Method for detecting bright stop and dark start of hazardous chemical enterprises

Abstract

The invention discloses a dangerous chemical enterprise bright-stop and dark-start detection method, and belongs to the technical field of safety production, and the method comprises the steps: analyzing two types of historical data of the number of production devices and power utilization indexes, selecting sample enterprises with large-area production stop and recovery, carrying out the clustering separation of the historical data of each sample enterprise, and carrying out the calculation of the historical data of each sample enterprise; obtaining a production halt maximum power utilization index Emax of the sample enterprise; obtaining a model by using a transverse normalization method and a regression training method; on the basis of the dangerous chemical enterprise operating rate counted by the early warning system, comparing the models obtained by the two methods, determining one of the models to judge the production stop states of all major dangerous source enterprises, verifying the accuracy and effectiveness of a conclusion obtained by model mapping, and further judging a list of enterprises suspected to be in bright stop and dark start in combination with safety commitment data. According to the invention, a large number of false alarm conditions of production and stop states of an original system can be eliminated through enterprise desensitized total power consumption data, and the accuracy of enterprise operation state identification is effectively improved.

Description

TECHNICAL FIELD

[0001] The application discloses a method for detecting open production in disguise of closed production of a hazardous chemical enterprise, and belongs to the technical field of safety production. BACKGROUND

[0002] The risk of open production in disguise of closed production of a hazardous chemical enterprise refers to that some hazardous chemical enterprises declare to stop production (closed production) on the surface when they stop production or rectify, but actually continue to produce or operate in part through some hidden ways (open production in disguise). In the production process of a hazardous chemical enterprise, high-risk substances and complex process flow are involved. If the production is stopped on the surface but actually operated, the equipment is easy to be in an unstable state for a long time. These hidden dangers may not be revealed for a period of time, but once an accident occurs, the consequences can be very serious. The hazardous chemical enterprise must report the stop production according to law and conduct stop production inspection according to relevant regulations.

[0003] In recent years, with the increasing demand for safety production of dangerous chemicals, the construction of a dangerous chemical safety production risk monitoring and early warning platform (hereinafter referred to as an early warning system) and a power-assisted emergency enterprise monitoring system (hereinafter referred to as a power-assisted emergency system) has gradually improved. The latter can provide about 3800 major hazard source enterprises under joint supervision with multiple state types related to safety production such as emergency production shutdown, sudden stop and night production, overload production, short-term shutdown, long-term shutdown and shutdown and production, and provide certain methods and means for the management department to supervise the enterprise level. However, the above-mentioned state has low accuracy in state judgment on the one hand, and the state dimension is not sufficient on the other hand. Since the power-assisted emergency system only provides a daily electricity index of a major hazard source enterprise, and the index is derived from the desensitization data of the State Grid, in addition to production, it also includes life office, public works and other types of electricity, so it cannot directly determine the production and shutdown state of the enterprise by the high and low of the electricity consumption, and it cannot further confirm whether there is open production in disguise of closed production. SUMMARY

[0004] The purpose of the present application is to provide a method for detecting open production in disguise of closed production of a hazardous chemical enterprise, to solve the problem that it is difficult to find open production in disguise of closed production of a hazardous chemical enterprise in the prior art.

[0005] A method for detecting open production in disguise of closed production of a hazardous chemical enterprise, comprising:

[0006] S1, analyze two types of historical data of production device set number and electricity index, select sample enterprises that have appeared large-area shutdown and production, respectively cluster and separate the historical data of each sample enterprise, and obtain the maximum electricity index E of the sample enterprise in shutdown max ;

[0007] S2, use horizontal normalization and regression training to obtain a model, and the regression training includes multiple regression methods;

[0008] S3, taking the dangerous chemical enterprise operating rate counted by the early warning system as a benchmark, comparing the models obtained by the two methods to determine one of them for judging the production and shutdown state of all major hazard source enterprises;

[0009] S4, through the dangerous chemical enterprise operating rate trend counted by the early warning system, combined with manual enterprise production and shutdown state confirmation, verifying the accuracy and effectiveness of the conclusion obtained by the model mapping;

[0010] S5, further judge the list of suspected enterprises that stop production and start production secretly in combination with safety commitment data.

[0011] The clustering separation includes Kmeans clustering processing based on Euclidean distance, locking small cluster center category samples, and taking the maximum value of the electricity index of the enterprise as the E max .

[0012] The horizontal normalization includes horizontal normalization of E max , the maximum electricity index, and the minimum electricity index, normalizing the maximum electricity index and the minimum electricity index of each sample enterprise to the interval [0, 1] to obtain E max The mean value of E max at the position in the interval is taken as the threshold value for evaluating the production and shutdown of the enterprise.

[0013] Taking the E max of the sample enterprise as the target value, and taking the maximum electricity index and the minimum electricity index of all sample enterprises as characteristic values for regression training analysis, 80% of the sample enterprises are taken as the training set for regression training, and one of the multiple linear regression, ridge regression, Lasso regression and gradient boosting methods is used for regression. The remaining 20% of the sample enterprises are used as a test set for model accuracy verification.

[0014] The predicted E max of the method with the highest regression accuracy in the multiple regression methods is taken as the threshold value for evaluating the production and shutdown of the enterprise.

[0015] On the basis of judging the production and shutdown state of the enterprise, when the enterprise is in the production state, the number of production devices is 0, and it is not in the trial production state, it is considered that the enterprise is in the suspected state of stopping production and starting production secretly.

[0016] Compared with the prior art, the present application has the following beneficial effects: the present application can exclude a large number of false reports of the original system production stop state by using the overall power consumption data of enterprise desensitization, effectively improving the accuracy of enterprise operation state recognition; the present application can consider the relationship between public auxiliary life power consumption and production power consumption from the business and technical aspects, separate the maximum power consumption index of enterprise stop production, that is, the production stop state power consumption threshold, through the method based on Euclidean distance clustering; the present application trains the model through various machine learning algorithms such as normalization and regression, respectively verifies the effectiveness of the model from the business and technical aspects, and finally maps the model to all major hazard source enterprises included in the power emergency response system, and has a clear verification process, proving that the business conclusion obtained by the model has good accuracy; the present application further increases the identification of suspected open state on the basis of judging the production stop state of the enterprise, providing targeted services for the safety production supervision of the management department to the enterprise. BRIEF DESCRIPTION OF DRAWINGS

[0017] Figure 1 The technical roadmap of the present application is shown in the figure;

[0018] Figure 2 The production device number and power consumption index logical matrix diagram is shown in the figure;

[0019] Figure 3 The daily power consumption index diagram before clustering algorithm is shown in the figure;

[0020] Figure 4 The daily power consumption index diagram after clustering algorithm is shown in the figure;

[0021] Figure 5 The horizontal normalization result diagram is shown in the figure;

[0022] Figure 6 The regression training comparison diagram is shown in the figure;

[0023] Figure 7 The regression training result diagram is shown in the figure;

[0024] Figure 8 The in-production rate comparison diagram of the model and the early warning system is shown in the figure;

[0025] Figure 9 The actual verification result comparison diagram of the model and the early warning system is shown in the figure;

[0026] Figure 10 The suspected open alarm condition verification feedback reason diagram is shown in the figure. DETAILED DESCRIPTION

[0027] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention are described clearly and completely below. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0028] A method for detecting the covert operation of hazardous chemical enterprises during apparent shutdowns includes:

[0029] S1. Analyze two types of historical data: the number of production units and the electricity consumption index. Select sample enterprises that have experienced large-scale shutdowns and restarts. Cluster the historical data of each sample enterprise to obtain the maximum electricity consumption index E during shutdowns for each sample enterprise. max ;

[0030] S2. The model is obtained using two methods: lateral normalization and regression training. Regression training includes various regression methods.

[0031] S3. Based on the operating rate of hazardous chemical enterprises as statistically obtained by the early warning system, compare the models obtained by the two methods and determine one of them to judge the production and shutdown status of all major hazard source enterprises.

[0032] S4. By combining the trend of the operating rate of hazardous chemical enterprises statistically analyzed by the early warning system with the confirmation of the production and shutdown status of enterprises via manual electrical communication, the accuracy and effectiveness of the conclusions drawn from the model mapping are verified.

[0033] S5. Further determine the list of enterprises suspected of operating covertly while ostensibly shut down, by combining safety commitment data.

[0034] The clustering separation includes performing K-means clustering based on Euclidean distance, identifying the clusters with the smallest cluster centers, and using the maximum value of their electricity consumption index as the enterprise's E. max .

[0035] Lateral normalization includes E max The maximum and minimum electricity consumption indices are horizontally normalized to the interval [0, 1] for each sample enterprise, resulting in E. max Take E at the position within this interval. max The mean value is used as the threshold for evaluating whether a company is in production or shutting down.

[0036] Taking the example company E maxUsing the maximum and minimum electricity consumption indices of all sample enterprises as the target values, regression training analysis was performed. 80% of the sample enterprises were used as the training set for regression training. One of the regression methods was adopted: multiple linear regression, ridge regression, Lasso regression, and gradient boosting. The remaining 20% ​​of the sample enterprises were used as the test set to verify the accuracy of the model.

[0037] The method with the highest accuracy among various regression methods predicts E. max This serves as a threshold for evaluating whether a company is in production or not.

[0038] Based on the assessment of a company's production status, if a company is in production but has zero sets of production equipment and is not in trial production, it is considered to be in a state of suspected apparent shutdown but covert operation.

[0039] The purpose of this invention is to study a method for detecting the apparent shutdown but actual operation status of hazardous chemical enterprises. It mainly combines daily electricity consumption index data and safety commitment data (including the number of production units, the number of trial production units, the number of maintenance and repair units, etc.) of hazardous chemical enterprises, which are respectively built into the power emergency response system and the early warning system. It proposes a clustering and regression method based on machine learning to analyze and judge the production and shutdown status of specific major hazard source enterprises, thereby providing a list of enterprises suspected of apparent shutdown but actual operation, so that management departments can carry out targeted supervision.

[0040] The technical flowchart of this invention is as follows: Figure 1 As shown, the first step involved cross-referencing the early warning system and the power emergency response system to confirm that the overall start-up and shutdown status of an enterprise cannot be directly evaluated based on its power consumption when no equipment is running. The second step involved analyzing historical data on the number of production units and the power consumption index. For nearly 2,000 enterprises with significant hazard sources exhibiting logical inconsistencies, nearly 400 enterprises that had experienced substantial shutdowns and restarts were manually selected, primarily concentrated around the Lunar New Year period. For each sample enterprise, historical data including the Lunar New Year period was clustered to obtain the enterprise's maximum power consumption index during shutdowns, denoted as E. max Step 3: Set the value of each sample company to E. max The maximum and minimum electricity consumption indices are horizontally normalized, and the mean of the normalized values ​​for all sample enterprises is taken. The maximum and minimum electricity consumption indices for all sample enterprises are used as characteristic values, E. maxThe regression training is performed as a target value. Based on the national chemical enterprise opening rate counted by the early warning system, the above two methods are compared, and one of them is determined to judge the production stop state of all major hazard source enterprises. Fourth step: verify the accuracy and effectiveness of the conclusion obtained by the model through the national chemical enterprise opening rate trend counted by the early warning system, manual electric enterprise production stop state confirmation and other methods. Fifth step: on the basis of production stop state confirmation, further judge the list of suspected enterprises with stop production and start operation in combination with safety commitment data.

[0041] In the specific implementation of the present application, the historical electricity index data of the major hazard source enterprise to which the power emergency system belongs is read, and the safety commitment data of the early warning system is associated. It is not difficult to find from the scatter plot that there is a logical contradiction between the production device number and the electricity index of a large number of enterprises, and the current production stop state of the enterprise cannot be simply judged by the historical minimum electricity index of the enterprise as a threshold, Figure 2 For example, an enterprise.

[0042] Through observation of the scatter plot of the historical data of a large number of enterprises, it is found that the electricity index of most enterprises is significantly reduced during the period from the end of January to the beginning of February 2025 (i.e. during the Spring Festival), and after verification, it is found that almost all enterprises are in a stop production state during this period. However, on the one hand, the time period of stop production of each enterprise is not completely consistent, and on the other hand, the reported number of production device operation of the enterprise is biased, so the maximum electricity index of stop production of each enterprise cannot be extracted by traditional statistical logic, and the real production state of the enterprise cannot be judged. Based on this, nearly 400 enterprises that have appeared obvious large-scale stop production and production are manually selected, and each example enterprise is processed by Kmeans clustering based on Euclidean distance, the cluster center of the smaller category sample is locked, the maximum value of the electricity index of the enterprise is taken as the maximum electricity index of the stop production of the enterprise, and is recorded as E max The daily electricity index before clustering algorithm is shown in Figure 3 , and the daily electricity index after clustering algorithm is shown in Figure 4 . As can be seen from the schematic diagram of the example enterprise, the E max of the enterprise after the clustering algorithm processing is far lower than the orange point position of the confused electricity index before the processing, which is a stop production maximum electricity index closer to the actual situation.

[0043] In order to be able to identify all the major hazard source enterprises included in the power emergency, the E max of the example enterprise and the original maximum and minimum electricity index provided by the power emergency system need to be combined for training, and the obtained model is mapped to the remaining major hazard source enterprises. In this way, two methods are selected for mapping:

[0044] (1) The three characteristics of the example enterprise are horizontally normalized, and the maximum and minimum electricity index of each enterprise is normalized to 0-1, and the Emax The location of the interval is shown in FIG. 5. As can be seen from the figure, the normalized values of most sample enterprises Emax are between 0 and 0.5, and the standard deviation is about 0.11. From a statistical point of view, the data is mostly concentrated around the mean value, and the change is small. Taking the mean value 0.25 as the threshold for evaluating the production suspension of enterprises, the value is basically consistent with the proportion of public auxiliary and living electricity obtained from the previous enterprise investigation and communication. It is found through subsequent continuous multi-day calculation that the mean value of the enterprise production rate is about 70.3%.

[0045] (2) Taking Emax of the sample enterprise as the target value and the original maximum and minimum electricity index as the characteristic value for regression training analysis, the sample enterprises are randomly selected according to the "80 / 20 rule". Multiple linear regression, ridge regression, Lasso regression and gradient boosting regression methods are selected to perform regression training on 80% of the samples as the training set, and the remaining 20% of the samples are used as the test set for model accuracy verification. The accuracy is shown in FIG. 6. As can be seen from the figure, the accuracy of the four regression methods on the test set is ideal, among which the accuracy of Lasso regression is the highest, about 92%. The comparison between the original value and the predicted value of Emax is shown in FIG. 7. Figure 6 Figure 7 Taking the Emax value of the Lasso regression model mapping the power emergency response system major hazard source enterprise as the threshold for evaluating the production suspension of enterprises, it is found through subsequent continuous multi-day calculation that the mean value of the enterprise production rate is about 73.5%.

[0046] To verify the accuracy of the enterprise production suspension state, on the one hand, the enterprise production rate output by the early warning system and the model in the past month is compared. Among the above two methods, the enterprise production rate obtained by the regression method of the power emergency response system major hazard source enterprise is closer to the actual situation. The comparison between the production rates of the model and the early warning system is shown in FIG. 8. Figure 8

[0047] On the other hand, the accuracy rate of the enterprise state derived daily is confirmed by sampling, which has been greatly improved compared with the original enterprise state of the power emergency response system. The actual verification results of the model and the early warning system are compared as shown in FIG. 9. Figure 9 Figure 10

[0048] ​​​​The above examples are only used for illustrating the technical solutions of the present application, and are not intended to limit the present application. Although the present application has been described in detail with reference to the foregoing examples, it should be understood by those skilled in the art that the technical solutions recorded in the foregoing examples can be modified, or some or all of the technical features can be replaced by equivalent replacements, and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for detecting the covert operation of hazardous chemical enterprises during apparent shutdowns, characterized in that, include: S1. Analyze two types of historical data: the number of production units and the electricity consumption index. Select sample enterprises that have experienced large-scale shutdowns and restarts. Cluster the historical data of each sample enterprise to obtain the maximum electricity consumption index E during shutdowns for each sample enterprise. max ; S2. The model is obtained using two methods: lateral normalization and regression training. Regression training includes various regression methods. S3. Based on the operating rate of hazardous chemical enterprises as statistically obtained by the early warning system, compare the models obtained by the two methods and determine one of them to judge the production and shutdown status of all major hazard source enterprises. S4. By combining the trend of the operating rate of hazardous chemical enterprises statistically analyzed by the early warning system with the confirmation of the production and shutdown status of enterprises via manual electrical communication, the accuracy and effectiveness of the conclusions drawn from the model mapping are verified. S5. Further determine the list of enterprises suspected of operating covertly while ostensibly shut down, by combining safety commitment data.

2. The method for detecting the apparent shutdown and covert operation of hazardous chemical enterprises according to claim 1, characterized in that, The clustering separation includes performing K-means clustering based on Euclidean distance, identifying the clusters with the smallest cluster centers, and using the maximum value of their electricity consumption index as the enterprise's E. max .

3. The method for detecting the apparent shutdown and covert operation of hazardous chemical enterprises according to claim 1, characterized in that, Lateral normalization includes E max The maximum and minimum electricity consumption indices are horizontally normalized to the interval [0, 1] for each sample enterprise, resulting in E. max Take E at the position within this interval. max The mean value is used as the threshold for evaluating whether a company is in production or shutting down.

4. The method for detecting the apparent shutdown and covert operation of hazardous chemical enterprises according to claim 1, characterized in that, Taking the example company E max Using the maximum and minimum electricity consumption indices of all sample enterprises as the target values, regression training analysis was performed. 80% of the sample enterprises were used as the training set for regression training. One of the regression methods was adopted: multiple linear regression, ridge regression, Lasso regression, and gradient boosting. The remaining 20% ​​of the sample enterprises were used as the test set to verify the accuracy of the model.

5. The method for detecting the apparent shutdown and covert operation of hazardous chemical enterprises according to claim 1, characterized in that, The method with the highest accuracy among various regression methods predicts E. max This serves as a threshold for evaluating whether a company is in production or not.

6. The method for detecting the apparent shutdown and covert operation of hazardous chemical enterprises according to claim 1, characterized in that, Based on the assessment of a company's production status, if a company is in production but has zero sets of production equipment and is not in trial production, it is considered to be in a state of suspected apparent shutdown but covert operation.

Images