Animal epidemic disease early warning method based on big data analysis

Abstract

The application discloses an animal epidemic disease early warning method based on big data analysis and relates to the technical field of animal epidemic disease early warning, which comprises the following steps: based on an abnormal candidate item set, time connection and space connection are carried out, a multi-layer propagation chain graph is constructed, source area locking and expansion direction checking are performed, and a propagation risk graph is generated; enhanced inhibition candidate determination is performed on abnormal content in the propagation risk graph, parallel arbitration is carried out, and an animal epidemic disease early warning item set is generated; the animal epidemic disease early warning item set is written back to a breeding unit, directional sampling arrangement and review order arrangement are performed on leading abnormal items, and continuation tracking and dynamic review are carried out on standby abnormal items, and an animal epidemic disease early warning table is generated. The application realizes time-space deconstruction of a propagation path, shortens early warning response time and optimizes prevention and control resource investment.

Description

Technical Field

[0001] This invention relates to the field of animal disease early warning technology, and in particular to an animal disease early warning method based on big data analysis. Background Technology

[0002] In the evolution of animal disease prevention and control technologies, the widespread adoption of IoT technology has driven the large-scale deployment of monitoring equipment for livestock farming environments, enhancing the real-time acquisition capabilities of internal and external environmental parameters, biomarker data, and external meteorological and geographic information within farming units. Big data analytics methods, particularly spatiotemporal data mining models, are being deeply integrated into early warning systems, providing a scientific basis for early disease identification. The industry is actively integrating satellite remote sensing, mobile terminals, and edge computing devices to build a data sensing network covering the entire livestock farming chain, promoting the development of early warning technologies towards real-time and precision, effectively enhancing the foresight and response efficiency of disease prevention and control, and laying a technological foundation for the safe operation of the livestock industry.

[0003] However, existing animal disease early warning technologies still have limitations: First, the lack of unified object identification and spatial positioning algorithms for multi-source heterogeneous data makes it difficult to achieve accurate spatiotemporal alignment of data inside and outside the breeding unit, resulting in frequent data gaps and missing correlation links, which seriously weakens the integrity and credibility of the early warning basis; Second, the anomaly detection process relies on static threshold models and does not introduce matching with the same working conditions and counterfactual mirror analysis, making it impossible to dynamically construct a reference baseline for working condition similarity, which makes anomaly identification susceptible to environmental noise interference, with a high false alarm rate and delayed response, making it difficult to capture the early dynamic characteristics of disease transmission. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides an animal disease early warning method based on big data analysis to solve the problems of insufficient integration of multi-source heterogeneous data and delayed early warning response.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: This invention provides an animal disease early warning method based on big data analysis, comprising: collecting multi-source data from inside and outside the breeding unit; performing object unification and spatial positioning on data from different sources; filling gaps and associating data aligned within the same window to generate a multi-source spatiotemporal disease evidence stream; based on the multi-source spatiotemporal disease evidence stream, performing same-condition matching and mirror reference extraction to construct a same-condition counterfactual mirror baseline, and performing attachment verification and suppression processing to generate an abnormal candidate item set; based on the abnormal candidate item set, performing time-series connection and spatial adjacency connection to construct a multi-layer propagation chain diagram, and performing source area locking and expansion direction verification to generate a propagation risk map; performing enhanced suppression candidate judgment on abnormal content in the propagation risk map and performing parallel arbitration to generate an animal disease early warning item set; writing the animal disease early warning item set back to the breeding unit, performing targeted sampling and review priority arrangement on the dominant abnormal items, and continuously tracking and dynamically reviewing the backup abnormal items to generate an animal disease early warning table.

[0007] As a preferred embodiment of the animal disease early warning method based on big data analysis described in this invention, the steps of performing object unification and spatial positioning on data from different sources are as follows: Collect data from multiple sources inside and outside the breeding unit, perform subject extraction and heteronym unification of breeding objects, and generate a unified base table of objects; Based on the unified collection base table of objects, a unified time stamp conversion, hierarchical spatial principal position selection, and internal and external evidence folding are performed to generate a window-aligned base table.

[0008] As a preferred embodiment of the animal disease early warning method based on big data analysis described in this invention, the steps for generating a multi-source spatiotemporal disease evidence stream are as follows: Perform time-series gap checking, neighbor mirroring completion, and outer reference completion on the peer alignment base table to generate peer alignment data; Cross-source symptom chain merging, sequential writing of related data, and spatial reachability connection are performed on the data aligned with the same window to generate a multi-source spatiotemporal epidemic evidence stream.

[0009] As a preferred embodiment of the animal disease early warning method based on big data analysis described in this invention, the steps for constructing a counterfactual mirror baseline under the same working conditions are as follows: Based on the multi-source spatiotemporal epidemic evidence flow, we perform same-condition constraint folding and condition principal position supplementation to generate a same-condition matching base table; Extract continuous and stable normal evidence segments from the same working condition matching base table, and perform head-to-tail concatenation to obtain the mirror reference entry chain. At the same time, perform reference time sequence resetting to generate the same working condition counterfactual mirror baseline.

[0010] As a preferred embodiment of the animal disease early warning method based on big data analysis described in this invention, the steps for generating the abnormal candidate item set are as follows: The baseline of the counterfactual mirror under the same working condition is matched with the real-time evidence segment for verification, and a set of matching verification items is generated. Based on the counterfactual mirror baseline under the same operating conditions and the set of items for close verification, disturbance suppression diversion and deviation retention registration are performed to generate a set of candidate items for anomalies.

[0011] As a preferred embodiment of the animal disease early warning method based on big data analysis described in this invention, the steps for constructing a multi-layered transmission chain diagram are as follows: Perform time-sequence folding and concatenation on the abnormal candidate entry set to obtain time-sequence candidate segments, and perform layered supplementation of the succession position to generate a time-sequence expanded entry set; Based on the time-series expansion of the entry set, spatial adjacency hierarchy is performed and cross-regional connectivity is verified, generating a multi-layered propagation chain graph.

[0012] As a preferred embodiment of the animal disease early warning method based on big data analysis described in this invention, the steps for generating the transmission risk map are as follows: The temporal succession relationship and spatial adjacency relationship in the multi-layer propagation chain diagram are chained together, and the source region first-order locking is performed to generate a source region locking entry set; Perform extended direction link verification on the source region lock entry set to obtain valid propagation chains, and reorganize abnormal chain segments to generate a propagation risk map.

[0013] As a preferred embodiment of the animal disease early warning method based on big data analysis described in this invention, the step of performing enhanced suppression candidate determination on abnormal content in the transmission risk map includes the following steps: Based on the propagation risk map, abnormal content is expanded item by item, and a dual-track diversion of enhancement and suppression is performed according to the continuity of expansion and interpretability, dividing abnormal content into enhancement candidate positions and suppression candidate positions; Cross-chain consistency verification and source persistence verification are performed on enhancement candidate bits and suppression candidate bits respectively, and abnormal content is refined into priority enhancement bits, general enhancement bits, strong suppression bits and weak suppression bits to generate an enhancement suppression classification table.

[0014] As a preferred embodiment of the animal disease early warning method based on big data analysis described in this invention, the steps for generating the animal disease early warning item set are as follows: The enhanced inhibition grading table is arranged in parallel according to time order and spatial location, and evidence comparison and priority sorting are performed to generate a set of arbitration candidate items; Parallel arbitration is performed on the set of candidate arbitration entries, and the dominant abnormal entries and backup abnormal entries are distinguished. At the same time, risk level classification and spread range marking are carried out to generate a set of animal disease early warning entries.

[0015] As a preferred embodiment of the animal disease early warning method based on big data analysis described in this invention, the steps for generating the animal disease early warning table are as follows: The animal disease early warning item set is spatially written back according to the breeding unit, and risk location markers are added to form an early warning location item set; Based on the early warning location set, targeted sampling is performed around the source area locking location, propagation receiving location, and terminal affected location, and combined with spatial connectivity, a dominant anomaly sampling arrangement table is generated; The dominant anomaly sampling arrangement table is sorted by time order for review and the review trigger conditions are added to generate the dominant anomaly review order table. Continue to track and dynamically review the backup abnormal entries, and merge and register them with the primary abnormal review priority list to generate an animal disease early warning table.

[0016] The beneficial effects of this invention are as follows: by folding under the same working condition constraint and supplementing the working condition principal position, continuous normal evidence segments are extracted to generate a mirror reference chain, realizing the dynamic construction of counterfactual benchmarks, avoiding noise interference and reducing false alarm rate; by folding and connecting in time and connecting in spatial adjacency hierarchy, the abnormal item set is transformed into a multi-layer propagation chain graph, realizing the spatiotemporal deconstruction of propagation path, shortening the early warning response time and optimizing the allocation of prevention and control resources. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart of an animal disease early warning method based on big data analysis.

[0019] Figure 2 This is a comparison chart of early warning scores and responses under short-term disturbance scenarios.

[0020] Figure 3 The graph shows the relationship between the accuracy of identifying the initial location of the source area and the lead time of the first warning.

[0021] Figure 4 This is a comparison chart of effective sampling hit rates under different single-round sampling capabilities. Detailed Implementation

[0022] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0023] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0024] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0025] Reference Figures 1-4 This is one embodiment of the present invention, which provides an animal disease early warning method based on big data analysis, including the following steps: S1: Collect multi-source data inside and outside the breeding unit, perform object unification and spatial positioning on data from different sources, and fill gaps and link data aligned in the same window to generate a multi-source spatiotemporal disease evidence stream; S1.1: Collect multi-source data from inside and outside the breeding unit, perform subject extraction and heteronym unification of breeding objects, and generate a unified data collection table for objects; Furthermore, data from multiple sources both inside and outside the breeding unit are collected one by one according to the source category, and source markers are registered (source categories can be divided into sources collected at the breeding site, sources filled in by manual inspection, sources of quarantine and testing, sources of circulation supervision, and sources of external public information). The name, number, location description, time description, and symptom description of the breeding object in each collection record are separated and aligned. The name or number that can directly indicate the breeding batch, species category, and pen location is used as the main position of the breeding object. Records with different spellings, number formats, and colloquial terms but consistent batch, species, and location are placed in the same row for verification. The corresponding content is uniformly written as the same breeding object name and attached to the same collection row. The source marker and collection order are written synchronously. For content with missing names but which can be identified by other records in the same row, a unified name is written. Records with the same name but different positions or the same position but different names are retained and attached in parallel. The contents in the table are arranged according to the source order and the breeding object order to generate a unified collection base table for objects.

[0026] It should be noted that the multi-source data inside and outside the breeding unit includes records of changes in feeding and drinking water within the pens, records of changes in weather conditions, and records of external media activity.

[0027] S1.2: Based on the unified collection base table of objects, perform unified time stamp conversion, hierarchical spatial principal position selection and internal and external evidence folding to generate a window-aligned base table; Furthermore, the time descriptions in the unified collection table are expanded line by line, and the dates, times, shifts, and sequence descriptions are converted into unified time markers. For records that only record time periods or only record dates, the corresponding unified time markers are added according to the collection sequence. The location descriptions are arranged hierarchically by pen, zone, passage, and outer landing point. The location content that can directly indicate the collection landing point is selected as the hierarchical spatial primary position, and the other location content is written in parallel appendices (for example, if the same record simultaneously states "No. 3 breeding area, No. 2 pen, and east passage entrance", then "No. 2 pen" is determined as the hierarchical spatial primary position, and "No. 3 breeding area" and "east passage entrance" are written in parallel appendices). Then, the records in the table are merged according to the breeding object primary position, unified time marker, and hierarchical spatial primary position. Evidence within the breeding unit and evidence outside the breeding unit within the same time window are folded to the same aligned row position. The sequence markers of time conflict records are retained, and a window-aligned base table is generated.

[0028] It should be noted that the unified time stamp is converted according to the example time range of 5min, 10min, 30min or 1h. The hierarchical spatial principal position is selected in the order of "pen location first, zone location second, passage location third, and peripheral landing point supplement". The content of the remaining positions is written as parallel appendages, so as to merge the evidence within the breeding unit and the evidence outside the breeding unit within the same time range into the same aligned row position.

[0029] S1.3: Perform time-series gap checking, neighbor mirroring completion, and outer reference completion on the peer-aligned base table to generate peer-aligned data; Furthermore, the data is compared row by row around the subject position of the aquaculture object, the unified time marker, and the hierarchical spatial subject position in the peer alignment table. Situations where the same position has a record in the previous time window, a missing record in the current time window, and a record reappears in the next time window are identified as time-series gaps and the gap position is recorded. Alignment rows with consistent symptom content, consistent direction of change, and continuous records from adjacent time windows before and after the gap position are selected as the basis for neighborhood mirroring and completion. Content appearing before and after the gap is filled into the missing position. If neighborhood mirroring still cannot complete the gap, corresponding records with the same peripheral landing point, the same time window, and a stable source are selected as the basis for peripheral reference completion. Content that directly corresponds to the missing position is written into the corresponding alignment row, and the completion source and completion order are simultaneously recorded to generate peer alignment data.

[0030] It should be noted that the neighborhood mirroring completion uses consecutive records within the same time mark range of adjacent records as the neighborhood (e.g., one time mark before and after, or extended outward to three time marks before and after). Priority is given to selecting corresponding records with consistent symptom content, consistent direction of change, and within the allowable time interval as the basis for completion (e.g., time interval not exceeding 10 minutes, 30 minutes, or 1 hour). If the aforementioned content is insufficient, peripheral reference records with the same peripheral landing point, the same time range, and continuous source are called, and the source of completion, the order of completion, and the position of completion are registered in sequence.

[0031] S1.4: Perform cross-source symptom chain merging, sequential writing of related data, and spatial reachability connection on the data aligned with the same window to generate a multi-source spatiotemporal epidemic evidence stream; Furthermore, along a unified time marker, the records in the aligned data are expanded according to the subject of the aquaculture object and the subject of the hierarchical space. First, records from sources such as on-site aquaculture, manual inspection and reporting, quarantine and testing, circulation supervision, and external public information are arranged in the same comparison row. Then, each record is compared to see if the occurrence time of each source is within an adjacent time window, whether the symptom name or symptom manifestation is consistent, whether the direction of change is consistent, and whether the symptom record in one source can be corroborated by the physical signs, environmental records, or peripheral records in another source. For the content that meets the above correspondence, cross-source symptom chain merging is performed, and records from different sources pointing to the same abnormal evolution process are grouped into the same cross-source symptom chain starting segment or continuation segment. Content from different sources with adjacent occurrence times, consistent symptom descriptions, or mutual corroboration is checked in parallel. Parallel checking refers to merging records from different sources that are related to the same abnormal evolution process. Records belonging to the same aquaculture species, at the same spatial level, and within adjacent time windows should be placed in the same verification position. Each record should be checked to verify whether the symptom names correspond, whether the symptom strengths are similar, whether the direction of change is consistent, and whether the abnormal manifestations in one record can be corroborated by the inspection results, test results, environmental changes, or peripheral activity records in another record. Records pointing to the same abnormal evolution process should be linked end to end into the same cross-source symptom chain, and trigger positions, continuation positions, and change positions should be added according to the order of appearance. Each record should be checked to verify whether there is a spatial reachability relationship between the previous record and the next record, such as pen connection, passageway accessibility, adjacent zones, or corresponding peripheral landing points. Records that meet the spatial reachability relationship should be attached to the same link. Records that cannot be directly attached but still retain reference value should be written into the parallel appended link position. At the same time, the source mark, unified time mark, hierarchical spatial primary position, and association order should be synchronously written into the corresponding row position in the chain to generate a multi-source spatiotemporal disease evidence stream.

[0032] It should be noted that spatial reachability refers to the existence of spatial conditions between the locations corresponding to two records, enabling them to be actually connected, adjacently transmitted, or have corresponding paths, so that abnormal content appearing at the previous location can continue to the next location along the location path.

[0033] S2: Based on the multi-source spatiotemporal epidemic evidence stream, perform same-condition matching and mirror reference extraction, construct the same-condition counterfactual mirror baseline, and perform attachment verification and suppression processing to generate an abnormal candidate item set; S2.1: Based on the multi-source spatiotemporal epidemic evidence flow, perform same-condition constraint folding and condition principal position supplementation to generate the same-condition matching base table; Furthermore, the evidence is expanded sequentially along the multi-source spatiotemporal disease evidence stream, including the subject position of the farmed object, the unified time marker, the hierarchical spatial subject position, and the position of the evidence chain. Records with consistent species category, growth stage, farming status, environmental range, and external input background are arranged side by side. Records with complete working condition content are directly retained. For records with missing or inconsistent working condition content, the working condition content that appears repeatedly in adjacent records within the same evidence chain and points to the same position is added to the current row. Records with consistent working condition content and comparable symptom changes are folded into the same matching group. Records with the same name but different conditions are split and rearranged while retaining the parallel marker. The working condition name that can directly summarize the common time period and common state of the records in the same group is written at the beginning of the group as the working condition subject position, and the matching order is synchronously written to the corresponding row position to generate a working condition matching base table.

[0034] S2.2: Extract continuous and stable normal evidence segments from the same working condition matching base table, and perform head-to-tail concatenation to obtain the mirror reference entry chain. At the same time, perform reference time sequence resetting to generate the same working condition counterfactual mirror baseline. Furthermore, the working condition principal position, unified time marker, and matching order are verified line by line in the matching table for the same working condition. Content with stable symptoms, convergent changes, continuous record connection, and no abnormal jumps is identified as a continuous and stable normal evidence segment (for example, records of the same breeding batch showing stable feeding records, no abnormalities in physical examinations, and environmental record fluctuations within the normal range under multiple consecutive unified time markers, with no sudden changes between them, can be identified as a continuous and stable normal evidence segment). The continuous and stable normal evidence segments with consistent working condition principal positions and consistent change trends within adjacent time windows are connected end to end according to their sequential connection relationship to form a mirror reference item chain. According to the unified time marker, each record in the mirror reference item chain is rearranged. Missing time points are filled in with common content from consecutive records and inserted into the corresponding time sequence position. Records with misplaced order are adjusted back to their corresponding row positions according to time sequence, so that each time sequence position under the same working condition retains continuous and normal evolution content, generating a counterfactual mirror baseline for the same working condition.

[0035] It should be noted that the sequential relationship refers to the order of the previous record and the next record on the same time mark, as well as the succession order in the evolution of symptoms. It comes from the verification of the time arrangement, changes and continuation content, and whether there are gaps or jumps in the records of adjacent time windows under the same working condition.

[0036] The continuous and stable normal evidence segment only retains the content with stable feeding records, no abnormalities in vital signs inspections, environmental records within the daily fluctuation range of the same working conditions, and continuous connection between the preceding and following time windows. Items with sudden increases or decreases, jumps across windows, abnormal changes in symptoms, or discontinuous local records are not included in the mirror reference item chain.

[0037] S2.3: Verify the alignment between the counterfactual mirror baseline under the same working conditions and the real-time evidence segment, and generate an alignment verification item set; Furthermore, each baseline record in the counterfactual mirror baseline of the same working condition is arranged into the corresponding verification row position according to the working condition subject position, unified time mark, and hierarchical spatial subject position. Records with consistent time position and consistent spatial landing point in the real-time evidence segment are grouped into the same position. Each record is verified to determine whether the time of occurrence, duration, symptom trend and intensity of change are consistent with the baseline record. For content that is continuously consistent and has a consistent trend, a consistency mark is registered. For content that is earlier, later, turning point, enhanced or weakened, a deviation mark is registered. For content with discontinuous correspondence, partial gaps or landing point deviations, the discontinuity position, gap position and corresponding landing point are added. All the content of the correlation verification is organized according to the unified time mark and matching order to generate a correlation verification item set.

[0038] It should be noted that a real-time evidence segment refers to the actual occurrence evidence content extracted from the multi-source spatiotemporal disease evidence stream under the current unified time marker and organized according to the working condition, time position, and spatial landing point. For example, the feed intake decline record, abnormal vital sign record, and environmental fluctuation record corresponding to the breeding batch under the current unified time marker are organized into a segment of actual occurrence content according to the same working condition and the same pen landing point.

[0039] S2.4: Based on the same working condition counterfactual mirror baseline and the set of check items, perform disturbance suppression diversion and deviation retention registration to generate an abnormal candidate item set; Furthermore, focusing on the baseline positions in the counterfactual mirror baseline under the same working conditions and the corresponding positions in the set of check items, the content of the registered deviation markers is arranged in the order of working condition principal position, unified time marker, and hierarchical spatial principal position. It is verified whether the deviation content corresponds to immunization operations, herding changes, feeding adjustments, or short-term environmental fluctuations. For content that occurs close to the time of the disturbance event, has a short duration, and does not continue along the adjacent time window, disturbance suppression and diversion are performed, and the content is included in the suppression position to retain the check trace. For content that has a persistent deviation, continues across adjacent time windows, or recurs at the same landing point, deviation retention registration is performed, and the deviation starting point, deviation continuation position, and corresponding landing point are recorded. The entries are then aggregated according to the unified time marker and matching order to generate an abnormal candidate item set.

[0040] Figure 2 The horizontal axis represents the unified time marker, and the vertical axis represents the normalized score. The four curves correspond to the actual disturbance intensity, the warning score of the invention scheme, the warning score of control scheme A, and the warning score of control scheme B, respectively. Expanding along the unified time marker, it can be seen that during the period of concentrated short-term disturbance, the actual disturbance intensity forms a significant peak. The warning scores of control scheme A and control scheme B rise synchronously with the disturbance, with continuous high-level fluctuations in some local periods. The warning score of the invention scheme remains at a low level with slow changes, showing a limited rise only during the period of the disturbance center, and then rapidly falling back in subsequent time positions. After verifying the counterfactual mirror baseline of the same working condition with the real-time evidence segment, the disturbance suppression diversion and deviation retention registration are then performed. This can suppress local deviations that are close to immunization operations, group changes, feeding adjustments, or short-term environmental fluctuations, and retain only the deviations that persist, continue across adjacent time windows, or recur at the same point in the abnormal candidate item set, thereby reducing the influence of noise disturbance on the warning judgment.

[0041] Among them, the actual disturbance intensity represents the input intensity of the short-term disturbance itself in the simulation scenario. Unlike the three warning scores, it is the amount of external interference rather than the output of the scheme.

[0042] Invention scheme warning score: It represents the anomaly judgment strength obtained after the invention scheme has undergone the same working condition counterfactual mirror baseline, close verification and disturbance suppression diversion. Unlike the warning scores of the two comparison schemes, it emphasizes the preservation of the real anomaly after suppressing short-term disturbances.

[0043] The warning score for scheme A represents the abnormal warning intensity directly formed based on the magnitude of symptom changes and local deviations under the current unified time marker, using only real-time evidence segments after object unification, spatial positioning, and window alignment. Because it does not involve the counterfactual mirror baseline of the same working condition and disturbance suppression diversion, it is more susceptible to short-term disturbances and thus more likely to increase.

[0044] The warning score of Comparison Scheme B indicates the intensity of the abnormal warning formed after comparing the deviation between the real-time evidence segment and the reference content based on the object uniformity, spatial positioning and window alignment, combined with the matching results of the same working conditions. Although it can better constrain local fluctuations than Comparison Scheme A, its ability to filter short-term disturbances is still weaker than the invention scheme because it does not perform complete disturbance suppression diversion and deviation retention registration.

[0045] S3: Based on the set of abnormal candidate entries, perform time-sequence concatenation and spatial adjacency connection to construct a multi-layer propagation chain diagram, and perform source region locking and expansion direction verification to generate a propagation risk map; S3.1: Perform time concatenation and folding on the abnormal candidate entry set, obtain the time continuation candidate segment, and perform layered filling of the continuation position to generate the time continuation expanded entry set; Furthermore, the abnormal candidate entries are arranged according to the subject of the aquaculture, the unified time marker, and the hierarchical spatial subject. It is checked whether the deviations in adjacent time positions can be continued in terms of the order of occurrence, the trend of symptoms, the strength of changes, and the corresponding landing point. Content that can be continuously extended along the time progression is folded into the same time succession candidate segment. Content with gaps, reversals, sudden changes in strength, or separation of position is split into another time succession candidate segment. The position of each record is verified in turn around each time succession candidate segment. The earliest occurrence is added as the succession start position, the content that continues to unfold in the previous order is added as the succession continuation position, and the content that shows enhancement, turning point, or outward shift is added as the succession change position. The entries are organized according to the unified time marker and succession order to generate a time succession expansion entry set.

[0046] S3.2: Based on time-series expansion of the item set, perform spatial adjacency hierarchical attachment and cross-regional connectivity verification to generate a multi-layer propagation chain graph; Furthermore, based on the time sequence of entries, the spatial hierarchy of each record is located in its respective column, section, passage, entrance / exit, and peripheral landing point. Records located in adjacent positions within the same column, consecutive positions within the same section, or sequential positions within the same passage are spatially adjacent and linked to the same level. Records located in different sections but connected by a connecting passage, corresponding entrance / exit, or a clear transfer landing point are linked across levels. The existence of an actual transfer path between the previous and next records, whether it passes through a clear relay position, or is interrupted by a closed position is verified around the cross-section location. Content with continuous transfer paths is considered cross-section connected and merged into the same link; content without continuous transfer paths is separated and linked according to spatial hierarchy and sequence of inheritance to generate a multi-layered propagation chain diagram.

[0047] It should be noted that content with a continuous transfer path refers to content where there is a connected location between the corresponding position of the previous record and the corresponding position of the next record that can be passed sequentially according to the actual space, so that abnormal content can be continuously transmitted along the enclosure, zone, passage, entrance and exit or the outer landing point.

[0048] Spatial hierarchy refers to the hierarchical arrangement of location content according to the order of placement from near to far, from inside to outside, or from local to peripheral. For example, the hierarchical order between enclosure location, zone location, passageway location, and peripheral landing point.

[0049] A multi-layered propagation chain diagram is a propagation path structure formed by connecting abnormal candidate items according to the order of time, spatial adjacency level, and cross-regional connectivity. It is used to represent the gradual expansion process of abnormal content among buildings, zones, channels, entrances and exits, and peripheral landing points.

[0050] S3.3: Merge the temporal succession and spatial adjacency relationships in the multi-layer propagation chain diagram in a chain, and perform source region first-time locking to generate a source region locking entry set; Furthermore, the temporal continuity and spatial adjacency relationships between records are verified one by one along each link in the multi-layer propagation chain diagram. Records with continuous unified time markers, consecutive successive successive positions, and sequentially reachable hierarchical spatial principal positions are grouped into the same chain position. Records with only temporal continuity but no spatial adjacency are separated into another list, and records with only spatial adjacency but no temporal continuity are stopped from being chained together. Around each link that has completed chain merging, each link is compared according to the order of unified time markers and the successive succession order within the chain. The content of the abnormal record that appears first in the link and has no identical position, adjacent position, or can reach the current position along the same expansion path in its previous time position is identified as a record not covered by the previous position. The corresponding position is locked as the source region's initial position, and the subsequent successive positions, terminal diffusion positions, and intra-chain order are organized and assigned to the corresponding entries to generate a source region locked entry set.

[0051] Figure 3The horizontal axis represents the accuracy rate of locking the initial position of the source region, and the vertical axis represents the lead time for the first warning. Each scatter point corresponds to a simulation batch of the invention scheme. The scatter points are generally distributed in the range of high locking accuracy and show a clear clustering trend in the upper right direction, indicating that as the accuracy rate of locking the initial position of the source region increases, the lead time for the first warning also increases. After performing chain-like merging of the temporal succession and spatial adjacency relationships in the multi-layer propagation chain diagram, the initial locking of the source region is then performed on records that appear early, are not covered by the preceding position, and can lead to the continuous extension of subsequent adjacent positions. This can identify the starting point of abnormal expansion earlier, compress the judgment range required for subsequent expansion direction link verification and effective propagation chain identification, and enable the propagation risk map to be formed faster. The more accurate the initial position of the source region is locked, the easier it is for the subsequent propagation succession position and the terminal diffusion position to be sequentially continued, and the earlier the warning triggering time can be moved, thus demonstrating the supporting role of the spatiotemporal deconstruction of the propagation path in shortening the warning response time.

[0052] It should be noted that temporal succession refers to the sequential connection between two records on a unified time marker, where the abnormal content continues, strengthens, or changes over time.

[0053] Spatial adjacency refers to the fact that the corresponding positions of two records are adjacent to each other, connected, or sequentially reachable in terms of enclosures, zones, passages, entrances, or peripheral landing points, thus possessing the conditions for spatial transmission and continuation.

[0054] S3.4: Perform extended direction link verification on the source region lock entry set, obtain valid propagation chains, reorganize abnormal chain segments, and generate a propagation risk map; Furthermore, the initial position, subsequent receiving position, and terminal diffusion position of the source area lock position entry set are sequentially connected in the chain order. The expansion direction link verification is performed to verify whether the previous position to the next position continuously moves outward along the adjacent direction of the fence, the direction of the partition connection, the direction of the channel extension, the direction of the entrance and exit transfer, or the direction of the peripheral landing point. At the same time, it is verified whether the unified time mark advances sequentially with the link advancement. Links with consistent expansion direction, continuous time advancement, and no reverse back, jump connection, or local swing are included in the effective propagation chain. Links with reversed direction, broken position, time disconnection, or terminal backflow are cut off and separated. Records that can still be connected are re-merged according to spatial hierarchy and expansion order to complete the abnormal link reorganization and generate a propagation risk map.

[0055] It should be noted that an effective propagation chain refers to a continuous abnormal link that starts from the initial position of the source region, moves outward along the spatial location, and progresses sequentially with a unified time mark, without any reversal of direction, breakage of position, jump connection, or return flow at the end.

[0056] Before abnormal chain segment reorganization, the continuity of the unified time stamp of the previous and subsequent records and the sequential connectivity of spatial positions are used as the conditions for continuation. For chain segments that exceed the continuous continuation range or have obvious spatial breaks, truncation and separation are performed (e.g., the time interval exceeds 2 time windows, or the positions before and after cross more than 1 unconnected level positions). Then, the remaining records that can still be continued are re-merged according to the expansion direction and spatial level.

[0057] S4: Perform enhanced suppression candidate determination on abnormal content in the transmission risk map and conduct parallel arbitration to generate a set of animal disease early warning entries; S4.1: Based on the propagation risk map, unfold the abnormal content one by one, and perform dual-track diversion of enhancement and suppression according to the continuity of expansion and interpretability, dividing the abnormal content into enhancement candidate positions and suppression candidate positions; Furthermore, along each effective transmission chain in the transmission risk map, abnormal content corresponding to the initial position, subsequent receiving position, and terminal diffusion position of the source region is extracted. These abnormal contents are then arranged into the same verification row according to the order within the chain. It is examined whether the abnormal contents continuously move outward between adjacent positions, whether the symptom trend continues, and whether the unified time markers advance continuously. Abnormal contents that appear continuously along the same expansion direction, maintain receiving across multiple positions, and are not interrupted midway are classified into the enhancement judgment track. At the same time, it is verified whether the abnormal contents can be explained by immunization operations, population changes, feeding adjustments, short-term environmental fluctuations, or local collection disturbances. Abnormal contents that appear in a limited location, have a short duration, do not continue to expand along the link, and have a clearly explained source are classified into the suppression track. Enhancement candidate positions and suppression candidate positions are registered according to the position within the chain and the expansion order.

[0058] It should be noted that anomalous content refers to symptom records that appear along the effective transmission chain in the transmission risk map, deviate from the counterfactual mirror baseline of the same working condition, and have the value of further expansion or need to be suppressed for verification.

[0059] An enhanced candidate position refers to the position of abnormal content that appears continuously in the same direction of expansion in an effective propagation chain, is continuously advanced with a unified time marker, has a symptom progression that is sequential and can lead to the continued expansion of subsequent positions.

[0060] Suppression candidate sites refer to locations where abnormal content is limited to a local area, lasts for a short period of time, does not continue to spread along the effective transmission chain, and can be explained by immunization operations, regrouping changes, feeding adjustments, short-term environmental fluctuations, or local collection disturbances.

[0061] S4.2: Perform cross-chain consistency verification and source persistence verification on enhancement candidate bits and suppression candidate bits respectively, and refine the abnormal content into priority enhancement bits, general enhancement bits, strong suppression bits and weak suppression bits, and generate an enhancement suppression classification table; Furthermore, enhancement and suppression candidate positions are arranged one by one according to the subject of aquaculture, unified time stamp, hierarchical spatial subject, and link position. When performing cross-chain consistency verification on enhancement candidate positions, the corresponding positions of the same abnormal content in different effective propagation chains are checked side by side to check whether the expansion direction is consistent, whether the order of appearance is connected, and whether the symptom trend is continuous. When performing source persistence verification on enhancement candidate positions, it is checked whether the abnormal content appears continuously and extends backward along the same source and adjacent time windows. Content that meets both cross-chain consistency and source persistence is designated as priority enhancement position, and content that meets only one of them is designated as general enhancement position. When performing cross-chain consistency verification on suppression candidate positions, it is checked whether the content can explain whether it is limited to a single link or local position. When performing source persistence verification on suppression candidate positions, it is checked whether the source appears briefly and then disappears or only repeats locally. Content with obvious limitation and short duration is designated as strong suppression position, and content with incomplete limitation but still not separated from the local position is designated as weak suppression position. The enhancement and suppression classification table is compiled in order of position.

[0062] S4.3: Arrange the enhanced inhibition grading table in parallel according to time order and spatial location, and perform evidence comparison and priority sorting to generate a set of arbitration candidate items; Furthermore, the priority enhancement positions, general enhancement positions, strong suppression positions, and weak suppression positions in the enhancement and suppression grading table are arranged in chronological order according to a unified time marker. Spatial positions are then mapped to specific column locations, partition locations, channel locations, and peripheral landing points according to the hierarchical spatial principal position. Contents with the same chronological order and identical or adjacent spatial positions are grouped into the same parallel position. Evidence comparison is performed item by item around each parallel position to verify the satisfaction of priority enhancement positions and strong suppression positions, and general enhancement positions and weak suppression positions in terms of extension continuity, cross-chain consistency, source persistence, and interpretability. The degree of dominance is calculated based on the strength and number of satisfying contents, and then prioritized according to the degree of dominance. Candidate anomaly entries with higher dominance are ranked first, and those with lower dominance are ranked last. Candidate anomaly entries with similar dominance that still require further arbitration are retained in parallel order, generating an arbitration candidate entry set.

[0063] It should be noted that dominance refers to the degree to which one content in the same parallel position satisfies more or less of the four criteria of extension continuity, cross-chain consistency, source persistence, and interpretability than another content. For example, if a certain priority enhancement bit satisfies extension continuity, cross-chain consistency, and source persistence, while the corresponding strong suppression bit only satisfies interpretability, then the priority enhancement bit has a higher degree of dominance.

[0064] The formula for calculating dominance is: ; in, Indicates the first position in a parallel sequence The degree of dominance of each content Indicates the first The value that a piece of content satisfies in terms of extended continuity. Indicates the first The value of content in terms of cross-chain consistency Indicates the first The content satisfies the requirement of source continuity. Indicates the first The content satisfies the interpretability requirement.

[0065] The continuity satisfaction value is derived from the verification results of whether the abnormal item continuously moves outward between adjacent positions, maintains intra-chain continuity, and is not interrupted midway. The cross-chain consistency satisfaction value is derived from the verification results of whether the expansion direction, order of appearance, and symptom progression of the abnormal item are consistent between different effective transmission chains. The source continuity satisfaction value is derived from the verification results of whether the abnormal item appears continuously and extends backward within the same source and adjacent time windows. The interpretability satisfaction value is derived from the verification results of whether the abnormal item can be explained by immunization operations, herding changes, feeding adjustments, short-term environmental fluctuations, or local collection disturbances. The above satisfaction values ​​are written into the arbitration candidate positions of the corresponding abnormal items and participate in the dominance calculation to characterize the strength of the abnormal item's dominance in parallel comparisons.

[0066] S4.4: Perform parallel arbitration on the set of candidate arbitration entries, distinguish between dominant abnormal entries and backup abnormal entries, and simultaneously classify risk levels and define the scope of spread to generate a set of animal disease early warning entries; Furthermore, the preceding, following, and parallel content are simultaneously developed around each parallel position in the arbitration candidate item set. Parallel arbitration is performed on competing anomalous content under the same unified time marker and the same spatial principal position. The dominant direction of extension continuity, cross-chain consistency, source persistence, and interpretability is determined for each item. Content with a continuously maintained dominant direction that can drive the further expansion of adjacent positions is designated as the dominant anomalous item, while content with insufficient dominant direction, still requiring tracking, or only valid in local positions is designated as the backup anomalous item. Then, the initial position of the source region corresponding to the dominant and backup anomalous items is determined. The subsequent receiving location and the final diffusion location are used to determine the actual coverage level. Content whose coverage is limited to the same pen or a locally adjacent location and has not formed a cross-area expansion is classified as low risk level. Content that has expanded across pens or zones but has not yet extended to passageways or peripheral landing points is classified as medium risk level. Content that has crossed passageways, reached entrances / exits or peripheral landing points, and is still maintaining an outward expansion trend is classified as high risk level. The transmission boundary is delineated according to the actual pen location, zone location, passageway location, entrance / exit location, and peripheral landing point location of the abnormal content, and the transmission range is marked to generate an animal disease early warning item set.

[0067] It should be noted that the propagation range refers to the spatial boundary of the enclosure location, zone location, passage location, and outer landing point location actually covered by the dominant or backup abnormal entry after it expands outward from the initial location of the source area.

[0068] Risk level refers to the degree of danger of abnormal content after parallel arbitration, based on the actual expansion range, spatial coverage level and continuous outward movement trend. It is used to distinguish whether the abnormal content is limited to a local location, has formed cross-regional expansion, or has reached a higher risk of spread, such as a channel, entrance / exit or peripheral landing point.

[0069] S5: Collect animal disease early warning items and write them back to the breeding unit, implement targeted sampling and review order arrangement for the dominant abnormal items, and continue to track and dynamically review the backup abnormal items to generate an animal disease early warning table; S5.1: Spatial write-back of the animal disease early warning item set according to breeding unit, and add risk location markers to form an early warning location item set; Furthermore, based on the identification of the breeding unit, the dominant and backup abnormal items in the animal disease early warning item set are assigned to the corresponding breeding location. According to the initial location of the source area, the subsequent receiving location, the terminal diffusion location, and the spread range, each item is mapped to the specific pen location, zone location, passage location, and peripheral landing point location. Items with different time sequences at the same location are arranged into the corresponding rows. For multiple early warning items appearing at the same location, the landing order is sorted according to risk level and spread sequence. A corresponding risk landing mark is added to each landing location, so that each early warning item corresponds to a clear spatial location and its sequence within the location, forming an early warning landing item set.

[0070] S5.2: Based on the early warning location entry set, perform directional sampling around the source area locking location, propagation receiving location and terminal affected location, and generate a dominant anomaly sampling arrangement table in combination with spatial connectivity; Furthermore, for each of the dominant anomaly items in the early warning location set, the specific location corresponding to the source area lock position, propagation receiving position, and terminal affected position is verified. The location of the source area lock position is determined as the first sampling position, the location of the propagation receiving position is arranged as the relay sampling position according to the expansion sequence, and the location of the terminal affected position is determined as the boundary sampling position. Combining the spatial connectivity between fences, zones, channels, and peripheral landing points, the reachable paths and arrival order between each sampling position are verified. Positions that can be connected by a straight path are arranged sequentially, and positions with blockages or detours are listed separately. The sampling positions are organized into a table according to the location order and path order to generate the dominant anomaly sampling arrangement table.

[0071] Figure 4The horizontal axis represents single-round sampling capability, and the vertical axis represents effective sampling hit rate. Each group of bars corresponds to the hit rate of the invented scheme, control scheme A, and control scheme B under different single-round sampling capabilities. As the single-round sampling capability gradually increases from low to high, the effective sampling hit rate of all three schemes shows an upward trend. However, the invented scheme maintains a higher hit rate at all sampling capability levels and maintains a stable advantage in the medium-to-high sampling capability range. This figure shows that by implementing targeted sampling around the source area locking position, propagation receiving position, and terminal affected position of the early warning landing item concentration, and generating a dominant anomaly sampling arrangement table based on spatial connectivity, the sampling action can prioritize covering key landing positions among the initial sampling position, relay sampling position, and boundary sampling position, reducing the dispersed deployment to non-critical positions. This allows limited sampling capability to fall more on the highly relevant positions corresponding to the dominant anomaly items, thereby improving the effective hit rate per unit sampling input and demonstrating the focus of prevention and control resource deployment and the effectiveness of path arrangement.

[0072] The invention scheme represents the result of directional sampling arrangement using source region locking position, propagation receiving position and terminal affected position. Unlike the two comparative schemes, it can concentrate the sampling points on the key landing position.

[0073] Comparison Scheme A: This represents the sampling results when the directional sampling logic is not fully utilized. The difference from the invention scheme is that the sampling points are more dispersed and the coverage efficiency of key locations is lower.

[0074] Comparison Scheme B: This represents the sampling results when the sampling logic of the invention scheme is partially optimized but not fully adopted. It differs from Comparison Scheme A in that it has a stronger hit rate, but is still weaker than the invention scheme.

[0075] It should be noted that spatial connectivity refers to the existence of path conditions between different landing locations that allow for actual passage, sequential arrival, or direct transfer, enabling the source region locking position, propagation receiving position, and terminal affected position to be arranged continuously according to spatial location.

[0076] S5.3: Perform review order arrangement on the dominant anomaly sampling arrangement table according to time order, and write the review trigger conditions to generate the dominant anomaly review order table; Furthermore, the initial sampling position, intermediate sampling position, and boundary sampling position in the dominant anomaly sampling arrangement table are arranged in order of sampling occurrence, propagation and expansion sequence, and path arrival sequence to determine the review order. The sampling position that first contacts the source region locking position is listed as the first review position, the sampling position that continues to advance along the propagation direction is listed as the subsequent review position, and the sampling position corresponding to the affected position at the end is listed as the boundary review position. Sampling positions that appear in parallel within the same time window are distinguished in order of risk level and spatial proximity. The review trigger conditions are supplemented by recording the continuous anomaly of the first review position, the same-direction extension of the subsequent review position, and the outward expansion signs of the boundary review position. The switching and stopping trigger content is supplemented simultaneously for sampling positions where the anomaly weakens, is interrupted, or changes direction. The results are organized according to the review order and corresponding position to generate the dominant anomaly review order table.

[0077] It should be noted that the review triggering condition refers to the specific basis for initiating the review action of the corresponding sampling position in the dominant anomaly sampling arrangement table. For example, when the initial sampling position is continuously abnormal, the subsequent sampling position extends in the same direction, or the boundary position shows signs of outward expansion, the review of the corresponding sequence is triggered.

[0078] S5.4: Perform continuous tracking and dynamic review of standby abnormal entries, and merge and register them with the leading abnormal review priority table to generate an animal disease early warning table; Furthermore, based on a unified time stamp and hierarchical spatial position, the records before and after the backup anomaly item are continuously checked. Items that appear continuously at the same location, have shortened intervals, or extend from a local location to adjacent locations are included in the continuous tracking sequence. Items that reappear after a short period of disappearance, have shifted location, or are close to the propagation direction of the dominant anomaly item are included in the dynamic review sequence. The tracking order and review order are then added according to the order of appearance. The continuous tracking sequence and dynamic review sequence are merged with the first review position, subsequent review position, and boundary review position in the dominant anomaly review priority table according to a unified time stamp, hierarchical spatial position, and risk level. Items that can continue the dominant anomaly item are registered in parallel, while items that remain locally independent are registered separately, generating an animal disease early warning table.

[0079] In summary, this invention achieves dynamic construction of counterfactual benchmarks by: folding under the same working condition constraint and supplementing the working condition principal position, extracting continuous normal evidence segments to generate a mirror reference chain, avoiding noise interference and reducing the false alarm rate; and transforming the abnormal item set into a multi-layer propagation chain graph by folding and connecting with time-series continuation and connecting with spatial adjacency levels, realizing the spatiotemporal deconstruction of the propagation path, shortening the early warning response time and optimizing the allocation of prevention and control resources.

[0080] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for early warning of animal diseases based on big data analysis, characterized in that, include: Collect multi-source data from inside and outside the breeding unit, unify the objects and spatial positions of data from different sources, and fill gaps and link data aligned in the same window to generate a multi-source spatiotemporal disease evidence stream; Based on multi-source spatiotemporal epidemic evidence streams, we perform matching and mirror reference extraction under the same working conditions, construct a counterfactual mirror baseline under the same working conditions, and perform close verification and suppression processing to generate a set of abnormal candidate items. Based on the set of abnormal candidate entries, time sequence and spatial adjacency are linked to construct a multi-layer propagation chain diagram, and source region locking and expansion direction verification are performed to generate a propagation risk map. Enhanced suppression candidate determination is performed on abnormal content in the transmission risk map, and parallel arbitration is conducted to generate a set of animal disease early warning entries; Animal disease early warning items are collected and written back to the breeding unit. Targeted sampling and review are carried out for the dominant abnormal items, and backup abnormal items are continuously tracked and dynamically reviewed to generate an animal disease early warning table.

2. The animal disease early warning method based on big data analysis as described in claim 1, characterized in that, The steps for performing object unification and spatial positioning on data from different sources are as follows: Collect data from multiple sources inside and outside the breeding unit, perform subject extraction and heteronym unification of breeding objects, and generate a unified base table of objects; Based on the unified collection base table of objects, a unified time stamp conversion, hierarchical spatial principal position selection, and internal and external evidence folding are performed to generate a window-aligned base table.

3. The animal disease early warning method based on big data analysis as described in claim 2, characterized in that, The steps for generating the multi-source spatiotemporal epidemic evidence stream are as follows: Perform time-series gap checking, neighbor mirroring completion, and outer reference completion on the peer alignment base table to generate peer alignment data; Cross-source symptom chain merging, sequential writing of related data, and spatial reachability connection are performed on the data aligned with the same window to generate a multi-source spatiotemporal epidemic evidence stream.

4. The animal disease early warning method based on big data analysis as described in claim 3, characterized in that, The steps for constructing a counterfactual mirror baseline under the same operating conditions are as follows: Based on the multi-source spatiotemporal epidemic evidence flow, we perform same-condition constraint folding and condition principal position supplementation to generate a same-condition matching base table; Extract continuous and stable normal evidence segments from the same working condition matching base table, and perform head-to-tail concatenation to obtain the mirror reference entry chain. At the same time, perform reference time sequence resetting to generate the same working condition counterfactual mirror baseline.

5. The animal disease early warning method based on big data analysis as described in claim 1, characterized in that, The steps for generating the set of candidate abnormal entries are as follows: The baseline of the counterfactual mirror under the same working condition is matched with the real-time evidence segment for verification, and a set of matching verification items is generated. Based on the counterfactual mirror baseline under the same operating conditions and the set of items for close verification, disturbance suppression diversion and deviation retention registration are performed to generate a set of candidate items for anomalies.

6. The animal disease early warning method based on big data analysis as described in claim 5, characterized in that, The steps for constructing the multi-layer propagation chain diagram are as follows: Perform time-sequence folding and concatenation on the abnormal candidate entry set to obtain time-sequence candidate segments, and perform layered supplementation of the succession position to generate a time-sequence expanded entry set; Based on the time-series expansion of the entry set, spatial adjacency hierarchy is performed and cross-regional connectivity is verified, generating a multi-layered propagation chain graph.

7. The animal disease early warning method based on big data analysis as described in claim 6, characterized in that, The steps for generating the propagation risk map are as follows: The temporal succession relationship and spatial adjacency relationship in the multi-layer propagation chain diagram are chained together, and the source region first-order locking is performed to generate a source region locking entry set; Perform extended direction link verification on the source region lock entry set to obtain valid propagation chains, and reorganize abnormal chain segments to generate a propagation risk map.

8. The animal disease early warning method based on big data analysis as described in claim 7, characterized in that, The steps for performing enhanced suppression candidate determination on abnormal content in the propagation risk map are as follows: Based on the propagation risk map, abnormal content is expanded item by item, and a dual-track diversion of enhancement and suppression is performed according to the continuity of expansion and interpretability, dividing abnormal content into enhancement candidate positions and suppression candidate positions; Cross-chain consistency verification and source persistence verification are performed on enhancement candidate bits and suppression candidate bits respectively, and abnormal content is refined into priority enhancement bits, general enhancement bits, strong suppression bits and weak suppression bits to generate an enhancement suppression classification table.

9. The animal disease early warning method based on big data analysis as described in claim 1 or 8, characterized in that, The steps for generating the animal disease early warning item set are as follows: The enhanced inhibition grading table is arranged in parallel according to time order and spatial location, and evidence comparison and priority sorting are performed to generate a set of arbitration candidate items; Parallel arbitration is performed on the set of candidate arbitration entries, and the dominant abnormal entries and backup abnormal entries are distinguished. At the same time, risk level classification and spread range marking are carried out to generate a set of animal disease early warning entries.

10. The animal disease early warning method based on big data analysis as described in claim 9, characterized in that, The steps for generating the animal disease early warning table are as follows: The animal disease early warning item set is spatially written back according to the breeding unit, and risk location markers are added to form an early warning location item set; Based on the early warning location set, targeted sampling is performed around the source area locking location, propagation receiving location, and terminal affected location, and combined with spatial connectivity, a dominant anomaly sampling arrangement table is generated; The dominant anomaly sampling arrangement table is sorted by time order for review and the review trigger conditions are added to generate the dominant anomaly review order table. Continue to track and dynamically review the backup abnormal entries, and merge and register them with the primary abnormal review priority list to generate an animal disease early warning table.

Images