Grain depot safety prediction method and system based on edge computing and time series analysis

Abstract

The application provides a grain depot safety prediction method and system based on edge computing and time series analysis, which comprises the following steps: determining an abnormal sub-region based on the time series relationship between the grain depot environment state sequence and the equipment operation state sequence; performing abnormal evolution based on the space-time distribution characteristics of the abnormal monitoring points in each abnormal sub-region to obtain an abnormal space-time evolution track; performing diffusion prediction based on the abnormal propagation evolution trend of the abnormal space-time evolution track to obtain an abnormal risk propagation chain; coupling space and time based on the spatial topological distance between the abnormal sub-regions and the transmission direction of the abnormal risk propagation chain to obtain a correlated abnormal region pair; performing topological reorganization on the abnormal sub-regions that exist interactive influence based on the abnormal propagation evolution trend of the correlated abnormal region pair to obtain an abnormal risk connected domain; and extending the space-time boundary of the abnormal risk connected domain based on the spatial extension path of each abnormal risk propagation chain to obtain an abnormal risk region. The application improves the pertinence of the grain depot emergency disposal.

Description

Technical Field

[0001] This invention relates to the field of computer technology, and in particular to a method and system for predicting grain depot safety based on edge computing and time series analysis. Background Technology

[0002] Grain depots are core facilities for the safe storage and transfer of grain. Their internal environment is characterized by strong enclosure, complex spatial structure, and significant dynamic changes. During grain storage, factors such as temperature and humidity fluctuations, dust accumulation, grain mold growth, and pest infestation interact and can easily lead to safety accidents.

[0003] Currently, grain depot safety monitoring mainly relies on distributed monitoring networks to collect environmental parameters and transmit the data to a central platform for centralized processing. Existing mainstream methods typically employ threshold alarm mechanisms or static statistical models based on historical data to identify anomalies. For example, when the temperature and humidity at a certain point exceed a preset safety threshold, it is determined that there is a safety hazard at that point. Some solutions introduce simple spatial interpolation algorithms to estimate the overall condition based on data from adjacent monitoring points.

[0004] However, existing methods lack fine-grained spatiotemporal correlation analysis of the dynamic evolution of anomalies, resulting in coarse-grained anomaly localization and a high false alarm rate. Specifically, existing methods often treat each monitoring point as an isolated entity or perform simple linear spatial correlations, ignoring the nonlinear propagation characteristics of disasters such as grain heating and mold at the microscale, as well as the coupled impact of equipment operating status on environmental changes. This coarse-grained approach cannot distinguish between a transient malfunction of a local sensor and the spread of a real internal problem within the grain, nor can it accurately define the specific boundaries and diffusion direction of the anomaly source, making it difficult to accurately locate high-risk areas in the early stages. Summary of the Invention

[0005] This invention provides a grain depot safety prediction method and system based on edge computing and time series analysis. It aims to solve the boundary ambiguity problem caused by the rough positioning of abnormal areas and the lack of dynamic evolution analysis in the background technology, so as to achieve accurate definition of the spatial range of grain depot safety hazards, realize precise control from "post-event alarm" to "pre-event warning", and thus improve the pertinence of grain depot emergency response.

[0006] In a first aspect, the present invention provides a grain depot safety prediction method based on edge computing and time series analysis, including: Based on the temporal relationship between the grain depot environmental state sequence and the equipment operation state sequence on the edge side, the grain depot is located and partitioned into regions to obtain abnormal sub-regions; Anomaly evolution is performed based on the spatiotemporal distribution characteristics of anomaly monitoring points within each anomaly sub-region to obtain anomaly spatiotemporal evolution trajectory. Furthermore, diffusion prediction is performed based on the anomaly propagation trend of the anomaly spatiotemporal evolution trajectory within each anomaly sub-region to obtain the anomaly risk propagation chain within each anomaly sub-region. Spatiotemporal coupling is performed based on the spatial topological distance between abnormal sub-regions and the transmission direction of their respective abnormal risk propagation chains to obtain associated abnormal region pairs. Based on the abnormal propagation evolution of the associated abnormal region pairs, the abnormal sub-regions with interactive influences are topologically reorganized to obtain the abnormal risk connected domain. Based on the spatial extension path of the abnormal risk propagation chain of each abnormal sub-region in the abnormal risk connected domain, the spatiotemporal boundary of the abnormal risk connected domain is extended to obtain the abnormal risk region.

[0007] Secondly, the present invention also provides a grain depot safety prediction system based on edge computing and time series analysis, for implementing the grain depot safety prediction method based on edge computing and time series analysis as described in the first aspect; the grain depot safety prediction system based on edge computing and time series analysis includes: The regional positioning and partitioning module is used to perform regional positioning and regional partitioning of grain depots based on the temporal relationship between the grain depot environmental state sequence and the equipment operation state sequence on the edge side, and to obtain abnormal sub-regions; The anomaly propagation prediction module is used to perform anomaly evolution based on the spatiotemporal distribution characteristics of anomaly monitoring points in each anomaly sub-region, obtain the anomaly spatiotemporal evolution trajectory, and perform propagation prediction based on the anomaly propagation trend of the anomaly spatiotemporal evolution trajectory in each anomaly sub-region, thereby obtaining the anomaly risk propagation chain in each anomaly sub-region. The abnormal spatiotemporal coupling module is used to perform spatiotemporal coupling based on the spatial topological distance between abnormal sub-regions and the transmission direction of their respective abnormal risk propagation chains to obtain associated abnormal region pairs. Based on the abnormal propagation evolution of the associated abnormal region pairs, the abnormal sub-regions with interactive influences are topologically reorganized to obtain abnormal risk connected domains. The risk area positioning module is used to extend the spatiotemporal boundary of the abnormal risk connected domain based on the spatial extension path of the abnormal risk propagation chain of each abnormal sub-region in the abnormal risk connected domain, so as to obtain the abnormal risk area.

[0008] Thirdly, the present invention also provides an electronic device, comprising: a memory for storing computer software programs; and a processor for reading and executing the computer software programs, thereby realizing the grain depot safety prediction method based on edge computing and time series analysis as described above.

[0009] Fourthly, the present invention also provides a non-transitory computer-readable storage medium storing a computer software program, which, when executed by a processor, implements the grain depot safety prediction method based on edge computing and time-series analysis as described above.

[0010] Fifthly, the present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the above-mentioned grain depot safety prediction method based on edge computing and time series analysis.

[0011] The grain depot safety prediction method based on edge computing and temporal analysis provided in this invention locates and partitions the grain depot by temporal correlation between the environmental state sequence and equipment operation state sequence on the edge side, obtaining abnormal sub-regions. This achieves preliminary and accurate division of abnormal regions, breaking the limitation of existing methods that treat each monitoring point as an isolated individual. By analyzing the spatiotemporal distribution characteristics of abnormal monitoring points within each abnormal sub-region, an abnormal spatiotemporal evolution trajectory is obtained. Based on the propagation trend of this trajectory, diffusion prediction is performed to obtain the abnormal risk propagation chain for each abnormal sub-region. This captures the nonlinear propagation characteristics of disasters such as grain heating and mold, as well as the coupled influence of equipment operation status on environmental changes, solving the problem of existing methods lacking fine-grained analysis of abnormal dynamic evolution. By using the spatial topological distance and propagation chain transmission direction between abnormal sub-regions for spatiotemporal coupling, associated abnormal region pairs are obtained. Furthermore, topological reorganization is performed on abnormal sub-regions with interactive influences to obtain abnormal risk connected domains, achieving the association and integration of abnormal regions and avoiding one-sided judgments on isolated abnormal points. By extending the spatiotemporal boundaries according to the spatial extension path of the abnormal risk propagation chain in each abnormal sub-region, the abnormal risk area is obtained, which accurately defines the specific boundary and diffusion direction of the abnormal source, distinguishes between the instantaneous failure of local sensors and the actual internal spread of hidden dangers in grain, and solves the boundary ambiguity problem caused by the rough positioning of abnormal areas and the lack of dynamic evolution analysis in the background technology. It realizes the accurate definition of the spatial range of safety hazards in grain depots, and realizes the precise control from "post-event alarm" to "pre-event warning", thereby improving the pertinence of emergency response in grain depots. Attached Figure Description

[0012] Figure 1 This is a flowchart illustrating the grain depot safety prediction method based on edge computing and time series analysis provided in this embodiment of the invention. Figure 2 This is a schematic diagram of the structure of the grain depot safety prediction system based on edge computing and time series analysis provided in an embodiment of the present invention; Figure 3 An embodiment diagram of the electronic device provided in this invention; Figure 4 An embodiment diagram of a computer-readable storage medium provided in accordance with the present invention. Detailed Implementation

[0013] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0014] Optionally, see Figure 1 , Figure 1 This is a flowchart illustrating the grain depot safety prediction method based on edge computing and time series analysis provided by the present invention. In this embodiment of the invention, the execution entity of the grain depot safety prediction method based on edge computing and time series analysis is the safety prediction system. Therefore, the method of this embodiment includes: Step 10: Based on the temporal relationship between the grain depot environmental state sequence and the equipment operation state sequence on the edge side, the grain depot is located and partitioned into regions to obtain abnormal sub-regions.

[0015] Optionally, the safety prediction system acquires the grain depot environmental state sequence and equipment operation state sequence collected from the edge side (computing nodes deployed at the grain depot site, close to the monitoring equipment). The grain depot environmental state sequence is a set of environmental parameters from various monitoring points arranged in chronological order. This set of environmental parameters includes, but is not limited to, temperature, humidity, and dust concentration values ​​collected by industrial sensors, reflecting the changes in grain depot environmental parameters over time. The equipment operation state sequence is a set of operating parameters from various types of equipment arranged in chronological order. This set of operating parameters includes, but is not limited to, the on / off status, operating power values, and fan speed values ​​recorded by equipment such as ventilation, refrigeration, or dehumidification equipment within the grain depot, reflecting the changes in equipment operation status over time.

[0016] The safety prediction system analyzes the temporal relationship between the environmental state sequence and the equipment operation state sequence of the grain depot, that is, the correlation between the environmental state and the equipment operation state of the grain depot at the same time point, to determine the approximate area range where anomalies may exist inside the grain depot and complete the area positioning.

[0017] The safety prediction system divides the approximate abnormal area into multiple smaller sub-regions according to the spatial layout of the grain depot, the distribution of monitoring points, and the grain stacking situation. Each sub-region contains at least one monitoring point. The system then filters out abnormal sub-regions that may have safety anomalies, as described in steps 101 to 104. Each abnormal sub-region contains at least one abnormal monitoring point, which is a monitoring point where the data exceeds the threshold or deviates from historical patterns.

[0018] Step 20: Based on the spatiotemporal distribution characteristics of abnormal monitoring points in each abnormal sub-region, perform abnormal evolution to obtain the abnormal spatiotemporal evolution trajectory, and perform diffusion prediction based on the abnormal propagation evolution trend of the abnormal spatiotemporal evolution trajectory in each abnormal sub-region to obtain the abnormal risk propagation chain in each abnormal sub-region.

[0019] Optionally, for each abnormal sub-region, the safety prediction system extracts the spatiotemporal distribution characteristics of all abnormal monitoring points within the region. These spatiotemporal distribution characteristics include, but are not limited to, spatial distribution location and density, as well as temporal occurrence order, duration, and trend of change. In other words, it clarifies the specific location of each abnormal monitoring point within the sub-region, the number of abnormal monitoring points per unit space, the order in which each monitoring point first exhibits an anomaly, the duration of the anomaly, and the direction and rate of change of the abnormal data.

[0020] The safety prediction system performs anomaly evolution analysis based on spatiotemporal distribution characteristics. The specific analysis of anomaly evolution analysis is as follows: taking the time sequence of anomaly monitoring points as a clue, combined with the spatial distribution location, it simulates the complete process of anomaly from its initial generation to its gradual development, analyzes the changing patterns of anomaly at different time nodes and different spatial locations, and reconstructs the development history of anomaly. During this process, the spatial location and anomaly degree corresponding to each time node are recorded simultaneously to obtain the spatiotemporal evolution trajectory of anomaly that reflects the location of anomaly generation, propagation path, development speed, and changing trend.

[0021] The safety prediction system analyzes the spatiotemporal evolution trajectory of anomalies to determine the possible future propagation direction, speed, and range of anomalies. The propagation range is the trend of anomaly propagation evolution. Based on the trend of anomaly propagation evolution, the system predicts the future spread of anomalies, clarifies the order of anomalies spreading from their source, the location of each diffusion node, and the degree of anomaly, and obtains the anomaly risk propagation chain for each anomaly sub-region, as shown in steps 201 to 204.

[0022] Step 30: Based on the spatial topological distance between abnormal sub-regions and the transmission direction of their respective abnormal risk propagation chains, spatiotemporal coupling is performed to obtain associated abnormal region pairs. Based on the abnormal propagation evolution of the associated abnormal region pairs, the abnormal sub-regions with interactive influences are topologically reorganized to obtain abnormal risk connected domains.

[0023] Optionally, the safety prediction system calculates the spatial topological distance between each abnormal sub-region, where the spatial topological distance is the straight-line distance between the center of the grain depot's spatial topology. At the same time, it extracts the transmission direction of the abnormal risk propagation chain in each abnormal sub-region, where the transmission direction is the direction of abnormal diffusion, reflecting the propagation trend of the abnormal sub-region.

[0024] The security prediction system combines spatial topological distance and transmission direction to perform spatiotemporal coupling analysis to obtain pairs of associated abnormal regions, as detailed in steps 301 to 303.

[0025] The security prediction system analyzes the abnormal propagation and evolution of each pair of related regions, that is, it clarifies the abnormal propagation speed, abnormal propagation range, abnormal propagation degree of each of the two abnormal sub-regions, as well as the interaction trend between them, and screens out abnormal sub-regions with interactive influence. The existence of interactive influence means that the abnormal spread of one abnormal sub-region will affect the abnormal development of another abnormal sub-region, and vice versa.

[0026] For these anomalous sub-regions with interactive influences, the original boundaries are broken, and topological reorganization is carried out in combination with the anomalous propagation and evolution trend. The spatial range of each anomalous sub-region is integrated to ensure that the anomalous regions in the reorganized region can propagate and be related to each other, thus obtaining the anomalous risk connected domain. Therefore, the anomalous risk connected domain can be understood as a connected whole domain composed of multiple interactive anomalous sub-regions.

[0027] Step 40: Extend the spatiotemporal boundary of the abnormal risk connected domain based on the spatial extension path of the abnormal risk propagation chain of each abnormal sub-region in the abnormal risk connected domain to obtain the abnormal risk region.

[0028] Optionally, the safety prediction system extracts the spatial extension path of the abnormal risk propagation chain in each abnormal sub-region of the abnormal risk connectivity domain. The spatial extension path is a spatial path that, combined with the abnormal propagation evolution trend, clearly indicates the future spread of the abnormal from the abnormal risk connectivity domain to the surrounding normal area.

[0029] The safety prediction system, based on spatial extension paths, considers key factors such as the propagation speed, propagation time, and severity of anomalies to extend the boundaries of the connected domains of anomaly risk in both time and space. Temporally, it predicts the potential spread of anomalies at different time points based on the propagation speed. Spatially, according to the spatial extension path, it gradually incorporates surrounding normal areas that the anomaly might cover in the future, achieving precise boundary extension. After the extension operation, the connected domains of anomaly risk and the extended surrounding areas are integrated to obtain the anomaly risk area, as detailed in steps 401 to 404.

[0030] The embodiments of this invention accurately define the specific boundaries and diffusion direction of the anomaly source, distinguish between the instantaneous failure of a local sensor and the actual diffusion of hidden dangers inside the grain, solve the problem of boundary ambiguity caused by the rough positioning of the anomaly area and the lack of dynamic evolution analysis in the background technology, realize the accurate definition of the spatial range of safety hazards in grain depots, realize the precise control from "post-event alarm" to "pre-event warning", and thus improve the pertinence of emergency response in grain depots.

[0031] Optionally, the process of steps 101 to 104 includes: Step 101: Construct a state association map based on the time-series synchronization mapping relationship between the grain depot environmental state sequence and the equipment operation state sequence.

[0032] Optionally, the safety prediction system performs time-series synchronization processing on the grain depot environmental state sequence and the equipment operation state sequence, clarifying the time-series synchronization mapping relationship. This time-series synchronization mapping relationship means matching the grain depot environmental state sequence and the equipment operation state sequence according to the same time node, ensuring a one-to-one correspondence between environmental parameters and equipment operation parameters at the same time node, eliminating misalignment problems caused by differences in collection time between the two sequences. After completing time-series synchronization, the safety prediction system uses each time node as a basis, treating the corresponding environmental parameters and equipment operation parameters as associated nodes, and the mutual influence relationships between nodes as associated edges, constructing a state association graph. Therefore, the state association graph presents the correlation between the grain depot environmental state and the equipment operation state at different time nodes, as well as the interaction patterns between various parameters.

[0033] In one embodiment, environmental status data and equipment operation status data are collected every hour at the edge of the grain depot for 6 consecutive hours, resulting in an environmental status sequence (including temperature and humidity data at each monitoring point) and an equipment operation status sequence (including operating parameters of ventilation and refrigeration equipment). The safety prediction system synchronizes and matches the two sequences according to time nodes, determining the time-series synchronization mapping relationship as follows: environmental data of the first hour corresponds to equipment operation data of the first hour, environmental data of the second hour corresponds to equipment operation data of the second hour, and so on. Subsequently, taking each hour as a time node, the temperature and humidity data of that hour are associated with the ventilation equipment speed and refrigeration equipment power as association nodes. If an increase in the ventilation equipment speed leads to a decrease in humidity, an association edge is established between the ventilation equipment speed node and the humidity node, constructing a state association graph containing 6 time nodes, each node corresponding to multiple parameters and association relationships.

[0034] Step 102: Determine the abnormal monitoring points based on the temporal deviation of the state of each abnormal monitoring point in the state association graph.

[0035] Optionally, the safety prediction system extracts the state data of each monitoring point at various time nodes from the state correlation graph. The state data of a monitoring point refers to the environmental parameter data corresponding to the monitoring point. Combining this data with the normal temporal correlation pattern of the monitoring point in the state correlation graph, the system calculates the temporal deviation degree of the state of each monitoring point. The temporal deviation degree refers to the degree of deviation between the state data of the monitoring point at the current time node and the standard state data of the corresponding time node under the normal correlation pattern. The greater the deviation degree, the more obvious the temporal anomaly of the monitoring point. The preset temporal deviation threshold is the maximum allowable deviation value determined based on the grain depot safety storage standards and the historical normal operation data of the monitoring points. When the temporal deviation degree of a certain monitoring point exceeds the preset temporal deviation threshold, the safety prediction system determines that the monitoring point is an abnormal monitoring point.

[0036] Continuing with the embodiment of step 101, temperature data for six time points of a monitoring point (responsible for collecting temperature) are extracted from the state correlation graph: 23℃, 24℃, 28℃, 29℃, 30℃, and 31℃. Based on the normal temporal correlation pattern of this monitoring point in the state correlation graph, its normal temperature range is determined to be 23℃-25℃, with a preset temporal deviation threshold of 2℃. The temporal deviation degree of each time point of this monitoring point is calculated: deviation of 0℃ in the 1st hour, 0℃ in the 2nd hour, 3℃ in the 3rd hour, 4℃ in the 4th hour, 5℃ in the 5th hour, and 6℃ in the 6th hour. The temporal deviation degrees from the 3rd to the 6th hour all exceed the preset threshold; therefore, the safety prediction system determines this monitoring point to be an abnormal monitoring point.

[0037] Step 103: Cluster the monitoring points based on the spatial adjacency relationship of the abnormal monitoring points and the duration of the abnormal state in the time series to obtain the range of the target abnormal area.

[0038] Optionally, spatial adjacency refers to the spatial location association between anomaly monitoring points; that is, when the distance between two anomaly monitoring points is less than a preset spatial distance threshold, they are determined to have a spatial adjacency relationship. Anomaly duration refers to the length of time an anomaly monitoring point is in a temporally abnormal state. The security prediction system clusters monitoring points based on their spatial adjacency and the duration of the temporally abnormal state to obtain the target anomaly area range, as detailed in steps 1031 to 1034.

[0039] Step 104: Based on the consistency of the temporal change trend of monitoring points within the target anomaly area, the target anomaly area is divided into regional partitions to obtain anomaly sub-regions.

[0040] Optionally, the safety prediction system extracts the temporal change trend of all monitoring points (including abnormal monitoring points and normal monitoring points that are not judged as abnormal but are located within the range) within the target abnormal area. The temporal change trend refers to the direction and rate of change of the status data of each monitoring point over time.

[0041] The safety prediction system analyzes the consistency of the temporal change trends of various monitoring points within the target anomaly area. The consistency criterion is that the change direction of the state data of each monitoring point is the same, and the deviation of the change rate is within a preset deviation range. Based on this consistency criterion, the safety prediction system divides the target anomaly area into multiple sub-regions. Within each sub-region, the temporal change trends of all monitoring points are consistent, and there are significant differences in the temporal change trends of monitoring points between different sub-regions. These divided sub-regions are the anomaly sub-regions.

[0042] In one embodiment, suppose the target abnormal area obtained in step 103 covers the entire area of ​​warehouse No. 3 in the grain depot, including 8 monitoring points, of which 5 are abnormal monitoring points determined in step 102 and 3 are normal monitoring points. Temperature change data from these 8 monitoring points over a continuous 6-hour period are extracted, and their temporal trends are analyzed: the temperatures of 4 monitoring points (3 abnormal, 1 normal) show a continuous upward trend, with a change rate of approximately 1°C per hour, exhibiting a consistent temporal trend. The temperatures of the other 4 monitoring points (2 abnormal, 2 normal) show a fluctuating upward trend, with a change rate between 0.5°C and 0.8°C per hour, also exhibiting a consistent temporal trend, but significantly different from the trends of the first 4 monitoring points. Therefore, the safety prediction system divides this target abnormal area into two abnormal sub-regions: one containing the 4 monitoring points with a continuously rising temporal trend, and the other containing the 4 monitoring points with a fluctuating upward temporal trend.

[0043] The embodiments of the present invention enable precise location and zoning of abnormal areas in grain depots, resulting in abnormal sub-regions with clear boundaries and consistent internal conditions. This allows for precise definition of the spatial scope of potential safety hazards in grain depots and improves the targeted nature of emergency response.

[0044] Optionally, the process of steps 1031 to 1034 includes: Step 1031: Based on the spatial coordinates of the abnormal monitoring points in the three-dimensional space of the grain depot and the heat conduction radius of the grain pile medium, a neighbor connectivity determination is made to obtain connected monitoring point pairs. Based on the abnormal duration of each abnormal monitoring point in the connected monitoring point pair and the minimum steady-state time threshold for disaster formation, the target monitoring point is determined.

[0045] Optionally, the safety prediction system extracts the spatial coordinates of each abnormal monitoring point in the three-dimensional space of the grain depot. The spatial coordinates refer to the specific location of the abnormal monitoring point in the three-dimensional space inside the grain depot, which can accurately identify the spatial location of the monitoring point. At the same time, the system obtains the thermal conductivity radius of the grain pile medium. The thermal conductivity radius of the grain pile medium refers to the maximum distance that heat can be naturally conducted inside the grain pile, which is calculated and determined based on the actual storage conditions such as the type of grain, stacking density, and moisture content.

[0046] The safety prediction system calculates the straight-line distance between any two anomaly monitoring points based on their three-dimensional spatial coordinates. This straight-line distance is then compared to the thermal conductivity radius of the grain pile medium to determine proximity connectivity. Specifically, if the straight-line distance between two anomaly monitoring points is less than or equal to the thermal conductivity radius of the grain pile medium, they are considered to be adjacent and connected, forming a connected monitoring point pair. If the straight-line distance is greater than the thermal conductivity radius of the grain pile medium, they are not considered to be adjacent and not connected, thus not forming a connected monitoring point pair.

[0047] The minimum steady-state time threshold for disaster formation refers to the shortest time required for disasters such as grain heating and mold to form and stabilize, calculated based on grain depot safe storage experience and disaster evolution patterns. The safety prediction system extracts the duration of anomalies from each abnormal monitoring point in each connected monitoring point pair. The duration of anomalies refers to the length of time an abnormal monitoring point is in a temporally abnormal state. The system compares the duration of anomalies at each abnormal monitoring point in each connected monitoring point pair with the minimum steady-state time threshold for disaster formation. If the duration of anomalies at a particular monitoring point is greater than or equal to this minimum steady-state time threshold, then that monitoring point is determined to be a target monitoring point. If the duration of anomalies is less than this minimum steady-state time threshold, then it is not determined to be a target monitoring point. Finally, all target monitoring points are selected.

[0048] Continuing with the above embodiment, assuming there are 5 abnormal monitoring points determined in step 102, the three-dimensional spatial coordinates of the grain depot for each abnormal monitoring point are extracted, and the heat conduction radius of the grain pile medium is calculated to be 8 meters. The straight-line distance between any two abnormal monitoring points is calculated. For example, if the distance between monitoring point 1 and monitoring point 2 is 6 meters, the distance between monitoring point 2 and monitoring point 3 is 7 meters, and the distance between monitoring point 4 and monitoring point 5 is 9 meters, and the distance between the remaining monitoring points is greater than 8 meters, then the connected monitoring point pairs are (monitoring point 1, monitoring point 2) and (monitoring point 2, monitoring point 3). The minimum steady-state time threshold is 4 hours. The abnormal duration of the abnormal monitoring points in each connected monitoring point pair is extracted: monitoring point 1 is 5 hours, monitoring point 2 is 6 hours, monitoring point 3 is 3 hours, monitoring point 4 is 5 hours, and monitoring point 5 is 4 hours. Among them, the abnormal duration of monitoring points 1, 2, 4, and 5 is greater than or equal to 4 hours, therefore these 4 monitoring points are determined to be target monitoring points.

[0049] Step 1032: Based on the spatiotemporal intersection matching of the target monitoring point and the connected monitoring point pair, the target connected unit is obtained, and based on the spatial sequence continuity of the monitoring points in the target connected unit, the topological link is traced to obtain the linear anomaly chain reflecting the local diffusion path of the disaster.

[0050] Optionally, spatiotemporal intersection matching refers to associating target monitoring points with connected monitoring point pairs. Therefore, the specific process of the safety prediction system performing spatiotemporal intersection matching based on all target monitoring points and connected monitoring point pairs is as follows: Connected monitoring point pairs containing target monitoring points are selected, and these pairs are taken as target connected units, ensuring that each target connected unit contains at least one target monitoring point. Connected monitoring point pairs that do not contain any target monitoring points are excluded. The safety prediction system extracts the spatial sequence continuity of all monitoring points in each target connected unit. Spatial sequence continuity means that after the spatial positions of each monitoring point in the target connected unit are arranged in a certain order, adjacent monitoring points satisfy the proximity connectivity relationship, and the whole forms a continuous spatial distribution. Based on the spatial sequence continuity, the safety prediction system performs topological link tracing. Topological link tracing refers to starting from any target monitoring point in the target connected unit and sequentially associating adjacent monitoring points according to the spatial sequence continuity to form a continuous monitoring point link. The monitoring point link can reflect the diffusion path of disasters such as grain heating and mold in a local area. The obtained continuous monitoring point link is a linear anomaly chain reflecting the local diffusion path of the disaster.

[0051] Continuing from step 1031, the target monitoring points obtained are monitoring point 1, monitoring point 2, monitoring point 4, and monitoring point 5. The connected monitoring point pairs are (monitoring point 1, monitoring point 2), (monitoring point 2, monitoring point 3), and (monitoring point 4, monitoring point 5). Spatiotemporal intersection matching filters out the connected monitoring point pairs containing the target monitoring points, namely (monitoring point 1, monitoring point 2) and (monitoring point 4, monitoring point 5), and these two connected monitoring point pairs are taken as target connected units. The spatial sequence continuity of each target connected unit is analyzed: In the target connected unit (monitoring point 1, monitoring point 2), the distance between monitoring point 1 and monitoring point 2 is 6 meters, satisfying the proximity connectivity relationship, and the spatial sequence is continuous. In the target connected unit (monitoring point 4, monitoring point 5), the distance between monitoring point 4 and monitoring point 5 is 9 meters, which is greater than the heat conduction radius of 8 meters, but both are target monitoring points and there are no other adjacent monitoring points, so the spatial sequence can be considered continuous. Topological link tracing is performed, starting from monitoring point 1 and associating with monitoring point 2, forming a linear anomaly chain 1 (monitoring point 1 → monitoring point 2). Starting from monitoring point 4 and linking it to monitoring point 5, a linear anomaly chain 2 is formed (monitoring point 4 → monitoring point 5).

[0052] Step 1033: Based on the deviation of the distance between adjacent monitoring points in the linear anomaly chain from the thermal conduction radius, breakpoint detection is performed to obtain discontinuous chain segments with medium obstruction or monitoring blind spots, and topological truncation is performed based on the breakpoint positions of the discontinuous chain segments to obtain the initial anomaly cluster.

[0053] Optionally, the safety prediction system extracts the distance between adjacent monitoring points in each linear anomaly chain and calculates the deviation between the distance between adjacent monitoring points and the heat conduction radius. The deviation refers to the absolute value of the difference between the distance between adjacent monitoring points and the heat conduction radius; the larger the absolute value, the weaker the heat conduction correlation between adjacent monitoring points. The safety prediction system performs breakpoint detection based on the deviation. Breakpoint detection means that when the deviation between the distance between adjacent monitoring points and the heat conduction radius exceeds a preset deviation threshold, a breakpoint is determined to exist between the adjacent monitoring points. The area corresponding to this breakpoint is a region with medium obstruction or monitoring blind zone. Medium obstruction refers to the presence of impurities, clumps, or other substances inside the grain pile that hinder heat conduction, while monitoring blind zone refers to an area inside the grain depot where no monitoring equipment is deployed and environmental parameters cannot be collected. After breakpoint detection, discontinuous chain segments with medium obstruction or monitoring blind zones are obtained, i.e., segments in the linear anomaly chain that are broken and cannot form a continuous heat conduction correlation.

[0054] The safety prediction system performs topological truncation segmentation based on the breakpoint locations of discontinuous chain segments. Topological truncation segmentation refers to cutting the linear abnormal chain at the breakpoint location and treating each discontinuous chain segment as an independent cluster, which is the initial abnormal cluster.

[0055] Continuing from step 1032, the linear abnormal chain 1 is monitoring point 1 → monitoring point 2 → monitoring point 6 (supplementary monitoring point 6 is the extension point of the original chain), and the linear abnormal chain 2 is monitoring point 4 → monitoring point 5 → monitoring point 7 (supplementary monitoring point 7 is the extension point of the original chain). The heat conduction radius of the grain pile medium is 8 meters, and the preset deviation threshold is 2 meters. The distances between adjacent monitoring points are extracted: the distance between monitoring point 1 and monitoring point 2 is 6 meters (deviation of 2 meters), the distance between monitoring point 2 and monitoring point 6 is 11 meters (deviation of 3 meters), the distance between monitoring point 4 and monitoring point 5 is 9 meters (deviation of 1 meter), and the distance between monitoring point 5 and monitoring point 7 is 12 meters (deviation of 4 meters). Breakpoint detection is performed. The deviations between monitoring point 2 and monitoring point 6, and between monitoring point 5 and monitoring point 7, both exceed the preset threshold. These two points are determined to be breakpoints, and the corresponding discontinuous chain segments are (monitoring point 1 → monitoring point 2), (monitoring point 6), (monitoring point 4 → monitoring point 5), and (monitoring point 7).

[0056] Based on the breakpoint location, topological truncation segmentation is performed, and the four non-continuous chain segments are treated as independent clusters to obtain four initial abnormal clusters: cluster 1 (monitoring point 1, monitoring point 2), cluster 2 (monitoring point 6), cluster 3 (monitoring point 4, monitoring point 5), and cluster 4 (monitoring point 7).

[0057] Step 1034: Based on the spatial inclusion relationship between the outer boundary monitoring points of the initial anomaly cluster and the remaining unaggregated anomaly monitoring points, perform edge fusion expansion to obtain the target anomaly region range.

[0058] Optionally, unaggregated anomaly monitoring points refer to those anomaly monitoring points determined in step 1031 that have not been included in any initial anomaly cluster. The security prediction system extracts the peripheral boundary monitoring points for each initial anomaly cluster. The peripheral boundary monitoring points refer to the monitoring points located on the outermost side of the initial anomaly cluster that are closest to other initial anomaly clusters or unaggregated anomaly monitoring points.

[0059] The safety prediction system calculates the spatial containment relationship between the outer boundary monitoring point of each initial anomaly cluster and the remaining unaggregated anomaly monitoring points. The spatial containment relationship refers to whether the unaggregated anomaly monitoring point is located within the spatial coverage of the initial anomaly cluster, or whether the spatial distance between the unaggregated anomaly monitoring point and the outer boundary monitoring point of the initial anomaly cluster is less than or equal to the heat conduction radius of the grain pile medium. If any of the above conditions are met, it is determined that the unaggregated anomaly monitoring point has a spatial containment relationship with the initial anomaly cluster.

[0060] Edge fusion expansion is based on spatial inclusion relationships. This means the security prediction system incorporates unaggregated anomaly monitoring points that have spatial inclusion relationships with the initial anomaly cluster into the corresponding initial anomaly cluster, expanding the spatial range of the initial anomaly cluster. All the expanded initial anomaly clusters are then integrated. If two or more initial anomaly clusters have spatial overlap or adjacent connectivity after fusion expansion, they are merged into a larger cluster. The final spatial range obtained by integrating all clusters is the target anomaly region range.

[0061] Continuing with step 1033, four initial anomaly clusters are obtained. Let the anomaly monitoring point that was not aggregated in step 1031 be monitoring point 8. Extract the outer boundary monitoring points of each initial anomaly cluster: the outer boundary monitoring point of cluster 1 is monitoring point 2, the outer boundary monitoring point of cluster 2 is monitoring point 6, the outer boundary monitoring point of cluster 3 is monitoring point 5, and the outer boundary monitoring point of cluster 4 is monitoring point 7.

[0062] The spatial distance between the outer boundary monitoring points and the unaggregated monitoring point 8 is calculated: the distance between monitoring point 2 and monitoring point 8 is 7 meters (less than the heat conduction radius of 8 meters), and the distance between the remaining outer boundary monitoring points and monitoring point 8 is greater than 8 meters. This indicates that monitoring point 8 has a spatial containment relationship with cluster 1. Monitoring point 8 is merged into cluster 1, completing the edge fusion expansion. The merged cluster 1 consists of (monitoring point 1, monitoring point 2, and monitoring point 8). Other initial anomalous clusters have no unaggregated monitoring points that meet the conditions and remain unchanged. Subsequently, all clusters are integrated, with no spatial overlap or adjacent connectivity between clusters. The final spatial range of all integrated clusters is the target anomalous area range, encompassing the entire space covered by clusters 1, 2, 3, and 4.

[0063] This invention focuses on the thermal conductivity characteristics of grain piles and the evolution law of disasters. Through multi-dimensional clustering, breakpoint detection and fusion expansion, it achieves precise positioning of abnormal areas and obtains a target abnormal area range with clear boundaries that fits the actual spread range of the disaster. This provides support for the subsequent division of abnormal sub-regions and the accurate analysis of abnormal risks in the entire grain depot, thereby improving the accuracy of abnormal monitoring in grain depots.

[0064] Optionally, the processes of steps 201 to 204 include: Step 201: For each anomalous sub-region, based on the spatial coordinates of the monitoring points corresponding to the continuous time slices in the anomalous spatiotemporal evolution trajectory, calculate the displacement vector of the centroid position between adjacent time slices, and take the propagation direction vector with the highest directional distribution frequency as the anomalous propagation direction reference, and construct a spatial search area extending along the anomalous propagation direction reference axis with the centroid position of the anomalous sub-region as the starting point.

[0065] Optionally, the security prediction system acquires the spatiotemporal evolution trajectory of each anomalous sub-region. The spatiotemporal evolution trajectory refers to the path formed by the anomalous event in the time and space dimensions during its evolution, which can clearly reflect the location of the anomalous event, its propagation path, its development speed, and its changing trend.

[0066] The safety prediction system performs time-slicing processing on the spatiotemporal evolution trajectory of anomalies. Time slicing refers to dividing the spatiotemporal evolution trajectory of anomalies into multiple consecutive time segments according to a preset time interval. Each time slice corresponds to a fixed time node and contains the spatial coordinates of all monitoring points covered by the anomaly at that time node. The safety prediction system extracts the spatial coordinates of all monitoring points corresponding to each consecutive time slice and calculates the centroid position of each time slice based on these spatial coordinates. The centroid position refers to the average position of the spatial coordinates of all monitoring points within that time slice. The calculation method is to take the arithmetic mean of the values ​​of each dimension of the spatial coordinates of all monitoring points to obtain the centroid coordinates of that time slice.

[0067] The safety prediction system calculates the displacement vector of the centroid position between two adjacent time slices. The displacement vector is the vector from the centroid position of the previous time slice to the centroid position of the next time slice, which can reflect the direction and distance of the propagation of the anomaly between two adjacent time slices.

[0068] The safety prediction system collects displacement vectors between all adjacent time slices, counts the propagation direction corresponding to each displacement vector, and calculates the distribution frequency of each propagation direction. The distribution frequency refers to the proportion of the number of displacement vectors corresponding to a certain propagation direction to the total number of all displacement vectors. The propagation direction vector with the highest distribution frequency is determined as the benchmark for the abnormal propagation direction. This benchmark can accurately reflect the main propagation direction of the anomaly.

[0069] The safety prediction system obtains the centroid position of the abnormal sub-region. The centroid position of the abnormal sub-region refers to the average position of the spatial coordinates of all monitoring points in the abnormal sub-region. Starting from the centroid position, the system extends a preset width to both sides of the abnormal propagation direction along the axis corresponding to the reference axis of the abnormal propagation direction to form a long strip-shaped region. This region is the spatial search area extending along the reference axis of the abnormal propagation direction.

[0070] In one embodiment, suppose the spatiotemporal evolution trajectory of the anomalous sub-region is divided into 5 consecutive time slices with a time interval of 1 hour, and each time slice corresponds to the anomalous state for 1 hour. The spatial coordinates of the monitoring points within each time slice are extracted, and the centroid positions of each time slice are calculated as (10, 20, 5), (12, 20, 5), (14, 21, 5), (16, 21, 5), and (18, 22, 5). The displacement vectors of the centroid positions between adjacent time slices are calculated as (2, 0, 0), (2, 1, 0), (2, 0, 0), and (2, 1, 0), respectively. The frequency distribution of the propagation direction is statistically analyzed, and the displacement vectors along the positive x-axis (horizontally eastward) have the highest frequency (4 vectors), which are determined as the reference for the anomalous propagation direction. The centroid position of this anomalous sub-region is (14, 20, 8, 5). Starting from this position, a spatial search area with a width of 6 meters is constructed by extending 3 meters to both sides along the positive x-axis.

[0071] Step 202: Search for monitoring points whose spatial coordinates are distributed within the spatial search area to obtain potential spatial impact monitoring points, and construct candidate topological paths based on the topological connection relationship between each potential spatial impact monitoring point and the boundary monitoring points of the abnormal sub-region.

[0072] Optionally, the safety prediction system searches for the spatial coordinates of all monitoring points inside the grain depot, filters out monitoring points whose spatial coordinates are distributed within the spatial search area, and identifies these monitoring points as potential spatial impact monitoring points. Potential spatial impact monitoring points refer to monitoring points that may be affected by the abnormal spread of the abnormal sub-region, and their spatial location is on the potential path of abnormal propagation.

[0073] The safety prediction system extracts the boundary monitoring points of the abnormal sub-region. The boundary monitoring points refer to the monitoring points located on the outermost side of the abnormal sub-region that are closest to the spatial potential impact monitoring points, and can directly reflect the outer boundary position of the abnormal sub-region.

[0074] The safety prediction system analyzes the topological connection between each potential spatial impact monitoring point and the boundary monitoring point of the abnormal sub-region. The topological connection refers to whether there is a continuous spatial path between two monitoring points, and whether the grain pile medium on the path can conduct abnormalities such as heat and mold. The judgment criteria are that the spatial straight distance between the two monitoring points is less than or equal to the heat conduction radius of the grain pile medium, and there is no obvious medium barrier between the two points (such as grain caking, impurity accumulation, or other substances that hinder the conduction of abnormalities).

[0075] For each potential spatial impact monitoring point, if it has a topological connection with a boundary monitoring point, the safety prediction system forms an initial path starting from that boundary monitoring point and ending at the potential spatial impact monitoring point. If the potential spatial impact monitoring point has topological connections with multiple boundary monitoring points, multiple initial paths are formed. The safety prediction system filters all initial paths, eliminating paths with breakpoints (i.e., no topological connection between two adjacent monitoring points). The remaining paths are candidate topological paths, which refer to potential paths where anomalies may propagate.

[0076] Continuing within the spatial search area constructed in step 201, four potential spatial impact monitoring points were found: Monitoring Point A, Monitoring Point B, Monitoring Point C, and Monitoring Point D. The boundary monitoring points for this anomaly sub-region are Monitoring Point 1 and Monitoring Point 2. Analysis of topological connections: Monitoring Point A is 7 meters away from Monitoring Point 1 (less than the 8-meter thermal conductivity radius of the grain pile medium, with no medium obstruction), indicating a topological connection. Monitoring Point B is 6 meters away from Monitoring Point 1 and 7.5 meters away from Monitoring Point 2, both satisfying the condition, indicating a topological connection. Monitoring Point C is 8 meters away from Monitoring Point 2, indicating a topological connection. Monitoring Point D is more than 8 meters away from both boundary monitoring points, indicating no topological connection. Based on this, initial paths are formed: Monitoring Point 1 → Monitoring Point A, Monitoring Point 1 → Monitoring Point B, Monitoring Point 2 → Monitoring Point B, and Monitoring Point 2 → Monitoring Point C. After eliminating paths without topological connections, four candidate topological paths are obtained.

[0077] Step 203: Determine the time lag value based on the first start time when each monitoring point in each candidate topology path is in an abnormal state and the second start time of the abnormal sub-region.

[0078] Optionally, the safety prediction system extracts the abnormal state information of each monitoring point in each candidate topology path, and determines the first start time of each monitoring point's abnormal state. The first start time refers to the specific time when the monitoring point first collects abnormal environmental parameter data and is identified as an abnormal monitoring point. This time is calculated from the initial time node of the abnormal spatiotemporal evolution trajectory. At the same time, the safety prediction system obtains the second start time of the abnormal sub-region. The second start time refers to the specific time when the abnormal sub-region first appears abnormal and is identified as an abnormal sub-region, that is, the time when the first monitoring point in the abnormal sub-region exhibits an anomaly.

[0079] For each candidate topology path, the security prediction system calculates the time difference between each monitoring point and the second start time of the abnormal sub-region based on the first start time of each monitoring point in the path. This time difference is the time lag of the monitoring point relative to the start of the abnormal sub-region.

[0080] The safety prediction system performs statistical analysis on the time lag of all monitoring points in each candidate topology path, and takes the average of all time lags as the time lag value of the candidate topology path. The time lag value can reflect the average time delay of an anomaly propagating from the anomaly sub-region to each monitoring point on the path.

[0081] In one embodiment, the second start time of a certain abnormal sub-region is the 2nd hour, and a candidate topology path obtained in step 202 is monitoring point 1 → monitoring point A → monitoring point B. The first start time of monitoring point 1 is the 2nd hour, the first start time of monitoring point A is the 3rd hour, and the first start time of monitoring point B is the 4th hour. The time lag of each monitoring point is calculated: the time lag of monitoring point 1 is 0 hours (from the 2nd to the 2nd hour), the time lag of monitoring point A is 1 hour (from the 3rd to the 2nd hour), and the time lag of monitoring point B is 2 hours (from the 4th to the 2nd hour). The average is calculated as (0+1+2) / 3=1 hour, meaning the time lag value of the candidate topology path is 1 hour.

[0082] Similarly, calculate the time lag values ​​for other candidate topological paths, for example, 1.2 hours, 0.8 hours, and 1.5 hours respectively.

[0083] Step 204: Based on the time lag value and total path length of each candidate topological path, perform diffusion prediction to obtain the anomalous risk propagation chain for each anomalous sub-region.

[0084] Optionally, the total path length refers to the total straight-line distance in space from the starting point (anomaly sub-region boundary monitoring point) to the ending point (spatial potential impact monitoring point) in the candidate topology path. The security prediction system performs diffusion prediction based on the time lag value of each candidate topology path and the total path length to obtain the anomaly risk propagation chain for each anomaly sub-region, as described in steps 2041 to 2044.

[0085] Based on the abnormal spatiotemporal evolution trajectory, this invention analyzes the situation from both spatial and temporal dimensions, accurately capturing the nonlinear propagation characteristics of disasters such as grain heating and mold, as well as the coupled impact of equipment operating status on environmental changes. This yields an abnormal risk propagation chain that fully reflects the abnormal diffusion pattern, providing support for the subsequent spatiotemporal coupling, topological reorganization, and precise definition of abnormal risk areas in abnormal sub-regions. This improves the accuracy of grain depot anomaly early warning, promotes the transformation of grain depot safety management from "post-event alarm" to "pre-event early warning," and further enhances the pertinence of grain depot emergency response.

[0086] Optionally, the processes of steps 2041 to 2044 include: Step 2041: Based on the ratio of the time lag value of each candidate topology path to the total path length, the propagation rate is obtained, and candidate topology paths with propagation rates greater than a preset rate threshold are eliminated to obtain the target topology path.

[0087] Optionally, for each candidate topology path, the security prediction system calculates the ratio of its time lag value to the total path length. This ratio is the propagation rate of the candidate topology path. The propagation rate reflects the average propagation speed of the anomaly on the path. The larger the value, the faster the anomaly propagates.

[0088] The preset rate threshold is the maximum reasonable propagation rate of anomalies, calculated based on the thermal conductivity of the grain pile medium, grain storage conditions, and historical anomaly propagation data. This threshold is used to filter paths that conform to the natural propagation patterns of anomalies. The safety prediction system compares the propagation rate of each candidate topology path with the preset rate threshold. If the propagation rate of a candidate topology path is greater than the preset rate threshold, the path is determined to not conform to the natural propagation patterns of anomalies and is eliminated. If the propagation rate is less than or equal to the preset rate threshold, the path is retained. All retained paths are the target topology paths.

[0089] Continuing with step 202, four candidate topological paths are obtained, with time lag values ​​of 1 hour, 1.2 hours, 0.8 hours, and 1.5 hours, respectively. Assume the total path lengths of the four candidate topological paths are 8 meters, 9.6 meters, 4 meters, and 15 meters, respectively. Calculate the propagation rate of each path: The propagation speed of the first path is 8 meters per hour (8 m / h), the second is 9.6 meters per hour (9.6 m / 1.2 m / h), the third is 4 meters per hour (4.8 m / h) and the fourth is 15 meters per hour (1.5 m / h) and the fifth is 10 meters per hour (15 m / h). The preset speed threshold is 9 meters per hour. Comparing each propagation speed with the threshold, the fourth path, with a speed of 10 meters per hour, exceeds the threshold and is therefore discarded. The remaining three paths are retained, resulting in the target topology path.

[0090] Step 2042: Based on the first state change time of the preceding monitoring point and the second state change time of the subsequent monitoring point on each target topology path, determine the time lag difference of each pair of adjacent monitoring points.

[0091] Optionally, the security prediction system extracts all monitoring points on each target topology path and divides these monitoring points into precursor monitoring points and successor monitoring points according to the order of anomaly propagation. Precursor monitoring points refer to the monitoring points located at the front of the target topology path where the anomaly propagates first, while successor monitoring points refer to the monitoring points located behind the precursor monitoring points where the anomaly propagates later. Adjacent precursor and successor monitoring points form adjacent monitoring point pairs. The security prediction system extracts the first state change time of the precursor monitoring points and the second state change time of the successor monitoring points on each target topology path. The first state change time refers to the specific time when the abnormal environmental parameter data of the precursor monitoring point first shows a significant change and deviates from the normal fluctuation range, and the second state change time refers to the specific time when the abnormal environmental parameter data of the successor monitoring point first shows a significant change and deviates from the normal fluctuation range.

[0092] The safety prediction system calculates the difference between the second state change time of the successor monitoring point and the first state change time of the predecessor monitoring point in each pair of adjacent monitoring points. This difference is the time lag difference of each pair of adjacent monitoring points, which can reflect the time delay difference of anomalies propagating from the predecessor monitoring point to the successor monitoring point.

[0093] Continuing from step 2041, the resulting target topology path is Monitoring Point 1 → Monitoring Point A → Monitoring Point B, with adjacent monitoring point pairs being (Monitoring Point 1, Monitoring Point A) and (Monitoring Point A, Monitoring Point B). The state transition times for each monitoring point are extracted: the first state transition time for Monitoring Point 1 is at hour 2, the second state transition time for Monitoring Point A is at hour 2.8, and the second state transition time for Monitoring Point B is at hour 3.9. The time lag difference between adjacent monitoring point pairs is calculated: the time lag difference for (Monitoring Point 1, Monitoring Point A) is 2.8 - 2 hours = 0.8 hours. The time lag difference for (Monitoring Point A, Monitoring Point B) is 3.9 - 2.8 hours = 1.1 hours. Similarly, the time lag difference between adjacent monitoring point pairs for other target topology paths is calculated.

[0094] Step 2043: Based on the time lag difference of each adjacent monitoring point pair on each target topology path, perform discrete prediction to obtain the discreteness of each target topology path, and eliminate target topology paths with a discreteness greater than a preset discrete threshold to obtain the causal propagation path.

[0095] Optionally, for each target topology path, the security prediction system performs discrete prediction based on the time lag difference of all adjacent monitoring point pairs. Discrete prediction refers to calculating the degree of dispersion of all time lag differences on the target topology path. The degree of dispersion refers to the degree of deviation of each time lag difference from the average of all time lag differences, which is used to reflect the stability of anomaly propagation on the path. The smaller the degree of dispersion, the more stable the anomaly propagation, and the more the path can reflect the true anomaly propagation pattern. The calculation method is as follows: calculate the arithmetic mean of all time lag differences on the target topology path, then calculate the absolute value of the difference between each time lag difference and the average value, and take the arithmetic mean of all absolute values, which is the degree of dispersion of the target topology path.

[0096] The preset discrete threshold is determined based on the normal discrete range of time lag difference in historical abnormal propagation data, and is used to screen out target topological paths with stable propagation.

[0097] The security prediction system compares the dispersion of each target topology path with a preset dispersion threshold. If the dispersion of a target topology path is greater than the preset dispersion threshold, the path is determined to be unstable and abnormally propagated, and is therefore eliminated. If the dispersion is less than or equal to the preset dispersion threshold, the path is retained. All retained paths are considered causal propagation paths, which accurately reflect the causal relationship of abnormal propagation and ensure the accuracy of subsequent propagation chains.

[0098] Continuing with step 2041, three target topology paths are obtained. The time lag difference between adjacent monitoring points on one of these paths is 0.8 hours and 1.1 hours. The dispersion of this path is calculated as follows: the average value is (0.8 + 1.1) / 2 = 0.95 hours. The absolute values ​​of each difference and the average value are then calculated to be 0.15 hours and 0.15 hours respectively. Finally, the average of the absolute values ​​is (0.15 + 0.15) / 2 = 0.15 hours. The preset dispersion threshold is 0.2 hours. Since the dispersion of this path (0.15 hours) is less than the threshold, it is retained.

[0099] If the time lag difference of another target topology path is 0.5 hours and 1.3 hours, and the calculated dispersion is 0.4 hours, which is greater than the preset threshold, it will be discarded. The dispersion of the third path is 0.18 hours, which is less than the threshold, so it will be retained, resulting in two causal propagation paths.

[0100] Step 2044: Based on the spatial overlap and monitoring point sharing relationships of each causal propagation path, identify target path segments with common precursor or successor monitoring points, and splice and merge the target path segments at the shared monitoring points to obtain the abnormal risk propagation chain.

[0101] Optionally, the safety prediction system extracts the spatial coordinate distribution and monitoring point information of each causal propagation path, and analyzes the spatial overlap and monitoring point sharing relationships between the various causal propagation paths. Spatial overlap refers to whether there are partially or completely overlapping areas between two or more causal propagation paths. Monitoring point sharing refers to whether two or more causal propagation paths contain the same monitoring points.

[0102] Based on the two relationships mentioned above, the safety prediction system identifies target path segments that share common precursor or successor monitoring points. Target path segments refer to the parts of the causal propagation path that share common precursor or successor monitoring points with other causal propagation paths.

[0103] The security prediction system performs head-to-tail splicing and branch merging of these target path segments at shared monitoring points. Head-to-tail splicing refers to taking two path segments with a common successor monitoring point, using the end point of the previous path segment (the shared monitoring point) as the starting point, and splicing the next path segment. Branch merging refers to merging two path segments with a common precursor monitoring point as subsequent branch paths after the common precursor monitoring point. After splicing and merging, one or more path chains are formed that can comprehensively reflect the spread of anomalies from the anomaly sub-region to the surrounding areas. This path chain is the anomaly risk propagation chain for each anomaly sub-region.

[0104] Continuing with step 2043, two causal propagation paths are obtained: Path 1 is Monitoring Point 1 → Monitoring Point A → Monitoring Point B, and Path 2 is Monitoring Point 1 → Monitoring Point C → Monitoring Point B. The safety prediction system analysis reveals a shared monitoring point relationship between the two paths: a common preceding monitoring point (Monitoring Point 1) and a common following monitoring point (Monitoring Point B). The target path segments are identified as (Monitoring Point 1 → Monitoring Point A), (Monitoring Point A → Monitoring Point B), (Monitoring Point 1 → Monitoring Point C), and (Monitoring Point C → Monitoring Point B). At the shared monitoring point Monitoring Point 1, the segments (Monitoring Point 1 → Monitoring Point A) and (Monitoring Point 1 → Monitoring Point C) are merged into a branch path. At the shared monitoring point Monitoring Point B, the segments (Monitoring Point A → Monitoring Point B) and (Monitoring Point C → Monitoring Point B) are concatenated to the end of their respective branch paths, ultimately forming the abnormal risk propagation chain: Monitoring Point 1 → Monitoring Point A → Monitoring Point B and Monitoring Point 1 → Monitoring Point C → Monitoring Point B, fully reflecting the two propagation paths of the anomaly from Monitoring Point 1 to Monitoring Point B.

[0105] Based on candidate topological paths, time lag values, and total path length, this invention accurately characterizes the anomaly propagation pattern through difference analysis and path integration, obtaining a complete and reliable anomaly risk propagation chain. It precisely captures the nonlinear propagation characteristics of disasters such as grain heating and mold, as well as the coupled impact of equipment operating status on environmental changes. This provides support for the subsequent spatiotemporal coupling, topological reorganization, and precise definition of anomaly risk areas in anomaly sub-regions, improving the accuracy of grain depot anomaly early warning, promoting the transformation of grain depot safety management from "post-event alarm" to "pre-event early warning," and thus enhancing the pertinence of grain depot emergency response.

[0106] Optionally, the processes of steps 301 to 304 include: Step 301: Based on the spatial coordinates and state evolution rate of each monitoring point in the anomaly risk propagation chain of each anomaly sub-region, identify monitoring points with a state evolution rate greater than zero and located on the boundary of the minimum spatial convex hull, and obtain the propagation front monitoring points of each anomaly sub-region.

[0107] Optionally, for each abnormal sub-region, the security prediction system obtains the abnormal risk propagation chain of the abnormal sub-region. The abnormal risk propagation chain is obtained based on step 20 and refers to the path chain that can completely reflect the spread of the abnormality from the abnormal sub-region to the surrounding area, including the order of the abnormal spread, the location of each spread node, the degree of abnormality, and the time required for the spread.

[0108] The safety prediction system extracts the spatial coordinates and state evolution rate of each monitoring point in the abnormal risk propagation chain. The spatial coordinates refer to the specific location of the monitoring point in the three-dimensional space of the grain depot, accurately identifying its spatial position. The state evolution rate refers to the rate of change of the abnormal state parameters (such as temperature and humidity) of the monitoring point over time, reflecting the changing trend of the abnormality level. A state evolution rate greater than zero indicates that the abnormality level of the monitoring point is continuously intensifying, while a state evolution rate less than or equal to zero indicates that the abnormality level is stabilizing or weakening. Based on the spatial coordinates of each monitoring point, the safety prediction system constructs the minimum spatial convex hull of the abnormal risk propagation chain for this abnormal sub-region. The minimum spatial convex hull is the smallest convex polygon (two-dimensional space) or convex polyhedron (three-dimensional space) that can completely enclose all monitoring points in the abnormal risk propagation chain. Its boundary is formed by the outermost monitoring point, reflecting the spatial range boundary of the abnormal propagation.

[0109] The safety prediction system selects monitoring points with a state evolution rate greater than zero that are located on the boundary of the minimum convex hull of the space. These monitoring points are not only at the forefront of anomaly propagation but also show a trend of anomalous intensification. They can accurately reflect the frontier position of anomaly propagation in anomaly sub-regions. These monitoring points are identified as the frontier monitoring points of propagation in each anomaly sub-region.

[0110] In one embodiment, suppose the anomalous risk propagation chain of an anomalous sub-region contains 5 monitoring points with the following spatial coordinates and state evolution rates: Monitoring point 1 (10, 20, 5), state evolution rate 0.8℃ / hour; Monitoring point 2 (12, 20, 5), state evolution rate 0.6℃ / hour; Monitoring point 3 (14, 21, 5), state evolution rate 0.9℃ / hour; Monitoring point 4 (16, 21, 5), state evolution rate 0.4℃ / hour; Monitoring point 5 (14, 19, 5), state evolution rate 0.3℃ / hour. A minimum spatial convex hull is constructed based on the spatial coordinates, and the boundary monitoring points of the convex hull are monitoring points 1, 3, 4, and 5. Monitoring points with a state evolution rate greater than zero and located on the boundary of the convex hull are selected. Since the state evolution rate of all boundary monitoring points is greater than zero, these four monitoring points are determined as the propagation front monitoring points of this anomalous sub-region.

[0111] Step 302: Based on the distance between monitoring points of each propagation front monitoring point in any two abnormal sub-regions, the propagation front monitoring points whose distance is less than the preset spatial coupling threshold are identified as candidate interactive monitoring point pairs, and a bridging path is constructed based on the monitoring points and connecting edges on the shortest connected path of the candidate interactive monitoring point pairs.

[0112] Optionally, the safety prediction system selects any two abnormal sub-regions as a group of analysis objects. For the two abnormal sub-regions in each group, it extracts all propagation front monitoring points of the first abnormal sub-region and all propagation front monitoring points of the second abnormal sub-region. It calculates the spatial straight-line distance between the propagation front monitoring points of any one abnormal sub-region and the propagation front monitoring points of the other abnormal sub-region, which is the monitoring point distance, used to reflect the degree of spatial correlation between the two propagation front monitoring points.

[0113] The preset spatial coupling threshold is the maximum spatial distance calculated based on the heat conduction radius of the grain pile medium, the abnormal propagation characteristics, and the spatial layout of the grain depot. When the distance between two monitoring points at the propagation front is less than this threshold, it indicates that there is a possibility of abnormal interactive propagation between the two monitoring points.

[0114] The safety prediction system identifies propagation front monitoring points whose distance to each other is less than a preset spatial coupling threshold as candidate interactive monitoring point pairs. A candidate interactive monitoring point pair refers to a pair of two propagation front monitoring points that may exhibit abnormal interactive propagation. After identifying the candidate interactive monitoring point pairs, the system searches for the shortest connected path between each pair. The shortest connected path is defined as the path with the shortest straight-line distance between two monitoring points and no significant media obstructions (such as grain clumping, impurity accumulation, or other substances that hinder abnormal transmission). All monitoring points (including the candidate interactive monitoring point pair itself) and edges connecting adjacent monitoring points are extracted from this shortest connected path. These monitoring points and connecting edges are then used to construct a bridging path, which reflects the abnormal propagation channel between the two candidate interactive monitoring point pairs.

[0115] In one embodiment, abnormal sub-regions A and B are selected as a set of analysis objects. The propagation front monitoring points of abnormal sub-region A are monitoring point 1 and monitoring point 2, and the propagation front monitoring points of abnormal sub-region B are monitoring point 3 and monitoring point 4. The distances between the monitoring points are calculated: the distance between monitoring point 1 and monitoring point 3 is 7 meters, the distance between monitoring point 1 and monitoring point 4 is 10 meters, the distance between monitoring point 2 and monitoring point 3 is 8 meters, and the distance between monitoring point 2 and monitoring point 4 is 9 meters. A preset spatial coupling threshold of 8 meters is set, and candidate interactive monitoring point pairs with a distance less than the threshold are selected, namely (monitoring point 1, monitoring point 3) and (monitoring point 2, monitoring point 3). Subsequently, the shortest connected path for each candidate interactive monitoring point pair is found: the shortest connected path for (monitoring point 1, monitoring point 3) is monitoring point 1 → monitoring point 5 → monitoring point 3, the monitoring points on the path are monitoring point 1, monitoring point 5, and monitoring point 3, and the connecting edges are monitoring point 1 and monitoring point 5, and monitoring point 5 and monitoring point 3. The shortest connected path between (monitoring point 2 and monitoring point 3) is monitoring point 2 → monitoring point 3, with monitoring points 2 and 3 on the path and the connecting edge between monitoring point 2 and monitoring point 3. Based on this, two bridging paths are constructed, each corresponding to one of the two candidate interactive monitoring point pairs.

[0116] Step 303: Based on the angle between the propagation direction vector of the abnormal risk propagation chain of the abnormal sub-region where the precursor monitoring point is located in each candidate interactive monitoring point pair and the direction vector of the starting segment of the bridging path, the candidate interactive monitoring point pairs with the angle value greater than the preset direction threshold are determined as target interactive monitoring point pairs.

[0117] Optionally, for each candidate interactive monitoring point pair, the security prediction system clarifies the predecessor-successor relationship between the two monitoring points in the pair, extracts the abnormal risk propagation chain of the abnormal sub-region where the predecessor monitoring point is located, and obtains the propagation direction vector of the abnormal risk propagation chain at the predecessor monitoring point. The propagation direction vector refers to the direction vector of the abnormality propagating to the surrounding area at the predecessor monitoring point, which can reflect the direction of the abnormality propagation.

[0118] Meanwhile, the safety prediction system extracts the starting segment direction vector of the bridging path corresponding to the candidate interactive monitoring point pair. The starting segment of the bridging path refers to the road segment in the bridging path that starts from the preceding monitoring point and extends to the first adjacent monitoring point on the path. The starting segment direction vector refers to the direction vector from the preceding monitoring point to the first adjacent monitoring point on the path, which can reflect the initial propagation direction of the bridging path.

[0119] The safety prediction system calculates the angle between the propagation direction vector and the initial segment direction vector. This angle reflects the consistency between the anomaly propagation direction and the bridging path propagation direction. The smaller the angle, the more consistent the two directions are, and the greater the likelihood that the anomaly will propagate through the bridging path. Conversely, the larger the angle, the more divergent the two directions are, and the lower the likelihood that the anomaly will propagate through the bridging path.

[0120] The preset direction threshold is the maximum angle calculated based on the directional patterns of anomaly propagation and the conductivity characteristics of the grain pile medium. It is used to screen candidate interactive monitoring point pairs with consistent propagation directions. The safety prediction system compares the angle value of each candidate interactive monitoring point pair with the preset direction threshold. If the angle value of a candidate interactive monitoring point pair is greater than the preset direction threshold, it is determined that the propagation direction of the candidate interactive monitoring point pair deviates too much, and the possibility of the anomaly propagating through the bridging path is extremely low, so it is discarded. If the angle value is less than or equal to the preset direction threshold, the candidate interactive monitoring point pair is retained, becoming the target interactive monitoring point pair.

[0121] Continuing with step 302, two candidate interaction monitoring point pairs are obtained: (Monitoring Point 1, Monitoring Point 3) and (Monitoring Point 2, Monitoring Point 3). In (Monitoring Point 1, Monitoring Point 3), Monitoring Point 1 is the preceding monitoring point (located in anomaly sub-region A), and Monitoring Point 3 is the succeeding monitoring point (located in anomaly sub-region B). In (Monitoring Point 2, Monitoring Point 3), Monitoring Point 2 is the preceding monitoring point (located in anomaly sub-region A), and Monitoring Point 3 is the succeeding monitoring point (located in anomaly sub-region B). The propagation direction vector of the abnormal risk propagation chain in anomaly sub-region A at Monitoring Point 1 is due east, and the direction vector of the corresponding bridging path starting segment is southeast. The angle between the two is calculated to be 30 degrees. The propagation direction vector at Monitoring Point 2 is due east, and the direction vector of the corresponding bridging path starting segment is northeast. The angle between the two is calculated to be 45 degrees. The preset direction threshold is 40 degrees. Comparing the angle value with the threshold, the angle value of (Monitoring Point 1, Monitoring Point 3) is 30 degrees, which is less than the threshold, and is retained. The angle between (monitoring point 2 and monitoring point 3) is 45 degrees, which is greater than the threshold, so it is removed, and finally one target interaction monitoring point pair (monitoring point 1 and monitoring point 3) is obtained.

[0122] Step 304: Based on the time difference of the start time of the abnormal state when each target interaction monitoring point is in its abnormal sub-region as the predecessor and successor, spatiotemporal coupling is performed to obtain the associated abnormal region pair.

[0123] Optionally, the security prediction system extracts two abnormal sub-regions where each target interaction monitoring point pair is located, and uses these two abnormal sub-regions as a combination of abnormal sub-regions to be coupled.

[0124] The safety prediction system obtains the start time of the two abnormal sub-regions entering the abnormal state, designates one of the abnormal sub-regions as the predecessor and the other as the successor, and calculates the time difference between their start times. Then, the predecessor and successor relationships are swapped, and another set of time differences is calculated. Based on the time difference of the start time of the abnormal state for each target interaction monitoring point when its respective abnormal sub-region is used as the predecessor and successor, spatiotemporal coupling is performed to obtain the associated abnormal region pair, as detailed in steps 3041 to 3044.

[0125] This invention focuses on the frontier monitoring points and the propagation chain of abnormal risks in abnormal sub-regions. It performs spatiotemporal coupling analysis from three dimensions: spatial distance, propagation direction, and temporal correlation. This accurately identifies pairs of abnormal sub-regions with abnormal interactive propagation associations, realizing the correlation and integration of abnormal regions. It avoids one-sided judgments on isolated abnormal points, captures the interactive influence patterns between abnormal sub-regions, improves the accuracy of abnormal risk analysis in grain depots, and enhances the pertinence of emergency response in grain depots.

[0126] Optionally, the processes of steps 3041 to 3044 include: Step 3041: Determine the temporal dominant relationship based on the time difference of each target interaction monitoring point pair, and determine the candidate coupled monitoring point pair based on the temporal dominant relationship of each target interaction monitoring point pair.

[0127] Optionally, the security prediction system obtains the start time when each target interactive monitoring point is in an abnormal state in the two abnormal sub-regions it belongs to. The start time refers to the specific time when the abnormal sub-region first appears abnormal and is determined to be an abnormal sub-region, that is, the time when the first monitoring point in the abnormal sub-region appears abnormal.

[0128] For each pair of target interaction monitoring points, the security prediction system calculates the time difference between the two abnormal sub-regions as predecessors and successors: The first abnormal sub-region is considered the predecessor, and the second the successor, and the difference between the start time of the successor and the predecessor is calculated. Then, the second abnormal sub-region is considered the predecessor, and the first the successor, and another set of time differences is calculated. Based on these two sets of time differences, the temporal dominance relationship is determined. The temporal dominance relationship refers to the dominant propagation relationship between the two abnormal sub-regions in the time dimension. The criterion is: if one set of time differences is positive and less than a preset temporal threshold, and the other is negative or greater than the preset temporal threshold, then in the set with the smaller positive time difference, the predecessor abnormal sub-region is the temporal dominance region, and the successor abnormal sub-region is the temporal subordinate region. This means that the corresponding target interaction monitoring point pair is determined to be a monitoring point pair with a temporal dominance relationship. If the time difference between the two sets of values ​​is positive and less than the preset time series threshold, it is determined that the two abnormal sub-regions have no obvious time series dominance relationship, but there is still a time series correlation.

[0129] The security prediction system identifies all target interaction monitoring point pairs with temporal dominant or temporal correlation as candidate coupled monitoring point pairs.

[0130] Continuing with step 303, two target interaction monitoring point pairs are obtained: (Monitoring Point 1, Monitoring Point 3) and (Monitoring Point 2, Monitoring Point 4). The corresponding abnormal sub-regions are Abnormal Sub-Region A and Abnormal Sub-Region B, respectively. The start time of Abnormal Sub-Region A is the 2nd hour, and the start time of Abnormal Sub-Region B is the 3rd hour. A preset time series threshold of 2 hours is used to calculate the time difference between the two sets: For (Monitoring Point 1, Monitoring Point 3), Abnormal Sub-Region A is the predecessor and B is the successor, with a time difference of 3-2=1 hour (positive and less than the threshold). Abnormal Sub-Region B is the predecessor and A is the successor, with a time difference of 2-3=-1 hour (negative). Therefore, Abnormal Sub-Region A is determined to be the time series dominant region, and this monitoring point pair is a candidate coupled monitoring point pair. For (Monitoring Point 2, Monitoring Point 4), the two time differences are 1 hour and -1 hour respectively, and are similarly determined to be candidate coupled monitoring point pairs, ultimately resulting in two candidate coupled monitoring point pairs.

[0131] Step 3042: For each candidate coupling monitoring point pair, if there exists a path from the first abnormal sub-region to the second abnormal sub-region, and also a path from the second abnormal sub-region to the first abnormal sub-region, then it is determined to be a bidirectional interaction mode. If only a unidirectional path exists or there is no direct topological path, then it is determined to be a homogeneous interaction mode.

[0132] Optionally, the security prediction system identifies two anomalous sub-regions (denoted as the first anomalous sub-region and the second anomalous sub-region) for each candidate coupled monitoring point pair. For each candidate coupled monitoring point pair, it searches for a path from the first anomalous sub-region to the second anomalous sub-region. This path must satisfy the following conditions: starting from the propagation front monitoring point in the first anomalous sub-region and ending at the propagation front monitoring point in the second anomalous sub-region; all monitoring points along the path must have a topological connection; and the path direction must be consistent with the anomalous propagation direction of the first anomalous sub-region. Simultaneously, it searches for a path from the second anomalous sub-region to the first anomalous sub-region, using the same search criteria as the path described above, only with the starting and ending points interchanged.

[0133] If there exists simultaneously a path from the first abnormal sub-region to the second abnormal sub-region and a path from the second abnormal sub-region to the first abnormal sub-region, the security prediction system determines that the candidate coupled monitoring point is in a two-way interaction mode with the corresponding two abnormal sub-regions. The two-way interaction mode means that the two abnormal sub-regions can propagate anomalies to each other and influence each other.

[0134] If there is only one path in one direction (one-way path), or if there is no topological path directly connecting the two abnormal sub-regions, the security prediction system determines that the candidate coupling monitoring point and the corresponding two abnormal sub-regions are in the same source interaction mode. The same source interaction mode means that the anomalies in the two abnormal sub-regions may originate from the same anomaly source, or there is only one-way influence without bidirectional interactive propagation.

[0135] Continuing with step 3041, two candidate coupled monitoring point pairs are obtained, corresponding to abnormal sub-regions A and B. Search paths: There exists a path from monitoring point 1 in abnormal sub-region A to monitoring point 3 in abnormal sub-region B (monitoring point 1 → monitoring point 5 → monitoring point 3), and there also exists a path from monitoring point 3 in abnormal sub-region B to monitoring point 1 in abnormal sub-region A (monitoring point 3 → monitoring point 5 → monitoring point 1). Therefore, this candidate coupled monitoring point pair (monitoring point 1, monitoring point 3) is a bidirectional interaction mode. For the candidate coupled monitoring point pair (monitoring point 2, monitoring point 4), there is only a path from monitoring point 2 in abnormal sub-region A to monitoring point 4 in abnormal sub-region B; there is no reverse path. Therefore, it is determined to be a same-source interaction mode.

[0136] Step 3043: For the first candidate coupling monitoring point pair in the bidirectional interaction mode, calculate the consistency index of the propagation rate of the bidirectional path. For the second candidate coupling monitoring point pair in the same-source interaction mode, calculate the parallelism index of the propagation direction vectors of the two abnormal sub-regions and the synchronization ratio at the start time.

[0137] Optionally, the security prediction system classifies candidate coupling monitoring point pairs, designating those belonging to the bidirectional interaction mode as the first candidate coupling monitoring point pair and those belonging to the same-source interaction mode as the second candidate coupling monitoring point pair.

[0138] For each first candidate coupling monitoring point pair, the security prediction system extracts its corresponding two bidirectional paths (the path from the first anomalous sub-region to the second anomalous sub-region, and the path from the second anomalous sub-region to the first anomalous sub-region), and calculates the propagation rate of each path. The propagation rate is the ratio of the path's time lag value to the total path length. Then, a propagation rate consistency index is calculated. This is done by calculating the absolute value of the difference between the propagation rates of the two paths, and then subtracting the ratio of this absolute value to the average propagation rate of the two paths from 1. The result is the propagation rate consistency index, ranging from 0 to 1. The closer the value is to 1, the more consistent the propagation rates of the two bidirectional paths, and the more stable the bidirectional interaction.

[0139] For each second candidate coupled monitoring point pair, the security prediction system calculates two metrics: First, there's the parallelism index of the propagation direction vectors. This involves extracting the propagation direction vectors of the anomalous risk propagation chains from the first and second anomalous sub-regions, calculating the angle between the two vectors, and subtracting the ratio of this angle to 90 degrees from 1 to obtain the parallelism index. The value ranges from 0 to 1; the closer the value is to 1, the closer the two propagation directions are to being parallel. Second, there's the synchronization ratio at the start time. This involves calculating the absolute value of the difference between the start times of the two anomalous sub-regions, subtracting the ratio of this absolute value to a preset time series threshold from 1 to obtain the synchronization ratio at the start time. The value ranges from 0 to 1; the closer the value is to 1, the more synchronized the start times of the two anomalous sub-regions are, and the more likely they originated from the same anomalous source.

[0140] In one embodiment, for the first candidate coupling monitoring point and its corresponding two bidirectional paths, assuming the propagation rates are 8 m / h and 7.2 m / h respectively, the propagation rate consistency index is calculated: the absolute value of the difference is 0.8 m / h, the average value is 7.6 m / h, and then 1 is subtracted from the ratio of 0.8 to 7.6 (approximately 0.105), resulting in a propagation rate consistency index of approximately 0.895. For the second candidate coupling monitoring point and its corresponding two anomalous sub-regions, the angle between the propagation direction vectors is 30 degrees, and the parallelism index is calculated: 1 is subtracted from the ratio of 30 degrees to 90 degrees (0.333), resulting in approximately 0.667. With a preset time series threshold of 2 hours and a starting time difference of 1 hour between the two anomalous sub-regions, the starting time synchronization ratio is calculated: 1 is subtracted from the ratio of 1 to 2 (0.5), resulting in 0.5.

[0141] Step 3044: If the propagation rate consistency index of the first candidate coupling monitoring point pair is greater than the preset consistency threshold, or if the parallelism index and the synchronization ratio at the start time of the second candidate coupling monitoring point pair are both greater than the corresponding preset thresholds, then the two abnormal sub-regions associated with the first candidate coupling monitoring point pair and the second candidate coupling monitoring point pair are determined as an associated abnormal region pair.

[0142] Optionally, the preset consistency threshold, parallelism threshold, and synchronization ratio threshold are determined based on the abnormal propagation pattern of grain piles and historical monitoring data, and are used to screen out abnormal sub-region combinations with real interactive relationships.

[0143] For each first candidate coupling monitoring point pair, the security prediction system compares its propagation rate consistency index with a preset consistency threshold. If the propagation rate consistency index is greater than the preset consistency threshold, it indicates that the two abnormal sub-regions of this bidirectional interaction mode have stable interaction and a real correlation. For each second candidate coupling monitoring point pair, the security prediction system compares its parallelism index with a preset parallelism threshold and its initial synchronization ratio with a preset initial synchronization ratio threshold. If both indices are greater than their corresponding preset thresholds, it indicates that the two abnormal sub-regions of this homogeneous interaction mode have a homogeneous correlation or a one-way effective correlation.

[0144] The safety prediction system identifies the two abnormal sub-regions associated with the first candidate coupled monitoring point pair and the second candidate coupled monitoring point pair that meet the above conditions as associated abnormal region pairs. Finally, it integrates all combinations of abnormal sub-regions that meet the conditions to obtain associated abnormal region pairs.

[0145] In one embodiment, the preset consistency threshold is 0.8, the parallelism threshold is 0.6, and the synchronization ratio threshold is 0.4. The propagation rate consistency index of the first candidate coupling monitoring point pair is 0.895, which is greater than 0.8, and its associated abnormal sub-regions A and B are determined as an associated abnormal region pair. The parallelism index of the second candidate coupling monitoring point pair is 0.667 (greater than 0.6), and the synchronization ratio at the initial time is 0.5 (greater than 0.4). Its associated abnormal sub-regions A and B are also determined as an associated abnormal region pair. Finally, after integration, one set of associated abnormal region pairs (abnormal sub-region A, abnormal sub-region B) is obtained.

[0146] Based on target interaction monitoring point pairs, this invention performs spatiotemporal coupling analysis from three levels: time series correlation, interaction mode, and indicators. It accurately identifies abnormal sub-region pairs with real interaction and propagation correlation, realizes accurate correlation and integration of abnormal regions, avoids one-sided judgment of isolated abnormal sub-regions, captures the interaction and influence patterns between abnormal sub-regions, and improves the accuracy of grain depot abnormal risk analysis.

[0147] Optionally, the processes of steps 401 to 404 include: Step 401: For the abnormal risk propagation chain of each abnormal sub-region in the abnormal risk connected domain, take the monitoring point at the end of the propagation in each abnormal risk propagation chain as the terminal diffusion source point, and obtain the duration of the abnormality based on the time difference between the state abnormal start time of each terminal diffusion source point and the current time.

[0148] Optionally, the abnormal risk connected domain refers to a connected whole composed of multiple abnormal sub-regions that have interactive influences, including the entire spatial range and abnormal correlation of each abnormal sub-region. The security prediction system extracts the abnormal risk propagation chain of each abnormal sub-region in the abnormal risk connected domain. The abnormal risk propagation chain refers to the path chain that can completely reflect the spread of the abnormal from the abnormal sub-region to the surrounding area, including the order of abnormal spread, the location of each spread node, the degree of abnormality, and the time required for spread.

[0149] For each abnormal risk propagation chain, the safety prediction system identifies the monitoring point at the end of the propagation. The monitoring point at the end of the propagation refers to the outermost monitoring point in the abnormal risk propagation chain where the abnormality is last to spread. This monitoring point is the starting point for the abnormality to spread further to the surrounding area, and therefore it is identified as the end diffusion source point.

[0150] The safety prediction system extracts the anomaly initiation time for each terminal diffusion source point. The anomaly initiation time refers to the specific time when the terminal diffusion source point first collects abnormal environmental parameter data and is identified as an abnormal monitoring point, which is consistent with the timing standard of the first initiation time mentioned above. At the same time, the current time is obtained, which refers to the specific time when the safety prediction system performs this spatiotemporal boundary extension operation.

[0151] The safety prediction system calculates the time difference between the start time of the abnormal state of each terminal diffusion source point and the current time, which is the duration of the abnormality at each terminal diffusion source point. The duration of the abnormality reflects the duration of the abnormality at that terminal diffusion source point. The longer the duration, the larger the potential range of the abnormality spread.

[0152] In one embodiment, the abnormal risk connectivity domain includes two abnormal sub-regions, namely abnormal sub-region A and abnormal sub-region B. The abnormal risk propagation chain of abnormal sub-region A is monitoring point 1 → monitoring point 2 → monitoring point 3, with monitoring point 3 being the end point of the propagation, and its abnormal state starting time is the 3rd hour. The abnormal risk propagation chain of abnormal sub-region B is monitoring point 4 → monitoring point 5 → monitoring point 6, with monitoring point 6 being the end point of the propagation, and its abnormal state starting time is the 4th hour. The current time is set to the 8th hour, and the duration of the abnormality at the two end propagation source points is calculated: the duration of the abnormality at monitoring point 3 is 8 - 3 hours = 5 hours. The duration of the abnormality at monitoring point 6 is 8 - 4 hours = 4 hours.

[0153] Step 402: Based on the abnormal duration of each terminal diffusion source point and the historical average propagation rate of its corresponding abnormal sub-region, determine the potential influence radius of each terminal diffusion source point.

[0154] Optionally, the historical average propagation rate refers to the average propagation speed of the abnormal risk propagation chain in the abnormal sub-region during the historical abnormal propagation process. It is calculated as the arithmetic mean of the propagation rates of each abnormal propagation in the history of the abnormal sub-region. The propagation rate is the ratio of the total path length of the abnormal propagation to the corresponding time lag value. The historical average propagation rate can reflect the normal speed of abnormal propagation in the abnormal sub-region.

[0155] The safety prediction system calculates the potential influence radius of each terminal diffusion source point based on its anomaly duration and the historical average propagation rate of its corresponding anomaly sub-region. The potential influence radius refers to the maximum spatial distance that an anomaly at the terminal diffusion source point can spread to its surroundings within the current anomaly duration. It is calculated by multiplying the anomaly duration of the terminal diffusion source point by its corresponding historical average propagation rate; the product is the potential influence radius of that terminal diffusion source point. The potential influence radius accurately quantifies the potential spread range of an anomaly at the terminal diffusion source point.

[0156] Continuing with the two terminal diffusion source points obtained in step 401, the duration of the anomaly at monitoring point 3 (corresponding to anomalous sub-region A) is 5 hours, and the historical average propagation rate of anomalous sub-region A is 8 m / h. The duration of the anomaly at monitoring point 6 (corresponding to anomalous sub-region B) is 4 hours, and the historical average propagation rate of anomalous sub-region B is 7 m / h. Calculate the potential radius of influence: The potential radius of influence for monitoring point 3 is 5 hours * 8 m / h = 40 meters. The potential radius of influence for monitoring point 6 is 4 hours * 7 m / h = 28 meters.

[0157] Step 403: Using the spatial coordinates of each terminal diffusion source point as the starting point, the potential influence radius as the radius, and the propagation direction vector as the extension direction, construct a potential risk coverage area. Based on the spatial union of the potential risk coverage areas and the overlapping or adjacent potential risk coverage areas, obtain the spatiotemporal extension region of the abnormal risk connected domain.

[0158] Optionally, the safety prediction system extracts the propagation direction vector of the anomaly risk propagation chain at each terminal diffusion source point. The propagation direction vector refers to the direction vector of the anomaly propagating from the terminal diffusion source point to the surrounding area, reflecting the main propagation direction of the anomaly. For each terminal diffusion source point, the safety prediction system constructs a potential risk coverage area with the spatial coordinates of the terminal diffusion source point as the starting point, the potential influence radius as the radius, and the propagation direction vector as the extension direction. The potential risk coverage area refers to the spatial range to which the anomaly at the terminal diffusion source point may spread within the current anomaly duration. This area is a fan-shaped region extending along the propagation direction vector with the starting point as the center and the potential influence radius as the radius, encompassing both the spatial range of the anomaly propagation and reflecting the directional characteristics of the anomaly propagation.

[0159] The safety prediction system analyzes the spatial relationships of each potential risk coverage area. Potential risk coverage areas that have spatial overlap (two or more potential risk coverage areas partially overlap in space) or spatial adjacency (the boundary distance between two potential risk coverage areas is less than or equal to the heat conduction radius of the grain pile medium, with no obvious spatial interval) are merged. The spatial union of all potential risk coverage areas is calculated, which is the smallest spatial range that contains all potential risk coverage areas. This spatial range is the spatiotemporal extension region of the abnormal risk connectivity domain. The spatiotemporal extension region can reflect the spatial range to which the anomaly may spread in the future within the abnormal risk connectivity domain.

[0160] Continuing with step 401, two terminal diffusion source points are obtained. Assume the spatial coordinates of monitoring point 3 are (18, 22, 5), with a potential influence radius of 40 meters and a propagation direction vector of east. Assume the spatial coordinates of monitoring point 6 are (25, 30, 5), with a potential influence radius of 28 meters and a propagation direction vector of northeast.

[0161] Starting from two terminal diffusion source points, fan-shaped potential risk coverage areas extending along their corresponding propagation directions were constructed: the potential risk coverage area for monitoring point 3 is a fan-shaped region extending due eastward with a radius of 40 meters centered at (18, 22, 5). The potential risk coverage area for monitoring point 6 is a fan-shaped region extending northeastward with a radius of 28 meters centered at (25, 30, 5). Analysis revealed partial spatial overlap between the two potential risk coverage areas; therefore, they were merged, and the spatial union of the two regions was calculated to obtain the spatiotemporal extension region of the abnormal risk connectivity domain.

[0162] Step 404: Extend the spatiotemporal boundary based on the spatial coordinates of each monitoring point in the spatiotemporal extension region and the abnormal risk connected domain to obtain the abnormal risk region.

[0163] Optionally, the spatiotemporal boundary is extended based on the spatial coordinates of each monitoring point in the spatiotemporal extension region and the abnormal risk connectivity domain to obtain the abnormal risk region, as in steps 4041 to 4043.

[0164] This invention focuses on the abnormal risk connectivity domain and the abnormal risk propagation chain, accurately calculating the potential spread range of anomalies from both temporal and spatial dimensions. It extends the spatiotemporal boundaries of the abnormal risk connectivity domain, precisely defines the specific boundaries and spread direction of the anomaly source, distinguishes between instantaneous sensor failures and the actual spread of internal grain hazards, and achieves precise definition of the spatial range of grain depot safety hazards, thereby improving the pertinence and effectiveness of grain depot emergency response.

[0165] Optionally, the processes of steps 4041 to 4043 include: Step 4041: Based on the spatial coordinates of each monitoring point in the abnormal risk connected domain, the monitoring points located inside or on the boundary of the spatiotemporal extension region are identified as potential risk monitoring points.

[0166] Optionally, the spatiotemporal extension region refers to the spatial extent to which anomalies may spread in the future within the anomaly risk connectivity domain, consisting of the merged potential risk coverage area. The spatial coordinates of each monitoring point within the anomaly risk connectivity domain are obtained. The anomaly risk connectivity domain refers to a connected whole composed of multiple anomaly sub-regions with interactive influences. Its internal monitoring points include monitoring points in each anomaly sub-region and monitoring points on bridging paths. The spatial coordinates refer to the specific location of the monitoring point in the three-dimensional space of the grain depot, enabling precise identification of the monitoring point's spatial location.

[0167] For each monitoring point in the abnormal risk connectivity domain, the safety prediction system determines whether the spatial coordinates of the monitoring point are located inside or on the boundary of the spatiotemporal extension region. The criteria are as follows: if the location corresponding to the spatial coordinates of the monitoring point is completely within the spatial range covered by the spatiotemporal extension region, it is determined to be located inside the spatiotemporal extension region. If the location corresponding to the spatial coordinates of the monitoring point falls exactly on the boundary line of the spatiotemporal extension region, it is determined to be located on the boundary line of the spatiotemporal extension region. All monitoring points located inside or on the boundary of the spatiotemporal extension region are identified as potential risk monitoring points. Potential risk monitoring points refer to monitoring points that may be affected by the abnormal spread and are within the potential spread range of the abnormality.

[0168] In one embodiment, there are a total of 8 monitoring points in the abnormal risk connectivity domain, and the spatiotemporal extension area obtained in step 403 covers an area in the eastern part of the grain depot. The spatial coordinates of the 8 monitoring points are determined one by one: the spatial coordinates of 5 monitoring points are located inside the spatiotemporal extension area, the spatial coordinates of 2 monitoring points are located on the boundary of the spatiotemporal extension area, and the spatial coordinates of 1 monitoring point are located outside the spatiotemporal extension area. Therefore, these 5 internal monitoring points and 2 boundary monitoring points, a total of 7 monitoring points, are identified as potential risk monitoring points.

[0169] Step 4042: Based on the topological hop count between the potential risk monitoring point and the nearest terminal diffusion source point, remove potential risk monitoring points whose topological hop count exceeds the preset maximum propagation level to obtain the target risk monitoring point.

[0170] Optionally, the safety prediction system calculates the topological hop count between each potential risk monitoring point and the nearest end diffusion source. The topological hop count refers to the number of times adjacent monitoring points are connected on the shortest path between two monitoring points. That is, how many adjacent monitoring points need to be passed from the potential risk monitoring point to reach the nearest end diffusion source. The fewer the topological hop count, the closer the potential risk monitoring point is to the end diffusion source, and the greater the possibility of being affected by abnormal diffusion.

[0171] The preset maximum propagation level is a maximum number of topological hops calculated based on the thermal conductivity of the grain pile medium, the abnormal propagation rate, and the safety and control requirements of the grain depot. This maximum number is used to screen out effective potential risk monitoring points affected by abnormal diffusion, eliminating monitoring points that are too far from the terminal diffusion source or exceed the reasonable propagation range. The safety prediction system compares the topological hop count of each potential risk monitoring point with the preset maximum propagation level. If the topological hop count of a potential risk monitoring point exceeds the preset maximum propagation level, the monitoring point is determined to be outside the reasonable propagation range and is eliminated. If the topological hop count is less than or equal to the preset maximum propagation level, the monitoring point is retained. All retained monitoring points are the target risk monitoring points.

[0172] Continuing with the 7 potential risk monitoring points from step 4041, step 401 yielded 2 terminal propagation source points (monitoring point 3 and monitoring point 6). The topological hop count between each potential risk monitoring point and its nearest terminal propagation source point is calculated: 3 monitoring points have a topological hop count of 1 with monitoring point 3, 2 monitoring points have a topological hop count of 2 with monitoring point 6, 1 monitoring point has a topological hop count of 4 with monitoring point 3, and 1 monitoring point has a topological hop count of 5 with monitoring point 6. With a preset maximum propagation level of 3, each topological hop count is compared to a threshold. The two monitoring points with topological hop counts of 4 and 5 exceed the threshold and are removed. The remaining 5 monitoring points are retained to obtain the target risk monitoring points.

[0173] Step 4043: Calculate the minimum convex polygon that encloses each target risk monitoring point based on the spatial coordinates of the target risk monitoring points to obtain the abnormal risk area.

[0174] Optionally, the target risk monitoring point refers to an effective monitoring point that is within the reasonable range of abnormal propagation and is affected by the abnormal spread, and its spatial coordinates can accurately reflect the key nodes of the potential spread range of the abnormality.

[0175] The security prediction system calculates the minimum convex polygon that can enclose each target risk monitoring point based on the spatial coordinates of all target risk monitoring points. The minimum convex polygon is the smallest convex shape that can completely enclose all target risk monitoring points; its boundary is formed by the outermost target risk monitoring points. It accurately covers the spatial range of all target risk monitoring points while also considering the spatial continuity of anomaly propagation. The entire spatial range covered by this minimum convex polygon is the anomaly risk region, which includes the connected domain of the anomaly risk and the surrounding effective area that the anomaly may spread to in the future.

[0176] Continuing with step 4042, five target risk monitoring points are obtained, with spatial coordinates of (18, 22, 5), (25, 22, 5), (30, 28, 5), (25, 34, 5), and (18, 34, 5). Based on these spatial coordinates, the smallest convex polygon that can completely enclose these five monitoring points is calculated. The vertices of this convex polygon are the four outermost monitoring points, namely (18, 22, 5), (25, 22, 5), (30, 28, 5), (25, 34, 5), and (18, 34, 5), forming a closed convex polygon region, which is the abnormal risk region.

[0177] This invention focuses on monitoring points in the spatiotemporal extension region and the abnormal risk connectivity domain. Through spatial boundary fitting, it achieves precise spatiotemporal boundary extension of the abnormal risk connectivity domain, accurately defines the specific boundary and diffusion direction of the abnormal source, distinguishes between local sensor instantaneous failures and the actual internal spread of hidden dangers in grain, and realizes precise definition of the spatial range of safety hazards in grain depots. This promotes the transformation of grain depot safety management from "post-event alarm" to "pre-event warning," and improves the pertinence and effectiveness of emergency response in grain depots.

[0178] Furthermore, the grain depot safety prediction system based on edge computing and time series analysis provided by the present invention will be described below. The grain depot safety prediction system based on edge computing and time series analysis described below can be referred to in correspondence with the grain depot safety prediction method based on edge computing and time series analysis described above.

[0179] Figure 2 This is a schematic diagram of the structure of the grain depot safety prediction system based on edge computing and time series analysis provided by the present invention. The grain depot safety prediction system based on edge computing and time series analysis includes a regional positioning and partitioning module 210, an anomaly diffusion prediction module 220, an anomaly spatiotemporal coupling module 230, and a risk area positioning module 240.

[0180] The embodiments of the present invention improve the targeted nature of emergency response in grain depots.

[0181] Please see Figure 3 , Figure 3An embodiment diagram of an electronic device provided in accordance with the present invention. For example... Figure 3 As shown, an embodiment of the present invention provides an electronic device 300, including a memory 310, a processor 320, and a computer program 311 stored in the memory 310 and executable on the processor 320. When the processor 320 executes the computer program 311, it implements the processes of steps 10 to 40.

[0182] Please see Figure 4 , Figure 4 An embodiment diagram of a computer-readable storage medium provided in accordance with an embodiment of the present invention is shown. Figure 4 As shown, this embodiment provides a computer-readable storage medium 400 on which a computer program 311 is stored. When the computer program 311 is executed by a processor, it implements the processes of steps 10 to 40.

[0183] On the other hand, the present invention also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer is able to execute the grain depot security prediction method based on edge computing and time series analysis provided by the above methods, which includes steps 10 to 40.

Claims

1. A grain depot safety prediction method based on edge computing and time series analysis, characterized in that, include: Based on the temporal relationship between the grain depot environmental state sequence and the equipment operation state sequence on the edge side, the grain depot is located and partitioned into regions to obtain abnormal sub-regions; Anomaly evolution is performed based on the spatiotemporal distribution characteristics of anomaly monitoring points within each anomaly sub-region to obtain anomaly spatiotemporal evolution trajectory. Furthermore, diffusion prediction is performed based on the anomaly propagation trend of the anomaly spatiotemporal evolution trajectory within each anomaly sub-region to obtain the anomaly risk propagation chain within each anomaly sub-region. Spatiotemporal coupling is performed based on the spatial topological distance between abnormal sub-regions and the transmission direction of their respective abnormal risk propagation chains to obtain associated abnormal region pairs. Based on the abnormal propagation evolution of the associated abnormal region pairs, the abnormal sub-regions with interactive influences are topologically reorganized to obtain the abnormal risk connected domain. Based on the spatial extension path of the abnormal risk propagation chain of each abnormal sub-region in the abnormal risk connected domain, the spatiotemporal boundary of the abnormal risk connected domain is extended to obtain the abnormal risk region.

2. The grain depot safety prediction method based on edge computing and time series analysis according to claim 1, characterized in that, The steps for determining the abnormal risk area include: For the abnormal risk propagation chain of each abnormal sub-region in the abnormal risk connected domain, the monitoring point at the end of each abnormal risk propagation chain is taken as the terminal diffusion source point. Based on the time difference between the state abnormal start time of each terminal diffusion source point and the current time, the duration of the abnormality is obtained. Based on the abnormal duration of each terminal diffusion source point and the historical average propagation rate of its corresponding abnormal sub-region, the potential influence radius of each terminal diffusion source point is determined. Starting from the spatial coordinates of each terminal diffusion source point, with the potential influence radius as the radius and the propagation direction vector as the extension direction, a potential risk coverage area is constructed. Based on the spatial union of the potential risk coverage area and the overlapping or adjacent potential risk coverage areas, the spatiotemporal extension region of the abnormal risk connectivity domain is obtained. The spatiotemporal boundary is extended based on the spatial coordinates of each monitoring point in the spatiotemporal extension region and the abnormal risk connectivity domain to obtain the abnormal risk region.

3. The grain depot safety prediction method based on edge computing and time series analysis according to claim 2, characterized in that, The process of extending the spatiotemporal boundary based on the spatial coordinates of each monitoring point in the spatiotemporal extension region and the abnormal risk connectivity domain to obtain the abnormal risk region includes: Based on the spatial coordinates of each monitoring point in the abnormal risk connectivity domain, the monitoring points located inside or on the boundary of the spatiotemporal extension region are identified as potential risk monitoring points. Based on the topological hop count between each potential risk monitoring point and the nearest terminal propagation source, potential risk monitoring points with a topological hop count exceeding the preset maximum propagation level are eliminated to obtain the target risk monitoring points; The minimum convex polygon enclosing each target risk monitoring point is calculated based on the spatial coordinates of the target risk monitoring points to obtain the abnormal risk area.

4. The grain depot safety prediction method based on edge computing and time series analysis according to claim 1, characterized in that, The steps involved in determining the anomaly risk propagation chain for each anomaly sub-region include: For each anomalous sub-region, based on the spatial coordinates of the monitoring points corresponding to the continuous time slices in the anomalous spatiotemporal evolution trajectory, the displacement vector of the centroid position between adjacent time slices is calculated, and the propagation direction vector with the highest directional distribution frequency is used as the anomalous propagation direction reference. Starting from the centroid position of the anomalous sub-region, a spatial search area extending along the anomalous propagation direction reference axis is constructed. The spatial coordinates of the monitoring points distributed within the spatial search area are searched to obtain the spatial potential impact monitoring points. Based on the topological connection relationship between each spatial potential impact monitoring point and the boundary monitoring points of the abnormal sub-region, a candidate topological path is constructed. The time lag value is determined based on the first start time when each monitoring point in each candidate topology path is in an abnormal state and the second start time of the abnormal sub-region. Based on the time lag value and total path length of each candidate topological path, diffusion prediction is performed to obtain the anomalous risk propagation chain for each anomalous sub-region.

5. The grain depot safety prediction method based on edge computing and time series analysis according to claim 4, characterized in that, The diffusion prediction based on the time lag value and total path length of each candidate topological path yields the anomaly risk propagation chain for each anomaly sub-region, including: The propagation rate is obtained by comparing the time lag value of each candidate topology path with the total path length. Candidate topology paths with propagation rates greater than a preset rate threshold are then eliminated to obtain the target topology path. Based on the first state change time of the preceding monitoring point and the second state change time of the subsequent monitoring point on each target topology path, the time lag difference of each pair of adjacent monitoring points is determined. Discrete prediction is performed based on the time lag difference between each pair of adjacent monitoring points on each target topology path to obtain the dispersion of each target topology path. Target topology paths with a dispersion greater than a preset dispersion threshold are then eliminated to obtain the causal propagation path. Based on the spatial overlap and monitoring point sharing relationships of various causal propagation paths, target path segments with common precursor or successor monitoring points are identified, and the target path segments are spliced ​​together and branched at the shared monitoring points to obtain the abnormal risk propagation chain of each abnormal sub-region.

6. The grain depot safety prediction method based on edge computing and time series analysis according to claim 1, characterized in that, The steps for determining the associated abnormal region pairs include: Based on the spatial coordinates and state evolution rate of each monitoring point in the anomaly risk propagation chain of each anomaly sub-region, the monitoring points with state evolution rates greater than zero and located on the boundary of the minimum spatial convex hull are identified, thus obtaining the propagation front monitoring points of each anomaly sub-region. Based on the distance between monitoring points of each propagation front monitoring point in any two abnormal sub-regions, the propagation front monitoring points whose distance is less than the preset spatial coupling threshold are identified as candidate interactive monitoring point pairs, and a bridging path is constructed based on the monitoring points and connecting edges on the shortest connected path of the candidate interactive monitoring point pairs. Based on the angle between the propagation direction vector of the abnormal risk propagation chain in the abnormal sub-region where the precursor monitoring point is located in each candidate interactive monitoring point pair and the direction vector of the starting segment of the bridging path, candidate interactive monitoring point pairs with an angle value greater than a preset direction threshold are determined as target interactive monitoring point pairs. Spatiotemporal coupling is performed based on the time difference at the start time of the abnormal state when each target interactive monitoring point acts as the predecessor and successor of its respective abnormal sub-region, to obtain the associated abnormal region pair.

7. The grain depot safety prediction method based on edge computing and time series analysis according to claim 6, characterized in that, Spatiotemporal coupling is performed based on the time difference of each target interactive monitoring point pair to obtain associated anomaly region pairs, including: The temporal dominant relationship is determined based on the time difference of each target interaction monitoring point pair, and candidate coupled monitoring point pairs are determined based on the temporal dominant relationship of each target interaction monitoring point pair; For each candidate pair of coupling monitoring points, if there is a path from the first abnormal sub-region to the second abnormal sub-region and a path from the second abnormal sub-region to the first abnormal sub-region, it is determined to be a bidirectional interaction mode; if there is only a unidirectional path or no direct topological path, it is determined to be a homogeneous interaction mode. For the first candidate coupling monitoring point pair in the bidirectional interaction mode, calculate the consistency index of the propagation rate of the bidirectional path; for the second candidate coupling monitoring point pair in the same source interaction mode, calculate the parallelism index of the propagation direction vectors of the two abnormal sub-regions and the synchronization ratio at the start time. If the propagation rate consistency index of the first candidate coupling monitoring point pair is greater than the preset consistency threshold, or if the parallelism index and the synchronization ratio at the start time of the second candidate coupling monitoring point pair are both greater than the corresponding preset threshold, then the two abnormal sub-regions associated with the first candidate coupling monitoring point pair and the second candidate coupling monitoring point pair are determined as an associated abnormal region pair.

8. The grain depot safety prediction method based on edge computing and time series analysis according to any one of claims 1 to 7, characterized in that, The steps for determining the abnormal sub-region include: Based on the time-series synchronization mapping relationship between the grain depot environmental state sequence and the equipment operation state sequence, a state association map is constructed; Based on the temporal deviation of the state of each abnormal monitoring point in the state correlation map, the abnormal monitoring points are determined; Based on the spatial adjacency relationship of the abnormal monitoring points and the duration of the abnormal temporal state, the monitoring points are clustered to obtain the range of the target abnormal area. Based on the consistency of the temporal change trend of the monitoring points within the target anomaly area, the target anomaly area is divided into regional partitions to obtain the anomaly sub-regions.

9. The grain depot safety prediction method based on edge computing and time series analysis according to claim 8, characterized in that, The monitoring points are clustered based on their spatial adjacency and the duration of their temporal anomalies to obtain the target anomaly region, which includes: Based on the spatial coordinates of the abnormal monitoring points in the three-dimensional space of the grain depot and the heat conduction radius of the grain pile medium, a proximity connectivity determination is made to obtain connected monitoring point pairs. Based on the abnormal duration of each abnormal monitoring point in the connected monitoring point pair and the minimum steady-state time threshold for disaster formation, the target monitoring point is determined. Based on the spatiotemporal intersection matching of the target monitoring point and the connected monitoring point pair, a target connected unit is obtained, and based on the spatial sequence continuity of the monitoring points in the target connected unit, topological link tracing is performed to obtain a linear abnormal chain reflecting the local diffusion path of the disaster. Based on the deviation of the distance between adjacent monitoring points in the linear anomaly chain from the thermal conduction radius, breakpoint detection is performed to obtain discontinuous chain segments with medium obstruction or monitoring blind spots. Based on the breakpoint positions of the discontinuous chain segments, topological truncation and segmentation are performed to obtain the initial anomaly cluster. Edge fusion expansion is performed based on the spatial containment relationship between the outer boundary monitoring points of the initial anomaly cluster and the remaining unaggregated anomaly monitoring points to obtain the range of the target anomaly region.

10. A grain depot safety prediction system based on edge computing and time series analysis, characterized in that, Used to implement the grain depot safety prediction method based on edge computing and time series analysis as described in any one of claims 1 to 9; The grain depot safety prediction system based on edge computing and time series analysis includes: The regional positioning and partitioning module is used to perform regional positioning and regional partitioning of grain depots based on the temporal relationship between the grain depot environmental state sequence and the equipment operation state sequence on the edge side, and to obtain abnormal sub-regions; The anomaly propagation prediction module is used to perform anomaly evolution based on the spatiotemporal distribution characteristics of anomaly monitoring points in each anomaly sub-region, obtain the anomaly spatiotemporal evolution trajectory, and perform propagation prediction based on the anomaly propagation trend of the anomaly spatiotemporal evolution trajectory in each anomaly sub-region, thereby obtaining the anomaly risk propagation chain in each anomaly sub-region. The abnormal spatiotemporal coupling module is used to perform spatiotemporal coupling based on the spatial topological distance between abnormal sub-regions and the transmission direction of their respective abnormal risk propagation chains to obtain associated abnormal region pairs. Based on the abnormal propagation evolution of the associated abnormal region pairs, the abnormal sub-regions with interactive influences are topologically reorganized to obtain abnormal risk connected domains. The risk area positioning module is used to extend the spatiotemporal boundary of the abnormal risk connected domain based on the spatial extension path of the abnormal risk propagation chain of each abnormal sub-region in the abnormal risk connected domain, so as to obtain the abnormal risk area.

Images