CN109522962B - Chemical plant safety quantitative evaluation method - Google Patents
Chemical plant safety quantitative evaluation method Download PDFInfo
- Publication number
- CN109522962B CN109522962B CN201811409826.8A CN201811409826A CN109522962B CN 109522962 B CN109522962 B CN 109522962B CN 201811409826 A CN201811409826 A CN 201811409826A CN 109522962 B CN109522962 B CN 109522962B
- Authority
- CN
- China
- Prior art keywords
- safety
- chemical plant
- learning
- factors
- bayesian network
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
- G06Q10/063—Operations research, analysis or management
- G06Q10/0635—Risk analysis of enterprise or organisation activities
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/29—Graphical models, e.g. Bayesian networks
Abstract
The invention relates to a safety quantitative evaluation method for a chemical plant. The invention establishes a Bayesian network-based chemical plant safety quantitative analysis research model. In the model, detailed analysis is mainly carried out from the perspective of human factor reliability in seven aspects of organization, information, work design, human-computer system interface, task environment, workplace design and operator characteristics, an questionnaire and an expert judgment method are used for establishing a safety index system of a chemical plant, and a Bayesian network is used for training samples. Finally, Bayes is used for processing and modeling, the safety quantitative value of the chemical plant is finally estimated, and the safety level of the chemical plant is judged. The invention has the characteristics of accurate reasoning, quick calculation, comprehensive evaluation and the like.
Description
Technical Field
The invention belongs to the field of chemical safety, relates to an automatic application technology, and particularly relates to a safety quantitative evaluation method for a chemical plant.
Background
The chemical industry is an important basic industry of national economy of all countries, and makes outstanding contribution to the development of the economy of all countries. However, due to the complex process in the chemical industry, the materials are dangerous, and there are high (low) temperature, high pressure, inflammable, explosive and corrosive operating environments, which makes them an industry with a large potential danger, and once a safety production accident occurs, it often causes serious economic loss, casualties and environmental damage. However, in the last two decades, the number of major accidents has been significantly reduced with the development of society and the advancement of technology. However, once a safety accident occurs in a chemical plant, the cost is still high. Therefore, the reduction of the accident rate of the chemical plant is always the direction of efforts in the chemical industry.
At present, the production technology of the chemical industry is updated day by day, and most of major accidents are caused by human errors. Statistics of data show that industrial accidents during chemical production and storage range from 60-90% due at least in part to human involvement, which is a terrorist rate. In addition, in the petrochemical industry, for example, in oil refineries with high degree of automation, human errors in the cause of industrial accidents account for 50%. It is therefore clear that the relative number of accidents due to human error is increasing, while the relative number of accidents due to technical failures is decreasing. This is caused by two factors. First, improvements and innovations in technology and design are emphasized. Most engineers are interested in developing process plants with high reliability of the equipment, and they focus on the risk caused by technical failures. To this end, it is the equipment and facility aspects that influence operator performance that are the focus of our research, and the analysis and assessment of these aspects are in the area of human factors. Technical, management and human factors should be closely matched to improve the performance of the plant. Second, most work on human error has focused on the phenomenon of human error rather than the root cause. The percentage ranges given above (60-90%) are clearly a very large distribution. This can be explained by the uncertainty that constitutes human error. Some analysts, primarily in accident surveys, attribute an accident as human error. They consider human error to be the failure of a front-line operator to perform the operation correctly or to ignore the operation. This idea is wrong and natural, and is too simplistic. This is like the phenomenon of identifying a disease, without examining the root cause. After all, in most systems today it is not possible to identify a single person causing an accident.
Disclosure of Invention
The invention provides a chemical plant safety quantitative evaluation method aiming at the defects of the prior art.
The invention aims to establish a Bayesian network-based chemical plant safety quantitative analysis research model aiming at some problems in chemical plant safety evaluation. In the model, detailed analysis is mainly performed from the perspective of human factors in seven aspects of organization, information, work design, human-computer system interface, task environment, workplace design and operator characteristics, an questionnaire and an expert judgment method are used for establishing a safety index system of a chemical plant, and a Bayesian network is used for training samples. Finally, Bayes is used for processing and modeling, the safety quantitative value of the chemical plant is finally estimated, and the safety level of the chemical plant is judged, so that the method is used as a method for quantitatively evaluating the safety of the chemical plant.
The method specifically comprises the following steps:
A chemical plant safety analysis research model is built around personnel factors, and the main contents of the model comprise seven aspects of organization, information, work design, human-computer system interface, task environment, workplace design and operator characteristics. Organization is a drive, information is a bridge, work design is a method, a human-computer system interface is a key point, a task environment is a support, workplace design is a guarantee, and operator characteristics are a foundation. They collectively affect the security analysis results.
Step 2: establishing algorithm application flow
The actual modeling process of the bayesian network should be regarded as an overall process to establish the process.
And step 3: determination of Bayesian network nodes for chemical plant risk analysis
A target node of the Bayesian network structure is a chemical plant risk analysis and is defined as W, 30 nodes are divided according to organization, information, work design, human-computer system interface, task environment, workplace design and operator characteristic factors, and the nodes are numbered.
1 Bayesian network assessment criteria definition
And establishing a network structure through expert knowledge and machine learning according to the selected nodes. The influence degrees of all nodes on the safety risk in each accident are different, the influence factors are identified and evaluated in combination with survey report description, the relative influence degrees of all the factors are evaluated according to evaluation standards to give scores, therefore, a Likten five-point scale is selected, the least important factor is 1, and the most important factor is 5.
2 Bayesian network data collection
(1) Investigation of influence degree of safety factors of chemical plant
Questionnaires were prepared against the evaluation criteria for the degree of influence. The reviewer is required to score factors at each level based on their perceptual importance to the cause of the human error. The reviewers or respondents of the questionnaires have background in terms of safety and chemistry. Of all the interviewees, 23 judges were selected.
(2) State estimation survey of various factors of chemical plant safety
A questionnaire was prepared against the evaluation criteria for state estimation. 23 experts in the chemical industry were invited to investigate the state estimation of the chemical plant under study.
3 chemical plant safety level factor estimation
And obtaining data of the degree of influence of each factor on the safety of the chemical plant and the state estimation from the questionnaire survey results. And obtaining a safety evaluation value through the questionnaire data value coordinates of the influence degree and the state evaluation value, and carrying out normalized processing on the data by utilizing a safety level matrix so as to obtain the safety level of each risk.
The safety conditions are classified into 5 levels of S1, S2, S3, S4, and S5, which represent very unsafe, generally safe, safer, and very safe, respectively. According to an algorithm like risk estimation, 25 safety values (of which there are a large number of repetitions) are obtained. The processing of questionnaire data normalized by the security rank matrix can greatly reduce the workload of calculating security assessment values, and all security factors after processing can be measured by 5 ranks (S1, S2, S3, S4, and S5). And counting the proportion of each safety grade in each safety factor, and providing necessary data for establishing a Bayesian network model.
4-pair Bayesian network structure learning
(1) Pre-editing of structural background knowledge of a Bayesian network structure
Theoretically, it is objective and feasible to construct a target network through sample data learning, and the network can be generated only by reasonably defining an evaluation function for evaluating the quality of the target network and operating with software. In order to make the topological structure simple and clear and the calculation quick, the measures such as expert knowledge, report analysis and the like are fully utilized, the variable sequence is determined after the result according to the reason before and the cause, the causal network is established, sample data is imported on the basis to learn, and the hidden association relation between the nodes is further mined. And carrying out preliminary judgment, summary and carding on the causal relationship among the nodes of the established Bayesian network, and then carrying out pre-editing on the structure learning background knowledge by means of Bayesian software.
(2) Bayesian network structure learning
Bayesian application software is used for learning and perfecting a network structure, and 24 evaluation samples are imported into a network as machine learning data. The learning method of the bayesian network structure mainly comprises a statistical test-based method and a search score-based method. A representative algorithm based on search scores is a K2 algorithm, and the method mainly comprises the steps of firstly defining a measure function for evaluating the quality of a network model, then selecting a node with the maximum posterior probability as a father node of the node from an initial network according to a node sequence determined in advance, traversing all the nodes in sequence, and gradually adding the best father node for each variable.
On the basis of editing of network structure background knowledge, machine learning can better mine all possible potential relation node pairs, so that directed arcs among network structure nodes are richer. Due to the complexity of the learning network, the node arcs are checked and selected according to the independence principle of node selection, and a learning structure is obtained.
The accuracy of the network structure of machine learning is closely related to the number of learning samples, and the more 'real' network needs more sample data. Because the research only provides 24 samples, the data for training and learning is limited, and a real 'correct' and concise network structure cannot be obtained, further adjustment and optimization are needed.
Optimization of network structure by Bayesian application software
(1) Causal correlation analysis
In order to further analyze the causal relationship among the risk factors of each factor, the relevance among the factors is judged by means of expert knowledge, and the causal relationship among the risk factors is adjusted according to the judgment result. All the factors are listed for the factors without cause and effect relationship. Potential causal relationships among the nodes are shown through sample data learning. It should be noted that the newly added wired arc is mined from the sample data, and expresses a certain relationship between data, but does not necessarily have a logical relationship between nodes in a true sense, and the connection relationship between the nodes needs to be checked and judged. The result of the causal relationship analysis can be used to reduce the complexity of the network and optimize the network structure.
(2) Optimized editing of structural background knowledge of bayesian network structures
According to the causal correlation analysis result, editing of background knowledge is optimized, irrelevant factors are listed in advance, and the Bayesian network structure after data production optimization can be imported.
(3) Optimized Bayesian network structure
And analyzing the correlation among all safety factors, importing optimized background knowledge editing, and finally obtaining the optimized Bayesian network structure.
Each safety factor comprises five safety states, namely R1, R2, R3, R4 and R5, and before parameter learning, the probability of each network node variable needs to be initialized, namely, each node variable is initialized and assigned according to uniform distribution.
At present, two Bayesian network parameter learning methods are commonly used, namely Bayesian statistics-based estimation (Bayesian estimation) and Maximum likelihood estimation (Maximum likelihood estimation), wherein the Bayesian statistics-based estimation refers to that parameters are regarded as random variables, prior probability can be considered during operation, and the Maximum likelihood estimation refers to that the parameters are regarded as unknown quantification without considering the prior probability. The Bayesian statistics-based estimation is adopted for parameter learning, and prior probability needs to be considered. And obtaining a safety evaluation value aiming at the questionnaire data value coordinates of the influence degree and the state evaluation value, and calculating to obtain the prior probability of all root nodes by utilizing a safety level matrix. And after the sample learning database is imported, parameter learning is carried out, and the prior probabilities of the other root nodes are manually input. After all the probability parameters are input, probability updating is realized, and learning updating of all the node network parameters can be realized.
Step 6 Bayesian network model for chemical plant safety analysis
The parameter learning is established on the basis that the network topology is constructed and optimized, the purpose of the parameter learning is to quantitatively describe the connection strength between nodes of the existing network topology, and the learning result is the final Bayesian network structure constructed by a Bayesian network model of chemical plant safety analysis.
Step 7 quantizes the calculation result
The safety level of the chemical plant is divided into five levels of 1, 2, 3, 4 and 5, which correspond to five conditions of very unsafe, generally safe, safer and very safe of the whole condition of the chemical plant. As can be seen from the final bayesian network structure diagram, the safety level of the chemical plant corresponds to 2% of the 1-level probability, 15% of the 2-level probability, 27% of the 3-level probability, 46% of the 4-level probability, and 11% of the 5-level probability.
The model operation result only represents the safety probability (namely the probability of risk occurrence) of each level of the safety of the chemical plant under the general condition, when the safety early warning is carried out on a specific chemical plant, relevant data of the chemical plant needs to be collected firstly, the background of the chemical plant is known, the safety level of each factor is analyzed according to the actual situation and the characteristics of the chemical plant, the analysis result is brought into the Bayesian network structure model, and the management risk of the chemical plant is early warned by calculating the level value of the safety of the chemical plant. The overall safety level expectation value of the chemical plant is calculated, wherein the safety level 1 of the chemical plant is calculated according to 2%, the safety level 2 of the chemical plant is calculated according to 15%, the safety level 3 of the chemical plant is calculated according to 27%, the safety level 4 of the chemical plant is calculated according to 46%, and the safety level 5 of the chemical plant is calculated according to 11%. Taking the above model operation result as an example, the overall level of the safety of the chemical plant is as follows: 1 × 2% + 15% +3 × 27% +4 × 46% +5 × 11% ═ 3.5. Namely, the safety level of the chemical plant is between 3 and 4 levels, and the method belongs to the approximate comparative safety range.
The invention has the beneficial effects that: the invention provides a Bayesian network-based chemical plant safety quantitative prediction model based on the consideration of various human error influences by combining the characteristics of a chemical plant, wherein the model comprehensively considers seven main safety factors of human error influences in the aspects of organization, information, work design, man-machine system interface, task environment, workplace design and operator characteristics, adopts a Bayesian network modeling method, and utilizes a Bayesian network application simulation to predict the safety quantitative value and the safety level of the chemical plant. By applying the model, the safe quantitative proportion of each influence factor of the chemical plant can be rapidly calculated, and the safety quantitative value and the safety level are passed; the invention has the characteristics of accurate reasoning, quick calculation, comprehensive evaluation and the like.
Drawings
FIG. 1 is a schematic diagram of influence factors of a risk analysis model of a chemical plant.
FIG. 2 is a Bayesian network modeling flow chart.
Fig. 3 is a security level matrix.
Detailed Description
The invention is further described below with reference to the accompanying drawings.
The method specifically comprises the following steps:
The safety production risk management system not only needs to consider the safety of people, but also needs to consider the influence of production, systems, equipment, environment and the like on people. With the progress of various aspects of automation, intelligence and systematization, the accidents of the chemical plants are almost rarely caused by only system, equipment and environment, basically caused by comprehensive reasons, wherein the influence of personnel is not lacked, and the management is performed by personnel, so that the safety analysis research model of the chemical plants is built around personnel factors, and the main contents of the model comprise seven aspects of organization, information, work design, human-computer system interface, task environment, work place design and operator characteristics. Organization is a drive, information is a bridge, work design is a method, a human-computer system interface is a key point, a task environment is a support, workplace design is a guarantee, and operator characteristics are a foundation. They collectively influence the risk analysis results, as shown in fig. 1.
In practice, there are a number of factors to consider when applying an algorithm. The actual modeling process of the bayesian network should be viewed as an integrated process. Because in practice the definition of variables, the selection and processing of data, the selection of algorithms, and the actual modeling all involve many problems, see fig. 2.
And step 3: determination of Bayesian network nodes for chemical plant risk analysis
1 Bayesian network node selection
According to the analysis, a plurality of factors influencing the safety of the chemical plant can be easily found, and the influence factors are reasonably selected to be modeled by combining the characteristics of the Bayesian network according to necessary principles: firstly, a representative principle is adopted, and the selected nodes can reflect comprehensive information embodied by the safety management of the chemical plant and represent the internal characteristics of each element so as to avoid information omission or redundancy. And secondly, an independence principle is adopted, and the information contained between the selected nodes does not have inclusion or cross relation, so that the mutual independence on logics is ensured. And thirdly, an effectiveness principle is adopted, the selected nodes can be extracted and refined from the dangerous goods accident investigation report, and effective acquisition of data information is guaranteed. Therefore, 30 nodes are finally determined through global consideration by combining expert knowledge and analyzing accident investigation reports.
The target node of the bayesian network structure is a chemical plant risk analysis, which is defined as W, and 30 nodes are divided according to organization, information, work design, human-computer system interface, task environment, workplace design and operator characteristic factors, and are numbered as shown in table 1.
TABLE 1 chemical plant safety analysis influencing factors
State definition for 2-Bayesian network nodes
Since each node represents different connotations, it is necessary to explain the node state. For the sake of easy network implementation and operation, expert opinions are consulted, node states are defined in a uniform manner, and there are 5 states for 30 factors, as shown in table 2.
TABLE 2 State estimation of nodes
| |
1 | 2 | 3 | 4 | 5 |
| State of node | Very poor | Is poor | In general | Is preferably used | Is very good |
1 Bayesian network assessment criteria definition
And establishing a network structure through expert knowledge and machine learning according to the selected nodes. The influence degrees of all nodes on the safety risk in each accident are different, the influence factors are identified and evaluated in combination with survey report description, the relative influence degrees of all the factors are evaluated according to evaluation standards to give scores, therefore, a Likter five-point scale is selected, the least important factor is the lowest, the most important factor is the highest, and the evaluation standards are shown in a table 2.
TABLE 3 node evaluation criteria
| Evaluating |
1 | 2 | 3 | 4 | 5 |
| Degree of influence on safety | Is of no great importance | Is not important | In general | Of greater importance | Is very important |
| Chemical plant security level | Is very unsafe | Is not safe | General safety | Is safer | Is very safe |
For the purposes of research convenience and expert scoring uniformity, the rating of S is defined: 5 is taken as very safe, 4 is taken as safer, 3 is taken as safe, 2 is taken as unsafe, and 1 is taken as very unsafe.
2 Bayesian network data collection
(1) Investigation of influence degree of safety factors of chemical plant
Questionnaires were prepared against the evaluation criteria for the degree of influence. The reviewer is required to score factors at each level based on their perceptual importance to the cause of the human error. The reviewers or respondents of the questionnaires have background in terms of safety and chemistry. Of all the interviewees, 23 judges were selected.
TABLE 4 results List of factor influence questionnaires
(2) State estimation survey of various factors of chemical plant safety
A questionnaire was prepared against the evaluation criteria for state estimation. 23 experts in the chemical industry are invited to investigate the state evaluation of the chemical plant studied by the experts, and the investigation result is as follows.
TABLE 5 results List of factor State evaluation questionnaires
3 chemical plant safety level factor estimation
And obtaining data of the degree of influence of each factor on the safety of the chemical plant and the state estimation from the questionnaire survey results. The security assessment value is obtained by coordinating the questionnaire data values of the degree of influence and the state assessment, and the data is normalized using a security level matrix (as shown in fig. 3) to derive a security level for each risk.
In fig. 3, the safety conditions are divided into 5 levels of S1, S2, S3, S4, and S5, which represent very unsafe, generally safe, safer, and very safe, respectively. From fig. 3, 25 safety values (of which there are a large number of repetitions) can be derived according to an algorithm similar to the risk estimation. The processing of questionnaire data normalized by the security rank matrix can greatly reduce the workload of calculating security assessment values, and all security factors after processing can be measured by 5 ranks (S1, S2, S3, S4, and S5). The statistics of the proportion of each security level in each security factor can provide necessary data for establishing a bayesian network model, and the statistics of the security factors involved in the method are shown in table 6.
TABLE 6 statistical results of Security level
4-pair Bayesian network structure learning
(1) Pre-editing of structural background knowledge of a Bayesian network structure
Theoretically, it is objective and feasible to construct a target network through sample data learning, and the network can be generated only by reasonably defining an evaluation function for evaluating the quality of the target network and operating with software. In order to make the topological structure simple and clear and the calculation quick, the measures such as expert knowledge, report analysis and the like are fully utilized, the variable sequence is determined after the result according to the reason before and the cause, the causal network is established, sample data is imported on the basis to learn, and the hidden association relation between the nodes is further mined. And carrying out preliminary judgment, summary and carding on the causal relationship among the nodes of the established Bayesian network, and then carrying out pre-editing on the structure learning background knowledge by means of Bayesian software.
(2) Bayesian network structure learning
Bayesian application software is used for learning and perfecting a network structure, and 24 evaluation samples are imported into a network as machine learning data. The learning method of the bayesian network structure mainly comprises a statistical test-based method and a search score-based method. A representative algorithm based on search scores is a K2 algorithm, and the method mainly comprises the steps of firstly defining a measure function for evaluating the quality of a network model, then selecting a node with the maximum posterior probability as a father node of the node from an initial network according to a node sequence determined in advance, traversing all the nodes in sequence, and gradually adding the best father node for each variable.
On the basis of editing of network structure background knowledge, machine learning can better mine all possible potential relation node pairs, so that directed arcs among network structure nodes are richer. Due to the complexity of the learning network, the node arcs are checked and selected according to the independence principle of node selection, and a learning structure is obtained.
The accuracy of the network structure of machine learning is closely related to the number of learning samples, and the more 'real' network needs more sample data. Because the research only provides 24 samples, the data for training and learning is limited, and a real 'correct' and concise network structure cannot be obtained, further adjustment and optimization are needed.
Optimization of network structure by Bayesian application software
(1) Causal correlation analysis
In order to further analyze the causal relationship among the risk factors of each factor, the relevance among the factors is judged by means of expert knowledge, and the causal relationship among the risk factors is adjusted according to the judgment result. All the factors are listed for the factors without cause and effect relationship. Potential causal relationships among the nodes are shown through sample data learning. It should be noted that the newly added wired arc is mined from the sample data, and expresses a certain relationship between data, but does not necessarily have a logical relationship between nodes in a true sense, and the connection relationship between the nodes needs to be checked and judged. The result of the causal relationship analysis can be used to reduce the complexity of the network and optimize the network structure.
TABLE 7 factor relationship Table based on structural background knowledge
(2) Optimized editing of structural background knowledge of bayesian network structures
According to the causal correlation analysis result, editing of background knowledge is optimized, irrelevant factors are listed in advance, and the Bayesian network structure after data production optimization can be imported.
(3) Optimized Bayesian network structure
And analyzing the correlation among all safety factors, importing optimized background knowledge editing, and finally obtaining the optimized Bayesian network structure.
Each safety factor comprises five safety states, namely R1, R2, R3, R4 and R5, and before parameter learning, the probability of each network node variable needs to be initialized, namely, each node variable is initialized and assigned according to uniform distribution.
At present, two Bayesian network parameter learning methods are commonly used, namely Bayesian statistics-based estimation (Bayesian estimation) and Maximum likelihood estimation (Maximum likelihood estimation), wherein the Bayesian statistics-based estimation refers to that parameters are regarded as random variables, prior probability can be considered during operation, and the Maximum likelihood estimation refers to that the parameters are regarded as unknown quantification without considering the prior probability. The Bayesian statistics-based estimation is adopted for parameter learning, and prior probability needs to be considered. And (4) obtaining a safety evaluation value aiming at the questionnaire data value coordinates of the influence degree and the state evaluation value, and calculating to obtain the prior probability of all root nodes by using a safety level matrix, which is specifically shown in table 6.
And after the sample learning database is imported, parameter learning is carried out, and the prior probabilities of the other root nodes are manually input. After all the probability parameters are input, probability updating is realized, and learning updating of all the node network parameters can be realized.
Step 6 Bayesian network model for chemical plant safety analysis
The parameter learning is established on the basis that the network topology is constructed and optimized, the purpose of the parameter learning is to quantitatively describe the connection strength between nodes of the existing network topology, and the learning result is the final Bayesian network structure constructed by a Bayesian network model of chemical plant safety analysis.
Step 7 quantizes the calculation result
The safety level of the chemical plant is divided into five levels of 1, 2, 3, 4 and 5, which correspond to five conditions of very unsafe, generally safe, safer and very safe of the whole condition of the chemical plant. As can be seen from the final bayesian network structure diagram, the safety level of the chemical plant corresponds to 2% of the 1-level probability, 15% of the 2-level probability, 27% of the 3-level probability, 46% of the 4-level probability, and 11% of the 5-level probability.
The model operation result only represents the safety probability (namely the probability of risk occurrence) of each level of the safety of the chemical plant under the general condition, when the safety early warning is carried out on a specific chemical plant, relevant data of the chemical plant needs to be collected firstly, the background of the chemical plant is known, the safety level of each factor is analyzed according to the actual situation and the characteristics of the chemical plant, the analysis result is brought into the Bayesian network structure model, and the management risk of the chemical plant is early warned by calculating the level value of the safety of the chemical plant. The overall safety level expectation value of the chemical plant is calculated, wherein the safety level 1 of the chemical plant is calculated according to 2%, the safety level 2 of the chemical plant is calculated according to 15%, the safety level 3 of the chemical plant is calculated according to 27%, the safety level 4 of the chemical plant is calculated according to 46%, and the safety level 5 of the chemical plant is calculated according to 11%. Taking the above model operation result as an example, the overall level of the safety of the chemical plant is as follows: 1 × 2% + 15% +3 × 27% +4 × 46% +5 × 11% ═ 3.5. Namely, the safety level of the chemical plant is between 3 and 4 levels, and the method belongs to the approximate comparative safety range.
Claims (1)
1. A chemical plant safety quantitative evaluation method is characterized by comprising the following steps:
step 1, analysis of influence factors of chemical plant safety
Establishing a chemical plant safety analysis research model around personnel factors, wherein the main contents of the model comprise seven aspects of organization, information, work design, human-computer system interface, task environment, workplace design and operator characteristics; organization is driving, information is a bridge, work design is a method, a human-computer system interface is a key point, a task environment is supporting, workplace design is guaranteeing, and operator characteristics are basic; they influence the risk analysis results together;
step 2: establishing algorithm application flow
The actual modeling process of the Bayesian network is regarded as an integral process, and the process is established;
and step 3: determination of Bayesian network nodes for chemical plant risk analysis
The target node of the Bayesian network structure is a chemical plant risk analysis and is defined as W, 30 nodes are divided according to organization, information, work design, human-computer system interface, task environment, workplace design and operator characteristic factors, and the nodes are numbered;
step 4, establishing a Bayesian network structure for risk analysis of chemical plants
1 Bayesian network assessment criteria definition
Establishing a network structure through expert knowledge and machine learning according to the selected nodes; the influence degrees of all nodes on the safety risk in each accident are different, all influence factors are identified and evaluated in combination with survey report description, the relative influence degrees of all factors are evaluated according to evaluation standards to give scores, a Likter five-point scale is selected, the least important =1 is the lowest, and the most important =5 is the highest;
2 Bayesian network data collection
(1) Investigation of influence degree of safety factors of chemical plant
Making a questionnaire according to the evaluation standard of the influence degree; the judges are required to score factors at all levels according to the perceptual importance of human error causes; the reviewers or the interviewees of the questionnaires have the background in the aspects of safety and chemical engineering; of all the interviewees, 23 panelists were selected;
(2) state estimation survey of various factors of chemical plant safety
Making a questionnaire by contrasting the state evaluation standard; inviting 23 experts in the chemical industry to investigate the state estimation of the chemical plant;
3 chemical plant safety level factor estimation
Data of the degree of influence of each factor on the safety of the chemical plant and the state estimation can be obtained from questionnaire survey results; obtaining a safety evaluation value through questionnaire data value coordinates of the influence degree and the state evaluation value, and carrying out normalized processing on data by utilizing a safety level matrix so as to obtain the safety level of each risk;
4-pair Bayesian network structure learning
(1) Pre-editing of structural background knowledge of a Bayesian network structure
In order to make the topological structure simple and clear and calculate quickly, expert knowledge and report analysis means are fully utilized, the variable sequence is determined after the result according to the reason before and the reason before, a causal network is established, sample data is imported on the basis for learning, and the hidden association relationship between the nodes is further mined; preliminarily judging and summarizing causal relationships among the nodes of the established Bayesian network, and editing the structure learning background knowledge in advance by means of Bayesian software;
(2) bayesian network structure learning
Learning and perfecting a network structure by using Bayesian application software, and importing 24 evaluation samples into a network as machine learning data;
on the basis of editing of network structure background knowledge, machine learning can better mine all possible potential relation node pairs, so that directed arcs among network structure nodes are enriched; due to the complexity of the learning network, the node arcs are checked and selected according to the independence principle of node selection to obtain a learning structure;
optimization of network structure by Bayesian application software
(1) Causal correlation analysis
In order to further analyze the causal relationship among the risk factors of each factor, the relevance among the factors is judged in an expert knowledge mode, and the causal relationship among the risk factors is adjusted according to the judgment result; all the factors without causal relationship are listed; potential causal relationships among the nodes are shown through sample data learning;
(2) optimized editing of structural background knowledge of bayesian network structures
Optimizing background knowledge editing according to the causal correlation analysis result, listing irrelevant factors in advance, and importing the data into a Bayesian network structure after data production optimization;
(3) optimized Bayesian network structure
Analyzing according to the correlation among all safety factors, importing optimized background knowledge for editing, and finally obtaining an optimized Bayesian network structure;
step 5 learning of Bayesian network parameters for chemical plant risk analysis
R1, R2, R3, R4, R5, before parameter learning, carry on the initialization operation to the probability of every network node variable, namely carry on the initialization assignment to every node variable according to the uniform distribution;
carrying out parameter learning by adopting Bayesian statistics-based estimation, wherein prior probability needs to be considered; obtaining a safety evaluation value aiming at the questionnaire data value coordinates of the influence degree and the state evaluation value, and calculating to obtain the prior probability of all root nodes by utilizing a safety level matrix; after a sample learning database is imported, parameter learning is carried out, and the prior probability of other root nodes is manually input; after all the probability parameters are input, probability updating is realized, namely learning updating of all the node network parameters can be realized;
step 6 Bayesian network model for chemical plant safety analysis
The parameter learning is established on the basis that the network topological structure is constructed and optimized, the purpose of the parameter learning is to quantitatively describe the connection strength between nodes of the existing network topological structure, and the learning result is actually the final Bayesian network structure constructed by a Bayesian network model for the safety analysis of a chemical plant;
step 7 quantizes the calculation result
The safety level of the chemical plant is divided into five levels of 1, 2, 3, 4 and 5, which correspondingly reflects five conditions of very unsafe, generally safe, safer and very safe of the whole condition of the chemical plant.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811409826.8A CN109522962B (en) | 2018-11-23 | 2018-11-23 | Chemical plant safety quantitative evaluation method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811409826.8A CN109522962B (en) | 2018-11-23 | 2018-11-23 | Chemical plant safety quantitative evaluation method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109522962A CN109522962A (en) | 2019-03-26 |
| CN109522962B true CN109522962B (en) | 2021-07-13 |
Family
ID=65779039
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811409826.8A Active CN109522962B (en) | 2018-11-23 | 2018-11-23 | Chemical plant safety quantitative evaluation method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109522962B (en) |
Families Citing this family (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109740950A (en) * | 2019-01-09 | 2019-05-10 | 哈尔滨理工大学 | A kind of safety behavior analysis method based on the equation of structure |
| CN110942260B (en) * | 2019-12-12 | 2024-02-13 | 长安大学 | College traffic safety evaluation method based on Bayesian maximum entropy |
| CN111178404B (en) * | 2019-12-17 | 2023-06-30 | 中国航天标准化研究所 | Human-computer interaction system task reliability modeling method considering path dependency effect |
| CN113505979A (en) * | 2021-07-01 | 2021-10-15 | 国网山西省电力公司运城供电公司 | Power safety behavior assessment method used in smart grid environment |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105550486A (en) * | 2016-03-07 | 2016-05-04 | 天津工业大学 | Petrochemical enterprise deluge system reliability evaluating method |
| CN107480895A (en) * | 2017-08-19 | 2017-12-15 | 中国标准化研究院 | A kind of reliable consumer goods methods of risk assessment based on Bayes enhancing study |
| CN108345986A (en) * | 2018-01-19 | 2018-07-31 | 杭州电子科技大学 | A kind of chemical industry danger source dynamic quantization appraisal procedure |
| CN108776804A (en) * | 2018-04-24 | 2018-11-09 | 山东电力设备有限公司 | A kind of transformer specialization repair method based on portable terminal |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| DE102005056265A1 (en) * | 2005-11-14 | 2007-05-16 | Pilz Gmbh & Co Kg | Device and method for monitoring a room area, in particular for securing a danger zone of an automated system |
| US9218233B2 (en) * | 2012-07-24 | 2015-12-22 | Paul Venditti | Systems and methods for control reliability operations |
-
2018
- 2018-11-23 CN CN201811409826.8A patent/CN109522962B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105550486A (en) * | 2016-03-07 | 2016-05-04 | 天津工业大学 | Petrochemical enterprise deluge system reliability evaluating method |
| CN107480895A (en) * | 2017-08-19 | 2017-12-15 | 中国标准化研究院 | A kind of reliable consumer goods methods of risk assessment based on Bayes enhancing study |
| CN108345986A (en) * | 2018-01-19 | 2018-07-31 | 杭州电子科技大学 | A kind of chemical industry danger source dynamic quantization appraisal procedure |
| CN108776804A (en) * | 2018-04-24 | 2018-11-09 | 山东电力设备有限公司 | A kind of transformer specialization repair method based on portable terminal |
Non-Patent Citations (1)
| Title |
|---|
| 危险化学品化工厂风险评价研究;冯云;《中国优秀硕士学位论文全文数据库 工程科技Ⅰ辑》;20121115(第11期);第B026-8页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109522962A (en) | 2019-03-26 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109522962B (en) | Chemical plant safety quantitative evaluation method | |
| CA2843276A1 (en) | Dynamic outlier bias reduction system and method | |
| Pani et al. | A survey of data treatment techniques for soft sensor design | |
| CN110619482A (en) | Driving behavior scoring method based on logistic regression and single-level analysis weighting method | |
| CN103577676A (en) | Grey weighting method for sewage treatment process comprehensive evaluation | |
| CN105975797A (en) | Product early-fault root cause recognition method based on fuzzy data processing | |
| Peng et al. | Reliability modelling and assessment of a heterogeneously repaired system with partially relevant recurrence data | |
| EP4080429A1 (en) | Technology readiness level determination method and system based on science and technology big data | |
| Eskandari et al. | Offering a preventive solution to defects in commercial building facility system using BIM | |
| Yu et al. | A framework to identify and respond to weak signals of disastrous process incidents based on FRAM and machine learning techniques | |
| Isah et al. | Assessment of risk impact on road project using deep neural network | |
| Tripathy et al. | Explaining Anomalies in Industrial Multivariate Time-series Data with the help of eXplainable AI | |
| Meritxell et al. | On the evaluation, management and improvement of data quality in streaming time series | |
| Rosli et al. | The design of a software fault prone application using evolutionary algorithm | |
| Aboshady et al. | A fuzzy risk management framework for the Egyptian real estate development projects | |
| CN111553581A (en) | Equipment maintainability evaluation model based on entropy value | |
| CN101527016A (en) | Screening method of compliant conditions of entry and exit inspection and quarantine | |
| CN116384732A (en) | Intelligent assessment method, system, storage medium and computing device for station pipeline risk | |
| Huang et al. | Confidence-based quantitative risk analysis for offshore accidental hydrocarbon release events | |
| CN114548494A (en) | Visual cost data prediction intelligent analysis system | |
| CN112184040A (en) | Platform for evaluating software engineering capability based on behavior and learning data | |
| CN114048592A (en) | Finish rolling whole-flow distributed operation performance evaluation and non-optimal reason tracing method | |
| Prasad et al. | Predictive maintenance in forging industry | |
| Pouri et al. | Spatial analysis and failure management in water distribution networks using fuzzy inference system | |
| Apollo et al. | Observation value analysis–integral part of Bayesian diagnostics |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| CB03 | Change of inventor or designer information |
Inventor after: Jiang Peng Inventor after: Zheng Song Inventor after: Liu Jun Inventor after: Song Qiusheng Inventor after: Xu Huan Inventor after: Men Jinkun Inventor after: Zhou Shuo Inventor before: Jiang Peng Inventor before: Song Qiusheng Inventor before: Zheng Song Inventor before: Kong Yaguang Inventor before: Wang Ling Inventor before: Yao Xuefei Inventor before: Lu Jian |
|
| CB03 | Change of inventor or designer information | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |