A method and system for intelligent factory safety management

Abstract

The application provides a kind of intelligent factory safety management method and system, related to production safety analysis technical field.The method comprises: collecting the historical production safety record data of multiple production safety events, and extracting the historical equipment state data corresponding to each production safety event respectively;Parameter evolution anomaly detection is carried out on each group of historical equipment state data, a plurality of abnormal evolution units in the historical equipment state data are determined, and an evolution feature vector of each abnormal evolution unit is constructed;Evolution correlation analysis is carried out on the plurality of abnormal evolution units, and the evolution correlation parameters between any two abnormal evolution units are calculated;An evolution chain graph of the plurality of abnormal evolution units is constructed, and a plurality of target evolution links are determined;According to the plurality of target evolution links, the production state safety analysis of real-time production safety monitoring data is carried out, and the production safety detection result of the target production equipment is generated.The application realizes intelligent safety analysis and early warning of production equipment.

Description

Technical Field

[0001] This invention relates to the field of production safety analysis technology, and in particular to an intelligent factory safety management method and system. Background Technology

[0002] With the continuous advancement of industrial automation and intelligent manufacturing, smart factories are gradually becoming a key direction for the transformation and upgrading of modern manufacturing. Against this backdrop, safety management during factory operations is receiving increasing attention. Especially in scenarios involving high-risk processes or complex operating conditions of critical equipment, how to achieve timely detection and effective response to potential safety hazards has become a crucial issue that urgently needs to be addressed in the construction of smart factories.

[0003] Some factory safety management systems use methods based on fixed threshold settings or static model analysis to monitor equipment operating status and issue real-time alarms for key parameters. While these methods are effective in addressing some known anomalies, they have limitations in practical applications. For example, relying solely on single-point indicator triggers or static rule-based judgments may fail to capture early, subtle signs of equipment status evolution in a timely manner, leading to delayed risk identification.

[0004] In fact, many production safety incidents often undergo a latent evolutionary process before they occur, during which equipment operating parameters may exhibit a certain non-linear trend. Without in-depth analysis of these subtle changes, it may be difficult to reveal the potential development path of abnormal states in a timely manner, thus affecting the foresight and accuracy of early warnings. Therefore, it is urgent to introduce more process-understanding analytical mechanisms into factory safety management to support the continuous perception and trend identification of changes in equipment operating status.

[0005] Therefore, in the context of smart factories, how to make full use of a large amount of historical monitoring data, explore the potential evolutionary patterns before safety incidents occur, construct a knowledge structure that reflects the dynamic evolution characteristics of equipment, and effectively use it for intelligent comparison and trend prediction of real-time data has become one of the technical problems that urgently need to be solved in the field of factory safety management. Summary of the Invention

[0006] To address the aforementioned technical problems, this invention proposes an intelligent factory safety management method and system. Based on in-depth analysis of a large amount of historical monitoring data, it captures characteristic information reflecting the potential abnormal evolution trend of equipment, thereby achieving intelligent safety analysis and early warning of real-time data.

[0007] The first aspect of this invention provides an intelligent factory safety management method, comprising:

[0008] Collect historical production safety record data of target production equipment in the factory regarding multiple production safety events, and extract the historical equipment status data corresponding to each production safety event from the historical production safety record data;

[0009] For each set of historical equipment status data, parameter evolution anomaly detection is performed to identify multiple abnormal evolution units in the historical equipment status data, and the evolution feature vector of each abnormal evolution unit is constructed based on the historical equipment status data.

[0010] By combining historical equipment status data and evolution feature vectors, evolutionary correlation analysis is performed on multiple anomalous evolutionary units, and the evolutionary correlation parameters between any two anomalous evolutionary units are calculated.

[0011] Based on the evolutionary correlation parameters, an evolutionary chain map of multiple anomalous evolutionary units is constructed. Multiple state evolution links are extracted from the evolutionary chain map. Link activity analysis is performed on multiple state evolution links to determine multiple target evolution links.

[0012] After acquiring real-time production safety monitoring data of the target production equipment, the production status safety analysis is performed on the real-time production safety monitoring data based on multiple target evolution links to generate production safety detection results for the target production equipment.

[0013] Preferably, the evolutionary feature vector of each abnormal evolutionary unit is constructed based on historical equipment status data, including:

[0014] For abnormal evolution units, by performing sliding window processing on historical equipment state data, multiple evolution characteristic parameters of historical equipment state data within each sliding window are extracted, including drift characteristic parameters, energy characteristic parameters and state characteristic parameters;

[0015] Based on multiple evolutionary characteristic parameters, parameter evolution anomaly detection is performed on historical equipment status data, and multiple abnormal evolutionary units in historical equipment status data are determined from multiple sliding windows;

[0016] Multiple parameter state difference features of each abnormal evolution unit are extracted from historical equipment status data. Combined with multiple evolution characteristic parameters of the abnormal evolution unit, an evolution feature vector of each abnormal evolution unit is constructed.

[0017] Preferably, evolutionary correlation analysis is performed on multiple anomalous evolutionary units by combining historical equipment status data and evolutionary feature vectors to calculate the evolutionary correlation parameters between any two anomalous evolutionary units, including:

[0018] Determine the timestamp data of each abnormal evolution unit, locate the evolution trajectory window of each abnormal evolution unit based on the timestamp data of the abnormal evolution unit, and extract the local time-series data of each abnormal evolution unit from the historical equipment status data based on the evolution trajectory window;

[0019] The state correlation parameters between any two abnormal evolution units are calculated based on the evolution feature vector of the abnormal evolution units. The parameter evolution trend vector of the abnormal evolution units is extracted from the local time series data. The trend matching parameters between any two abnormal evolution units are calculated based on the parameter evolution trend vector. The abnormal dominant parameters of each abnormal evolution unit are determined based on the local time series data. Abnormal coupling matching is performed on any two abnormal evolution units to generate abnormal coupling parameters between any two abnormal evolution units. Evolution correlation analysis is performed on any two abnormal evolution units based on the timestamp data of the abnormal evolution units, including time nearest neighbor fusion of the state correlation parameters, trend matching parameters and abnormal coupling parameters between the two abnormal evolution units. The evolution correlation parameters between any two abnormal evolution units under each production safety event are calculated.

[0020] Preferably, multiple state evolution links are extracted from the evolutionary chain map, and link activity analysis is performed on these multiple state evolution links to determine multiple target evolution links, including:

[0021] Based on the temporal relationship, multiple abnormal evolutionary units in the evolutionary chain graph are traversed. Temporal nearest neighbor test and evolutionary association test are performed on two temporally adjacent abnormal evolutionary units. If the temporal nearest neighbor test and evolutionary association test pass, the two abnormal evolutionary units are chained together. The evolutionary chain graph contains multiple abnormal evolutionary units under one set of historical device state data. The edge between any two abnormal evolutionary units is constructed according to the evolutionary association parameters between the two abnormal evolutionary units. By traversing the evolutionary chain graph, multiple state evolution links corresponding to each set of historical device state data are extracted.

[0022] The structural length index of each state evolution link is extracted. Based on the abnormal dominant parameters of the abnormal evolution unit, the link evolution trend analysis is performed on the state evolution link. This includes statistically analyzing the dominant frequency of each device state parameter in the state evolution link, generating the evolution trend index of the state evolution link based on the dominant frequency of each device state parameter, calculating the link activity parameter of the state evolution link based on the structural length index and the evolution trend index, and selecting multiple target evolution links corresponding to each group of historical device state data based on the link activity parameter.

[0023] Preferably, production status safety analysis is performed on real-time production safety monitoring data based on multiple target evolution links, including:

[0024] Multiple real-time evolution links are extracted from real-time production safety monitoring data. Each real-time evolution link is matched with multiple target evolution links to obtain the link matching parameters between the real-time evolution link and each target evolution link.

[0025] Based on multiple link matching parameters, the trend correlation data between real-time production safety monitoring data and multiple production safety events is determined. Based on the trend correlation data, the production status safety analysis of real-time production safety monitoring data is performed to obtain the production safety detection results of the target production equipment.

[0026] Preferably, trend correlation data between real-time production safety monitoring data and multiple production safety events is determined based on multiple link matching parameters, including:

[0027] Based on multiple real-time evolution links of real-time production safety monitoring data, a link matching window is determined. By traversing multiple target evolution links under each production safety event through the link matching window, local matching parameters between real-time production safety monitoring data and production safety events under each link matching window are calculated. Based on multiple local matching parameters, global matching parameters between real-time production safety monitoring data and production safety events are determined, and trend correlation data between real-time production safety monitoring data and multiple production safety events is obtained.

[0028] A second aspect of the present invention provides an intelligent factory safety management system for implementing the above-described intelligent factory safety management method, comprising:

[0029] The data acquisition module is used to collect historical production safety record data of target production equipment in the factory regarding multiple production safety events, and extract the historical equipment status data corresponding to each production safety event from the historical production safety record data;

[0030] The evolution anomaly analysis module is used to detect parameter evolution anomalies in each group of historical equipment status data, identify multiple abnormal evolution units in the historical equipment status data, and construct the evolution feature vector of each abnormal evolution unit based on the historical equipment status data.

[0031] The evolution correlation module is used to perform evolution correlation analysis on multiple abnormal evolution units by combining historical equipment status data and evolution feature vectors, and to calculate the evolution correlation parameters between any two abnormal evolution units.

[0032] The evolution link generation module is used to construct an evolution chain map of multiple anomalous evolutionary units based on evolutionary association parameters, extract multiple state evolution links from the evolution chain map, and perform link activity analysis on multiple state evolution links to determine multiple target evolution links.

[0033] The production safety detection module is used to perform production status safety analysis on the real-time production safety monitoring data of the target production equipment after acquiring the real-time production safety monitoring data, based on multiple target evolution links, and generate production safety detection results for the target production equipment.

[0034] The present invention has the following beneficial effects:

[0035] This invention collects and analyzes historical data on various production safety events from production equipment, extracts multiple anomaly evolution units from the perspective of micro-anomalies, and analyzes the characteristics of state drift, energy, and state mutation of different anomaly evolution units to construct evolutionary feature vectors for the anomaly evolution units. Combining timestamp positioning and local time-series segment analysis, it calculates the evolutionary correlation between different evolution units, constructs a chain graph reflecting the dynamic correlation of anomaly evolution, extracts active evolutionary links based on temporal nearest neighbor and correlation tests, and generates target evolutionary paths by quantifying link structure length and parameter dominance frequency. During the real-time monitoring phase, it dynamically matches real-time evolutionary links with historical target links, and combines local window matching and global trend correlation analysis to accurately identify the potential correlation between equipment status and safety hazard events. This enables the analysis of evolutionary patterns from implicit parameter drift to explicit safety events. By integrating historical evolution patterns with real-time data feature matching, it provides factory production equipment with multi-stage early warning capabilities covering the anomaly bud stage to the pre-outbreak stage, significantly improving the timeliness and accuracy of safety hazard identification. Attached Figure Description

[0036] Figure 1 This is a flowchart illustrating an intelligent factory safety management method provided by the present invention.

[0037] Figure 2 This is a schematic diagram of the structure of an intelligent factory safety management system provided by the present invention. Detailed Implementation

[0038] To enable those skilled in the art to better understand the technical solutions of this invention, the technical solutions of the embodiments of this invention will be clearly and completely described below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are only for explaining the invention and are not intended to limit the invention.

[0039] Please see Figure 1 The present invention provides an intelligent factory safety management method, comprising steps S1 to S5. Specifically:

[0040] Step S1: Collect historical production safety record data of the target production equipment in the factory regarding multiple production safety events, and extract the historical equipment status data corresponding to each production safety event from the historical production safety record data.

[0041] Understandably, the operational status and safety events of production equipment in a factory are crucial components of safety monitoring. The collected monitoring records of multiple historical production safety events for target production equipment—which can be processing equipment such as chemical reactors and heat treatment furnaces, or operating equipment such as industrial motors and compressors—have a critical impact on the safety and stability of the entire production line. During daily factory operations, these devices may trigger production safety events due to factors such as load changes, sensor drift, and the accumulation of localized faults. Examples include automatic meltdowns caused by excessive temperature or bearing damage due to abnormal vibration, which can lead to production interruptions or other safety accidents in severe cases. Some historical production safety events, such as those caused by gradual anomalies like abnormal motor current fluctuations leading to overloads, are often preceded by subtle fluctuations or abnormal trends in equipment status parameters, changes that may be difficult to identify in a timely manner using traditional monitoring methods.

[0042] To identify potential evolutionary patterns before a safety incident occurs, this embodiment can acquire equipment status-related data from multiple historical production safety events associated with the target production equipment, based on data sources such as equipment monitoring systems, production logs, and alarm records. Historical equipment status data can include a continuous sequence of equipment status parameters prior to the event, such as monitoring data corresponding to equipment status parameters like temperature, pressure, current, voltage, vibration frequency, rotational speed, and running time. This data serves as the basis for subsequent evolutionary modeling, reflecting the dynamic characteristics of the equipment during fault evolution or anomaly formation.

[0043] Step S2: Perform parameter evolution anomaly detection on each group of historical equipment status data, identify multiple abnormal evolution units in the historical equipment status data, and construct the evolution feature vector of each abnormal evolution unit based on the historical equipment status data.

[0044] Understandably, considering that the operating status of production equipment may gradually change over time, evolutionary anomaly detection is used to capture potential risks in equipment status changes. For example, equipment may experience minor, gradual failures due to component aging or long-term operation. Such failures are easily overlooked in traditional threshold monitoring strategies and are considered normal phenomena, leading to an inability to accurately analyze and capture them in the early stages of safety events. In this embodiment, parameter evolutionary anomaly detection is performed on each set of historical equipment status data to locate some minor anomalies in the early stages of safety events, identify multiple anomaly evolution units, and further analyze the evolutionary status of each anomaly evolution unit to construct an evolutionary feature vector that can characterize the evolutionary characteristics of equipment status parameters in a time series.

[0045] In some implementation processes, the construction of abnormal evolution units involves extracting multiple evolution characteristic parameters of historical equipment status data within each sliding window by performing sliding window processing on historical equipment status data.

[0046] Specifically, in this embodiment, the evolution characteristics of data within different windows are comprehensively analyzed from three dimensions: drift trend, change conformity, and state mutation. Multiple evolution characteristic parameters within each sliding window are calculated, including drift characteristic parameters, energy characteristic parameters, and state characteristic parameters.

[0047] The drift trend is characterized by changes in the mean values ​​of parameters within a window. For example, the window is divided into a first half and a second half, and a multidimensional mean vector is constructed from the mean values ​​of various device state parameters within each window based on historical device state data. The distance between the mean vectors of the first and second half windows is calculated to obtain the drift characteristic parameter, which is used to analyze whether the overall mean value of the parameters within the window has a trend of drift. The magnitude of change is characterized by the covariance volume of parameters within the window. A parameter matrix is ​​constructed by using the values ​​of different device state parameters at multiple time steps within the sliding window, and the corresponding covariance matrix is ​​calculated. The trace of the covariance matrix is ​​used as an energy characteristic parameter to characterize the overall fluctuation energy, reflecting the overall magnitude of changes in multivariate parameters. Sudden jumps in the data are described by the instantaneous offset of parameters. Specifically, the distance between device states at any two adjacent time steps is calculated, for example, by using Euclidean distance. The maximum value of multiple distances within the window is then used as a state characteristic parameter to analyze sudden anomalies within the window.

[0048] Then, parameter evolution anomaly detection is performed on historical equipment status data based on multiple evolution characteristic parameters. Specifically, the analysis is conducted from three dimensions: drift trend, change conformity, and state change. By using the thresholds corresponding to the three pre-set dimensions, if at least one of the drift characteristic parameters, energy characteristic parameters, and status characteristic parameters within the detection window triggers the threshold, the sliding window is marked as an abnormal evolution unit. This indicates that a minor anomaly has occurred in the equipment status within the window period, which can be regarded as a micro-event point, representing a minor anomaly that existed in the early stage of a safety event.

[0049] This method identifies multiple anomalous evolutionary units from historical equipment state data through multiple sliding windows. Furthermore, it extracts multiple parameter state difference features for each anomalous evolutionary unit from the historical equipment state data. These features include the difference values ​​of each equipment state parameter within the sliding window, such as the parameter values ​​at the end and beginning of the window. Finally, the multiple parameter state difference features are combined with multiple evolutionary characteristic parameters of the anomalous evolutionary unit to construct an evolutionary feature vector for each anomalous evolutionary unit, representing the characteristics of equipment state changes during different anomalous evolution processes.

[0050] Step S3: Combine historical equipment status data and evolution feature vectors to perform evolutionary correlation analysis on multiple abnormal evolutionary units, and calculate the evolutionary correlation parameters between any two abnormal evolutionary units.

[0051] Understandably, the significance of performing evolutionary correlation analysis on abnormal evolutionary units lies in analyzing the correlation between different minor anomalies before a safety event occurs. For example, in the early stages of a safety event, equipment temperature may be slightly higher than normal, followed by abnormal vibration, ultimately leading to damage to equipment components. By mining the evolutionary relationships of equipment states before a major failure occurs, and quantitatively calculating the evolutionary correlation parameters between any two abnormal evolutionary units, the analysis can be conducted.

[0052] In some implementation processes, the calculation of evolutionary correlation parameters between anomalous evolutionary units specifically includes:

[0053] Determine the timestamp data for each abnormal evolution unit, locate the evolution trajectory window of each abnormal evolution unit based on the timestamp data, and extract the local time-series data of each abnormal evolution unit from the historical equipment status data based on the evolution trajectory window.

[0054] Specifically, time-series localization is performed based on the timestamp data of the abnormal evolution units to determine the evolution trajectory window of each abnormal evolution unit. For example, the time range within 2 minutes before and after the center of the abnormal evolution unit is recorded as the evolution trajectory window, so as to fully capture the behavioral transition area before and after the abnormal evolution. The data in the evolution trajectory window in the historical equipment status data is used as the local time-series segment data of the abnormal evolution unit to extract local context information from the original data.

[0055] This study evaluates the evolutionary correlation characteristics between two anomalous evolutionary units from two perspectives: evolutionary trend and anomaly coupling characteristics, combined with contextual information. In this process, from the perspective of evolutionary trend, parameter evolutionary trend vectors of the anomalous evolutionary units are extracted from local time-series data, and trend matching parameters between any two anomalous evolutionary units are calculated based on these vectors. The parameter evolutionary trend vectors characterize the changing trends of different device state parameters in the local time-series data; for example, trends such as growth, stability, and decline are quantified using parameters such as 1, 0, and -1, respectively. Then, the trend matching between the parameter evolutionary trend vectors of two anomalous evolutionary units is analyzed; for example, the angle between the parameter evolutionary trend vectors is calculated as the trend matching parameter between the two anomalous evolutionary units. From the perspective of abnormal coupling characteristics, the dominant abnormal parameter of each abnormal evolution unit is first determined based on the local time-series data, that is, the device state parameter with the largest deviation in the local time-series data. Then, abnormal coupling matching is performed on any two abnormal evolution units, including determining whether they share a dominant abnormal channel, such as whether they are both dominated by temperature, that is, whether the temperature deviation is the largest in both abnormal evolution units. If they meet the criteria, it is recorded as 1; otherwise, it is recorded as 0. In this way, the abnormal coupling parameters between any two abnormal evolution units are generated.

[0056] Furthermore, by combining the characteristics of the anomalous evolutionary unit itself, namely the evolutionary feature vector of the anomalous evolutionary unit, the overall correlation between the evolutionary feature vectors of any two anomalous evolutionary units can be analyzed and calculated. For example, distance measurement can be performed to calculate the state correlation parameters between any two anomalous evolutionary units.

[0057] Finally, based on the timestamp data of the abnormal evolution units, an evolutionary correlation analysis is performed on any two abnormal evolution units to calculate the evolutionary correlation parameters between any two abnormal evolution units under each production safety event.

[0058] Specifically, the state association parameters, trend matching parameters, and anomaly coupling parameters between two anomalous evolution units are fused using time nearest neighbor fusion. This includes weighting the state association parameters, trend matching parameters, and anomaly coupling parameters according to pre-set fusion weights, and then applying a time decay function. Make corrections, among which The time difference between two anomalous evolutionary units is calculated based on the timestamp data of the anomalous evolutionary units. This is used to calculate the evolutionary correlation parameters between the two anomalous evolutionary units, thereby measuring the evolutionary correlation between them.

[0059] Step S4: Construct an evolutionary chain map of multiple anomalous evolutionary units based on evolutionary correlation parameters, extract multiple state evolutionary links from the evolutionary chain map, and perform link activity analysis on multiple state evolutionary links to determine multiple target evolutionary links.

[0060] Understandably, for historical production safety records of different production safety events, corresponding evolutionary chain graphs can be constructed. These graphs contain multiple anomalous evolutionary units under historical equipment state data as nodes, and the edges between any two anomalous evolutionary units are constructed based on the evolutionary correlation parameters between them. By considering the temporal correlation and evolutionary property correlation between different anomalous evolutionary units, multiple state evolutionary links are extracted. Through activity analysis of different state evolutionary links, several representative target evolutionary links in each evolutionary chain graph are finally selected to characterize the potential evolutionary characteristics of different parameters of the production equipment before the safety event occurs.

[0061] In some implementation processes, multiple state evolution links are extracted from the evolutionary chain graph, and link activity analysis is performed on these links to identify multiple target evolution links, specifically including:

[0062] Based on temporal relationships, multiple anomalous evolutionary units in the evolutionary chain graph are traversed. For temporally adjacent anomalous evolutionary units, temporal nearest neighbor and evolutionary association checks are performed. Specifically, the temporal nearest neighbor check verifies whether the time difference between two anomalous evolutionary units is less than a preset temporal nearest neighbor threshold, such as 5 minutes. If it is less, the temporal nearest neighbor check passes, indicating a strong temporal association between the anomalous evolutionary units. The evolutionary association check verifies whether the evolutionary association parameter between two anomalous evolutionary units is less than a preset evolutionary association threshold. If it is greater, the evolutionary association check passes, indicating a strong evolutionary association between the anomalous evolutionary units. If both the temporal nearest neighbor and evolutionary association checks pass, the two anomalous evolutionary units are chained together. By traversing the evolutionary chain graph, multiple state evolution links corresponding to each set of historical device state data are finally extracted. Each link represents a state change process with a continuous evolutionary trend, composed of multiple anomalous evolutionary units.

[0063] The structural length index of each state evolution link is extracted. Based on the abnormal dominant parameters of the abnormal evolution unit, the link evolution trend analysis is performed on the state evolution link to generate the evolution trend index of the state evolution link. The link activity parameter of the state evolution link is calculated based on the structural length index and the evolution trend index.

[0064] Specifically, the structural length index of the state evolution link refers to the number of anomalous evolutionary units it contains. The process of analyzing the evolution trend of the state evolution link includes statistically analyzing the dominance frequency of each device state parameter within the link. Based on the dominance frequency of each device state parameter, an evolution trend index is calculated, specifically the percentage of the dominance frequency of the device state parameter with the highest dominance frequency. Then, based on the structural length index and the evolution trend index, the ratio between the evolution trend index and the structural length index is recorded as the link activity parameter. A smaller structural length index with a larger evolution trend index indicates a stronger overall trend and greater representativeness throughout the link. Finally, multiple target evolution links corresponding to each set of historical device state data are selected based on the link activity parameter.

[0065] Step S5: After acquiring the real-time production safety monitoring data of the target production equipment, perform production status safety analysis on the real-time production safety monitoring data based on multiple target evolution links, and generate the production safety detection results of the target production equipment.

[0066] Understandably, real-time production safety monitoring data includes parameter monitoring data related to the real-time operating status of the target production equipment. Using the target evolution link constructed in the preceding steps, the real-time collected production data is analyzed to conduct a safety assessment of the production status. By matching the data with the evolution link to identify any abnormal patterns in the equipment's state evolution, the system assesses whether the equipment's current state is on a dangerous link, thus predicting potential faults in advance and obtaining production safety detection results for the target production equipment. These results may indicate whether the equipment is safe, whether there are any hidden faults, or provide early warnings for high-probability faults.

[0067] In some implementation processes, production status safety analysis is performed on real-time production safety monitoring data based on multiple target evolution links, specifically including:

[0068] Multiple real-time evolutionary links are extracted from real-time production safety monitoring data. Specifically, the process is similar to the aforementioned state evolutionary link construction process: multiple minor abnormal evolutionary events existing in the real-time production safety monitoring data are identified and linked together to form evolutionary links. Each real-time evolutionary link is matched with multiple target evolutionary links to obtain link matching parameters between the real-time evolutionary link and each target evolutionary link. During the link matching process, a feature matrix for each evolutionary link is constructed based on the evolutionary feature vectors of different abnormal evolutionary units within the link. Distance analysis is then performed on the feature matrix to calculate the link matching parameters between the real-time evolutionary link and each target evolutionary link. Considering the potential differences in the length of different evolutionary links, a dynamic time warping algorithm is used for analysis. For example, the feature matrix is ​​expanded, and a sequence is constructed based on the temporal relationship of the abnormal evolutionary units, thereby calculating the link matching parameters between evolutionary links in scenarios with inconsistent link lengths.

[0069] Then, based on multiple link matching parameters, the trend correlation data between real-time production safety monitoring data and multiple production safety events is determined. Based on the trend correlation data, the production status safety analysis of real-time production safety monitoring data is performed to obtain the production safety detection results of the target production equipment.

[0070] Specifically, the generation process for the aforementioned trend-related data includes: determining a link matching window based on multiple real-time evolution links of the real-time production safety monitoring data (specifically, the number of evolution links included); traversing multiple target evolution links under each production safety event through the link matching window; calculating the local matching parameters between the real-time production safety monitoring data and the production safety event under each link matching window (i.e., the sum of the link matching parameters between the multiple target evolution links under the link matching window and each real-time evolution link); and then determining the global matching parameters between the real-time production safety monitoring data and the production safety event based on the multiple local matching parameters. In this embodiment, the maximum value of the local matching parameters is recorded as the global matching parameter between the real-time production safety monitoring data and the production safety event. The global matching parameter characterizes the matching situation between the real-time production safety monitoring data and different production safety events, thereby obtaining trend-related data between the real-time production safety monitoring data and multiple production safety events. This data characterizes the current equipment state evolution trend exhibited by the real-time production safety monitoring data and its matching situation with different safety production events, thereby enabling early warning of potential safety issues that may arise in production equipment.

[0071] Please see Figure 2 The present invention also provides an intelligent factory safety management system for implementing the above-mentioned intelligent factory safety management method, comprising:

[0072] The data acquisition module is used to collect historical production safety record data of target production equipment in the factory regarding multiple production safety events, and extract the historical equipment status data corresponding to each production safety event from the historical production safety record data;

[0073] The evolution anomaly analysis module is used to detect parameter evolution anomalies in each group of historical equipment status data, identify multiple abnormal evolution units in the historical equipment status data, and construct the evolution feature vector of each abnormal evolution unit based on the historical equipment status data.

[0074] The evolution correlation module is used to perform evolution correlation analysis on multiple abnormal evolution units by combining historical equipment status data and evolution feature vectors, and to calculate the evolution correlation parameters between any two abnormal evolution units.

[0075] The evolution link generation module is used to construct an evolution chain map of multiple anomalous evolutionary units based on evolutionary association parameters, extract multiple state evolution links from the evolution chain map, and perform link activity analysis on multiple state evolution links to determine multiple target evolution links.

[0076] The production safety detection module is used to perform production status safety analysis on the real-time production safety monitoring data of the target production equipment after acquiring the real-time production safety monitoring data, based on multiple target evolution links, and generate production safety detection results for the target production equipment.

[0077] The above are merely specific embodiments of the present invention, enabling those skilled in the art to understand or implement the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art. Parts not described in detail in this specification are prior art known to those skilled in the art.

Claims

1. An intelligent plant safety management method, characterized by, Comprise: Collecting historical production safety record data of target production equipment in a factory about multiple production safety events, extracting historical equipment state data corresponding to each production safety event from the historical production safety record data; Performing parameter evolution anomaly detection on each set of historical equipment state data to determine multiple abnormal evolution units in the historical equipment state data, and constructing an evolution feature vector of each abnormal evolution unit based on the historical equipment state data, wherein for the abnormal evolution unit, multiple evolution characteristic parameters of the historical equipment state data in each sliding window are extracted by performing sliding window processing on the historical equipment state data, including drift characteristic parameters, energy characteristic parameters, and state characteristic parameters; Wherein the drift trend is represented by the change of the mean value of the parameters in the window, the window is divided into a first half window and a second half window, and a multi-dimensional mean vector composed of the mean values of multiple equipment state parameters of the historical equipment state data in the window is extracted, the distance between the mean vectors of the first half window and the second half window is calculated to obtain the drift characteristic parameters, which are used to analyze whether the overall mean value of the parameters in the window has a trend drift; the change amplitude characteristic is represented by the covariance volume of the parameters in the window, which can be constructed by the values of different equipment state parameters at multiple time steps in the sliding window, and the covariance matrix corresponding to the parameter matrix is calculated, and the trace of the covariance matrix is calculated as the energy characteristic parameter to represent the overall fluctuation energy and reflect the overall amplitude of the multivariate parameter change; the instantaneous offset of the parameters is used to describe the sudden jump of the data, which is specifically calculated by the distance between any two adjacent time steps of the equipment state, which is measured by the Euclidean distance, and then the maximum value of multiple distances in the window is calculated as the state characteristic parameter to analyze the sudden abnormal phenomenon in the window; Performing parameter evolution anomaly detection on the historical equipment state data according to the multiple evolution characteristic parameters to determine multiple abnormal evolution units in the historical equipment state data from multiple sliding windows; Extracting multiple parameter state difference features of each abnormal evolution unit from the historical equipment state data, and constructing an evolution feature vector of each abnormal evolution unit by combining the multiple evolution characteristic parameters of the abnormal evolution unit; Performing evolution correlation analysis on the multiple abnormal evolution units based on the historical equipment state data and the evolution feature vector, and calculating the evolution correlation parameters between any two abnormal evolution units, including determining the timestamp data of each abnormal evolution unit, locating the evolution trajectory window of each abnormal evolution unit according to the timestamp data of the abnormal evolution unit, and extracting local time sequence segment data of each abnormal evolution unit from the historical equipment state data according to the evolution trajectory window; The state correlation parameter between any two abnormal evolution units is calculated according to the evolution feature vector of the abnormal evolution unit, the parameter evolution trend vector of the abnormal evolution unit is extracted from the local time sequence fragment data, and the trend matching parameter between any two abnormal evolution units is calculated according to the parameter evolution trend vector, wherein the parameter evolution trend vector is used to represent the change trend of different equipment state parameters in the local time sequence fragment data, and the growth, stable and decline trends are quantitatively represented by 1, 0 and -1 parameters respectively, and then the trend matching between the parameter evolution trend vectors of the two abnormal evolution units is analyzed, the included angle between the parameter evolution trend vectors is calculated as the trend matching parameter between the two abnormal evolution units, the abnormal dominant parameter of each abnormal evolution unit is determined according to the local time sequence fragment data, the abnormal coupling matching of any two abnormal evolution units is performed, the abnormal coupling parameter between any two abnormal evolution units is generated, the evolution correlation analysis of any two abnormal evolution units is performed according to the timestamp data of the abnormal evolution unit, including the time near neighbor fusion of the state correlation parameter, the trend matching parameter and the abnormal coupling parameter between the two abnormal evolution units, including the state correlation parameter, the trend matching parameter and the abnormal coupling parameter, after weighted fusion according to the pre-set fusion weight, the time decay function is used for correction, wherein is the time difference between the two abnormal evolution units, which is calculated according to the timestamp data of the abnormal evolution unit, and the evolution correlation parameter between any two abnormal evolution units under each production safety event is calculated to measure the evolution correlation relationship between the two abnormal evolution units. The evolution chain graph of the plurality of abnormal evolution units is constructed according to the evolution correlation parameter, and a plurality of state evolution links are extracted from the evolution chain graph; the link activity analysis is performed on the plurality of state evolution links to determine a plurality of target evolution links, including traversing the plurality of abnormal evolution units in the evolution chain graph based on the time sequence relationship, performing time sequence neighbor inspection and evolution correlation inspection on two abnormal evolution units adjacent in time sequence, and combining the two abnormal evolution units in a chain if the time sequence neighbor inspection and the evolution correlation inspection pass; the evolution chain graph comprises a plurality of abnormal evolution units under one set of historical equipment state data, and an edge between any two abnormal evolution units is constructed according to the evolution correlation parameter between the two abnormal evolution units; and the plurality of state evolution links corresponding to each set of historical equipment state data are extracted by traversing the evolution chain graph. The structure length index of each state evolution link is extracted, the link evolution trend analysis is performed on the state evolution link according to the abnormal dominant parameter of the abnormal evolution unit, including counting the dominant frequency of each device state parameter in the state evolution link, generating the evolution trend index of the state evolution link according to the dominant frequency of each device state parameter, calculating the link activity parameter of the state evolution link based on the structure length index and the evolution trend index, and screening the plurality of target evolution links corresponding to each set of historical equipment state data according to the link activity parameter. After obtaining the real-time production safety monitoring data of the target production equipment, the production state safety analysis is performed on the real-time production safety monitoring data according to the plurality of target evolution links, including extracting a plurality of real-time evolution links from the real-time production safety monitoring data, performing link matching between each real-time evolution link and the plurality of target evolution links, and obtaining the link matching parameter between the real-time evolution link and each target evolution link. The trend correlation data between the real-time production safety monitoring data and the plurality of production safety events is determined according to the plurality of link matching parameters, the production state safety analysis is performed on the real-time production safety monitoring data according to the trend correlation data, and the production safety detection result of the target production equipment is generated.

2. The intelligent plant safety management method of claim 1, wherein, The trend correlation data between the real-time production safety monitoring data and the plurality of production safety events is determined according to the plurality of link matching parameters, including: The link matching window is determined according to the plurality of real-time evolution links of the real-time production safety monitoring data, the plurality of target evolution links under each production safety event are traversed through the link matching window, the local matching parameter between the real-time production safety monitoring data and the production safety event under each link matching window is calculated, the global matching parameter between the real-time production safety monitoring data and the production safety event is determined according to the plurality of local matching parameters, and the trend correlation data between the real-time production safety monitoring data and the plurality of production safety events is obtained.

3. An intelligent plant safety management system, characterized by, The system is used to implement the intelligent factory safety management method of any one of claims 1-2, including: A data acquisition module is configured to acquire historical production safety record data of target production equipment in a factory with respect to a plurality of production safety events, and extract historical equipment state data corresponding to each production safety event from the historical production safety record data. an evolution anomaly analysis module, configured to perform parameter evolution anomaly detection on each set of historical equipment state data, determine a plurality of abnormal evolution units in the historical equipment state data, and construct an evolution feature vector of each abnormal evolution unit based on the historical equipment state data; an evolution correlation module, configured to perform evolution correlation analysis on the plurality of abnormal evolution units in combination with the historical equipment state data and the evolution feature vector, and calculate evolution correlation parameters between any two abnormal evolution units; an evolution link generation module, configured to construct an evolution chain graph of the plurality of abnormal evolution units according to the evolution correlation parameters, extract a plurality of state evolution links from the evolution chain graph, and perform link active analysis on the plurality of state evolution links to determine a plurality of target evolution links; a production safety detection module, configured to, after obtaining real-time production safety monitoring data of a target production equipment, perform production state safety analysis on the real-time production safety monitoring data according to the plurality of target evolution links, and generate a production safety detection result of the target production equipment.

Images