Smoke exhaust valve detection data edge calculation processing method and system

Abstract

The invention relates to the technical field of edge calculation, in particular to a smoke exhaust valve detection data edge calculation processing method and system. Comprising the following steps: acquiring operation state parameters in edge nodes; comparing the operation state parameter with a corresponding preset normal range, a change amplitude threshold and a duration threshold, and identifying an abnormal operation state of the edge node when the operation state parameter exceeds the preset normal range, or the change amplitude of the operation state parameter exceeds the change amplitude threshold and the duration reaches the duration threshold; generating summary information according to the abnormal operation state; and sending the summary information to a central control room. The problems that in a distributed smoke exhaust system, dependency of edge nodes on sensor data quality is high, slow drifting or local degradation of sensor performance is difficult to find, local decision logic is affected, a central control room is difficult to effectively monitor internal health conditions of the edge nodes, and the edge nodes have non-fatal and intermittent faults are solved.

Description

Technical Field

[0001] This application relates to the field of edge computing technology, and more specifically, to a method and system for edge computing processing of smoke exhaust valve detection data. Background Technology

[0002] Traditional centralized smoke extraction systems in large commercial complexes are prone to network congestion and server bottlenecks during fires due to centralized data uploads, leading to slow responses or even system failures. While edge computing-based distributed smoke extraction systems improve local response speed and reliability, their judgment is highly dependent on the quality of sensor data. In practice, smoke and temperature sensors may experience decreased sensitivity, increased false alarms, and data drift due to changes in fumes, steam, environmental conditions, or long-term operation, thus affecting the local decisions of edge nodes and causing smoke extraction delays, malfunctions, or threshold misalignments. Meanwhile, the central control room typically only receives summary information processed by edge nodes, making it difficult to grasp the status of internal sensors, decision-making biases, and intermittent communication or hardware failures, creating monitoring blind spots. The long-term accumulation of these problems can render the system appear normal on the surface but harbor significant hidden dangers, potentially leading to smoke extraction system failure in a real fire.

[0003] To address the aforementioned issues, existing technologies urgently need improvement. Summary of the Invention

[0004] This application discloses an edge computing processing method and system for smoke exhaust valve detection data, which aims to solve the problems of edge nodes in distributed smoke exhaust systems being highly dependent on sensor data quality, slow drift or local degradation of sensor performance being difficult to detect, local decision-making logic being affected, difficulty for the central control room to effectively monitor the internal health status of edge nodes, and the existence of non-fatal or intermittent faults in the edge nodes themselves.

[0005] The technical solution of this application is as follows: In a first aspect, this application discloses an edge computing processing method for smoke exhaust valve detection data, applied to edge nodes in a distributed smoke exhaust system. The method includes: Obtain the operating status parameters inside the edge node, including processor workload parameters, memory usage parameters, communication status parameters between the edge node and local sensors, power stability parameters, and / or internal ambient temperature parameters. The operating status parameters are compared with the corresponding preset normal range, change amplitude threshold and duration threshold respectively. When the operating status parameters exceed the preset normal range, or the change amplitude of the operating status parameters exceeds the change amplitude threshold and the duration reaches the duration threshold, the abnormal operating status of the edge node is identified. The abnormal operating status is used to characterize the edge node in a sub-healthy state of slow performance change, local degradation or intermittent failure. The sub-healthy state is the state in which the edge node has not completely failed but the operating performance deviates abnormally. Based on the abnormal operating status, generate summary information. The summary information includes at least the identification information of the edge nodes and the description of the abnormal type corresponding to the abnormal operating status. Summary information is sent to the central control room in a manner that is lower in priority than the transmission of smoke exhaust control commands, so that the central control room can obtain abnormal operating status information of the edge nodes without receiving the original operating data of the edge nodes.

[0006] Secondly, this application also discloses an edge computing processing system for smoke exhaust valve detection data, applied to edge nodes in a distributed smoke exhaust system. The system includes: The acquisition module is used to acquire the operating status parameters inside the edge node. The operating status parameters include processor workload parameters, memory usage parameters, communication status parameters between the edge node and local sensors, power stability parameters, and / or internal ambient temperature parameters. The analysis module is used to compare the operating status parameters with the corresponding preset normal range, change amplitude threshold and duration threshold respectively. When the operating status parameters exceed the preset normal range, or the change amplitude of the operating status parameters exceeds the change amplitude threshold and the duration reaches the duration threshold, the abnormal operating status of the edge node is identified. The abnormal operating status is used to characterize the edge node in a sub-healthy state of slow performance change, local degradation or intermittent failure. The sub-healthy state is the state in which the edge node has not completely failed but the operating performance deviates abnormally. The generation module is used to generate summary information based on the abnormal operating state. The summary information includes at least the identification information of the edge nodes and the description of the abnormal type corresponding to the abnormal operating state. The sending module is used to send summary information to the central control room in a manner that is lower than the transmission priority of smoke exhaust control commands, so that the central control room can obtain abnormal operating status information of the edge nodes without receiving the original operating data of the edge nodes.

[0007] Beneficial Effects: The edge computing processing method for smoke exhaust valve detection data disclosed in this application can effectively identify sub-healthy states where edge nodes are experiencing slow performance changes, local degradation, or intermittent failures. This identification of sub-healthy states overcomes the limitations of traditional binary "working / faulting" judgments, enabling the capture of slow drift in sensor performance or hidden lesions within edge nodes. This solves the problem in existing technologies of difficulty in detecting sensor performance degradation and non-fatal failures of edge nodes. Therefore, this application significantly improves the fire response speed and system reliability of distributed smoke exhaust systems, effectively reducing potential safety hazards. Attached Figure Description

[0008] Figure 1 This is a flowchart illustrating a method for edge computing processing of smoke exhaust valve detection data provided in this application.

[0009] Figure 2 A flowchart of an edge computing processing system for smoke exhaust valve detection data provided in this application.

[0010] In the diagram: 1. Acquisition module; 2. Analysis module; 3. Generation module; 4. Sending module. Detailed Implementation

[0011] The technical solutions of this application will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of this application, and not all embodiments. The components of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0012] It should be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. Furthermore, in the description of this application, terms such as "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0013] An edge computing processing method for smoke exhaust valve detection data is proposed and applied to edge nodes in a distributed smoke exhaust system. Traditional distributed smoke exhaust systems struggle to address slow sensor performance drift, local degradation, or intermittent faults in edge nodes using simple "working / failed" binary states. This results in data received by edge nodes no longer accurately reflecting the actual situation, thus impacting the accuracy of local decision-making. Furthermore, the central control room struggles to effectively monitor the internal health of edge nodes at a macroscopic level, failing to promptly identify and correct potential risks.

[0014] In this regard, refer to Figure 1 This application proposes a method for edge computing processing of smoke exhaust valve detection data, including: S1000: Acquire the operating status parameters inside the edge node, including processor workload parameters, memory usage parameters, communication status parameters between the edge node and local sensors, power stability parameters, and / or internal ambient temperature parameters. S2000: The operating status parameters are compared with the corresponding preset normal range, change amplitude threshold and duration threshold respectively. When the operating status parameters exceed the preset normal range, or the change amplitude of the operating status parameters exceeds the change amplitude threshold and the duration reaches the duration threshold, the abnormal operating status of the edge node is identified. The abnormal operating status is used to characterize the edge node in a sub-healthy state of slow performance change, local degradation or intermittent failure. The sub-healthy state is the state in which the edge node has not completely failed but the operating performance deviates abnormally. S3000: Generate summary information based on the abnormal operating status. The summary information includes at least the identification information of the edge nodes and the description of the abnormal type corresponding to the abnormal operating status. S4000: Sends summary information to the central control room in a manner with a lower priority than the transmission of smoke exhaust control commands, so that the central control room can obtain abnormal operating status information of the edge nodes without receiving the original operating data of the edge nodes.

[0015] In this application, "edge node" refers to a computing unit deployed in each fire protection zone of a distributed smoke extraction system. It possesses local data processing, analysis, and decision-making capabilities, enabling it to collect, process, and respond to operational data within its area without real-time intervention from the central control room. "Operating status parameters" refer to various indicators characterizing the internal working status of the edge node, such as processor workload, memory usage, communication status with local sensors, power stability, and internal ambient temperature. "Sub-health state" refers to a state where the edge node has not yet completely failed, but its operational performance has deviated abnormally, potentially manifesting as slow performance changes, localized degradation, or intermittent failures. "Summary information" refers to information that highly summarizes and generalizes the abnormal operating status of the edge node, including at least the edge node's identification information and a description of the anomaly type, aiming to report key anomalies to the central control room in a concise manner. This method is primarily applied to edge nodes in distributed smoke extraction systems. By performing intelligent analysis at the edge, it reduces the burden on the central control room, improves system response efficiency, and achieves effective monitoring of the edge node's health status without transmitting large amounts of raw operational data.

[0016] Specifically, the edge computing processing method for smoke exhaust valve detection data in this application includes the following main features: First, this method acquires the internal operational status parameters of the edge node. These parameters form the basis for assessing the health of the edge node and serve as direct input for identifying abnormal operating states. For example, processor workload parameters reflect the computational pressure on the edge node, while memory usage parameters indicate its resource consumption. Communication status parameters between the edge node and local sensors reveal the reliability of data transmission, power stability parameters concern the continuity of hardware operation, and internal ambient temperature parameters can be used to monitor potential problems such as overheating. As an implementation approach, the edge node can incorporate multiple monitoring modules. For instance, a processor monitoring module can collect processor workload data in real time, a memory monitoring module can record memory occupancy, a communication monitoring module can track packet loss rate and latency with local sensors, a power monitoring module can detect voltage and current fluctuations, and a temperature sensor can measure the internal ambient temperature. These modules can periodically store the collected data in the edge node's local memory for subsequent processing. To ensure the accuracy of subsequent comparisons and analyses, these operational status parameters can be recorded according to a uniform sampling period or a uniform timestamp, making different types of parameters comparable over time.

[0017] Secondly, this method compares the operating status parameters with corresponding preset normal ranges, variation thresholds, and duration thresholds to identify abnormal operating states of edge nodes. When an operating status parameter exceeds the preset normal range, or when the variation amplitude of the operating status parameter exceeds the variation threshold and the duration reaches the duration threshold, the edge node is considered to be in an abnormal operating state. This abnormal operating state does not require the edge node to be completely failed, but rather characterizes the edge node as being in a sub-healthy state characterized by slow performance changes, localized degradation, or intermittent failures. For example, a normal range (e.g., 10%~80%), a variation threshold (e.g., fluctuations exceeding 20% ​​within 5 minutes), and a duration threshold (e.g., a duration of 10 minutes) can be set for the processor workload. If the processor workload remains above 80% for 10 minutes, or suddenly spikes from 20% to 70% within 5 minutes and continues for 10 minutes, it can be determined as abnormal. As one implementation, the edge node can be configured with a parameter monitoring and comparison unit. This unit can pre-store the normal range, variation threshold, and duration threshold for each operating status parameter. Upon receiving real-time operational status parameters, the unit compares each parameter in parallel. For example, a state machine model can be used to track parameter changes. When a parameter first exceeds the normal range or its change exceeds a threshold, a timer is started and continuously monitored to see if it reaches a duration threshold. Once any anomaly detection condition is met, an abnormal operational state is identified. When multiple operational status parameters simultaneously meet the anomaly detection conditions, the identification results of multiple abnormal parameters can be combined as a comprehensive abnormal operational state of the edge node.

[0018] Furthermore, this method generates summary information based on the identified abnormal operating states. The summary information includes at least the edge node's identification information and an anomaly type description corresponding to the abnormal operating state. This summary information generation aims to reduce data volume, avoid transmitting large amounts of raw operating data to the central control room, and enable the central control room to quickly understand the main content and source of the anomaly. For example, if the edge node identifies a persistently high processor workload, the anomaly type description could be "processor overload." If the communication module experiences intermittent packet loss, the anomaly type description could be "unstable communication link." As one implementation, the edge node can include a summary information generation unit. After receiving the identification result of the abnormal operating state, this unit converts the specific deviation of abnormal parameters into a standardized anomaly type description according to a preset anomaly type mapping table. Simultaneously, this unit extracts the edge node's unique identification information (such as device ID, deployment location, etc.) from its configuration information and encapsulates this information with the anomaly type description to form structured summary information. Without changing the basic structure of the summary information, which includes at least the identification information of edge nodes and the description of the anomaly type, the summary information can also include supplementary fields such as the time of anomaly occurrence, anomaly level (i.e., anomaly severity information) or corresponding parameter identifiers, as needed, so that the central control room can conduct more detailed anomaly analysis.

[0019] Finally, this method sends summary information to the central control room with a lower priority than that given to smoke exhaust control commands. This priority setting ensures that smoke exhaust control commands are transmitted first when network resources are limited, while summary information of abnormal states is transmitted without affecting core business operations. The central control room can obtain information about abnormal operating statuses of edge nodes without receiving raw operational data from the edge nodes. For example, in a distributed smoke exhaust system, smoke exhaust control commands (such as "start smoke exhaust fan" and "close fire damper") have the highest transmission priority to ensure a rapid response in the event of a fire. While summary information of the health status of edge nodes is important, its timeliness requirements are relatively low, so it can be transmitted with a lower priority. As one implementation method, edge nodes can be configured with a communication management module. When sending summary information, this module will tag the data packets of summary information with lower priority according to a preset priority strategy. When the network is congested, network devices will prioritize smoke exhaust control commands with high-priority tags, while summary information will be transmitted when the network is idle or resources allow. The central control room is equipped with a summary information receiving and parsing module. This module is specifically designed to receive and parse summary information from each edge node and display it to operators, enabling them to promptly understand the health status of the edge nodes. As a result, the central control room can summarize and monitor abnormal operating states of each edge node without continuously receiving and processing large amounts of raw operational data.

[0020] The edge computing processing method for smoke exhaust valve detection data in this application, through real-time monitoring and intelligent analysis of operating status parameters at edge nodes, can promptly detect and identify sub-health states such as slow performance changes, local degradation, or intermittent failures that may exist at edge nodes. Compared to the traditional binary "working / faulting" judgment, this application can more precisely capture abnormal deviations of edge nodes, thereby achieving earlier warnings and interventions. Furthermore, by generating concise summary information and transmitting it to the central control room with lower priority, this application effectively avoids network congestion and processing burdens caused by transmitting large amounts of raw operating data to the central control room. This allows the central control room to comprehensively grasp the health status of each edge node without increasing additional resource consumption. This approach not only improves the overall reliability and safety of the distributed smoke exhaust system but also provides more timely and accurate information support for system maintenance and fault diagnosis.

[0021] In another embodiment of this application, the step of identifying abnormal operating states of edge nodes is further proposed to include: S2100: Acquire real-time data from multiple sensors in the local sensor and perform time correlation on the real-time data; S2200: When the edge node is in a normal state, learn and store the correlation pattern of time correlation between real-time data of multiple sensors, and determine the dynamic baseline of each sensor in the current environment based on the correlation pattern. The correlation pattern is a pattern used to characterize the correlation between real-time data of multiple sensors, and the dynamic baseline is a dynamic reference value or dynamic reference range used to characterize the normal reading state of each sensor in the current environment. S2300: When the reading of any of the multiple sensors deviates from the corresponding dynamic baseline, acquire the current correlation between the real-time data of the multiple sensors; S2400: Match the current association with the association pattern to obtain the matching result; S2500: Based on the matching results, determine whether the abnormal reading of the sensor whose reading deviates from the corresponding dynamic baseline is caused by environmental factors or by the performance degradation of the sensor, and obtain the abnormal operating status of the edge node.

[0022] Specifically, local sensors can be understood as various physical sensors deployed near or directly connected to edge nodes, such as temperature sensors, humidity sensors, smoke sensors, air pressure sensors, and wind speed sensors. These sensors are used to monitor key physical quantities of the operating environment or equipment of the smoke extraction system in real time. Real-time data refers to the instantaneous measurements collected by these sensors at continuous points in time. Time correlation of real-time data refers to the correspondence and analysis of readings from different sensors at the same point in time or within similar time periods according to a unified time reference, in order to reveal possible synchronicities, lags, or causal relationships between them.

[0023] The period when edge nodes are in a normal state refers to the period after the system has undergone initialization, calibration, or long-term stable operation and is confirmed to be healthy and performing normally. During this period, real-time data from multiple sensors is continuously collected and used as the basis for learning and updating correlation patterns. Correlation patterns can be understood as models describing the interdependencies between real-time data from multiple sensors; these can be statistical correlation models (such as covariance matrices and correlation coefficients), machine learning models (such as neural networks and support vector machines), or rule-based models. Based on the learned correlation patterns, the system determines a dynamic baseline for each sensor in the current environment. The dynamic baseline is not a fixed threshold but a reference value or range that can adaptively adjust according to environmental changes and system operating status, more accurately reflecting the expected readings of the sensor under normal operating conditions. In other words, correlation patterns characterize the normal correlation relationships between multiple sensors, while the dynamic baseline characterizes the expected reading status of a single sensor under these normal correlation relationships; these correspond to multi-sensor relationship judgment and single-sensor deviation judgment, respectively.

[0024] In practical applications, when the reading of any of the multiple sensors deviates from its corresponding dynamic baseline, the system immediately acquires real-time data from multiple sensors within the current moment or a recent time window and calculates the current correlation between them. The current correlation here refers to the instantaneous or short-term relationship between the real-time data of multiple sensors when the sensor readings deviate. Subsequently, the acquired current correlation is matched with pre-learned and stored correlation patterns. The matching process aims to evaluate the similarity or difference between the relationship between the current sensor readings and the correlation patterns under normal operating conditions. The matching result can be a similarity score, a distance metric, or a classification judgment. Therefore, the current correlation is the input for current anomaly detection, the correlation pattern is the reference for comparison, and the matching result is the direct basis for subsequent anomaly cause determination.

[0025] Furthermore, based on the matching results, the system can determine the root cause of sensor reading deviations. If the current correlation matches the correlation pattern highly, indicating that the relationship between multiple sensor readings is consistent with the normal state, the deviation is likely caused by a holistic change in environmental factors, such as a general increase in ambient temperature causing all related sensor readings to rise synchronously. Conversely, if the current correlation matches the correlation pattern low, indicating a significant change in the relationship between sensor readings, the deviation is more likely caused by the performance degradation of a single sensor or local component, such as a malfunction in a temperature sensor causing its reading to be abnormal, while the readings of other related sensors remain normally correlated. Thus, the system can not only identify whether edge nodes are in an abnormal operating state, but also further distinguish whether the abnormal operating state is more likely to be caused by environmental factors or by sensor performance degradation, thereby achieving a more accurate classification of abnormal operating states.

[0026] The solution proposed in this application overcomes the limitations of traditional methods that rely solely on a single parameter threshold for judgment by introducing time correlation analysis of real-time data from local sensors.

[0027] In another embodiment of this application, a method for learning and storing temporal correlation patterns among real-time data from multiple sensors is further proposed, including: S2210: Acquire the first real-time data from multiple sensors within a preset time window; S2220: Calculate the statistical correlation index of the first real-time data within a preset time window. The statistical correlation index includes the covariance matrix. S2230: Perform a first comparison between the covariance matrix and the historically stored association patterns. The first comparison includes calculating the degree of difference between the covariance matrix and the historically stored association patterns. S2240: When the difference is less than the preset difference threshold, update the association pattern of the historical storage. The update includes incorporating the covariance matrix into the association pattern of the historical storage in a moving average manner.

[0028] Specifically, when the edge nodes are in a normal state, in order to establish and maintain an accurate correlation pattern reflecting the interactions between sensors, it is first necessary to acquire the first real-time data from multiple sensors within a preset time window. This preset time window can be set according to system characteristics and data sampling frequency, for example, it can be a period of several minutes, several hours, or longer, to ensure that the collected data is sufficiently representative. Here, the first real-time data refers to the set of real-time data from multiple sensors collected within the preset time window during the current correlation pattern learning or update process. Subsequently, based on these first real-time data, their statistical correlation index is calculated. The covariance matrix is ​​a statistic that can effectively characterize the degree of linear correlation between multiple random variables; its elements reflect the trend and strength of synchronous changes in the readings of different sensors. By calculating the covariance matrix, the interdependence between sensor data can be quantified, thereby obtaining a snapshot of the current correlation pattern reflecting the correlation between multiple sensors within the current preset time window. This snapshot of the current correlation pattern is then used as the basis for subsequent comparisons with historically stored correlation patterns.

[0029] To ensure that the stored association patterns can adapt to dynamic changes in the environment, this application introduces an update mechanism. Specifically, a first comparison is made between the newly calculated covariance matrix and the currently stored historical association patterns. This first comparison refers to the process of comparing the degree of difference between the covariance matrix calculated within the current preset time window and the historically stored association patterns. This first comparison aims to assess the degree of difference between the new and old patterns, which can be quantified, for example, by calculating the Euclidean distance, Frobenius norm, or other suitable similarity or difference measures. When the calculated degree of difference is less than a preset difference threshold, it indicates that there is a high degree of consistency between the new data snapshot and the historical patterns. At this point, the system can be considered to be in a stable state, and the new data provides an effective supplement to the existing patterns. Under this condition, the historically stored association patterns will be updated. The update process includes incorporating the newly calculated covariance matrix into the historically stored association patterns using a moving average. The moving average is a smoothing technique that can gradually integrate new information into existing patterns while preserving the stability of historical information and avoiding drastic changes in patterns due to short-term fluctuations. When the difference is not less than the preset difference threshold, the association pattern of the current historical storage remains unchanged, or the covariance matrix is ​​temporarily not updated as data to be observed, so as to avoid abnormal shifts in the association pattern of the historical storage caused by sudden disturbances, abnormal fluctuations or short-term distortions.

[0030] This application's solution captures the latest state of correlation patterns between sensors by periodically acquiring real-time sensor data and calculating its statistical correlation index. By comparing newly calculated correlation patterns with historically stored correlation patterns and determining whether to update them based on the degree of difference, the adaptability of the correlation patterns is ensured. When the difference is small, a moving average method is used for updating, allowing the correlation patterns to smoothly absorb new data features and gradually adapt to subtle changes in the environment without drastic deviations due to instantaneous noise or short-term fluctuations. This mechanism enables edge nodes to continuously learn and maintain a correlation pattern highly matched to the current operating environment, and further determines the dynamic baselines corresponding to multiple sensors in the current environment based on the updated historically stored correlation patterns, thus providing a more solid and accurate foundation for subsequent dynamic baseline determination and anomaly identification.

[0031] In some preferred embodiments, an edge node is assumed to be connected to three sensors: a temperature sensor, a humidity sensor, and a differential pressure sensor, to monitor the environment within the exhaust duct. During normal system operation, the edge node periodically (e.g., every 5 minutes) acquires real-time data from these three sensors over the past minute. A 3x3 covariance matrix is ​​then calculated based on this data, reflecting the correlation between the three sensor readings over the current minute. For example, if a temperature increase is typically accompanied by a slight decrease in differential pressure, the corresponding elements in the covariance matrix will reflect this negative correlation. This newly calculated covariance matrix is ​​then compared to currently stored historical correlation patterns (also a 3x3 covariance matrix), and the degree of difference between them is calculated. If the degree of difference is less than a preset difference threshold (e.g., by calculating the difference in the Frobenius norm of the two matrices), the current environmental change is considered stable, and the new data can be used to optimize the historical pattern. At this point, the new covariance matrix is ​​incorporated into the historically stored correlation patterns using a moving average method. For example, the new historical pattern = (old historical pattern * (N-1) + new covariance matrix) / N, where N is the window size of the moving average. In this way, the historical correlation patterns can be continuously and smoothly updated, always reflecting the actual sensor correlation relationships under the current environment. Based on this, the dynamic baselines corresponding to the temperature sensor, humidity sensor, and differential pressure sensor are continuously corrected, thereby providing an accurate dynamic baseline for subsequent anomaly detection.

[0032] In another embodiment of this application, step S2300 is further proposed to include: S2310: For each of the multiple sensors, set a short-time deviation counter and a deviation duration threshold; S2320: When the reading of any sensor deviates from the corresponding dynamic baseline for the first time, the short-time deviation counter corresponding to that sensor is activated. The short-time deviation counter is used to record the cumulative number of times or the cumulative number of sampling points in which the reading of the corresponding sensor deviates from the corresponding dynamic baseline within a preset short time window. S2330: Within a preset short time window, when the reading of the sensor continuously deviates from the corresponding dynamic baseline, the count value of the short-time deviation counter corresponding to the sensor is incremented; S2340: When the sensor reading recovers to the corresponding dynamic baseline, reset the short-time deviation counter corresponding to the sensor; S2350: When the duration of the deviation of the sensor reading from the corresponding dynamic baseline exceeds the deviation duration threshold, or when the short-time deviation counter corresponding to the sensor reaches the count value of the preset counting threshold, the current correlation between the real-time data of multiple sensors is obtained.

[0033] Specifically, to more accurately determine whether a deviation in sensor readings constitutes an anomaly requiring further analysis, this application sets two key parameters for each local sensor connected to an edge node in the distributed smoke extraction system: a short-time deviation counter and a deviation duration threshold. The short-time deviation counter records the cumulative number of times or sampling points a corresponding sensor reading deviates from its dynamic baseline within a preset short time window. In this application, to ensure this parameter has a clear and actionable meaning, the short-time deviation counter is preferably used to record the cumulative number of sampling points or cumulative detections where the reading continuously deviates from the dynamic baseline within the preset short time window. For example, a short time window can be set to 5 seconds, with sampling once per second; if the reading deviates, the counter increments. The deviation duration threshold is used to set the minimum time required for a sensor reading to continuously deviate from the dynamic baseline, for example, it can be set to 10 seconds. Therefore, the short-time deviation counter reflects the frequency of deviations occurring within a short period, and the deviation duration threshold reflects the continuous duration of a single deviation; both constrain the deviation from the dimensions of frequency and persistence, respectively.

[0034] When the reading of any sensor deviates from its corresponding dynamic baseline for the first time, the short-term deviation counter for that sensor is activated. Here, "first deviation" refers to the state where, within the current preset short-term window, the sensor's reading first moves from within the dynamic baseline range to outside the dynamic baseline range. Subsequently, within the preset short-term window, if the sensor's reading continues to deviate from the dynamic baseline, the short-term deviation counter's count value is incremented. More specifically, within the preset short-term window, each time the sensor's current sampling reading is detected to still deviate from the dynamic baseline, the short-term deviation counter is incremented by 1. Conversely, if the sensor's reading returns to the dynamic baseline range, the corresponding short-term deviation counter is reset to ensure that only continuous or frequent deviations within the short-term window are recorded. Therefore, activating the short-term deviation counter marks the beginning of a deviation, incrementing the short-term deviation counter accumulates the degree of deviation, and resetting the short-term deviation counter terminates the current deviation statistics and prevents historical deviations from interfering with subsequent judgments.

[0035] The system will only trigger the acquisition of the current correlation between real-time data from multiple sensors when the duration of a sensor reading deviating from the corresponding dynamic baseline exceeds a preset deviation duration threshold, or when the short-term deviation counter corresponding to the sensor reaches a preset counting threshold. In other words, the system does not immediately perform multi-sensor correlation analysis when any sensor experiences a momentary deviation. Instead, it first filters the deviation using the short-term deviation counter and the deviation duration threshold. Only when the deviation meets the persistence or frequency conditions is further analysis deemed necessary. For example, if a temperature sensor reading continuously deviates from the dynamic baseline for more than 10 seconds, or if 80% of the sampling points deviate from the dynamic baseline within 5 seconds (assuming the preset counting threshold is 80% of the sampling points), then the deviation is considered to have sufficient persistence or frequency, requiring further analysis of its correlation with other sensor data. Therefore, triggering the acquisition of the current correlation between real-time data from multiple sensors is based on the aforementioned duration or counting judgment results as prerequisites.

[0036] The solution proposed in this application effectively solves the problem of oversensitivity to instantaneous or slight sensor reading deviations in traditional methods by introducing a short-time deviation counter and a deviation duration threshold.

[0037] In some preferred embodiments, a specific example is illustrated below. Assume a temperature sensor and a humidity sensor are connected to an edge node of a distributed smoke extraction system. To monitor their operational status, the system sets a short-term deviation counter and a deviation duration threshold for each sensor. Specifically, the short-term deviation counter is configured to record the number of sampling points where the sensor reading deviates from the dynamic baseline within a preset short time window of 10 seconds, with a preset counting threshold of 8 sampling points (assuming a sampling frequency of 1Hz). Simultaneously, the deviation duration threshold is set to 5 seconds. In this embodiment, the short-term deviation counter is used to count the number of sampling points deviating from the dynamic baseline within 10 seconds, and the deviation duration threshold is used to determine whether a single deviation has persisted for a duration requiring further analysis.

[0038] When a temperature sensor reading first deviates from its dynamic baseline, its short-term deviation counter is activated. If the temperature reading continues to deviate for the next 3 seconds but then returns to normal in the 4th second, the short-term deviation counter is reset, and the system does not trigger the acquisition of the current correlation. This indicates that the deviation may only be a momentary fluctuation. The reason why the acquisition of the current correlation is not triggered here is that the deviation has neither reached the preset deviation duration threshold of 5 seconds nor the preset counting threshold of 8 sampling points.

[0039] However, if the temperature sensor reading continuously deviates from its dynamic baseline, and this deviation persists for 6 seconds, the system will immediately trigger the acquisition of the current correlation between the real-time data of the temperature and humidity sensors. This is because the deviation duration exceeds the preset deviation duration threshold of 5 seconds, in order to further analyze whether this persistent deviation has an abnormal correlation with data from other sensors. In this case, the direct basis for triggering the acquisition of the current correlation is that the deviation duration exceeds the deviation duration threshold, rather than the short-term deviation counter reaching the preset counting threshold.

[0040] For example, if nine sampling points deviate from the dynamic baseline within a preset short time window of 10 seconds, even if the duration of each deviation does not exceed 5 seconds, the system will still trigger the acquisition of the current correlation between real-time data from multiple sensors because the short-term deviation counter has reached the preset counting threshold of eight sampling points. This mechanism ensures that even intermittent but frequent deviations can be captured and analyzed in a timely manner, thus preventing potential sub-optimal conditions from being overlooked. In this case, the direct basis for triggering the acquisition of the current correlation is that the short-term deviation counter reaches the preset counting threshold, rather than the duration of a single deviation exceeding the deviation duration threshold. Therefore, the deviation duration threshold and the short-term deviation counter correspond to the continuous deviation trigger path and the frequent deviation trigger path, respectively; satisfying either one is sufficient to trigger subsequent analysis.

[0041] In another embodiment of this application, it is further proposed that, when acquiring the current correlation between real-time data from multiple sensors, the method further includes: S2351: Within a preset short time window, continuously monitor the deviation of the readings of multiple sensors from their respective dynamic baselines. S2352: Determine whether the readings of at least two of the multiple sensors deviate from their respective dynamic baselines simultaneously or within a preset time interval, and whether either sensor reaches its respective preset counting threshold or deviation duration threshold, and obtain the determination result. S2353: When the judgment result is yes, determine whether the deviation direction and deviation magnitude of the readings of at least two sensors meet the preset cooperative change mode. The preset cooperative change mode is a mode used to characterize the preset correspondence between the deviation direction and deviation magnitude of the readings of multiple sensors. S2354: When the deviation direction and deviation magnitude of the readings of at least two sensors meet the preset cooperative change mode, acquire the current correlation between the real-time data of multiple sensors.

[0042] Specifically, within a preset short time window, edge nodes are configured to continuously monitor the readings of multiple connected local sensors and compare them with corresponding dynamic baselines to identify any deviations. "Continuous monitoring" here means acquiring readings from multiple local sensors continuously according to a preset sampling period within the preset short time window, and comparing each sampled reading with its corresponding dynamic baseline. The preset short time window is a relatively short time period designed to capture instantaneous or short-term changes in sensor readings. The dynamic baseline is dynamically determined based on current environmental conditions and historical data, serving as a dynamic reference value or range characterizing the normal reading status of each sensor under the current environment. Therefore, the purpose of continuously monitoring multiple local sensors within this preset short time window is to identify whether multiple sensors exhibit temporally correlated deviations, rather than simply identifying isolated deviations by a single sensor.

[0043] Furthermore, the system is configured to determine whether the readings of at least two of the multiple sensors deviate from their respective dynamic baselines simultaneously or within a preset time interval. Here, "simultaneously or within a preset time interval" means that the deviation events of these sensors are closely correlated in time, such as occurring within the same sampling period or within a very short interval. Importantly, this determination also requires that none of these deviating readings reach their respective preset counting thresholds or deviation duration thresholds. In other words, the system focuses on identifying a situation in this step where "the deviation of a single sensor is insufficient to trigger anomaly analysis alone, but the deviations of multiple sensors are correlated in time." This means that the deviation of a single sensor is insufficient to be judged as an anomaly alone, but its coordinated deviation with other sensors may indicate a potential problem. Therefore, whether multiple sensors deviate simultaneously or within a preset short time window constitutes a prerequisite for subsequent determination of whether the direction and magnitude of the deviations of multiple sensors meet a preset coordinated change pattern.

[0044] In a preferred implementation, when the determination result is yes, the system will further determine whether the deviation direction and deviation magnitude of the readings of at least two sensors satisfy a preset coordinated change pattern. Here, "deviation direction" refers to the direction of increase or decrease of the sensor reading relative to the corresponding dynamic baseline, and "deviation magnitude" refers to the amount of deviation between the current sensor reading and the corresponding dynamic baseline. The preset coordinated change pattern is a predefined pattern used to characterize a preset correspondence between the deviation directions and deviation magnitudes of multiple sensor readings. For example, in a smoke extraction system, when a component begins to degrade, it may manifest as a slight increase in the temperature sensor reading and a slight decrease in the air pressure sensor reading, with a specific proportional relationship between the magnitudes of these changes. This coordinated change pattern can be established based on historical data, expert experience, or physical models. Therefore, only when the deviations of multiple sensors simultaneously satisfy both the time correlation condition and the coordinated change pattern condition will the system identify the deviations of these multiple sensors as a coordinated deviation event with further analytical value.

[0045] In some preferred embodiments, it is assumed that an edge node in a smoke exhaust system is connected to three sensors: a temperature sensor, a pressure sensor, and a wind speed sensor. Under normal operating conditions, there is a specific correlation pattern and dynamic baseline between the readings of these three sensors. When a fan bearing in the smoke exhaust system begins to wear slightly, it may cause a slight decrease in fan efficiency. At this time, the wind speed sensor reading may decrease slightly, the temperature sensor reading may increase slightly, while the pressure sensor reading may remain relatively stable or fluctuate slightly. The deviation of these individual sensors may be very small, insufficient to trigger their respective preset counting thresholds or deviation duration thresholds. However, the solution of this application continuously monitors the readings of these three sensors. Once a slight decrease in the wind speed sensor reading and a slight increase in the temperature sensor reading are detected, and these two deviations occur within a preset time interval, and their deviation direction (wind speed decrease, temperature increase) and deviation magnitude (e.g., wind speed decrease of 0.5%, temperature increase of 0.2°C) meet the preset "fan efficiency decrease" co-change pattern (e.g., wind speed and temperature are negatively correlated, and the change magnitude is within a specific range), the system will immediately obtain the current correlation between the real-time data of these sensors and identify it as an abnormal operating state. In this embodiment, the reason why subsequent analysis is triggered even when the deviation of each individual sensor is insufficient to individually trigger the preset counting threshold or deviation duration threshold is because the deviations of the wind speed sensor and the temperature sensor simultaneously satisfy the time correlation condition and the "wind turbine efficiency decline" co-change mode condition. This approach allows the system to issue an early warning in the early stages of wind turbine bearing wear, much earlier than when the readings of a single sensor reach a severely abnormal threshold.

[0046] In another embodiment of this application, a method for recording and aggregating deviation events after acquiring the current correlation between real-time data from multiple sensors is further proposed, specifically including: S2355-1: Assign an independent deviation state recorder to each of the multiple sensors. The deviation state recorder is used to record deviation events where the corresponding sensor reading deviates from the corresponding dynamic baseline. The deviation event includes at least the sensor identifier, deviation start time, deviation end time and deviation magnitude. S2355-2: When the reading of one of the multiple sensors deviates from the corresponding dynamic baseline, the deviation status recorder corresponding to that sensor is activated, and the timestamp of the current moment of that sensor is recorded as the deviation start time. S2355-3: During the period when the sensor reading continuously deviates from the corresponding dynamic baseline, continuously update the deviation magnitude in the deviation state recorder; S2355-4: When the sensor reading recovers to the corresponding dynamic baseline, or when the deviation magnitude changes more than the preset event segmentation threshold relative to the recorded deviation magnitude of the current deviation event, the timestamp of the current moment is recorded as the deviation end time, and the recording of the current deviation event is completed. S2355-5: Periodically scan the deviation state recorders corresponding to multiple sensors and read the deviation events that have been recorded; S2355-6: Based on the preset time window overlap rule and deviation amplitude superposition rule, the deviation events of different sensors are aggregated to form an aggregated deviation event sequence. The time window overlap rule is used to determine whether the deviation events of different sensors belong to the same aggregated event in time, and the deviation amplitude superposition rule is used to calculate the aggregated deviation intensity of multiple deviation events in the same aggregated event.

[0047] Specifically, a deviation state recorder can be understood as a data structure or storage unit configured specifically to record detailed information about a single sensor reading deviating from its dynamic baseline. This "detailed information about a single sensor reading deviating from its dynamic baseline" refers to the structured record content surrounding a complete deviation event. This information includes at least the sensor's unique identifier to distinguish which sensor deviated; the deviation start time, i.e., the time when the sensor reading first deviated from the dynamic baseline; the deviation end time, i.e., the time when the sensor reading returned to normal or the deviation characteristics changed significantly; and the deviation magnitude, used to quantify the degree to which the reading deviated from the dynamic baseline. Therefore, the role of the deviation state recorder is not only to record whether a deviation occurred, but also to fully describe the evolution of the deviation event from beginning to end.

[0048] When a sensor's reading first deviates from its corresponding dynamic baseline, the deviation status recorder for that sensor is activated. At this point, the system records the current timestamp as the start time of the deviation event. During the period when the sensor reading continues to deviate from the dynamic baseline, the deviation status recorder continuously updates the deviation magnitude; for example, it can record the maximum deviation value, the average deviation value, or the trend of deviation value changes. That is, after the deviation status recorder is activated, the system does not only record the start time of the deviation but continuously tracks the change in the degree of deviation during the duration of the event. When the sensor reading returns to within the dynamic baseline, or when the deviation magnitude changes significantly relative to the currently recorded deviation magnitude and exceeds a preset event segmentation threshold, the system records the current timestamp as the end time of the deviation and completes the recording of the current deviation event. The preset event segmentation threshold aims to distinguish between fluctuations within the same deviation event and new, independent deviation events. For example, when the deviation magnitude suddenly increases or decreases significantly, it may mean the end of an old deviation event and the beginning of a new one. Therefore, the function of the preset event segmentation threshold is to set clear boundaries for deviation events, thereby avoiding mistaking deviation processes with significantly changed characteristics as the same continuous event.

[0049] To process and analyze these deviation events promptly, the system periodically scans the deviation state recorders corresponding to multiple sensors to read all recorded deviation events. Here, "recorded deviation events" refers to those with clearly defined start and end times and magnitudes. These independent deviation events are then aggregated according to preset time window overlap and deviation magnitude superposition rules. The time window overlap rule determines whether deviation events from different sensors have sufficient temporal overlap to be considered part of the same aggregated event. For example, if deviation events from two sensors overlap by more than a certain percentage or duration, they may be aggregated. The deviation magnitude superposition rule calculates the aggregated deviation intensity of multiple deviation events within the same aggregated event. For example, the aggregated intensity can be obtained by weighted summation or taking the maximum value of the deviation magnitudes of each deviation event. In this way, an aggregated deviation event sequence can be formed, reflecting more macroscopic and complex anomaly patterns. In other words, periodic scanning is used to obtain event inputs that can participate in aggregation, time window overlap rules are used to determine which deviation events should be merged into the same aggregation event, deviation magnitude superposition rules are used to quantify the overall intensity of the aggregation event, and the aggregation deviation event sequence is the result of the above processing.

[0050] In some preferred embodiments, the following specific example illustrates the situation: Suppose an edge node is connected to three sensors: temperature sensor A, pressure sensor B, and humidity sensor C, each monitoring key parameters of a smoke exhaust valve. First, an offset logger is assigned to each of sensors A, B, and C. When the reading of temperature sensor A first deviates from its dynamic baseline at a certain moment (e.g., T1), the offset logger for sensor A is activated, and T1 is recorded as the start time of the deviation.

[0051] Over the following time period, if the temperature continues to deviate, the deviation magnitude will be continuously updated. Assuming that the temperature reading returns to normal at time T2, T2 is recorded as the deviation end time, completing the first deviation event recording for sensor A (event A1). Subsequently, at time T3, the reading of pressure sensor B begins to deviate, its deviation state recorder is activated, and T3 is recorded as the deviation start time. At time T4, the reading of humidity sensor C also begins to deviate, its recorder is activated, and T4 is recorded as the deviation start time. Assuming that the reading of pressure sensor B continues to deviate between T3 and T5, but at time T5, its deviation magnitude suddenly increases significantly, exceeding the preset event segmentation threshold.

[0052] At this point, the system records T5 as the end time of the first deviation event (event B1) of pressure sensor B, and immediately starts a new deviation event (event B2), recording T5 as the new deviation start time to distinguish between two deviations with different characteristics. The system periodically scans these recorders, reading completed deviation events (e.g., events A1, B1). When the deviation events of sensors B and C (e.g., events B2 and C1) overlap in time, the system applies a preset time window overlap rule. For example, if the overlap time of events B2 and C1 exceeds 50% of their respective durations, they are determined to belong to the same aggregated event. Then, according to a preset deviation amplitude superposition rule (e.g., adding the normalized deviation amplitudes), the aggregated deviation intensity of this aggregated event is calculated.

[0053] Ultimately, these aggregated events form an aggregated deviation event sequence. This sequence more comprehensively reflects anomalies that may occur simultaneously or collaboratively across multiple sensors at the edge node within a specific time period. For example, simultaneous anomalies in temperature, pressure, and humidity may indicate a more serious malfunction in the smoke exhaust valve. In this embodiment, events A1, B1, B2, and C1 are all individual sensor deviation events, while the result formed by events B2 and C1 according to the time window overlap rule and the deviation magnitude superposition rule constitutes an aggregated deviation event. This aggregated deviation event is further used as a component unit in the aggregated deviation event sequence for subsequent anomaly analysis.

[0054] In another embodiment of this application, for the aggregation deviation event sequence, it is further proposed that: S2356-1: Identify the source region and type of deviation events of individual sensors contained in each aggregated event in an aggregated deviation event sequence; S2356-2: Based on the identified source region and type, aggregated events are classified into homogeneous coordinated deviation events, heterogeneous overlapping deviation events, or composite deviation events. Homogeneous coordinated deviation events are aggregated events containing multiple single sensor deviation events originating from the same region and of the same type or belonging to a preset association type combination, and exhibiting coordinated changes in deviation direction or deviation magnitude. Heterogeneous overlapping deviation events are aggregated events containing multiple single sensor deviation events originating from different regions or of types that do not meet the preset association type combination, and overlapping only in time. Composite deviation events are aggregated events that simultaneously contain homogeneous coordinated deviation event components and heterogeneous overlapping deviation event components. S2356-3: For homogeneous collaborative deviation events, the aggregation deviation intensity is compared with a preset aggregation intensity threshold, and the duration of the homogeneous collaborative deviation event is compared with a preset duration threshold to determine whether the homogeneous collaborative deviation event represents an abnormal operating state. The duration is determined by the time length between the earliest deviation start time and the latest deviation end time among the deviation events belonging to the same aggregation event, so as to obtain the corresponding anomaly analysis results. S2356-4: For heterogeneous overlapping deviation events, the deviation amplitude and duration of each individual sensor's deviation event are compared with the corresponding sensor's preset anomaly judgment conditions to determine whether each individual sensor's deviation event represents an abnormal operating state, so as to obtain the corresponding anomaly analysis results. S2356-5: For compound deviation events, simultaneously perform judgment and processing on co-originating deviation events and heterogeneous overlapping deviation events to obtain the corresponding anomaly analysis results; S2356-6: Based on the anomaly analysis results, determine the anomaly type description corresponding to the abnormal operating status of the edge node.

[0055] Specifically, upon receiving the aggregated deviation event sequence, each aggregated event first requires in-depth analysis. "Receiving the aggregated deviation event sequence" here means that the edge node has completed the aggregation processing of multiple individual sensor deviation events, forming an aggregated deviation event sequence that can be used for subsequent classification and judgment. This includes identifying the specific source region of the individual sensor deviation events contained within each aggregated event and their corresponding sensor types. For example, an aggregated event may contain deviation events from multiple temperature and pressure sensors from the same exhaust valve region, or it may contain deviation events from independent sensors (such as a temperature sensor and a vibration sensor) from different regions. Therefore, the starting point for in-depth analysis of each aggregated event is to first clarify which individual sensor deviation events constitute the aggregated event, and where these individual sensor deviation events originate and what type they belong to.

[0056] Based on the identification of the source region and type of individual sensor deviation events, aggregated events are classified into three main types: co-originating deviation events, heterogeneous overlapping deviation events, or composite deviation events. Co-originating deviation events refer to aggregated events in which multiple individual sensor deviation events originate from the same physical region, and these sensors are of the same type or belong to a pre-defined combination of related types (e.g., temperature and humidity sensors in the same region). Furthermore, these deviation events exhibit coordinated changes in deviation direction (e.g., simultaneous increase or simultaneous decrease) or deviation magnitude. This coordination usually indicates a common, systematic cause of the anomaly. Heterogeneous overlapping deviation events refer to aggregated events in which multiple individual sensor deviation events originate from different physical regions, or even if they originate from the same region, their types do not meet the pre-defined combination of related types. These events only overlap in time, but do not exhibit significant coordination in deviation direction or magnitude. Such events may indicate multiple independent local anomalies that happen to overlap in time, or they may be caused by non-systematic factors such as environmental noise. A composite deviation event refers to an aggregated event that simultaneously contains characteristics of both co-originating and heterogeneous overlapping deviation events. In other words, some sensor deviations exhibit synergy, while others show independence. Therefore, co-originating deviation events emphasize a common source and synergistic relationship, heterogeneous overlapping deviation events emphasize temporal overlap but lack synergy, and composite deviation events emphasize the simultaneous existence of both types of deviation components with different properties within the same aggregated event.

[0057] After classifying aggregated events, differentiated judgment and processing logic is applied to different types of aggregated events. For homogeneous coordinated deviation events, the judgment logic focuses on wholeness and coordination. Specifically, the aggregated deviation intensity of the event is compared with a preset aggregated intensity threshold, and its duration is compared with a preset duration threshold. The duration is defined as the time length between the earliest deviation start time and the latest deviation end time among the deviation events belonging to the same aggregated event. In this way, it can be determined whether the coordinated deviation is significant and sustained enough to characterize the abnormal operating state of the edge node. For heterogeneous overlapping deviation events, due to their lack of coordination, the judgment logic focuses more on the independent anomaly of a single sensor. Therefore, the deviation amplitude and duration of each individual sensor deviation event are compared with the preset anomaly judgment conditions corresponding to that sensor. This method can independently evaluate the anomaly of each sensor, avoiding the obscuring or amplification of the true state of a single sensor due to aggregation. For composite deviation events, since they combine the characteristics of both types of events, the judgment and processing for homogeneous coordinated deviation events and heterogeneous overlapping deviation events are performed simultaneously. This means that both the overall coordinated deviation strength and duration are assessed, as well as individual sensor deviation events exhibiting independence are evaluated independently. Ultimately, based on the analysis results of different types of anomalies, the anomaly type description corresponding to the abnormal operating state of the edge node can be determined more accurately and meticulously. In other words, the classification result of the aggregated events directly determines which judgment and processing logic is used subsequently, and the output results of different judgment and processing logics are further used to determine the anomaly type description corresponding to the abnormal operating state.

[0058] The solution proposed in this application effectively solves the problem that simply aggregating events cannot accurately diagnose the nature of anomalies. This is achieved by refining the classification and differential judgment of aggregated deviation event sequences.

[0059] In another embodiment of this application, it is further proposed that, based on the identified source region and type, aggregation events be classified into homogeneous co-deviation events, heterogeneous overlapping deviation events, or composite deviation events, including: S2356-21: Extract the event features of the aggregated events, which shall include at least the aggregated deviation intensity, the duration of the aggregated events, the combination of sensor types involved, and the time series distribution characteristics of the deviation events; S2356-22: Calculate the similarity between the aggregated event and each category based on the event characteristics. The categories include homologous co-existing deviation events, heterologous overlapping deviation events, and composite deviation events. S2356-23: Determine the classification confidence level corresponding to the aggregated event based on similarity; S2356-24: When the classification confidence level is lower than the preset confidence threshold, the fuzzy classification processing procedure is initiated; S2356-25: In the fuzzy classification process, extract the key features of the aggregated event and compare the key features with the typical features corresponding to multiple related categories to obtain the comparison results; S2356-26: Based on the comparison results, adjust the comparison thresholds between the aggregation deviation intensity and the preset aggregation intensity threshold, the duration of the aggregation event and the preset duration threshold, and the deviation event of each individual sensor and the preset abnormal judgment condition of the corresponding sensor, or simultaneously perform judgments corresponding to multiple related categories and weight the multiple judgment results to generate a comprehensive diagnostic result for the aggregation event. S2356-27: Based on the comprehensive diagnostic results, update the classification results corresponding to the aggregated events, and determine the corresponding anomaly analysis results based on the updated classification results.

[0060] Specifically, when classifying aggregated events, the first step is to extract the event features. Here, "event features" refers to a set of structured features that characterize the overall attributes of the aggregated event and can be used for category differentiation. These event features are key information describing the essential attributes of the aggregated event, and may include at least the intensity of the aggregated deviation, which characterizes the overall degree of anomaly; the duration of the aggregated event, which reflects the duration of the anomaly; the combination of sensor types involved, which reveals the types of sensors involved in the anomaly and their interrelationships; and the time-series distribution characteristics of the deviation event, which describes the evolution of the deviation event over time. Therefore, the results of event feature extraction constitute the direct input for subsequent similarity calculations and classification confidence assessments.

[0061] Based on the extracted event features, the system calculates the similarity between the aggregated event and each preset category. These categories include co-originating deviation events, hetero-originating overlapping deviation events, and composite deviation events. Similarity calculation can employ various mathematical or statistical methods, such as distance metrics based on feature vectors or probabilistic model matching, to quantify the closeness of the aggregated event to each typical category pattern. In other words, the higher the similarity, the closer the aggregated event is to the typical pattern of the corresponding preset category at the event feature level.

[0062] Subsequently, based on the calculated similarity, the classification confidence level corresponding to the aggregated event can be determined. Classification confidence level is an indicator of the reliability of classification results, typically expressed as the probability or score of the aggregated event belonging to a specific category. Therefore, the classification confidence level is not generated independently, but rather further derived from the similarity calculation results between the aggregated event and each preset category.

[0063] When the classification confidence level falls below a preset confidence threshold, it indicates a high degree of uncertainty in the current classification result. At this point, the system will initiate a fuzzy classification process. This process aims to handle aggregated events with ambiguous boundaries and difficult-to-categorize characteristics, avoiding arbitrary judgments due to insufficient information. In other words, the fuzzy classification process is not executed for all aggregated events, but only triggered when the classification confidence level cannot support a clear classification conclusion.

[0064] In the fuzzy classification process, key features of aggregated events are further extracted. Here, "key features" refer to those features that, compared to general event features, better reflect category differences and are more suitable for in-depth comparison within the current fuzzy classification process. These key features are attributes with greater discriminative and diagnostic value than general event features; for example, abnormal patterns of specific sensor combinations or dynamic trends in deviation magnitude. Subsequently, the key features are compared with typical features corresponding to multiple related categories to obtain comparison results. This comparison process aims to deeply analyze the intrinsic attributes of aggregated events and determine their deep correlation with the patterns of each typical category. Therefore, key feature extraction and typical feature comparison constitute the core analytical steps of the fuzzy classification process, and the comparison results serve as the basis for subsequent adjustments to judgment criteria or the fusion of judgment results.

[0065] Based on the comparison results, the system can adopt two strategies to generate a comprehensive diagnostic result for aggregated events. One strategy involves adjusting the comparison thresholds between the aggregated deviation intensity and a preset aggregated intensity threshold, the duration of the aggregated event and a preset duration threshold, and the deviation events of each individual sensor and the preset anomaly judgment conditions of the corresponding sensor. This means the system dynamically adjusts the judgment criteria according to the fuzzy characteristics of the event, making it more adaptable to the actual situation of the current event. The other strategy involves simultaneously performing judgments corresponding to multiple related categories and weighting the results of these judgments. For example, if an event simultaneously possesses certain characteristics of both homogeneous collaboration and heterogeneous overlap, the system can assign different weights to the judgment results of different categories based on the comparison results of its key features, thereby generating a more comprehensive and detailed diagnostic result. Therefore, the comprehensive diagnostic result either originates from a re-judgment after adjusting various judgment thresholds or from a comprehensive output after weighting and combining the judgment results of multiple related categories.

[0066] Finally, based on the comprehensive diagnostic results, the system updates the classification results corresponding to the aggregated events and determines the corresponding anomaly analysis results based on the updated classification results. Here, "updating the classification results corresponding to the aggregated events" means correcting the original classification conclusions based on initial similarity and classification confidence to classification conclusions processed by a fuzzy classification process; "corresponding anomaly analysis results" refers to the judgment results that match the updated classification results. These updated classification results are no longer simple binary judgments, but rather more precise descriptions reflecting the complexity and ambiguity of the events, thus providing a more reliable basis for subsequent anomaly handling. Therefore, the comprehensive diagnostic results are not the final output itself, but rather an intermediate basis used to update the classification results and further determine the anomaly analysis results.

[0067] In another embodiment of this application, it is further proposed that summary information be generated based on the abnormal operating state, including: S3100: Extract the identification information, anomaly occurrence time information, anomaly severity information, and anomaly type description of the edge nodes corresponding to the abnormal operating state; S3200: Encapsulates the identification information of edge nodes, the time of anomaly occurrence, the severity of anomalies, and the description of anomaly types according to a preset summary data format to generate summary information.

[0068] Specifically, after identifying the abnormal operating state of an edge node, it is necessary to extract key information from this abnormal operating state. Here, "key information" refers to structured information content that can characterize the core attributes of the abnormal operating state and can be directly used to generate summary information. This key information includes at least the edge node's identification information, such as its unique identifier or name, so that the central control room can accurately identify the malfunctioning device; the anomaly occurrence time information, used to record the specific time when the anomaly began or was identified, which is crucial for subsequent fault tracing and analysis; the anomaly severity information, used to quantify the severity or urgency of the anomaly, for example, it can be divided into minor, moderate, and severe levels, so that the central control room can prioritize high-severity anomalies; and an anomaly type description, used to detail the specific nature of the anomaly, such as slow performance changes, localized degradation, or intermittent failure, which helps the central control room quickly understand the problem and take targeted measures. Therefore, the purpose of extracting key information from the abnormal operating state is to convert the abnormal operating state, originally used for local identification and judgment of the edge node, into summary information content that can be directly received, identified, and processed by the central control room.

[0069] Furthermore, to ensure that this critical information can be effectively received and parsed by the central control room, the extracted edge node identification information, anomaly occurrence time information, anomaly severity information, and anomaly type description need to be encapsulated according to a preset summary data format. The preset summary data format can be a standardized data structure, such as JSON, XML, or other custom lightweight data protocols, aiming to unify data representation and facilitate automated parsing and processing by the central control room system. Through this encapsulation, scattered anomaly information can be organized into a structured, easily transmitted, and understandable summary. In other words, key information extraction corresponds to determining the content of the summary information, while encapsulation according to the preset summary data format corresponds to unifying the expression form of the summary information; both together constitute the process of generating the summary information.

[0070] Reference Figure 2 The specific embodiments of this application also disclose an edge computing processing system for smoke exhaust valve detection data, which is applied to edge nodes in a distributed smoke exhaust system. The system includes: The acquisition module 1 is used to acquire the operating status parameters inside the edge node. The operating status parameters include processor workload parameters, memory usage parameters, communication status parameters between the edge node and the local sensor, power stability parameters, and / or internal ambient temperature parameters. Analysis module 2 is used to compare the operating status parameters with the corresponding preset normal range, change amplitude threshold and duration threshold respectively. When the operating status parameters exceed the preset normal range, or the change amplitude of the operating status parameters exceeds the change amplitude threshold and the duration reaches the duration threshold, the abnormal operating status of the edge node is identified. The abnormal operating status is used to characterize the edge node in a sub-healthy state of slow performance change, local degradation or intermittent failure. The sub-healthy state is the state in which the edge node has not completely failed but the operating performance deviates abnormally. The generation module 3 is used to generate summary information based on the abnormal operating state. The summary information includes at least the identification information of the edge nodes and the description of the abnormal type corresponding to the abnormal operating state. The sending module 4 is used to send summary information to the central control room in a manner that is lower than the transmission priority of the smoke exhaust control command, so that the central control room can obtain the abnormal operating status information of the edge node without receiving the original operating data of the edge node.

[0071] The edge computing processing system for smoke exhaust valve detection data in this application aims to address the shortcomings of traditional distributed smoke exhaust systems in monitoring the sub-health status of edge nodes. This system achieves real-time, intelligent monitoring and analysis of its own operational status parameters by deploying functional modules within the edge nodes.

[0072] The above description is merely an embodiment of this application and is not intended to limit the scope of protection of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A method for edge computing processing of smoke exhaust valve detection data, characterized in that, The method, applied to edge nodes in a distributed smoke extraction system, includes: The operating status parameters inside the edge node are obtained, including processor workload parameters, memory usage parameters, communication status parameters between the edge node and the local sensor, power stability parameters, and / or internal ambient temperature parameters. The operating status parameters are compared with the corresponding preset normal range, change amplitude threshold, and duration threshold. When the operating status parameters exceed the preset normal range, or the change amplitude of the operating status parameters exceeds the change amplitude threshold and the duration reaches the duration threshold, the abnormal operating status of the edge node is identified. The abnormal operating status is used to characterize the edge node as being in a sub-healthy state with slow performance changes, local degradation, or intermittent failures. The sub-healthy state is a state in which the edge node has not completely failed but its operating performance deviates abnormally. Based on the abnormal operating state, a summary information is generated, which includes at least the identification information of the edge node and the description of the abnormal type corresponding to the abnormal operating state. The summary information is sent to the central control room in a manner that is lower than the transmission priority of the smoke exhaust control command, so that the central control room can obtain the abnormal operating status information of the edge node without receiving the original operating data of the edge node.

2. The edge computing processing method for smoke exhaust valve detection data according to claim 1, characterized in that, The process of identifying the abnormal operating state of the edge nodes includes: Acquire real-time data from multiple sensors in the local sensor suite, and perform time correlation on the real-time data; When the edge node is in a normal state, the association pattern of the time correlation between the real-time data of multiple sensors is learned and stored, and the dynamic baseline corresponding to each of the multiple sensors in the current environment is determined based on the association pattern. The association pattern is a pattern used to characterize the correlation between the real-time data of multiple sensors, and the dynamic baseline is a dynamic reference value or dynamic reference range used to characterize the normal reading state of each sensor in the current environment. When the reading of any of the plurality of sensors deviates from the corresponding dynamic baseline, the current correlation between the real-time data of the plurality of sensors is obtained; The current association is matched with the association pattern to obtain the matching result; Based on the matching results, it is determined whether the abnormal reading of the sensor whose reading deviates from the corresponding dynamic baseline is caused by environmental factors or by the performance degradation of the sensor, thus obtaining the abnormal operating state of the edge node.

3. The edge computing processing method for smoke exhaust valve detection data according to claim 2, characterized in that, The learning and storage of association patterns among the real-time data from multiple sensors, including: Acquire the first real-time data from the multiple sensors within a preset time window; Calculate the statistical correlation index of the first real-time data within the preset time window, wherein the statistical correlation index includes the covariance matrix; The covariance matrix is ​​compared with the historically stored association patterns in a first comparison, the first comparison including calculating the degree of difference between the covariance matrix and the historically stored association patterns. When the difference is less than a preset difference threshold, the association pattern of the historical storage is updated. The update includes incorporating the covariance matrix into the association pattern of the historical storage in a moving average manner.

4. The edge computing processing method for smoke exhaust valve detection data according to claim 2, characterized in that, The step of obtaining the current correlation between the real-time data of the plurality of sensors when the reading of any of the plurality of sensors deviates from the corresponding dynamic baseline includes: For each of the plurality of sensors, a short-term deviation counter and a deviation duration threshold are set; When the reading of any sensor deviates from the corresponding dynamic baseline for the first time, the short-time deviation counter corresponding to that sensor is activated. The short-time deviation counter is used to record the cumulative number of times or the cumulative number of sampling points when the reading of the corresponding sensor deviates from the corresponding dynamic baseline within a preset short time window. Within a preset short time window, when the reading of the sensor continuously deviates from the corresponding dynamic baseline, the count value of the short-time deviation counter corresponding to the sensor is incremented; When the sensor reading recovers to the corresponding dynamic baseline, the short-time deviation counter corresponding to the sensor is reset; When the duration of the deviation of the sensor reading from the corresponding dynamic baseline exceeds the deviation duration threshold, or when the short-term deviation counter corresponding to the sensor reaches the count value of a preset counting threshold, the current correlation between the real-time data of the multiple sensors is obtained.

5. The edge computing processing method for smoke exhaust valve detection data according to claim 4, characterized in that, The step of acquiring the current correlation between the real-time data of the multiple sensors also includes: Within the preset short time window, the deviation of the readings of each of the multiple sensors from the corresponding dynamic baseline is continuously monitored; Determine whether the readings of at least two of the plurality of sensors deviate from their respective dynamic baselines simultaneously or within a preset time interval, and whether neither reaches their respective preset counting thresholds or deviation duration thresholds, and obtain the determination result; When the determination result is yes, it is determined whether the deviation direction and deviation magnitude of the readings of the at least two sensors meet the preset coordinated change mode. The preset coordinated change mode is a mode used to characterize the preset correspondence between the deviation direction and deviation magnitude of the readings of multiple sensors. When the deviation direction and magnitude of the readings of at least two sensors satisfy the preset coordinated change mode, the current correlation between the real-time data of the multiple sensors is obtained.

6. The edge computing processing method for smoke exhaust valve detection data according to claim 5, characterized in that, After acquiring the current correlation between the real-time data of the plurality of sensors, the method further includes: Each of the plurality of sensors is assigned an independent deviation state recorder, which is used to record deviation events where the corresponding sensor reading deviates from the corresponding dynamic baseline. The deviation event includes at least the sensor identifier, deviation start time, deviation end time and deviation magnitude. When the reading of one of the plurality of sensors deviates from the corresponding dynamic baseline, the deviation state recorder corresponding to that sensor is activated, and the timestamp of the current moment of that sensor is recorded as the deviation start time. During the period when the sensor reading continuously deviates from the corresponding dynamic baseline, the deviation magnitude in the deviation state recorder is continuously updated; When the sensor reading recovers to the corresponding dynamic baseline, or when the deviation amplitude changes more than a preset event segmentation threshold relative to the recorded deviation amplitude of the current deviation event, the timestamp of the current moment is recorded as the deviation end time, and the recording of the current deviation event is completed. Periodically scan the deviation state recorders corresponding to each of the multiple sensors and read the deviation events that have been recorded; According to the preset time window overlap rule and deviation magnitude superposition rule, the deviation events of different sensors are aggregated to form an aggregated deviation event sequence. The time window overlap rule is used to determine whether the deviation events of different sensors belong to the same aggregated event in time, and the deviation magnitude superposition rule is used to calculate the aggregated deviation intensity of multiple deviation events in the same aggregated event.

7. The edge computing processing method for smoke exhaust valve detection data according to claim 6, characterized in that, For the aforementioned aggregated deviation event sequence, it also includes: Identify the source region and type of the deviation event of the individual sensor contained in each aggregated event in the aggregated deviation event sequence; Based on the identified source region and type, the aggregated events are classified into homogeneous coordinated deviation events, heterogeneous overlapping deviation events, or composite deviation events. Homogeneous coordinated deviation events are aggregated events comprising multiple single sensor deviation events originating from the same region, of the same type, or belonging to a preset association type combination, and exhibiting coordinated changes in deviation direction or magnitude. Heterogeneous overlapping deviation events are aggregated events comprising multiple single sensor deviation events originating from different regions or of types that do not satisfy the preset association type combination, and overlapping only in time. Composite deviation events are aggregated events that simultaneously contain components of homogeneous coordinated deviation events and heterogeneous overlapping deviation events. For the homogeneous collaborative deviation event, the aggregation deviation intensity is compared with a preset aggregation intensity threshold, and the duration of the homogeneous collaborative deviation event is compared with a preset duration threshold to determine whether the homogeneous collaborative deviation event represents an abnormal operating state. The duration is determined by the time length between the earliest deviation start time and the latest deviation end time among the deviation events belonging to the same aggregation event, so as to obtain the corresponding anomaly analysis result. For the heterogeneous overlapping deviation event, the deviation amplitude and duration of the deviation event of each individual sensor are compared with the preset anomaly judgment conditions of the corresponding sensor to determine whether the deviation event of each individual sensor represents an abnormal operating state, so as to obtain the corresponding anomaly analysis results. For the composite deviation event, the judgment and processing of the homogeneous cooperative deviation event and the heterogeneous overlapping deviation event are performed simultaneously to obtain the corresponding anomaly analysis results; Based on the anomaly analysis results, determine the anomaly type description corresponding to the abnormal operating state of the edge node.

8. The edge computing processing method for smoke exhaust valve detection data according to claim 7, characterized in that, The process of classifying the aggregated events into homogeneous co-existing deviation events, heterogeneous overlapping deviation events, or composite deviation events based on the identified source region and type includes: Extract the event features of the aggregated events, which include at least the aggregated deviation intensity, the duration of the aggregated events, the combination of sensor types involved, and the time series distribution features of the deviation events; Based on the event characteristics, the similarity between the aggregated event and each category is calculated, where the categories include the homologous co-existing deviation event, the heterologous overlapping deviation event, and the composite deviation event. Based on the similarity, determine the classification confidence level corresponding to the aggregated event; When the classification confidence level is lower than the preset confidence threshold, the fuzzy classification process is initiated. In the fuzzy classification process, key features of the aggregated event are extracted, and the key features are compared with typical features corresponding to multiple related categories to obtain comparison results. Based on the comparison results, the comparison thresholds between the aggregation deviation intensity and the preset aggregation intensity threshold, the duration of the aggregation event and the preset duration threshold, and the deviation event of each individual sensor and the preset abnormal judgment condition of the corresponding sensor are adjusted respectively. Alternatively, multiple judgments corresponding to related categories are performed simultaneously, and the multiple judgment results are weighted and combined to generate a comprehensive diagnostic result of the aggregation event. Based on the comprehensive diagnostic results, the classification results corresponding to the aggregated events are updated, and the corresponding anomaly analysis results are determined based on the updated classification results.

9. The edge computing processing method for smoke exhaust valve detection data according to claim 1, characterized in that, The step of generating summary information based on the abnormal operating state includes: Extract the identification information, anomaly occurrence time information, anomaly severity information, and anomaly type description of the edge nodes corresponding to the abnormal operating state; The edge node identification information, the anomaly occurrence time information, the anomaly severity information, and the anomaly type description are encapsulated according to a preset summary data format to generate the summary information.

10. An edge computing processing system for smoke exhaust valve detection data, characterized in that, Edge nodes are used in a distributed smoke extraction system, the system comprising: The acquisition module is used to acquire the operating status parameters inside the edge node, including processor workload parameters, memory usage parameters, communication status parameters between the edge node and the local sensor, power stability parameters, and / or internal ambient temperature parameters. The analysis module is used to compare the operating status parameters with the corresponding preset normal range, change amplitude threshold, and duration threshold, respectively. When the operating status parameters exceed the preset normal range, or the change amplitude of the operating status parameters exceeds the change amplitude threshold and the duration reaches the duration threshold, the abnormal operating status of the edge node is identified. The abnormal operating status is used to characterize the edge node as being in a sub-healthy state with slow performance changes, local degradation, or intermittent failures. The sub-healthy state is a state in which the edge node has not completely failed but its operating performance deviates abnormally. The generation module is used to generate summary information based on the abnormal operating state. The summary information includes at least the identification information of the edge nodes and the description of the abnormal type corresponding to the abnormal operating state. The sending module is used to send the summary information to the central control room in a manner that is lower than the transmission priority of the smoke exhaust control command, so that the central control room can obtain the abnormal operating status information of the edge node without receiving the original operating data of the edge node.

Images