Intelligent reasoning method and system for chemical production safety risk and storage medium

Abstract

The invention relates to the technical field of chemical production safety, in particular to a chemical production safety risk intelligent reasoning method and system and a storage medium, and the method comprises the steps: building a multi-level flow model according to a chemical production flow chart, and constructing a fault tree analysis qualitative model for the multi-level flow model through a danger and operability analysis technology; fusing and correcting the multiple groups of evidence sources according to the D-S evidence theory to obtain the initial abnormal probability of each basic event; and establishing a dynamic Bayesian network quantitative model in combination with a logic gate rule, and performing forward reasoning and backward reasoning. According to the invention, forward reasoning and backward reasoning can be carried out on abnormal event nodes in the chemical production process, causal traceability is completed, an abnormal propagation path is explored, and an effective management means is provided for safe production in the chemical process.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of chemical production safety, in particular to a chemical production safety risk intelligent reasoning method and system and a storage medium. BACKGROUND

[0002] Chemical production process involves complex physical and chemical reactions, high temperature and high pressure operation and the use of various dangerous substances, with high risk and complexity. With the advent of the digital information age, new generation technologies such as artificial intelligence, big data and industrial internet are increasingly combined with safety management, reflecting the new trend of industrial safety production management level.

[0003] Chemical production safety risk reasoning refers to identifying and assessing potential risks in chemical production process, predicting the root cause of failure, possible consequences and their transmission path, such as identifying potential risk nodes such as heat exchanger corrosion and sensor failure in advance, predicting how abnormal conditions such as pressure anomaly affect the transmission direction of downstream storage tanks and other abnormal conditions through pipelines, so as to take appropriate intervention measures in time and effectively, prevent accidents at the embryonic stage, and provide help for process recovery and process optimization, thereby significantly improving the safety and efficiency of chemical production.

[0004] However, the existing chemical production safety risk reasoning method still has the following defects: first, it mainly relies on expert experience or safety regulations for reasoning, which is difficult to quantify risk probability, resulting in the inability to scientifically judge the possibility of accidents at each risk node, and lack of data support for risk prevention and control, which may miss the early intervention of high probability risk events; second, it cannot effectively integrate multi-source evaluation information, resulting in one-sided evaluation of risks and insufficient accuracy and reliability of reasoning; third, it cannot effectively identify key risk nodes and analyze abnormal transmission paths, and when an abnormality occurs in a certain link, it cannot timely predict its impact on the entire production system, resulting in delayed accident emergency response. SUMMARY

[0005] Therefore, the technical problem to be solved by the present application is to overcome the problem that the existing technology cannot effectively ensure the safety of chemical production due to the dependence on experience, the difficulty in quantifying risk probability, the inability to integrate multi-source information and the inability to identify key nodes and abnormal transmission paths.

[0006] To solve the above technical problems, the present application provides a chemical production safety risk intelligent reasoning method, which comprises:

[0007] A multi-level flow model is established according to the chemical production flow chart; the nodes in the multi-level flow model are divided into basic events, intermediate events and top events by using hazard and operability analysis technology, and a fault tree analysis model is constructed according to the relationship between each event;

[0008] a plurality of sets of evaluation information of basic event abnormality are collected, and each piece of evaluation information in each set of evaluation information is converted into a fuzzy risk probability according to an evaluation index, and a plurality of sets of evidence sources are constructed correspondingly;

[0009] The fuzzy risk probabilities of the plurality of sets of evidence sources are fused to obtain initial abnormal probabilities of the basic events;

[0010] Each event is taken as a node, and a dynamic Bayesian network model is constructed according to conditional transition probabilities between nodes in the same time slice, state transition probabilities of each node between adjacent time slices and the initial abnormal probabilities of the basic events.

[0011] Forward reasoning is performed on the dynamic Bayesian network model to obtain abnormal probabilities of each node event in each time slice.

[0012] Reverse reasoning is performed on the dynamic Bayesian network model, the abnormal probability of a target node event is set to 1, and a basic event causing the abnormal occurrence of the target node event and an abnormal probability of the basic event are obtained.

[0013] Preferably, a plurality of sets of evaluation information of basic event abnormality are collected, and each piece of evaluation information in each set of evaluation information is converted into a fuzzy risk probability according to an evaluation index, and a plurality of sets of evidence sources are constructed correspondingly.

[0014] The evaluation index of each basic event in each set of evaluation information is converted into a trapezoidal fuzzy number, and the fuzzy likelihood of each basic event is solved by using an area barycenter method, and a formula is as follows:

[0015]

[0016] wherein, P S is the fuzzy likelihood of the basic event, and (a, b, c, d) is the trapezoidal fuzzy number of the basic event;

[0017] The fuzzy risk probability of each basic event is calculated according to the fuzzy likelihood of each basic event, and a formula is as follows:

[0018]

[0019] wherein, P is the fuzzy risk probability of the basic event.

[0020] Preferably, the fuzzy risk probabilities of the plurality of sets of evidence sources are fused to obtain initial abnormal probabilities of the basic events, and the method comprises the following steps:

[0021] The support degree between each two sets of evidence sources is calculated, the weight of each set of evidence sources is calculated according to the support degree between each two sets of evidence sources, and the initial abnormal probability of each basic event is obtained by fusing the fuzzy risk probability of each basic event in the evidence sources according to the D-S evidence theory after weighting.

[0022] Preferably, the support between each two groups of evidence sources is calculated by using the formula of the Lantian distance:

[0023]

[0024] s ij =1-d ij (E i ,E j )

[0025] where E i and E j are the ith group of evidence sources and the jth group of evidence sources respectively, d ij (E i ,E j ) is the Lantian distance between E i and E j , N0 is the number of the power set of basic events, P i (A k ) and P j (A k ) are the fuzzy risk probability of the kth focus element in the ith group of evidence sources and the jth group of evidence sources respectively; s ij is the support between the ith group of evidence sources and the jth group of evidence sources.

[0026] Preferably, the support between each two groups of evidence sources is calculated by using the formula of the cosine theorem:

[0027]

[0028] where s ij is the support between the ith group of evidence sources and the jth group of evidence sources, N0 is the number of the power set of basic events, P i (A k ) and P j (A k ) are the fuzzy risk probability of the kth focus element in the ith group of evidence sources and the jth group of evidence sources respectively.

[0029] Preferably, the weight of each group of evidence sources is calculated by using the support between each two groups of evidence sources, and the formula is:

[0030]

[0031] where s i is the average similarity of the ith group of evidence sources, s ij is the support between the ith group of evidence sources and the jth group of evidence sources, M is the number of evidence sources, and w i is the weight of the ith group of evidence sources.

[0032] Preferably, the nodes in the fault tree analysis model are connected according to the logic gate rules to construct a conditional probability table and obtain the conditional transition probabilities between nodes within the same time slice.

[0033] Preferably, in the dynamic Bayesian network model, the anomaly probability of the nth node in the t-th time slice is expressed as:

[0034]

[0035] in, and These are the nth node in the t-th time slice and the nth node in the (t-1)-th time slice, respectively. Let N be the probability of an anomaly of the nth node in the t-th time slice, and N be the total number of nodes. for The parent node, for The parent node; Let represent the conditional transition probability of the nth node within time slice t. This represents the state transition probability of the nth node between time slice t-1 and time slice t.

[0036] This invention also provides an intelligent reasoning system for safety risks in chemical production, comprising:

[0037] The fault tree analysis model building module is used to establish a multi-level flow model based on the chemical production process diagram; using hazard and operability analysis techniques, the nodes in the multi-level flow model are divided into basic events, intermediate events, and top events, and a fault tree analysis model is built based on the relationships between the events.

[0038] The evidence source construction module is used to collect multiple sets of evaluation information on basic event anomalies, and convert each piece of evaluation information in each set of evaluation information into fuzzy risk probability according to the evaluation index, thereby constructing a set of evidence sources.

[0039] The initial probability calculation module is used to fuse the fuzzy risk probabilities of multiple evidence sources to obtain the initial anomaly probability of each basic event.

[0040] The Dynamic Bayesian Network Model Building Module is used to construct a dynamic Bayesian network model by treating each event as a node, using the conditional transition probability between nodes within the same time slice, the state transition probability between nodes in adjacent time slices, and the initial anomaly probability of basic events.

[0041] The forward reasoning module is used to perform forward reasoning using a dynamic Bayesian network model to obtain the anomaly probability of events at each node in each time slice.

[0042] The reverse reasoning module is used for reverse reasoning in a dynamic Bayesian network model, sets an abnormal probability of a target node event as 1, and obtains a basic event causing the abnormal occurrence of the target node event and an abnormal probability thereof.

[0043] The application further provides a computer readable storage medium, wherein a computer program is stored on the computer readable storage medium, and the computer program is executed by a processor to implement steps of the intelligent reasoning method for chemical production safety risk.

[0044] The above technical scheme of the application has the following beneficial effects compared with the prior art:

[0045] The intelligent reasoning method for chemical production safety risk converts language-form evaluation information into fuzzy risk probability in the form of fuzzy numbers, forms multiple sets of evidence sources, and fuses the fuzzy risk probability of the multiple sets of evidence sources to obtain the initial abnormal probability of each basic event. BRIEF DESCRIPTION OF DRAWINGS

[0046] In order to make the content of the application more easily understood, the application will be further described in detail below according to specific embodiments of the application and in conjunction with the drawings, in which:

[0047] Figure 1 is a flowchart of the intelligent reasoning method for chemical production safety risk of the application;

[0048] Figure 2 is a chemical production flowchart of an embodiment of the application;

[0049] Figure 3 This is a schematic diagram of the MFM modeling language according to an embodiment of the present invention;

[0050] Figure 4 This is a schematic diagram of the multi-level flow model in this embodiment;

[0051] Figure 5 This is a schematic diagram of the fault tree analysis model in this embodiment;

[0052] Figure 6 This is a schematic diagram of the dynamic Bayesian network model in this embodiment;

[0053] Figure 7 This is a forward reasoning result diagram of the target node event in an embodiment of the present invention, wherein... Figure 7 In the diagram, (a) is the anomaly probability graph of node C1. Figure 7 In the diagram, (b) is the anomaly probability graph for node M6. Figure 7 In the diagram, (c) represents the anomaly probability map of node M7. Figure 7 In the graph, (d) represents the anomaly probability map of node M8;

[0054] Figure 8 This is a diagram showing the reverse reasoning result of the target node event in an embodiment of the present invention;

[0055] Figure 9 This is a schematic diagram of the fault chain of the target node event in an embodiment of the present invention. Detailed Implementation

[0056] The present invention will be further described below with reference to the accompanying drawings and specific embodiments, so that those skilled in the art can better understand and implement the present invention. However, the embodiments described are not intended to limit the present invention.

[0057] Reference Figure 1 As shown, this invention provides an intelligent reasoning method for safety risks in chemical production, comprising:

[0058] S1: Establish a multilevel flow model (MFM) based on the chemical production process flow diagram; use Hazard and Operability Study (HAZOP) technology to divide the nodes in the multilevel flow model into basic events, intermediate events and top events, and construct a fault tree analysis (FTA) model based on the relationships between the events.

[0059] Figure 2A chemical production flow chart is provided for the present embodiment. According to the transmission path of the material flow and the energy flow, the MFM modeling language is introduced to reconstruct the chemical production flow chart, forming a multi-level flow model. Figure 3 A schematic diagram of the MFM modeling language is provided.

[0060] The meanings, applicable scenarios and examples of each function in the MFM modeling language are shown in Table 1.

[0061] Table 1, meanings, applicable scenarios and examples of each function in the MFM modeling language

[0062]

[0063] The target in the MFM modeling language represents the expected state of a system or subsystem, refers to the set value of a process parameter or safety requirement, and is usually a representation of a quantity, such as the amount of material discharged or the amount of leakage.

[0064] The structure in the MFM modeling language represents a physical device or component, refers to a physical carrier that implements a function, such as a fractionating module, an exhaust module, a precipitated extract module, etc.

[0065] The meanings, applicable scenarios and examples of each relationship in the MFM modeling language are shown in Table 2.

[0066] Table 2, meanings, applicable scenarios and examples of each relationship in the MFM modeling language

[0067]

[0068] The MFM modeling process specifically includes: first, dividing the nodes in the chemical production flow chart, which can be understood as devices (such as condensers, fractionating columns, etc.) contained in the flow; then using the MFM graphical modeling language, replacing the corresponding devices in the chemical production flow chart according to the functions of different nodes, such as a feed pump which can be represented by a source graphically, indicating the function of providing raw materials, and a condenser which mainly functions to transport material flow, which can be replaced by a transmission graph. Figure 4 A schematic diagram of the multi-level flow model established according to the chemical production flow chart of the present embodiment is provided.

[0069] Establishing a multi-level flow model is in the category of qualitative analysis. In the specific application of a flow, the selected function representation is not unique, as long as it meets certain physical meaning, can reasonably represent the function contained in the original component, and does not produce serious impact during subsequent analysis.

[0070] All nodes in the multi-level flow model are identified and subjected to qualitative risk reasoning using the hazard and operability analysis technique, each time a target node is specified, the causes, consequences and countermeasures about the node obtained through HAZOP analysis are recorded, and finally a HAZOP record table for the flow is formed.

[0071] According to the HAZOP record table, the nodes in the multi-level flow model are further divided into basic events, intermediate events and top events, and a fault tree analysis model is constructed according to the relationship between the events, so as to present the relationship between the nodes. Both the multi-level flow model and the fault tree analysis model are qualitative models.

[0072] Specifically, the basic event is the bottom layer of the physical node that can be faulted in the multi-level flow model, such as valve X19. The intermediate event is the abnormality of a node, such as "condenser failure". The top event is a system-level fault, such as "combustible material leakage". For example, the combustible material leakage is taken as the top event in this embodiment, and the nodes related to the top event in the multi-level flow model are analyzed as the heavy reboiler tank, the light reboiler tank and the chimney. The leakage of the heavy reboiler tank and the light reboiler tank and the abnormality of the chimney gas discharge can all cause the occurrence of the top event, so the three are taken as the lower intermediate events of the top event and marked as M6, M7 and M8. According to the multi-level flow model, the fault tree analysis model is finally obtained by backtracking, as shown in Figure 5

[0073] Specifically, the basic events, intermediate events and top events obtained by the division in this embodiment and their meanings are shown in Table 3. Wherein X represents the basic event, M and Y represent the intermediate event, and C represents the top event.

[0074] Table 3, meanings of basic events, intermediate events and top events

[0075]

[0076] S2: Collect multiple sets of evaluation information of the abnormality of the basic event, and convert each piece of evaluation information in each set of evaluation information into a fuzzy risk probability according to the evaluation index, so as to correspondingly construct a set of evidence sources.

[0077] Since the probability of occurrence of abnormality in the real situation is small, and once it occurs, it is a major accident, it is difficult to collect a considerable number of numerical values, therefore, the evaluation information in this embodiment adopts the language quantity evaluation index of the experts for the abnormality of the node, such as the over-high and over-low events of the X1 fractionating tower temperature, and the experts will give the judgment according to the evaluation index, such as very low, low-very low, etc., as the evaluation information of the event abnormality. The evaluation information in this embodiment comes from multiple expert groups, so as to weaken the subjectivity of the information of a single expert group.

[0078] This embodiment selects the trapezoidal fuzzy number (a, b, c, d) as the fuzzy number of the evaluation index, and the corresponding conversion relationship is shown in Table 4.

[0079] Table 4, conversion of evaluation index and trapezoidal fuzzy number

[0080]

[0081] ​After converting the evaluation indicators of each basic event in each set of evaluation information into trapezoidal fuzzy numbers, the fuzzy likelihood of each basic event is solved using the area centroid method, with the following formula:

[0082]

[0083] Among them, P S Let (a,b,c,d) be the fuzzy likelihood of the basic events, and let (a,b,c,d) be the trapezoidal fuzzy number of the basic events.

[0084] The fuzzy risk probability of each basic event is calculated based on its fuzzy likelihood, using the following formula:

[0085]

[0086] Where P is the fuzzy risk probability of the basic event.

[0087] S3: By fusing the fuzzy risk probabilities from multiple evidence sources, the initial anomaly probabilities of each basic event are obtained, including:

[0088] Calculate the support between every two sets of evidence sources, and use the support between every two sets of evidence sources to calculate the weight of each set of evidence sources; after weighting the fuzzy risk probabilities of each basic event in the evidence sources with the weights, fuse them according to the DS (Dempster-Shafer) evidence theory to obtain the initial anomalous probability of each basic event.

[0089] The Focal Element is a core concept in the DS evidence theory, used to describe "the set of propositions supported by evidence" in uncertain reasoning.

[0090] Preferably, the support between evidence sources can be measured using methods such as the Langevin distance and the law of cosines.

[0091] Suppose there exist M sets of evidence sources {E1, E2, ..., E...} i E j ,…,E M}, randomly select two independent evidence vectors E i ={P i (X1),P i (X2),…,P i (X G )} and E j ={P j (X1),P j (X2),…,P j (X G )}, where G is the total number of basic events.

[0092] The support between any two sets of evidence sources can be calculated using the Langone distance or the law of cosines, with the following formula:

[0093]

[0094] s ij =1-d ij (E i ,E j )

[0095] wherein E i and E j are the ith group of evidence sources and the jth group of evidence sources respectively, d ij (E i ,E j ) is the Mahalanobis distance between E i and E j , N0 is the number of power set of basic events, and the power set of all evidence sources is the same; P i (A k ) and P j (A k ) are the fuzzy risk probability of the kth focus element in the ith group of evidence sources and the jth group of evidence sources respectively; s ij is the support degree between the ith group of evidence sources and the jth group of evidence sources.

[0096] The support degree between each two groups of evidence sources is calculated by using the cosine theorem, and the formula is as follows:

[0097]

[0098] wherein s ij is the support degree between the ith group of evidence sources and the jth group of evidence sources, N0 is the number of power set of basic events, and P i (A k ) and P j (A k ) are the fuzzy risk probability of the kth focus element in the ith group of evidence sources and the jth group of evidence sources respectively.

[0099] The weight of each group of evidence sources is calculated by using the support degree between each two groups of evidence sources, and the formula is as follows:

[0100]

[0101] wherein s i is the average similarity of the ith group of evidence sources, s ij is the support degree between the ith group of evidence sources and the jth group of evidence sources, M is the number of evidence sources, and w i is the weight of the ith group of evidence sources.

[0102] After the fuzzy risk probability of each basic event in the evidence sources is weighted by using the weight, the initial abnormal probability of each basic event is obtained by using the classical D-S evidence theory fusion.

[0103] In this example, the fuzzy risk probabilities of the weighted basic events cannot be simply added together as the corrected probability. Instead, they need to be fused using the classic DS combination rule. This is because when there is a significant conflict between the evidence, direct addition will mask the conflict and lead to distorted results.

[0104] S4: Treat each event as a node, and construct a dynamic Bayesian network model using the conditional transition probabilities between nodes within the same time slice, the state transition probabilities between nodes in adjacent time slices, and the initial anomaly probabilities of basic events.

[0105] Preferably, the nodes in the fault tree analysis model are connected according to the logic gate rules to construct a Conditional Probability Table (CPT) and obtain the conditional transition probabilities between nodes within the same time slice.

[0106] Logic gates typically refer to AND-OR gates. The pass condition for an AND gate is that all input events occur simultaneously; the pass condition for an OR gate is that any one input event occurs.

[0107] Suppose that the current fault tree analysis model has G basic events, and the state f of each basic event is a binary variable, taking two states: 0 and 1, where 0 indicates that the fault has not occurred and 1 indicates that the fault has occurred.

[0108] Since the top event is derived step-by-step from the basic events, the state of the top event... It is the basic event vector F = (f1, f2, ..., f G The function of ) can be expressed as The structure function representing the fault tree.

[0109] For an AND gate fault tree, the top event indicates an event has occurred when all the state values ​​of the basic events are 1; for an OR gate fault tree, the top event indicates an event has occurred when the state value of any basic event is 1.

[0110] The fault tree constructed in this embodiment mainly uses OR gates to connect causal events, indicating that any combination of causal events below will cause the result event above to occur. The detailed combinations of causal events are too complex and extensive to be fully depicted, but will be shown in CPT. The specific conditional transition probabilities between event combinations are also converted from empirical language to probabilities through fuzzy processing. For example, for nodes M3, X10, Y1, and X11, different permutations and combinations of 0 / 1 (i.e., occurrence and non-occurrence) for X10, Y1, and X11 are performed, and then an empirical judgment is given for the 0 / 1 case of M3.

[0111] For a set of random variables [x1, x2, ..., x...]N The established dynamic Bayesian network defines the initial anomaly probability of each node in the initial time slice. and conditional transition probability n = 1, 2, ..., N This refers to the state transition probability between adjacent time slices of the same node.

[0112] Therefore, in the dynamic Bayesian network model, the probability of the nth node occurring in the t-th time slice is expressed as:

[0113]

[0114] in, and These are the nth node in the t-th time slice and the nth node in the (t-1)-th time slice, respectively. Let N be the probability of the nth node occurring in the t-th time slice, and N be the total number of nodes. for The parent node, for The parent node; Let represent the conditional transition probability of the nth node within time slice t. This represents the state transition probability of the nth node between time slice t-1 and time slice t.

[0115] Using Bayes' theorem, the abnormal probabilities of intermediate and top events in the initial time slice are calculated using the conditional transition probabilities between nodes within the same time slice and the initial abnormal probabilities of basic events. Furthermore, by combining the state transition probabilities of each node between adjacent time slices, the abnormal probabilities of events of each node in subsequent time slices are calculated.

[0116] The dynamic Bayesian network model described is a quantitative model. The dynamic Bayesian network model constructed in this embodiment refers to... Figure 6 As shown.

[0117] S5: Use a dynamic Bayesian network model for forward reasoning to obtain the anomaly probability of events at each node in each time slice.

[0118] Specify the target node to focus on, use the DBN quantitative model to perform forward reasoning, deduce how the anomaly probability of the target node changes over time, and further deduce the result chain.

[0119] S6: Perform reverse reasoning using a dynamic Bayesian network model, set the abnormal probability of the target node event to 1, and obtain the basic events that cause the abnormal occurrence of the target node event and their abnormal probabilities.

[0120] The target node to be focused is specified, the target node is taken as the center, the abnormal probability is set to 1, the reverse reasoning is carried out through the DBN quantitative model, the reason chain is derived, and the root cause and other sensitive factors with strong influence on causing the abnormality of the current target node are found.

[0121] Specifically, the reverse reasoning includes:

[0122] The abnormal probability of the target node event is set to 1: for example, combustible leakage C1 = 1 indicates that the observation of 'leakage occurrence' is observed;

[0123] Bayesian update: the posterior probability of the parent node is reversely calculated by using the conditional probability table (CPT) of the DBN:

[0124]

[0125] Wherein P(C1 = 1 | X19) is from CPT, P(X19) is the initial probability of the basic event X19, and P(C1 = 1) is the total probability of occurrence obtained by the total probability formula under the influence of all parent nodes, which is essentially the result of forward reasoning.

[0126] According to the fault chain, the basic event is continuously reversely calculated.

[0127] Meanwhile, the forward and reverse reasoning are carried out through the DBN quantitative model, and the above operation is repeated for the maximum probability of each layer, so that the fault propagation path is derived, thereby providing certain guidance for the safety of the chemical production process.

[0128] Figure 7 is a forward reasoning result graph of the target node event in the embodiment of the application, wherein Figure 7 (a) in the figure is an abnormal probability graph of the node C1, Figure 7 (b) in the figure is an abnormal probability graph of the node M6, Figure 7 (c) in the figure is an abnormal probability graph of the node M7, Figure 7 (d) in the figure is an abnormal probability graph of the node M8. In the embodiment, the number of time slices is set to 10, and the abnormal state risk evolution probability of the top event C1 and its parent nodes M6, M7 and M8 can be observed. Figure 7 It can be seen that the abnormal probability of C1 is 2.14% at the initial moment, and increases with time, and the abnormal change trend of the parent nodes M6, M7 and M8 is the same, which further verifies the rationality of the model reasoning result.

[0129] Figure 8 is a reverse reasoning result graph of the target node event in the embodiment of the application. From Figure 8 It can be seen that when the target node event is abnormal, the posterior probability of each basic event is sorted as follows: X 19 >X 14 >X10 X 18 X 11 , which indicates that the basic node X 19 - the valve between the top and blowdown drum - over opening is the most likely cause of C1 abnormality with a probability of 17.36%, and is the root cause of C1 abnormality, which needs to be focused on first. According to the size comparison of ROV (Ratio of Variation) in the figure, although the posterior probabilities of basic events X8, X 12 , X 16 , etc. are low, their ROV values are high, especially the ROV value of X8 reaches 0.8676, which indicates that the top event state changes are sensitive to X8, and once X8 produces an abnormality, it will cause a great change in the top event, so it still needs to be focused on.

[0130] Figure 9 is a fault chain diagram of the target node event in the embodiment of the present application. Assuming that C1 occurs, the posterior probability of M8 among the three parent nodes M6, M7 and M8 is the highest through reverse reasoning, which indicates that the abnormality of M8 has the most significant influence on the abnormality of C1, and then M8 is taken as the target node, and assuming that M8 occurs abnormally, the same reasoning process is performed to deduce that the posterior probability of M5 among the three parent nodes is the highest, and so on until the basic node is deduced, and finally the most likely fault chain causing C1 abnormality is inferred as shown by the red line in Figure 9 . The remaining two blue and green fault chains shown in the figure also have a great influence, and when they occur abnormally, they still need to be appropriately focused on.

[0131] In summary, the chemical production safety risk intelligent reasoning method provided by the application constructs a fault tree analysis model for the chemical production process, obtains multiple sets of evaluation information based on the basic events in the fault tree analysis model, converts the language form evaluation information into fuzzy risk probability in the form of fuzzy numbers, constitutes multiple sets of evidence sources, fuses the fuzzy risk probability of multiple sets of evidence sources to obtain the initial abnormal probability of each basic event, further performs dynamic Bayesian network modeling based on the initial abnormal probability of the basic event, so that the application combines the clarity of the fault tree in the logical structure modeling and the quantitative ability of the dynamic Bayesian network for time sequence uncertainty, can accurately calculate the probability and evolution trend of various risk events based on the initial abnormal probability of the basic event, provide quantitative basis for risk early warning and preventive measures in the production process, and can also trace back the key inducements and their probabilities leading to the results from the abnormal events or risk consequences by means of backward reasoning, complete the cause and effect tracing and explore the abnormal propagation path, and accurately locate the risk source. The bidirectional quantitative analysis ability of the dynamic Bayesian network model constructed by the application not only overcomes the limitations of the traditional fault tree in dealing with dynamic changes and uncertainty, but also makes up for the deficiency of the single dynamic Bayesian network in expressing logical causal relationship, so that the analysis of the chemical production safety risk is improved from static and qualitative analysis to dynamic and fully quantitative level, which helps enterprises to more comprehensively and accurately master the risk situation in the production process, identify potential hidden dangers in advance and optimize the prevention and control strategy, thereby significantly improving the safety and reliability of the chemical production, and reducing the probability of accidents and the loss caused by the accidents.

[0132] Based on the above-mentioned chemical production safety risk intelligent reasoning method, the application further provides a chemical production safety risk intelligent reasoning system, comprising:

[0133] A fault tree analysis model construction module is used to establish a multi-level flow model according to a chemical production process diagram, divide the nodes in the multi-level flow model into basic events, intermediate events and top events by using a hazard and operability analysis technology, and construct a fault tree analysis model according to the relationship between the events.

[0134] An evidence source construction module is used to collect multiple sets of evaluation information of the abnormality of the basic events, convert each piece of evaluation information in each set of evaluation information into a fuzzy risk probability according to an evaluation index, and construct a set of evidence sources.

[0135] An initial probability calculation module is used to fuse the fuzzy risk probability of multiple sets of evidence sources to obtain the initial abnormal probability of each basic event.

[0136] A dynamic Bayesian network model construction module is used to construct a dynamic Bayesian network model by taking each event as a node, taking the conditional transition probability between the nodes in the same time slice, the state transition probability between the nodes in adjacent time slices and the initial abnormal probability of the basic event as parameters.

[0137] The forward reasoning module is used for forward reasoning in the dynamic Bayesian network model to obtain the abnormal probability of each node event in each time slice;

[0138] The backward reasoning module is used for backward reasoning in the dynamic Bayesian network model, sets the abnormal probability of the target node event as 1, and obtains the basic event and the abnormal probability thereof causing the abnormal occurrence of the target node event.

[0139] The application further provides a computer readable storage medium, wherein the computer readable storage medium stores a computer program, and the computer program is executed by a processor to realize the steps of the chemical production safety risk intelligent reasoning method.

[0140] Those skilled in the art should understand that embodiments of the present application can be provided as a method, a system, or a computer program product. Therefore, the present application can adopt a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can adopt a computer program product in the form of being implemented on one or more computer usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer usable program codes.

[0141] The present application is described with reference to flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to embodiments of the present application. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, and the combination of the flows and / or blocks in the flowcharts and / or block diagrams can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing devices generate a means for implementing the functions specified in the flowcharts and / or block diagrams. Figure 1 The functions specified in one or more flows and / or blocks Figure 1 The functions specified in one or more flows and / or blocks

[0142] These computer program instructions can also be stored in a computer readable storage medium capable of guiding a computer or other programmable data processing devices to work in a specific manner, so that the instructions stored in the computer readable storage medium produce a product including instruction means, which implements the functions specified in the flowcharts and / or block diagrams. Figure 1 The functions specified in one or more flows and / or blocks Figure 1 The functions specified in one or more flows and / or blocks

[0143] These computer program instructions can also be loaded into a computer or other programmable data processing devices, so that a series of operational steps are performed on the computer or other programmable data processing devices to generate a computer implemented process, so that the instructions executed on the computer or other programmable data processing devices provide steps for implementing the function specified in the flowchart Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0144] Obviously, the above-mentioned embodiments are only examples for clearly illustrating the present application, and are not intended to limit the embodiments. Based on the above description, other different forms of changes or variations can be made by those skilled in the art. Here, all the embodiments are not required to be exhausted, and the obvious changes or variations derived therefrom are still within the protection scope of the present application.

Claims

1. An intelligent inference method for safety risk in chemical production, characterized in that, The method comprises the following steps: According to the chemical production flow chart, a multi-level flow model is established; the nodes in the multi-level flow model are divided into basic events, intermediate events and top events by using the hazard and operability analysis technology, and a fault tree analysis model is constructed according to the relationship between the events; A plurality of sets of evaluation information of the basic event abnormalities are collected, and each piece of evaluation information in each set of evaluation information is converted into a fuzzy risk probability according to an evaluation index, and a corresponding set of evidence sources is constructed; The fuzzy risk probabilities of the plurality of sets of evidence sources are fused to obtain initial abnormal probabilities of the basic events; In the dynamic Bayesian network model, the abnormal probability of the nth node in the tth time slice is represented as: The method comprises the following steps: A fault tree analysis model construction module is configured to establish a multi-level flow model according to a chemical production flow chart; 2.The chemical production safety risk intelligent reasoning method according to claim 1, characterized in that, The nodes in the multi-level flow model are divided into basic events, intermediate events and top events by using the hazard and operability analysis technology, and a fault tree analysis model is constructed according to the relationship between the events; An evidence source construction module is configured to collect a plurality of sets of evaluation information of basic event abnormalities, and convert each piece of evaluation information in each set of evaluation information into a fuzzy risk probability according to an evaluation index, and construct a corresponding set of evidence sources; where P S is the fuzzy likelihood of the basic event, and (a, b, c, d) is the trapezoidal fuzzy number of the basic event. An initial probability calculation module is configured to fuse the fuzzy risk probabilities of the plurality of sets of evidence sources to obtain initial abnormal probabilities of the basic events; In the dynamic Bayesian network model, the abnormal probability of the nth node in the tth time slice is represented as:

3. The intelligent reasoning method for safety risk of chemical production according to claim 1, characterized in that, ​ ​ 4. The chemical production safety risk intelligent reasoning method according to claim 3, characterized in that, ​ s ij = 1 - d ij (E i , E j ) Among them, E i and E j These are the i-th and j-th evidence sources, respectively, d ij (E i E j ) is E i and E j The Langstroth distance between them, N0 is the number of power sets of elementary events, P i (A k ) and P j (A k ) represent the fuzzy risk probabilities of the k-th focal element in the i-th and j-th evidence sources, respectively; s ij Let be the support between the i-th evidence source and the j-th evidence source.

5. The chemical production safety risk intelligent reasoning method according to claim 3, characterized in that, ​ where s ij is the support between the ith group of evidence sources and the jth group of evidence sources, N0is the number of the power set of the basic events, P i (A k ) and P j (A k ) are the fuzzy risk probabilities of the kth focal element in the ith group of evidence sources and the jth group of evidence sources, respectively.

6. The chemical production safety risk intelligent reasoning method according to claim 3, characterized in that, ​ where s i is the average similarity of the i-th group of evidence sources, s ij is the support between the i-th group of evidence sources and the j-th group of evidence sources, M is the number of evidence sources, and w i is the weight of the i-th group of evidence sources.

7. The chemical production safety risk intelligent reasoning method according to claim 1, characterized in that, ​ 8.The chemical production safety risk intelligent reasoning method according to claim 1, characterized in that, ​ wherein, and is the nth node in the tth time slice and the (t-1)th time slice, respectively, is the abnormal probability of the nth node in the tth time slice, and N is the total number of nodes, is the parent node of is the parent node of represents the conditional transition probability of the nth node in the time slice t, represents the state transition probability of the nth node between the (t-1)th time slice and the tth time slice.

9. A chemical production safety risk intelligent inference system, characterized in that, ​ ​ ​ ​ ​ A dynamic Bayesian network model construction module is configured to construct a dynamic Bayesian network model by taking each event as a node, and taking the conditional transition probability between nodes in the same time slice, the state transition probability of each node between adjacent time slices, and the initial abnormal probability of the basic event as the initial abnormal probability of the dynamic Bayesian network model; A forward reasoning module is configured to perform forward reasoning by using the dynamic Bayesian network model to obtain the abnormal probability of each node event in each time slice; A backward reasoning module is configured to perform backward reasoning by using the dynamic Bayesian network model, set the abnormal probability of the target node event as 1, and obtain the basic event and the abnormal probability thereof that causes the abnormal occurrence of the target node event.

10. A computer-readable storage medium, characterized in that, The computer readable storage medium stores a computer program, and the computer program is executed by the processor to implement the steps of the intelligent reasoning method for chemical production safety risk according to any one of claims 1 to 8.

Images