An industrial field multi-class risk linkage alarm disposal method and system

Abstract

The application provides an industrial field multi-class risk linkage alarm handling method and system, the method comprises the following steps: converting industrial multi-dimensional time sequence data into discrete risk events, relying on event level configuration time decay rate, fusing basic weight and decay factor to obtain event comprehensive weight, screening high weight events to form frequent item set; relying on candidate item set iteration mechanism to generate high-order candidate item set, calculating space-time fusion degree according to event space-time proximity, matching evolution template to obtain time sequence consistency score; combining comprehensive weight, time sequence consistency score and linkage bias coefficient to solve linkage support degree, comparing the linkage support degree with alarm threshold, and screening to obtain all levels of frequent item set; iteratively identifying all risk linkage rules, generating corresponding alarm handling strategies relying on the obtained linkage relationship, which can identify industrial multi-risk chain conduction rules, reduce false alarm probability, and improve the real-time performance and matching degree of field safety handling.

Description

Technical Field

[0001] This application pertains to the field of industrial sites, and in particular relates to a method and system for handling multi-risk linkage alarms in industrial sites. Background Technology

[0002] With the continuous expansion of modern industry and the increasing level of intelligence, industrial site safety management and risk prevention face challenges. Industrial sites typically deploy massive numbers of sensors and monitoring devices, generating multi-dimensional time-series data. This data contains potential risk information that could lead to safety accidents, including equipment operating status, environmental changes, and production processes. In actual production environments, most risk events do not occur in isolation; they exhibit strong correlation and chain reaction characteristics. A local anomaly or minor risk can spread over time and space, evolving and triggering other types of risks. Traditional industrial site alarm mechanisms are mostly based on single-dimensional, fixed threshold judgments, primarily handling independent risk events in a discrete manner. This isolated alarm model struggles to discern the inherent correlations between different risk events behind massive amounts of data, and it is difficult to identify the linkage effects between multiple types of risks. This leaves safety management personnel in a passive response state, hindering proactive risk prevention and comprehensive coordinated control.

[0003] Most existing analytical methods fail to recognize the attenuation of risk event severity over time, making it difficult to scientifically calculate the impact weights of different types of risks. This results in crucial risk clues being easily buried in massive amounts of redundant data. Secondly, risk linkages in industrial settings heavily rely on the close integration of time and space, and risk outbreaks typically follow specific temporal evolutionary logic. However, existing methods struggle to integrate spatiotemporal proximity with temporal evolution consistency, making it difficult to identify the degree of linkage between diverse risk combinations. Furthermore, due to the significant differences in triggering conditions for risk linkages across different industrial scenarios, using static, fixed alarm thresholds leads to numerous false alarms or missed alarms, and it is difficult to adjust them according to the spatiotemporal characteristics of real-world environments. This limits the accuracy of assessing multi-type risk linkage relationships, making it difficult for the system to extract truly reliable linkage rules, resulting in inadequate scientific risk linkage alarm handling strategies for dynamic industrial settings. Summary of the Invention

[0004] According to one aspect of the present invention, a method for handling multi-risk linkage alarms in industrial sites is proposed to address the problem that existing methods cannot integrate spatiotemporal proximity with temporal evolution consistency and cannot identify the degree of linkage between variable risk combinations. The method includes the following steps: Acquire multidimensional time-series data from industrial sites, convert it into a discrete set of risk events, and configure basic risk weights for each risk event type; for any risk event instance, adjust the basic time decay rate according to the severity level; calculate the time decay factor based on the interval between the occurrence time and the current time; merge the basic risk weights and the time decay factor to obtain a comprehensive weight; filter risk event instances whose comprehensive weights exceed a first preset threshold to form a frequent one set; Candidate k-th itemsets are generated based on frequent item set iterative combinations, where k is an integer not less than two; spatial proximity and temporal proximity are calculated for any two risk event combinations within the candidate k-th itemset, and the spatial-temporal fusion degree is obtained by aggregation; temporal consistency scores are obtained by matching the occurrence order of each risk event instance with a pre-set evolution sequence template. The linkage bias coefficient and basic linkage alarm threshold are obtained from the risk linkage matrix; the basic linkage alarm threshold is adjusted according to the spatial-temporal fusion degree to generate an alarm threshold; the linkage support is calculated by combining the comprehensive weight, temporal consistency score and linkage bias coefficient; when the linkage support is greater than the alarm threshold, the candidate k-th itemset is determined as the frequent k-th itemset; the value of k is incremented and the candidate itemset generation, fusion degree calculation, temporal matching, threshold adjustment, support calculation and frequent itemset determination are repeated until the preset maximum itemset length is reached or no new candidate itemsets are generated; the obtained frequent itemsets are set as linkage rules to generate a linkage alarm handling strategy.

[0005] According to another aspect of the present invention, an industrial site multi-risk linkage alarm and handling system is proposed, comprising the following modules: The adjustment module is used to acquire multi-dimensional time-series data from industrial sites, convert it into a set of discrete risk events, and configure basic risk weights for each risk event type; for any risk event instance, the basic time decay rate is adjusted according to the severity level; the time decay factor is calculated based on the interval between the occurrence time and the current time; the basic risk weights and the time decay factor are merged to obtain a comprehensive weight; and risk event instances whose comprehensive weights exceed a first preset threshold are selected to form a frequent item set. The generation module is used to generate candidate k-th itemsets based on frequent itemset iterative combinations, where k is an integer not less than two; calculate the spatial proximity and temporal proximity of any two risk event combinations in the candidate k-th itemset, and aggregate them to obtain the spatial-temporal fusion degree; and match the occurrence order of each risk event instance with a pre-set evolution sequence template to obtain the temporal consistency score. The calculation module is used to obtain the linkage bias coefficient and basic linkage alarm threshold from the risk linkage matrix; and adjust the basic linkage alarm threshold according to the spatial-temporal fusion degree to generate an alarm threshold; calculate the linkage support by combining the comprehensive weight, temporal consistency score and linkage bias coefficient; when the linkage support is greater than the alarm threshold, the candidate k-th itemset is determined as the frequent k-th itemset; the k value is incremented and the candidate itemset generation, fusion degree calculation, temporal matching, threshold adjustment, support calculation and frequent itemset determination are repeated until the preset maximum itemset length is reached or no new candidate itemsets are generated; the obtained frequent itemsets are set as linkage rules to generate linkage alarm handling strategies.

[0006] This invention generates a comprehensive weight by integrating basic risk weights with a time decay factor adjusted according to severity level, thus measuring the timeliness and importance of risks, reducing interference from outdated data, and improving the accuracy of basic event screening. When identifying risk correlations, the spatial and temporal proximity of risk combinations is considered to aggregate them into a spatiotemporal fusion degree, and the matching degree between the event occurrence sequence and preset templates is evaluated, revealing the hidden evolutionary patterns among multiple types of risks. Furthermore, the basic alarm judgment criteria are flexibly adjusted according to spatiotemporal characteristics, and the linkage support degree is calculated by combining multi-dimensional indicators such as temporal consistency score and linkage bias coefficient, making the selected linkage relationship rules more aligned with actual business scenarios. This reduces the probability of false alarms and false negatives in risk monitoring, constructs a linkage alarm handling strategy that fits the real-world situation, and enhances the efficiency and defense level of collaborative handling of multiple concurrent risks in industrial settings. Attached Figure Description

[0007] Figure 1 A flowchart of the first embodiment; Figure 2 A bar chart showing the basic risk weighting calculation; Figure 3 A schematic diagram of the time decay coefficient curves under different severity levels; Figure 4 A bar chart is constructed for the spatiotemporal fusion components. Detailed Implementation

[0008] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0009] In the first embodiment, the present invention proposes a method for handling multi-risk linkage alarms in industrial sites, such as... Figure 1 It includes the following steps: S1. Acquire multi-dimensional time-series data from the industrial site, convert it into a discrete set of risk events, and configure basic risk weights for each risk event type; for any risk event instance, adjust the basic time decay rate according to the severity level; calculate the time decay factor based on the interval between the occurrence time and the current time; merge the basic risk weights and the time decay factor to obtain a comprehensive weight; filter risk event instances whose comprehensive weights exceed a first preset threshold to form a frequent item set.

[0010] Multidimensional continuous time-series data from temperature and pressure sensors are acquired using a message queue telemetry transmission protocol. The acquired data is read and discretized. An equal-frequency binning strategy is used to map the discretized data into a discrete set of risk events, each containing at least an event identifier and occurrence time. When generating this discrete set of risk events, the system synchronously reads the sensor node identifier that triggered the risk event and, based on this identifier, retrieves the corresponding spatial latitude and longitude coordinates, absolute altitude, and spatial region association code from the industrial site 3D geographic information system database or equipment ledger. Thus, each risk event instance in the discrete set is associated with an event identifier, event type identifier, occurrence time, sensor node identifier, spatial latitude and longitude coordinates, absolute altitude, and spatial region association code for subsequent spatial proximity calculations, time-series matching, and linkage alarm handling strategy matching.

[0011] Establish a key-value mapping table between risk event types and weights, and use a dictionary data structure to pre-assign a basic risk weight value greater than 0 to each type of risk event.

[0012] The severity level label carried in the risk event instance log is read and mapped to a severity level value from 1 to 5, where the larger the value, the higher the severity. A base time decay rate constant is preset in the program. The base time decay rate constant is divided by the severity level value to obtain the adjusted time decay rate, so that the higher the severity level of the risk event, the slower the decay.

[0013] Obtain the current system timestamp and extract the occurrence timestamp from the risk event instance data packet. Calculate the time interval by subtracting the occurrence timestamp from the current timestamp. Use the absolute value of the time interval as the independent variable and the adjusted time decay rate as the decay parameter. Substitute these values ​​into the natural exponential decay model to calculate the product of the negative time interval and the time decay rate at the base of the natural logarithm, yielding a time decay factor that varies between 0 and 1.

[0014] Extract the basic risk weights and the calculated time decay factor from the mapping table. Using a weighted multiplication algorithm, multiply the basic risk weights by the time decay factor to output the current comprehensive weight value of the risk event instance. This comprehensive weight value decreases over time.

[0015] An empty list is initialized in memory to store frequent itemsets. A first preset threshold is set to a floating-point number. A for loop iterates through all risk event instances in the discrete risk event set, reading their corresponding composite weight values. The composite weight is compared with the first preset threshold using a greater-than relational operator. When the composite weight value is greater than the first preset threshold, the risk event instance is added to the empty list. After iteration, the list is deduplicated to obtain frequent itemsets consisting of single high-weight risk event instances.

[0016] In an optional embodiment, the step of acquiring multidimensional time-series data from the industrial site, converting it into a discrete set of risk events, and configuring basic risk weights for each risk event type includes: Extract discrete time nodes from the multidimensional time-series data of the industrial site and arrange them in chronological order to generate a time-series axis containing all time nodes; Deduplication and merging of similar risk events occurring in the same spatial region or the same sensor node at each discrete time node of the time series axis yields a discrete risk event set. Obtain the average historical accident economic loss recorded in the knowledge base for each risk event type, and divide the average historical accident economic loss by the set benchmark economic loss value to obtain the corresponding ratio, which serves as the basic risk weight for the risk event type.

[0017] By deploying various sensors in industrial sites, such as chemical workshops, to collect multidimensional time-series data such as temperature, pressure, and gas concentration at a frequency of 1 to 10 times per second, and by setting threshold conditions such as temperature > 85℃ and pressure > 1.5MPa, the continuous data is converted into discrete time nodes with timestamps, such as 2023-10-25 10:00:00, and arranged in ascending order of time to generate a unified time-series axis. For multiple similar risk events occurring on the timeline, such as "reactor temperature too high" triggered three times consecutively, deduplication and merging are performed within a preset time sliding window, preferably between 30 and 300 seconds, for example, within 120 seconds. This integrates multiple similar risk events into a single independent risk event instance, reducing redundant events caused by sensor jitter or high-frequency sampling, and resulting in a structured set of discrete risk events. During the deduplication and merging process, the system retains the sensor node identifier that triggered the risk event and reads the spatial latitude and longitude coordinates, absolute height, and spatial region association code corresponding to the sensor node from the industrial site 3D geographic information system database or equipment ledger. When multiple sensor nodes in the same spatial region are merged into the same risk event instance, the center coordinates of the triggering node, the representative equipment coordinates, or the preset region coordinates are used as the spatial location attributes of the risk event instance.

[0018] The historical accident knowledge base is accessed to query the average economic loss for each type of risk event. For example, the average loss for an accident caused by "overheating of the reactor" is 500,000 yuan, and for "minor pipeline leakage" it is 20,000 yuan. A unified benchmark economic loss value is set, preferably within the range of 10,000 to 100,000 yuan, with an example value of 100,000 yuan. The extracted average historical loss is then divided by this benchmark value. For example, dividing 500,000 yuan by 100,000 yuan yields a ratio of 5.0. This ratio of 5.0 is set as the basic risk weight for the "overheating of the reactor" risk event type. The calculation results of the historical average loss and corresponding basic risk weight for the four types of risk events are as follows: Figure 2 As shown.

[0019] In an optional embodiment, adjusting the base time decay rate according to the severity level for any risk event instance; calculating the time decay factor based on the interval between the occurrence time and the current time includes: The adjusted time decay rate is obtained by dividing the preset base time decay rate constant by the severity level value of any risk event instance. The adjusted time decay rate has a dimension that is the reciprocal of the time dimension. Obtain the system timestamp of the current time, subtract the timestamp of the occurrence time of the risk event instance, and obtain the time interval difference with time dimension; The time decay factor is calculated using a negative exponential decay method. Specifically, the time interval difference is multiplied by the adjusted time decay rate, the negative of the product is taken as the exponent, and the power value corresponding to the exponent is calculated with the natural constant as the base, thus obtaining the time decay factor.

[0020] Obtain a preset base time decay rate constant, preferably ranging from 0.01 to 0.1, with an example value of 0.05 and a unit of 1 / minute. Then, read the severity level value of the specific risk event instance, set to an integer from 1 to 5, where 5 represents the highest severity and 1 represents the lowest severity; the example is severity level 4. Divide the base time decay rate constant 0.05 by the severity level value 4 to calculate the adjusted time decay rate v, which is 0.0125, with a unit of 1 / minute. This ensures that the time decay rate is relatively slower for events with higher severity levels, resulting in a relatively longer information retention period.

[0021] Get the timestamp of the current time, for example, a timestamp value of 1698201600 corresponds to 2023-10-25 10:40:00. Extract the occurrence timestamp recorded when the risk event instance was generated, for example, 1698200400 corresponds to 2023-10-25 10:20:00. Subtract the two to get the time interval difference. The time interval is 1200 seconds, converted to 20 minutes. Substituting this into the negative exponential decay formula, the time interval difference is calculated. Multiplying 20 minutes by the adjusted time decay rate v (0.0125) yields 0.25. Then, using the natural constant e (approximately 2.71828) as the base, we calculate e to the power of -0.25 to obtain the current time decay factor for this risk event instance. Approximately 0.7788, this calculation result represents the time decay state of the risk event at the current moment, such as... Figure 3 As shown.

[0022] S2 generates candidate k-th itemsets based on iterative combinations of frequent itemsets, where k is an integer not less than two; calculates the spatial proximity and temporal proximity of any two risk event combinations within the candidate k-th itemset, and aggregates them to obtain the spatial-temporal fusion degree; matches each risk event instance with a pre-set evolutionary sequence template according to the order of occurrence to obtain the temporal consistency score.

[0023] A candidate itemset iterative generation logic is employed, initializing k to 2. Risk event instances from frequent itemsets are combined using k elements to generate candidate itemsets of length k. In each iteration, the generated candidate k-th itemset undergoes fusion calculation, temporal matching, threshold adjustment, support calculation, and frequent itemset determination. The candidate itemset generation process does not use the assumption that all subsets of length k-1 are frequent itemsets as the sole pruning condition. Instead, it can use engineering constraints based on the maximum time window, spatial region range, or a preset maximum itemset length for selection, retaining the remaining itemsets to form the candidate k-th itemset.

[0024] The latitude and longitude coordinates of the sensors corresponding to any two risk events in the candidate k-th item set are queried using a 3D geographic information system database, and the spherical distance between them is calculated. In implementations that need to consider height differences, the latitude and longitude coordinates and absolute height can also be converted to a unified local spatial coordinate system to calculate the Euclidean spatial distance, and the resulting spatial distance is then dimensionless. The spatial proximity is obtained by taking the reciprocal of the dimensionless spatial distance plus 1. The absolute difference between the timestamps of the two risk events is calculated, and the absolute difference is divided by a preset maximum time window to obtain the dimensionless time difference. The reciprocal of the dimensionless time difference plus 1 is taken to obtain the temporal proximity. The spatial proximity and the temporal proximity are directly added to obtain a single-pair fusion value. The single-pair fusion values ​​of all risk event combinations in the candidate k-th item set are accumulated and normalized to generate a spatial-temporal fusion value representing the strength of the spatial and temporal correlation of risk events, such as... Figure 4 As shown.

[0025] Read the standard evolution matching character sequence corresponding to the preset evolution sequence template, extract all risk event instances in the candidate k-th itemset and sort them according to their occurrence time to generate the target time series character sequence; calculate the longest common subsequence between the target time series character sequence and the standard evolution matching character sequence, and divide the number of nodes in the longest common subsequence by the total number of risk events in the candidate k-th itemset to obtain the time series consistency score.

[0026] In an optional embodiment, the step of calculating the spatial proximity and temporal proximity of any two risk event pairs within the candidate k-th itemset and aggregating them to obtain the spatial-temporal fusion degree includes: Extract any two risk event pairs within the candidate k-th itemset, obtain the absolute value of the time difference between the timestamps of the occurrence of the event pairs and the spherical distance of the spatial location latitude and longitude coordinates, and perform dimensionless processing on each pair. The spatial-temporal fusion degree of the candidate k-th itemset is generated based on the aggregation calculation model. The spatial-temporal fusion degree is calculated as follows: The reciprocal of the sum of the dimensionless spatial latitude and longitude spherical distances between the corresponding event pairs and a constant 1 is obtained. The reciprocal of the sum of the absolute values ​​of the time differences of the occurrence timestamps between the corresponding event pairs and a constant 1 is also obtained. The sum of these two reciprocals yields the single-pair fusion value for the risk event pair. ; According to the formula The spatial-temporal fusion degree is obtained.

[0027] For a candidate k-th itemset containing k risk event instances (e.g., k=3), iterate through all pairs of events to generate a total of k(k-1) / 2 pairs (3 pairs when k=3). For each pair, calculate the absolute time difference between the timestamps of events A and B (e.g., a time difference of 150 seconds), and calculate the spherical distance between the spatial latitude and longitude coordinates of the sensor nodes bound to them (e.g., a spatial distance of 25 meters). To remove the influence of different dimensions, a normalization method is used, dividing the absolute time difference by a preset maximum allowable time window (e.g., 300 seconds) to obtain a dimensionless result. For example, 150 / 300=0.5. Divide the spherical distance by the preset maximum monitoring radius, such as 100 meters, to obtain a dimensionless value. For example, 25 / 100 = 0.25.

[0028] After obtaining the dimensionless parameters for all event pairs, perform aggregate calculations: for the aforementioned events A and B, first calculate... That is, 0.25. Adding the reciprocal of 1 gives the spatial proximity, which is 1 / 1.25 = 0.8. Then calculate... The factor is 0.5. Adding the reciprocal of 1 gives the time proximity, which is approximately 1 / 1.5 ≈ 0.667. Adding these two together gives the single-pair fusion value of this combination, which is 0.8 + 0.667 = 1.467. The single-pair fusion values ​​of all three combinations in the candidate set are accumulated. Assuming the calculated values ​​for the other two pairs are 1.2 and 1.5, the total sum is 4.167. Simultaneously, the coefficient is calculated by dividing the constant 2 by k squared minus k, i.e., -3=6, resulting in a coefficient of 2 / 6=0.333; multiplying this coefficient 0.333 by the cumulative sum 4.167 yields the spatial-temporal fusion degree. It is 1.388.

[0029] In an optional embodiment, the step of matching the occurrence order of each risk event instance with a pre-set evolution sequence template to obtain a temporal consistency score includes: The risk event type identifiers in the candidate k-th itemset are sorted according to their occurrence time to construct the target time series character sequence, and the standard evolution matching character sequence corresponding to the preset evolution sequence template is read. Calculate the longest common subsequence between the target temporal character sequence and the standard evolutionary matching character sequence, and count the number of nodes contained in the longest common subsequence; The temporal consistency score is obtained by dividing the number of nodes by the total number of risk events in the candidate k-th itemset.

[0030] Retrieve all risk event instances within the candidate k-th itemset, sort them in ascending order based on their timestamps, and extract the corresponding risk event type identifiers. Type identifiers are encoded as uppercase English letters, such as A for pressure anomaly, B for valve leakage, and C for temperature exceeding limits. Concatenate the sorted identifiers to obtain the target time-series character sequence string, for example, the actual occurrence sequence "ABCE". Simultaneously, access the configured knowledge base and read the corresponding standard evolution matching character sequence according to the triggering rules; for example, the preset standard accident chain sequence is "ABDCF".

[0031] A dynamic programming algorithm is used to find the longest common subsequence between the target temporal character sequence "ABCE" and the standard evolutionary matching character sequence "ABDCF". After comparing and backtracking character by character, the longest common subsequence "ABC" is identified. The extracted common subsequence is traversed and counted, revealing that it contains 3 nodes, namely nodes A, B, and C. The number of nodes 3 is divided by the total number of risk events in the candidate k-th item set. In this example, the total number of events k=4, namely events A, B, C, and E. The resulting temporal consistency score is 3 / 4 = 0.75. This value, ranging from 0 to 1, represents the degree of matching between the actual risk sequence and the pre-set evolutionary sequence template.

[0032] S3: Obtain the linkage bias coefficient and basic linkage alarm threshold from the risk linkage matrix; adjust the basic linkage alarm threshold according to the spatial-temporal fusion degree to generate an alarm threshold; calculate the linkage support by combining the comprehensive weight, temporal consistency score and linkage bias coefficient; when the linkage support is greater than the alarm threshold, determine the candidate k-th itemset as the frequent k-th itemset; increment the k value and repeat the candidate itemset generation, fusion degree calculation, temporal matching, threshold adjustment, support calculation and frequent itemset determination until the preset maximum itemset length is reached or no new candidate itemsets are generated; set the obtained frequent itemsets as linkage rules to generate a linkage alarm handling strategy.

[0033] A risk linkage matrix of DataFrame structure is constructed, where rows and columns represent different types of risk events, and the matrix intersections store prior linkage bias coefficients and basic linkage alarm thresholds obtained from training a historical association rule recognition algorithm. When the candidate k-th itemset contains two risk event types, these two risk event types are directly used as row and column indices to extract the corresponding linkage bias coefficients and basic linkage alarm thresholds. When the candidate k-th itemset contains three or more risk event types, all pairwise risk event type combinations within the candidate k-th itemset are traversed, and the linkage bias coefficients and basic linkage alarm thresholds corresponding to each combination are read from the risk linkage matrix. The arithmetic mean of the read linkage bias coefficients is then performed to obtain the linkage bias coefficient corresponding to the candidate k-th itemset, and the arithmetic mean of the read basic linkage alarm thresholds is then performed to obtain the basic linkage alarm threshold corresponding to the candidate k-th itemset.

[0034] The basic linkage alarm threshold obtained from the risk linkage matrix and the aggregated spatial-temporal fusion degree are extracted. A scaling adjustment method negatively correlated with the spatial-temporal fusion degree is used to adjust the basic linkage alarm threshold. Specifically, a constant 1 is added to the spatial-temporal fusion degree to obtain an adjustment factor. The constant 1 is divided by the adjustment factor to obtain a decay scaling coefficient. The basic linkage alarm threshold is then multiplied by the decay scaling coefficient to obtain the changed alarm threshold. This results in lower alarm thresholds for candidate sets with higher spatial-temporal fusion degrees, reducing the threshold for triggering alarms from strongly correlated events.

[0035] Extract the comprehensive weights of all risk event instances within the candidate k-th itemset, and select the minimum comprehensive weight value. Read the calculated temporal consistency score and the obtained linkage bias coefficient, and multiply the minimum comprehensive weight value, temporal consistency score, and linkage bias coefficient together to obtain the linkage support of the candidate k-th itemset. This linkage support is not limited to a numerical range of 0 to 1; its numerical scale corresponds to the set scale of the comprehensive weight and alarm threshold.

[0036] The greater than relational operator is used to compare the linkage support value with the reduced alarm threshold. When the linkage support is greater than the alarm threshold, it indicates that the candidate k-th itemset has sufficiently strong risk intensity and rule correlation. The add method of the set is called to store the candidate k-th itemset into a hash set structure specifically used to store frequent k-th itemsets, thus completing the confirmation and retention of high-support linkage risk combinations.

[0037] Set up a while loop structure, and set the loop termination condition as k reaching the preset maximum itemset length, or failing to generate a new candidate k-th itemset under preset engineering constraints; automatically increment the value of k by 1 before the end of each loop, and generate the next-order candidate itemset based on the frequent 1-itemset, the previous round of screening results and the preset engineering constraints. Repeat the above processes of generating candidate sets, calculating proximity, matching time sequences, adjusting thresholds and comparing support degrees continuously. When k reaches the preset maximum itemset length, or fails to generate a new candidate k-th itemset under the preset engineering constraints, the while loop condition is determined to be false, the loop is immediately exited, and the iterative calculation of the algorithm is stopped.

[0038] Create a linked alarm handling strategy configuration file in JSON format, traverse all non-empty frequent itemsets generated during the iteration process, and convert the frequent itemsets into association rule objects composed of antecedents and consequents. Serialize these association rule objects by calling the dumps function, query and match the corresponding control instructions or manual inspection dispatch templates for each rule object in the handling strategy database, package and encapsulate them into an executable linked alarm handling strategy file, and push it to the distributed industrial control terminal to realize collaborative defense against on-site risks.

[0039] In an optional embodiment, the step of adjusting the basic linked alarm threshold according to the space-time fusion degree to generate an alarm threshold includes: Add the space-time fusion degree to the constant 1 to obtain an adjustment factor; Divide the constant 1 by the adjustment factor to obtain an attenuation scaling coefficient; Read the corresponding basic linked alarm threshold and multiply it by the attenuation scaling coefficient to calculate a truncation comparison value; Set the truncation comparison value as the alarm threshold.

[0040] Extract the calculated value of the space-time fusion degree. Assuming the calculated value is 1.50, perform an addition operation to add this value to the constant 1, and the calculated adjustment factor is 2.50. Through this operation, it is ensured that the adjustment factor is always greater than 1, avoiding the situation that the denominator is 0 in the subsequent division calculation. Divide the constant 1 by the obtained adjustment factor 2.50, and the calculated attenuation scaling coefficient is 0.40; the attenuation scaling coefficient represents the adjustment ratio of the space-time tightness to the alarm triggering threshold, and the higher the space-time fusion degree, the smaller the attenuation scaling coefficient.

[0041] The basic linkage alarm threshold set for this type of device group is read, preferably ranging from 1.0 to 10.0, with an example value set to 8.0. The read basic linkage alarm threshold of 8.0 is then multiplied by the aforementioned attenuation scaling factor of 0.40 to calculate a truncation comparison value of 3.2. This calculated truncation comparison value of 3.2 is set as the alarm threshold; this adjustment mechanism allows small, spatiotemporally clustered anomalies to be detected through the reduced threshold.

[0042] In an optional embodiment, the calculation of linkage support by combining comprehensive weights, temporal consistency scores, and linkage bias coefficients includes: The linkage support of the candidate k-th itemset is calculated according to the linkage support calculation method. Specifically, the minimum comprehensive weight value is selected from the comprehensive weights of each risk event instance in the candidate k-th itemset, and the minimum comprehensive weight value, the temporal consistency score, and the linkage bias coefficient are multiplied in sequence to obtain the linkage support of the candidate k-th itemset.

[0043] Iterate through all k risk events contained in the current candidate k-th itemset. For example, if k=3, the events included... , , Read the previously calculated composite weights for each event. ,For example The weight is 4.5. It is 3.2. The minimum overall weight is set to 5.1, and the minimum overall weight value in the set is found. In this example, the minimum value is 3.2. The minimum overall weight in the set is used as the representative value so that the evaluation result of the entire risk event set represents the event with the lowest weight, avoiding the situation where high-weight events mask the impact of low-weight events.

[0044] Read the time-series consistency score of this itemset For example, 0.85 and the linkage bias coefficient α obtained from the configuration file are preferably set between 1.0 and 2.0, with an example configuration value of 1.2. The calculation is performed according to the aforementioned linkage support calculation method, multiplying the extracted minimum comprehensive weight of 3.2, the temporal consistency score of 0.85, and the linkage bias coefficient of 1.2 sequentially to obtain a result of 3.264. This result is then used as the linkage support. That is, 3.264. This multiplication mechanism integrates multiple parameters into a single evaluation metric.

[0045] In an optional embodiment, setting the obtained frequent itemsets as linkage rules and generating a linkage alarm handling strategy includes: Traverse each generated frequent k-th itemset, combine causal chain rules and confidence constraints, extract the individual risk event at the end of the causal chain, set the event as the successor result event of the target triggered by the specific itemset, and set all other risk events in the frequent k-th itemset as the condition-driven cause event set; Extract the spatial region association code within the condition-driven cause event set, match it with the preset linkage response instruction library, generate a linkage alarm handling strategy pointing to the subsequent result event, and write it into the system's configuration record.

[0046] Before generating the aforementioned linkage rules, the system pre-constructs a causal chain rule base for equipment. This rule base is generated from historical accident records, equipment process constraints, expert-annotated accident evolution paths, and on-site handling records. Each causal chain rule includes at least a preceding risk event, a subsequent risk event, the sequence of events, the applicable spatial area, and a confidence level field. The confidence level field characterizes the reliability of the causal chain rule in historical accident samples or expert verification results, and can be determined based on the frequency of occurrence of the corresponding causal chain in historical samples, manual annotation confirmation results, or the confidence level set by the rule base maintenance personnel. Therefore, the frequent k-th itemset is used to filter risk combinations with statistical linkage relationships, and the equipment causal chain rule base is used to determine the causal event and subsequent result event within the risk combination, avoiding the direct equation of ordinary correlation with causal relationship.

[0047] Once a set of frequent itemsets (k-th itemsets) is generated, iterate through each frequent itemset. For example, an itemset might contain three events: "cooling pump shutdown," "reactor pressure rise," and "safety valve activation." Obtain the preset equipment causal chain rules and, combined with confidence constraints (e.g., setting the confidence level to be greater than 80%), sort the temporal causal relationships of the events within the itemset. Identify the single risk event at the end of the causal chain through this sorting; in this example, it's "safety valve activation." Designate this event as the successor result event of the specific itemset's trigger target. Simultaneously, set the remaining events, "cooling pump shutdown" and "reactor pressure rise," as condition-driven cause event sets.

[0048] The system extracts the spatial region association codes corresponding to each sensor from the set of condition-driven cause events, such as "Region Code: Area-05-ReactorA". Using this region association code and event type as conditions, it queries a pre-set linkage response command library to match the corresponding handling command, such as "Activate backup cooling water circuit" or "Issue area evacuation broadcast". The matched handling command is combined with the identified subsequent result events to generate a linkage alarm handling strategy, such as "Execute handling command to prevent subsequent result events when condition-driven events occur". This handling strategy is written into the system's configuration record, ensuring that a response is executed when the same preconditions are monitored.

[0049] In the second embodiment, the present invention also proposes an industrial site multi-risk linkage alarm and handling system, including the following modules: The adjustment module is used to acquire multi-dimensional time-series data from industrial sites, convert it into a set of discrete risk events, and configure basic risk weights for each risk event type; for any risk event instance, the basic time decay rate is adjusted according to the severity level; the time decay factor is calculated based on the interval between the occurrence time and the current time; the basic risk weights and the time decay factor are merged to obtain a comprehensive weight; and risk event instances whose comprehensive weights exceed a first preset threshold are selected to form a frequent item set. The generation module is used to generate candidate k-th itemsets based on frequent itemset iterative combinations, where k is an integer not less than two; calculate the spatial proximity and temporal proximity of any two risk event combinations in the candidate k-th itemset, and aggregate them to obtain the spatial-temporal fusion degree; and match the occurrence order of each risk event instance with a pre-set evolution sequence template to obtain the temporal consistency score. The calculation module is used to obtain the linkage bias coefficient and basic linkage alarm threshold from the risk linkage matrix; and adjust the basic linkage alarm threshold according to the spatial-temporal fusion degree to generate an alarm threshold; calculate the linkage support by combining the comprehensive weight, temporal consistency score and linkage bias coefficient; when the linkage support is greater than the alarm threshold, the candidate k-th itemset is determined as the frequent k-th itemset; the k value is incremented and the candidate itemset generation, fusion degree calculation, temporal matching, threshold adjustment, support calculation and frequent itemset determination are repeated until the preset maximum itemset length is reached or no new candidate itemsets are generated; the obtained frequent itemsets are set as linkage rules to generate linkage alarm handling strategies.

[0050] The above description represents the preferred embodiments of the present invention. It should be noted that, for those skilled in the art, various improvements and modifications can be made without departing from the principles of the present invention, and these improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A method for handling multi-risk linkage alarms in industrial sites, characterized in that, include: Acquire multidimensional time-series data from industrial sites, convert it into a set of discrete risk events, and configure basic risk weights for each risk event type; For any instance of a risk event, adjust the base time decay rate according to the severity level; Calculate the time decay factor based on the interval between the occurrence time and the current time; The basic risk weight is combined with the time decay factor to obtain the comprehensive weight; Instances of risk events whose overall weight exceeds a first preset threshold are selected to form a frequent item set. Candidate k-th itemsets are generated based on frequent item set iterative combinations, where k is an integer not less than two. For any two risk event pairs in the candidate k-th itemset, calculate the spatial proximity and temporal proximity, and aggregate them to obtain the spatial-temporal fusion degree; match the occurrence order of each risk event instance with the pre-set evolution sequence template to obtain the temporal consistency score; Obtain the linkage bias coefficient and basic linkage alarm threshold from the risk linkage matrix; And based on the spatial-temporal fusion degree, the basic linkage alarm threshold is adjusted to generate alarm thresholds; The linkage support is calculated by combining the comprehensive weight, the temporal consistency score, and the linkage bias coefficient. When the linkage support is greater than the alarm threshold, the candidate k-th itemset is determined as the frequent k-th itemset. The value of k is incremented and the candidate itemset generation, fusion calculation, temporal matching, threshold adjustment, support calculation, and frequent itemset determination are repeated until the preset maximum itemset length is reached or no new candidate itemsets are generated. The obtained frequent itemsets are set as linkage rules to generate linkage alarm handling strategies.

2. The method according to claim 1, characterized in that, The process of acquiring multidimensional time-series data from industrial sites, converting it into a discrete set of risk events, and configuring basic risk weights for each risk event type includes: Extract discrete time nodes from the multidimensional time-series data of the industrial site and arrange them in chronological order to generate a time-series axis containing all time nodes; Deduplicatively merge similar risk events occurring in the same spatial region or the same sensor node at each discrete time node of the time series axis to obtain a discrete risk event set. Obtain the average historical accident economic loss recorded in the knowledge base for each risk event type, and divide the average historical accident economic loss by the set benchmark economic loss value to obtain the corresponding ratio, which serves as the basic risk weight for the risk event type.

3. The method according to claim 1, characterized in that, For any instance of a risk event, the base time decay rate is adjusted according to the severity level. The time decay factor is calculated based on the interval between the occurrence time and the current time, including: The adjusted time decay rate is obtained by dividing the preset base time decay rate constant by the severity level value of any risk event instance. The adjusted time decay rate has a dimension that is the reciprocal of the time dimension. Obtain the system timestamp of the current time, subtract the timestamp of the occurrence time of the risk event instance, and obtain the time interval difference with time dimension; The time decay factor is calculated using a negative exponential decay method. Specifically, the time interval difference is multiplied by the adjusted time decay rate, the negative of the product is taken as the exponent, and the power value corresponding to the exponent is calculated with the natural constant as the base, thus obtaining the time decay factor.

4. The method according to claim 1 or 2, characterized in that, The calculation of spatial and temporal proximity for any two risk event pairs within the candidate k-th itemset, and the aggregation to obtain the spatial-temporal fusion degree, includes: Extract any two risk event pairs within the candidate k-th itemset, obtain the absolute value of the time difference between the timestamps of the occurrence of the event pairs and the spherical distance of the spatial location latitude and longitude coordinates, and perform dimensionless processing on each pair. The spatial-temporal fusion degree of the candidate k-th itemset is generated based on the aggregation calculation model. The spatial-temporal fusion degree is calculated as follows: The reciprocal of the sum of the dimensionless spatial latitude and longitude spherical distances between the corresponding event pairs and a constant 1 is obtained. The reciprocal of the sum of the absolute values ​​of the time differences of the occurrence timestamps between the corresponding event pairs and a constant 1 is also obtained. The sum of these two reciprocals yields the single-pair fusion value for the risk event pair. ; According to the formula The spatial-temporal fusion degree is obtained.

5. The method according to claim 1, characterized in that, The step of matching the occurrence order of each risk event instance with a pre-set evolution sequence template to obtain a temporal consistency score includes: The risk event type identifiers in the candidate k-th itemset are sorted according to their occurrence time to construct the target time series character sequence, and the standard evolution matching character sequence corresponding to the preset evolution sequence template is read. Calculate the longest common subsequence between the target temporal character sequence and the standard evolutionary matching character sequence, and count the number of nodes contained in the longest common subsequence; The temporal consistency score is obtained by dividing the number of nodes by the total number of risk events in the candidate k-th itemset.

6. The method according to claim 1, characterized in that, The process of adjusting the basic linkage alarm threshold based on the spatial-temporal fusion degree to generate alarm thresholds includes: Adding a constant 1 to the spatial-temporal fusion degree yields the adjustment factor; Divide the constant 1 by the adjustment factor to obtain the attenuation scaling factor; Read the corresponding basic linkage alarm threshold and multiply it by the attenuation scaling factor to calculate the truncation comparison value; Set the truncation comparison value as the alarm threshold.

7. The method according to claim 1, characterized in that, The calculation of linkage support by combining comprehensive weights, temporal consistency scores, and linkage bias coefficients includes: The linkage support of the candidate k-th itemset is calculated according to the linkage support calculation method. Specifically, the minimum comprehensive weight value is selected from the comprehensive weights of each risk event instance in the candidate k-th itemset, and the minimum comprehensive weight value, the temporal consistency score, and the linkage bias coefficient are multiplied in sequence to obtain the linkage support of the candidate k-th itemset.

8. The method according to claim 1, characterized in that, The step of setting the obtained frequent itemsets as linkage rules and generating linkage alarm handling strategies includes: Traverse each generated frequent k-th itemset, combine causal chain rules and confidence constraints, extract the individual risk event at the end of the causal chain, set the event as the successor result event of the target triggered by the specific itemset, and set all other risk events in the frequent k-th itemset as the condition-driven cause event set; Extract the spatial region association code within the condition-driven cause event set, match it with the preset linkage response instruction library, generate a linkage alarm handling strategy pointing to the subsequent result event, and write it into the system's configuration record.

9. An industrial site multi-risk linkage alarm and handling system, characterized in that, include: The adjustment module is used to acquire multi-dimensional time-series data from industrial sites, convert it into a set of discrete risk events, and configure basic risk weights for each risk event type. For any instance of a risk event, adjust the base time decay rate according to the severity level; Calculate the time decay factor based on the interval between the occurrence time and the current time; The basic risk weight is combined with the time decay factor to obtain the comprehensive weight; Instances of risk events whose overall weight exceeds a first preset threshold are selected to form a frequent item set. The generation module is used to generate candidate k-th itemsets based on iterative combinations of frequent itemsets, where k is an integer not less than two. For any two risk event pairs in the candidate k-th itemset, calculate the spatial proximity and temporal proximity, and aggregate them to obtain the spatial-temporal fusion degree; match the occurrence order of each risk event instance with the pre-set evolution sequence template to obtain the temporal consistency score; The calculation module is used to obtain the linkage bias coefficient and the basic linkage alarm threshold from the risk linkage matrix; And based on the spatial-temporal fusion degree, the basic linkage alarm threshold is adjusted to generate alarm thresholds; The linkage support is calculated by combining the comprehensive weight, the temporal consistency score, and the linkage bias coefficient. When the linkage support is greater than the alarm threshold, the candidate k-th itemset is determined as the frequent k-th itemset. The value of k is incremented and the candidate itemset generation, fusion calculation, temporal matching, threshold adjustment, support calculation, and frequent itemset determination are repeated until the preset maximum itemset length is reached or no new candidate itemsets are generated. The obtained frequent itemsets are set as linkage rules to generate linkage alarm handling strategies.

10. The system according to claim 9, characterized in that, The process of acquiring multidimensional time-series data from industrial sites, converting it into a discrete set of risk events, and configuring basic risk weights for each risk event type includes: Extract discrete time nodes from the multidimensional time-series data of the industrial site and arrange them in chronological order to generate a time-series axis containing all time nodes; Deduplicatively merge similar risk events occurring in the same spatial region or the same sensor node at each discrete time node of the time series axis to obtain a discrete risk event set. Obtain the average historical accident economic loss recorded in the knowledge base for each risk event type, and divide the average historical accident economic loss by the set benchmark economic loss value to obtain the corresponding ratio, which serves as the basic risk weight for the risk event type.

Images