SIS safety integrity level dynamic verification method and system considering uncertainty

Abstract

This invention belongs to the technical field of Safety Integrity Level (SIL) verification for Safety Instrumented Systems (SIS), specifically disclosing a dynamic verification method and system for SIL considering uncertainty. Addressing the uncertainty issues in SIL verification, this invention introduces stochastic uncertainty theory and fuzzy uncertainty theory, combining stochastic uncertainty theory with LSTM networks and fuzzy uncertainty theory with fuzzy logic to propose a novel dynamic SIL verification method. It also performs source tracing analysis on the SIL verification results and uses a random forest method for feature importance analysis. The SIL level is dynamically updated as data is updated. This invention solves the problems of uncertainty and static lag in traditional SIL verification, eliminates the fuzzy uncertainty of expert judgment and the static nature of field data, and enables the final verification results to be dynamic and up-to-date.

Description

Technical Field

[0001] This invention belongs to the technical field of safety integrity level verification for safety instrumented systems, and specifically relates to a dynamic verification method and system for SIS safety integrity level considering uncertainty. Background Technology

[0002] According to IEC-61508 and IEC-61511 standards, a Safety Instrumented System (SIS) consists of a sensor subsystem, a controller subsystem, and an actuator subsystem, and it can perform one or more Safety Instrumented Functions (SIFs). The Safety Integrity Level (SIL) refers to the probability that the SIS can perform its required safety instrumented functions under specified conditions and within a specified time. After assigning a Safety Integrity Level to an in-service Safety Instrumented System through hazard analysis and risk assessment, SIL compliance verification is required to confirm whether the current loop structure and selected components meet the required SIL level, ensuring that the SIS meets the necessary safety functions. Commonly used SIL verification calculation models include the reliability block diagram method, the fault tree method, and the Markov model method. All three models use static, fixed constant data as input for SIL verification calculations, and the SIL verification results are not updated with subsequent new input data.

[0003] Furthermore, SIL (Safety Integrity Detection) verification for SIS (System-Independent Components) typically involves numerous uncertainties, such as incomplete historical failure data for parameters, low reliability of collected historical data, inconsistent parameter values ​​assessed by experts, and significant fluctuations in parameter values ​​between verifications. Since SIL verification is usually conducted over a 3-5 year cycle, these uncertainties can cause the verification results to fail to accurately reflect the SIS status over that period. In other words, the static verification results deviate from the actual on-site SIS status over 3-5 years. Because the parameter data from SIL verification fluctuates within this timeframe, a fixed constant cannot represent the actual state over the entire period. This situation severely hinders employees' application and maintenance of the SIS and poses significant potential safety risks to production activities. Therefore, eliminating uncertainties in the SIL verification process and dynamically updating SIL verification results is essential.

[0004] Uncertainty in SIL validation is difficult to quantify, and the types and data uncertainties of the parameters involved are key aspects of upgrading SIL validation efforts. While some existing studies analyze the uncertainty of a single parameter during SIL validation and quantify the uncertainty of some parameters to a certain extent, they do not quantify and categorize the uncertainty of all parameters throughout the entire process. This results in static SIL validation results. However, the methods for handling parameters under different uncertainties differ significantly. Only by comprehensively considering and classifying all parameters, and by dynamically updating and updating SIL validation results, can we better reflect reality and guide fieldwork.

[0005] Meanwhile, research on dynamic upgrades during SIL verification is scarce. This is partly because dynamic upgrades in SIL verification rely on uncertainty analysis, which is difficult to achieve. Secondly, current SIL verification data is based on expert experience or constant data from the international EXIDA database, which does not support dynamic updates. Therefore, achieving dynamic and updated SIL verification requires investigating a large amount of SIS failure data, using dynamic data processing algorithms to uncover data patterns, predict future trends, and thus dynamically update the SIL verification results. Furthermore, in safety instrumented systems, various instruments exhibit uncertainties in failure modes, testing, maintenance, and other aspects. For example, many uncertainties arise during instrument selection and use, ultimately leading to failure. Since safety instrumented systems generally have high reliability, these uncertainties significantly impact instrument failure rates, resulting in random uncertainty in failure rates. Another example is the design of testing intervals (T1) and mean recovery time (MTTR) during component testing and maintenance. The subjective interpretation of time data reading and recording by personnel leads to ambiguity and uncertainty. Summary of the Invention

[0006] The purpose of this invention is to propose a dynamic verification method for SIS security integrity level that takes into account uncertainty, so as to solve the uncertainty problem in the SIL verification process and the problem that static SIL verification results are lagging and cannot be represented in real time.

[0007] To achieve the above objectives, the present invention adopts the following technical solution:

[0008] The dynamic verification method for SIS security integrity levels considering uncertainties includes the following steps:

[0009] Step 1. Classify SIS parameters into three categories: random uncertainty parameters, fuzzy uncertainty parameters, and constant parameters; and collect time-series failure rate data based on field failure records, and integrate fuzzy parameter data by evaluating multiple experts.

[0010] Step 2. Build a stochastic uncertainty parameter prediction model based on LSTM network, and train the stochastic uncertainty parameter prediction model on time based on the collected time-series failure rate data; then use the trained stochastic uncertainty parameter prediction model to dynamically predict and update the stochastic uncertainty parameters.

[0011] Step 3. Based on the fuzzy parameter data obtained by integrating the evaluations of multiple experts, construct the relation matrix of the expert evaluation data, calculate the associated fuzzy frequency based on the non-negative matrix eigenvector theory, and select the optimal value of each fuzzy uncertainty parameter;

[0012] Step 4. Based on the random uncertainty parameters at each time point in Step 2, the fuzzy uncertainty parameters at each time point in Step 3, and the constant parameters at each time point obtained on-site, calculate the mean failure probability (PFD) of the SIS system when it meets the requirements. avg By combining the safety failure score (SFF), the timing-based SIL verification results are obtained.

[0013] Step 5. Combine the random forest algorithm to perform source analysis on the SIL validation results, evaluate the influence weight of the random uncertainty parameters of each subsystem in the SIS system on the SIL validation results, and generate feature importance ranking.

[0014] Furthermore, based on the dynamic verification method for SIS security integrity levels that considers uncertainty, this invention also proposes a corresponding dynamic verification system for SIS security integrity levels that considers uncertainty, which adopts the following technical solution:

[0015] A dynamic verification system for SIS security integrity levels considering uncertainties includes the following modules:

[0016] The parameter classification and data acquisition module is used to classify SIS parameters into three categories: random uncertainty parameters, fuzzy uncertainty parameters, and constant parameters; and to collect time-series failure rate data based on field failure records and integrate fuzzy parameter data by multiple expert evaluations.

[0017] The random uncertainty parameter data processing module is used to build a random uncertainty parameter prediction model based on LSTM network, and to train the random uncertainty parameter prediction model on time based on the collected time-series failure rate data; then, the trained random uncertainty parameter prediction model is used to dynamically predict and update the random uncertainty parameters.

[0018] The fuzzy uncertainty parameter data processing module is used to construct a relation matrix of expert evaluation data based on the fuzzy parameter data obtained by fusing multiple expert evaluations, calculate the associated fuzzy frequency based on the non-negative matrix eigenvector theory, and select the optimal value of each fuzzy uncertainty parameter.

[0019] The dynamic SIL verification module is used to calculate the average probability of failure (PFD) of the SIS system when it meets the requirements, based on random uncertainty parameters and fuzzy uncertainty parameters at various time points, combined with constant parameters obtained on-site. avg By combining the Security Failure SFF with the time-series SIL verification results, we obtain the SIL verification results corresponding to each time point.

[0020] The module also includes a source analysis module, which combines the random forest algorithm to perform source analysis on the SIL verification results, evaluate the influence weight of the random uncertainty parameters of each subsystem in the SIS system on the SIL verification results, and generate a feature importance ranking.

[0021] Furthermore, based on the dynamic verification method for SIS security integrity levels that takes into account uncertainty, this invention also proposes a computer device that includes a memory and one or more processors.

[0022] Executable code is stored in memory. When the processor executes the executable code, it is used to implement the aforementioned dynamic verification method for SIS security integrity levels that takes into account uncertainties.

[0023] Furthermore, based on the dynamic verification method for SIS security integrity levels that considers uncertainties, this invention also proposes a computer-readable storage medium on which a program is stored. When executed by a processor, this program is used to implement the aforementioned dynamic verification method for SIS security integrity levels that considers uncertainties.

[0024] The present invention has the following advantages:

[0025] As described above, this invention discloses a dynamic verification method for Safety Integrity Level (SIS) considering uncertainty. This method addresses the uncertainty issues in SIL verification by introducing stochastic uncertainty theory and fuzzy uncertainty theory. It combines stochastic uncertainty theory with LSTM networks and fuzzy uncertainty theory with the newly proposed Fuzzy-Logic Method (FLM), thus proposing a novel dynamic SIL verification method. Furthermore, it performs source tracing analysis on the SIL verification results and uses random forests for feature importance analysis. This invention solves the problems of uncertainty and static lag in traditional SIL verification, eliminates the fuzzy uncertainty of expert judgment and the static nature of field data, and enables dynamic and up-to-date final verification results. It transforms the uncertainty of data and parameters into data fusion and pattern prediction, which is beneficial for achieving intelligent monitoring and early warning of process safety in SIS systems under the major trend of Industrial Internet + safe production. Attached Figure Description

[0026] Figure 1 This is a flowchart of the dynamic verification method for SIS security integrity level considering uncertainties in an embodiment of the present invention;

[0027] Figure 2 A simplified P&ID diagram of the n-hexane buffer tank in this invention example;

[0028] Figure 3 This is a block diagram of the high-level interlock SIS in the hexane buffer tank in an example of the present invention;

[0029] Figure 4 This is a graph showing the failure rate of the SIS element in an example of the present invention, obtained through LSTM training and prediction.

[0030] Figure 4 (a) shows the results for the radar level gauge (LSH-95), (b) shows the results for the intermediate element (safety barrier), (c) shows the results for the logic controller, (d) shows the results for the solenoid valve, (e) shows the results for the shut-off valve (RBV), and (f) shows the results for the regulating valve (LV-90).

[0031] Figure 5 PFD of the three subsystems in the example of this invention avg Calculation results graph;

[0032] Figure 6 The diagram shows the SFF calculation results of the three subsystems in this invention example;

[0033] Figure 7 This is a diagram showing the SIL dynamic verification results of SIS in an example of the present invention;

[0034] Figure 8This is a graph showing the results of the feature importance analysis of the SIS system in the example of this invention;

[0035] Figure 8 (a) shows the importance of the SFF features of the sensing subsystem, (b) shows the importance of the SFF features of the logic control subsystem, (c) shows the importance of the SFF features of the execution subsystem, and (d) shows the PFD of the SIS system. avg Feature importance. Detailed Implementation

[0036] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments:

[0037] Example 1

[0038] Traditional SIS SIL verification methods typically have the following drawbacks:

[0039] 1. The validation results of the model are mostly fixed values ​​that remain unchanged within a SIL validation cycle (3-5 years). However, in reality, as time goes on, the components of the SIS will age or even fail. It cannot be guaranteed that the failure rate or other parameters will not change within a SIL validation cycle. Changes in parameter values ​​will lead to changes in the SIL validation results. Therefore, using a static SIL validation result within a SIL validation cycle is too idealistic and does not conform to reality.

[0040] 2. Current SIL verification work generally uses fixed failure rate data from databases (such as the Exida database). Although the data sample in the database is large and has high acceptance, for a specific SIS, its failure rate should reflect its own variation (such as bathtub curve description) rather than a fixed average value from a large sample of components from different manufacturers and different environments. Therefore, the failure rate should be a dynamic value that changes over time, rather than a fixed value.

[0041] 3. In the current SIL verification method, the Mean Time To Repair (MTTR) is set as a fixed value estimated by experts, which is highly subjective and a static value. In fact, MTTR is greatly affected by the field maintenance conditions and is a fluctuating value. It should be quantitatively calculated based on the start and end times in the maintenance forms in the database and set as a dynamic average value, which is more in line with the actual field situation and makes the SIL verification calculation results more accurate.

[0042] 4. Existing SIL validation work rarely considers dynamics. Even when dynamics are considered, it only uses a data-driven approach to statistically predict existing data without performing uncertainty analysis on the parameters of SIL validation and then selecting different data fusion methods based on different uncertainties. This easily leads to a "one-size-fits-all" approach to data, resulting in a mismatch between data and methods.

[0043] 5. Uncertainty analysis for SIL verification is almost non-existent. Without considering uncertainty analysis, SIL verification becomes highly dependent on expert experience and field data. If the expert evaluation is not objective or the field data is missing, it is difficult to conduct high-reliability SIL verification.

[0044] To address the problems of inconsistent and complex parameters in traditional SIL verification models, static and unupdated parameter data and verification results, and the inability to trace which parameter needs adjustment after obtaining SIL verification results, this embodiment 1 describes a dynamic verification method for SIS safety integrity level that considers uncertainty. This method improves the reliability of SIL verification work in guiding on-site safety production in real time, and enables the device to cut off SIS in time and provide early warning before a safety accident occurs.

[0045] The general process of this invention is as follows: 1) Classify the parameters of SIL verification into random and fuzzy uncertainties; 2) Process the data of random uncertain parameters; 3) Propose and process new methods for fuzzy uncertain parameters; 4) Dynamic SIL verification; 5) Source tracing analysis of dynamic SIL verification, and use random forest to analyze the importance of parameter features.

[0046] like Figure 1 As shown, the dynamic verification method for SIS security integrity level considering uncertainty includes the following steps:

[0047] Step 1. Classify SIS parameters into three categories: random uncertainty parameters, fuzzy uncertainty parameters, and constant parameters; and collect time-series failure rate data based on field failure records, and integrate fuzzy parameter data by evaluating multiple experts.

[0048] The SIS is structurally divided into three functionally independent subsystems: the sensor subsystem, the logic control subsystem, and the actuator subsystem. Each of these three subsystems contains dedicated components adapted to its function.

[0049] The sensor subsystem integrates key detection equipment (such as radar level transmitters, pressure sensors, and temperature sensors) and signal conditioning modules (such as intrinsically safe barriers and signal isolators) to accurately detect process parameters and convert them into standardized input signals (such as 4–20mA and digital protocols). The logic control subsystem encompasses a programmable logic controller (PLC), redundant voting units, and communication interfaces (such as HART and Modbus). It executes predefined safety logic (such as 1002 voting) to parse sensor inputs and generate execution instructions. It typically integrates diagnostic modules and fail-safe mechanisms (such as watchdog timers) to achieve real-time fault detection and functional integrity assurance. The actuator subsystem contains final control elements (such as solenoid valves, electric valves, and pneumatic / hydraulic actuators). It uses physical intervention to bring the process into a safe state, and redundant design and partial stroke test (PST) devices enhance reliability.

[0050] For each subsystem's specific failure mode, and in conjunction with the SIL verification process, the parameters involved in each subsystem are described below:

[0051] 1. The Beta parameter, abbreviated as β, is the undetectable common cause failure score in diagnostic testing. The calculation and data acquisition process is as follows: the proportion of common cause dangerous failures with common causes that have not been detected (1-DC, where DC is the diagnostic detection rate), which can be obtained by statistical field data.

[0052] 2. The common cause failure fraction (Beta-D) parameter, abbreviated as β, is detectable in diagnostic testing. D The calculation and data acquisition process is as follows: failure scores that have a common cause and have been detected by diagnostic tests can be obtained by statistically analyzing on-site data.

[0053] 3. The voting method within the group, abbreviated as MooN, involves the following calculation and data acquisition process: a fixed value selected manually during SIS setup is set and remains unchanged.

[0054] 4. Failure rate, abbreviated as λ, is calculated and acquired as follows: It is calculated based on the statistical number of component failures over a period of time. If a component fails, its failure rate will change. Depending on the failure mode of the component, it can be divided into λ... DD , λ DU , λ SD , λ SU The four types of failure rates will each participate in the calculation of SIL verification.

[0055] 5. Mean Time to Recovery (MTTR) is calculated and collected as follows: The expected value of recovery time is a measure of product maintainability. The MTTR should be statistically analyzed based on the actual field data in the initial data to obtain the real field data.

[0056] 6. The test interval, abbreviated as T1, is calculated and data acquired as follows: It is the time interval between tests performed to discover undetected faults in the safety instrumented system (SIL), i.e., the functional test interval. Functional testing aims to discover dangerous system failures. In current SIL verification work, T1 is generally given a fixed value based on previous test results, such as 48 months or 36 months.

[0057] 7. Inspection and testing coverage, abbreviated as PTC, is calculated and data collected as follows: the ratio of hazardous failures detected to potential hazardous failures in inspection and testing, determined based on the company's maintenance level records.

[0058] 8. Service life, abbreviated as T2, is calculated and data collected as follows: the specific value is given on the equipment certificate of the component.

[0059] As can be seen from the above, the voting method MooN and the service life T2 are values ​​that were determined in the early stage of SIS design and will not change or be uncertain during use. Therefore, the parameters MooN and T2 are determined to be constant parameters.

[0060] The calculation of five parameters—Beta, Beta-D, Mean Time to Recovery (MTTR), Test Interval (T1), and Test Coverage (PTC)—involves subjective uncertainty based on the work status and efficiency of on-site personnel, leading to ambiguity in the final results. Therefore, β and β-D are subject to certain subjective uncertainties. D The five parameters MTTR, T1, and PTC are determined as fuzzy uncertainty parameters.

[0061] The failure rate λ is considered a stochastic uncertainty parameter because components may fail randomly during use due to objective weather conditions or human error. Failure rates include four types, with the most detectable hazardous failure rate λ being... DD Undetectable dangerous failure rate λ DU Detectable safety failure rate λ SD Undetectable safety failure rate λ SU Therefore, these four types of failure rates are considered as random uncertainty parameters.

[0062] The stochastic uncertainty parameter of the SIS system is λ for each subsystem. DD , λ DU , λ SD , λ SU The fuzzy uncertainty parameters are β and β0 for each subsystem. D MTTR, T1, PTC, and constant parameters are MooN and T2 for each subsystem.

[0063] Based on the random uncertainty parameter failure rate (λ)DD , λ DU , λ SD , λ SU Based on the data characteristics, the LSTM method is used for training and prediction of various failure rate data. An implicit law function for failure rate is directly established and predicted using field data to eliminate the uncertainty caused by the incompatibility of using EXIDA data or the randomness of field failures. Failure rate data is sampled based on actual SIS field operation data and used for SIL verification, which can accurately reflect the safety integrity level of the SIS.

[0064] Data acquisition primarily utilizes failure and maintenance records from the SIS (System-on-a-Side) field, using time nodes as a foundation. The failure rate is recorded for each SIS component failure and subsequent maintenance, and the failure rate at each time node is collected into an array to form the dynamic failure rate. This dynamic failure rate is then used to train and predict the model based on the LSTM (Laser-Based Memory) method. The dynamic failure rate prediction model is continuously updated based on the increasing number of failure and maintenance records during field operation. As the time span of the dynamic failure rate data expands, the underlying data of the LSTM training model is continuously expanded and updated, resulting in more accurate failure rate predictions and achieving dynamic update prediction of the failure rate.

[0065] Specifically, in this embodiment, the random uncertainty parameters of the three functionally independent subsystems—the sensing subsystem, the logic control subsystem, and the execution subsystem—are of the same type, all being the failure rate λ. DD , λ DU , λ SD , λ SU After collecting failure and maintenance records for the three types of subsystems, the dynamic failure rate λ of the three types of subsystems was determined. DD , λ DU , λ SD , λ SU The calculation formula is uniformly expressed as:

[0066] λ i =n i / τ i (1)

[0067] Where i represents the time point in time when failure and maintenance records exist; λ i Let λ represent the failure rate of any class, i.e., the failure rate at the i-th time point. DD , λ DU , λ SD and λ SU The data are represented by λ. DDi , λ DUi , λ SDi , λ Sui Indicates; n i τ represents the number of component failures up to the i-th time point; iThis represents the cumulative service time of the element up to time node i.

[0068] Parameter λ i It is a follower of τ i The changing curve, λ, exists as long as there are failure and maintenance records. i The value will then change.

[0069] Step 2. Build a stochastic uncertainty parameter prediction model based on an LSTM network, and train the model on a time-series basis using collected temporal failure rate data. Then, use the trained model to dynamically predict and update the stochastic uncertainty parameters. The LSTM training data consists of historical data from each SIS element in each subsystem, which is used by the LSTM network after training to predict parameters for future time periods.

[0070] For each SIS element in any of the subsystems—sensing, logic control, and execution—a stochastic uncertainty parameter prediction model based on an LSTM network is built. Historical failure rate data at multiple consecutive time points for each SIS element, collected and calculated, are used as input data for the corresponding LSTM network to train it. The historical failure rate data at each time point includes four categories, namely λ... DD , λ DU , λ SD , λ SU After the LSTM network is trained, the trained stochastic uncertainty parameter prediction models based on the LSTM network are applied to the corresponding subsystems. Using the current time as a baseline, historical failure rate data from multiple consecutive time points prior to the current time are used as input to the trained stochastic uncertainty parameter prediction models to predict the λ of each SIS element in the corresponding subsystem. DD , λ DU , λ SD , λ SU The results at several points in the future.

[0071] Taking the radar level gauge (i.e., SIS element) in the sensing subsystem as an example, a stochastic uncertainty parameter prediction model based on LSTM network is first built for the radar level gauge. The historical failure rate data of the radar level gauge at multiple consecutive time nodes (e.g., 36 time nodes, representing 36 months) is collected and calculated as input data of LSTM network. LSTM network predicts the failure rate data of several time nodes (e.g., 4 time nodes, representing 4 months; of course, the number of predicted time nodes can be adjusted as needed). Then, the failure rate data predicted by LSTM network is compared with the actual historical failure rate data, and the LSTM network is optimized in reverse until the accuracy requirement is met or the maximum number of iterations is reached. This kind of LSTM network has the optimal network parameters, and the LSTM network training is completed.

[0072] After the LSTM network is trained, the trained stochastic uncertainty parameter prediction model for the radar level gauge is applied to the sensing subsystem to predict the failure rate data of the radar level gauge over a future period. Specifically, using the current time as a reference, the historical failure rate data from multiple consecutive time points (e.g., 36 time points) collected before the current time are used as input to the trained stochastic uncertainty parameter prediction model for the radar level gauge. This prediction yields the λ value of the radar level gauge in the sensing subsystem. DD , λ DU , λ SD , λ SU The results at several points in the future.

[0073] Based on the increasing failure and maintenance records during field operation, the prediction model is continuously updated. The time span of dynamic failure rate data is constantly expanding, and the underlying data of the LSTM network is continuously expanded and updated, resulting in more accurate failure rate predictions.

[0074] Step 3. Based on the fuzzy parameter data obtained by integrating the evaluations of multiple experts, construct the relation matrix of the expert evaluation data, calculate the associated fuzzy frequency based on the non-negative matrix eigenvector theory, and select the optimal value for each fuzzy uncertainty parameter.

[0075] In the SIL verification calculation, the β, β' of the SIS elements of each subsystem in the SIS system are... DThe five parameters—MTTR, T1, PTC, and T2—were identified as fuzzy uncertainty parameters. Due to the fuzziness in the data acquisition process for these five parameters, this invention employs a newly proposed Fuzzy-Logic Method (FLM) for data fusion calculation. Based on field data, several experts from the SIL verification working group calculated and evaluated the values ​​of the five parameters for each of the three subsystems of each SIS, obtaining their respective parameter results. These parameter results serve as the foundation for subsequent FLM data fusion.

[0076] Assume β, β D The true value of a certain fuzzy uncertainty parameter among MTTR, T1, and PTC is x. t Furthermore, we have obtained evaluation data from n experts, where the evaluation data of the s-th expert is x. s The corresponding variance is σ s .

[0077] Define fuzzy support distance, whose physical meaning is the mutual support among expert-assessed data; where x s With x j The formula for calculating the fuzzy support distance between them is shown in Equation (2).

[0078]

[0079] Where, x j Let s,j∈[1,n] represent the j-th expert evaluation data; d sj For x s With x j The fuzzy support distance between them, d js For x j With x s The fuzzy support distance between them; σ j Let be the variance corresponding to the j-th expert evaluation data.

[0080] definition or The formula for calculating the error function fr(z) is as follows:

[0081]

[0082] Define the fuzzy frequency membership function as follows:

[0083] h sj =h js =d sj ×(d max -d min ) -1 (4)

[0084] In the formula, h sj and hjs For fuzzy frequencies, d max =max{d sj}, d min =min{d sj}, h sj ∈[0,1].

[0085] h sj The larger the value, the closer it is to the actual state, and the less ambiguous it is. When d sj =max(d sj When ), then d is considered to be s With d j They do not support each other at all, at this time h sj =0; at the same time, d ss =0,h ss =1.

[0086] Based on the fuzzy frequency h of the data evaluated by various experts sj The relationship matrix H of the expert evaluation data is constructed as follows:

[0087]

[0088] To represent the expert evaluation data x s To emphasize the importance of correlation fuzzy frequency p in FLM data fusion, this invention introduces the associated fuzzy frequency p. s , by h s1 ,h s2 ,…,h sn This is reflected in the matrix R. p is solved based on the matrix R. s Given a set of nonnegative numbers That is, q1, q2, ..., q n , so that:

[0089]

[0090] Equation (6) can be rewritten in matrix form as U = HV, where H is a non-negative symmetric matrix.

[0091] Where, U=[p1,p2,…,p n ] T V = [q1, q2, ..., q n ] T .

[0092] As can be seen from the matrix properties, there exists a matrix H with a maximum eigenvalue γ by modulus and its corresponding eigenvector, and all eigenvectors are positive, such that γV = HV. Therefore, γV can be used as a measure of the correlation fuzziness frequency of expert evaluation data, where γ > 0.

[0093] Considering q i Constraints The correlation fuzzy frequency of the s-th expert evaluation data is:

[0094]

[0095] The power method is used to find the modulo-maximal eigenvalue and its corresponding eigenvector of matrix H.

[0096] The sequence constructed in the k-th iteration:

[0097]

[0098] Among them, γ1, γ2, γ n This represents the maximum eigenvalue of the s-th expert evaluation data by modulus. Indicates an intermediate variable.

[0099] The correlation fuzzy frequency of several expert evaluation data is calculated. By comparing and optimizing the correlation fuzzy frequency, the maximum value among all correlation fuzzy frequencies is taken as the optimal value of each fuzzy uncertainty parameter.

[0100] Step 4. Based on the random uncertainty parameters at each time point in Step 2, the fuzzy uncertainty parameters at each time point in Step 3, and the constant parameters at each time point obtained on-site, calculate the Average Probability of Failure on Demand (PFD) of the SIS system at the required time. avg The SIL verification results are obtained by combining the Safety Failure Fraction (SFF) with the time-series SIL verification results, i.e., the SIL verification results at each corresponding time point.

[0101] For random uncertainty parameters, the final parameter values ​​are a time-series sequence, while for fuzzy uncertainty parameters and constant parameters, the final parameter values ​​are single values. Therefore, if the parameter values ​​of random uncertainty parameters at different time points are used as independent parameters, then time-series SIL verification results can be obtained, that is, dynamically predicted SIL verification results.

[0102] Specifically, based on historical data of the random uncertainty parameters of each SIS element at multiple time points, predicted data of several time points after the current time, single-value data of the fuzzy uncertainty parameters of each SIS element, and constant parameters of each subsystem of the SIS determined by the basic information, the SIL verification calculation is performed on the entire SIS.

[0103] The SIL dynamic verification process includes the following steps:

[0104] I. A reliability block diagram (PFD) is used to construct the three subsystems: the sensing subsystem, the logic control subsystem, and the execution subsystem.avg The PFD of the SIS system is calculated separately. avg PFD for three subsystems avg sum.

[0105] Taking the MooN of the sensing subsystem as a loo1 structure as an example, the PFD is calculated using formula (10). avg1 :

[0106]

[0107] Taking the MooN of the logic control subsystem as an example with a 1oo1 structure, the PFD is calculated using formula (11). avg2 :

[0108]

[0109] Taking the execution subsystem with a MooN of 1oo2 structure as an example, PFD is calculated using formulas (12) to (14). avg3 :

[0110]

[0111] Among them, the PFD of the SIS system avg PFD for three subsystems avg1 PFD avg2 PFD avg3 sum.

[0112] II. Structural Constraint Calculation. The safety integrity of structural constraints is jointly determined by the instrument type, the safety failure fraction (SFF), and the hardware failure margin (HFT); the safety failure fraction is calculated using the following formula:

[0113]

[0114] The SFFs of the sensing subsystem, logic control subsystem, and execution subsystem are calculated using formula (15).

[0115] Hardware fault margin HFT = NM. When HFT ≥ 2, HFT is 2.

[0116] Where N and M are the voting forms MooN within each group; based on the structural constraint table of Class A / B safety-related subsystems in IEC61508, the SIL of each subsystem's structural constraint determination is obtained by combining SFF and HFT.

[0117] Among them, the SIL determined by the structural constraints of SIS is the minimum value of the structural constraint results of the three subsystems.

[0118] III. Dynamic SIL Verification Results. Determining the SIL level based on the PFD value:

[0119] When PFD avg If ∈[0,1,1), then it is SIL0; when PFD avg If ∈ [0.01, 0.1), then it is SIL1; when PFD avg If ∈ [0.001, 0.01), then it is SIL2; when PFD avg If ∈ [0.0001, 0.001), then it is SIL3; when PFD avg If it is less than 0.0001, it is SIL4.

[0120] The final SIL of the SIS system is PFD. avg The smaller of the SIL determined by the structural constraints and the SIL determined by the structural constraints.

[0121] During the calculation, the data of the fuzzy uncertainty parameter and the constant parameter corresponding to the data of each random uncertainty parameter at each time node are the same constant, so the dynamic SIL verification results of multiple time nodes can be calculated.

[0122] Step 5. Combine the random forest algorithm to perform source analysis on the SIL validation results, evaluate the influence weight of the random uncertainty parameters of each subsystem in the SIS system on the SIL validation results, and generate feature importance ranking.

[0123] Specifically, based on the SIL verification results, the feature importance analysis of the parameters is performed using the random forest algorithm to analyze which parameters are closely related to the current SIL verification results, thereby providing data support for the preventive maintenance of SIS.

[0124] The calculation of dynamic SIL verification involves multiple parameters, and changes in different parameters will have varying degrees of impact on the results of dynamic SIL verification. To better quantify the impact of different parameter changes on dynamic SIL verification, this invention employs the random forest algorithm to calculate the feature importance of changing parameters, thereby better understanding the impact of parameter data changes and directly controlling the dynamic verification results. For feature importance analysis, only parameters whose data can change can be analyzed; if the parameter data is constant, then the data does not change, violating the premise for performing feature importance analysis.

[0125] Therefore, this embodiment only performs feature importance analysis on the random uncertainty parameters of SIS elements in each subsystem, and analyzes the impact of parameter changes on SIL verification results based on the random forest algorithm and sorts them by importance.

[0126] To rank the importance of SFFs for each subsystem, input historical data of the random uncertainty parameters of the SIS elements in the corresponding subsystem and the SFF values. This will yield the ranking of the importance of SFFs for each subsystem.

[0127] PFD for each subsystem avg Importance ranking, input historical data of random uncertainty parameters of SIS elements in the corresponding subsystem and PFD avg The value obtained is the PFD for each subsystem. avg Ranking by importance.

[0128] For the importance ranking of SIS systems, the input consists of historical data on the random uncertainty parameters of SIS components in all subsystems and the PFD of the SIS system. avg The result is a PFD for SIS systems. avg Ranking by importance.

[0129] Random forest (RF) is an ensemble machine learning method that uses bootstrap random resampling and random node splitting techniques to construct multiple decision trees, and then obtains the final classification result through voting. As a decision tree-based ensemble learning algorithm, random forest can handle regression tasks with continuous data well, providing an interpretable global parameter importance measure for each parameter. It can be directly implemented using computer languages ​​such as MATLAB.

[0130] The importance of a feature is determined by randomly changing its value and calculating the out-of-bag error of the test dataset before and after the change. Assuming there are N trees in the random forest, the feature importance I... RF The calculation formula is as follows:

[0131]

[0132] In the formula, N w errOOB2 is the number of times feature w appears in the random forest, errOOB2 is the out-of-bag error calculated again after adding noise interference to the feature values ​​in N decision trees, and errOOB1 is the out-of-bag error calculated from the out-of-bag data corresponding to the feature values ​​of N decision trees.

[0133] Step 6. Dynamically update the SIL level.

[0134] As production data is updated in real time, the supplementary data is processed in steps 2 to 5, continuously expanding the data storage file and making the fitted prediction results closer to reality. This achieves dynamic updates and continuous corrections, enabling intelligent data management and dynamic updates, predictions, and early warnings for SIL verification.

[0135] This invention establishes a traceability mechanism for dynamic SIL verification, providing real-time values ​​for the characteristic importance of parameters, analyzing and obtaining SIS parameters with high importance under the current SIL verification results in real time, and adjusting the SIS operation based on these parameters to continuously improve the reliability and security of the SIS system; and issuing early warnings if the predicted SIL verification results at a certain time point fail to meet the requirements, or if the SIS parameter with the highest importance in the SIL verification results is identified.

[0136] Furthermore, to verify the effectiveness of the method proposed in this invention, the following specific examples are provided:

[0137] Hexane from the upstream process unit enters the hexane buffer tank T-101. This tank typically holds a large quantity of hexane; if an overflow occurs, the fire dike cannot contain the spilled material, leading to widespread hexane diffusion and potential fire, damage, and death. Therefore, a separate SIF (Self-Isolating Flood Filter) is installed to detect and prevent overflows, as shown in the attached diagram. Figure 2 As shown.

[0138] Based on LOPA analysis, a SIF loop was added to the n-hexane storage tank: the high-high level (LSH-95) interlock shut-off feed valve RBV / LV-90 (1oo2) of the n-hexane buffer tank T101. The required SIL level for this SIF loop is SIL2. The block diagram of the high-high level interlock SIS for the n-hexane buffer tank is attached. Figure 3 As shown in Table 1, the specific components of the three subsystems are as follows.

[0139] Table 1. Specific components of the n-hexane buffer tank high-level interlock SIS.

[0140]

[0141] Step 1. Parameter classification and data collection.

[0142] SIS parameters are divided into three categories: random uncertainty parameters, fuzzy uncertainty parameters, and constant parameters; temporal failure rate data are collected based on field failure records, and fuzzy parameter data are obtained based on expert evaluation.

[0143] The parameters of the SIS are classified using the classification method detailed in step 1 above. Then, time-series data are collected for the random uncertainty parameters of all SIS elements, and training and prediction are performed based on the LSTM algorithm.

[0144] Because SIS instrument components have high reliability and few failures, and the amount of data accumulated in the field is insufficient, failure data of similar components in the field are used as supplementary data to uniformly statistically analyze the failures of instrument components, serving as the basis for training and prediction. The fuzzy uncertainty parameters of all components are verified by SIL (System Indeterminate) and evaluated by experts. The expert evaluation data is then fused using the FLM (Fluorescent Least Mean) method. Simultaneously, constant parameter data for all components are also collected. Taking the radar level gauge (LSH-95) as an example, its random uncertainty parameter data (36 months) is shown in Table 2, and the fuzzy uncertainty data is shown in Table 3.

[0145] Table 2. Random uncertainty parameter data for radar level gauge (LSH-95)

[0146]

[0147] Table 3. Expert Data Collection of Fuzzy Uncertainty Parameters

[0148]

[0149] Step 2. Data processing of random uncertainty parameters. In this example, an LSTM network is used to perform time-series training and prediction of random uncertainty parameters, including model training and dynamic prediction updates.

[0150] The batch size of the LSTM network was set to 8 to balance computational efficiency and gradient stability. Stochastic gradient descent with momentum (SGDM) was used as the optimizer (initial learning rate 0.001, momentum 0.9), and a total of 1024 training epochs were performed. The network architecture contains a single hidden layer (128 units), where the Sigmoid function controls the gating operations (input gate, forget gate, output gate), the Tanh function regulates cell state updates, and the output layer uses a linear activation function to predict the successive failure rate (output dimension is 1). To ensure numerical stability, gradient clipping (threshold = 100) was used during backpropagation to prevent gradient explosion. If the validation loss tended to flatten, the learning rate was dynamically adjusted using a decay factor of 0.1, and convergence was monitored based on a threshold method (training was terminated if the loss improvement was less than 1e-5 after 50 consecutive epochs). Simultaneously, to achieve more accurate prediction results, the commonly used input dim configuration was optimized. An input dim of 3 resulted in the smallest error and the best fitting effect; therefore, an input dim parameter of 3 was used in this model.

[0151] Historical data on the four types of failure rates of this SIS element were collected and calculated. Prediction was based on LSTM training. The historical data consisted of data from the previous three years, i.e., 36 months (36 time points in monthly units). After training, this data was used to predict the failure rate for the next four time points (i.e., four months) starting from the current moment. The results are attached. Figure 4 As shown.

[0152] Figure 4 In the diagram, hollow dots represent historical data, while solid dots represent LSTM prediction data.

[0153] According to the LSTM training prediction results, the failure rates of the six types of components, except for the failure modes for which there is no failure data, all show an increasing trend to varying degrees, but the increasing trend and degree of increase are different for each component.

[0154] Step 3. Processing of fuzzy uncertainty parameter data.

[0155] β, β of each subsystem D The five parameters MTTR, T1, and PTC were determined to be fuzzy uncertainty parameters. Based on field data, the experts of the SIL verification working group evaluated the values ​​of the five parameters of each of the three subsystems of each SIS and obtained their respective parameter results. Finally, the FLM algorithm was used to calculate each parameter to obtain the final parameter data fusion result.

[0156] The evaluation results of the fuzzy uncertainty parameters of the SIS element by the SIL verification experts are shown in Table 4.

[0157] Table 4. Expert Evaluation Results of Fuzzy Uncertainty Parameter SIL Verification

[0158]

[0159] The data in Table 4 are normalized and the confidence distance and relation matrix H are calculated. The eigenvector V corresponding to the largest eigenvalue of the relation matrix H by modulus is obtained. Finally, the comprehensive fuzzy support matrix U between the expert evaluation data of each SIS element is calculated, and the comprehensive fuzzy support u is selected. imax The data constitute the optimal values ​​of the fuzzy uncertain parameters, as shown in Table 5.

[0160] Table 5. Overall fuzzy support degree u of each SIS element imax data

[0161]

[0162] The calculation results in Table 5 are the optimal values ​​for the six types of components considering fuzzy uncertainty.

[0163] Step 4. Dynamic SIL verification.

[0164] The MooN and T2 of each subsystem of this SIS have been determined by the basic information, see appendix. Figure 3Based on Table 1, and using 40 data points (including 36 historical data points and 4 predicted data points) for each random uncertainty parameter obtained in step 2, and single-valued data (constants) for each fuzzy uncertainty parameter obtained in step 3, the SIL verification calculation is performed on the entire SIS.

[0165] During the calculation, the data of the fuzzy uncertainty parameter and the constant parameter data corresponding to the 40 data points of each random uncertainty parameter are all the same constant, so the dynamic SIL verification results at 40 time points can be calculated.

[0166] I.PFD avg calculate.

[0167] This case study uses the reliability block diagram method for PFD. avg Calculation. Select the corresponding calculation formula based on the structure of each subsystem.

[0168] The sensing subsystem MooN has a 1oo1 structure and adopts formula (10); the logic control subsystem MooN has a 1oo1 structure and adopts formula (11); the execution subsystem MooN has a 1oo2 structure and adopts formulas (12) to (14).

[0169] Calculations show that the PFDs of the sensing subsystem, logic control subsystem, and execution subsystem... avg As attached Figure 5 As shown. PFD of SIS system. avg PFD consists of three subsystems: sensing subsystem, logic control subsystem, and execution subsystem. avg sum.

[0170] II. Structural constraint calculation.

[0171] The safety integrity of structural constraints is determined by the instrument type, the safety failure fraction (SFF), and the hardware fault tolerance (HFT). The safety failure fraction is calculated using equation (15). The SFFs for the sensing element, logic controller, and actuator are calculated separately, as shown in the appendix. Figure 6 As shown.

[0172] Hardware fault margin HFT = NM. When HFT ≥ 2, HFT is 2.

[0173] Where N and M represent the voting form MooN within each group. Based on the structural constraint table for Class A / B safety-related subsystems in IEC 61508, the SIL determined by the structural constraints of each subsystem can be obtained by combining SFF and HFT. The SIL determined by the structural constraints of the SIS is the minimum value of the structural constraint results for the sensing subsystem, logic control subsystem, and execution subsystem.

[0174] III. Dynamic SIL verification results.

[0175] Determining the SIL level based on the PFD value:

[0176] When PFD avg If ∈[0,1,1), then it is SIL0; when PFD avg If ∈ [0.01, 0.1), then it is SIL1; when PFD avg If ∈ [0.001, 0.01), then it is SIL2; when PFD avg If ∈ [0.0001, 0.001), then it is SIL3, when PFD avg If it is less than 0.0001, it is SIL4.

[0177] The final SIL of the SIS system is PFD. avg The smaller of the SIL determined by the structural constraints and the SIL determined by the structural constraints. (See attached...) Figure 7 As shown in the figure. According to the dynamic verification results of this SIS, the final verification level of the SIS is SIL2, and the verification meets the requirements.

[0178] Step 5. Source tracing analysis.

[0179] By combining the random forest algorithm to perform source analysis on the SIL validation results, we can evaluate the influence weight of the random uncertainty parameters of each subsystem in the SIS system on the SIL validation results and generate a feature importance ranking.

[0180] The random forest model is configured with 100 decision trees, and robustness is improved through ensemble averaging. Each tree is trained on bootstrap sampled data with replacement to ensure diversity. The tree depth is limited to a maximum of 10 layers to prevent overfitting, while retaining the ability to capture key interactions and allowing a minimum of 1 leaf to accommodate high-dimensional Security Integrity Level (SIL) data.

[0181] When a node splits, each tree randomly selects... There are 24 total features (n = 24 features). The Gini impurity criterion is used for classification tasks, and the mean squared error (MSE) is used for regression tasks. Regularization balances model complexity and generalization ability by requiring a minimum of 5 split samples in internal nodes and limiting the maximum number of features per tree to 12 (50% of the total features).

[0182] Out-of-bag error (errOOB) estimation is used to evaluate generalization performance without requiring an independent validation set.

[0183] Based on the random forest method, PFD of the three subsystems SFF and SIS was performed using MATLAB. avg Feature importance analysis was performed, and the results are attached. Figure 8 As shown. Figure 8 λ XXi The 'i' refers to the serial number of the element in the table.

[0184] According to the appendix Figure 8 It can be seen that the feature importance of the SFF feature parameters of the sensing subsystem, ordered from largest to smallest, is: λ DD1 >λ DU2 >λ SU1 >λ DD2 >λ SD1 >λ DU1 The feature importance of the SFF feature parameters of the logic control subsystem, ranked from largest to smallest, is as follows: λ DU3 >λ SD3 >λ SU3 >λ DD3 The feature importance of the SFF feature parameters of the execution subsystem, sorted from largest to smallest, is as follows: λ DU6 >λ DU4 >λ SU4 >λ DU5 >λ SU5 SIS system PFD avg The feature importance of the feature parameters, sorted from largest to smallest, is as follows: λ SU4 >λ DU4 >λ SU1 >λ DU6 >λ DU1 >λ DD2 >λ DU2 >λ DD3 >λ SU3 >λ SD3 >λ DD1 >λ SD1 >λ DU5 >λ DU3 >λ SU5 .

[0185] In an SIS system, the sensing subsystem, logic control subsystem, execution subsystem, and the SIS system can all pay different levels of attention to the data of parameters based on the characteristic importance of different parameters. Parameters with high characteristic importance should be given special control.

[0186] Step 5. Visualization and early warning of SIL dynamic updates.

[0187] PFD for real-time computing SIS systemsavg Along with SFF, the SIL level is dynamically updated, and warning signals are output through a visual interface. As production data is updated in real time, the same operations in steps 2 and 5 are performed on the supplementary data, continuously expanding the data volume of the data storage file, making the fitted prediction results closer to reality, achieving the effect of dynamic updating and continuous correction, realizing intelligent data management and dynamic updating, prediction and early warning of SIL verification.

[0188] From the appendix Figure 5 It can be seen that the three subsystems PFD avg The trends are all upward, but the degree of increase varies. The execution subsystem saw the fastest growth, followed by the sensing subsystem, and finally the logic control subsystem. The sensing subsystem's PFD performance over three years... avg It increased by 1.63 times, compared to the predicted four-month PFD. avg The final result can reach 9.30×10 -4 It is the highest among the three subsystems. The logic control subsystem has a three-year timeframe for PFD. avg It increased by 1.16 times, compared to the predicted four-month PFD. avg The final result can reach 6.57×10 -4 It is PFD in three subsystems avg The slowest developing, and ultimately the PFD avg This is also the lowest, which to some extent reflects that the reliability of the logic control subsystem is relatively high and its stability is relatively good. The execution subsystem PFD has been running for three years. avg It increased by 2.78 times, compared to the predicted four-month PFD. avg The final result can reach 8.88×10 -4 It is PFD in three subsystems avg The fastest-growing subsystem indicates that the execution subsystem experiences higher daily wear and tear or a faster rate of failure compared to other subsystems. This results in the execution subsystem, which was originally a PFD (Programmable Component Optimizer) among the three subsystems. avg The smallest can easily become the largest of the three subsystems. This means that field maintenance personnel need to pay attention to and strengthen the maintenance of the execution subsystem to prevent it from exceeding 10 subsystems in a short time if this trend continues. -3 The magnitude of the scale is a significant issue that cannot be ignored in terms of the overall security and reliability of SIS.

[0189] From the appendix Figure 6It can be seen that the SFF (Signal Frame Failure) of the sensing subsystem and the logic control subsystem are relatively similar, but the difference with the execution subsystem is significant. The final predicted SFF values ​​for the sensing subsystem and the logic control subsystem are 0.9775 and 0.9914, respectively. This indicates that the proportion of undetectable hazardous failures in these two subsystems is very low. Therefore, apart from these types of failures, the remaining failures are all detectable or safe failures, which are controllable or acceptable for the system. This means that the field has a high level of understanding, detection, and handling of failure modes in these two subsystems, making them safer than the execution subsystem. Among the two subsystems, the logic control subsystem is safer. For the execution subsystem, the predicted SFF is 0.4589, less than half that of the other two subsystems, indicating a relatively low SFF level. Furthermore, its SFF trend is gradually decreasing, indicating that undetectable hazardous failures develop relatively quickly. The field's ability to control and maintain undetectable hazardous failures in the execution system is not high, which poses a significant hidden danger to the overall safety of the system. Field intervention, management, and maintenance control of these failure modes in the execution subsystem are needed to avoid dangerous consequences.

[0190] The SIL validation results show that the overall PFD avg The predicted value reached 2.48 × 10⁻⁶. -3 It increased 1.71 times within three years. This was driven by the total PFD. avg The determined SIL level and the SIL level determined by structural constraints are both SIL2. The prediction results show that the PFD of this SIS will be low in the short term. avg There will be no change in order of magnitude, i.e., the total PFD. avg The determined SIL level is unlikely to change in the short term. Furthermore, the SIL verification level determined by structural constraints, based on the SFF trends of the three subsystems, is also unlikely to change significantly. Therefore, the SIL verification level of this SIS is unlikely to change in the short term.

[0191] After tracing the source, the primary parameter affecting the SFF of the sensing subsystem is λ. DD1 The primary parameter affecting the SFF logic control subsystem is λ. DU3 The primary parameter affecting the execution subsystem SFF is λ. DU6 The impact of PFD on SIS systems avg The primary parameter is λ SU4 In response to these parameter warnings, corresponding preventative maintenance strategies should be implemented to improve the SIL level.

[0192] Example 2

[0193] This embodiment 2 describes a dynamic verification system for SIS security integrity level considering uncertainty, which is based on the same inventive concept as the dynamic verification method for SIS security integrity level considering uncertainty described in embodiment 1 above.

[0194] A dynamic verification system for SIS security integrity levels considering uncertainties includes the following modules:

[0195] The parameter classification and data acquisition module is used to classify SIS parameters into three categories: random uncertainty parameters, fuzzy uncertainty parameters, and constant parameters; and to collect time-series failure rate data based on field failure records and integrate fuzzy parameter data by multiple expert evaluations.

[0196] The random uncertainty parameter data processing module is used to build a random uncertainty parameter prediction model based on LSTM network, and to train the random uncertainty parameter prediction model on time based on the collected time-series failure rate data; then, the trained random uncertainty parameter prediction model is used to dynamically predict and update the random uncertainty parameters.

[0197] The fuzzy uncertainty parameter data processing module is used to construct a relation matrix of expert evaluation data based on the fuzzy parameter data obtained by fusing multiple expert evaluations, calculate the associated fuzzy frequency based on the non-negative matrix eigenvector theory, and select the optimal value of each fuzzy uncertainty parameter.

[0198] The dynamic SIL verification module is used to calculate the average probability of failure (PFD) of the SIS system when it meets the requirements, based on random uncertainty parameters and fuzzy uncertainty parameters at various time points, combined with constant parameters obtained on-site. avg By combining the Security Failure SFF with the time-series SIL verification results, we obtain the SIL verification results corresponding to each time point.

[0199] The module also includes a source analysis module, which combines the random forest algorithm to perform source analysis on the SIL verification results, evaluate the influence weight of the random uncertainty parameters of each subsystem in the SIS system on the SIL verification results, and generate a feature importance ranking.

[0200] It should be noted that the implementation process of the functions and roles of each functional module in the system of this embodiment 2 is detailed in the implementation process of the corresponding steps of the dynamic verification method for SIS security integrity level considering uncertainty in the above embodiment 1, and will not be repeated here.

[0201] Example 3

[0202] This embodiment 3 describes a computer device. The computer device includes a memory and one or more processors. Executable code is stored in the memory. When the processor executes the executable code, it implements the steps of the SIS security integrity level dynamic verification method considering uncertainties described in embodiment 1 above.

[0203] In this embodiment, the computer device can be any device or apparatus with data processing capabilities, and will not be described in detail here.

[0204] Example 4

[0205] This embodiment 4 describes a computer-readable storage medium storing a program that, when executed by a processor, is used to implement the steps of the dynamic verification method for SIS security integrity level considering uncertainty in embodiment 1 above.

[0206] The computer-readable storage medium can be an internal storage unit of any device or apparatus with data processing capabilities, such as a hard disk or memory, or an external storage device of any device with data processing capabilities, such as a plug-in hard disk, smart media card (SMC), SD card, flash card, etc.

[0207] Compared to traditional SIL verification methods for SIS (System-Independent Equipment), this invention firstly addresses the uncertainty of SIL verification parameter data, dynamically monitors and updates SIL verification results, grasps the dynamic trends of SIL verification results, and makes timely adjustments to SIS operating conditions to ensure safe and stable SIS production. Secondly, dynamically predicted SIL verification results can identify SIS anomalies in advance, providing early warnings and alarms, effectively preventing safety accidents caused by SIS anomalies. Thirdly, dynamic SIL verification enables real-time updating and collection of production data, transmitting production information and facilitating intelligent data management and application in the era of intelligent manufacturing. Fourthly, considering the dynamic changing trends of uncertain SIL verification results, it helps enterprises to focus on key variables, forming a standardized SIS anomaly handling process, enabling timely response and problem-solving of SIS anomalies, and ensuring the safe and stable operation of SIS and equipment. Fifthly, the dynamic SIL verification method and system can reduce SIS downtime. Through predictive maintenance with human-machine interaction, anomalies can be identified and prevented before problems occur, reducing maintenance costs and economic losses from downtime. Sixth, the dynamic SIL verification system can automatically update calculations after a verification operation, eliminating the need for manual verification by experts every few years, thus greatly saving manpower, material resources, and financial resources. The method and system of this invention can be used for functional safety analysis and risk warning of petrochemical plant equipment and SIS systems, dynamically monitoring SIL verification results and issuing warnings. This effectively solves the uncertainty problem in the SIL verification process, as well as the problem of the lag and inability to represent static SIL verification results in real time, achieving closed-loop management of SIS. This ensures that the SIS meets requirements and functions effectively when needed, truly achieving timely forecasting, preventative maintenance, and suppression of safety accidents.

[0208] Of course, the above description is only a preferred embodiment of the present invention. The present invention is not limited to the above-described embodiments. It should be noted that any equivalent substitutions or obvious modifications made by those skilled in the art under the guidance of this specification fall within the scope of this specification and should be protected by the present invention.

Claims

1. A dynamic verification method for SIS security integrity levels considering uncertainty, characterized in that, Includes the following steps: Step 1. Classify SIS parameters into three categories: random uncertainty parameters, fuzzy uncertainty parameters, and constant parameters; and collect time-series failure rate data based on field failure records, and integrate fuzzy parameter data by evaluating multiple experts. Step 2. Build a stochastic uncertainty parameter prediction model based on LSTM network, and train the stochastic uncertainty parameter prediction model on time based on the collected time-series failure rate data; then use the trained stochastic uncertainty parameter prediction model to dynamically predict and update the stochastic uncertainty parameters. Step 3. Based on the fuzzy parameter data obtained by integrating the evaluations of multiple experts, construct the relation matrix of the expert evaluation data, calculate the associated fuzzy frequency based on the non-negative matrix eigenvector theory, and select the optimal value of each fuzzy uncertainty parameter; The process for handling the fuzzy uncertainty parameters of each SIS element in each subsystem is as follows: Suppose the true value of a certain fuzzy uncertainty parameter for each SIS element in each subsystem is And it has already been obtained. The evaluation data of the experts, of which the first The expert evaluation data is as follows The corresponding variance is ; The fuzzy support distance is defined, and its physical meaning is the mutual support between expert-assessed data; among which... and The formula for calculating the fuzzy support distance between them is shown below; ； in, Indicates the first Data from expert evaluations ; for and Fuzzy support distance between them for and The fuzzy support distance between them; For the first The variance corresponding to the expert evaluation data; definition ,or Then the error function The calculation formula is as follows: ； Define the fuzzy frequency membership function as follows: In the formula, and For fuzzy frequencies, , , ; Based on the fuzziness frequency of the data evaluated by various experts A relational matrix of expert evaluation data was constructed. : ； Introducing related fuzzy frequencies To represent expert evaluation data The importance of FLM data fusion, according to the matrix Solve Given a set of nonnegative numbers Right now , so that: ,and ; Rewrite the above equation in matrix form. ,in It is a non-negative symmetric matrix; in, , ; According to the properties of matrices, there exists a matrix The largest eigenvalue by modulus And its corresponding eigenvectors, and all eigenvectors are positive, such that Therefore, using As a measure of the frequency of fuzzy associations in expert-assessed data, ; Considering Constraints , No. The frequency of fuzzy associations among the expert evaluation data is: ； Solving matrices using the power method The modulo-maximal eigenvalue and its corresponding eigenvector; After calculating the correlation fuzzy frequency of several expert evaluation data, the maximum value among all correlation fuzzy frequencies is selected as the optimal value of each fuzzy uncertainty parameter by comparing and optimizing each correlation fuzzy frequency. Step 4. Based on the random uncertainty parameters at each time point in Step 2, the fuzzy uncertainty parameters at each time point in Step 3, and the constant parameters at each time point obtained on-site, calculate the average failure probability of the SIS system when it meets the requirements. By combining the safety failure score (SFF), the timing-based SIL verification results are obtained. Step 5. Use the random forest algorithm to perform source analysis on the SIL validation results to evaluate the influence weight of the random uncertainty parameters of each subsystem in the SIS system on the SIL validation results and generate a feature importance ranking.

2. The dynamic verification method for SIS security integrity level considering uncertainty according to claim 1, characterized in that, In step 1, the SIS system is structurally divided into a sensing subsystem, a logic control subsystem, and an execution subsystem. Each subsystem has three types of parameters: random uncertainty parameters, fuzzy uncertainty parameters, and constant parameters. The random uncertainty parameter is the failure rate. This includes detectable hazardous failure rates. Undetectable dangerous failure rate Detectable safety failure rate And undetectable security failure rate Four failure rate parameters; Fuzzy uncertainty parameters include the fraction of common cause failures that are undetectable in diagnostic tests. Common cause failures detectable in diagnostic tests Average recovery time Test interval Inspection and testing coverage Five parameters; constant parameters include the voting method within the group. and service life Two parameters.

3. The dynamic verification method for SIS security integrity level considering uncertainty according to claim 2, characterized in that, In step 1, the process of collecting time-series failure rate data based on on-site failure records is as follows: Collect failure and maintenance records of SIS in the field, and use time points as a basis to analyze the SIS components in each subsystem. , , as well as Historical data on four types of failure rates were collected; after collecting failure and maintenance records for three types of subsystems, the dynamic failure rate of each SIS component was determined. , , , The unified representation is: ; in, The time points at which failure and maintenance records exist; Refers to any type of failure rate, i.e., the first Each time point , , as well as Data, respectively using , , , express; As of the end of the The number of failures of SIS components at each time point; As of the deadline The cumulative service time of time-based SIS node components; parameter It is a casual The changing curve, as long as there are failure and maintenance records, The value will then change.

4. The dynamic verification method for SIS security integrity level considering uncertainty according to claim 3, characterized in that, In step 2, the process of training and dynamically updating the random uncertainty parameter prediction model is as follows: For each SIS element in any of the subsystems—sensing, logic control, and execution—a stochastic uncertainty parameter prediction model based on an LSTM network is built. Historical failure rate data from multiple consecutive time points for each SIS element, collected and calculated, are used as input data to train the corresponding LSTM network. After the LSTM network training is complete, the trained stochastic uncertainty parameter prediction models are applied to their respective subsystems. Using the current time as a baseline, historical failure rate data from multiple consecutive time points prior to the current time are used as input to the trained stochastic uncertainty parameter prediction models to predict the stochastic uncertainty parameter prediction parameters for each SIS element in the corresponding subsystem. , , , The results at several points in the future.

5. The dynamic verification method for SIS security integrity level considering uncertainty according to claim 2, characterized in that, In step 4, based on the historical data of multiple time nodes of the random uncertainty parameters of each SIS element, the predicted data of several time nodes after the current time, the single-value data of the fuzzy uncertainty parameters of each SIS element, and the constant parameters of each subsystem of the SIS determined by the basic information, the SIL verification calculation is performed on the entire SIS. The SIL dynamic verification process includes the following steps: I. A reliability block diagram (PFD) is used to construct the three subsystems: the sensing subsystem, the logic control subsystem, and the execution subsystem. avg The PFD of the SIS system is calculated separately. avg PFD for three subsystems avg sum; II. Structural constraint calculation; The safety integrity of structural constraints is determined by the instrument type, safety failure fraction (SFF), and hardware failure margin (HFT). Calculate the SFF of the sensing subsystem, logic control subsystem, and execution subsystem respectively; Hardware fault margin HFT=NM, when HFT≥2, HFT is 2; Where N and M are the voting forms MooN within each group; based on the structural constraint table of Class A / B safety-related subsystems in IEC61508, the SIL determined by the structural constraints of each subsystem is obtained by combining SFF and HFT. Among them, the SIL determined by the structural constraints of SIS is the minimum value of the structural constraint results of the three subsystems; III. Dynamic SIL verification results; determining the SIL level based on the PFD value: When PFD avg If ∈ [0.1, 1), then it is SIL0; when PFD avg If ∈ [0.01, 0.1), then it is SIL1; when PFD avg If ∈ [0.001, 0.01), then it is SIL2; when PFD avg If ∈ [0.0001, 0.001), then it is SIL3; when PFD avg If it is less than 0.0001, it is SIL4; The final SIL of the SIS system is PFD. avg The smaller of the SIL determined by the structural constraints and the SIL determined by the structural constraints.

6. The dynamic verification method for SIS security integrity level considering uncertainty according to claim 1, characterized in that, In step 4, feature importance analysis is performed only on the random uncertainty parameters of SIS components in each subsystem. The influence of parameter changes on the SIL verification results is analyzed based on the random forest algorithm, and the importance is ranked. To rank the importance of SFFs for each subsystem, input historical data of the random uncertainty parameters of the SIS elements in the corresponding subsystem and the SFF values. The result is a ranking of the importance of SFFs for each subsystem. PFD for each subsystem avg Importance ranking, input historical data of random uncertainty parameters of SIS elements in the corresponding subsystem and PFD avg The value obtained is the PFD for each subsystem. avg Importance ranking; For the importance ranking of SIS systems, the input consists of historical data on the random uncertainty parameters of SIS components in all subsystems and the PFD of the SIS system. avg The result is a PFD for SIS systems. avg Ranking by importance.

7. The dynamic verification method for SIS security integrity level considering uncertainty according to claim 1, characterized in that, Step 5 is followed by: Step 6. Dynamically update the SIL level; as production data is updated in real time, the supplementary data will also be processed in steps 2 to 5, thereby expanding the data volume of the data storage file and realizing the dynamic updating of the SIL verification of the data.

8. A dynamic verification system for SIS security integrity levels considering uncertainty, used to implement the dynamic verification method for SIS security integrity levels considering uncertainty as described in any one of claims 1 to 7, characterized in that, The dynamic verification system for SIS security integrity level considering uncertainty includes the following modules: The parameter classification and data acquisition module is used to classify SIS parameters into three categories: random uncertainty parameters, fuzzy uncertainty parameters, and constant parameters; and to collect time-series failure rate data based on field failure records and integrate fuzzy parameter data by multiple expert evaluations. The random uncertainty parameter data processing module is used to build a random uncertainty parameter prediction model based on LSTM network, and to train the random uncertainty parameter prediction model on time based on the collected time-series failure rate data; then, the trained random uncertainty parameter prediction model is used to dynamically predict and update the random uncertainty parameters. The fuzzy uncertainty parameter data processing module is used to construct a relation matrix of expert evaluation data based on the fuzzy parameter data obtained by fusing multiple expert evaluations, calculate the associated fuzzy frequency based on the non-negative matrix eigenvector theory, and select the optimal value of each fuzzy uncertainty parameter. The dynamic SIL verification module is used to calculate the average probability of failure (PFD) of the SIS system when it meets the requirements, based on random uncertainty parameters and fuzzy uncertainty parameters at various time points, combined with constant parameters obtained on-site. avg By combining the Security Failure SFF with the time-series SIL verification results, we obtain the SIL verification results corresponding to each time point. The module also includes a source analysis module, which combines the random forest algorithm to perform source analysis on the SIL verification results, evaluate the influence weight of the random uncertainty parameters of each subsystem in the SIS system on the SIL verification results, and generate a feature importance ranking.

9. A computer device comprising a memory and one or more processors; executable code is stored in the memory; characterized in that, When the processor executes the executable code, it performs the steps of the dynamic verification method for SIS security integrity level considering uncertainty as described in any one of claims 1 to 7.

Images