes. and This allows S3 to extract time window matching factors and spatial proximity evidence; When duplicate edges appear in the candidate propagation edge set, with the same starting node identifier, ending node identifier, edge type, and standard timestamp of edge occurrence, the system merges the duplicate edges into a single edge and performs a union deduplication on the edge support record set. At the same time, it keeps the spatial distance, time difference, batch quantity, quarantine certificate number, and vehicle route identifier in the edge attributes consistent with the values ​​in the merged group or determined values ​​selected according to the record priority table. Thus, it outputs a structured input containing the heterogeneous entity set and the candidate propagation edge set for step S3 to further generate a dynamic heterogeneous propagation graph.

[0023] In this specific embodiment, S3 includes: The system takes the candidate transmission edge set as input and foot-and-mouth disease as the evaluation disease to establish a disease parameter table. The disease parameter table stores the lower limit of the incubation period as 2 days, the upper limit of the incubation period as 14 days, the infectious period as 10 days, and the immunity period as 180 days. At the same time, the "day" is uniformly converted to milliseconds to align with the standard timestamp. The system reads the edge support record set for each candidate propagation edge one by one, and extracts the event time, spatial distance, number of animal batches, same-vehicle transport relationship, immunization date, vaccine type, quarantine certificate status, test results, and number of abnormal deaths related to the nodes at both ends of the edge from the edge support record set. Among them, the event time is taken from the standard timestamp of the edge bound to the candidate propagation edge, the spatial distance is taken from the spatial distance field written in step S2 or calculated by the coordinates of the nodes at both ends of the edge according to the great circle distance algorithm, the number of animal batches is taken from the quantity field in the transportation record or the aggregated result of the transport task, the same-vehicle transport relationship is taken from the consistency mark of the transport task key when the edge type is the same-vehicle transport edge, the immunization date and vaccine type are taken from the latest immunization event record associated with the node, the quarantine certificate status is taken from the consistency verification result of the release result status and certificate number of the quarantine release event record, the test results are taken from the positive or negative results of the laboratory test event record, and the number of abnormal deaths is taken from the abnormal event node or the farm node's abnormal death cluster event count within the statistical period. Based on this, the system generates an evidence vector for each candidate propagation edge. The evidence vector consists of a contact intensity factor, a time window matching factor, an immune protection attenuation factor, a quarantine detection reliability factor, and an abnormal death triggering factor, with each factor normalized to 0 to 1. The contact intensity factor is determined by the same vehicle transport marker, the number of animal batches, and the spatial distance, with linear attenuation applied to the spatial distance from 0 to 5000 meters. The time window matching factor is determined by the time difference between the standard timestamp of the edge occurrence and the standard timestamp of the most recent risk triggering event at the source node, with the risk triggering event limited to a laboratory positive result event or an abnormal death. The clustering event and the immune protection attenuation factor are determined by the ratio of the time interval between the edge occurrence standard timestamp and the last immunization date to the 180-day immunization validity period, and are set to 0 when the immunization validity period is exceeded. The quarantine detection credibility factor is jointly determined by the quarantine certificate status, the test result, and the data credibility factor of the event record in the edge support record set, and is set to 1 when the test result is positive and 0 when the quarantine certificate status is consistent and the test result is negative. The abnormal death triggering factor is determined by whether the abnormal death clustering event count occurs within 7 days before the edge occurrence standard timestamp, and is set to 1 when it occurs and 0 when it does not occur. To ensure traceability of evidence and conflict correction, the system binds a source record identifier, a record timestamp, and a credibility identifier to each factor in the evidence vector. The source record identifier is taken from the source record identifier in the set of edge support records that the factor depends on. The record timestamp is taken from the standard timestamp of the corresponding source record. The credibility identifier is fixedly mapped to three levels according to the source system type and corresponds to values ​​of 1.0, 0.7, and 0.4 respectively. When calculating the evidence vector, each factor is multiplied by the value corresponding to its credibility identifier to suppress the influence of low credibility sources on edge weights. The system generates evidence conflict markers based on the edge support record set of candidate propagation edges. The evidence conflict markers include four types of conflict items: inconsistency between quarantine certificate status and vehicle route, immunization date later than the transportation start time, missing laboratory sample number, and abnormal death reporting time exceeding the 24-hour reporting time limit. When any conflict item is established, the candidate propagation edge is marked as having a conflict and the number of conflict items is recorded. The system generates gating values ​​for candidate transmission edges based on the disease parameter table. The basic judgment of the gating value is whether the time difference falls within the foot-and-mouth disease transmission time window. The transmission time window is 24 days, which is the sum of the lower limit of the incubation period (2 days) to the upper limit of the incubation period (14 days) and the infectious period (10 days). When the time difference is not within the transmission time window or the time difference is negative, the gating value is set to 0. When the time difference is within the transmission time window, the gating value is set to 1. When there is a conflict of evidence marker, the gating value is multiplied by the conflict attenuation coefficient of 0.4 and then multiplied by the edge data credibility factor to complete the joint correction of conflict and missing data. The edge data credibility factor is the minimum value of the data credibility factor of all event records in the edge support record set to reflect the weakest link effect. The system integrates evidence vectors, gate values, and edge data credibility factors into dynamic edge weights and writes them into candidate propagation edges. The dynamic edge weights are calculated according to the formula: ; in Indicates from node Pointing to node The dynamic edge weights have values ​​ranging from 0 to... Indicates the node identifier of the starting point of the candidate propagation edge. Indicates the endpoint node identifier of the candidate propagation edge. This represents the gate value, which ranges from 0 to 1 and is determined jointly by the propagation time window and collision attenuation. This represents the edge data reliability factor, which takes a value between 0 and 1 and is obtained by aggregating the data reliability factors of event records within the edge support record set. This represents the Sigmoid function that maps any real number to 0. This represents the evidence factor weight vector and has a fixed value. These factors correspond to the contact intensity factor, time window matching factor, immune protection attenuation factor, quarantine detection reliability factor, and abnormal death triggering factor, respectively. Representing vectors transpose, The evidence vector representing the candidate propagation edge is fixed as a five-dimensional column vector and arranged in the order of the five types of factors mentioned above; The system constructs row and column indices of the propagation credibility matrix using all node identifiers, and generates a sparse propagation credibility matrix using dynamic edge weights as non-zero elements of the matrix. At the same time, it encapsulates "node set, edge set with edge type and dynamic edge weight, propagation credibility matrix and edge occurrence standard timestamp" into a dynamic heterogeneous propagation graph for S5 to iteratively read and infer updates by time slice.

[0024] In this specific embodiment, S4 includes: The system takes the event dataset, the heterogeneous entity set, and the node risk status output and persisted by step S5 in the previous round as input, and constructs the node time series feature sequence with the day as the time granularity. The time granularity is to divide the standard timestamp according to the natural day of Beijing time and use the standard timestamp corresponding to the zero point of the day as the time index of that day. The system summarizes the events of the day for each node by time index and generates node time-series features. The node time-series features consist of the stock size, immunization status, time interval from the last immunization date, number of transfers in, number of transfers out, number of quarantine releases, test result category, degree of abnormal mortality, and risk status of the previous round. Among them, the stock size is the number of stocks or batches registered at the end of the day in the farm. The immunization status is a binary label, and it is set to 1 when the last immunization event corresponding to the node meets the requirements of the vaccine type field being foot-and-mouth disease and the immunization date being no more than 180 days from the time index of the day, otherwise it is set to 0. The time interval from the last immunization date is expressed in days and is set to 999 when there is no immunization record. The number of transfers in and out are respectively counted as the number of transport transmission edges with the node as the endpoint or starting point on the day. The number of quarantine releases is counted as the number of quarantine release edges with the node as the endpoint on the day. The test result category is coded according to the test results of the associated laboratory samples on the day: 1 indicates positive, -1 indicates negative, and 0 indicates no test. The system calculates the degree of abnormal mortality by the difference between the number of abnormal deaths and the baseline number of deaths and writes it into the node time-series features on the same day. The baseline number of deaths is calculated as the daily average of death event records in the past 30 days for the same farm node and the same animal category. The number of abnormal deaths is calculated by summing the death event records in the window of the past 7 days for the current time index. The difference is obtained by subtracting the result of multiplying the baseline number of deaths by 7 from the 7-day number of deaths. When the difference exceeds the death number threshold of 5, the system generates an abnormal event node for the farm node in the heterogeneous entity set and marks it as an abnormal death cluster event. The abnormal death cluster event is written into the event dataset for subsequent steps to read. The system maintains risk memory vectors for farm nodes, animal batch nodes, and transport vehicle nodes, and uses these vectors as part of the node's input state. The risk memory vector is defined as follows: ; in Represents a node Risk memory vector, This indicates the node identifier and is consistent with the node identifier in step S2. This represents the exposure risk component and its value ranges from 0 to... This represents the infection risk component, with values ​​ranging from 0 to 1. This represents the transmission risk component, with values ​​ranging from 0 to... This represents the blocking risk component and its value ranges from 0 to 1; Upon initial runtime, the system initializes the risk memory vectors of all three types of nodes to all zeros and updates them daily by incrementing the time index. During updates, the system processes events in ascending order according to the standard timestamps of the event records related to that node on that day and reads the event credibility factor from the event records. The event credibility factor is taken from the data credibility factor field value generated in step S1 and recorded as the current event credibility factor. When a node receives a positive laboratory result event, it will... Increase Current event credibility factor, will Increase Current event credibility factor, will Increase Current event credibility factor; When a node receives an abnormal death aggregation event, Increase Current event credibility factor, will Increase Current event credibility factor, will Increase Current event credibility factor; When a node receives a call-in event, the system reads the previous risk score of the source node corresponding to the call-in event and compares it with the score threshold of 0.7. If the source node's risk score is not lower than 0.7, the receiving node's risk score will be used. Increase Current event credibility factor and will Increase The credibility factor of the current event, and when the incoming event has an associated transport vehicle node, the transport vehicle node will be... Increase The current event credibility factor reflects the risk of cross-entity transport and propagation; when a node receives a valid immune event, it will... Increase Current event credibility factor, and will Multiply (Current event credibility factor) and will Multiply (Current event credibility factor) When the node receives a negative detection event, Increase Current event credibility factor, and will Multiply (Current event credibility factor) and will Multiply (Current event credibility factor) When a node receives a lockdown / disinfection event or a quarantine / blockage event, it will... Increase Current event credibility factor, and will Multiply (Current event credibility factor) After each component update, the system trims the four risk components to the 0 to 1 range to ensure numerical stability. The updated risk memory vector and the node time-series features generated on the same day are written into the "node-day" index table to form a node input feature sequence that can be read by time slice in step S5. At the same time, abnormal death cluster events are written back to the event dataset as new event records to maintain data consistency in subsequent steps.

[0025] In this specific embodiment, S5 includes: The system reads the dynamic heterogeneous propagation graph, node temporal features, and risk memory vector daily according to the time index of S4, and uses the propagation credibility matrix generated in step S3 as the adjacency weight of the temporal graph inference model to participate in the inference update. Index of each day's time The system for each node Construct the node input feature vector as "node temporal features" With risk memory vector splicing, among which The fixed risk level is composed of the following factors in the order listed above: stock size, immunization status, time interval since the last immunization, number of transfers in, number of transfers out, number of quarantine releases, test result category, degree of abnormal mortality, and risk status from the previous round. It is fixed and consists of exposure risk component, infection risk component, transmission risk component and blocking risk component in the above order; The system determines the neighborhood set based on the node pairs with non-zero dynamic edge weights in the propagation credibility matrix, and constructs a time mask based on the 2-day incubation period, 10-day infectious period of the target disease, and the 24-day propagation time window corresponding to their sum. The time mask is applied to each edge based on the standard timestamp of the edge's occurrence and the current day's time index. The time difference is used to determine the message transmission, thereby limiting the message transmission to only occur on the edge of the propagation time window; In each day's reasoning process, the system executes three rounds of graph reasoning iterations. Each iteration first performs message aggregation based on dynamic edge weights and time masks, followed by gating update unit updates. The message aggregation is performed according to the formula: ; in Represents a node In time index The neighborhood aggregated message vector with dimension 1 Indicates the node identifier. Indicates the day-time index. Represents a node The neighboring node identifier, Indicates time index Corresponding to nodes in the dynamic heterogeneous propagation graph The set of neighboring nodes connected by candidate propagation edges and with dynamic edge weights not being zero. This represents a time mask with a value of 0 or 1, and is determined by whether the standard timestamp falls within the specified range. The endpoint is determined within a 24-day time window. Indicates from node Pointing to node The dynamic edge weights are output by S3 and their values ​​range from 0 to... Represents a node In time index Risk memory vector; The gated update unit uses a GRU gated recurrent unit and updates in each iteration by using " and The concatenated vector is used as input, and the hidden representation of the risk state of the node in the previous round is used as the input of the GRU hidden state, so as to output the hidden representation of the risk state of the node in the current round. The dimension of the GRU hidden state is fixed at 64 and the weight parameters are learned during the training phase. After three rounds of iteration, the system obtains inference output through three sets of output heads. One is a node risk state classification head, which outputs the probability distribution of four risk states for each node and uses the probability of the "infection" category as the risk score, while also applying a threshold. The first part is the risk score mapping to four risk levels, and the second part is the edge propagation probability regression head. For each edge, the final hidden representation of the nodes at both ends of the edge and the dynamic edge weight are input into a two-layer perceptron and the edge propagation probability is output through Sigmoid. The third part is the confidence interval output head. The lower quantile value of the risk score of 0.05 and the upper quantile value of 0.95 are output as the upper and lower bounds of the confidence interval, and the confidence interval width is output, using quantile regression. The time-series graph reasoning model is obtained through training. The training samples are constructed from historical animal disease treatment records. Each training sample contains a historical event dataset, a historical propagation credibility matrix, historical node time-series features, and a historical risk memory vector. Confirmed infected nodes, exposed nodes, blocked nodes, and uninfected nodes are used as node risk status labels. At the same time, the confirmed propagation relationship in the treatment records is used as the edge propagation probability supervision signal, and confirmed propagation edges are marked as 1, while other candidate propagation edges are marked as 0. The training objective is a weighted sum of four losses with fixed weights. The four losses are 0.20, namely node risk state classification cross-entropy loss, edge propagation probability mean square error loss, propagation chain sequence consistency loss, and confidence interval calibration loss. The propagation chain sequence consistency loss is constrained by the fact that the peak time of the predicted risk score of the upstream node in the confirmation propagation chain is no later than the peak time of the predicted risk score of the downstream node, and a penalty term is included for chain segments that violate the constraint. The confidence interval calibration loss is generated by calculating the non-consistency score between the predicted risk score and the confirmation label on the calibration sample and taking its 0.95 quantile value to generate the confidence interval width correction, so that the predicted confidence interval coverage is consistent with the 0.90 target coverage. The training uses the Adam optimizer with a fixed learning rate of 0.001, a fixed batch size of 256, and a fixed number of training rounds of 50. The model parameters with the highest F1 score for node classification in the validation set are used as the final model parameters. The system then solidifies these final model parameters for online inference and outputs the node risk status, edge propagation probability, risk score, risk level, and confidence interval each day, as well as the suspected propagation chains obtained by backtracking from high to low edge propagation probability.

[0026] In this specific embodiment, S6 includes: The system reads the risk score, risk level, side propagation probability, and upper and lower bounds of the confidence interval for each node, and generates verification decisions and deployment tasks on a node-by-node basis. The system will nodes Risk score is recorded as The value ranges from 0 to 1, and the confidence interval is denoted as... And among them Indicates the lower bound, Describe the upper bound and satisfy The system calculates the confidence interval width. ; The system fixes the scoring threshold at 0.70 and the width threshold at 0.25. and Time node This node has been identified as a pending verification point. The system checks each node. Construct a preset neighborhood radius and calculate the set of unverified nodes within the neighborhood. The preset neighborhood radius is a road network distance radius of 30km. The road network distance is calculated by using the edge weight of the road network graph as the road length and the Dijkstra algorithm. Unverified nodes are defined as nodes that do not have "laboratory sampling result event", "on-site verification result event" or "transportation route interception result event" within the most recent 7-day time window. The system counts the number of unverified nodes within a 30km radius of the road network and assigns risk scores for these unverified nodes to five intervals. , , , , Binning, and simultaneously checking each unverified node Calculate its distance to the node to be verified The road network distance is recorded as The unit is meters, and the decay is measured by distance. Calculate distance weights Used for nearest neighbor priority; The system calculates the distance-weighted local entropy based on this. And as a quantitative indicator of the degree of uncertainty diffusion, distance-weighted local entropy is calculated using the formula: ; in Indicates the node to be verified Distance-weighted local entropy, This indicates the risk score bin index. Indicates the total number of boxes and takes the value of Indicates the node to be verified Falling into the neighborhood of the first The distance-weighted percentage of unverified nodes in each sub-container, and ,in Indicates the node to be verified The set of unchecked nodes in the neighborhood of . Represents a set The medium-risk score fell to the first A subset of unverified nodes in each bin. Indicates unverified nodes Corresponding verification nodes Distance weights, Represented by natural constant The natural logarithm with base 0; The system is implemented with Items are treated as having a contribution of 0 to ensure numerical stability; the system then generates a quarantine control priority and a verification list for each node to be verified. The quarantine control priority is determined by a priority mapping table, which is based on... ,node Risk score ,node Maximum edge propagation probability among adjacent edges and nodes Processing capacity of quarantine stations associated with the administrative region For input, where Defined as all nodes This represents the maximum value of the edge propagation probability output by the edge that starts or ends in S5. Defined as the upper limit of the number of laboratory samples that can be processed in this administrative region on a given day and fixedly read from the capacity configuration table; The priority mapping table calculates priority scores using a tiered, cumulative method and sorts them accordingly. The tiering rule is fixed as follows: when... 2 points 3 points 4 points, when 1 point 2 points 3 points, when 1 point 2 points 3 points, when 1 point 2 points 3 points are awarded, and truncation is performed according to capacity constraints: for each administrative region, nodes to be verified are selected sequentially from highest to lowest priority score until the cumulative laboratory sampling volume reaches the target. ; The verification list outputs three types of tasks according to node type. The farm node outputs the "on-site verification + laboratory sampling" task and includes the sampling quantity equal to the square root of the stock size, rounded up, with a maximum of 30. The animal batch node outputs the "laboratory sampling" task and includes the quarantine certificate number and a list of associated transportation and transmission edges. The transport vehicle node outputs the "transport route interception" task and includes the vehicle license plate, route markings, and time windows corresponding to high transmission probability edges. The system writes the verification results back to the event dataset. During the write-back, the verification results are converted into new event records and written into the fields of unified subject identifier, standard timestamp, event type, verification conclusion, source record identifier, and event credibility factor. When the verification conclusion is a positive detection, an abnormal death cluster event found on site, or a confirmed contact event by interception of the transportation route, the event credibility factor is set to the first credibility factor of 0.95. When the verification conclusion is a negative detection, no abnormal death cluster event found on site, or no confirmed contact event by interception of the transportation route, the event credibility factor is set to the second credibility factor of 0.60. Both the first and second credibility factors are fixedly given by the preset credibility factor table and are used as inputs for the next round of updates in steps S1 to S5.

[0027] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

[0028] This invention transforms information that was originally scattered in the breeding, transportation, quarantine, detection and abnormal event reporting links into updatable node risk status and edge propagation probability by standardizing multi-source data, dynamic heterogeneous propagation graph and time series graph reasoning model. This allows risk scores, suspected propagation chains and quarantine control priorities to be obtained from the same propagation graph reasoning process.

[0029] This invention introduces evidence chain gating, propagation credibility matrix, risk memory vector, and active verification closed loop in the candidate propagation edge construction and updating stage, so that conflict records, underreported data, and nodes with wide confidence intervals can be identified and transformed into laboratory sampling, on-site verification, or transportation line interception tasks, thereby enabling the risk assessment results to have both propagation chain interpretability and quarantine control orientation.

Claims

1. A method for assessing the risk of animal diseases, characterized in that, include: S1. Acquire multi-source animal disease prevention and quarantine data, perform subject identification and time standardization, and generate an event dataset with unified subject identifiers and standard timestamps; S2. Generate a heterogeneous entity set with farms, animal batches, transport vehicles, quarantine stations, laboratory samples, administrative regions, and abnormal events as nodes according to the event dataset, and generate a candidate propagation edge set; S3. Extract evidence vectors for each candidate propagation edge, and perform gating weighting on the evidence vectors according to the incubation period, infectious period, and immunization validity period of the target disease to generate a dynamic heterogeneous propagation graph; S4. Generate node temporal features and risk memory vectors according to the node's stock size, immunization status, transportation frequency, test results, degree of mortality anomaly, and historical risk status; S5. Input the dynamic heterogeneous propagation graph, the node temporal features, and the risk memory vector into the trained temporal graph inference model, iteratively update the node risk status and edge propagation probability, and output the risk score, risk level, and suspected propagation chain; S6. Based on the confidence interval of the risk score and the distance-weighted local entropy of the unchecked nodes in the neighborhood, generate the quarantine control priority and the check list, and write the check results back to the event dataset as the input for the next round of updates.

2. The method according to claim 1, characterized in that, S1 includes: The fields of farm name, unified social credit code, farm area coordinates, animal batch ear tag or batch number, vehicle license plate, quarantine certificate number, laboratory sample number and reported event number are normalized; different records of the same entity are merged into a unified entity identifier using a subject identifier mapping table; and the immunization date, transportation start time, quarantine release time, sampling time, test completion time and abnormal death reporting time are converted into standard timestamps. Data credibility factors are generated for missing or conflicting fields, and these data credibility factors serve as input fields for the evidence vector in step S3.

3. The method according to claim 2, characterized in that, S2 include: The unified subject identifier is mapped to the node identifier, and the standard timestamp is mapped to the node attribute and event sequence. When an animal batch is transferred from the first farm to the second farm, a transfer transmission edge is generated. When two animal batches are transported in the same vehicle, a same-vehicle transport edge is generated. When a contact record appears in the same farm area time window, a same-farm contact edge is generated. When the animal is released at the quarantine station, a quarantine release edge is generated. When a laboratory sample tests positive, a test evidence edge is generated. When the administrative region or farm area meets the preset spatial distance threshold and the event time difference does not exceed the preset time threshold, a neighboring transmission edge is generated. A set of candidate propagation edges is generated from the node identifier, the node attributes, and each propagation edge.

4. The method according to claim 3, characterized in that, S3 includes: for each candidate propagation edge, reading the event time, spatial distance, number of animal batches, same-vehicle transport relationship, immunization date, vaccine type, quarantine certificate status, test results, and number of abnormal deaths at both ends of the edge, generating an evidence vector including contact intensity factor, time window matching factor, immune protection attenuation factor, quarantine test credibility factor, and abnormal death triggering factor; when the event time difference of the candidate propagation edge falls within the propagation time window corresponding to the incubation period and infectious period of the target disease, the candidate propagation edge is retained; otherwise, the gating value of the candidate propagation edge is set to zero; the evidence vector, the gating value, and the data credibility factor are weighted and fused to obtain dynamic edge weights, and the propagation credibility matrix and dynamic heterogeneous propagation graph are generated from the dynamic edge weights.

5. The method according to claim 4, characterized in that, S4 includes: Based on the corresponding inventory size, immunization status, time interval from the last immunization date, number of transfers in, number of transfers out, number of quarantine releases, test result category, difference between the number of abnormal deaths and the baseline number of deaths, and the risk status of the previous round, the node time sequence characteristics are generated. The baseline number of deaths is the average number of deaths of the same animal category in the same farm area within a preset historical statistical period. Events where the difference between the number of abnormal deaths and the baseline number of deaths exceeds a preset death number threshold are identified as abnormal death cluster events. Risk memory vectors are set for farm nodes, animal batch nodes, and transport vehicle nodes respectively. The risk memory vectors include exposure risk components, infection risk components, transmission risk components, and blocking risk components. When a node receives a positive laboratory result, an abnormal death cluster event, or an incoming event from a node whose risk score exceeds a preset scoring threshold, the corresponding risk component is increased according to the event credibility factor. When a node receives a valid immunization, negative test, blockade, disinfection, or quarantine event, it reduces the corresponding risk component according to a preset attenuation coefficient.

6. The method according to claim 5, characterized in that, S5 includes: The propagation credibility matrix is ​​used as the adjacency weight of the time sequence graph reasoning model, and the node time sequence features and the risk memory vector are used as node input features. In each round of inference, the risk memory vectors of adjacent nodes are aggregated based on dynamic edge weights, and the message transmission range is limited by the time mask generated by the incubation period, infectious period and immune validity period. The aggregation result is concatenated with the risk state of the node in the previous round and input into the gating update unit to output the risk state of the node in this round, edge propagation probability, risk score, risk level and confidence interval.

7. The method according to claim 6, characterized in that, S6 include: For each node, read the risk score and the confidence interval, and calculate the confidence interval width; When the risk score exceeds a preset score threshold and the confidence interval width exceeds a preset width threshold, the node is determined as a node to be verified. Within a preset neighborhood radius of the node to be verified, the number of unverified nodes, the risk score distribution of unverified nodes, and the road network distance from unverified nodes to the node to be verified are counted, and the distance-weighted local entropy is calculated. Based on the distance-weighted local entropy, the edge propagation probability, and the processing capacity of the quarantine station, a quarantine deployment priority and a list of laboratory sampling, on-site verification, or transportation route interception are generated. The verification results are converted into new event records and written into the event dataset.

8. The method according to claim 6, characterized in that, The time-series graph inference model is trained as follows: using confirmed infected nodes, exposed nodes, blocked nodes, and uninfected nodes from historical animal disease treatment records as label samples, a training sample is constructed that includes a historical event dataset, a historical propagation credibility matrix, historical node temporal features, and a historical risk memory vector; the training objective is a weighted sum of node risk state classification loss, side propagation probability regression loss, propagation chain sequence consistency loss, and confidence interval calibration loss; the confidence interval calibration loss determines the quantile value based on the inconsistency score between the predicted risk score and the confirmed label in the calibration sample, and maps the quantile value to a confidence interval width correction amount.

9. The method according to claim 4, characterized in that, The threshold value is determined as follows: a source record identifier, a record timestamp, and a credibility identifier are set for each factor in the evidence vector; when the quarantine certificate status is inconsistent with the vehicle route, the immunization date is later than the transportation start time, the laboratory sample number is missing, or the abnormal death reporting time exceeds the preset reporting time limit, an evidence conflict marker is generated. The gating value is modified by combining the evidence conflict marker, the data credibility factor, and the target epidemic time window, so that the dynamic edge weight of the candidate propagation edge recording the conflict in the propagation credibility matrix is ​​reduced by a preset conflict attenuation coefficient.

10. The method according to claim 7, characterized in that, The quarantine deployment priority is generated as follows: the distance-weighted local entropy, the edge propagation probability, the risk score, the confidence interval width, and the quarantine station processing capacity are input into the priority mapping table to obtain the node deployment score, route deployment score, and area deployment score; when the verification result is a positive detection, an abnormal death and gathering event is found on site, or a contact event is confirmed by interception of the transportation route, the event credibility factor of the corresponding newly added event record is set as the first credibility factor; When the verification result is negative, no abnormal death or gathering event is found on site, or no contact event is confirmed when the transportation route is intercepted, the event credibility factor of the corresponding newly added event record will be set as the second credibility factor. The first confidence factor is greater than the second confidence factor, and both are determined by a preset confidence factor table.

Images