AI brain early warning method for risk management and control of chemical industry park

Abstract

The invention discloses an AI brain early warning method for risk management and control of a chemical industrial park. The method comprises the following steps: collecting risk monitoring data of the chemical industrial park in real time based on Internet of Things equipment arranged in the chemical industrial park; transmitting the risk monitoring data to a background server through a transmission module; obtaining a target early warning demand of the chemical industry park, and transmitting the target early warning demand to a background server; the background server performs data processing on the risk monitoring data according to the target early warning demand to obtain a risk report; and performing risk grading management and control on the chemical industry park according to the risk report. The risk monitoring data of the chemical industry park are comprehensively collected, data processing is performed on the risk monitoring data based on the target early warning demand, the park safety data are deeply analyzed and applied, risk hierarchical management and control are performed on the chemical industry park, the efficiency and accuracy of risk management and control are improved, and waste of management and control resources is avoided.

Description

Technical Field

[0001] The present invention relates to the technical field of risk management and control, and in particular to an AI brain early warning method for risk management and control in a chemical park. Background Art

[0002] Chemical parks have the advantages of intensiveness and maximum benefits, which can promote the optimal allocation of resources. Enterprises in chemical parks are relatively concentrated, with a wide variety of chemicals, complex chemical processes, large-scale equipment, and harsh operating conditions. This leads to a large number of hazardous sources in chemical parks and a high risk level. Accidents occur from time to time and cause serious casualties, property losses, environmental pollution, and adverse social impacts. The safety management levels of various enterprises in chemical parks vary, and less consideration is given to the mutual influence between different hazardous units and between enterprises, as well as the impact of multiple disaster coupling on accidents; the safety management system of chemical parks has not been able to fully utilize safety information, and has not conducted in-depth analysis and application of park safety data. The existing safety management system cannot fully utilize existing information to provide support for risk control, and there is no hierarchical control of each hazardous unit, which is inefficient and causes a waste of resources. Summary of the invention

[0003] The present invention aims to solve at least one of the technical problems in the above-mentioned technologies to a certain extent. To this end, the purpose of the present invention is to propose an AI brain early warning method and method for risk management of chemical parks, comprehensively collect risk monitoring data of chemical parks, process risk monitoring data based on target early warning needs, conduct in-depth analysis and application of park safety data, conduct risk classification management of chemical parks, improve the efficiency and accuracy of risk management, and avoid waste of management resources.

[0004] To achieve the above objectives, the embodiment of the present invention proposes an AI brain early warning method for risk management and control in a chemical park, comprising:

[0005] Collect risk monitoring data of chemical parks in real time based on IoT devices installed in chemical parks;

[0006] Transmit the risk monitoring data to the backend server through the transmission module;

[0007] Obtain the target warning requirements of the chemical park and transmit them to the backend server;

[0008] The backend server processes the risk monitoring data according to the target warning requirements and obtains a risk report;

[0009] Conduct risk classification and control of chemical parks based on risk reports.

[0010] According to some embodiments of the present invention, transmitting the risk monitoring data to a background server through a transmission module includes:

[0011] The transmission module determines the number and location of IoT devices that acquire risk monitoring data;

[0012] According to the number and location of IoT devices, IoT devices are clustered based on the K-means clustering algorithm to obtain several cluster sets;

[0013] The IoT device corresponding to the cluster center in the cluster set is used as the cluster head node, and other IoT devices in the cluster set are used as member nodes to complete clustering;

[0014] All cluster head nodes and member nodes form a clustered wireless transmission network, and transmit risk monitoring data to the background server based on the wireless transmission network.

[0015] According to some embodiments of the present invention, the IoT devices include sensors, surveillance cameras, and drones;

[0016] The risk monitoring data includes first sensor data acquired by a sensor, first image data acquired by a surveillance camera, and second sensor data and second image data acquired by a drone.

[0017] According to some embodiments of the present invention, the backend server processes the risk monitoring data according to the target warning requirements to obtain a risk report, including:

[0018] Performing image enhancement processing on the first image data and the second image data to obtain target image data;

[0019] Performing data cleaning and data deduplication processing on the first sensor data and the second sensor data to obtain target sensor data;

[0020] The backend server performs image recognition on the target image data according to the target warning requirements to obtain employee behavior data, infrastructure data, and equipment action data;

[0021] The backend server analyzes the target sensor data according to the target warning requirements to obtain leakage emergency event data and environmental management event data;

[0022] Perform data normalization on employee behavior data, infrastructure data, equipment action data, leakage emergency event data, and environmental management event data to obtain normalized data;

[0023] According to the normalized data, the risk status value of each area in the chemical park is determined based on a preset algorithm, and a risk report is generated according to the risk status value of each area in the chemical park.

[0024] According to some embodiments of the present invention, determining the risk status value of each area of ​​the chemical park based on the normalized data based on a preset algorithm includes:

[0025] R=a×exp(-F)+b×log(1+P)+c×sqrt(C)+d×E 2 +e×(-M)

[0026] Among them, R is the risk status value of the area; F, P, C, E and M represent the normalized employee behavior data, infrastructure data, equipment action data, leakage emergency event data and environmental management event data respectively; a, b, c, d, e are the corresponding weight coefficients of the normalized employee behavior data, infrastructure data, equipment action data, leakage emergency event data and environmental management event data respectively; a+b+c+d+e=1.

[0027] According to some embodiments of the present invention, performing image enhancement processing on the first image data and the second image data to obtain target image data includes:

[0028] Performing a convolution operation on the first image data and the second image data based on a convolution kernel of edge detection, converting the convolved image into a processable two-dimensional matrix based on an image processing library, and calculating the variance of the image according to the two-dimensional matrix;

[0029] Screen out images whose variance is less than a preset threshold as images to be enhanced;

[0030] Calculate the grayscale histogram of the image to be enhanced, and determine the minimum grayscale level and the maximum grayscale level in the grayscale histogram;

[0031] The grayscale value of each pixel in the image to be enhanced is mapped to a new grayscale range according to the minimum grayscale level and the maximum grayscale level to obtain the target image data.

[0032] According to some embodiments of the present invention, obtaining target warning requirements of a chemical park includes:

[0033] Based on the chemical layout of the chemical park, the historical warning needs corresponding to the known chemical park with the highest matching degree with the chemical layout are obtained from the warning database, and the layout differences are obtained; the layout differences include the type, quantity, location, and process flow of the equipment;

[0034] The historical warning requirements are modified according to the layout differences to obtain the target warning requirements.

[0035] According to some embodiments of the present invention, risk classification management and control of a chemical park is performed according to a risk report, including:

[0036] Input the risk report into the pre-trained early warning model and output the risk classification control strategy; the risk classification control strategy includes determining whether there is a risk in the inspection item and determining the risk type when there is a risk, dividing the risk points, identifying the hazard sources included in the risk points, determining the levels of the hazard sources and risk points, formulating control measures and forming a risk control list;

[0037] According to the risk control list, hidden dangers in the chemical park are checked and managed; the hidden dangers check and management include the allocation of check content, hidden danger check and hidden danger rectification after approval and assignment.

[0038] According to some embodiments of the present invention, a method for obtaining an early warning model includes:

[0039] Obtain a training database, wherein the training database contains several sample risk reports, wherein each sample risk report contains S risk monitoring data and N early warning data of the chemical park, and the S risk monitoring data and the N early warning data are combined into an early warning matrix, wherein the early warning matrix has S rows and N columns, and the risk classification control strategy corresponding to each sample risk report is recorded at the same time to form a vector Y1, and the vector Y1 is removed after duplicate values ​​are formed;

[0040] Deep learning based on training database;

[0041] W1=rand(D,S)W2=rand(D,S)

[0042] f=W2*max(zero(D,1),W1X i )

[0043]

[0044] Among them, rand(D,S) generates a random matrix with D rows and S columns, and the value of each element in the matrix is ​​a random value between 0 and 1, W1 and W2 are deep learning adjustment coefficients, D is the number of values ​​in the vector Y, max() is the vector composed of the maximum values ​​of each row in the brackets, zero(D,1) generates a matrix of all 0s with D rows and 1 column, L is the activation function, X i is the value of the i-th column of the warning matrix X, and f is the value of i The vector function formed after mapping with W1 and W2; To X i The jth value in the vector function formed after mapping with W1 and W2, To X i The Y1th value in the vector function formed after mapping with W1 and W2; where j = 1, 2, 3, ... Y1 i -1, Y1 i +1...D, Y1i +1 is the position of the value in vector Y corresponding to the i-th value in vector Y1, W ii,k,t is the value of the kth row and tth column of the random matrix Wii;

[0045] Adjust the deep learning adjustment coefficient;

[0046]

[0047]

[0048] in, L to W ii,k,t Find the partial derivative, WS ii,k,t To obtain the value after partial derivation, K = 1, 2, 3...D, t = 1, 2, 3...S; after adjusting all elements in the deep learning adjustment coefficients W1 and W2, new deep learning adjustment coefficients W1 and W2 can be obtained;

[0049] According to the new deep learning adjustment coefficients W1, W2, iterate continuously until WS ii,k,t All are 0, and the target deep learning adjustment coefficient is obtained, and then the pre-trained early warning model is determined.

[0050] According to some embodiments of the present invention, the risk classification management process further includes:

[0051] For low-risk areas in chemical parks, normal signals are sent to staff, and risk monitoring data of chemical parks is continuously collected and updated in real time through IoT devices;

[0052] For medium-risk areas in chemical parks, early warning signals and risk monitoring data of risk areas are sent to staff, and warning notices are sent to people around risk areas;

[0053] For high-risk areas in chemical parks, alarm signals and risk monitoring data of risk areas are sent to staff, and emergency evacuation and avoidance notices are sent to people around the risk areas.

[0054] The present invention proposes an AI brain early warning method for risk management of chemical parks. It comprehensively collects risk monitoring data of chemical parks, processes the risk monitoring data based on target early warning needs, conducts in-depth analysis and application of park safety data, and conducts risk classification management of chemical parks, thereby improving the efficiency and accuracy of risk management and avoiding waste of management resources.

[0055] Other features and advantages of the present invention will be described in the following description, and partly become apparent from the description, or understood by practicing the present invention. The purpose and other advantages of the present invention can be realized and obtained by the structures particularly pointed out in the written description and the accompanying drawings.

[0056] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0058] Figure 1 It is a flow chart of an AI brain early warning method for risk management in a chemical park according to an embodiment of the present invention. DETAILED DESCRIPTION

[0059] The preferred embodiments of the present invention are described below in conjunction with the accompanying drawings. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present invention, and are not used to limit the present invention.

[0060] like Figure 1 As shown, the embodiment of the present invention proposes an AI brain early warning method for risk management and control in a chemical park, including steps S1-S5:

[0061] S1. Collect risk monitoring data of chemical parks in real time based on IoT devices installed in chemical parks;

[0062] S2. Transmit the risk monitoring data to the backend server through the transmission module;

[0063] S3, obtain the target warning requirements of the chemical park and transmit them to the backend server;

[0064] S4. The backend server processes the risk monitoring data according to the target warning requirements and obtains a risk report;

[0065] S5. Carry out risk classification and control of chemical parks based on risk reports.

[0066] The working principle of the above technical solution is as follows: IoT devices (such as sensors, cameras, etc.) are responsible for monitoring various parameters of the park, such as temperature, pressure, leakage, personnel activities, etc. Based on IoT devices, the monitoring scope is expanded to include more types of risk parameters to improve the comprehensiveness of early warning. The transmission module uses high-speed and stable network transmission technology and designs a data retransmission mechanism to deal with the problem of data loss caused by network instability. The management needs and safety goals of the park are converted into specific early warning indicators, and the target early warning needs are determined based on historical early warning needs. The background server processes the risk monitoring data according to the target early warning needs to obtain a risk report; the risk report is the safety data of the chemical park obtained after data processing, and realizes classification and regional processing. According to the risk report, the chemical park is risk-graded and controlled, and different control measures are implemented to improve the efficiency and accuracy of risk control and avoid waste of control resources.

[0067] The beneficial effects of the above technical solution are: comprehensive collection of risk monitoring data of chemical parks, data processing of risk monitoring data based on target early warning needs, in-depth analysis and application of park safety data, risk classification management of chemical parks, improving the efficiency and accuracy of risk management, and avoiding waste of management resources.

[0068] According to some embodiments of the present invention, transmitting the risk monitoring data to a background server through a transmission module includes:

[0069] The transmission module determines the number and location of IoT devices that acquire risk monitoring data;

[0070] According to the number and location of IoT devices, IoT devices are clustered based on the K-means clustering algorithm to obtain several cluster sets;

[0071] The IoT device corresponding to the cluster center in the cluster set is used as the cluster head node, and other IoT devices in the cluster set are used as member nodes to complete clustering;

[0072] All cluster head nodes and member nodes form a clustered wireless transmission network, and transmit risk monitoring data to the background server based on the wireless transmission network.

[0073] The working principle of the above technical solution is as follows: the transmission module first identifies and counts the number of all IoT devices involved in risk monitoring in the chemical park and their specific locations. The K-means clustering algorithm is used to group IoT devices. The K-means algorithm is a clustering algorithm that can divide data points into K clusters so that the data points in each cluster are as similar as possible, while the data points between different clusters are as different as possible. The K value, that is, the number of clusters desired, needs to be determined in advance. This value can be set based on factors such as the scale of the park, the distribution density of IoT devices, and the complexity of data processing. The algorithm iteratively adjusts the center position of each cluster until a certain convergence condition is reached (such as the cluster center no longer changes significantly). In each cluster set, the IoT device corresponding to the cluster center is selected as the cluster head node. The cluster head node is responsible for collecting the risk monitoring data of other member nodes in its cluster set and sending these data to the background server through the wireless transmission network. Other IoT devices in the cluster set serve as member nodes, and they send their respective risk monitoring data to the cluster head node. All cluster head nodes and member nodes together form a clustered wireless transmission network. This network can more effectively utilize wireless resources and improve the efficiency and reliability of data transmission. In this network, cluster head nodes and member nodes transmit data through wireless communication. The cluster head node then sends the collected data to the background server through a more efficient path (such as directly or through a relay node).

[0074] The beneficial effects of the above technical solution are as follows: through clustering processing, the redundancy and conflict of data transmission are reduced, and the throughput of the network is improved. As a transfer station for data collection, the cluster head node can reduce the communication burden of a single IoT device and improve the stability and reliability of the network. Clustering processing based on the number and location of IoT devices can more reasonably allocate wireless resources and improve resource utilization. Introducing the K-means clustering algorithm for clustering IoT devices can significantly improve the efficiency and reliability of risk monitoring data transmitted to the background server through the transmission module.

[0075] According to some embodiments of the present invention, the IoT devices include sensors, surveillance cameras, and drones;

[0076] The risk monitoring data includes first sensor data acquired by a sensor, first image data acquired by a surveillance camera, and second sensor data and second image data acquired by a drone.

[0077] Working principle of the above technical solution: Sensors can monitor various physical and chemical parameters in the environment in real time, such as temperature, pressure, humidity, gas concentration, etc. These sensors are installed around key equipment or areas to detect abnormal situations in time. The data obtained by the sensors (called the first sensing data) is crucial for evaluating the safety status of the park. Surveillance cameras can capture real-time images of the park and provide managers with intuitive monitoring pictures. Surveillance cameras are deployed in key areas or important channels to monitor personnel activities, vehicle entry and exit, and potential security threats. The data obtained by the camera (called the first image data) is very useful for identifying abnormal behaviors or events. Drones play an increasingly important role in risk management and control of chemical parks. They are able to fly over the park to obtain more extensive and comprehensive monitoring data. Drones are equipped with sensors and cameras, so they can simultaneously obtain second sensing data and second image data. These data are very helpful for evaluating the overall safety status of the park and discovering potential risk areas.

[0078] The beneficial effects of the above technical solutions: By combining various IoT devices such as sensors, surveillance cameras and drones, the chemical park can collect rich risk monitoring data in real time. These data provide a solid foundation for the analysis and early warning of the backend server, helping to timely discover potential risks, prevent accidents, and ensure the safe operation of the park.

[0079] According to some embodiments of the present invention, the backend server processes the risk monitoring data according to the target warning requirements to obtain a risk report, including:

[0080] Performing image enhancement processing on the first image data and the second image data to obtain target image data;

[0081] Performing data cleaning and data deduplication processing on the first sensor data and the second sensor data to obtain target sensor data;

[0082] The backend server performs image recognition on the target image data according to the target warning requirements to obtain employee behavior data, infrastructure data, and equipment action data;

[0083] The backend server analyzes the target sensor data according to the target warning requirements to obtain leakage emergency event data and environmental management event data;

[0084] Perform data normalization on employee behavior data, infrastructure data, equipment action data, leakage emergency event data, and environmental management event data to obtain normalized data;

[0085] According to the normalized data, the risk status value of each area in the chemical park is determined based on a preset algorithm, and a risk report is generated according to the risk status value of each area in the chemical park.

[0086] The working principle of the above technical solution is as follows: the background server first performs image enhancement processing on the first image data and the second image data to improve the clarity and contrast of the image, thereby obtaining the target image data. Image enhancement processing helps to improve image quality and make image recognition more accurate. The background server performs data cleaning and deduplication processing on the first sensor data and the second sensor data to eliminate duplicate data, abnormal data and invalid data, thereby obtaining the target sensor data. Ensure the accuracy of data analysis. According to the target warning requirements, the background server performs image recognition on the target image data to obtain employee behavior data, infrastructure data and equipment action data. Image recognition technology can identify key information in the image, such as employee behavior data, infrastructure data, and equipment action data. The background server performs data analysis on the target sensor data according to the target warning requirements to obtain leakage emergency event data and environmental management event data; the background server performs normalization processing on the employee behavior data, infrastructure data, equipment action data, leakage emergency event data and environmental management event data to obtain normalized data. Data normalization can eliminate the dimensional differences between different data, making the data easier to process and analyze. Finally, the backend server determines the risk status values ​​of each area in the chemical park based on the normalized data and the preset algorithm. Based on these risk status values, the backend server generates a risk report to provide intuitive risk assessment results and early warning information to park managers.

[0087] Beneficial effects of the above technical solution: By combining image recognition and data analysis technology, the backend server can comprehensively and accurately assess the risk status of the park. The backend server can process data from IoT devices in real time and quickly generate risk reports to provide timely decision support for park managers. The data processing flow of the backend server can be flexibly adjusted according to the target warning requirements, and supports the expansion of more data processing and analysis functions. Through a series of data processing processes, the backend server can comprehensively and accurately assess the risk status of the chemical park.

[0088] According to some embodiments of the present invention, determining the risk status value of each area of ​​the chemical park based on the normalized data based on a preset algorithm includes:

[0089] R=a×exp(-F)+b×log(1+P)+c×sqrt(C)+d×E 2 +e×(-M)

[0090] Among them, R is the risk status value of the area; F, P, C, E and M represent the normalized employee behavior data, infrastructure data, equipment action data, leakage emergency event data and environmental management event data respectively; a, b, c, d, e are the corresponding weight coefficients of the normalized employee behavior data, infrastructure data, equipment action data, leakage emergency event data and environmental management event data respectively; a+b+c+d+e=1.

[0091] The working principle of the above technical solution: R: represents the risk status value of a certain area in the chemical park. This is a quantitative indicator used to measure the risk level of the area. F: Normalized employee behavior data. Employee behavior data includes information on employee operation norms, safety awareness, etc. P: Normalized infrastructure data. Infrastructure data covers the park's building structure, road conditions, safety protection facilities, etc. C: Normalized equipment action data. Equipment action data involves the operating status, maintenance records, failure rate, etc. of the equipment. E: Normalized leakage emergency event data. This includes the number, severity, and impact range of leakage events. M: Normalized environmental management event data. Environmental management data includes the park's environmental quality, pollutant emissions, and the implementation effect of environmental protection measures. ; a, b, c, d, e are the corresponding weight coefficients of normalized employee behavior data, infrastructure data, equipment action data, leakage emergency event data, and environmental management event data, respectively, and the values ​​are (0,1). The weight coefficients a, b, c, d, e can be determined by expert scoring, historical data analysis, questionnaire surveys, and other methods. These weighting factors should reflect the relative importance of each type of data in assessing the risk status value.

[0092] Beneficial effects of the above technical solution: This formula provides an effective quantitative assessment tool for risk management in chemical parks. By comprehensively considering a variety of data and information, the risk level of the park can be assessed more comprehensively and accurately, providing strong support for the safety management of the park.

[0093] According to some embodiments of the present invention, performing image enhancement processing on the first image data and the second image data to obtain target image data includes:

[0094] Performing a convolution operation on the first image data and the second image data based on a convolution kernel of edge detection, converting the convolved image into a processable two-dimensional matrix based on an image processing library, and calculating the variance of the image according to the two-dimensional matrix;

[0095] Screen out images whose variance is less than a preset threshold as images to be enhanced;

[0096] Calculate the grayscale histogram of the image to be enhanced, and determine the minimum grayscale level and the maximum grayscale level in the grayscale histogram;

[0097] The grayscale value of each pixel in the image to be enhanced is mapped to a new grayscale range according to the minimum grayscale level and the maximum grayscale level to obtain the target image data.

[0098] The working principle of the above technical solution is: the convolution kernel based on edge detection (such as the convolution kernel of the Sobel, Prewitt or Canny edge detection operator) helps to identify edge features in the image. The convolution operation is completed by sliding the convolution kernel on the image and calculating the dot product of the convolution kernel and the local area of ​​the image at each position. After the convolution operation, the image data is converted into a processable two-dimensional matrix form. This two-dimensional matrix represents the pixel values ​​or eigenvalues ​​of the image. Based on the two-dimensional matrix, the variance of the image is calculated. Variance is an indicator of the degree of fluctuation of the pixel values ​​of the image, which helps to identify whether the image is clear or blurred. A preset variance threshold is set. If the variance of the image is less than this threshold, it means that the image is relatively blurred or has low contrast, so it is screened as an image to be enhanced. For the screened image to be enhanced, its grayscale histogram is calculated. The grayscale histogram shows the number of pixels at each grayscale level in the image, which can help understand the grayscale distribution of the image. In the grayscale histogram, find the minimum grayscale level (i.e., the grayscale level with the smallest pixel value) and the maximum grayscale level (i.e., the grayscale level with the largest pixel value). According to the minimum grayscale and the maximum grayscale, the grayscale value of each pixel in the image to be enhanced is mapped to a new grayscale range to improve the contrast and clarity of the image. After the grayscale value mapping, the enhanced image data, i.e., the target image data, is obtained.

[0099]

[0100] Among them, new g ray is the grayscale value of the target image data; old g ray is the original grayscale value; min g ray is the minimum gray level; max g ray is the maximum grayscale. The image enhancement process is applied to the video surveillance system of the chemical park to improve the clarity and contrast of the surveillance image, so as to more accurately identify employee behavior, equipment status and potential safety risks.

[0101] The beneficial effects of the above technical solution are as follows: through edge detection and grayscale histogram adjustment, the details and contrast of the image can be enhanced to make the image clearer. The enhanced image data can improve the recognition ability and accuracy of the video surveillance system, which helps to timely discover and deal with potential security risks.

[0102] According to some embodiments of the present invention, obtaining target warning requirements of a chemical park includes:

[0103] Based on the chemical layout of the chemical park, the historical warning needs corresponding to the known chemical park with the highest matching degree with the chemical layout are obtained from the warning database, and the layout differences are obtained; the layout differences include the type, quantity, location, and process flow of the equipment;

[0104] The historical warning requirements are modified according to the layout differences to obtain the target warning requirements.

[0105] The working principle of the above technical solution is as follows: the chemical layout of the target chemical park is analyzed in detail, including key information such as the type, quantity, location and process flow of equipment. In the early warning database, the known chemical park with the highest matching degree with the chemical layout is searched, which involves comparing the historical records in the database to find the case that is most similar to the layout of the target park. From the matching known chemical parks, the corresponding historical early warning needs are obtained. These needs include specific safety monitoring indicators, early warning thresholds, emergency response measures, etc. The layout of the target chemical park is compared with that of the matching park, and the differences in equipment types, quantities, locations and process flows between the two are identified. An impact assessment is conducted on the identified layout differences, and the impact of these differences on the safety risks, early warning needs and emergency response of the chemical park is analyzed. According to the layout differences and their impact assessment results, the historical early warning needs are revised to obtain the target early warning needs applicable to the target chemical park. This involves adjusting the safety monitoring indicators, early warning thresholds, emergency response measures and other aspects. In the process of revising the early warning needs, it is necessary to comprehensively consider factors such as the overall safety risks of the target chemical park, historical accident records, policy and regulatory requirements and industry best practices. The revised early warning needs are verified to ensure that they can accurately reflect the safety risk status of the target chemical park.

[0106] The beneficial effect of the above technical solution is that based on the chemical layout of the chemical park, the historical warning requirements corresponding to the known chemical park with the highest matching degree with the chemical layout can be obtained from the warning database, and the target warning requirements applicable to the target chemical park can be finally obtained according to the layout differences. This helps the chemical park to better carry out safety risk management and warning work and improve the overall safety level of the park.

[0107] According to some embodiments of the present invention, risk classification management and control of a chemical park is performed according to a risk report, including:

[0108] Input the risk report into the pre-trained early warning model and output the risk classification control strategy; the risk classification control strategy includes determining whether there is a risk in the inspection item and determining the risk type when there is a risk, dividing the risk points, identifying the hazard sources included in the risk points, determining the levels of the hazard sources and risk points, formulating control measures and forming a risk control list;

[0109] According to the risk control list, hidden dangers in the chemical park are checked and managed; the hidden dangers check and management include the allocation of check content, hidden danger check and hidden danger rectification after approval and assignment.

[0110] The working principle of the above technical solution: the collected risk reports are used as input data. These reports should cover all kinds of potential risk information in the chemical park. The risk reports are analyzed using the pre-trained early warning model. The model has the ability to identify, classify and evaluate risks. Generation of risk classification control strategy: Inspection item risk judgment: The model first determines whether there are risks in each inspection item. Risk type and risk point classification: If there is a risk, further determine the specific type of risk and divide the risk point. Hazard source identification: Conduct an in-depth analysis of each risk point to identify the hazard source contained therein. Level determination: According to the nature and potential consequences of the hazard source, determine the level (such as high, medium, and low) for the hazard source and risk point respectively. Control measures formulation: For each risk point, formulate corresponding control measures to reduce or eliminate risks. Formation of risk control list: Summarize all risk points, hazard sources, levels and corresponding control measures to form a detailed risk control list. Referring to the risk control list, assign specific investigation content to the relevant responsible departments or personnel. The responsible department or personnel conduct on-site investigation according to the assigned content to identify and record potential hidden dangers. Evaluate the hidden dangers found and determine their severity and urgency. Based on the evaluation results, approve and assign the corresponding rectification tasks to the responsible units or individuals. The responsible units or individuals implement the rectification measures according to the rectification requirements. Regularly track and check the rectification progress to ensure the effectiveness and timeliness of the rectification work.

[0111] The beneficial effects of the above technical solution are: it can realize the comprehensive identification, assessment, control and management of risks in chemical parks, thereby effectively ensuring the safe production and stable operation of the parks.

[0112] According to some embodiments of the present invention, a method for obtaining an early warning model includes:

[0113] Obtain a training database, wherein the training database contains several sample risk reports, wherein each sample risk report contains S risk monitoring data and N early warning data of the chemical park, and the S risk monitoring data and the N early warning data are combined into an early warning matrix, wherein the early warning matrix has S rows and N columns, and the risk classification control strategy corresponding to each sample risk report is recorded at the same time to form a vector Y1, and the vector Y1 is removed after duplicate values ​​are formed;

[0114] Deep learning based on training database;

[0115] W1=rand(D,S) W2=rand(D,S)

[0116] f=W2*max(zero(D,1),W1X i )

[0117]

[0118] Among them, rand(D,S) generates a random matrix with D rows and S columns, and the value of each element in the matrix is ​​a random value between 0 and 1, W1 and W2 are deep learning adjustment coefficients, D is the number of values ​​in the vector Y, max() is the vector composed of the maximum values ​​of each row in the brackets, zero(D,1) generates a matrix of all 0s with D rows and 1 column, L is the activation function, X i is the value of the i-th column of the warning matrix X, and f is the value of i The vector function formed after mapping with W1 and W2; To X i The jth value in the vector function formed after mapping with W1 and W2, To X i The Y1th value in the vector function formed after mapping with W1 and W2; where j = 1, 2, 3, ... Y1 i -1, Y1 i +1...D, Y1 i +1 is the position of the value in vector Y corresponding to the i-th value in vector Y1, W ii,k,t is the value of the kth row and tth column of the random matrix Wii;

[0119] Adjust the deep learning adjustment coefficient;

[0120]

[0121] in, L to W ii,k,t Find the partial derivative, WS ii,k,t To obtain the value after partial derivation, K = 1, 2, 3...D, t = 1, 2, 3...S; after adjusting all elements in the deep learning adjustment coefficients W1 and W2, new deep learning adjustment coefficients W1 and W2 can be obtained;

[0122] According to the new deep learning adjustment coefficients W1, W2, iterate continuously until WS ii,k,t All are 0, and the target deep learning adjustment coefficient is obtained, and then the pre-trained early warning model is determined.

[0123] The working principle of the above technical solution is as follows: obtain a training database, which contains several sample risk reports, each of which contains S risk monitoring data and N warning data of the chemical park. The S risk monitoring data and N warning data are combined into a warning matrix X with S rows and N columns. Form a vector Y1, and after removing duplicate values, form a vector Y as the output target of the model. Use rand(D,S) to generate two random matrices W1 and W2 with D rows and S columns as the initial weights of the deep learning model. Define the mapping function f: This function maps the column vector Xi of the warning matrix X through W1 and W2 to obtain a new vector. The max function and zero matrix are used to introduce nonlinearity in the mapping process. Define the loss function L: The loss function consists of two parts, one is the interval loss of multi-classification (similar to the hinge loss of SVM), and the other is the regularization term of the weight to prevent overfitting. Calculate the gradient of the loss function: Calculate the partial derivative WS(ii,k,t) of the loss function L with respect to W1 and W2 by the chain rule. Update the deep learning adjustment coefficient: According to the gradient descent method, update each element in W1 and W2. The update rule determines whether to increase or decrease the value of the current element based on the positive or negative gradient, and multiplies it by a small learning rate (here is 0.001). Iterative training: Repeat the above steps until the gradient WS(ii,k,t) is all 0. At this time, the obtained W1 and W2 are the target deep learning adjustment coefficients. Using the trained deep learning adjustment coefficients W1 and W2, and the defined mapping function f, a pre-trained early warning model can be constructed. The model can receive new risk monitoring data as input and output the predicted risk grading control strategy through the mapping function f.

[0124] The beneficial effects of the above technical solution are: an early warning model for risk classification and control of chemical parks can be obtained, which can automatically output risk classification and control strategies based on risk monitoring data, providing strong support for the safety management of chemical parks.

[0125] According to some embodiments of the present invention, the risk classification management process further includes:

[0126] For low-risk areas in chemical parks, normal signals are sent to staff, and risk monitoring data of chemical parks is continuously collected and updated in real time through IoT devices;

[0127] For medium-risk areas in chemical parks, early warning signals and risk monitoring data of risk areas are sent to staff, and warning notices are sent to people around risk areas;

[0128] For high-risk areas in chemical parks, alarm signals and risk monitoring data of risk areas are sent to staff, and emergency evacuation and avoidance notices are sent to people around the risk areas.

[0129] The working principle of the above technical solution: For low-risk areas in the chemical park, the system should send a normal signal to the staff, indicating that the area is currently in a safe state, and obtain data in real time and update it to the database, so as to timely discover potential risks and prevent accidents. When an area in the chemical park is judged to be medium-risk, the system should immediately send an early warning signal to the staff and send the risk monitoring data of the area at the same time. These data include key factors that may lead to risk escalation, so that the staff can quickly judge and take preventive measures. In addition to sending early warnings to the staff, warning notices should also be sent to people around the risk area. The content of the notice should include the nature of the risk, the scope of impact, and the recommended actions (such as staying away from the risk area) to ensure the safety of the surrounding people. For high-risk areas, the system should immediately send an alarm signal to the staff and send detailed risk monitoring data at the same time. These data should be able to fully reflect the risk situation, including possible sources of danger, risk diffusion trends, etc., so that the staff can quickly formulate an emergency response plan. In addition to sending alarms to the staff, emergency evacuation and avoidance notices should also be sent immediately to the people around the risk area. The content of the notification should be clear and concise, including key information such as evacuation routes, avoidance locations, emergency contact information, etc., to ensure that people around can evacuate the danger zone quickly and orderly.

[0130] The beneficial effects of the above technical solution are: it can achieve effective hierarchical management and control of risks in chemical parks, reduce the probability and consequences of accidents, and ensure the safe and stable operation of the park.

[0131] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. An AI brain early warning method for risk management in chemical parks, characterized in that: include: Collect risk monitoring data of chemical parks in real time based on IoT devices installed in chemical parks; Transmit the risk monitoring data to the backend server through the transmission module; Obtain the target warning requirements of the chemical park and transmit them to the backend server; The backend server processes the risk monitoring data according to the target warning requirements and obtains a risk report; Conduct risk classification and control of chemical parks based on risk reports.

2. The AI ​​brain early warning method for risk management in a chemical park according to claim 1, characterized in that: The risk monitoring data is transmitted to the backend server through the transmission module, including: The transmission module determines the number and location of IoT devices that acquire risk monitoring data; According to the number and location of IoT devices, IoT devices are clustered based on the K-means clustering algorithm to obtain several cluster sets; The IoT device corresponding to the cluster center in the cluster set is used as the cluster head node, and other IoT devices in the cluster set are used as member nodes to complete clustering; All cluster head nodes and member nodes form a clustered wireless transmission network, and transmit risk monitoring data to the background server based on the wireless transmission network.

3. The AI ​​brain early warning method for risk management in a chemical park according to claim 1, characterized in that: The IoT devices include sensors, surveillance cameras, and drones; The risk monitoring data includes first sensor data acquired by a sensor, first image data acquired by a surveillance camera, and second sensor data and second image data acquired by a drone.

4. The AI ​​brain early warning method for risk management in a chemical park according to claim 3, characterized in that: The backend server processes the risk monitoring data according to the target warning requirements and obtains a risk report, including: Performing image enhancement processing on the first image data and the second image data to obtain target image data; Performing data cleaning and data deduplication processing on the first sensor data and the second sensor data to obtain target sensor data; The backend server performs image recognition on the target image data according to the target warning requirements to obtain employee behavior data, infrastructure data, and equipment action data; The backend server analyzes the target sensor data according to the target warning requirements to obtain leakage emergency event data and environmental management event data; Perform data normalization on employee behavior data, infrastructure data, equipment action data, leakage emergency event data, and environmental management event data to obtain normalized data; According to the normalized data, the risk status value of each area in the chemical park is determined based on a preset algorithm, and a risk report is generated according to the risk status value of each area in the chemical park.

5. The AI ​​brain early warning method for risk management in a chemical park according to claim 4, characterized in that: According to the normalized data, the risk status value of each area in the chemical park is determined based on the preset algorithm, including: R=a×exp(-F)+b×log(1+P)+c×sqrt(C)+d×E 2 +e×(-M)wherein R is the risk status value of the area; F, P, C, E and M represent the normalized employee behavior data, infrastructure data, equipment action data, leakage emergency event data and environmental management event data respectively; a, b, c, d, e are the corresponding weight coefficients of the normalized employee behavior data, infrastructure data, equipment action data, leakage emergency event data and environmental management event data respectively; a+b+c+d+e=1.

6. The AI ​​brain early warning method for risk management in a chemical park according to claim 4, characterized in that: Performing image enhancement processing on the first image data and the second image data to obtain target image data includes: Performing a convolution operation on the first image data and the second image data based on a convolution kernel of edge detection, converting the convolved image into a processable two-dimensional matrix based on an image processing library, and calculating the variance of the image according to the two-dimensional matrix; Screen out images whose variance is less than a preset threshold as images to be enhanced; Calculate the grayscale histogram of the image to be enhanced, and determine the minimum grayscale level and the maximum grayscale level in the grayscale histogram; The grayscale value of each pixel in the image to be enhanced is mapped to a new grayscale range according to the minimum grayscale level and the maximum grayscale level to obtain the target image data.

7. The AI ​​brain early warning method for risk management in a chemical park according to claim 1, characterized in that: Obtain the target early warning needs of the chemical park, including: Based on the chemical layout of the chemical park, the historical warning needs corresponding to the known chemical park with the highest matching degree with the chemical layout are obtained from the warning database, and the layout differences are obtained; the layout differences include the type, quantity, location, and process flow of the equipment; The historical warning requirements are modified according to the layout differences to obtain the target warning requirements.

8. The AI ​​brain early warning method for risk management in a chemical park according to claim 1, characterized in that: Conduct risk classification management and control of chemical parks based on risk reports, including: Input the risk report into the pre-trained early warning model and output the risk classification control strategy; the risk classification control strategy includes determining whether there is a risk in the inspection item and determining the risk type when there is a risk, dividing the risk points, identifying the hazard sources included in the risk points, determining the levels of the hazard sources and risk points, formulating control measures and forming a risk control list; According to the risk control list, hidden dangers in the chemical park are checked and managed; the hidden dangers check and management include the allocation of check content, hidden danger check and hidden danger rectification after approval and assignment.

9. The AI ​​brain early warning method for risk management in a chemical park according to claim 8, characterized in that: The method for obtaining the early warning model includes: Obtain a training database, wherein the training database contains several sample risk reports, wherein each sample risk report contains S risk monitoring data and N early warning data of the chemical park, and the S risk monitoring data and the N early warning data are combined into an early warning matrix, wherein the early warning matrix has S rows and N columns, and the risk classification control strategy corresponding to each sample risk report is recorded at the same time to form a vector Y1, and the vector Y1 is removed after duplicate values ​​are formed; Deep learning based on training database; W1=rand(D,S)W2=rand(D,S) f=W2*max(zero(D,1),W1X i ) Among them, rand(D,S) generates a random matrix with D rows and S columns, and the value of each element in the matrix is ​​a random value between 0 and 1, W1 and W2 are deep learning adjustment coefficients, D is the number of values ​​in the vector Y, max() is the vector composed of the maximum values ​​of each row in the brackets, zero(D,1) generates a matrix of all 0s with D rows and 1 column, L is the activation function, X i is the value of the i-th column of the warning matrix X, and f is the value of i The vector function formed after mapping with W1 and W2; To X i The jth value in the vector function formed after mapping with W1 and W2, To X i The Y1th value in the vector function formed after mapping with W1 and W2; where j = 1, 2, 3, ... Y1 i -1, Y1 i +1...D, Y1 i +1 is the position of the value in vector Y corresponding to the i-th value in vector Y1, W ii,k,t is the value of the kth row and tth column of the random matrix Wii; Adjust the deep learning adjustment coefficient; in, L to W ii,k,t Find the partial derivative, WS ii,k,t To obtain the value after partial derivation, K = 1, 2, 3...D, t = 1, 2, 3...S; after adjusting all elements in the deep learning adjustment coefficients W1 and W2, new deep learning adjustment coefficients W1 and W2 can be obtained; According to the new deep learning adjustment coefficients W1, W2, iterate continuously until WS ii,k,t All are 0, and the target deep learning adjustment coefficient is obtained, and then the pre-trained early warning model is determined.

10. The AI ​​brain early warning method for risk management in a chemical park according to claim 3, characterized in that: The risk classification and control process also includes: For low-risk areas in chemical parks, normal signals are sent to staff, and risk monitoring data of chemical parks is continuously collected and updated in real time through IoT devices; For medium-risk areas in chemical parks, early warning signals and risk monitoring data of risk areas are sent to staff, and warning notices are sent to people around risk areas; For high-risk areas in chemical parks, alarm signals and risk monitoring data of risk areas are sent to staff, and emergency evacuation and avoidance notices are sent to people around the risk areas.

Images