Monitoring and early warning analysis method based on artificial intelligence and server

Abstract

The invention provides a monitoring early warning analysis method based on artificial intelligence and a server, and the method comprises the steps: firstly collecting a multi-source monitoring data stream of a target monitoring area, which comprises at least two kinds of real-time monitoring data and time-space label information, so as to determine a collection position and a timestamp, and then carrying out the spatial-temporal feature extraction, the method comprises the following steps: generating a spatial-temporal characteristic matrix with a hierarchical association relationship, analyzing a dynamic mode of the matrix through a preset anomaly recognition network, determining a potential anomaly event and an anomaly propagation path, and generating an adaptive dynamic early warning strategy including a differentiated trigger condition and a response instruction according to a path topological structure and event attribute parameters. And finally, optimizing and adjusting the strategy in real time by utilizing historical early warning feedback data, and outputting the strategy to a terminal equipment cluster, thereby comprehensively and accurately analyzing the monitoring data, and realizing efficient and intelligent monitoring and early warning.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology. Specifically, it relates to a monitoring and early warning analysis method and a server based on artificial intelligence. Background Art

[0002] In the operation and management of various fields today, the effective monitoring and early warning analysis of various target monitoring areas are crucial, covering many fields such as industrial production, urban security, and environmental monitoring. However, the existing monitoring and early warning analysis methods have many deficiencies and are difficult to meet the growing complex monitoring requirements.

[0003] For example, the related technologies do not fully consider the time and space information contained in the data. Moreover, the related technologies mainly rely on pre-set fixed thresholds, which are too rigid when facing complex and changeable actual situations, difficult to adapt to dynamic environments, and lack effective prediction capabilities for potential abnormal events that have not yet been significantly manifested.

[0004] In addition, the related technologies usually adopt unified and static early warning rules, which do not take into account the differences in different positions and different time stages within the target monitoring area and cannot give targeted early warnings according to the actual situation. Summary of the Invention

[0005] In view of the above-mentioned problems, in combination with the first aspect of the present invention, embodiments of the present invention provide a monitoring and early warning analysis method based on artificial intelligence, and the method includes: Collect multi-source monitoring data streams within a target monitoring area, where the multi-source monitoring data streams include at least two types of real-time monitoring data and corresponding spatio-temporal tag information, and the spatio-temporal tag information is used to describe the collection location and timestamp of the real-time monitoring data within the target monitoring area; Extract spatio-temporal features from the multi-source monitoring data streams to generate a spatio-temporal feature matrix with hierarchical association relationships, where each element in the spatio-temporal feature matrix corresponds to the monitoring index feature of a target location within the target time window in the target monitoring area; Based on a preset anomaly recognition network, perform dynamic pattern analysis on the spatio-temporal feature matrix to determine potential abnormal events existing within the target monitoring area and their corresponding abnormal propagation paths, and the abnormal propagation paths are used to indicate the diffusion direction and influence range of the potential abnormal events within the target monitoring area; Generate a dynamic early warning strategy adapted to the target monitoring area according to the topological structure of the abnormal propagation path and the attribute parameters of the potential abnormal events, and the dynamic early warning strategy includes differential early warning trigger conditions and response instruction sets for different positions and different time stages; Based on the historical warning feedback data, the dynamic warning strategy is optimized and adjusted in real time, and the optimized dynamic warning strategy is output to the terminal device cluster corresponding to the target monitoring area.

[0006] In another aspect, an embodiment of the present invention further provides a server, including a processor and a machine-readable storage medium. The machine-readable storage medium is connected to the processor. The machine-readable storage medium is used to store programs, instructions or codes, and the processor is used to execute the programs, instructions or codes in the machine-readable storage medium to implement the above method.

[0007] Based on the above aspects, the embodiments of the present application collect multi-source monitoring data streams including at least two types of real-time monitoring data and corresponding spatio-temporal tag information, extract spatio-temporal features from the multi-source monitoring data streams and generate a spatio-temporal feature matrix with hierarchical association relationships, mining the potential spatio-temporal connections between the data. Each element in the spatio-temporal feature matrix corresponds to the monitoring index features within a specific location and time window, enabling the multi-source data to be presented in a structured, hierarchical and easily analyzable form, improving the efficiency and depth of data processing. Furthermore, based on a preset anomaly recognition network, dynamic pattern analysis is performed on the spatio-temporal feature matrix, which can not only accurately determine the potential anomaly events existing in the target monitoring area, but also depict the anomaly propagation path, clearly indicating the diffusion direction and influence range of the potential anomaly events in the area. Then, according to the topological structure of the anomaly propagation path and the attribute parameters of the potential anomaly events, a dynamic warning strategy adapted to the target monitoring area is generated. This dynamic warning strategy takes into account the differences in different locations and different time stages, and the set of differential warning trigger conditions and response instructions can be accurately implemented according to the actual situation, significantly improving the accuracy and effectiveness of the warning, ensuring that warnings can be issued and actions can be taken in a timely and appropriate manner in different scenarios. Finally, based on the historical warning feedback data, the dynamic warning strategy is optimized and adjusted in real time. By continuously absorbing historical feedback information, it can dynamically adapt to the complex and changeable situations in the target monitoring area, continuously optimize the warning strategy, and ensure that the warning strategy output to the terminal device cluster always maintains the optimal performance. BRIEF DESCRIPTION OF THE DRAWINGS

[0008] Figure 1 is a schematic execution flow diagram of the monitoring and warning analysis method based on artificial intelligence provided by the embodiment of the present invention.

[0009] Figure 2 is a schematic hardware architecture diagram of the server provided by the embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0010] The present invention will be specifically described below with reference to the accompanying drawings of the specification. Figure 1It is a schematic flowchart of a monitoring and early warning analysis method based on artificial intelligence provided by an embodiment of the present invention. The monitoring and early warning analysis method based on artificial intelligence will be introduced in detail below.

[0011] Step S110: Collect multi-source monitoring data streams in the target monitoring area. The multi-source monitoring data streams include at least two types of real-time monitoring data and corresponding spatio-temporal tag information. The spatio-temporal tag information is used to describe the collection location and timestamp of the real-time monitoring data in the target monitoring area.

[0012] In this embodiment, taking an industrial park as an example of the target monitoring area, there are various types of real-time monitoring data sources in the industrial park. First, there are environmental monitoring sensors distributed in every corner of the park, including air quality sensors, temperature sensors, and humidity sensors. The air quality sensors can monitor the concentration of various pollutants in the air, such as sulfur dioxide, nitrogen oxides, and particulate matter; the temperature sensors and humidity sensors measure the temperature and humidity of the surrounding environment respectively. Among them, the above sensors can collect data every 10 minutes, and record the collection location (for example, through the positioning module of the sensor itself or a pre-set location identifier, such as near Building 3 in Area A) and the timestamp (accurate to seconds) while collecting data.

[0013] In addition, some monitoring devices related to production equipment can also be configured. For example, in each factory workshop, there are sensors for monitoring the operating status of production equipment, and these sensors can monitor key operating parameters such as the vibration frequency, current intensity, and oil temperature of the equipment. For large power equipment, data is collected every 5 minutes, while for some relatively small auxiliary equipment, data is collected every 15 minutes. Similarly, these equipment monitoring data also carry the collection location (such as Production Line 5 in Workshop B) and timestamp information.

[0014] In addition, a security monitoring system can also be configured. Cameras are distributed at the entrances and exits, main channels, and some key areas of the park. The cameras can not only collect video image data in real time, but also obtain the collection time of the images and the location information of the cameras. For example, the camera at the park gate collects one frame of image data every 1 second. The above different types of monitoring data from the environment, production equipment, and security monitoring, as well as their corresponding spatio-temporal tag information, together constitute the multi-source monitoring data streams in the target monitoring area.

[0015] Step S120: Extract spatio-temporal features from the multi-source monitoring data streams to generate a spatio-temporal feature matrix with hierarchical association relationships. Each element in the spatio-temporal feature matrix corresponds to the monitoring index feature of the target location in the target monitoring area within the target time window.

[0016] Specifically, for the multi-source monitoring data stream collected from the industrial park, it is first divided into multiple spatio-temporal data blocks according to the spatio-temporal tag information. For example, taking one hour as a time interval, all the sensor data in Area A are divided into a spatio-temporal data block, which corresponds to the monitoring data set of Area A within this one hour.

[0017] Feature extraction is performed for each spatio-temporal data block. Taking the environmental monitoring spatio-temporal data block in Area A as an example, there is a coupling relationship among three different types of monitoring data, namely air quality, temperature, and humidity, under the same spatio-temporal dimension. For instance, in the case of high temperature and high humidity, the chemical reaction rate of some pollutants may increase, thus affecting air quality. By analyzing a large amount of historical data and real-time data, this cross-modal correlation feature is extracted. At the same time, the time series fluctuation characteristics of the monitoring data within this spatio-temporal data block are considered, such as the sudden increase in temperature or the sharp deterioration of air quality within a certain time period. The cross-modal correlation feature and the time series fluctuation feature are fused to generate the local spatio-temporal feature vector corresponding to this spatio-temporal data block.

[0018] Based on the physical connection relationships among various regions in the industrial park (such as road connections, pipeline connections, etc.) and historical event propagation records (for example, the situation where a fire once spread along the ventilation pipeline), a global spatio-temporal association map of the entire industrial park is constructed. The nodes in the global spatio-temporal association map represent each preset sub-region (such as each workshop, warehouse, office building, etc.), and the edges represent the event propagation probability and association strength between sub-regions. For example, Workshop A and Workshop B are connected by a material transportation channel. According to historical data statistics, there is a 30% probability that a fire in Workshop A will spread to Workshop B through this channel. This 30% is the event propagation probability, and the association strength may also be affected by factors such as the width of the channel and fire prevention facilities.

[0019] Based on this global spatio-temporal association map, spatial weight allocation and time series alignment are performed for each local spatio-temporal feature vector. For example, for the area near the center of the park, due to its relatively high density of personnel and equipment, a higher spatial weight may be assigned; for time series alignment, ensure the temporal consistency of different spatio-temporal data blocks to generate a hierarchical structure of the spatio-temporal feature matrix, which includes a basic feature layer, a spatio-temporal association layer, and an event prediction layer. In the basic feature layer, the original monitoring index statistical values of each sub-region are stored, such as the average temperature and air quality index in Area A; the spatio-temporal association layer stores the spatio-temporal dependence relationship coefficients between sub-regions, such as the correlation coefficient between Workshop A and adjacent workshops; the event prediction layer stores the predicted values of the occurrence probability of abnormal events trained based on historical event data, for example, predicting the probability of a fire occurring in Area A in the future according to past fire records.

[0020] Step S130, perform dynamic pattern analysis on the spatio-temporal feature matrix based on a preset anomaly recognition network to determine potential anomaly events existing in the target monitoring area and their corresponding anomaly propagation paths, where the anomaly propagation path is used to indicate the diffusion direction and influence range of the potential anomaly event in the target monitoring area.

[0021] Input the spatio-temporal feature matrix of the industrial park into a preset anomaly recognition network, which includes a spatio-temporal convolution module, a long short-term memory module, and a graph neural network module.

[0022] First, the spatio-temporal feature matrix is sliced along the time dimension into multiple segments and respectively input into the spatio-temporal convolution module for feature dimensionality reduction and pattern extraction. For example, for a spatio-temporal feature matrix containing monitoring data within a day, it is sliced according to hourly data segments and then input into the spatio-temporal convolution module. The spatio-temporal convolution module can extract local spatio-temporal patterns of the monitoring data. For instance, within a certain time period, the air quality index in Area A shows a specific fluctuation pattern, which may be related to abnormal production emissions from a nearby factory.

[0023] After reorganizing the output of the spatio-temporal convolution module along the spatial dimension, input it into the graph neural network module to generate spatial correlation feature vectors for each sub-region. For example, the graph neural network module can analyze the spatial correlation features between Workshop A and Warehouse B because there is a material transportation relationship between them, and this relationship may affect the propagation of anomaly events between them.

[0024] Input the spatial correlation feature vectors in chronological order into the long short-term memory module to generate a global spatio-temporal state representation of the industrial park. The long short-term memory module can capture long-term dependencies in the time series. For example, it can find that within several consecutive days, the power consumption in the park shows a gradually increasing trend, which may be due to the continuous abnormal operation of some large equipment.

[0025] Fuse the output features of the spatio-temporal convolution module, the graph neural network module, and the long short-term memory module to generate a comprehensive discriminant index for anomaly event recognition. Suppose it is found through this comprehensive discriminant index that the oil temperature of a certain large production equipment in Area A is too high, and this anomaly has persisted for some time, and at the same time, the temperature in the surrounding area also shows an upward trend, then this is determined as a potential anomaly event.

[0026] Then determine the anomaly propagation path. Dynamically allocate weights to the anomaly pattern features through the attention allocation module of the anomaly recognition network to determine the anomaly confidence of each sub-region in the industrial park and the correlation strength between anomaly events. For example, the confidence of the anomaly event of the too-high oil temperature of the equipment in Area A is 0.8, and it is found that there is a relatively high correlation strength with the adjacent Area B because the ventilation system in Area B is connected to Area A.

[0027] Construct an abnormal event propagation network based on the abnormal confidence level and the association strength, where nodes represent sub-regions where the detected abnormal confidence level exceeds a set confidence level (such as 0.6), and edges represent the propagation possibility of abnormal events between sub-regions. For example, there is an edge between Area A and Area B, indicating that the abnormal event of excessive oil temperature of the equipment has a high possibility of propagating from Area A to Area B.

[0028] Based on the node degree distribution and edge weight distribution of the abnormal event propagation network, identify the core occurrence location of potential abnormal events (such as the location of the large production equipment in Area A) and the key nodes of the propagation path (such as the entrance of the ventilation duct connecting Area A and Area B, etc.).

[0029] Finally, optimize the path connection of the key nodes of the propagation path by combining the time continuity of the spatio-temporal tag information. For example, extract the timestamp sequence in the spatio-temporal tag information corresponding to the key nodes. Assume that the timestamp of the abnormal oil temperature of the equipment in Area A is 10 am, and the timestamp of the temperature rise in Area B is 10:15 am. Arrange these key nodes in the order of timestamps. Calculate the timestamp interval between adjacent nodes and find that the time interval from Area A to Area B is 15 minutes, which is less than the preset time continuity threshold (such as 30 minutes), then divide them into the same time continuous group. Traverse the spatial coordinate information of the nodes in this time continuous group, calculate the spatial distance matrix between the nodes in the group, and find that the spatial distance between the entrance of the ventilation duct connecting Area A and Area B and the temperature sensor in Area B is less than the preset spatial proximity threshold (such as 10 meters), then mark them as a pair of spatially adjacent nodes, and generate a spatial proximity relationship graph based on the spatial distance matrix. Based on this spatial proximity relationship graph, linearly connect the equipment in Area A, the entrance of the ventilation duct, and the temperature sensor in Area B in the order of timestamps to form an initial propagation path segment. Detect the spatio-temporal conflicts between the initial propagation path segments generated by different time continuous groups. For example, it is found that another propagation path segment has a timestamp overlap conflict with the current path segment in Area B. According to the rules, preferentially retain the path segment with a later timestamp sorting and remove the overlapping part. After a series of adjustments, generate the complete topological structure of the abnormal propagation path, which includes path branch information (such as the equipment in Area A may also affect Area C through the power line), the node connection order (such as passing through the entrance of the ventilation duct first and then reaching the temperature sensor in Area B), and the path propagation direction weight parameter (such as the propagation weight from Area A to Area B is higher).

[0030] Step S140, generate a dynamic early warning strategy adapted to the target monitoring area according to the topological structure of the abnormal propagation path and the attribute parameters of the potential abnormal event, where the dynamic early warning strategy includes differential early warning trigger conditions and response instruction sets for different locations and different time stages.

[0031] Specifically, calculate the priority index of each sub-region in the potential abnormal event propagation chain according to the topological structure of the abnormal propagation path in the industrial park. For example, for the abnormal event of excessively high oil temperature of the equipment in Area A, the priority index of the sub-regions directly connected to the equipment (such as Area B) is relatively high because they are affected more urgently.

[0032] Determine the trigger threshold range of the dynamic early warning strategy based on the attribute parameters of the potential abnormal event. Suppose the abnormal intensity of the abnormal event of excessively high oil temperature of the equipment is that the oil temperature exceeds the normal range by 30%, the propagation speed is that it may affect the area within 10 meters around per hour, and the influence duration is expected to be 2 hours. According to these attribute parameters, set the early warning trigger threshold range. For example, when the oil temperature exceeds the normal range by 20%, enter the early warning preparation stage.

[0033] Generate multi-level early warning trigger rules according to the priority index and the trigger threshold range. For Area B with a high priority, when the oil temperature exceeds the normal range by 20%, trigger a first-level early warning, the early warning level is high, and the response delay time is 0 minutes; while for Area C with a relatively far distance and a lower priority, when the oil temperature exceeds the normal range by 25%, trigger a second-level early warning, the early warning level is medium, and the response delay time is 5 minutes.

[0034] Construct a set of response instructions that match the multi-level early warning trigger rules. For the first-level early warning, the equipment control instruction may be to immediately stop the operation of the relevant equipment in Area A, the personnel scheduling instruction is to dispatch maintenance personnel and security personnel to Area A and Area B, and the data reporting protocol requires reporting the equipment status and environmental data every 30 seconds; for the second-level early warning, the equipment control instruction may be to reduce the operating power of the relevant equipment in Area C, the personnel scheduling instruction is to notify the staff in Area C to pay attention to observing the equipment status, and the data reporting protocol requires reporting data every 5 minutes.

[0035] Dynamically bind the multi-level early warning trigger rules and the set of response instructions in the time sequence of the abnormal propagation path to form an execution logic chain of the dynamic early warning strategy. For example, first execute the instruction to stop the operation of the equipment in Area A, then perform subsequent operations according to the situation in Area B, and then execute the corresponding instructions in sequence according to the early warning situation in Area C.

[0036] Step S150, perform real-time optimization and adjustment on the dynamic early warning strategy based on the historical early warning feedback data, and output the optimized dynamic early warning strategy to the terminal device cluster corresponding to the target monitoring area.

[0037] Specifically, the feedback log data generated when the terminal device cluster in the industrial park executes the historical warning strategy can be obtained. For example, during the last abnormal event of high oil temperature of the equipment in Area A, the warning response delay time in Area B was actually 2 minutes (while it was set to 0 minutes). The equipment execution status showed that after the equipment in Area A stopped running, it had a certain additional impact on the entire production line (such as some semi-finished products were affected). The actual impact range of the abnormal event was slightly larger than predicted and affected a small warehouse next to Area B.

[0038] Extract the policy execution effect indicators from the feedback log data, such as the warning accuracy rate (it is found that there is a deviation between the actual warning situation and the expectation, and the accuracy rate has decreased), the response efficiency deviation (the response in Area B was delayed by 2 minutes), and the resource consumption rate (the equipment stop operation led to some additional resource waste, such as the waste of raw materials, etc.).

[0039] Compare the policy execution effect indicators with the preset optimization objective function. Assume that the optimization objective function includes a warning coverage efficiency term (positively correlated with the warning accuracy rate), a resource consumption penalty term (negatively correlated with the energy consumption and computing resource occupancy when the equipment executes the warning instruction), and a response timeliness reward term (positively correlated with the warning response speed). It is found that the warning accuracy rate decreases, the resource consumption increases, and the response speed slows down. Determine the trigger condition parameters and response instruction priorities that need to be adjusted in the dynamic warning strategy.

[0040] Iteratively correct the trigger threshold range and response delay time of the dynamic warning strategy according to the gradient direction of the optimization objective function. For example, adjust the first-level warning trigger condition in Area B to the oil temperature exceeding the normal range by 22%, and adjust the response delay time to 1 minute.

[0041] Based on the corrected parameters, recalculate the coverage range and resource allocation weights of each level of warning in the multi-level warning trigger rule, and generate an optimized dynamic warning strategy. For example, re-evaluate the warning levels and resource allocation situations of different regions under the new trigger conditions to ensure that when a similar abnormal event occurs next time, the warning strategy can be executed more effectively and accurately. Then output the optimized dynamic warning strategy to each relevant terminal device cluster in the industrial park (such as the controllers of production equipment, security monitoring equipment, etc.).

[0042] Based on the above steps, the embodiments of the present application collect a multi-source monitoring data stream including at least two types of real-time monitoring data and corresponding spatio-temporal tag information, extract spatio-temporal features from the multi-source monitoring data stream and generate a spatio-temporal feature matrix with hierarchical association relationships, mining the potential spatio-temporal connections between the data. Each element in the spatio-temporal feature matrix corresponds to the monitoring index features within a specific location and time window, enabling the multi-source data to be presented in a structured, hierarchical and easy-to-analyze form, improving the efficiency and depth of data processing. Furthermore, based on a preset anomaly recognition network, dynamic pattern analysis is performed on the spatio-temporal feature matrix, which can not only accurately determine the potential anomaly events existing in the target monitoring area, but also depict the anomaly propagation path, clearly indicating the diffusion direction and influence range of the potential anomaly events in the area. Then, according to the topological structure of the anomaly propagation path and the attribute parameters of the potential anomaly events, a dynamic early warning strategy adapted to the target monitoring area is generated. This dynamic early warning strategy takes into account the differences in different positions and different time stages, and the set of differential early warning trigger conditions and response instructions can be accurately implemented according to the actual situation, significantly improving the accuracy and effectiveness of the early warning, ensuring that warnings can be issued and actions can be taken in a timely and appropriate manner in different scenarios. Finally, based on the historical warning feedback data, the dynamic early warning strategy is optimized and adjusted in real time. By continuously absorbing historical feedback information, it can dynamically adapt to the complex and changeable situations in the target monitoring area, continuously optimize the early warning strategy, and ensure that the early warning strategy output to the terminal device cluster always maintains the optimal performance.

[0043] In a possible implementation manner, step S120 includes: Step S121, dividing the multi-source monitoring data stream into multiple spatio-temporal data blocks according to the spatio-temporal tag information, and each spatio-temporal data block corresponds to a monitoring data set of a preset sub-region in the target monitoring area within a continuous time series.

[0044] Specifically, each part in the industrial park is set as a preset sub-region, such as each workshop, warehouse, office building, as well as the roads and greening areas in the park. The spatio-temporal tag information details the collection location and time stamp of the monitoring data. Taking Workshop A as an example, the data collected by various sensors inside it (such as equipment operation status sensors, environmental sensors, etc.) within a period of time (such as 8 hours on a working day) can be divided into a spatio-temporal data block according to the spatio-temporal tags. This spatio-temporal data block contains all the monitoring data sets of Workshop A within these 8 hours, covering operation parameters such as the vibration frequency, current intensity, and oil temperature of the equipment, as well as environmental data such as the temperature, humidity, and air quality inside the workshop. Similarly, other sub-regions such as Warehouse B and Office Building C also have their corresponding spatio-temporal data blocks, which are divided according to the spatio-temporal tags of the data collected by the sensors within their respective regions.

[0045] Step S122: For each of the spatio-temporal data blocks, extract the cross-modal correlation features among the monitoring data included in the spatio-temporal data block, where the cross-modal correlation features are used to describe the coupling relationship between different types of monitoring data in the same spatio-temporal dimension.

[0046] Still taking the spatio-temporal data block of Workshop A as an example, there are cross-modal correlation features between the equipment operation status data and the environmental data in the workshop. During the operation of the equipment, the increase in oil temperature may cause the temperature in the workshop to rise, which is the coupling relationship between the equipment operation status and the environmental temperature. When the vibration frequency of the equipment increases, it may cause changes in the air flow around, thus affecting the air quality, which reflects the correlation between the equipment operation status and the air quality. Moreover, when the humidity in the workshop is relatively high, condensation may occur on the surface of the equipment, which will affect the electrical performance of the equipment and cause fluctuations in the current intensity. This is also the coupling relationship between different types of monitoring data (humidity and current intensity) in the same spatio-temporal dimension.

[0047] Step S123: Integrate the cross-modal correlation features with the time series fluctuation features of the spatio-temporal data block to generate the local spatio-temporal feature vector corresponding to the spatio-temporal data block.

[0048] Specifically, in the spatio-temporal data block of Workshop A, the time series fluctuation features are very obvious. For example, during the period from 9 am to 10 am, due to the concentrated arrangement of production tasks, the operation intensity of the equipment increases, and parameters such as the oil temperature and vibration frequency of the equipment will show an upward trend. At the same time, the temperature in the workshop also rises, and the air quality may slightly decrease due to the increase in equipment emissions. Integrate this time series fluctuation feature with the previously extracted cross-modal correlation features. For example, the cross-modal correlation feature that the increase in equipment oil temperature leads to the rise in temperature is more significant during the period from 9 am to 10 am. Combine the intensity, trend and other information of this cross-modal correlation feature in this time period with the time series fluctuation features to generate the local spatio-temporal feature vector corresponding to the spatio-temporal data block of Workshop A. This vector can comprehensively describe the comprehensive features of Workshop A in a specific time period, including both the internal correlations between different types of monitoring data and the changes of these correlations over time.

[0049] Step S124: Based on the physical connection relationship between the preset sub-regions and the historical event propagation records, construct the global spatio-temporal correlation map of the target monitoring area. The nodes in the global spatio-temporal correlation map represent the preset sub-regions, and the edges represent the event propagation probability and correlation intensity between the preset sub-regions.

[0050] Specifically, in an industrial park, there are various physical connection relationships among sub-regions. There is a material transportation channel between Workshop A and Warehouse B, which constitutes a physical connection relationship. According to the historical event propagation records, there was once a small fire in Workshop A. Due to the failure to close the material transportation channel in time, the smoke spread to Warehouse B through the channel. Based on the above physical connection relationship and historical event propagation records, a global spatio-temporal association graph is constructed. In this global spatio-temporal association graph, nodes represent preset sub-regions such as Workshop A, Warehouse B, and Office Building C. Edges represent the event propagation probability and association strength between preset sub-regions. For the edge between Workshop A and Warehouse B, its event propagation probability may be determined as 20% (hypothetical) based on the statistical data of historical fire events and other similar events. The association strength may be affected by factors such as the fire prevention facilities of the material transportation channel, the length and width of the channel. If the channel has good fire prevention isolation facilities, the association strength will be relatively low; if the channel is narrow and has no fire prevention facilities, the association strength will be high. For Workshop A and Office Building C, since there is no direct material transportation channel between them, but there is an electrical line connection, according to the historical records of power failure propagation, the event propagation probability is 10%, and the association strength is affected by factors such as the insulation performance and load condition of the electrical line.

[0051] Step S125, perform spatial weight assignment and time series alignment on the local spatio-temporal feature vectors according to the global spatio-temporal association graph to generate the hierarchical structure of the spatio-temporal feature matrix.

[0052] Specifically, based on the global spatio-temporal correlation map of the industrial park, the local spatio-temporal feature vectors of each sub-region (such as Workshop A, Warehouse B, etc.) are processed. For spatial weight allocation, considering that Workshop A is the main production area with intensive equipment and frequent personnel activities, once an abnormal event occurs, the impact range is relatively large, so a higher spatial weight is assigned. In contrast, for Office Building C, the density of personnel and equipment is relatively low, and a lower spatial weight is assigned. For time series alignment, ensure the temporal consistency of the spatio-temporal data blocks of each sub-region. For example, the spatio-temporal data blocks of Workshop A, Warehouse B, and Office Building C are all divided by hour in terms of time, so that the monitoring data of different sub-regions can be analyzed under the same time scale. Through such spatial weight allocation and time series alignment, a hierarchical structure of the spatio-temporal feature matrix is generated. The basic feature layer in this hierarchical structure stores the statistical values of the original monitoring indicators of each sub-region, such as the average oil temperature and average vibration frequency of the equipment in Workshop A, the average temperature and average humidity in Warehouse B, etc.; the spatio-temporal correlation layer stores the spatio-temporal dependence coefficient between sub-regions, such as the coefficient determined based on the previously calculated correlation strength and propagation probability between Workshop A and Warehouse B; the event prediction layer stores the predicted values of the probability of abnormal events occurring based on historical event data. For example, according to the frequency and propagation of abnormal events such as equipment failures and fires in Workshop A in the past, predict the probability of similar abnormal events occurring in Workshop A in the future. This hierarchical structure can comprehensively and systematically describe the spatio-temporal characteristics in the industrial park, providing a strong data basis for subsequent abnormal identification, early warning strategy generation, etc.

[0053] In a possible implementation manner, step S130 includes: Step S131, input the spatio-temporal feature matrix into the multi-scale feature extraction module of the abnormal identification network to obtain the abnormal pattern features of the target monitoring area at different time granularities and spatial resolutions.

[0054] Specifically, for an industrial park, the time granularity can be in hours, minutes, or even seconds, and the spatial resolution can be at the workshop level, equipment level, or smaller regional units. Taking Workshop A as an example, at the hourly time granularity, it may be found that the overall operating parameters of the equipment in the workshop show abnormal fluctuations within a certain hour, which is an abnormal pattern feature at a coarser time granularity. At the minute-level time granularity, it may be found that the oil temperature of a certain key equipment rises sharply within a minute, which is an abnormal pattern feature at a finer time granularity. From the perspective of spatial resolution, at the workshop level, it may be found that the power consumption of the entire Workshop A exceeds the normal range within a certain period of time, which is an abnormal pattern feature at the workshop-level spatial resolution; at the equipment-level spatial resolution, it may be determined that the abnormal increase in the current of a certain large production equipment leads to the abnormal power consumption at the workshop level. At the same time, between different sub-regions such as workshops and warehouses, such as between Workshop A and Warehouse B, abnormal pattern features in aspects such as material transportation volume and environmental parameters may be found at different time and spatial resolutions. These abnormal pattern features cover information at all levels from macro to micro in the industrial park, providing a comprehensive data basis for subsequent analysis.

[0055] Step S132: Dynamically assign weights to the abnormal pattern features through the attention allocation module of the abnormal recognition network, and determine the abnormal confidence levels of each sub-region within the target monitoring area and the correlation strength between abnormal events.

[0056] Specifically, in the industrial park, for the abnormal pattern feature of the abnormal rise in the oil temperature of a certain equipment in Workshop A, the attention allocation module will assign weights according to various factors. If this equipment is a key production equipment, and the increase in oil temperature is large, and the surrounding environmental temperature is also significantly affected, then a higher value, such as 0.8, will be assigned to the abnormal confidence level of the sub-region of Workshop A. For Warehouse B, if only a slight fluctuation in humidity is found and the correlation with other sub-regions is small, its abnormal confidence level may be assigned as 0.2. For the correlation strength between abnormal events, for example, the correlation between the abnormal oil temperature of the equipment in Workshop A and Warehouse B, if there is a material transportation channel and a ventilation system connection between Workshop A and Warehouse B, then the correlation strength between them may be relatively high, set as 0.6; if there is only a weak power correlation between Office Building C and Workshop A, its correlation strength may be only 0.1.

[0057] Step S133: Construct an abnormal event propagation network based on the abnormal confidence levels and correlation strengths. The nodes in the abnormal event propagation network represent the sub-regions where the detected abnormal confidence levels exceed the set confidence level, and the edges represent the possibility of abnormal event propagation between sub-regions.

[0058] Specifically, in the abnormal event propagation network, a node represents a sub-region where the detected abnormal confidence exceeds a set confidence level (such as 0.5). In the industrial park, Workshop A becomes a node because the abnormal confidence of the equipment oil temperature is 0.8, which exceeds the set value. If Warehouse B becomes a node because its comprehensive abnormal confidence exceeds 0.5 due to its association with Workshop A. An edge represents the possibility of abnormal event propagation between sub-regions. The edge between Workshop A and Warehouse B represents the possibility of the equipment oil temperature abnormal event propagating from Workshop A to Warehouse B, and this possibility is determined based on factors such as the previously calculated association strength and historical event propagation data, such as 0.6. For the connection between Workshop A and Office Building C, since the association strength is low, the possibility of the abnormal event propagating from Workshop A to Office Building C may be only 0.1, which is also represented by the edge between them.

[0059] Step S134, based on the node degree distribution and edge weight distribution of the abnormal event propagation network, identify the core occurrence location of the potential abnormal event and the key nodes of the propagation path.

[0060] Specifically, in the abnormal event propagation network of the industrial park, the node degree represents the number of edges connected to the node. If the node degree of Workshop A is high, it indicates that it is associated with multiple sub-regions, which may mean that Workshop A is the core occurrence location of the potential abnormal event because the abnormal may propagate in multiple directions from here. For the key nodes of the propagation path, for example, the entrance of the material transportation channel between Workshop A and Warehouse B, the connection point of the ventilation system, etc. may become key nodes because these locations play a key role in connecting and conducting during the abnormal event propagation. The edge weight distribution also helps to identify key nodes. If the weight of an edge is high, such as the edge weight from Workshop A to Warehouse B is 0.6, then the connection point on this edge (such as the entrance of the material transportation channel) is more likely to be a key node.

[0061] Step S135, combine the time continuity of the spatio-temporal label information to optimize the path connection of the key nodes of the propagation path, and generate the complete topological structure of the abnormal propagation path.

[0062] In a possible implementation manner, step S135 includes: Step S1351, extract the time stamp sequence in the spatio-temporal label information corresponding to the key nodes, and arrange the key nodes in a time-ordered node queue according to the chronological order of the time stamp sequence. Each node in the time-ordered node queue carries spatial coordinate information and the corresponding time stamp.

[0063] Specifically, assume that the timestamp of the abnormal oil temperature of the equipment in Workshop A is 10:00 am, the timestamp of the detected abnormal air flow change at the connection point of the ventilation system between this equipment and Warehouse B is 10:10 am, and the timestamp of the abnormal temperature increase in Warehouse B is 10:15 am. Arrange these key nodes in a time-ordered node queue according to the chronological order of the timestamp sequence. Each node carries spatial coordinate information and the corresponding timestamp. For example, the spatial coordinates (x1, y1, z1) of the equipment in Workshop A and the timestamp 10:00 am, the spatial coordinates (x2, y2, z2) of the connection point of the ventilation system and the timestamp 10:10 am, and the spatial coordinates (x3, y3, z3) of the temperature sensor in Warehouse B and the timestamp 10:15 am.

[0064] Step S1352, calculate the timestamp intervals between adjacent nodes in the time-ordered node queue, divide the adjacent nodes with timestamp intervals less than the preset time continuity threshold into the same time continuous group, and record the start timestamp and end timestamp of each time continuous group.

[0065] Assume that the time continuity threshold is 15 minutes. In the above example, the timestamp intervals between the equipment in Workshop A, the connection point of the ventilation system, and the temperature sensor in Warehouse B are 10 minutes and 5 minutes respectively, both of which are less than 15 minutes. Therefore, they can be divided into the same time continuous group, with the start timestamp being 10:00 am and the end timestamp being 10:15 am.

[0066] Step S1353, traverse the node spatial coordinate information within each time continuous group, calculate the spatial distance matrix between the nodes in the group, mark the node pairs with spatial distances less than the preset spatial proximity threshold as spatially adjacent node pairs, and generate a spatial proximity relationship graph based on the spatial distance matrix.

[0067] Specifically, the spatial distance between the equipment in Workshop A and the connection point of the ventilation system is d1, and the spatial distance between the connection point of the ventilation system and the temperature sensor in Warehouse B is d2. Calculate the spatial distance matrix based on the spatial distances. Mark the node pairs with spatial distances less than the preset spatial proximity threshold (such as 10 meters) as spatially adjacent node pairs, and generate a spatial proximity relationship graph based on the spatial distance matrix. If the distance d1 between the equipment in Workshop A and the connection point of the ventilation system is less than 10 meters, and the distance d2 between the connection point of the ventilation system and the temperature sensor in Warehouse B is also less than 10 meters, then these two pairs of nodes are both marked as spatially adjacent node pairs, and a spatial proximity relationship graph is generated based on this information.

[0068] Step S1354, based on the spatial proximity relationship graph, linearly connect the nodes with spatial proximity relationships within the time continuous group in timestamp order to form an initial propagation path segment, and the initial propagation path segment includes path direction information and path connection strength parameters.

[0069] For example, the equipment in Workshop A, the connection points of the ventilation system, and the temperature sensors in Warehouse B are linearly connected in the order of the time stamps. The path direction is from the equipment in Workshop A to the temperature sensors in Warehouse B. The path connection strength parameter can be determined according to factors such as the correlation strength and spatial distance calculated previously.

[0070] Step S1355: Detect the spatio-temporal conflicts between the initial propagation path segments generated by different consecutive time groups. The spatio-temporal conflicts include the time stamp overlap conflict of the path segments in the same spatial region and the topological contradiction that the path direction does not match the spatial proximity relationship map.

[0071] For example, assume that there is another propagation path segment involving the power supply system in Workshop A, from the power distribution room in Workshop A to the power equipment in Office Building C. This path segment may have spatio-temporal conflicts with the previous path segment from the equipment in Workshop A to Warehouse B. The spatio-temporal conflicts include the time stamp overlap conflict of the path segments in the same spatial region and the topological contradiction that the path direction does not match the spatial proximity relationship map. If events occur at a certain time point in the area near Warehouse B for both of these path segments, there is a time stamp overlap conflict; if, according to the spatial proximity relationship map, the connection direction of a certain path segment does not match the actual physical layout or logical relationship, there is a topological contradiction.

[0072] Step S1356: Dynamically adjust the conflicting path segments according to the detected spatio-temporal conflict types to generate the adjusted initial propagation path segments, specifically including: for the time stamp overlap conflict, preferentially retain the path segment with a later time stamp order and remove the overlapping part. For the topological contradiction, recalculate the connection direction of the path segment based on the spatial proximity relationship map.

[0073] Assume that the time stamp of the path segment from the power distribution room to the power equipment in Office Building C near Warehouse B is 10:20 am, and the time stamp of the path segment from the equipment in Workshop A to the temperature sensors in Warehouse B near Warehouse B is from 10:15 am to 10:25 am. Then, preferentially retain the part of the path segment from the power distribution room to the power equipment in Office Building C near Warehouse B, and remove the overlapping part of the path segment from the equipment in Workshop A to the temperature sensors in Warehouse B after 10:20 am. For the topological contradiction, recalculate the connection direction of the path segment based on the spatial proximity relationship map. If the original connection direction of a certain path segment is opposite to the reasonable direction shown in the spatial proximity relationship map, re-determine the correct connection direction according to the map.

[0074] Step S1357: Globally sort the adjusted initial propagation path segments according to the start time stamps of the time consecutive groups, splice the path segments head to tail according to the sorting result, and verify whether the time stamp interval and spatial jump distance of adjacent path segments meet the time continuity threshold and spatial proximity threshold during the splicing process.

[0075] Step S1358: When the time - stamp interval or spatial jump distance between adjacent path segments exceeds the threshold, insert a transition node and update the spatial proximity relationship map. The spatial coordinates of the transition node are generated by interpolating the spatial coordinates of the end nodes of the adjacent path segments, and the time - stamp is linearly assigned according to the time - stamp interval between the adjacent path segments.

[0076] Suppose there are three path segments after adjustment, starting at 10:00 am, 10:30 am, and 11:00 am respectively. After global sorting according to the start time - stamps, first splice the path segments starting at 10:00 am and 10:30 am end - to - end, and verify whether the time - stamp interval between them is less than 15 minutes and the spatial jump distance is less than 10 meters. If the conditions are met, continue to splice the next path segment; if not, when the time - stamp interval or spatial jump distance between adjacent path segments exceeds the threshold, insert a transition node and update the spatial proximity relationship map. For example, the time - stamp interval between the path segment starting at 10:30 am and the path segment starting at 11:00 am is 30 minutes, which exceeds 15 minutes, and the spatial jump distance is 15 meters, which exceeds 10 meters. At this time, insert a transition node. The spatial coordinates of the transition node are generated by interpolating the spatial coordinates of the end nodes of the adjacent path segments. Suppose the spatial coordinates of the end node of the path segment at 10:30 am are (x4, y4, z4), and the spatial coordinates of the start node of the path segment at 11:00 am are (x5, y5, z5). Calculate the spatial coordinates (x6, y6, z6) of the transition node according to the interpolation algorithm, and the time - stamp is linearly assigned as 10:45 am according to the time - stamp interval between the adjacent path segments.

[0077] Step S1359: Based on the finally - spliced path - segment set and the inserted transition nodes, generate the complete topological structure of the abnormal propagation path. The complete topological structure includes path - branch information, node connection order, and path - propagation direction weight parameters.

[0078] For example, starting from the equipment in Workshop A, through the ventilation - system connection point to the temperature sensor in Warehouse B is one path - branch, and from the power distribution room in Workshop A to the power equipment in Office Building C is another path - branch. The node connection order is determined according to the previous splicing order, such as first the equipment in Workshop A, then the ventilation - system connection point, and then the temperature sensor in Warehouse B, etc. The path - propagation direction weight parameters are determined according to factors such as the previously calculated correlation strength and spatial distance. For example, the propagation - direction weight from the equipment in Workshop A to the ventilation - system connection point is 0.8, and the propagation - direction weight from the ventilation - system connection point to the temperature sensor in Warehouse B is 0.6, etc. This complete topological structure can accurately describe the propagation path of potential abnormal events in the industrial park.

[0079] In a possible implementation manner, step S140 includes: Step S141: Calculate the priority index of each sub-region in the potential abnormal event propagation chain according to the topological structure of the abnormal propagation path. The priority index is used to quantify the urgency of the impact of the abnormal event on the sub-region.

[0080] Specifically, in the industrial park, continuing with the previous example of the abnormal oil temperature of the equipment in Workshop A, if the abnormal propagation path shows that it can spread from Workshop A to Warehouse B through the ventilation system connection point, and then to some small equipment storage areas adjacent to Warehouse B. As the source of the abnormal event, Workshop A's equipment is directly affected by the abnormal oil temperature, which may lead to equipment damage, production stagnation, or even serious consequences such as fire. Therefore, the priority index of Workshop A is the highest, set at 0.9. Since Warehouse B has a direct propagation path connection with Workshop A and the materials stored in the warehouse may be affected by fire or equipment failure, its priority index is the second highest, set at 0.7. For those small equipment storage areas, although they are at a slightly greater distance and the probability of being affected is relatively small, there is still a risk, and their priority index may be 0.5. The calculation of this priority index comprehensively considers the distance between the sub-region and the abnormal source, the tightness of the propagation path connection, and the importance of the sub-region itself (such as the value of the stored materials, the criticality of the equipment, etc.).

[0081] Step S142: Determine the trigger threshold range of the dynamic early warning strategy based on the attribute parameters of the potential abnormal event. The attribute parameters include abnormal intensity, propagation speed, and impact duration.

[0082] Specifically, for the potential abnormal event of the abnormal oil temperature of the equipment in Workshop A, its abnormal intensity can be measured by the percentage of the oil temperature exceeding the normal range. For example, when the oil temperature exceeds the normal range by 30%, it is a high-intensity abnormal. The propagation speed is estimated based on historical data and current monitoring data. Assuming that according to factors such as the air flow speed of the ventilation system and the material transportation speed, it is estimated that the abnormal event may affect the surrounding area within a range of 10 meters per hour. The impact duration is expected to be 2 hours, which is estimated based on the operating limit time of the equipment under abnormal oil temperature and the attenuation model of the abnormal event during the propagation process. Based on these attribute parameters such as abnormal intensity, propagation speed, and impact duration, the trigger threshold range of the early warning is set. For example, when the oil temperature exceeds the normal range by 15%, it enters the early warning preparation stage; when the oil temperature exceeds the normal range by 20%, the formal early warning is triggered.

[0083] Step S143: Generate a multi-level early warning trigger rule according to the priority index and the trigger threshold range. The multi-level early warning trigger rule defines the early warning level and response delay time of different priority sub-regions under different abnormal intensities.

[0084] Specifically, for Workshop A, since the priority index is 0.9, when the oil temperature exceeds the normal range by 20%, a first-level warning is triggered. The warning level is high, and the response delay time is 0 minutes, which means that once the threshold is reached, the operations related to the warning are immediately executed. For Warehouse B, the priority index is 0.7. When the oil temperature exceeds the normal range by 20%, a second-level warning is triggered. The warning level is medium, and the response delay time is 3 minutes. Since the urgency of the impact on Warehouse B is slightly lower than that of Workshop A, a 3-minute response delay time is given for preliminary assessment and preparation work. For the small equipment storage area, the priority index is 0.5. When the oil temperature exceeds the normal range by 25%, a third-level warning is triggered. The warning level is low, and the response delay time is 5 minutes. Here, due to the relatively low possibility and urgency of being affected, the warning trigger threshold is relatively high, and the response delay time is also relatively long.

[0085] Step S144, construct a set of response instructions that match the multi-level warning trigger rules, and the set of response instructions includes equipment control instructions, personnel scheduling instructions, and data reporting protocols for each warning level.

[0086] Specifically, for the first-level warning of Workshop A, the equipment control instruction is to immediately stop the operation of relevant equipment to prevent further damage to the equipment and the deterioration of abnormal events; the personnel scheduling instruction is to dispatch maintenance personnel, safety personnel, and technical experts to Workshop A. The maintenance personnel are responsible for inspecting and repairing the equipment, the safety personnel are responsible for on-site safety control, and the technical experts are responsible for comprehensively evaluating the abnormal events; the data reporting protocol requires reporting the equipment status (including oil temperature, vibration frequency, current intensity, etc.), environmental data (temperature, humidity, air quality, etc.), and operation records of personnel every 30 seconds. For the second-level warning of Warehouse B, the equipment control instruction is to pre-check the equipment that may be affected, such as checking ventilation equipment, fire prevention facilities, etc.; the personnel scheduling instruction is to notify the warehouse management personnel and some safety personnel to conduct inspections in the warehouse to ensure the safety of materials; the data reporting protocol requires reporting data such as temperature, humidity, and smoke concentration in the warehouse every 2 minutes. For the third-level warning of the small equipment storage area, the equipment control instruction is to simply observe the status of the equipment, such as checking whether there are abnormal sounds or odors in the equipment; the personnel scheduling instruction is to notify the nearby staff to pay attention to the equipment status; the data reporting protocol requires reporting the basic operation status data of the equipment every 5 minutes.

[0087] Step S145, dynamically bind the multi-level warning trigger rules and the set of response instructions in the chronological order of the abnormal propagation path to form the execution logic chain of the dynamic warning strategy.

[0088] Specifically, in the order of the anomaly spreading from Workshop A to Warehouse B and then to the small equipment storage area, first execute the equipment control instructions, personnel scheduling instructions, and data reporting protocols related to the first-level warning in Workshop A. Then, perform corresponding operations according to the second-level warning situation in Warehouse B. Finally, execute the corresponding instructions based on the third-level warning situation in the small equipment storage area. In this way, starting from the source of the abnormal event, as the anomaly spreads, warning and response operations are carried out orderly according to the pre-set rules and instructions, forming a complete dynamic warning strategy execution logic chain.

[0089] In a possible implementation manner, step S150 includes: Step S151, obtain the feedback log data generated when the terminal device cluster executes the historical warning strategy. The feedback log data includes the warning response delay time, the device execution status, and the actual impact range of the abnormal event.

[0090] For example, in the previous abnormal event of the equipment oil temperature in Workshop A, the warning response delay time in Workshop A was recorded as 1 minute in the record (although it was set to 0 minute, but due to actual factors such as communication and personnel response, a 1-minute delay occurred). After the device execution status showed that the device stopped running, some associated devices were impacted to a certain extent due to the sudden stop. For example, the cooling system connected to this device had a short-term pressure fluctuation. The actual impact range of the abnormal event was slightly larger than the previously predicted range, not only affecting Warehouse B but also a small tool room next to Warehouse B, and some tools inside were damaged due to high temperature.

[0091] Step S152, extract the policy execution effect indicators from the feedback log data. The policy execution effect indicators include the warning accuracy rate, the response efficiency deviation, and the resource consumption rate.

[0092] For example, in terms of the warning accuracy rate, since the actual impact range exceeded the predicted range, the warning accuracy rate decreased. It was originally predicted that only Workshop A and Warehouse B would be affected, but actually, the small tool room was also affected. The response efficiency deviation was manifested as a 1-minute response delay in Workshop A, which deviated from the set 0 minute. In terms of the resource consumption rate, the sudden stop of the device caused the cooling system to need to be readjusted, consuming additional electricity and manpower to restore normal operation. At the same time, the emergency scheduling of maintenance personnel and safety personnel also consumed a certain amount of human resources.

[0093] Step S153, compare the policy execution effect indicators with the preset optimization objective function to determine the trigger condition parameters and response instruction priorities that need to be adjusted in the dynamic warning strategy.

[0094] For example, assume that the optimization objective function includes an early warning coverage efficiency term (positively correlated with the early warning accuracy rate), a resource consumption penalty term (negatively correlated with the energy consumption and computing resource occupancy when the device executes the early warning instruction), and a response timeliness reward term (positively correlated with the early warning response speed). Due to the decrease in early warning accuracy rate and the increase in resource consumption, it is necessary to adjust the early warning trigger conditions. For example, the first-level early warning trigger condition for Workshop A is adjusted from the oil temperature exceeding the normal range by 20% to 18% to improve the timeliness and coverage of early warnings; at the same time, the priority of the instruction to stop the device is reduced, and a transition instruction for the device to operate at reduced power is added to reduce the impact of the sudden stop of the device on associated devices, thereby reducing resource consumption.

[0095] Step S154, iteratively correct the trigger threshold range and response delay time of the dynamic early warning strategy according to the gradient direction of the optimization objective function to obtain the corrected parameters.

[0096] For example, by analyzing the sensitivity of the optimization objective function to each parameter, it is found that the early warning trigger threshold has a greater impact on the early warning accuracy rate and resource consumption rate, and the response delay time has a greater impact on the response efficiency deviation. According to this analysis result, after adjusting the first-level early warning trigger threshold for Workshop A to 18%, re-evaluate its impact on the early warning accuracy rate and resource consumption rate; adjust the response delay time for Workshop A from 0 minutes to 30 seconds and consider its improvement effect on the response efficiency deviation. Through multiple iterative corrections, each parameter reaches a relatively optimal value.

[0097] Step S155, recalculate the coverage range and resource allocation weights of each level of early warning in the multi-level early warning trigger rule based on the corrected parameters, and generate an optimized dynamic early warning strategy.

[0098] For example, when the new early warning trigger threshold for Workshop A is 18%, re-evaluate the coverage range of each level of early warning under different abnormal intensities at this threshold. For example, when the oil temperature exceeds the normal range by 18% - 25%, it is the coverage range of the first-level early warning, and 25% - 35% is the coverage range of a higher level of early warning (if any). At the same time, re-allocate the resource allocation weights according to the resource consumption situation. For example, allocate more maintenance resources to the equipment inspection and maintenance of Workshop A during the first-level early warning, and reduce the resource allocation ratio of Warehouse B during the second-level early warning (because the adjusted strategy may reduce the probability of Warehouse B being affected). Through such adjustments, an optimized dynamic early warning strategy is generated, enabling it to play a more efficient and accurate role in subsequent abnormal event responses.

[0099] In a possible implementation manner, before step S110, the method further includes: Step S210: Divide multiple monitoring data acquisition units according to the geographical distribution characteristics of the target monitoring area, and each acquisition unit is configured with at least three heterogeneous sensors and a data preprocessing module.

[0100] Specifically, the geographical scope of the industrial park is large and includes areas with different functions, such as production workshops, warehouses, office buildings, and park roads. Acquisition units are divided according to the characteristics of these different areas. For example, Workshop A is divided into one acquisition unit. Since there are many devices and the production process is complex in the workshop, the heterogeneous sensors set here include device operation status sensors (such as monitoring device vibration frequency, oil temperature, current intensity, etc.), environmental sensors (monitoring temperature, humidity, air quality, etc.), and personnel activity monitoring sensors (such as personnel positioning sensors, personnel flow counters, etc.). Each acquisition unit is configured with a data preprocessing module. For the acquisition unit of Workshop A, the data preprocessing module is responsible for the preliminary processing of the data collected by the sensors.

[0101] Step S220: Set the data sampling frequency and transmission protocol of the heterogeneous sensors so that the data of different types of sensors within the same acquisition unit are transmitted through independent channels after the timestamps are aligned.

[0102] For example, within the acquisition unit of Workshop A, since the device operation status changes relatively fast, the data sampling frequency of the device operation status sensor is set to collect data once every 1 minute. The environmental sensor changes relatively slowly and is set to collect data once every 5 minutes. The personnel activity monitoring sensor is set to collect data once every 30 seconds according to the characteristics of personnel flow. At the same time, to ensure that the data of different types of sensors within the same acquisition unit are transmitted through independent channels after the timestamps are aligned, a unified time synchronization protocol is adopted. For example, based on the high-precision clock source in the park, all sensors are time-calibrated according to this clock source to ensure that the timestamps of the collected data are accurately consistent. The data of the device operation status sensor is transmitted through the industrial Ethernet channel, the environmental sensor data is transmitted through the ZigBee channel, and the personnel activity monitoring sensor data is transmitted through the Wi-Fi channel. In this way, the data of different types of sensors are transmitted in their respective independent channels, avoiding mutual interference and ensuring orderly transmission after the timestamps are aligned.

[0103] Step S230: Configure a noise filtering algorithm and data integrity verification rules for the data preprocessing module. The noise filtering algorithm uses the wavelet transform method with an adaptive threshold to eliminate the environmental interference components in the sensor signals.

[0104] For example, in the acquisition unit of Workshop A, the data preprocessing module uses the wavelet transform method with an adaptive threshold to eliminate the environmental interference components in the sensor signals. For the sensors monitoring the equipment operating status, electromagnetic interference and vibration interference may be generated by electrical equipment and mechanical operation in the workshop. The wavelet transform method with an adaptive threshold can automatically adjust the threshold according to the characteristics of the sensor signals, and filter these interference components as noise. For example, the sensor for the equipment vibration frequency may be interfered by the low-frequency vibration of other equipment in the workshop, and this interference can be accurately identified and filtered through the wavelet transform algorithm. At the same time, data integrity verification rules are configured to check the integrity of the collected data. For example, check whether the length of the data meets the requirements and whether there are missing values in the data. If incomplete data is found, the data preprocessing module will mark the data and try to repair it through methods such as interpolation or notify relevant personnel for inspection.

[0105] Step S240: Establish a two-way communication link between the acquisition unit and the central server, and the two-way communication link supports the real-time upload of the multi-source monitoring data stream and the synchronous issuance of control instructions.

[0106] For example, a two-way communication link is established between the acquisition unit of Workshop A and the central server, and this link supports the real-time upload of the multi-source monitoring data stream and the synchronous issuance of control instructions. During normal operation, the data collected by the sensors in the acquisition unit of Workshop A, after being processed by the data preprocessing module, is uploaded to the central server in real time through the two-way communication link. For example, the equipment oil temperature data collected by the equipment operating status sensor, the temperature data collected by the environmental sensor, and the personnel position data collected by the personnel activity monitoring sensor are all uploaded to the central server at the set time intervals and transmission protocols. At the same time, the central server can also issue control instructions to the acquisition unit of Workshop A through this two-way communication link. For example, if the central server needs to adjust the data sampling frequency of the equipment operating status sensor in Workshop A according to the overall operation of the park, it can issue the corresponding control instructions to the acquisition unit through this link, and the acquisition unit will adjust the data sampling frequency of the sensor as required after receiving the instructions.

[0107] Step S250: Periodically perform self-check and calibration on the acquisition unit, generate a sensor health status report, and dynamically adjust the data sampling frequency and noise filtering parameters.

[0108] In this embodiment, in the acquisition unit of Workshop A, self-check calibration is performed regularly (such as at 2 am every day). During the self-check process, the device operation status sensor will perform self-detection to check performance indicators such as the sensitivity and measurement range of the sensor. The environment sensor will check its own calibration parameters, and the personnel activity monitoring sensor will test the stability of signal transmission, etc. A sensor health status report is generated according to the self-check results. If the sensitivity of the device operation status sensor decreases, it may lead to inaccurate collected data. At this time, the data sampling frequency is dynamically adjusted according to the sensor health status report. For example, the sampling frequency is appropriately increased to obtain more data for analysis and compensation. At the same time, if it is found that the environmental interference in the workshop has changed (such as a new large device being added, resulting in increased electromagnetic interference), the noise filtering parameters are dynamically adjusted to ensure the accuracy of the data.

[0109] In a possible implementation manner, the hierarchical structure of the spatio-temporal feature matrix includes a basic feature layer, a spatio-temporal correlation layer, and an event prediction layer. The basic feature layer stores the original monitoring index statistical values of each sub-region. The spatio-temporal correlation layer stores the spatio-temporal dependence relationship coefficients between sub-regions. The event prediction layer stores the predicted values of the occurrence probabilities of abnormal events trained based on historical event data.

[0110] For example, in the scenario of a large industrial park, the basic feature layer stores the original monitoring index statistical values of each sub-region. Taking Workshop A as an example, the basic feature layer stores the original monitoring index statistical values such as the average oil temperature, average vibration frequency, highest temperature, and lowest humidity of the equipment in Workshop A. These statistical values are calculated by computing a large amount of real-time monitoring data uploaded by the acquisition unit of Workshop A. For example, 100 samples of equipment oil temperature data are collected within one hour, and statistical values such as the average value, maximum value, and minimum value of these samples are calculated and stored in the basic feature layer.

[0111] The spatio-temporal correlation layer stores the spatio-temporal dependence relationship coefficients between sub-regions. In the industrial park, Workshop A and Warehouse B are different sub-regions, and there is a certain spatio-temporal dependence relationship between them. For example, the production activities in Workshop A may affect the goods storage environment (such as temperature, humidity, etc.) in Warehouse B. By analyzing and computing a large amount of historical data and real-time data, the spatio-temporal dependence relationship coefficient between Workshop A and Warehouse B is obtained. This coefficient reflects the degree and possibility of the impact of changes in Workshop A on Warehouse B. If the production equipment in Workshop A increases production within a certain period of time, resulting in an increase in the temperature in the workshop, and Warehouse B is connected to Workshop A through a ventilation system, then a spatio-temporal dependence relationship coefficient is calculated based on factors such as the correlation of temperature changes and the flow rate of the ventilation system and stored in the spatio-temporal correlation layer.

[0112] The event prediction layer stores the predicted values of the occurrence probabilities of abnormal events trained based on historical event data. For an industrial park, according to past historical event data, such as events like fires and equipment failures that have occurred, the predicted values of the occurrence probabilities of abnormal events are trained and stored in the event prediction layer. Taking Workshop A as an example, if there have been multiple events of equipment failures caused by excessively high oil temperatures in the past, by analyzing the data related to these events (such as the trend of oil temperature changes, equipment operation time, environmental temperature, etc.) and training the model, the predicted value of the probability of equipment failure (excessively high oil temperature) in Workshop A under the current monitoring data conditions is obtained.

[0113] The method further includes: Step S310, updating the statistical values in the basic feature layer according to the real-time monitoring data, and triggering the dynamic recalculation of the dependence coefficient in the spatio-temporal association layer.

[0114] During the actual operation of Workshop A, the real-time monitoring data is continuously uploaded. For example, as production progresses, the equipment operation status sensor continuously collects the equipment oil temperature data. Every time a new set of oil temperature data is collected, the statistical value of the average equipment oil temperature in the basic feature layer is updated. When there is a large change in the equipment oil temperature data (such as a sudden increase in oil temperature), this change may affect the spatio-temporal dependence relationship between Workshop A and other sub-regions (such as Warehouse B). At this time, the dynamic recalculation of the dependence coefficient between Workshop A and Warehouse B in the spatio-temporal association layer is triggered. Because the increase in oil temperature may affect the environment of Warehouse B through the ventilation system or other means, it is necessary to re-evaluate the relationship coefficient between them.

[0115] Step S320, when the predicted value of the occurrence probability of the abnormal event in the event prediction layer exceeds the preset threshold, activating the incremental learning module of the abnormal recognition network, and updating the calculation model of the predicted value of the occurrence probability of the abnormal event based on the latest monitoring data.

[0116] Suppose in Workshop A, the preset probability threshold for equipment failure (due to excessively high oil temperature) is 30%. When the predicted value of the occurrence probability of equipment failure in the event prediction layer exceeds 30%, the incremental learning module of the abnormal recognition network is activated. This incremental learning module will retrain and update the calculation model of the predicted value of the occurrence probability of the abnormal event based on the latest collected equipment oil temperature data, equipment operation time, environmental temperature, etc. in Workshop A. For example, it may adjust the parameters in the model, consider new factors or re-evaluate the weights of existing factors, so as to more accurately predict the probability of equipment failure.

[0117] Step S330, dynamically maintaining the hierarchical structure of the spatio-temporal feature matrix through a sliding time window mechanism, retaining the feature data within a preset time length and removing the expired data blocks.

[0118] In the industrial park's monitoring system, a sliding time window is set to one day. For monitoring data from Workshop A, only feature data from the most recent day, including the basic feature layer statistics, the spatiotemporal correlation layer dependency coefficients, and the event prediction layer's predicted probability of abnormal events, is retained in the spatiotemporal feature matrix. As time passes, when a new day's data collection begins, data blocks from one day ago (including data corresponding to the basic feature layer, spatiotemporal correlation layer, and event prediction layer) are deemed expired and removed. This ensures that the data in the spatiotemporal feature matrix is always up-to-date and most relevant to the current situation, while also avoiding the waste of computing and storage resources caused by excessive data volumes.

[0119] In one possible embodiment, the anomaly recognition network includes a spatiotemporal convolution module, a long short-term memory module, and a graph neural network module. The spatiotemporal convolution module is used to extract local spatiotemporal patterns of monitoring data, the long short-term memory module is used to capture long-term dependencies in time series, and the graph neural network module is used to model spatial correlations between sub-regions.

[0120] The method further comprises: Step S410: dividing the spatiotemporal feature matrix into multiple segments according to the time dimension, and inputting each segment into the spatiotemporal convolution module for feature dimensionality reduction and pattern extraction.

[0121] In the context of an industrial park, the spatiotemporal feature matrix contains features related to monitoring data from various sub-areas (such as workshops and warehouses). Assume the spatiotemporal feature matrix covers a full day of monitoring data, segmented into hourly time segments. For Workshop A, the feature components, consisting of data such as equipment operating status (oil temperature, vibration frequency, etc.) and environmental parameters (temperature, humidity, etc.), are segmented into 24-hour segments. These segments are then fed into the spatiotemporal convolution module. The spatiotemporal convolution module uses a specific convolution kernel to perform convolution operations in both time and space to extract local spatiotemporal patterns in the monitoring data. For example, using the oil temperature data from equipment in Workshop A, the spatiotemporal convolution module can identify oil temperature fluctuation patterns during specific time periods (such as the morning peak production period). These fluctuations may be related to factors such as equipment load changes and ambient temperature fluctuations. Furthermore, the spatiotemporal convolution module can also extract local spatial patterns from oil temperature sensor data at different locations within the workshop. For example, oil temperature fluctuations at sensors near core equipment components may be more dramatic than those at edge locations. Therefore, the spatiotemporal convolution module performs feature dimensionality reduction and pattern extraction on the input data of each time segment, converting the high-dimensional raw data into more representative low-dimensional features.

[0122] Step S420: The output of the spatiotemporal convolution module is reorganized according to the spatial dimension and input into the graph neural network module to generate a spatial correlation feature vector for each sub-region.

[0123] Specifically, the output of the spatio-temporal convolution module after processing each time segment contains local spatio-temporal features, but these local spatio-temporal features need to be further integrated to reflect the spatial correlations between sub-regions. Taking Workshop A and Warehouse B as an example, the operating status of the equipment in Workshop A may be related to the environment and the status of stored items in Warehouse B through the ventilation system, material transportation channels, etc. The data related to Workshop A and Warehouse B output by the spatio-temporal convolution module is reorganized according to the spatial dimension, and the reorganized information is input into the graph neural network module. The graph neural network module constructs a graph structure based on the physical layout of the industrial park and the connection relationships between sub-regions, where the nodes represent sub-regions such as Workshop A and Warehouse B, and the edges represent the connection relationships between them (such as channels, lines, etc.). Through the calculation of the graph neural network module, a spatial correlation feature vector is generated for each sub-region. For Workshop A, its spatial correlation feature vector may contain information such as the spatial correlation strength with adjacent workshops and warehouses and the possibility of mutual influence. For example, if the ventilation system connectivity between Workshop A and Warehouse B is good, and historical data shows that the environmental changes in Workshop A have a greater impact on Warehouse B, then in the spatial correlation feature vector of Workshop A, the elements related to Warehouse B will reflect this strong correlation relationship.

[0124] Step S430: Input the spatial correlation feature vector into the long short-term memory module in chronological order to generate a global spatio-temporal state representation of the target monitoring area.

[0125] For example, the status of sub-regions in the industrial park changes over time, and the long short-term memory module can capture the long-term dependencies in this time series. The spatial correlation feature vectors of each sub-region obtained previously are input into the long short-term memory module in chronological order. Taking Workshop A as an example, within a period of time, information such as its equipment operating status and spatial correlations with other sub-regions forms a time series. The long short-term memory module can learn the dependency relationships of the status changes of Workshop A at different time points. For example, the maintenance situation of Workshop A's equipment the previous day may affect today's operating efficiency, and further affect the correlation relationships with other sub-regions. By processing the spatial correlation feature vectors of all sub-regions in the industrial park in the time series, the long short-term memory module generates a global spatio-temporal state representation of the target monitoring area (industrial park), which comprehensively combines the spatial correlations and status change information of each sub-region at different time points and can comprehensively reflect the operating status of the industrial park.

[0126] Step S440: Fuse the output features of the spatio-temporal convolution module, the graph neural network module, and the long short-term memory module to generate a comprehensive discrimination index for anomaly event recognition.

[0127] For example, the spatio-temporal convolution module provides local spatio-temporal pattern features, the graph neural network module gives spatial correlation features between sub-regions, and the long short-term memory module captures long-term dependencies in the time series. The output features of these three modules are fused. In an industrial park, for the identification of potential abnormal events, such as the judgment of equipment failures or fire risks, this fused comprehensive discrimination index is of great significance. Taking equipment failure as an example, the spatio-temporal convolution module may identify abnormal operation patterns of a certain piece of equipment in local time and space (such as abnormal fluctuations in oil temperature and limited around a certain component of the equipment), the graph neural network module reflects the spatial correlation between the equipment and surrounding sub-regions (such as a nearby warehouse may be threatened by a fire caused by the equipment failure), and the long short-term memory module takes into account the changes in the operating state of the equipment over a long period of time (such as frequent recent load changes may cause equipment fatigue). The discrimination index formed by integrating this information can more accurately judge whether there is a risk of equipment failure and the scope of influence of the possible failure and other abnormal event situations.

[0128] For example, in a possible implementation manner, the optimization objective function includes an early warning coverage efficiency term, a resource consumption penalty term, and a response timeliness reward term. The early warning coverage efficiency term is positively correlated with the early warning accuracy rate of the dynamic early warning strategy. The resource consumption penalty term is negatively correlated with the energy consumption and computing resource occupancy when the device executes the early warning instruction. The response timeliness reward term is positively correlated with the early warning response speed.

[0129] In an industrial park, the early warning accuracy rate is directly related to the timely discovery and accurate judgment of abnormal events. For example, for equipment failure early warning, if the early warning coverage efficiency is high, it means that the probability of accurately detecting equipment failures in different sub-regions (such as equipment in each workshop) is relatively large. If the early warning system can send out an early warning signal in time before a key piece of equipment in Workshop A has an excessively high oil temperature failure, then the early warning accuracy rate is relatively high. Correspondingly, the value of the early warning coverage efficiency term will also be relatively high. This requires comprehensive consideration of factors such as the accuracy of monitoring data, the performance of the abnormal event recognition network, and the rationality of the early warning strategy.

[0130] When executing the early warning instruction, the device may consume additional energy and computing resources. For example, after the early warning of excessively high oil temperature of the equipment in Workshop A is triggered, the device may need to adjust the operation mode or start an emergency cooling system, which will consume electrical energy. At the same time, devices such as servers in the monitoring system will also occupy computing resources when processing early warning-related data and instructions. If these energy consumptions and resource occupancies are too high, the value of the resource consumption penalty term will increase. To reduce this value, it is necessary to optimize the control instructions of the device and improve the utilization efficiency of computing resources. For example, adopt more energy-efficient device control algorithms and reasonably allocate the computing tasks of the server.

[0131] Quick response to abnormal events is crucial. When abnormalities occur in the equipment of Workshop A, if the early warning system can quickly send out early warning signals so that maintenance personnel, safety personnel, etc. can quickly arrive at the scene to take measures, the value of the response timeliness reward item will be relatively high. This depends on factors such as the transmission speed of the early warning signal, the response mechanism of personnel, and the operability of on-site equipment.

[0132] The method further includes: Step S510, dynamically adjusting the weight coefficients of each item in the optimization objective function according to the real-time resource status of the target monitoring area.

[0133] Step S520, when it is detected that the computing resource occupancy rate of the terminal device cluster exceeds a preset threshold, increasing the weight of the resource consumption penalty item and decreasing the weight of the response timeliness reward item.

[0134] Step S530, when the propagation speed of the potential abnormal event exceeds a preset safety threshold, increasing the weights of the early warning coverage efficiency item and the response timeliness reward item.

[0135] Step S540, generating a weight adjustment strategy for the optimization objective function based on the reinforcement learning algorithm, so that the dynamic early warning strategy achieves a comprehensive optimization effect under resource constraint conditions.

[0136] Dynamically adjust the weight coefficients of each item in the optimization objective function according to the real-time resource status of the target monitoring area. The resource status of the industrial park is constantly changing. For example, during the production peak period, the energy consumption of equipment is relatively large and the computing resources are relatively tight. At this time, it is detected that the computing resource occupancy rate of the terminal device cluster exceeds a preset threshold (such as 80%). In order to avoid excessive resource consumption, it is necessary to increase the weight of the resource consumption penalty item and decrease the weight of the response timeliness reward item. This means that in this case, more attention is paid to reducing resource consumption, and some response speed may be sacrificed appropriately. When the propagation speed of the potential abnormal event exceeds a preset safety threshold, such as the case where a fire spreads rapidly in the workshop, it is necessary to increase the weights of the early warning coverage efficiency item and the response timeliness reward item. This is because in such an emergency, quickly and accurately warning and taking measures to prevent the spread of the fire is the top priority, even if more resources may be consumed. Generate a weight adjustment strategy for the optimization objective function based on the reinforcement learning algorithm, so that the dynamic early warning strategy achieves a comprehensive optimization effect under resource constraint conditions. The reinforcement learning algorithm continuously interacts with the actual operation environment of the industrial park to learn the optimal weight adjustment strategy in different situations. For example, by simulating the early warning process in different equipment failure scenarios and different resource states multiple times, gradually determine how to adjust the weight coefficients in various situations to achieve a comprehensive balance of early warning accuracy, response speed, and resource consumption while meeting resource limitations.

[0137] For example, in a possible implementation, the method further includes: Step S610: Deploy edge computing nodes in the terminal device cluster, where the edge computing nodes store the simplified execution logic of the dynamic early warning policy and local emergency response rules.

[0138] In an industrial park, edge computing nodes are distributed near each sub-region, such as near Workshop A, Warehouse B, etc. The simplified execution logic of the dynamic early warning policy stored in these edge computing nodes is simplified based on the complete policy issued by the central server for rapid local execution. The local emergency response rules are special rules formulated for local possible abnormal events. For example, the local emergency response rule of the edge computing node in Workshop A may include immediately starting the local ventilation system and alarm device in the workshop when the smoke concentration in the workshop reaches a certain threshold, without waiting for instructions from the central server.

[0139] Step S620: When the communication link between the central server and the terminal device cluster is interrupted, activate the autonomous early warning mode of the edge computing nodes, and generate an emergency early warning instruction based on the most recently received complete dynamic early warning policy and local monitoring data.

[0140] Suppose due to a network failure, the communication link between the central server and the terminal device clusters in sub-regions such as Workshop A and Warehouse B is interrupted. The edge computing node in Workshop A generates an emergency early warning instruction based on the most recently received complete dynamic early warning policy (including early warning trigger conditions and response instructions for different abnormal events) and local monitoring data (such as oil temperature, temperature, smoke concentration, etc. data of the equipment in the workshop). If the local monitoring data shows that the oil temperature of a certain device in the workshop is too high and close to the early warning threshold, the edge computing node issues an instruction to reduce the device load according to the simplified execution logic and local emergency response rules, and notifies the staff in the workshop to conduct an inspection.

[0141] Step S630: After the communication link is restored, synchronize the execution logs and data difference records generated by the edge computing nodes in the autonomous early warning mode to the central server.

[0142] When the communication link returns to normal, the edge computing node in Workshop A synchronizes the execution logs (including issued instructions, execution times, response situations of the equipment, etc.) in the autonomous early warning mode and the difference records between the local monitoring data and the expected data of the central server (such as the difference between the actual change in oil temperature and the predicted value of the central server) to the central server.

[0143] Step S640: Perform consistency verification and policy version update on the local execution logic of the dynamic early warning policy according to the execution logs and data difference records.

[0144] Specifically, after receiving the execution logs and data difference records of the edge computing nodes in Workshop A, the central server performs consistency verification on the local execution logic of the dynamic early warning strategy for Workshop A. For example, it checks whether the instructions sent by the edge computing nodes conform to the overall early warning strategy principles and whether there are any misoperations or unreasonable instructions. At the same time, it updates the version of the dynamic early warning strategy according to the data difference records. If a large difference is found between the local monitoring data and the predicted data of the central server, some parameters in the early warning strategy (such as early warning thresholds, response instructions, etc.) may need to be adjusted to improve the accuracy and adaptability of the early warning strategy and ensure better response in case of similar situations in the future.

[0145] Figure 2 FIG. shows a schematic diagram of exemplary hardware and software components of a server 100 that can implement the concepts of the present application provided by some embodiments of the present application. For example, a processor 120 can be used on the server 100 and is used to execute the functions in the present application.

[0146] The server 100 can be a general-purpose server or a special-purpose server, both of which can be used to implement the artificial intelligence-based monitoring and early warning analysis method of the present application. Although only one server is shown in the present application, for convenience, the functions described in the present application can be implemented in a distributed manner on multiple similar platforms to balance the processing load.

[0147] For example, the server 100 can include a network port 110 connected to a network, one or more processors 120 for executing program instructions, a communication bus 130, and different forms of storage media 140, such as disks, ROM, or RAM, or any combination thereof. Exemplarily, the server 100 can also include program instructions stored in ROM, RAM, or other types of non-transitory storage media, or any combination thereof. The method of the present application can be implemented according to these program instructions. The server 100 also includes an input / output (I / O) interface 150 between the computer and other input / output devices.

[0148] For ease of illustration, only one processor is described in the server 100. However, it should be noted that the server 100 in the present application can also include multiple processors. Therefore, the steps executed by one processor described in the present application can also be jointly executed or separately executed by multiple processors. For example, if the processor of the server 100 executes steps A and B, it should be understood that steps A and B can also be jointly executed by two different processors or separately executed in one processor. For example, the first processor executes step A, the second processor executes step B, or the first processor and the second processor jointly execute steps A and B.

[0149] In addition, an embodiment of the present invention further provides a readable storage medium, in which computer-executable instructions are preset. When the processor executes the computer-executable instructions, the above-mentioned monitoring and early warning analysis method based on artificial intelligence is implemented.

[0150] It should be noted that, in order to simplify the description of the present invention disclosure and thus help the understanding of one or more embodiments of the invention, in the foregoing description of the embodiments of the present invention, sometimes multiple features are merged into one embodiment, drawing or description thereof.

Claims

1. An artificial intelligence-based monitoring and early warning analysis method, characterized in that The method includes: Collecting multi-source monitoring data streams within a target monitoring area, where the multi-source monitoring data streams include at least two types of real-time monitoring data and corresponding spatio-temporal tag information, and the spatio-temporal tag information is used to describe the collection location and timestamp of the real-time monitoring data within the target monitoring area; Performing spatio-temporal feature extraction on the multi-source monitoring data streams to generate a spatio-temporal feature matrix with hierarchical association relationships, where each element in the spatio-temporal feature matrix corresponds to the monitoring index feature of a target location within the target monitoring area within a target time window; Based on a preset anomaly recognition network, performing dynamic pattern analysis on the spatio-temporal feature matrix to determine potential anomaly events existing within the target monitoring area and their corresponding anomaly propagation paths, where the anomaly propagation paths are used to indicate the diffusion direction and influence range of the potential anomaly events within the target monitoring area; According to the topological structure of the anomaly propagation paths and the attribute parameters of the potential anomaly events, generating a dynamic early warning strategy adapted to the target monitoring area, where the dynamic early warning strategy includes differential early warning trigger conditions and response instruction sets for different locations and different time stages; Based on historical early warning feedback data, performing real-time optimization and adjustment on the dynamic early warning strategy, and outputting the optimized dynamic early warning strategy to the terminal device cluster corresponding to the target monitoring area.

2. The monitoring and early warning analysis method based on artificial intelligence according to claim 1, characterized in that The performing spatio-temporal feature extraction on the multi-source monitoring data streams to generate a spatio-temporal feature matrix with hierarchical association relationships includes: Dividing the multi-source monitoring data streams into multiple spatio-temporal data blocks according to the spatio-temporal tag information, where each spatio-temporal data block corresponds to a monitoring data set of a preset sub-area within the target monitoring area within a continuous time series; For each spatio-temporal data block, extracting cross-modal association features between the monitoring data included in the spatio-temporal data block, where the cross-modal association features are used to describe the coupling relationship between different types of monitoring data in the same spatio-temporal dimension; Fusing the cross-modal association features with the time series fluctuation features of the spatio-temporal data block to generate a local spatio-temporal feature vector corresponding to the spatio-temporal data block; Based on the physical connection relationships between the preset sub-areas and historical event propagation records, constructing a global spatio-temporal association graph of the target monitoring area, where the nodes in the global spatio-temporal association graph represent the preset sub-areas, and the edges represent the event propagation probability and association strength between the preset sub-areas; According to the global spatio-temporal association graph, performing spatial weight allocation and time series alignment on the local spatio-temporal feature vectors to generate the hierarchical structure of the spatio-temporal feature matrix.

3. The monitoring and early warning analysis method based on artificial intelligence according to claim 2, wherein, The based on a preset anomaly recognition network, performing dynamic pattern analysis on the spatio-temporal feature matrix to determine potential anomaly events existing within the target monitoring area and their corresponding anomaly propagation paths includes: Inputting the spatio-temporal feature matrix into the multi-scale feature extraction module of the anomaly recognition network to obtain the anomaly pattern features of the target monitoring area at different time granularities and spatial resolutions; Dynamically allocate weights to the abnormal pattern features through the attention allocation module of the abnormal recognition network to determine the abnormal confidence of each sub-region within the target monitoring area and the correlation strength between abnormal events; Construct an abnormal event propagation network based on the abnormal confidence and correlation strength. The nodes in the abnormal event propagation network represent sub-regions where the detected abnormal confidence exceeds the set confidence level, and the edges represent the propagation possibility of abnormal events between sub-regions; Based on the node degree distribution and edge weight distribution of the abnormal event propagation network, identify the core occurrence location of the potential abnormal event and the key nodes of the propagation path; Combine the time continuity of the spatio-temporal tag information to optimize the path connection of the key nodes of the propagation path, and generate the complete topological structure of the abnormal propagation path.

4. The monitoring and early warning analysis method based on artificial intelligence according to claim 3, wherein The combining the time continuity of the spatio-temporal tag information to optimize the path connection of the key nodes of the propagation path and generate the complete topological structure of the abnormal propagation path includes: Extract the timestamp sequence in the spatio-temporal tag information corresponding to the key nodes, and arrange the key nodes in a time-ordered node queue according to the chronological order of the timestamp sequence. Each node in the time-ordered node queue carries spatial coordinate information and the corresponding timestamp; Calculate the timestamp interval between adjacent nodes in the time-ordered node queue, divide adjacent nodes with a timestamp interval less than the preset time continuity threshold into the same time continuous group, and record the start timestamp and end timestamp of each time continuous group; Traverse the spatial coordinate information of the nodes within each time continuous group, calculate the spatial distance matrix between the nodes within the group, mark the node pairs with a spatial distance less than the preset spatial proximity threshold as spatially adjacent node pairs, and generate a spatial proximity relationship graph based on the spatial distance matrix; Based on the spatial proximity relationship graph, linearly connect the nodes with spatial proximity relationships within the time continuous group in timestamp order to form an initial propagation path segment, and the initial propagation path segment includes path direction information and path connection strength parameters; Detect the spatio-temporal conflicts between the initial propagation path segments generated by different time continuous groups. The spatio-temporal conflicts include timestamp overlap conflicts of path segments in the same spatial region and topological contradictions where the path direction does not match the spatial proximity relationship graph; According to the detected spatio-temporal conflict type, dynamically adjust the conflict path segments to generate adjusted initial propagation path segments, specifically including: for timestamp overlap conflicts, preferentially retain the path segment with a later timestamp order and remove the overlapping part; for topological contradictions, recalculate the connection direction of the path segment based on the spatial proximity relationship graph; Globally sort the adjusted initial propagation path segments according to the start timestamp of the time continuous group, splice the path segments according to the sorting result, and verify whether the timestamp interval and spatial jump distance between adjacent path segments meet the time continuity threshold and spatial proximity threshold during the splicing process; When the timestamp interval or spatial jump distance of adjacent path segments is detected to exceed the threshold, an intermediate node is inserted and the spatial proximity relationship map is updated. The spatial coordinates of the intermediate node are generated by interpolating the spatial coordinates of the end nodes of adjacent path segments, and the timestamp is linearly assigned according to the timestamp interval of adjacent path segments; Based on the finally stitched path segment set and the inserted intermediate nodes, the complete topological structure of the abnormal propagation path is generated. The complete topological structure includes path branch information, node connection order, and path propagation direction weight parameters.

5. The monitoring and early warning analysis method based on artificial intelligence according to claim 3, wherein Generating a dynamic early warning strategy adapted to the target monitoring area according to the topological structure of the abnormal propagation path and the attribute parameters of the potential abnormal event, including: Calculating the priority index of each sub-region in the potential abnormal event propagation chain according to the topological structure of the abnormal propagation path, where the priority index is used to quantify the urgency of the sub-region affected by the abnormal event; Determining the trigger threshold range of the dynamic early warning strategy based on the attribute parameters of the potential abnormal event, where the attribute parameters include abnormal intensity, propagation speed, and influence duration; Generating multi-level early warning trigger rules according to the priority index and the trigger threshold range, where the multi-level early warning trigger rules define the early warning levels and response delay times of different priority sub-regions under different abnormal intensities; Constructing a set of response instructions matching the multi-level early warning trigger rules, where the set of response instructions includes device control instructions, personnel scheduling instructions, and data reporting protocols for each early warning level; Dynamically binding the multi-level early warning trigger rules and the set of response instructions in the time sequence of the abnormal propagation path to form the execution logic chain of the dynamic early warning strategy.

6. The monitoring and early warning analysis method based on artificial intelligence according to claim 5, characterized in that The real-time optimization and adjustment of the dynamic early warning strategy based on historical early warning feedback data includes: Obtaining the feedback log data generated when the terminal device cluster executes the historical early warning strategy, where the feedback log data includes early warning response delay time, device execution status, and actual influence range of abnormal events; Extracting the strategy execution effect indicators from the feedback log data, where the strategy execution effect indicators include early warning accuracy rate, response efficiency deviation, and resource consumption rate; Comparing the strategy execution effect indicators with a preset optimization objective function to determine the trigger condition parameters and response instruction priorities to be adjusted in the dynamic early warning strategy; Iteratively correcting the trigger threshold range and response delay time of the dynamic early warning strategy according to the gradient direction of the optimization objective function to obtain the corrected parameters; Based on the corrected parameters, recalculating the coverage range and resource allocation weights of each level of early warning in the multi-level early warning trigger rules to generate an optimized dynamic early warning strategy.

7. The monitoring and early warning analysis method based on artificial intelligence according to claim 1, characterized in that Before collecting the multi-source monitoring data stream in the target monitoring area, the method further includes: Dividing a plurality of monitoring data collection units according to the geographical distribution characteristics of the target monitoring area, and each collection unit is configured with at least three heterogeneous sensors and a data preprocessing module; Set the data sampling frequency and transmission protocol of the heterogeneous sensors so that sensor data of different types within the same acquisition unit are transmitted through independent channels after timestamp alignment; Configure the noise filtering algorithm and data integrity verification rules for the data preprocessing module. The noise filtering algorithm uses the wavelet transform method with an adaptive threshold to eliminate the environmental interference components in the sensor signals; Establish a two-way communication link between the acquisition unit and the central server. The two-way communication link supports the real-time upload of the multi-source monitoring data stream and the synchronous issuance of control instructions; Periodically perform self-check and calibration on the acquisition unit, generate a sensor health status report, and dynamically adjust the data sampling frequency and noise filtering parameters.

8. The monitoring and early warning analysis method based on artificial intelligence according to claim 1, characterized in that The hierarchical structure of the spatio-temporal feature matrix includes a basic feature layer, a spatio-temporal correlation layer, and an event prediction layer. The basic feature layer stores the statistical values of the original monitoring indicators for each sub-region. The spatio-temporal correlation layer stores the spatio-temporal dependence relationship coefficients between sub-regions. The event prediction layer stores the predicted values of the occurrence probabilities of abnormal events trained based on historical event data; The method further includes: Update the statistical values in the basic feature layer according to the real-time monitoring data, and trigger the dynamic recalculation of the dependence relationship coefficients in the spatio-temporal correlation layer; When the predicted value of the occurrence probability of the abnormal event in the event prediction layer exceeds the preset threshold, activate the incremental learning module of the abnormal recognition network, and update the calculation model of the predicted value of the occurrence probability of the abnormal event based on the latest monitoring data; Dynamically maintain the hierarchical structure of the spatio-temporal feature matrix through a sliding time window mechanism, retain the feature data within a preset time length, and remove the expired data blocks.

9. The monitoring and early warning analysis method based on artificial intelligence according to claim 3, wherein The abnormal recognition network includes a spatio-temporal convolution module, a long short-term memory module, and a graph neural network module. The spatio-temporal convolution module is used to extract the local spatio-temporal patterns of the monitoring data. The long short-term memory module is used to capture the long-term dependence relationships in the time series. The graph neural network module is used to model the spatial associations between sub-regions; The method further includes: Slice the spatio-temporal feature matrix along the time dimension into multiple segments, and input them into the spatio-temporal convolution module respectively for feature dimensionality reduction and pattern extraction; Recombine the output of the spatio-temporal convolution module along the spatial dimension and input it into the graph neural network module to generate the spatial association feature vectors for each sub-region; Input the spatial association feature vectors into the long short-term memory module in chronological order to generate the global spatio-temporal state representation of the target monitoring area; Fuse the output features of the spatio-temporal convolution module, the graph neural network module, and the long short-term memory module to generate a comprehensive discrimination index for abnormal event recognition.

10. A server, characterized in that, The server includes a processor and a memory. The memory is connected to the processor. The memory is used to store programs, instructions, or codes. The processor is used to execute the programs, instructions, or codes in the memory to implement the artificial intelligence-based monitoring and early warning analysis method according to any one of claims 1-9 above.

Images