Research on Human Factors Reliability of Electric Power Enterprises Based on HFCRA Model

. Based on Bayesian network, this paper establishes a human factor analysis tool-Human Factor Classification and Reliability Analysis (HFCRA) model, and analyzes the key factors of human error in power enterprises by combining the human events and data of power safety over the years The research shows that the root causes of human error in electric power enterprises are mainly organizational defects of external human factors and poor personal ability of internal human factors Among them, poor education, training, organizational culture and organizational management quality are the key sub-causes of organizational defects. The lack of basic operation skills and experience knowledge of operators is the key sub-cause of poor personal ability. We should focus on these two kinds of human factors and take corrective measures The internal and external factors of human factors have the characteristics of mutual restriction and transformation, so the occurrence of operational errors in power enterprises should be avoided systematically.


Introduction
Power plant mainly includes three elements: technology, equipment, human-organization relationship and environmental impact system, and its corresponding subsystems are a complex social-technical system Compared with ordinary social security incidents, power safety accidents will not only cause direct losses of personnel and equipment, but also cause huge negative social impacts It is found that human error accounts for more than 70% of the accidents in electric power production safety. [1][2][3][4][5] Prevention and reduction of human accidents has become the key problem to be solved to ensure the safe operation of electric power enterprises. At present, the research on human factor reliability analysis of electric power enterprises is mainly divided into two categories: one is based on expert experience method such as CHANG [6], which studies the relationship between power operation operator behavior and human factor error through a large number of literature analysis and experience judgment; Xi Yongtao et al. [7] use expert technology system to build an intelligent hybrid system model; Li Long and others [8] use fuzzy analytic hierarchy process to divide the degree of human factors into different levels and analyze the relationship between various factors of human errors according to experience The other is the sample statistical method based on human error events, such as Chen Shuai and others [9], which uses primitive event analysis method and regression prediction to analyze and study human error in power enterprises; Rasmussern [10] proposed SRK model to solve the cognitive process patterns corresponding to different behavior types; Li Pengcheng et al. [11] established a data analysis model through event samples and analyzed the influencing factors of human error of power enterprise operators by studying related factors. The above methods provide theoretical basis for human reliability analysis of electric power enterprises, but they all have some limitations The expert experience method relies on expert experience to analyze and judge the human factors according to the statistics of actual events. It lacks subjectivity and uncertainty in analyzing data of historical events and potential human factors that have not been found; Sample statistic method relies on the historical statistic data of human factors that have occurred in the past to predict the potential human factors. It lacks all the data of actual events and can not evaluate the actual events. All human factors have certain dispersion and incompleteness Human factor classification and reliability analysisHFCRA (Human Factor Classification and Reliability AnalysisHFCRA) model is proposed in this paper to study the unsafe behavior of people in the operation of power plants. The model can be based on expert technology system and combined with the existing data of actual events. Based on Bayesian network, the key factors of human factors can be identified by calculation, and targeted corrective measures can be taken to avoid and reduce human errors and reduce the incidence of accidents in power plants.

HFCRA Model Theory
HFCRA is based on Bayesian network to identify the key risk factors and take measures to reduce the probability of risk occurrence by analyzing the contribution value of the interaction between human and system in the accident sequence and the total risk, and to identify the key risk factors and the process of the accident sequence affected by these factors. The analysis method is based on Bayesian network.

Bayesian Network Composition
Bayesian NetworksBN (Bayesian NetworksBN) was first proposed by Pearl [12] It has been widely used in the fields of finance, medical diagnosis, life science, industry, engineering control, aviation and so on. Now it has become an effective theoretical model tool for system uncertainty reasoning and data analysis [13][14][15][16][17][18] The network consists of two parts: Bayesian network graph and Bayesian network parameter Bayesian network is an acyclic directed acyclic graph whose mathematical expression is: Where DAG represents the network structure PT is a set of conditional probability distributions, also known as node probability table , the joint probability distribution of multivariate variables in the system is: Where the marginal probability of Xi is: The example diagram of reasoning model based on Bayesian network is shown in Figure 1 below. As shown in fig. 1, the network nodes are divided into hypothetical nodes (H nodes) and event nodes (E nodes).
Hypothetical nodes represent people's subjective understanding of events and event nodes represent objective facts that occur in a certain time and space range Events can be divided into two categories: events that can be directly observed through event clues and events that cannot be directly observed through event clues The directed connection between nodes in Fig. 1 indicates the causal relationship between the hypothetical node and the event node. Generally, the conditional probability matrix is used to describe the degree of association between the two nodes For a directed connection x-y, the conditional probability matrix is defined as: Each item in the matrix ) ( x y P is represented as:

Bayesian Network Parameters
Bayesian method uses Bayesian formula to solve network parameters. The prior information and sample data are organically combined to improve the estimation accuracy of probability parameters. The calculation formula is as follows: tree Bayesian network is used as the reasoning model, and its structural feature is that each node has only one parent node at most Figure 2 below shows a typical human factor analysis tree Bayesian network structure.  Figure 2 contains four parameter variables: human external factors (node factor HEF), human internal factors (node factor HIF), human behavior (node HA) and behavior consequences (node Hg) From the above Bayesian formula, it can be concluded that the HA probability of nodes is estimated as follows: According to this formula, the latest estimation of network nodes can be completed After completing the node probability calculation, we can mark the conditional probability table on the network to determine the influence between nodes step by step, finally determine which node is the largest influence point and take corrective measures.

Establishment of HFCRA Model
The HFCRA model based on Bayesian network is based on the inverse derivation function of Bayesian network to know the parent node and its posterior probability of occurrence. According to the parent node, the posterior probability of each child node is deduced to analyze the sensitivity of the parent node to each child node In HFCRA model, human behavior is defined as the result of the interaction of human internal factors and human external factors. The HFCRA model is mainly composed of four parts: (1) collecting event human factor data and establishing event human factor Bayesian network nodes through expert knowledge and historical human factor experience feedback; (2) Constructing Bayesian network according to nodes to determine the probability density table of event conditions and the probability distribution function of model parameters of nodes; Using Bayesian formula Bayesian inference theory and algorithm to solve the conditional probability of each node network parameter; (3) According to the probability information of network parameters of each node, the influence value of human factors is sorted to determine the main factors; (4) According to the calculation and analysis results, the priority solution of human factors is formulated. Based on the tree Bayesian network topology method in Fig. 2, the model decomposes human internal factors into ability, mental state, physical state nodes, human ability nodes into experience, skills, learning ability and other physical state nodes into body function, discomfort, sensory characteristics and other mental state nodes into alert stress and so on Human external factors are decomposed into nodes such as organizational environment, environmental nodes are decomposed into tasks, team factors and organizational nodes are decomposed into organizational structure, organizational management and so on The nodes of each level can be decomposed and increased according to the actual events until the most sensitive factors of human factors are obtained.

Human Factor Composition of Power Enterprises
According to the human factor characteristics of power enterprises and the analysis of human factor composition structure of HFCRA model, a total of 645 human factor incident reports of power safety accidents from 2009 to 2019 decompose the human factor composition of power enterprises into the following Table 1. (1) The root causes of human error in power generation enterprises are mainly the organizational defects of human external factors and the poor personal ability of human internal factors, and their probability values reach 66% and 52% respectively.
(2) From the comprehensive consideration of internal and external factors of human factors, organizational defects have the greatest impact on accidents, indicating that organizational management factors are an important root cause of human errors, and should be given great attention; secondly, the lack of personal ability can easily cause accidents. The occurrence of accidents and major safety accidents in power generation enterprises are mainly due to the mutual constraints of organizational defects and personal capabilities.
(3) Secondly, the probability of human error caused by poor mental state is also high, reaching 39%. One should be alert to the negative emotions brought about by mental stress. Such factors will be transformed into false factors of personal lack of ability. (4) The probability of task factors, environmental factors and physical status is low, which is determined by the characteristics of power generation enterprises. It shows that with the improvement of technology and management level of power generation enterprises, equipment reliability, task division and work intensity have been greatly improved. Improvement is not the main reason for human errors in power generation companies, but it also needs to be paid enough attention to. (5) Corrective measures for organizational defects should focus on improving education and training, organizational culture, and organizational management quality, which are the key sub-reasons for human errors; corrective measures for personal capabilities should focus on improving the operator's basic operating skills and experience, through Improve practical operation skills through education and training, and avoid operational errors through multiple channels.

Conclusion
Based on Bayesian network, this paper establishes HFCRA model, which extracts all the existing data on the basis of expert system and historical event experience feedback, and adjusts the prior probability according to the posterior probability to analyze the key factors of human error; The model has the advantages of high actuarial accuracy, good scalability and comprehensive analysis The model is applied to the human reliability analysis of power generation enterprises, and the decomposition structure of human internal factors and external factors is established. By constructing Bayesian network and calculating the probability of node network, the key factors in human error of power generation enterprises are analyzed, and then the corrective measures are carried out pertinently The application of this model can get clear and intuitive results of human factor analysis, which improves the reliability of human factor analysis in power generation enterprises, and is beneficial to avoid and reduce human error in power generation enterprises and reduce the incidence of power safety accidents.