Electric power operation personnel behavior risk identification early warning method and system
The system quantitatively evaluates power workers' behavioral risks using a decision tree and random forest model, addressing inefficiencies in existing qualitative methods to enhance risk identification and warning accuracy.
Patent Information
- Application Number
- CN202410991204.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-23
- Publication Date
- 2024-12-13
AI Technical Summary
Existing methods for identifying behavioral risks of electric power workers are qualitative and inefficient, failing to provide quantitative assessments and accurately distinguish individual operator differences, leading to ineffective risk identification and warning.
A risk identification and early warning system using an information collection, reading, and analysis unit, coupled with a single decision tree scoring model and random forest model, to quantify operator behavior risks based on attributes like length of service, job title, and violation points, and update risk scores dynamically.
The system provides accurate, quantitative risk evaluation of power workers' behaviors, enhancing recognition accuracy and timely risk warnings by leveraging a random forest model to integrate scores from multiple decision trees.
Smart Images
Figure CN119130108A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of electric power operation early warning, and in particular to a method and system for identifying and warning electric power operator behavior risks. Background Art
[0002] The operating behavior of power workers during circuit faults or power installation operations not only affects the safe operation of power equipment, but also affects the safety of the workers themselves. Therefore, identifying and warning the safety of power workers' operating behaviors is an important means to protect power workers.
[0003] The current process of identifying behavioral risks of power workers is often based on the assessment results of the workers' physical or mental health conditions, and rarely on the workers' own work behaviors. Even if a small number of people's behaviors are identified to have certain risks, they are mostly based on qualitative analysis, and it is difficult to use quantitative assessment methods to give quantitative risk values, resulting in low efficiency in identifying behavioral risks of power workers and difficulty in timely warning of risky behaviors.
[0004] Although the document entitled "Illegal behaviors and accident prevention in power operations" discussed common illegal behaviors in power operations, such as failure to wear safety protection equipment as required, illegal operation of equipment, failure to perform pre-operation inspections, etc., and analyzed the causes of accidents caused by these illegal behaviors, and proposed some targeted preventive measures, such as strengthening safety training, improving operating procedures, and strictly implementing safety supervision, it did not provide a method for quantitatively evaluating the risks caused by personnel behavior before the operation, making it difficult to identify and warn of risky behaviors of operators.
[0005] The prior art closest to the present invention, "Quantitative Assessment Model for Personal Risk in Power Operations ([1] Yang Yuequn. (2015). Doctoral dissertation, South China University of Technology)", can obtain a certain degree of weight using traditional methods such as accident trees, but it cannot distinguish the behavioral differences of similar people, resulting in the inability to accurately identify the personal behavioral risks of different operators. To this end, we propose a method and system for identifying and warning the behavioral risks of power operators that can improve the recognition accuracy. Summary of the invention
[0006] The purpose of the present invention is to provide a method and system for identifying and warning of behavioral risks of power workers, so as to solve the problems raised in the above-mentioned background technology.
[0007] To achieve the above object, the present invention provides the following technical solutions:
[0008] On the one hand, a risk identification and early warning system for electric power workers' behavior is provided, the risk identification and early warning system comprising an information collection unit, an information reading unit, an information analysis unit and an information output unit;
[0009] The information collection unit collects attribute data that affects the safety of the operator's behavior, and the attribute data includes the operator's length of service, job title, violation penalty points, latest assessment score, and work certificate data;
[0010] The information reading unit reads and identifies the data collected by the information collecting unit, and sends the read data to the information analyzing unit;
[0011] The information analysis unit uses a single decision tree scoring model to divide leaf nodes, obtains a quantitative score of the operator's behavior risk, and sends the score result to the information output unit;
[0012] The information output unit outputs the received scoring result.
[0013] On the other hand, a method for identifying and warning the behavior risk of electric power workers is also provided, which is applied to the electric power worker behavior risk identification and warning system, and comprises the following steps:
[0014] S1: Collect attribute data that affects people's behavior;
[0015] S2: Divide the attribute data collected in step S1 into modeling segment data and verification segment data;
[0016] S3: All data were scored using expert scoring and analytic hierarchy process;
[0017] S4: Establish a personnel behavior risk model for the modeling segment data and calculate the test segment error with the test segment data to obtain the
[0018] The weight of the machine forest is used to determine the final risk value of personnel behavior;
[0019] S5: Using the personnel behavior risk value in the above step S4 as a behavior factor, assessing the behavior factor risk level.
[0020] As a further solution of the present invention: randomly extracting modeling segment data of N operators and filling them into a scoring table, N≥30; having M scorers respectively perform risk scoring on the N operators, and obtaining sample data of no less than N×M operator scoring results;
[0021] Table 1 shows the risk description and probability value caused by personnel operations:
[0022] Risk Level Risk Description Probability value 1 It can't happen 0.0002 2 Rarely occurs 0.002 3 Occasionally 0.02 4 Frequent 0.2 5 Frequent 0.9
[0023] As a further solution of the present invention: the sample data of the operator's scoring results is taken as set Y, and the attribute data of the operator is taken as feature variable X i (k) (k = length of service, job title, penalty points for violation, latest assessment score, work qualification certificate; i is the order mark of the operator, i≤N).
[0024] As a further solution of the present invention: each time W data from N×M are extracted as modeling data, and the remaining N×MW data are used as test data to construct a decision tree personnel quantitative scoring model;
[0025] The W data are used to construct a training set D, and the scoring scores of the raters are marked as y i , and iterates through all the feature values in the current set one by one by traversing the values one by one;
[0026] When the kth feature variable x in the training set i (k) Given a value s, i (k) As the segmentation variable, s is the segmentation point, and D can be divided into two regions R1(k,s) and R2(k,s) according to s, that is, R1(k,s)={(x i (k) ,y i )|x i (k) <=s},R2(k,s)={(x i (k) ,y i )|x i (k) >s}. The value of s should minimize the sum of square errors between the two regions, that is, satisfy the formula:
[0027]
[0028] in, and are y in R1(k,s) and R2(k,s) respectively i The mean of .
[0029] As a further solution of the present invention: according to the principle of minimizing the sum of squared errors, the optimal value is selected as the cutting point to obtain the minimum value s and the corresponding characteristic variable X i (k) ;in,
[0030] The minimum value s is the current optimal split point;
[0031] The corresponding characteristic variable X i (k)Then it is the optimal segmentation variable;
[0032] SE k The characteristic variable x i (k) is the minimum sum of squared errors obtained during partitioning.
[0033] As a further solution of the present invention: SE is calculated for each characteristic variable k , and record each SE k The corresponding s value is taken from all s values to find the minimum value and find the value that makes SE k The value of s when the result is the smallest and the corresponding characteristic variable x i (k) Take, according to x i (k) The values of and s are calculated to determine and The training set D is used as the parent node, and R1(k,s) and R2(k,s) are its leaf nodes, marking the depth of the decision tree as one layer.
[0034] As a further solution of the present invention: R1(k, s) and R2(k, s) are used as new training sets to continuously traverse the partitioning and solve and generate new leaf nodes multiple times until all samples are cut or the maximum depth of the given decision tree is reached, and a regression tree is generated, and the corresponding mean value in the sample set of the last layer of leaf nodes is This is the scoring result of the decision tree model.
[0035] As a further solution of the present invention: the root mean square error RMSE is calculated based on the scoring results and the scoring results of the scorers:
[0036]
[0037] According to the root mean square error results of multiple decision trees, calculate the error coefficient C t :
[0038]
[0039] As a further solution of the present invention: the error coefficient C t Multiply it by the score given to the person by the corresponding decision tree and add them together to get the final random forest person score result:
[0040] Compared with the prior art, the present invention has the following beneficial effects:
[0041] The present invention establishes a random forest model of personnel behavior risk by constructing a personnel attribute factor library. It can quantitatively evaluate the risk scores of different operators' work behaviors based on the different attributes of similar operators that affect the safety of their work behaviors. At the same time, it can update the risk value scores according to changes in the operators' own attributes, thus avoiding the qualitative and non-quantitative problems existing in traditional evaluation systems. BRIEF DESCRIPTION OF THE DRAWINGS
[0042] Figure 1 is a flowchart of the process of the present invention;
[0043] Figure 2 A single decision tree operator behavior risk scoring model diagram of the present invention;
[0044] Figure 3 It is a system module diagram of the present invention. DETAILED DESCRIPTION
[0045] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0046] like Figure 1 The flowchart of the processing of the present invention is shown, the modeling segment data of N operators are randomly selected and filled into the scoring table, N≥30; the risk scores of the N operators are respectively performed by M scorers, and sample data of the scoring results of no less than N×M operators are obtained;
[0047] Table 1 shows the risk description and probability value caused by personnel operations:
[0048] Risk Level Risk Description Probability value 1 It can't happen 0.0002 2 Rarely occurs 0.002 3 Occasionally 0.02 4 Frequent 0.2 5 Frequent 0.9
[0049] Preferably, the sample data of the operator's scoring results is taken as set Y, and the attribute data of the operator is taken as feature variable X i (k) (k = length of service, job title, penalty points for violation, latest assessment score, work qualification certificate; i is the order mark of the operator, i≤N).
[0050] Preferably, W data from N×M are extracted each time as modeling data, and the remaining N×MW data are used as test data to construct a decision tree personnel quantitative scoring model;
[0051] For example: the number of operators is N=30, the number of scorers is M=30, and the 30 scorers score the 30 operators in turn, generating a total of N×M=30×30=900 score sheets. Each time, W=720 samples are extracted for modeling, of which 720 samples are used as modeling data, and the remaining N×MW=900-720=180 samples are used as test data. The principle of a single decision tree scoring model is an algorithm that measures and divides leaf nodes by the sum of squared errors. Its essence is to divide the feature space of the sample period into two left and right nodes according to the condition of the minimum sum of squared errors. The two nodes correspond to different values, and the results of each node are then divided in turn, that is, a number of units are obtained. Each divided unit has a specific output. In the data application of the test period, as long as it is classified into a unit according to the characteristics, the corresponding corrected output value can be obtained. After the calculation is completed, a binary tree will be generated, and its structure has the advantage of being simple and easy to understand.
[0052] The W data are used to construct a training set D, and the scoring scores of the raters are marked as y i , and iterates through all the feature values in the current set one by one by traversing the values one by one;
[0053] When the kth feature variable x in the training set i (k) Given a value s, i (k) As the segmentation variable, s is the segmentation point, and D can be divided into two regions R1(k,s) and R2(k,s) according to s, that is, R1(k,s)={(x i (k) ,y i )|x i (k) <=s},R2(k,s)={(x i (k) ,y i )|x i (k) >s}. The value of s should minimize the sum of square errors between the two regions, that is, satisfy the formula:
[0054]
[0055] in, and are y in R1(k,s) and R2(k,s) respectively i The mean of .
[0056] Preferably, according to the principle of minimizing the sum of squared errors, the optimal value is selected as the split point to obtain the minimum value s and the corresponding characteristic variable X i(k) ;in,
[0057] The minimum value s is the current optimal split point;
[0058] The corresponding characteristic variable X i (k) Then it is the optimal segmentation variable;
[0059] SE k The characteristic variable x i (k) is the minimum sum of squared errors obtained during partitioning.
[0060] Preferably, SE is calculated for each feature variable k , and record each SE k The corresponding s value is taken from all s values to find the minimum value and find the value that makes SE k The value of s when the result is the smallest and the corresponding characteristic variable x i (k) Take, according to x i (k) The values of and s are calculated to determine and The training set D is used as the parent node, and R1(k,s) and R2(k,s) are its leaf nodes, marking the depth of the decision tree as one layer.
[0061] Preferably, R1(k, s) and R2(k, s) are used as new training sets to continuously traverse the partitioning and solve and generate new leaf nodes multiple times until all samples are cut or the maximum depth of the given decision tree is reached, and a regression tree is generated, and the corresponding mean value in the sample set of the last layer of leaf nodes is This is the scoring result of the decision tree model.
[0062] The root mean square error RMSE is calculated based on the scoring results and the scoring results of the scorers:
[0063]
[0064] According to the root mean square error results of multiple decision trees, calculate the error coefficient C t :
[0065]
[0066] Preferably, the error coefficient C t Multiply it by the score given to the person by the corresponding decision tree and add them together to get the final random forest person score result:
[0067] Preferably, the age of the operating personnel is used as the segmentation variable, and the sample set is divided into sub-sample sets R1(k,s) (i.e., personnel with an age less than or equal to this value) and R2(k,s) (i.e., personnel with an age greater than this value) by the value s greater than the minimum and maximum ages of the personnel in the sample. i Calculate the corresponding mean and Further calculation to obtain the error square sum SE k (In this case, k is the employee’s age);
[0068] Change the value of s, divide the sample set again, and calculate the error sum of squares SE again k (In this case, k is the employee's age), and the above steps are repeated for each attribute of the employee (i.e., k is changed to other attributes). Finally, the attribute and its partition value s that minimize the sum of squared errors in this round are obtained, which is the first-layer node of the decision tree model. The above method is continued until all samples are cut;
[0069] After forming a single decision tree based on the above method, 180 samples of the test segment are entered into the decision tree model to obtain the model scoring result. The root mean square error is calculated by comparing the result with the actual personnel score. The root mean square error formula is shown in 2, where is the result of the decision tree scoring in the test sample, y i To test the scores of people in the sample:
[0070]
[0071] According to the root mean square error results of multiple decision trees, calculate the error coefficient C t , where C t is the error coefficient of a single decision tree, t is 1, 2, 3, 4...20, representing 20 different decision tree identifiers, T is 20;
[0072] RMSE t The root mean square error calculated by formula 2 for the test period of different decision trees:
[0073]
[0074] Multiply the coefficient by the score given to the person by the corresponding decision tree and add them together to get the final random forest person score result as follows:
[0075]
[0076] like Figure 2The figure shows a single decision tree operator behavior risk scoring model diagram of the present invention, where "gl" represents length of service, "fs" represents the most recent assessment score, "zc" represents the job title, "gzz" represents the work permit, "cs" represents the number of violations, and "value" represents the risk score derived from logical judgment;
[0077] Taking the first level of analysis as an example, it means that if the input employee age is less than 0.65, the risk score of this person is 0.5. Then, the various attributes of the person are input into another decision tree to obtain the person score. After calculating the person scores of all 20 decision trees, the error coefficient of the corresponding decision tree is multiplied by the person score output by the corresponding decision tree, and the sum of the above results of the 20 decision trees is calculated to obtain the final person behavior risk score.
[0078] The coefficients corresponding to the 20 decision trees in the present invention are [0.0756, 0.0703, 0.0754, 0.0029, 0.0376, 0.0753, 0.0859, 0.036, 0.0039, 0.0917, 0.049, 0.0426, 0.0038, 0.0696, 0.093, 0.0455, 0.0029, 0.019, 0.0719, 0.0481].
[0079] like Figure 3 The system module diagram of the present invention is shown, wherein the risk identification and early warning system comprises an information collection unit, an information reading unit, an information analysis unit and an information output unit;
[0080] The information collection unit collects attribute data that affects the safety of the operator's behavior, and the attribute data includes the operator's length of service, job title, violation penalty points, latest assessment score, and work certificate data;
[0081] The information reading unit reads and identifies the data collected by the information collecting unit, and sends the read data to the information analyzing unit;
[0082] The information analysis unit uses a single decision tree scoring model to divide leaf nodes, obtains a quantitative score of the operator's behavior risk, and sends the score result to the information output unit;
[0083] The information output unit outputs the received scoring result.
[0084] The method for identifying and warning the behavior risk of electric power workers is applied to the system for identifying and warning the behavior risk of electric power workers, and comprises the following steps:
[0085] S1: Collect attribute data that affects people's behavior;
[0086] S2: Divide the attribute data collected in step S1 into modeling segment data and verification segment data;
[0087] S3: All data were scored using the analytic hierarchy process;
[0088] S4: Establish a personnel behavior risk model based on the modeling segment data, calculate the personnel behavior risk score for the test segment data, and
[0089] Calculate the root mean square error to calculate the coefficient of the decision tree and finally determine the risk value of personnel behavior;
[0090] S5: Using the personnel behavior risk value in the above step S4 as a behavior factor, assessing the behavior factor risk level.
[0091] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.
[0092] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.
Claims
1. A behavioral risk identification and early warning system for power workers. It is characterized in that The risk identification and early warning system includes an information collection unit, an information reading unit, an information analysis unit and an information output unit; The information collection unit collects attribute data that affects the safety of the operator's behavior, and the attribute data includes the operator's length of service, job title, violation penalty points, latest assessment score, and work certificate data; The information reading unit reads and identifies the data collected by the information collecting unit, and sends the read data to the information analyzing unit; The information analysis unit uses a single decision tree scoring model to divide leaf nodes, obtains a quantitative score of the operator's behavior risk, and sends the score result to the information output unit; The information output unit outputs the received scoring result.
2. A method for identifying and warning the behavior risk of electric power workers, applied to the electric power workers behavior risk identification and warning system, The following steps are included: Collect attribute data that influences people’s behavior; Dividing the attribute data collected in step S1 into modeling segment data and verification segment data; The analytic hierarchy process is used to score all data to obtain the personnel behavior risk score; Establish a personnel behavior risk model with modeling segment data, score multiple personnel behavior risk models with test segment data, obtain the optimal personnel behavior risk model, and determine the final personnel behavior risk value; The personnel behavior risk value in the above step S4 is used as a behavior factor to assess the behavior factor risk level.
3. According to claim 2, a method for identifying and warning of behavioral risks of electric power workers, Features: The data of N operators are randomly selected and filled into a scoring table, where N≥30; and the risk scores of the N operators are respectively performed by M scorers to obtain sample data of no less than N×M operator scoring results.
4. According to claim 3, a method for identifying and warning of behavioral risks of electric power workers, Features: The sample data of the operator's scoring results is taken as set Y, and the operator's attribute data is taken as feature variable X i (k) (k = length of service, job title, penalty points for violation, latest assessment score, work qualification certificate; i is the order mark of the operator, i≤N).
5. According to claim 3, a method for identifying and warning of behavioral risks of electric power workers, Features: Each time, W data from N×M are extracted as modeling data, and the remaining N×MW data are used as test data to build a decision tree personnel quantitative scoring model; The W data are used to construct a training set D, and the scoring scores of the raters are marked as y i , and iterates through all the feature values in the current set one by one by traversing the values one by one; When the kth feature variable x in the training set i (k) Given a value s, change x i (k) As the segmentation variable, s is used as the segmentation point, and D can be divided into two regions R according to s. 1 (k,s) and R 2 (k,s), that is, R 1 (k,s)={(x i (k) ,y i )|x i (k) <=s},R 2 (k,s)={(x i (k) ,y i )|x i (k) >s}. The value of s should minimize the sum of square errors between the two regions, that is, satisfy the formula: in, and R 1 (k,s) and R 2 (k,s) inner y i The mean of .
6. A method for identifying and warning of behavioral risks of electric power workers according to claim 5, Features: According to the principle of minimizing the sum of squared errors, the best value is selected as the cutting point to obtain the minimum value s and the corresponding characteristic variable X i (k) ;in, The minimum value s is the current optimal split point; The corresponding characteristic variable X i (k) Then it is the optimal segmentation variable; SE k The characteristic variable x i (k) is the minimum sum of squared errors obtained during partitioning.
7. A method for identifying and warning of behavioral risks of electric power workers according to claim 6, Features: Calculate SE for each feature variable k , and record each SE k The corresponding s value is taken from all s values to find the minimum value and find the value that makes SE k The value of s when the result is the smallest and the corresponding characteristic variable x i (k) Take, according to x i (k) The values of and s are calculated to determine and And take the training set D as the parent node, and R 1 (k,s) and R 2 (k,s) is its leaf node, marking the depth of the decision tree as one layer.
8. A method for identifying and warning of behavioral risks of electric power workers according to claim 7, Features: R 1 (k,s) and R 2 (k, s) is used as a new training set to continuously traverse the partition and solve and generate new leaf nodes multiple times until all samples are cut or the maximum depth of the given decision tree is reached, and a regression tree is generated, and the corresponding mean value in the sample set of the last layer of leaf nodes is This is the scoring result of the decision tree model.
9. A method for identifying and warning of behavioral risks of electric power workers according to claim 8, Features: The root mean square error RMSE is calculated based on the scoring results and the scoring results of the scorers: Calculate the error coefficient C based on the root mean square error results in multiple decision tree test data t :
10. A method for identifying and warning of behavioral risks of electric power workers according to claim 9, Features: The error coefficient C t Multiply it by the score given to the person by the corresponding decision tree and add them together to get the final random forest person score result: