A method, system, medium and product for safety hazard investigation of chemical enterprises

Abstract

The present application relates to the technical field of potential hazard investigation, and in particular, to a method, a system, a medium, and a product for investigating safety hazards in chemical enterprises. The method includes performing feature recognition on an image to be investigated according to a trained online recognition model to obtain online recognition features; comparing the online recognition features with a preset potential hazard feature form to obtain first recognition data; determining on-site selected potential hazard features based on on-site environmental parameters; obtaining observation images from the area to be investigated based on the on-site selected potential hazard features, and performing feature recognition on the observation images according to the trained online recognition model to obtain observation recognition features; comparing the online recognition features with the observation recognition features to obtain second recognition data; generating potential hazard investigation feedback data based on the first recognition data and the second recognition data, and feeding back the potential hazard investigation feedback data. The present application facilitates improving the accuracy of potential hazard investigation.

Description

Technical Field

[0001] The present application relates to the technical field of hidden danger investigation, and in particular to a method, a system, a medium and a product for investigating safety hidden dangers in chemical enterprises. Background Art

[0002] During the production process of chemical enterprises, due to their complex production processes, various types of equipment, and harsh operating conditions, etc., there may be many potential safety hazards hidden in the production process. If these hidden dangers are not detected and eliminated in time, it is very likely to cause serious safety accidents such as fires, explosions, poisonings, and leakages, which will not only cause huge economic losses to the enterprise, but also seriously threaten the lives of employees and cause inestimable damage to the surrounding environment.

[0003] Conventional technical means mainly rely on manual identification of potential safety hazards through means such as on-site observation, equipment inspection, and process analysis, or use various sensors installed in the production site, such as temperature sensors, pressure sensors, vibration sensors, etc., to monitor the operating status of equipment and the environment in real time and prompt possible safety hazards. However, the manual investigation method highly depends on the professional qualities, experience accumulation, and sense of responsibility of personnel. There are differences in the cognition and judgment criteria of hidden dangers among different investigation personnel, and it is very easy to have missed inspections and misjudgments. Moreover, although sensors can provide real-time feedback on some key parameters, the sensors themselves have problems such as accuracy errors and insufficient stability, and are greatly affected by complex environmental factors in the chemical industry site, such as strongly corrosive gases and electromagnetic interference. Therefore, it may lead to inaccurate monitoring data and cause false alarms or missed alarms. Summary of the Invention

[0004] In order to improve the accuracy of hidden danger investigation and reduce the probability of false alarms or missed alarms, the present application provides a method, a system, a medium and a product for investigating safety hidden dangers in chemical enterprises.

[0005] In a first aspect, the present application provides a method for investigating safety hidden dangers in chemical enterprises, adopting the following technical solutions:

[0006] A method for investigating safety hidden dangers in chemical enterprises includes:

[0007] Obtain an image to be investigated, perform feature recognition on the image to be investigated according to a trained online recognition model, and obtain online recognition features;

[0008] Compare the online recognition features with a preset hidden danger feature form, and obtain first recognition data. The preset hidden danger feature form contains multiple known hidden danger features, and the first recognition data is the known hidden danger features whose feature similarity with the online recognition features is higher than a first preset threshold;

[0009] Obtain the on-site environmental parameters corresponding to the area to be investigated, and determine the on-site selected potential hazard features based on the on-site environmental parameters;

[0010] Obtain the observation images from the area to be investigated based on the on-site selected potential hazard features, and perform feature recognition on the observation images according to the trained online recognition model to obtain the observation recognition features;

[0011] Compare the online recognition features with the observation recognition features to obtain the second recognition data, where the second recognition data is the observation recognition features whose feature similarity with the online recognition features is higher than the second preset threshold;

[0012] Generate potential hazard investigation feedback data based on the first recognition data and the second recognition data, and feedback the potential hazard investigation feedback data.

[0013] By adopting the above technical solution, by comparing the online recognition features of the images to be investigated with these known potential hazard features, it is convenient to quickly screen out the possible types of potential hazards in the area to be investigated. Then, combined with the potential hazard features selected by the on-site staff for targeted observation and comparison, it is convenient to discover more potential hazards and avoid omissions. By combining the results of dual feature comparisons to determine the final potential hazard investigation feedback data, it is convenient to analyze and identify potential hazards from different angles, thereby facilitating the reduction of deviations that may occur due to single comparison, and further facilitating the improvement of the accuracy of potential hazard investigation results.

[0014] In a possible implementation manner, the determining the on-site selected potential hazard features based on the on-site environmental parameters includes:

[0015] Construct a simulation production model based on the on-site environmental parameters, and drive the simulation generation model to perform simulation production based on the production plan data;

[0016] Record each simulated potential hazard feature that appears during the simulation production process, and the simulated potential hazard location corresponding to each simulated potential hazard feature, and generate a simulated potential hazard list based on the simulated potential hazard features;

[0017] According to each simulated potential hazard feature and the simulated potential hazard location corresponding to each simulated potential hazard feature, determine the concerned simulated potential hazard locations, where the simulated potential hazard feature values corresponding to the concerned simulated potential hazard features at the concerned simulated potential hazard locations are higher than the preset simulated feature threshold;

[0018] Determine the potential hazard positioning area based on the concerned simulated potential hazard locations and the corresponding simulated potential hazard feature values, and select the concerned on-site staff from the potential hazard positioning area;

[0019] Feed back the simulated potential hazard list to the on-site staff to be concerned, so as to remind the on-site staff to be concerned to select from the simulated potential hazard list and feedback the selection results;

[0020] Determine the on-site selected potential hazard features according to the selection feedback results of the on-site staff to be concerned.

[0021] By adopting the above technical solution, before targeted observation based on the potential hazard features selected by the on-site staff, a simulated production model is constructed according to the on-site environmental parameters, and the model is driven by the production plan data for simulated production, so as to simulate the equipment operation status and parameter changes under different production conditions. Then, the potential hazard features and locations occurring during the simulated production process are sorted into a list form for the on-site staff to be concerned to select, rather than providing an inherent selection list for the on-site staff to be concerned to select. A simulated potential hazard list is generated based on the potential hazards that may be encountered in the production work in the next period of time, which is convenient for improving the adaptability between the simulated potential hazard list and the actual potential hazard situation, and thus convenient for enhancing the effectiveness of on-site selection of potential hazard features.

[0022] In a possible implementation manner, obtaining the observation images from the area to be investigated based on the on-site selected potential hazard features includes:

[0023] Based on the mapping relationship between the on-site selected potential hazard features and the preset shooting strategy, determine the target shooting strategy corresponding to the on-site selected potential hazard features;

[0024] Based on the simulated potential hazard location corresponding to the on-site selected potential hazard features and the concerned simulated potential hazard location, determine the target shooting route;

[0025] Generate a target shooting instruction based on the target shooting strategy and the target shooting route, and control the shooting device to shoot according to the target shooting instruction to obtain the observation images from the area to be investigated.

[0026] By adopting the above technical solution, based on the mapping relationship between the on-site selected potential hazard features and the preset shooting strategy, it is convenient to allocate the most suitable target shooting strategy for different on-site selected potential hazard features, improve the integrity and accuracy of the captured images, avoid the loss of key information caused by improper shooting, and thus greatly improve the accuracy of potential hazard identification. Determine the target shooting route based on the simulated potential hazard location and the concerned simulated potential hazard location, so that the shooting device can directly go to the high-risk area of potential hazards for shooting, which is convenient to effectively avoid the situation of missing important potential hazard points caused by blind shooting.

[0027] In a possible implementation manner, generating the potential hazard investigation feedback data based on the first identification data and the second identification data includes:

[0028] Generate a hidden danger investigation list based on the first identification data and the second identification data, and determine an associated hidden danger data group from the first identification data and the second identification data based on a preset hidden danger association map;

[0029] Determine the hidden danger frequency, abnormal result image, and hidden danger source corresponding to the associated hidden danger data group based on historical hidden danger records, and superimpose the hidden danger frequency and the hidden danger source onto the abnormal result image to obtain a parameter integration image;

[0030] Superimpose the parameter integration image corresponding to the associated hidden danger data group onto the hidden danger investigation list to obtain the hidden danger investigation feedback data.

[0031] By adopting the above technical solution, the first identification data obtained by comparing the global hidden danger database and the second identification data obtained by on-site targeted observation are fused, which is convenient for verifying the existence of hidden danger characteristics and supplementing associated hidden dangers, avoiding omissions and misjudgments of a single data source. Based on a preset hidden danger association map, the scattered hidden danger data are organized into a structured associated hidden danger data group, which is convenient for improving the clarity of the investigation logic. By superimposing the hidden danger frequency and the hidden danger source of the associated hidden danger data group onto the corresponding abnormal result image, it is convenient for relevant staff to intuitively view the risk degree faced by the associated hidden danger data group, so as to improve the attention degree of relevant staff to the associated hidden danger data group when relevant staff browse the hidden danger investigation feedback data.

[0032] In a possible implementation manner, the method further includes:

[0033] Integrate the hidden danger investigation feedback data fed back within a preset time period, and identify the target hidden danger points and target hidden danger degrees corresponding to each hidden danger investigation feedback data;

[0034] Based on each target hidden danger degree and a preset influence area mapping relationship, determine the target influence area corresponding to each target hidden danger degree, and determine each target based on each target influence area;

[0035] The target hidden danger influence area corresponding to the hidden danger point;

[0036] Determine an influence intersection area based on each target hidden danger influence area, where the influence intersection area is the intersection area of at least two target hidden danger influence areas;

[0037] Determine a hidden danger surrounding area based on each target hidden danger point, and determine a focus investigation area based on the hidden danger surrounding area and the intersection area, where the focus investigation area is the area where the hidden danger surrounding area and the intersection area have an intersection;

[0038] Adjust the investigation strategy corresponding to the focus investigation area, where the investigation strategy includes the investigation frequency and the number of on-site investigation personnel.

[0039] By adopting the above technical solution, by identifying the target potential hazard points and target potential hazard levels corresponding to each potential hazard investigation feedback data, the vague potential hazard descriptions are transformed into specific and quantifiable information, which is convenient for relevant investigation personnel to clearly understand the specific locations and severity levels of each potential hazard, expand the influence of potential hazards from points to surfaces, facilitate a more comprehensive assessment of the possible affected ranges of potential hazards, facilitate the discovery of potential associations and interactions between different potential hazards by determining the intersection areas of influence, determine the areas of concern for investigation based on the potential hazard surrounding areas and intersection areas, facilitate the accurate positioning of the areas that need to be focused on, and by concentrating the investigation resources on the areas of concern for investigation, improve the resource utilization efficiency, avoid excessive investigation of low-risk areas, and achieve the optimal allocation of investigation resources.

[0040] In a possible implementation manner, the training process of the online recognition model includes:

[0041] Obtain a plurality of multi-modal sample data, and based on the data source types corresponding to each multi-modal sample data, perform modal splitting on all the multi-modal sample data, and perform normalization processing on each multi-modal sample data;

[0042] Fuse all the normalized multi-modal sample data to obtain a joint display space, and extract multi-source sample potential hazard features and multi-source sample annotations corresponding to each multi-source sample potential hazard feature from the joint display space;

[0043] Perform initial training on a preset recognition model based on the extracted multi-source sample potential hazard features and corresponding multi-source sample annotations to obtain an initial training result;

[0044] Update the parameters of the initial training result based on a model fine-tuning algorithm to obtain a target training result, and determine a loss value based on a preset loss function and the target training result. When the loss value is lower than a preset loss threshold, determine that the online recognition model is completed with training.

[0045] By adopting the above technical solution, in the embodiments of the present application, through modal splitting and joint display space construction, it is convenient for information complementarity between multi-modal sample data. In addition, based on the initial training result, the parameters are dynamically updated through a model fine-tuning algorithm. While retaining the general feature extraction ability, it is convenient to strengthen the recognition ability of potential hazard features unique to the chemical industry, thereby improving the model recognition accuracy.

[0046] In a second aspect, the present application provides an investigation system, adopting the following technical solution:

[0047] An investigation system, the investigation system includes:

[0048] At least one processor;

[0049] Memory;

[0050] At least one application program, wherein the at least one application program is stored in the memory and configured to be executed by at least one processor, and the at least one application program is configured to: execute the above-mentioned chemical enterprise safety hazard investigation method.

[0051] In a third aspect, the present application provides a computer-readable storage medium, adopting the following technical solution:

[0052] A computer-readable storage medium, comprising: a computer program stored therein that can be loaded and executed by a processor to execute the above-mentioned chemical enterprise safety hazard investigation method.

[0053] In a fourth aspect, the present application provides a computer program product, adopting the following technical solution:

[0054] A computer program product, comprising a computer program, wherein when the computer program is executed by a processor, it implements the above-mentioned chemical enterprise safety hazard investigation method.

[0055] In summary, the present application includes at least one of the following beneficial technical effects:

[0056] By comparing the online recognition features of the images to be investigated with these known hazard features, it is convenient to quickly screen out the possible hazard types in the area to be investigated. Then, combined with the hazard features selected by on-site staff for targeted observation and comparison, it is convenient to discover more potential hazards and avoid omissions. By combining the results of dual feature comparisons to determine the final hazard investigation feedback data, it is convenient to analyze and identify hazards from different angles, thereby reducing the deviation that may occur due to single comparison, and further improving the accuracy of hazard investigation results.

[0057] By identifying the target hazard points and target hazard levels corresponding to each hazard investigation feedback data, converting the fuzzy hazard description into specific and quantifiable information, it is convenient for relevant investigation personnel to clearly understand the specific locations and severity levels of each hazard, expanding the impact of hazards from points to areas, facilitating a more comprehensive assessment of the possible affected range of hazards. By determining the intersection area of impacts, it is convenient to discover the potential associations and interactions between different hazards. Based on the hazard surrounding area and the intersection area, determining the area of concern for investigation is convenient for accurately locating the areas that need to be focused on. By concentrating the investigation resources on the area of concern for investigation, it is convenient to improve the resource utilization efficiency, avoid excessive investigation of low-risk areas, and achieve the optimal allocation of investigation resources. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Figure 1 is a flowchart of a chemical enterprise safety hazard investigation method in an embodiment of the present application;

[0059] Figure 2 It is a schematic flowchart of a process for adjusting a troubleshooting strategy in an embodiment of the present application;

[0060] Figure 3 It is a schematic structural diagram of a troubleshooting system in an embodiment of the present application. Detailed implementation manners

[0061] The following will further elaborate on the present application in conjunction with the attached Figures 1 to 3 for a more detailed description.

[0062] After reading this specification, those skilled in the art can make modifications to this embodiment without creative contributions as needed, but as long as it is within the scope of the claims of the present application, it is protected by the patent law.

[0063] To make the objectives, technical solutions, and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are some, but not all, of the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts fall within the scope of protection of the present application.

[0064] It should be noted that in the alternative embodiments of the present application, for relevant data such as object information, when the embodiments in the present application are applied to specific products or technologies, object permission or consent needs to be obtained, and the collection, use, and processing of relevant data need to comply with relevant laws, regulations, and standards of relevant countries and regions. That is to say, if the embodiments in the present application involve data related to objects, it needs to be obtained under the authorization and consent of the objects, the authorization and consent of relevant departments, and in compliance with relevant laws, regulations, and standards of relevant countries and regions. If personal information is involved in the embodiments, the acquisition of all personal information needs to obtain the consent of the individual. If sensitive information is involved, the separate consent of the information subject needs to be obtained, and the embodiments also need to be implemented under the authorization and consent of the objects.

[0065] Specifically, the embodiments of the present application provide a method for troubleshooting potential safety hazards in chemical enterprises, which is executed by a troubleshooting system. The troubleshooting system can be a server or a terminal device. Among them, the server can be an independent physical server, a server cluster or a distributed system composed of multiple physical servers, or a cloud server providing cloud computing services. The terminal device can be a smart phone, a tablet computer, a laptop computer, a desktop computer, etc., but is not limited thereto. The terminal device and the server can be directly or indirectly connected through wired or wireless communication methods, and the embodiments of the present application do not limit this here.

[0066] Refer to Figure 1, Figure 1 It is a schematic flowchart of a method for investigating potential safety hazards in a chemical enterprise in an embodiment of the present application. The method includes steps S110 - S160, where:

[0067] Step S110: Obtain the image to be investigated, perform feature recognition on the image to be investigated according to the trained online recognition model, and obtain the online recognition features.

[0068] Specifically, the image to be investigated is an image of the area to be investigated, which can be collected by an image acquisition device set in the area to be investigated and then uploaded to the investigation system. The online recognition model can be used to identify each online recognition feature included in the image to be investigated. The online recognition features include but are not limited to the appearance features of the equipment, such as cracks, corrosion, or deformation in the appearance of the equipment; the vibration features of the equipment, such as the vibration mode and vibration frequency of the equipment during processing or production; the temperature features of the equipment, such as abnormal temperatures of the equipment during processing or production; the leakage detection features, such as signs of liquid or gas leakage at the joints of the equipment or pipelines, etc.; the operating state features of the equipment, such as the sound features or pressure state of the equipment during operation. The specific online recognition features are not specifically limited in the embodiment of the present application and can be uploaded to the investigation system in advance by relevant staff. Since the appearance of the equipment, the vibration of the equipment, the temperature of the equipment, the leakage of liquid or gas, and the operating state of the equipment are the core carriers for ensuring the stable operation of the equipment process, which can directly determine the safety and reliability of the equipment during production or processing, therefore, potential safety hazards of the equipment can be investigated by identifying and analyzing the online recognition features. Further, in order to improve the accuracy of model recognition, the embodiment of the present application provides a training process of the online recognition model, which specifically may include:

[0069] Obtain a plurality of multi-modal sample data, and based on the data source type corresponding to each multi-modal sample data, perform modal splitting on all the multi-modal sample data, and perform normalization processing on each multi-modal sample data; fuse all the normalized multi-modal sample data to obtain a joint display space, and extract multi-source sample potential hazard features and the multi-source sample annotations corresponding to each multi-source sample potential hazard feature from the joint display space; perform initial training on a preset recognition model based on the extracted multi-source sample potential hazard features and the corresponding multi-source sample annotations to obtain an initial training result; perform parameter update on the initial training result based on a model fine-tuning algorithm to obtain a target training result, and determine a loss value based on a preset loss function and the target training result. When the loss value is lower than a preset loss threshold, it is determined that the online recognition model is trained.

[0070] Specifically, the multi-modal sample data can be images, videos, texts, etc. that contain different sample hazard characteristics. It can be sorted out by relevant staff according to historical investigation information and then uploaded to the investigation system. The specific content of the multi-modal sample data is not specifically limited in the embodiments of this application. After obtaining multiple multi-modal sample data, data preprocessing can be performed on the multiple multi-modal sample data first to facilitate cleaning and screening of valid data from the multiple multi-modal sample data, and eliminating noise or irrelevant data, so as to improve the effectiveness of the multi-modal sample data in the model training process. The specific data preprocessing method is not specifically limited in the embodiments of this application.

[0071] Since the data sources corresponding to different multi-modal sample data may be different. For example, the data source may be an image, a video or a text. At this time, all the multi-modal sample data can be split by modality according to the data type of each multi-modal sample data. That is, the multi-modal sample data with the data source of video will be sent to a preset video encoder for processing; the multi-modal sample data with the data source of image will be sent to a preset image encoder for processing; the multi-modal sample data with the data source of text will be sent to a preset text encoder for processing. Through modality splitting, the multi-modal sample data belonging to different data sources can be converted into a format suitable for model training. Among them, the specific preset video encoder, preset image encoder and preset text encoder are not specifically limited in the embodiments of this application and can be determined by relevant staff according to historical experimental data and then uploaded to the investigation system. After modality splitting, normalization processing can be performed on each multi-modal sample data according to the training requirements, where the training requirements are adapted to the main application fields of the online recognition model. That is, in the embodiments of this application, the training requirements are the training requirements for chemical equipment. After normalizing each multi-modal sample data according to the training requirements, the adaptability between the multi-modal sample data and the industry requirements can be improved. That is, it is convenient for the online recognition model to recognize and understand the professional terms or expression methods of different data sources. Among them, the specific training requirements can be uploaded by relevant staff according to the actual needs of the area to be investigated and are not specifically limited in the embodiments of this application.

[0072] After the multi-modal sample data are processed by their respective encoders, the multi-modal sample data carry their respective unique information after being processed by their respective encoders. However, this information is often scattered and heterogeneous. Therefore, through a deep learning network and a preset weighted fusion strategy, these different multi-modal sample data can be fused to obtain a joint display space, so as to integrate and uniformly represent these scattered and heterogeneous multi-modal sample data, enabling the features from different data sources to be correlated and complementary to each other in the same space, forming a set that comprehensively reflects the hidden danger features of multi-source samples in the chemical production environment. The construction process of the joint display space is not specifically limited in the embodiments of the present application, as long as it can ensure that the hidden danger features of the multi-source samples after fusion are all mapped into the joint display space. In the joint display space, extract the hidden danger features of the multi-source samples and perform annotation. The annotation of the multi-source samples is usually provided by relevant staff according to historical investigation information. Then, based on the extracted hidden danger features of the multi-source samples and the corresponding annotations of the multi-source samples, the initial training of the model is carried out. During the training process, the parameters of the initial training result are updated based on a preset model fine-tuning algorithm to obtain the target training result, where the preset model fine-tuning algorithm can be a grid search algorithm, and the specific algorithm is not specifically limited in the embodiments of the present application. Finally, calculate the loss value corresponding to the target training result based on a preset loss function. When the loss value is not lower than the preset loss threshold, it is determined that the online recognition model still needs to continue to perform cyclic training until the loss value is lower than the preset loss threshold. Among them, the specific preset loss value is not specifically limited in the embodiments of the present application. Based on the initial training result, the parameters are dynamically updated through the model fine-tuning algorithm, which not only retains the general feature extraction ability but also facilitates strengthening the recognition ability of the hidden danger features unique to the chemical industry, thereby improving the model recognition accuracy.

[0073] Step S120: Compare the online recognition features with a preset hidden danger feature form to obtain first recognition data. The preset hidden danger feature form contains multiple known hidden danger features, and the first recognition data are the known hidden danger features whose feature similarity with the online recognition features is higher than the first preset threshold.

[0074] Specifically, the preset hidden danger feature form contains multiple known hidden danger features, which can be determined by relevant staff based on historical inspection information and then uploaded to the inspection system. Hidden danger features similar to the online recognition features can be identified from the preset hidden danger feature form according to the preset similarity algorithm. Among them, the preset similarity algorithm can be the cosine similarity algorithm, the Jaccard coefficient, etc. The specific similarity algorithm is not specifically limited in the embodiments of the present application, as long as it can calculate the feature similarity between the online recognition features and each known hidden danger feature in the preset hidden danger feature form. Finally, all known hidden danger features are screened according to the first preset threshold, and the known hidden danger features with feature similarity higher than the first preset threshold are selected from all known hidden danger features and determined as the first recognition data. The specific first preset threshold is not specifically limited in the embodiments of the present application.

[0075] Step S130: Obtain the on-site environmental parameters corresponding to the area to be inspected, and determine the on-site selected hidden danger features based on the on-site environmental parameters.

[0076] Specifically, the on-site environmental parameters can be collected by sensors set in the area to be inspected and uploaded to the inspection system. The on-site environmental parameters include, but are not limited to, temperature, humidity, air pressure, gas concentration, current and voltage, etc. The possible hidden danger situations are different under different on-site environmental parameters. For example, high temperature may cause equipment overheating, low temperature may cause equipment embrittlement, high humidity may cause electrical short circuits or metal corrosion, and low humidity may increase the risk of static electricity. Therefore, relevant staff in the area to be inspected can predict the possible on-site selected hidden danger features based on historical inspection experience on the basis of the on-site environmental parameters. When determining the on-site selected hidden danger features, the collected on-site environmental parameters can be fed back to the terminal devices of relevant staff, so that relevant staff can make reasonable inferences according to the actual situation. Relevant staff can search by entering keywords themselves. After the inspection system obtains the keywords entered by relevant staff, it can feedback a hidden danger list based on this for relevant staff to select on-site hidden danger features from the feedback hidden danger list. It can also be that the inspection system simulates the possible production or processing in the future period according to the on-site environmental parameters, generates a feedback hidden danger list according to the simulation results, and then relevant staff select on-site hidden danger features from the feedback hidden danger list. Specifically, it can include:

[0077] Construct a simulation production model based on on-site environmental parameters, and drive the simulation generation model based on production plan data to conduct simulation production; record each simulated hidden danger feature that appears during the simulation production process, and the corresponding simulated hidden danger location of each simulated hidden danger feature, and generate a simulated hidden danger list based on the simulated hidden danger features; determine the concerned simulated hidden danger locations according to each simulated hidden danger feature and the corresponding simulated hidden danger location of each simulated hidden danger feature, and the simulated hidden danger feature value corresponding to the concerned simulated hidden danger feature corresponding to the concerned simulated hidden danger location is higher than the preset simulation feature threshold; determine the hidden danger positioning area based on the concerned simulated hidden danger location and the corresponding simulated hidden danger feature value, and select concerned on-site staff from within the hidden danger positioning area; feedback the simulated hidden danger list to the concerned on-site staff to remind the concerned on-site staff to select from the simulated hidden danger list and feedback the selection result; determine the on-site selected hidden danger features according to the selection feedback result of the concerned on-site staff.

[0078] Specifically, a simulation production model corresponding to the actual production process can be constructed based on historical production data and on-site environmental parameters. The simulation generation model can also be verified by comparing the actual production data with the simulation results according to a preset cross-validation algorithm. The actual production or processing process of the equipment in the area to be investigated can be simulated through the simulation production model. The construction method and verification method of the simulation production model are not specifically limited in the embodiments of the present application. The production plan data is the data that needs to be produced in a future period of time. The production plan data can be the order quantity, production schedule, etc., and can be determined and uploaded to the investigation system by relevant staff according to the actual production situation. Input the production plan data into the simulation generation model to drive the simulation production model to enter the simulation production process, and real-time locate each simulated hidden danger feature, its simulated hidden danger location and timestamp that appear during the simulation production process. Locate the concerned simulated hidden danger locations during the simulation production process according to the preset simulation feature threshold. The simulated hidden danger feature value corresponding to the concerned simulated hidden danger feature corresponding to the concerned simulated hidden danger location is higher than the preset simulation feature threshold. Specifically, the preset simulation feature threshold is not specifically limited in the embodiments of the present application. The number of concerned simulated hidden danger locations included in the area to be investigated can be 1, 2, or multiple. The specific number is not specifically limited in the embodiments of the present application. Among them, the historical abnormal probability and severity of each simulated hidden danger feature can be obtained according to the historical abnormal records, and each simulated hidden danger feature can be quantitatively scored based on this to obtain the simulated hidden danger feature value corresponding to each simulated hidden danger feature. Specifically, the quantitative scoring standard is not specifically limited in the embodiments of the present application and can be determined by relevant staff according to historical experimental data and then uploaded to the investigation system.

[0079] When determining the potential hazard location area based on the simulated potential hazard locations of concern and the corresponding simulated potential hazard eigenvalue, first, according to the preset mapping relationship of the location range, the location range corresponding to each simulated potential hazard location of concern can be determined. The preset mapping relationship of the location range is the corresponding relationship between the simulated potential hazard eigenvalue and the location range, and the larger the simulated potential hazard eigenvalue, the larger the corresponding location range. After determining the location range corresponding to each potential hazard location of concern, the location observation area corresponding to each potential hazard location of concern can be determined based on the potential hazard location of concern and the corresponding location range. When there is only one simulated potential hazard location of concern in the area to be checked, the location observation area corresponding to this potential hazard location of concern can be directly determined as the potential hazard location area. When there are at least two simulated potential hazard locations of concern in the area to be checked, the overlapping location area of the location observation areas corresponding to the at least two simulated potential hazard locations of concern can be determined as the potential hazard location area. If there is no overlapping location area between the location observation areas corresponding to the at least two simulated potential hazard locations of concern, the location observation area with the largest location range is directly determined as the potential hazard location area. According to the image to be checked, identify the staff in the potential hazard location area, and determine any staff in the potential hazard location area as the on-site staff of concern. Since this on-site staff of concern is relatively close to the area with a more serious potential hazard situation, by feeding back the simulated potential hazard list to this on-site staff of concern, it is convenient to improve the adaptability between the potential hazards selected on-site and the actual potential hazard situation.

[0080] Step S140: Obtain an observation image from the area to be checked based on the on-site selected potential hazard features, and perform feature recognition on the observation image according to the trained online recognition model to obtain an observation recognition feature.

[0081] Specifically, the observation image is a partial area image containing the on-site selected potential hazard features. Compared with the image to be checked, the observation image is more targeted. For example, the image to be checked may be a panoramic image of the area to be checked, and the image to be checked contains various devices in the area to be checked, while the observation image may only contain a certain heating device in the area to be checked. The observation image is determined based on the on-site selected potential hazard features determined by the relevant staff to assist the inspection system under the current environmental parameters. Different on-site selected potential hazard features correspond to different observation images. For example, when the on-site selected potential hazard feature is "heating", the corresponding observation image contains at least the device with a heating potential hazard. The specific implementation process of performing feature recognition on the observation image according to the trained online recognition model to obtain the observation recognition feature can refer to the specific implementation process of performing feature recognition on the image to be checked according to the trained online recognition model to obtain the online recognition feature in the above embodiment, which will not be elaborated here. The observation recognition feature also includes but is not limited to device appearance features, device vibration features, leakage detection features, device operating state features, etc. The specific observation recognition feature is not specifically limited in the embodiments of the present application.

[0082] Step S150: Compare the online recognition features with the observed recognition features to obtain second recognition data, where the second recognition data are the observed recognition features with a feature similarity higher than a second preset threshold with respect to the online recognition features.

[0083] Specifically, compare the online recognition features with the observed recognition features in the same way as obtaining the first recognition data in the above embodiments, which will not be elaborated here. Since the determination process of the online recognition features refers to the subjective judgment of relevant staff, the determination conditions of the second recognition data can be appropriately adjusted, that is, the second preset threshold can be set higher than the first preset threshold. The specific second preset threshold can be determined by relevant staff according to historical experimental data and is not specifically limited in the embodiments of the present application as long as it can ensure that the second preset threshold is higher than the first preset threshold.

[0084] Step S160: Generate hidden danger investigation feedback data based on the first recognition data and the second recognition data, and feedback the hidden danger investigation feedback data.

[0085] Specifically, determine the final hidden danger investigation feedback data by combining the dual feature comparison results, which is convenient for analyzing and identifying hidden dangers from different angles. The hidden danger investigation feedback data is the integration result of the first recognition data and the second recognition data. After generating the hidden danger investigation feedback data, the hidden danger investigation feedback data can be fed back to relevant investigation personnel, so that relevant investigation personnel can take corresponding preventive measures in time to avoid the continuous deterioration of hidden danger features.

[0086] For the embodiments of the present application, by comparing the online recognition features of the image to be investigated with these known hidden danger features, it is convenient to quickly screen out the possible types of hidden dangers in the area to be investigated. Then, combined with the hidden danger features selected by on-site staff for targeted observation and comparison, it is convenient to discover more potential hidden dangers and avoid omission. By combining the dual feature comparison results to determine the final hidden danger investigation feedback data, it is convenient to analyze and identify hidden dangers from different angles, thereby reducing the deviation that may occur due to single comparison and improving the accuracy of the hidden danger investigation results.

[0087] Further, in order to enhance the intuitiveness of the feedback result, in the method provided by the embodiments of the present application, when generating the hidden danger investigation feedback data based on the first recognition data and the second recognition data, it may specifically include:

[0088] Generate a potential hazard investigation list based on the first identification data and the second identification data, and determine an associated hazard data group from the first identification data and the second identification data based on a preset hazard association graph; determine the hazard frequency, abnormal result image, and hazard source corresponding to the associated hazard data group based on historical hazard records, and superimpose the hazard frequency and hazard source onto the abnormal result image to obtain a parameter integration image; superimpose the parameter integration image corresponding to the associated hazard data group onto the potential hazard investigation list to obtain potential hazard investigation feedback data.

[0089] Specifically, since the first identification data and the second identification data are determined based on different preset thresholds, before obtaining the potential hazard investigation list, the first identification data and the second identification data can be normalized first to obtain the quantified hazard feature values of each known hazard feature in the first identification data and the quantified hazard feature values of each on-site selected hazard feature in the second identification data, so as to eliminate the dimensional difference between the first identification data and the second identification data. After sorting the quantified hazard feature values based on the quantified hazard feature values of each known hazard feature and the quantified hazard feature values of each on-site selected hazard feature, a potential hazard investigation list is obtained.

[0090] The preset hazard association graph stores the influence relationships between various hazard features. Among them, the influence relationships include but are not limited to causal relationships, concurrent relationships, spatial proximity relationships, etc. The specific influence relationships can be determined by relevant staff based on historical experimental data and then uploaded to the investigation system. The edges in the preset hazard association graph are used to represent the degree of association and are represented by 0-1 quantification values. The preset vector similarity algorithm can be used to calculate the vector similarity between each node, and the nodes with vector similarity higher than the preset vector threshold are determined as the associated hazard data group. Among them, the specific preset vector threshold is not specifically limited in the embodiments of the present application and can be set by relevant staff according to historical experimental data.

[0091] After determining the associated hidden danger data group, according to the known hidden danger features in the associated hidden danger data group and the hidden danger features selected on site, retrieve the same type of hidden danger records within the investigable time period from the historical hidden danger records, and extract the hidden danger frequency, that is, the number of occurrences per unit time; the abnormal result image, that is, the historical defect image corresponding to the actual abnormality; and the hidden danger source, that is, the direct cause of the hidden danger. Through the preset AR algorithm, the hidden danger frequency and the hidden danger source can be superimposed on the abnormal result image to obtain a parameter integration image. Among them, a dynamic heat map layer can be superimposed on the abnormal result image, and the color depth of the layer can be dynamically set according to the hidden danger frequency. For example, when the hidden danger frequency > 5 times / month, a red layer color is used, and when the hidden danger frequency is 2 - 5 times / month, a yellow layer color is used. The specific correspondence between the hidden danger frequency and the layer color can be determined by relevant staff according to historical experimental data. The hidden danger source can be superimposed on the abnormal result image for display by adding text labels to the edge area of the abnormal result image. The specific display form is not specifically limited in the embodiments of this application.

[0092] By superimposing the hidden danger frequency and the hidden danger source of the associated hidden danger data group onto the corresponding abnormal result image, it is convenient for relevant staff to directly view the risk level faced by the associated hidden danger data group, thereby facilitating the improvement of the attention of relevant staff to the associated hidden danger data group when they browse the hidden danger investigation feedback data.

[0093] Furthermore, in order to improve the integrity and accuracy of the captured images, the method provided in the embodiments of this application, when obtaining the observation images from the area to be investigated based on the hidden danger features selected on site, may specifically include:

[0094] Based on the mapping relationship between the hidden danger features selected on site and the preset shooting strategy, determine the target shooting strategy corresponding to the hidden danger features selected on site; based on the simulated hidden danger position corresponding to the hidden danger features selected on site and the attention to the simulated hidden danger position, determine the target shooting route; generate a target shooting instruction based on the target shooting strategy and the target shooting route, and control the shooting device to shoot according to the target shooting instruction to obtain the observation images from the area to be investigated.

[0095] Specifically, different on-site selection hazard characteristics are suitable for different target shooting strategies. Based on the preset shooting strategy mapping relationship, the target shooting strategy corresponding to any on-site selection hazard characteristic can be determined. Since the observation image is a targeted shot of the equipment with on-site selection hazard characteristics, it is necessary to reasonably formulate the target shooting strategy to improve the effectiveness of the observation image. Specifically, the preset shooting strategy mapping relationship can be determined by relevant staff according to historical experimental data and then uploaded to the troubleshooting system. For example, when the on-site selection hazard characteristic is a pipeline leakage hazard characteristic, the corresponding shooting strategy may focus on collecting microscopic images of the area around the leakage point, the diffusion range of the leaked substance, and the equipment connection points to capture the shape, scale, and potential source of the leakage; when the on-site selection hazard characteristic is an equipment surface corrosion hazard characteristic, the corresponding shooting strategy may focus on multi-angle and high-resolution surface imaging to clearly present the corrosion texture, depth, and distribution.

[0096] The on-site selection hazard characteristics are predicted by relevant staff in the area to be troubleshot based on historical troubleshooting experience and on-site environmental parameters, while the concerned simulated hazard characteristics are predicted and simulated by the troubleshooting system for the possible production situations in the future. There may be differences between the two. Therefore, the target shooting route can be determined based on both the simulated hazard location corresponding to the on-site selection hazard characteristic and the concerned simulated hazard location corresponding to the concerned simulated hazard characteristic. The method for determining the target shooting route is not specifically limited in the embodiments of the present application, as long as it can ensure that when shooting is carried out according to the target shooting route, the simulated hazard location corresponding to the on-site selection hazard characteristic and the concerned simulated hazard location corresponding to the concerned simulated hazard characteristic can be shot together. The shooting device can be a shooting robot or a drone. The specific shooting device is not specifically limited in the embodiments of the present application, as long as it can ensure that the shooting device can shoot according to the target shooting instruction. By allocating the most suitable target shooting strategy for different on-site selection hazard characteristics, the integrity and accuracy of the captured images are improved, and the loss of key information due to improper shooting is avoided, thereby facilitating a significant improvement in the accuracy of hazard identification.

[0097] Furthermore, to facilitate improving the resource utilization efficiency and avoid over-troubleshooting of low-risk areas, the method provided in the embodiments of the present application further includes steps S210 - S250, as Figure 2 shown, where:

[0098] Step S210: Integrate the hazard troubleshooting feedback data fed back within the preset time period, and identify the target hazard points and target hazard levels corresponding to each hazard troubleshooting feedback data.

[0099] Specifically, the preset time period is a period of time before the current moment. The duration corresponding to the preset time period can be 7 days or 5 days. The specific duration is not specifically limited in the embodiments of the present application and can be set by relevant staff according to actual investigation requirements. The target potential hazard point is the potential hazard location where relevant maintenance staff actually respond and take relevant measures according to the potential hazard investigation feedback data. The target potential hazard degree is the actual severity of the target potential hazard point, which can be actually measured and uploaded to the investigation system by relevant maintenance personnel during the actual response and taking relevant measures.

[0100] Step S220: Based on the mapping relationship between each target potential hazard degree and the preset influence area, determine the target influence area corresponding to each target potential hazard degree, and based on each target influence area, determine the target potential hazard influence area corresponding to each target potential hazard point.

[0101] Step S230: Determine the influence intersection area based on each target potential hazard influence area. The influence intersection area is the intersection area of at least two target potential hazard influence areas.

[0102] Specifically, the target influence areas corresponding to different target potential hazard degrees are different. The higher the target potential hazard degree, the larger the corresponding target influence area. Based on the preset influence area mapping relationship, the target influence area corresponding to any target potential hazard degree can be determined. The specific preset influence area mapping relationship can be determined by relevant staff according to historical data and then uploaded to the investigation system. The target potential hazard influence area is a circular area with the target potential hazard point as the center and an area consistent with the target influence area. By using the above method, the target potential hazard influence area corresponding to each target potential hazard point can be obtained. By performing an intersection process on the target potential hazard influence areas corresponding to all target potential hazard points, the influence intersection area can be obtained.

[0103] Step S240: Determine the potential hazard surrounding area based on each target potential hazard point. Based on the potential hazard surrounding area and the intersection area, determine the area to be focused on for investigation. The area to be focused on for investigation is the area where the potential hazard surrounding area and the intersection area have an intersection.

[0104] Specifically, import each target potential hazard point into a preset coordinate system to obtain the coordinate of the potential hazard point corresponding to each target potential hazard point, and then connect each target potential hazard point according to the coordinate of each potential hazard point to obtain the potential hazard surrounding area. After determining the potential hazard surrounding areas and intersection areas corresponding to all target potential hazard points, the area where the potential hazard surrounding area and the intersection area have an intersection can be determined as the area to be focused on for investigation. The number of areas to be focused on for investigation may be 1 or multiple. The specific number is not specifically limited in the embodiments of the present application.

[0105] Step S250: Adjust the investigation strategy corresponding to the area to be focused on for investigation. The investigation strategy includes the investigation frequency and the number of on-site investigation personnel.

[0106] Specifically, after determining the area to be focused on for investigation, the investigation strategy corresponding to the area to be focused on can be adjusted by increasing the investigation frequency and the number of on-site investigation personnel. The specific adjustment method is not specifically limited in the embodiments of the present application, as long as the investigation frequency corresponding to the area to be focused on is higher than that of other non-focused areas, or the number of on-site investigation personnel corresponding to the area to be focused on is higher than that of other non-focused areas.

[0107] In the embodiments of the present application, by identifying the target potential hazard points and target potential hazard levels corresponding to each potential hazard investigation feedback data, the vague potential hazard descriptions are converted into specific and quantifiable information, which facilitates relevant investigation personnel to clearly understand the specific locations and severity levels of various potential hazards, expands the influence of potential hazards from points to surfaces, facilitates a more comprehensive assessment of the possible affected ranges of potential hazards, facilitates the discovery of potential associations and interactions between different potential hazards by determining the influence intersection area, determines the area to be focused on for investigation based on the potential hazard surrounding area and the intersection area, facilitates the accurate positioning of the areas that need to be focused on, and by concentrating the investigation resources on the areas to be focused on for investigation, it is conducive to improving the resource utilization efficiency, avoiding excessive investigation of low-risk areas, and realizing the optimal allocation of investigation resources.

[0108] An investigation system is provided in the embodiments of the present application, as Figure 3 shown. Figure 3 The investigation system 300 shown includes: a processor 301 and a memory 303. Among them, the processor 301 and the memory 303 are connected, such as connected through a bus 302. Optionally, the investigation system 300 may further include a transceiver 304. It should be noted that in practical applications, the transceiver 304 is not limited to one, and the structure of the investigation system 300 does not constitute a limitation to the embodiments of the present application.

[0109] The processor 301 may be a CPU (Central Processing Unit), a general-purpose processor, a DSP (Digital Signal Processor), an ASIC (Application Specific Integrated Circuit), an FPGA (Field Programmable Gate Array), or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. It can implement or execute various exemplary logic blocks, modules, and circuits described in combination with the disclosure of the present application. The processor 301 may also be a combination that realizes computing functions, such as a combination including one or more microprocessors, a combination of a DSP and a microprocessor, etc.

[0110] The bus 302 may include a path for transmitting information between the above components. The bus 302 can be a PCI (Peripheral Component Interconnect) bus, an EISA (Extended Industry Standard Architecture) bus, or the like. The bus 302 can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 3 only one line is shown in Figure 3 , but it does not mean that there is only one bus or one type of bus.

[0111] The memory 303 can be a ROM (Read Only Memory) or other types of static storage devices that can store static information and instructions, a RAM (Random Access Memory) or other types of dynamic storage devices that can store information and instructions, or it can also be an EEPROM (Electrically Erasable Programmable Read Only Memory), a CD-ROM (Compact Disc Read Only Memory), or other optical disc storage, optical disc storage (including compact discs, laser discs, optical discs, digital versatile discs, Blu-ray discs, etc.), magnetic disk storage media, or other magnetic storage devices, or any other medium that can be used to carry or store the desired program code in the form of instructions or data structures and can be accessed by a computer, but is not limited thereto.

[0112] The memory 303 is used to store the application program code for executing the solution of this application, and is controlled by the processor 301 for execution. The processor 301 is used to execute the application program code stored in the memory 303 to implement the content shown in the foregoing method embodiments.

[0113] Among them, the troubleshooting system includes but is not limited to: mobile terminals such as mobile phones, laptop computers, digital broadcast receivers, PDAs (Personal Digital Assistants), PADs (Tablet Computers), PMPs (Portable Multimedia Players), in-vehicle terminals (such as in-vehicle navigation terminals), etc., and fixed terminals such as digital TVs, desktop computers, etc. It can also be a server, etc. Figure 3 The troubleshooting system shown is only an example and should not impose any limitations on the functions and usage scope of the embodiments of this application.

[0114] The embodiments of this application provide a computer-readable storage medium, on which a computer program is stored. When it runs on a computer, it enables the computer to execute the corresponding content in the foregoing method embodiments.

[0115] An embodiment of the present application provides a computer program product, which includes a computer program that, when executed by a processor, implements the method in any of the above embodiments.

[0116] It should be understood that although the steps in the flowchart of the accompanying drawings are shown in sequence according to the indication of the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise clearly stated in this specification, the execution of these steps is not strictly limited in order and can be executed in other orders. Moreover, at least a part of the steps in the flowchart of the accompanying drawings may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily executed at the same time, but can be executed at different times, and their execution order is not necessarily sequential, but can be executed alternately or in turn with at least a part of other steps or sub-steps or stages of other steps.

[0117] The above are only some embodiments of the present application. It should be noted that for those of ordinary skill in the art, without departing from the principle of the present application, several improvements and refinements can be made, and these improvements and refinements should also be regarded as the protection scope of the present application.

Claims

1. A method for investigating potential safety hazards in chemical enterprises, characterized in that, Including: Obtain the image to be checked, perform feature recognition on the image to be checked according to the trained online recognition model to obtain online recognition features, and the online recognition features include device appearance features, device vibration features, device temperature features, leakage detection features, and device operation status features; Compare the online recognition features with a preset hidden danger feature form to obtain first recognition data. The preset hidden danger feature form contains multiple known hidden danger features, and the first recognition data is the known hidden danger features whose feature similarity with the online recognition features is higher than a first preset threshold; Obtain the on-site environmental parameters corresponding to the area to be checked, and determine on-site selected hidden danger features based on the on-site environmental parameters. The on-site environmental parameters include temperature, humidity, air pressure, gas concentration, current and voltage; Obtain observation images from the area to be checked based on the on-site selected hidden danger features, and perform feature recognition on the observation images according to the trained online recognition model to obtain observation recognition features; Compare the online recognition features with the observation recognition features to obtain second recognition data. The second recognition data is the observation recognition features whose feature similarity with the online recognition features is higher than a second preset threshold; Generate hidden danger investigation feedback data based on the first recognition data and the second recognition data, and feedback the hidden danger investigation feedback data; Among them, the determining the on-site selected hidden danger features based on the on-site environmental parameters includes: Construct a simulation production model based on the on-site environmental parameters, and drive the simulation generation model to perform simulation production based on production plan data. The production plan data is the data that needs to be produced in a future period of time, and the production plan data is the order quantity and production schedule; Record each simulated hidden danger feature that appears during the simulation production process, and the corresponding simulated hidden danger location for each simulated hidden danger feature, and generate a simulated hidden danger list based on the simulated hidden danger features; According to each simulated hidden danger feature and the corresponding simulated hidden danger location, determine the concerned simulated hidden danger location. The simulated hidden danger feature value corresponding to the concerned simulated hidden danger feature corresponding to the concerned simulated hidden danger location is higher than a preset simulated feature threshold. Obtain the historical abnormal probability and severity of each simulated hidden danger feature based on historical abnormal records, and perform quantitative scoring on each simulated hidden danger feature based on this to obtain the simulated hidden danger feature value corresponding to each simulated hidden danger feature; According to the preset positioning range mapping relationship, determine the positioning range corresponding to each concerned simulated hidden danger location. The preset positioning range mapping relationship is the corresponding relationship between the simulated hidden danger feature value and the positioning range; Determine the positioning observation area corresponding to each concerned hidden danger location according to the concerned hidden danger location and the corresponding positioning range; When there is only one concerned simulated hidden danger location in the area to be checked, determine the positioning observation area corresponding to the concerned hidden danger location as the hidden danger positioning area; When there are at least two concerned simulated potential hazard locations in the area to be investigated, the overlapping location area of the location observation areas corresponding to the at least two concerned simulated potential hazard locations is determined as the potential hazard location area. If there is no overlapping location area between the location observation areas corresponding to the at least two concerned simulated potential hazard locations, the location observation area with the largest location range is determined as the potential hazard location area; According to the image to be investigated, identify the staff members within the potential hazard location area, and determine any one of the staff members within the potential hazard location area as the concerned on-site staff member; Feed back the simulated potential hazard list to the concerned on-site staff member to remind the concerned on-site staff member to select from the simulated potential hazard list and feedback the selection result; Determine the on-site selected potential hazard characteristics according to the selection feedback result of the concerned on-site staff member; Among them, obtaining the observation image from the area to be investigated based on the on-site selected potential hazard characteristics includes: Based on the mapping relationship between the on-site selected potential hazard characteristics and the preset shooting strategy, determine the target shooting strategy corresponding to the on-site selected potential hazard characteristics. Different on-site selected potential hazard characteristics are suitable for different target shooting strategies; Determine the target shooting route based on the simulated potential hazard location corresponding to the on-site selected potential hazard characteristics and the concerned simulated potential hazard locations; ​ 2. A method for investigating potential safety hazards in chemical enterprises according to claim 1, characterized in that, ​ ​ ​ ​ 3. The safety hazard inspection method for a chemical enterprise according to claim 1, characterized in that ​ ​ ​ ​ ​ ​ ​ 4. A method for investigating potential safety hazards in a chemical enterprise according to claim 1, characterized in that, The training process of the online recognition model includes: Obtaining multiple multimodal sample data, and based on the data source type corresponding to each multimodal sample data, splitting all the multimodal sample data by modality, and normalizing each multimodal sample data; Fusing all the normalized multimodal sample data to obtain a joint display space, and extracting multi-source sample hidden danger features and multi-source sample annotations corresponding to each multi-source sample hidden danger feature from the joint display space; Initial training of a preset recognition model based on the extracted multi-source sample hidden danger features and corresponding multi-source sample annotations to obtain an initial training result; Updating the parameters of the initial training result based on a model fine-tuning algorithm to obtain a target training result, and determining a loss value based on a preset loss function and the target training result. When the loss value is lower than a preset loss threshold, it is determined that the online recognition model has completed training.

5. A troubleshooting system, characterized in that, The troubleshooting system includes: At least one processor; A memory; At least one application program, where the at least one application program is stored in the memory and is configured to be executed by at least one processor. The at least one application program is configured to: execute a method for troubleshooting safety hazards in a chemical enterprise according to any one of claims 1-4.

6. A computer-readable storage medium, characterized in that, Including: A computer program stored that can be loaded and executed by a processor to execute a method for troubleshooting safety hazards in a chemical enterprise according to any one of claims 1-4.

7. A computer program product, characterized in that, Including a computer program, and when the computer program is executed by a processor, the steps of a method for troubleshooting safety hazards in a chemical enterprise according to any one of claims 1-4 are implemented.

Images