A method for assessing safety risk of wax oil catalytic cracking unit based on ISM-BN model

Abstract

The application discloses a kind of wax oil catalytic cracking unit safety risk assessment methods based on ISM-BN model.First, the method combining mutual information analysis and partial correlation analysis is used to screen key risk variables;Then, the risk index system is constructed by principal component analysis, the hierarchical relationship of risk propagation is analyzed by establishing an interpretive structural model, and a Bayesian network is constructed to realize the probability reasoning and dynamic assessment of risk;Finally, combined with risk contribution analysis and risk matrix, the risk level division and key risk identification of FCC unit are realized.The application can realize the structural expression and probabilistic analysis of the risk propagation process of FCC unit, improve the interpretability of risk assessment and risk warning capability, and provide technical support for the safe operation and risk control of the unit.

Description

Technical Field

[0001] This invention relates to the field of safety risk assessment and intelligent early warning technology for petrochemical processes, specifically to a safety risk assessment method for FCC devices based on the ISM-BN model. Background Technology

[0002] Catalytic cracking (FCC) units, as core secondary processing units in the oil refining industry, are characterized by high temperature, high pressure, multivariate coupling, and dynamic nonlinearity. Their operation involves multiple subsystems, including reaction, regeneration, fractionation, and flue gas treatment, with complex process coupling relationships between these systems. In actual production, FCC units are susceptible to fluctuations in feedstock, changes in operating conditions, and equipment disturbances, leading to risks such as abnormal temperature, pressure fluctuations, decreased separation efficiency, and system instability.

[0003] Traditional risk assessment models based on experience or single statistical methods struggle to depict the structural transmission relationships between risks. While Bayesian network methods possess probabilistic reasoning capabilities, they lack a systematic structural construction mechanism, leading to a strong reliance on expert experience and insufficient scalability. Therefore, it is necessary to introduce structured modeling methods to hierarchically decompose and model the propagation of risk factors, thereby improving the interpretability and reliability of risk assessments. Summary of the Invention

[0004] To address the above problems, this invention provides a safety risk assessment method for wax oil catalytic cracking units based on ISM-BN fusion modeling, comprising:

[0005] S1: Collect historical operating data, process parameter data, and accident case data of FCC equipment to construct a risk assessment dataset;

[0006] S2: Based on the constructed risk assessment dataset, perform mutual information analysis and partial correlation analysis on the process variable data, calculate the correlation degree between process variables, establish a comprehensive correlation degree evaluation index, and screen key process variables of FCC devices;

[0007] S3: Principal component analysis is used to reduce the dimensionality and extract features from the screened variables, and a safety risk index system for FCC devices is constructed by combining process mechanism knowledge and control logic.

[0008] S4: Establish the adjacency matrix between risk indicators and calculate the reachability matrix to construct an FCC device interpretation structure model;

[0009] S5: Map the explanatory structural model to a Bayesian network structure, establish the probability propagation relationship between risk nodes, and construct a conditional probability table by combining historical operating data and expert knowledge;

[0010] S6: Use Bayesian networks to perform risk probability reasoning to obtain the occurrence probability of each risk node and the overall risk nodes of the device;

[0011] S7: Calculate the contribution of each risk node to the overall risk of the device, and combine it with the risk matrix to complete the risk level classification, realize the risk propagation analysis of FCC device, key risk identification and overall risk assessment.

[0012] Preferably, the FCC device risk assessment dataset in step S1 includes real-time monitoring data of the distributed control system (DCS), interlock status data of the safety instrumented system (SIS), on-site inspection and alarm log data, process parameter design value data, and text data of historical fault and accident cases.

[0013] The DCS real-time monitoring data and process parameter data include process variables such as reactor top pressure, riser outlet temperature, settler level, regenerator dense phase temperature, main air flow rate, fractionation tower top temperature, fractionation tower bottom level, stabilizer tower pressure, and acid water outlet flow rate; the SIS interlock status data includes safety interlock and protection action status; the on-site inspection and alarm log data includes equipment inspection records, abnormal alarm information, and alarm occurrence time; the process parameter design value data includes the design value, control target value, and allowable operating range of the process parameters; and the historical fault and accident case text data includes information such as the cause, evolution process, impact consequences, and handling measures of faults, abnormalities, and accidents.

[0014] Preferably, the formula for calculating the composite score in step S2 is as follows:

[0015]

[0016] In the formula, These are the weighting coefficients. Representative variable and The partial correlation coefficient between them is obtained by conditionally eliminating the elements of the inverse covariance matrix of the standardized dataset, and is used to characterize the net linear association strength after removing the collinearity mediation effect. Representative variable and The mutual information between variables is calculated using a nonparametric estimation method based on the nearest neighbor space, which is used to capture the nonlinear coupling characteristics between variables; The 90th percentile value of the upper triangular element in the total variable mutual information matrix is ​​used for scaling the nonlinear intensity. By retaining variables with a comprehensive correlation greater than a preset truncation threshold, redundant noise is filtered out, thus completing the intelligent selection of key process variables.

[0017] Preferably, the process of constructing the FCC device safety risk indicator system in step S3 is as follows:

[0018] First, principal component analysis (PCA) was performed on the screened key process variables, and principal components were extracted based on the principle that the cumulative variance contribution rate was greater than 90%. Then, the mechanism of each principal component was analyzed by combining the contribution weight of each variable in the principal component loading matrix. Based on the process mechanism relationship, control loop structure and risk propagation path of the FCC unit reaction system, regeneration system, fractionation system, absorption stabilization system and utility system, variables with the same engineering attributes were clustered and merged.

[0019] Finally, the principal components are mapped to risk indicator nodes with clear engineering physical meaning, forming several core risk indicators, and a top-level comprehensive node "device final risk" is added, thereby constructing the final FCC device safety risk indicator system.

[0020] Preferably, the construction of the adjacency matrix in step S4 adopts a dual determination mechanism that combines data correlation with process mechanism constraint verification:

[0021] First, based on the comprehensive correlation matrix obtained in step S2, risk indicator pairs that meet the correlation threshold requirements are screened. Then, the candidate correlation relationships are verified by combining the FCC unit process flow, material and energy transfer paths, control loop logic, and historical accident causal chains. A corresponding directed connection is established only when the risk indicators simultaneously satisfy both statistical correlation and process causal rationality. Finally, an adjacency matrix A that satisfies Boolean algebra operation rules is constructed.

[0022]

[0023] Among them, if factors right If there is a direct impact, then ,otherwise .

[0024] Add the adjacency matrix A to the identity matrix I, and iterate through matrix multiplication to perform exponentiation until the following equation is satisfied:

[0025]

[0026] in, It is the identity matrix. Let be the number of iterations, then the complete reachability matrix of the system is obtained. .

[0027] Determine the reachability set of each risk indicator based on the reachability matrix. and antecedent set :

[0028]

[0029]

[0030] When the following conditions are met:

[0031]

[0032] Then determine the risk factors If an element belongs to the current level, remove that level from the reachability matrix and repeat the level division process until a complete risk level structure is obtained.

[0033] Preferably, the method for mapping the explanatory structure model to a Bayesian network structure in step S5 is as follows:

[0034] The risk indicators in the Explanatory Structural Model are directly mapped to the variable nodes of a Bayesian network, and the directed edges between elements in the ISM are transformed into directed arcs of a Bayesian network, thereby constructing a directed acyclic graph (DAG) of a Bayesian network.

[0035] Preferably, the construction process of the conditional probability table (CPT) in step S5 is as follows: each risk node in the network is discretized into a dual state of "normal" and "abnormal".

[0036] For independent root nodes in the root cause layer, count their frequency of occurrence in historical operation datasets, equipment inspection logs, and abnormal alarm records, and directly assign them a prior probability.

[0037] For non-root nodes, the conditional frequency distribution of the child node's anomalies under all combinations of its parent node states is statistically analyzed. For state combinations not covered by historical data or with insufficient samples, the Delphi expert evaluation method is used for supplementation and correction. Subsequently, the obtained conditional probability parameters are normalized to form a complete conditional probability table. By fusing data statistics with expert knowledge, the inference accuracy and robustness of Bayesian networks in low-sample risk scenarios are improved.

[0038] Preferably, the mechanism for risk probability inference using Bayesian networks in step S6 is as follows: real-time acquisition of measurement point data from FCC devices and mapping to corresponding input nodes; using the conditional independence assumption and probability chain rule of Bayesian networks to perform joint probability inference across the entire network; the calculation formula is as follows:

[0039]

[0040] in, Represents the set of all nodes in the network. This represents the joint probability of the joint state of all nodes in a Bayesian network. For nodes The set of parent nodes, Indicates at node Given the state of the parent node, the node The conditional probability of being in the corresponding state. This formula, based on the conditional independence relationship of Bayesian networks, decomposes the joint probability distribution into the product of the conditional probabilities of each node under the conditions of its parent node, thereby realizing a structured expression of the joint probability distribution of complex multivariate risk systems.

[0041] Preferably, the key risk identification and overall risk assessment in step S7 specifically includes: when an abnormal deviation of the device is detected or a sudden change in the probability of the top consequence node occurs, the observation state of that node is set to an abnormal state and input as deterministic observation evidence into the Bayesian network. Based on the Bayesian network posterior inference mechanism, inverse probability updates are performed to dynamically update the posterior probabilities of each node in the root cause layer and intermediate layer. The calculation formula is as follows:

[0042]

[0043] Where E represents the observed evidence, i.e. the node state information formed by real-time running data or observed abnormal states; U represents the joint state of all risky nodes in the Bayesian network; P(U|E) represents the posterior probability of network state U given the observed evidence E; P(U,E) represents the joint probability of network state U and observed evidence E occurring simultaneously; and P(E) represents the marginal probability of observed evidence E. This represents the summation of the joint probabilities corresponding to all possible joint states U of the Bayesian network, used to normalize the posterior probability. Let... Let represent the prior probability of node Xi. This represents the conditional posterior probability given the system stop event. It also represents the risk contribution of each node. It is quantified by the difference between its posterior probability and prior probability, as shown below:

[0044]

[0045] Among them, contribution The larger the absolute value, the more significant the impact of the node on the overall risk status. Based on the risk contribution ranking results, key risk sources and main risk propagation paths can be identified.

[0046] Then, based on the occurrence probability obtained in real time from each risk node via a Bayesian network... A two-dimensional risk assessment matrix is ​​constructed by determining the probability level of risk occurrence and the severity level based on the severity criteria for accident consequences. The probability of risk occurrence is divided into five levels: L1 (P < 0.01%), L2 (0.01% ≤ P < 0.1%), L3 (0.1% ≤ P < 1%), L4 (1% ≤ P < 10%), and L5 (P ≥ 10%); the severity of risk consequences is divided into five levels: C1, C2, C3, C4, and C5.

[0047] According to the risk level calculation formula:

[0048]

[0049] Construct a 5×5 risk matrix. Based on the risk matrix, classify the risks of FCC devices into low risk, medium risk, high risk, and extremely high risk levels, thereby enabling FCC device risk level assessment and safety management decision support.

[0050] Beneficial effects of the present invention

[0051] This invention constructs a hierarchical propagation structure of risk factors through an interpretive structural model, revealing the formation and evolution mechanism of risks in FCC devices. Furthermore, it integrates the interpretive structural model with Bayesian networks to unify risk propagation relationship modeling and probabilistic reasoning analysis, overcoming the problems of traditional risk assessment methods that rely on experience and struggle to reflect complex causal relationships. Simultaneously, by combining risk contribution analysis and risk matrix assessment, it achieves the identification of key risk sources, the tracing of risk propagation paths, and the classification of risk levels, improving the scientific rigor, interpretability, and engineering application value of FCC device risk assessment. This provides technical support for FCC device safe operation monitoring, risk early warning, and safety decision-making. Attached Figure Description

[0052] Figure 1 This is a general framework diagram of the FCC device safety risk assessment method based on the ISM-BN model.

[0053] Figure 2 This is a diagram of the ISM hierarchy of the FCC device.

[0054] Figure 3 It is a Bayesian network model and a probability distribution diagram of normal operating conditions.

[0055] Figure 4 It is the inverse posterior probability inference graph triggered by S16.

[0056] Figure 5 This is a histogram comparing the risk contribution of each node.

[0057] Figure 6 This is a diagram showing the results of the FCC device risk matrix assessment. Detailed Implementation

[0058] The specific embodiments of the present invention will be further described in detail below with reference to the accompanying drawings. These embodiments are for illustrative purposes only and are not intended to limit the scope of protection of the present invention.

[0059] This embodiment takes a wax oil catalytic cracking (FCC) unit as the research object and establishes a safety risk assessment method for FCC units based on the ISM-BN model. The overall process is as follows: Figure 1 As shown.

[0060] First, historical operating data, process parameter data, alarm log data, and accident case data of the FCC unit are collected, and missing value processing, outlier removal, timestamp alignment, and normalization are completed.

[0061] Subsequently, a combination of partial correlation analysis and mutual information analysis was used to calculate the comprehensive correlation between variables, construct a variable correlation matrix, and screen key process variables according to a set threshold. After screening, 53 key process variables were retained for subsequent risk modeling.

[0062] Principal component analysis (PCA) was further used to reduce the dimensionality of key variables. Based on the principle that the cumulative variance contribution rate is greater than 90%, 15 principal components were extracted. Combined with the process mechanism knowledge of the FCC unit reaction system, regeneration system, fractionation system, absorption stabilization system and utility system, each principal component was given a clear engineering meaning, forming 15 core risk indicators, as shown in Table 1.

[0063] Table 1. Dimensionality reduction results of PCA

[0064]

[0065] Then, a top-level integrated node, "S16 Device Final Risk," is added to construct an FCC device safety risk indicator system containing 16 risk indicator nodes, as shown in Table 2.

[0066] Table 2 Risk Indicator System for FCC Units

[0067]

[0068] Candidate correlations were selected based on the comprehensive correlation evaluation results. Mechanism verification was performed by combining the FCC unit's process mechanism, control logic, and historical accident propagation paths. An adjacency matrix between risk indicators was constructed, and a reachability matrix was calculated. An Explanatory Structure Model (ISM) with eight levels was established, such as... Figure 2 As shown.

[0069] The underlying risk factors mainly include basic risk sources such as abnormal main air supply, abnormal steam system, and abnormal circulating cooling water system; the intermediate layer mainly includes the operating status of the reaction system, regeneration system, and fractionation system; and the top layer is the overall risk status of the FCC unit.

[0070] Subsequently, a Bayesian network model was established based on the ISM structure, mapping 16 risk indicators to Bayesian network nodes. The causal relationships in the ISM were converted into directed edges in the Bayesian network, forming the risk propagation network structure for the FCC device. Historical operating data was used to statistically analyze the node probability distribution, and a conditional probability table was constructed in conjunction with expert experience to complete the Bayesian network parameter learning. The final Bayesian network model and the probability distribution diagram for normal operating conditions are shown below. Figure 3 As shown.

[0071] Furthermore, the overall risk node S16 of the device is set as the target event. The probability changes of each risk node and the overall risk node are calculated through posterior probability inference. The posterior probability inference results are as follows: Figure 4 As shown.

[0072] Identify key risk sources and main risk propagation paths based on their risk contribution levels, for example, the contribution levels of each node. Figure 5 As shown.

[0073] Finally, a risk matrix was established by combining the probability of risk occurrence with the severity of accident consequences to classify the risk levels of FCC facilities. The risk status was then divided into four levels: low risk, medium risk, high risk, and extremely high risk. The final facility risk matrix assessment results are as follows: Figure 6 As shown in Table 3, the severity levels of risk events are classified according to the following criteria. The results indicate that the reaction-regeneration system and key utility systems are the main risk concentration areas of the FCC unit and should be the focus of monitoring and risk control. The fractionation system, absorption stabilization system, and flue gas treatment system are medium-risk areas and require regular inspections and alarm management to maintain safe operation. The product output control and compressor nodes have relatively low risks and can be managed with routine maintenance.

[0074] Table 3. Risk Event Severity Classification Standards

[0075]

[0076] The results show that the present invention can effectively reveal the risk propagation mechanism of FCC devices, realize the quantitative analysis of risk evolution process and key risk identification under complex operating conditions, and provide theoretical basis and technical support for risk classification and control, resource optimization and allocation and safety decision-making of FCC devices.

[0077] The specific embodiments described herein are merely illustrative of the spirit of the invention. Those skilled in the art to which this invention pertains may make various modifications or additions to the described specific embodiments or use similar methods to substitute them, without departing from the spirit of the invention or exceeding the scope defined by the appended claims.

Claims

1. A method for assessing the safety risks of FCC devices based on the ISM-BN model, characterized in that, Includes the following steps: S1: Collect historical operating data, process parameter data, and accident case data of FCC equipment to construct a risk assessment dataset; S2: Based on the constructed risk assessment dataset, perform mutual information analysis and partial correlation analysis on the process variable data, calculate the correlation degree between process variables, establish a comprehensive correlation degree evaluation index, and screen key process variables of FCC devices; S3: Principal component analysis is used to reduce the dimensionality and extract features of the screened key process variables, and a safety risk index system for FCC devices is constructed by combining process mechanism knowledge and control logic. S4: Establish the adjacency matrix between risk indicators and calculate the reachability matrix to construct an FCC device interpretation structure model; S5: Map the explanatory structural model to a Bayesian network structure, establish the probability propagation relationship between risk nodes, and construct a conditional probability table by combining historical operating data and expert knowledge; S6: Use Bayesian networks to perform risk probability reasoning to obtain the occurrence probability of each risk node and the overall risk nodes of the device; S7: Calculate the contribution of each risk node to the overall risk of the device, and combine it with the risk matrix to complete the risk level classification, realize the risk propagation analysis of FCC device, key risk identification and overall risk assessment.

2. The method according to claim 1, characterized in that: In step S2, a combination of partial correlation analysis and mutual information analysis is used to evaluate the linear and nonlinear relationships between process variables, and a comprehensive correlation evaluation index is constructed. Based on the magnitude of the comprehensive correlation, key process variables of the FCC device are selected.

3. The method according to claim 1, characterized in that: In step S3, principal component analysis is used to reduce the dimensionality of key process variables, and a safety risk index system for FCC devices is constructed based on the principal component loading matrix, process mechanism knowledge, and control logic.

4. The method according to claim 1, characterized in that: In step S4, based on the comprehensive correlation evaluation results, and combined with the knowledge of the FCC unit's process mechanism, control logic relationship and historical accident propagation path, an adjacency matrix between risk indicators is established to form a direct correlation structure between risk factors.

5. The method according to claim 1, characterized in that: In step S4, the reachability matrix M is calculated based on the adjacency matrix A. Boolean matrix operations are used to perform exponential iterations on (A+I). When the iteration result tends to be stable, the reachability matrix M is obtained. The reachable set and prior set of each risk indicator are determined based on the reachability matrix, and the hierarchical structure of risk factors is obtained through hierarchical division.

6. The method according to claim 1, characterized in that: In step S5, the directed relationships in the structural model are mapped to directed edges in a Bayesian network, the hierarchical nodes are mapped to Bayesian network nodes, and the conditional probability distribution between nodes is determined based on historical data and expert knowledge.

7. The method according to claim 1, characterized in that: In step S6, the state of the Bayesian network evidence nodes is continuously updated based on real-time running data, and the risk probability is dynamically corrected and evolved through forward inference and posterior update.

8. The method according to claim 1, characterized in that: In step S7, the overall risk nodes of the device are set to an abnormal state. The posterior probability of each risk node is calculated by Bayesian network posterior inference. The difference between the posterior probability and the prior probability of each risk node is used as the risk contribution index. Based on the risk contribution ranking results, key risk sources and main risk propagation paths are identified.

9. The method according to claim 1, characterized in that: In step S7, a risk matrix is ​​constructed based on the probability level of risk occurrence and the severity level of risk consequences. Based on the results of the risk matrix, the risks of FCC devices are divided into different risk levels to achieve risk level assessment and safety management decision support.

Images