Chemical industry park risk zoning and grading management and control method based on deep learning, chemical industry park risk zoning and grading platform and electronic equipment
Through deep learning technology and YOLOv8m network model, risk identification is carried out for chemical parks, combined with the 3D display and risk grading methods of Unity3D engine, the accuracy and personalized management of risk monitoring in chemical parks are solved, and efficient risk zoning and grading control is achieved.
Patent Information
- Application Number
- CN202411128647.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-16
- Publication Date
- 2024-11-15
AI Technical Summary
Risk monitoring in chemical parks relies on traditional means, has strong artificial dependence and low accuracy, and cannot manage different risk characteristics in real time and personalized, and is inaccurate in risk division and hierarchy.
The deep learning-based method is adopted to identify the risk of the three-dimensional data of the chemical park through the YOLOv8m network model, and display it in 3D visualization with the Unity3D engine, and risk hierarchical management is used using the work hazard analysis method, safety checklist analysis method and risk matrix analysis method.
It improves the accuracy and stability of risk identification, realizes accurate zoning and hierarchical control of risks in chemical parks, can display risk areas and situations in real time, and provide personalized risk management.
Smart Images

Figure CN118966780A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of chemical risk management, and is a chemical park risk zoning and grading control, platform and electronic equipment based on deep learning. Background Art
[0002] As one of the high-risk industries, the chemical industry involves many complex chemical reactions and material transformations in its production process, which poses potential safety hazards. At present, the chemical industry relies heavily on some traditional monitoring methods for risk monitoring, such as smoke sensors, infrared imagers, and traditional image processing algorithms, but traditional monitoring methods have certain disadvantages, such as difficulty in detecting in public places, strong dependence on manual labor, and low real-time performance. Furthermore, based on traditional risk monitoring methods, most of the existing chemical industry risk management and control platforms are built based on big data, with a single risk assessment and identification method, low risk assessment and identification accuracy, and focus on real-time information display of production process risks. It cannot complete personalized and differentiated management for different risk characteristics, and does not intuitively display risk areas, risk situations, and risk warnings. Summary of the invention
[0003] The present invention provides a chemical park risk zoning and grading management, platform and electronic equipment based on deep learning, which overcomes the shortcomings of the above-mentioned prior art and can effectively solve the problem of inaccurate risk zoning and grading management caused by the strong dependence on manual risk identification and low accuracy in the existing chemical park risk management.
[0004] One of the technical solutions of the present invention is achieved through the following measures: a risk zoning and grading control method for a chemical park based on deep learning, comprising: Import the 3D data of the chemical park into the Unity3D engine, create a 3D park space model of the chemical park, and visualize the layout of the chemical park in 3D, including buildings, equipment in buildings, and pipeline distribution; Obtain risk-related data to be identified, select one or more corresponding risk identification models in the risk identification model library according to the type of risk-related data to be identified, and input the risk-related data to be identified into the corresponding risk identification model, wherein each risk identification model in the risk identification model library is obtained by deep learning a YOLOv8m network model using a number of samples, and each of the several samples includes historical risk-related data and a corresponding risk identifier; Conduct risk assessment on the output results of each risk identification model, obtain risk zoning and grading results, and display them in the three-dimensional park space model.
[0005] The following are further optimizations and / or improvements to the above technical solutions: The process of constructing the risk identification model library includes obtaining historical risk-related data sets corresponding to various risks in the chemical park, using each historical risk-related data set to perform deep learning on the YOLOv8m network model to obtain the corresponding risk identification model, wherein the deep learning process of the YOLOv8m network model for each historical risk-related data set is the same, and the deep learning process of the risk identification model corresponding to a certain risk type specifically includes: Preprocess the historical risk-related data in a certain historical risk-related data set, and divide the historical risk-related data set into a training set, a validation set, and a test set in proportion, wherein the preprocessing includes: using a random combination of multiple enhancement methods in the enhancement method library to enhance the historical risk-related data, annotating the enhanced historical risk-related data using labelimage, saving the annotated data in xml format, and then converting it into a txt format file required by the YOLOv8m network model; Based on the preset loss function, the YOLOv8m network model is trained and adjusted using the training set and validation set until the loss function converges to obtain a risk identification model. The risk identification model is tested using the test set, and the test results are evaluated, and the output is a risk identification model whose evaluation results meet the requirements.
[0006] The above loss function includes classification loss BCE Loss and regression loss VFL Loss + CIOU Loss.
[0007] The above-mentioned risk assessment is performed on the output results of each risk identification model to obtain risk zoning and grading results, including: calling one or more risk grading methods in the risk grading method library to perform risk assessment on the output results of the risk identification model to obtain risk grading results, and issuing alarms through corresponding alarm levels, wherein the risk grading method library includes work hazard analysis method, safety checklist analysis method and risk matrix analysis method.
[0008] The above also includes scene roaming, data storage, and data visualization; Scene roaming, interacting with the three-dimensional park space model, and interactively viewing the real-time dynamic data and static data of the chemical park from the perspective of safety inspection; Data storage, which stores all platform data generated and received in the risk zoning and grading management of chemical parks, including risk-related data, operation data, and parameters; Data visualization display, calling visualization charts to display various types of stored data.
[0009] The second technical solution of the present invention is achieved through the following measures: a chemical park risk zoning and grading management and control platform based on deep learning, comprising: The data access unit obtains the external data required for the operation of the platform, including the three-dimensional data of the chemical park, the historical risk-related data set corresponding to various risks, the risk-related data to be identified at each risk point, and the external input parameter values; The 3D model creation unit imports the 3D data of the chemical park into the Unity3D engine, creates a 3D park space model of the chemical park, and displays the setting of the chemical park in 3D visualization, including buildings, equipment in buildings, and pipeline distribution; The risk identification unit obtains data related to the risks to be identified, selects one or more corresponding risk identification models in the risk identification model library according to the data type of the risks to be identified, and inputs the data related to the risks to be identified into the corresponding risk identification models, wherein each risk identification model in the risk identification model library is obtained by deep learning a YOLOv8m network model using a number of samples, and each of the several samples includes historical risk-related data and a corresponding risk identifier; The risk classification unit conducts risk assessment on the output results of each risk identification model, obtains risk zoning and classification results, and displays them in the three-dimensional park space model; The database stores all platform data generated and received in the risk zoning and grading management of chemical parks, including risk-related data, operation data, and parameters.
[0010] The following are further optimizations and / or improvements to the above technical solutions: The above-mentioned risk identification unit includes an identification module, a risk identification model library, and a model building module; The identification module obtains the risk-related data to be identified, selects one or more corresponding risk identification models in the risk identification model library according to the type of risk-related data to be identified, and inputs the risk-related data to be identified into the corresponding risk identification model to obtain the risk identification result; A risk identification model library stores multiple risk identification models, each of which is obtained using a model building unit; The model building module obtains the historical risk-related data sets corresponding to various risks in the chemical park, and uses each historical risk-related data set to perform deep learning on the YOLOv8m network model to obtain the corresponding risk identification model. The deep learning process of each historical risk-related data set for the YOLOv8m network model is the same, specifically including: Preprocess the historical risk-related data in a certain historical risk-related data set, and divide the historical risk-related data set into a training set, a validation set, and a test set in proportion, wherein the preprocessing includes: using a random combination of multiple enhancement methods in the enhancement method library to enhance the historical risk-related data, annotating the enhanced historical risk-related data using labelimage, saving the annotated data in xml format, and then converting it into a txt format file required by the YOLOv8m network model; Based on the preset loss function, the YOLOv8m network model is trained and adjusted using the training set and validation set until the loss function converges to obtain a risk identification model. The risk identification model is tested using the test set, and the test results are evaluated, and the output is a risk identification model whose evaluation results meet the requirements.
[0011] The above also includes: The scene roaming unit interacts with the three-dimensional park space model and interactively views the real-time dynamic and static data of the chemical park from the perspective of safety inspection; Data visualization display unit, calling visualization charts to display various types of stored data; The user management module implements platform user management, which includes permission management, identity management, and login management.
[0012] The present invention discloses a risk zoning and grading management and control method and platform for a chemical park based on deep learning. The chemical park is virtually displayed based on the Unity3D engine, and the risks in the chemical park are identified by using a risk identification model obtained by deep learning using a YOLOv8m network model. Compared with the risk identification method that combines sensor data collection, traditional image processing algorithms and empiricism, the accuracy and stability of risk identification are improved, and risk grading management is performed based on the risk identification results of the risk identification model, thereby accurately realizing risk zoning and grading management of chemical parks. BRIEF DESCRIPTION OF THE DRAWINGS
[0013] Attached Figure 1 It is a schematic diagram of the process of the risk zoning and grading management and control method of the present invention.
[0014] Attached Figure 2 It is a schematic diagram of the process of constructing the risk identification model in the present invention.
[0015] Attached Figure 3 It is a schematic diagram of the structure of the risk zoning and grading management and control platform of the present invention. DETAILED DESCRIPTION
[0016] The present invention is not limited by the following embodiments, and specific implementation methods can be determined based on the technical solution of the present invention and actual conditions.
[0017] The present invention will be further described below in conjunction with embodiments and drawings: Embodiment 1: As attached Figure 1 As shown, the embodiment of the present invention discloses a risk zoning and grading management method for a chemical park based on deep learning, including: Step S110, importing the three-dimensional data of the chemical park into the Unity3D engine, creating a three-dimensional park space model of the chemical park, and visually displaying the setting of the chemical park in 3D, including buildings, equipment in the buildings, and pipeline distribution; Step S120, obtaining risk-related data to be identified, selecting one or more corresponding risk identification models in the risk identification model library according to the type of risk-related data to be identified, and inputting the risk-related data to be identified into the corresponding risk identification model, wherein each risk identification model in the risk identification model library is obtained by deep learning a YOLOv8m network model using a number of samples, and each of the several samples includes historical risk-related data and a corresponding risk identifier; Step S130, performing risk assessment on the output results of each risk identification model, obtaining risk zoning and grading results, and displaying them in the three-dimensional park space model.
[0018] The above-mentioned data related to the risks to be identified are obtained, and one or more corresponding risk identification models are selected in the risk identification model library according to the type of the data related to the risks to be identified, and the data related to the risks to be identified are input into the corresponding risk identification model for risk identification. The risk identification model in the risk identification model library can identify whether the workers are wearing necessary protective equipment such as safety helmets and protective glasses, identify the workers' dangerous actions, identify the location and level of chemical leakage in the working environment, and identify the location and level of gas leakage.
[0019] The present invention discloses a risk zoning and grading management and control method for a chemical park based on deep learning. The chemical park is virtually displayed based on the Unity3D engine, and the risks in the chemical park are identified by using a risk identification model obtained by deep learning using a YOLOv8m network model. Compared with the risk identification method that combines sensor data collection, traditional image processing algorithms and empiricism, the accuracy and stability of risk identification are improved, thereby enabling accurate risk zoning and grading management and control of chemical parks.
[0020] Embodiment 2: The embodiment of the present invention discloses a risk zoning and grading management method for a chemical park based on deep learning, comprising: Step S210, importing the three-dimensional data of the chemical park into the Unity3D engine, creating a three-dimensional park space model of the chemical park, and 3D visualization to display the area settings of the chemical park, including buildings, equipment in buildings, and pipeline distribution.
[0021] In the above steps, the three-dimensional data of the chemical park is obtained through a number of sensors and measuring equipment preset in the chemical park.
[0022] Step S220, obtain the risk-related data to be identified, select one or more corresponding risk identification models in the risk identification model library according to the type of the risk-related data to be identified, and input the risk-related data to be identified into the corresponding risk identification model, wherein each risk identification model in the risk identification model library is obtained by deep learning the YOLOv8m network model using several samples, and each of the several samples includes historical risk-related data and a corresponding risk identifier.
[0023] In the above steps, the process of building a risk identification model library includes obtaining historical risk-related data sets corresponding to various risks in the chemical park, using each historical risk-related data set to perform deep learning on the YOLOv8m network model to obtain the corresponding risk identification model, where each historical risk-related data set has the same deep learning process for the YOLOv8m network model, as shown in the attached figure. Figure 2 As shown in the figure, the deep learning process of the risk identification model corresponding to a certain risk type specifically includes: Step S221, preprocessing the historical risk-related data in a certain historical risk-related data set, and dividing the historical risk-related data set into a training set, a validation set, and a test set in proportion, wherein the preprocessing includes: using a random combination of multiple enhancement methods in the enhancement method library to enhance the historical risk-related data, using labelimage to label the enhanced historical risk-related data for risk, saving it in xml format after the labeling is completed, and then converting it into a txt format file required by the YOLOv8m network model; Among them, the historical risk-related data set is related to the risk type to be identified, may involve multiple scenarios, and can be obtained by web crawling. The historical risk-related data are all image data.
[0024] Among them, a random combination of multiple enhancement methods in the enhancement method library is used to enhance the historical risk-related data to increase the diversity of the historical risk-related data, thereby improving the robustness of the model obtained by deep learning. The enhancement method library includes data enhancement methods such as Gaussian blur, horizontal flip, vertical flip, non-proportional scaling, random translation, perspective transformation and random cropping.
[0025] Among them, labelimage is used to annotate risks, and during the image annotation process, abnormal images can be deleted.
[0026] Step S222, based on a preset loss function, the YOLOv8m network model is trained and adjusted using the training set and the validation set until the loss function converges to obtain a risk identification model; The above YOLOv8m is a model variant of YOLOv8, including the following key features: Adaptive image size preprocessing: It can automatically adjust the input size of the model according to the input image size to optimize the model efficiency; Adaptive output layer: dynamically adjusts the number of channels of the output feature map to adapt to different task requirements; Path aggregation: Aggregate information through different paths to improve the ability of small target detection; Spatial Weighted Attention (SWA): Used to process spatial data, such as images, the pixels at each position are weighted. Spatial weighted attention is used to focus on certain specific areas in the image while ignoring other irrelevant or unimportant areas.
[0027] Model Scaling: Model efficiency can be further optimized through model scaling techniques.
[0028] The above preset loss functions include classification loss BCE Loss and regression loss VFL Loss + CIOU Loss.
[0029] The above classification loss BCE Loss is as follows: Where N is the number of samples, is the actual label of the i-th sample, which takes a value of 0 or 1. is the predicted value (identification value) of the sample, and its value range is between (0,1).
[0030] The above regression loss VFL Loss is as follows: Among them, q is the label, q is the IoU of bbox and gt for positive samples, and q=0 for negative samples The above regression loss CIOU Loss is as follows: Step S223, using the test set to test the risk identification model, and evaluating the test results, and outputting the risk identification model whose evaluation results meet the requirements.
[0031] The software and hardware conditions required for building the risk identification model in the above step 220 are: Windows 10 operating system, Python 3.7, PyTorch 1.9.1, Nvidia RTXA5000 GPU, 24GB RAM.
[0032] Taking fire risk identification as an example, 3,000 historical smoke data of multiple scenes were collected and annotated by labelimg, and divided into training set, validation set and test set. After 300 epochs of training, the test set was used for testing, and the precision, recall rate and mean average precision mAP50 indicators were used as evaluation criteria. The experimental results are shown in the following table: As can be seen from the table, the YOLOv8m model has achieved good performance in both recall and precision, and can meet the requirements of risk identification in chemical parks.
[0033] Step S230, performing risk assessment on the output results of each risk identification model, obtaining risk zoning and grading results, and displaying them in the three-dimensional park space model.
[0034] The above-mentioned output results of each risk identification model are subjected to risk assessment to obtain risk zoning and grading results, which specifically includes calling one or more risk grading methods in the risk grading method library to perform risk assessment on the output results of the risk identification model to obtain risk grading results, and issuing alarms through corresponding alarm levels, wherein the risk grading method library includes work hazard analysis method, safety checklist analysis method and risk matrix analysis method.
[0035] The work hazard analysis method, safety checklist analysis method and risk matrix analysis method are as follows: (1) Job Hazard Analysis (JHA) Job hazard analysis is a qualitative risk analysis and identification method. It is a risk identification technology based on work activities and is used to effectively identify unsafe human behaviors, unsafe conditions of objects, unsafe factors in the environment, and management defects.
[0036] (2) Safety Checklist Analysis Method (SCL) The safety checklist method is a qualitative risk analysis and identification method. It lists a series of items on a checklist for analysis to determine whether the status of the system and the site meets the safety requirements. It is a method to discover the risks in the system through inspection and propose improvement measures.
[0037] (3) Risk Matrix Analysis (LS) The risk matrix analysis method is a semi-quantitative risk assessment method. When conducting risk assessment, the severity of the consequences of risk events is relatively qualitatively divided into several levels, and the possibility of risk events is also relatively qualitatively divided into several levels. Then, a table is made with severity as columns and possibility as rows, and a qualitative weighted index is given at the intersection of rows and columns.
[0038] Embodiment 3: The embodiment of the present invention discloses a risk zoning and grading control method for a chemical park based on deep learning, which is a further optimization of the above embodiment and also includes scene roaming, data storage, and data visualization display; Scene roaming, interacting with the three-dimensional park space model, and interactively viewing the real-time dynamic data and static data of the chemical park from the perspective of safety inspection; Data storage, which stores all platform data generated and received in the risk zoning and grading management of chemical parks, including risk-related data, operation data, and parameters; Data visualization display, calling visualization charts to display various types of stored data.
[0039] Embodiment 4: As attached Figure 3 As shown, the embodiment of the present invention discloses a chemical park risk zoning and grading management and control platform based on deep learning, including: (I) Data access unit, which obtains external data required for platform operation, including three-dimensional data of chemical parks, historical risk-related data sets corresponding to various risks, risk-related data to be identified at each risk point, and external input parameter values.
[0040] (ii) A three-dimensional model creation unit, which imports the three-dimensional data of the chemical park into the Unity3D engine, creates a three-dimensional park space model of the chemical park, and displays the layout of the chemical park in 3D visualization, including buildings, equipment in buildings, and pipeline distribution.
[0041] (3) A risk identification unit, which obtains data related to the risks to be identified, selects one or more corresponding risk identification models from a risk identification model library according to the type of data related to the risks to be identified, and inputs the data related to the risks to be identified into the corresponding risk identification models, wherein each risk identification model in the risk identification model library is obtained by deep learning a YOLOv8m network model using a number of samples, and each of the several samples includes historical risk-related data and a corresponding risk identifier; The above-mentioned risk identification unit includes an identification module, a risk identification model library, and a model building unit; (1) An identification module, which obtains data related to the risks to be identified, selects one or more corresponding risk identification models from the risk identification model library according to the type of data related to the risks to be identified, and inputs the data related to the risks to be identified into the corresponding risk identification models to obtain risk identification results; (2) Risk identification model library, which stores multiple risk identification models, each of which is obtained using a model building unit; (3) Model building module, which obtains historical risk-related data sets corresponding to various risks in the chemical park, and uses each historical risk-related data set to perform deep learning on the YOLOv8m network model to obtain the corresponding risk identification model. The deep learning process of each historical risk-related data set for the YOLOv8m network model is the same, specifically including: Preprocess the historical risk-related data in a certain historical risk-related data set, and divide the historical risk-related data set into a training set, a validation set, and a test set in proportion, wherein the preprocessing includes: using a random combination of multiple enhancement methods in the enhancement method library to enhance the historical risk-related data, annotating the enhanced historical risk-related data using labelimage, saving the annotated data in xml format, and then converting it into a txt format file required by the YOLOv8m network model; Based on the preset loss function, the YOLOv8m network model is trained and adjusted using the training set and validation set until the loss function converges to obtain a risk identification model. The risk identification model is tested using the test set, and the test results are evaluated, and the output is a risk identification model whose evaluation results meet the requirements.
[0042] (iv) Risk grading unit, which conducts risk assessment on the output results of each risk identification model, obtains risk zoning and grading results, and displays them in the three-dimensional park space model; The above-mentioned risk grading unit includes a risk grading method library and a risk grading execution module; (1) Risk grading method library, which stores various risk grading methods; (2) Risk grading execution module, which calls one or more risk grading methods in the risk grading method library to conduct risk assessment on the output results of the risk identification model, obtains the risk grading results, and issues an alarm according to the corresponding alarm level. The risk grading method library includes work hazard analysis method, safety checklist analysis method and risk matrix analysis method.
[0043] (V) Scene roaming unit, which interacts with the three-dimensional park space model and interactively views the real-time dynamic data and static data of the chemical park from the perspective of safety inspection. When roaming, the perspective can be switched by using the direction keys to control the direction and the mouse to control the rotation of the lens.
[0044] (VI) Data visualization display unit, which calls visualization charts to display various types of stored data, and can also visualize risk warnings. Specific visualization charts can be selected as needed.
[0045] (VII) User management module, which implements platform user management, including permission management, identity management, and login management. The user management module is used to improve the security and management efficiency of the system.
[0046] (8) Database, which stores all platform data generated and received during the risk zoning and grading management and control of chemical parks, including risk-related data, operating data, and parameters.
[0047] The development environment of the platform in this embodiment can be specifically described as follows: 1. Development Environment (1) Hardware environment Processor: AMD Ryzen 7 4800U with Radeon Graphics; Memory: 16GB RAM; Hard disk: 1GB free hard disk space; Display card: High color display adapter supporting resolution of 1024×768 and above; (2) Software environment Recommended configuration: Windows 10 or later, macOS 10.14 or later; (3) Programming language Programming language: C#; Programming IDE: Visual Studio 2022; Database management: MySQL 8.0 Navicat Premium 12; Integrated development environment: Unity 2022.3.2f1c1; (II) Project development related software: (1) MySQL: MySQL is a relational database management system developed by the Swedish company MySQL AB and currently owned by Oracle. A relational database stores data in different tables instead of putting all data in one large warehouse, which increases speed and flexibility.
[0048] (2) Navicat Premium: Navicat Premium is a set of database development tools. When creating the table structure, you can directly fill in the relevant requirements, and the system will automatically generate the DDL language without having to write the SQL language yourself; when filling in the table data, you can directly fill in the table information similar to the Excel table structure. Every time you fill in a row of data, the system will automatically generate the SQL language; it is mainly divided into three parts: 1. Installation of MySQL 2. Installation of Navicat 3. Unity connects to MySQL for testing.
[0049] MySQL installation: MySQL 8.0 is installed Navicat installation: Navicat Premium12.0 Unity is connected to a MySQL database: data is entered into the database, read out when it needs to be displayed, and then printed in the Unity interface.
[0050] (3) Visual Studio 2022: Visual Studio is a powerful developer tool that allows you to complete the entire development cycle in one place. It is a comprehensive integrated development environment (IDE) that allows you to write, edit, debug, and build code, and then deploy your app. In addition to code editing and debugging, Visual Studio also includes compilers, code completion tools, source code management, extensions, and many other features to improve every stage of the software development process.
[0051] (III) Project related scripts (1) SQL database connection script (2) UIControl: This script is the UI control script of the main menu interface. It is mainly used to display the buttons of the main menu interface, the display of the scene interaction logo and the animation of the upper and lower title bars. The DOTween plug-in is used to achieve the animation effect. It is mounted on Canvas (1). For specific comments, see the script.
[0052] (3) Area1UIControll: This script is mainly the UI control script for the five risk areas, including camera switching, display of scene interaction signs, etc. It is installed on the canvas of each risk area. Specific comments can be found in the script. The same applies to the UI control scripts of other partitions.
[0053] (4) CameraControl: This script is mainly used to switch the camera perspective. It can set the angle and height of the camera rotation, so that there is an effect of zooming in and out. It is installed in Canvas (1) and is triggered by a button setting trigger event.
[0054] (5) First-person roaming: CameraController and playerControll. These two scripts are mainly the control scripts of the camera and the character when the roaming function is turned on. The parameters can also be adjusted directly in Unity. PlayerControll is installed on BlueSuitFree01. BlueSuitFree01 is a character model that simulates a person in the park. CameraController is installed on BlueSuitFree01->player camera.
[0055] (6) AlwaysFaceCamera: This script is installed on the small icon of scene interaction. Its function is to make the icon always face the camera, which is convenient for click interaction. This script was found on the Internet.
[0056] (7) Change: The change script is mainly used to switch the cameras in each risk area. The activation and deactivation of the camera can be set by clicking the interactive icon.
[0057] (8) OrbitCamera: This script implements the camera rotation and the scroll wheel to zoom in and out. OrbitCamera1 and OrbitCamera2 are each installed on different cameras. All cameras are managed in the same folder as the camera.
[0058] (9) Icon animation: realize the animation of the scene interaction logo icon floating up and down. Mounted on the scene interaction logo.
[0059] (10) Roaming function switch: Turn on the control of the roaming function UI interface, installed in Canvas (1)->Roaming function.
[0060] Embodiment 5: The embodiment of the present invention discloses an electronic device, including a processor and a memory, wherein the memory stores a computer program, and the computer program is loaded and executed by the processor to implement a risk zoning and grading management method for a chemical park based on deep learning.
[0061] The processor may be a central processing unit (CPU), a general purpose processor, a digital signal processor (DSP), an ASIC, an FPGA or other programmable logic device, a transistor logic device, a hardware component or any combination thereof. It may implement or execute various exemplary logic blocks, modules and circuits described in conjunction with the disclosure of this application. It may also be a combination that implements computing functions, such as a combination of one or more microprocessors, a combination of a DSP and a microprocessor, and the like. The memory may include, but is not limited to, various media that can store computer programs, such as a USB flash drive, a read-only memory, a mobile hard disk, a magnetic disk or an optical disk.
[0062] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present application may adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes. The schemes in the embodiments of the present application may be implemented in various computer languages, for example, object-oriented programming language Java and literal scripting language JavaScript, etc.
[0063] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.
[0064] These computer program instructions may also be stored in a computer readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture including an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.
[0065] The above technical features constitute the best embodiment of the present invention, which has strong adaptability and best implementation effect. Non-essential technical features can be added or reduced according to actual needs to meet the requirements of different situations.
Claims
1. A risk zoning and grading management method for chemical parks based on deep learning, characterized in that: include: Import the 3D data of the chemical park into the Unity3D engine, create a 3D park space model of the chemical park, and visualize the layout of the chemical park in 3D, including buildings, equipment in buildings, and pipeline distribution; Obtain risk-related data to be identified, select one or more corresponding risk identification models in the risk identification model library according to the type of risk-related data to be identified, and input the risk-related data to be identified into the corresponding risk identification model, wherein each risk identification model in the risk identification model library is obtained by deep learning a YOLOv8m network model using a number of samples, and each of the several samples includes historical risk-related data and a corresponding risk identifier; Conduct risk assessment on the output results of each risk identification model, obtain risk zoning and grading results, and display them in the three-dimensional park space model.
2. The method for risk zoning and grading management of chemical parks based on deep learning according to claim 1 is characterized in that: The process of building a risk identification model library includes obtaining historical risk-related data sets corresponding to various risks in the chemical park, using each historical risk-related data set to perform deep learning on the YOLOv8m network model to obtain the corresponding risk identification model, wherein the deep learning process of each historical risk-related data set on the YOLOv8m network model is the same, and the deep learning process of the risk identification model corresponding to a certain risk type specifically includes: Preprocess the historical risk-related data in a certain historical risk-related data set, and divide the historical risk-related data set into a training set, a validation set, and a test set in proportion, wherein the preprocessing includes: using a random combination of multiple enhancement methods in the enhancement method library to enhance the historical risk-related data, annotating the enhanced historical risk-related data using labelimage, saving the annotated data in xml format, and then converting it into a txt format file required by the YOLOv8m network model; Based on the preset loss function, the YOLOv8m network model is trained and adjusted using the training set and validation set until the loss function converges to obtain a risk identification model. The risk identification model is tested using the test set, and the test results are evaluated, and the output is a risk identification model whose evaluation results meet the requirements.
3. The method for risk zoning and grading management of chemical parks based on deep learning according to claim 2 is characterized in that: The loss function includes classification loss BCE Loss and regression loss VFL Loss + CIOU Loss.
4. The method for risk zoning and grading management of chemical parks based on deep learning according to any one of claims 1 to 3, characterized in that: The risk assessment of the output results of each risk identification model to obtain risk zoning and grading results includes: calling one or more risk grading methods in a risk grading method library to perform risk assessment on the output results of the risk identification model to obtain risk grading results, and issuing alarms through corresponding alarm levels, wherein the risk grading method library includes a work hazard analysis method, a safety checklist analysis method and a risk matrix analysis method.
5. The method for risk zoning and grading management of chemical parks based on deep learning according to any one of claims 1 to 5, characterized in that: It also includes scene roaming, data storage, and data visualization; Scene roaming, interacting with the three-dimensional park space model, and interactively viewing the real-time dynamic data and static data of the chemical park from the perspective of safety inspection; Data storage, which stores all platform data generated and received in the risk zoning and grading management of chemical parks, including risk-related data, operation data, and parameters; Data visualization display, calling visualization charts to display various types of stored data.
6. A chemical park risk zoning and grading management and control platform based on deep learning using the method as described in any one of claims 1 to 5, characterized in that: include: The data access unit obtains the external data required for the operation of the platform, including the three-dimensional data of the chemical park, the historical risk-related data set corresponding to various risks, the risk-related data to be identified at each risk point, and the external input parameter values; The 3D model creation unit imports the 3D data of the chemical park into the Unity3D engine, creates a 3D park space model of the chemical park, and displays the setting of the chemical park in 3D visualization, including buildings, equipment in buildings, and pipeline distribution; The risk identification unit obtains data related to the risks to be identified, selects one or more corresponding risk identification models in the risk identification model library according to the data type of the risks to be identified, and inputs the data related to the risks to be identified into the corresponding risk identification models, wherein each risk identification model in the risk identification model library is obtained by deep learning a YOLOv8m network model using a number of samples, and each of the several samples includes historical risk-related data and a corresponding risk identifier; The risk classification unit conducts risk assessment on the output results of each risk identification model, obtains risk zoning and classification results, and displays them in the three-dimensional park space model; The database stores all platform data generated and received in the risk zoning and grading management of chemical parks, including risk-related data, operation data, and parameters.
7. The chemical park risk zoning and grading management and control platform based on deep learning according to claim 6 is characterized in that: The risk identification unit includes an identification module, a risk identification model library, and a model building module; The identification module obtains the risk-related data to be identified, selects one or more corresponding risk identification models in the risk identification model library according to the type of risk-related data to be identified, and inputs the risk-related data to be identified into the corresponding risk identification model to obtain the risk identification result; A risk identification model library stores multiple risk identification models, each of which is obtained using a model building unit; The model building module obtains the historical risk-related data sets corresponding to various risks in the chemical park, and uses each historical risk-related data set to perform deep learning on the YOLOv8m network model to obtain the corresponding risk identification model. The deep learning process of each historical risk-related data set for the YOLOv8m network model is the same, specifically including: Preprocess the historical risk-related data in a certain historical risk-related data set, and divide the historical risk-related data set into a training set, a validation set, and a test set in proportion, wherein the preprocessing includes: using a random combination of multiple enhancement methods in the enhancement method library to enhance the historical risk-related data, annotating the enhanced historical risk-related data using labelimage, saving the annotated data in xml format, and then converting it into a txt format file required by the YOLOv8m network model; Based on the preset loss function, the YOLOv8m network model is trained and adjusted using the training set and validation set until the loss function converges to obtain a risk identification model. The risk identification model is tested using the test set, and the test results are evaluated, and the output is a risk identification model whose evaluation results meet the requirements.
8. The chemical park risk zoning and grading management and control platform based on deep learning according to claim 6 or 7, characterized in that: The risk grading unit includes a risk grading method library and a risk grading execution module: Risk grading method library, which stores various risk grading methods; The risk grading execution module calls one or more risk grading methods in the risk grading method library to conduct risk assessment on the output results of the risk identification model, obtains the risk grading results, and issues alarms through the corresponding alarm levels. The risk grading method library includes work hazard analysis method, safety checklist analysis method and risk matrix analysis method.
9. The chemical park risk zoning and grading management and control platform based on deep learning according to any one of claims 6 to 8, characterized in that: Also includes: The scene roaming unit interacts with the three-dimensional park space model and interactively views the real-time dynamic and static data of the chemical park from the perspective of safety inspection; Data visualization display unit, calling visualization charts to display various types of stored data; The user management module implements platform user management, which includes permission management, identity management, and login management.
10. An electronic device, characterized in that: It includes a processor and a memory, wherein a computer program is stored in the memory, and the computer program is loaded and executed by the processor to implement the chemical park risk zoning and grading management method based on deep learning as described in any one of claims 1 to 6.
