Chemical Industrial Park Safety Early Warning System Based on Multimodal Large Model

Abstract

The present invention discloses a chemical industrial park safety early warning system based on a multimodal large model, belonging to the technical field of chemical industrial parks, including: a park data collection module configured to collect multimodal data of the chemical industrial park; a data processing and analysis module configured to preprocess and analyze the multimodal data of the chemical industrial park to identify potential safety hazards in the chemical industrial park in advance; and a safety early warning management module configured to perform safety early warning management on the potential safety hazards in the chemical industrial park. The present invention solves the problems that the existing system cannot conduct real-time monitoring of chemical industrial parks, cannot identify potential safety hazards in chemical industrial parks in advance and perform safety early warning management, resulting in low accuracy and timeliness of safety early warning in chemical industrial parks. The present invention can conduct real-time monitoring of chemical industrial parks, can identify potential safety hazards in chemical industrial parks in advance and perform safety early warning management, effectively improving the accuracy and timeliness of safety early warning in chemical industrial parks and enhancing the safety of chemical industrial parks.

Description

Technical Field

[0001] The present invention relates to the technical field of chemical industrial parks, and particularly to a safety early warning system for chemical industrial parks based on a multimodal large model. Background Art

[0002] Chemical industrial parks are an important part of the development of the chemical industry. The products produced in chemical industrial parks are also closely related to people's daily lives. Multiple types of chemical enterprises are gathered in the chemical industrial park area, with concentrated hazard sources. There are mutual influences between hazard sources and between hazard sources and vulnerable targets. After an accident occurs, it is easy to cause serious consequences and chain reactions, and the safety problem is prominent.

[0003] Chinese Patent with Publication No. CN118735278B discloses a method for differential management of site pollution prevention and control in chemical industrial parks, including: determining the influencing factors for differential management of site pollution prevention and control in chemical industrial parks, and evaluating the contribution of each influencing factor to site pollution in chemical industrial parks; dimension reduction to identify the environmental characteristic index factors for differential management of site pollution prevention and control in chemical industrial parks; constructing a scoring system based on the quantification results of this factor in the chemical industrial park site and combining with the contribution rate, dividing grades, and assigning scores to each grade; based on the grade scoring results, combining with the investigation risks of the chemical industrial park site, etc., constructing a technical route for zoning the pollution prevention and control area for differential management; gradually judging the pollution risk, migration risk, potential risk, and surrounding risk of the enterprise site to determine the site classification of the enterprise site. However, this patent has the following defects:

[0004] The existing technologies cannot conduct real-time monitoring of chemical industrial parks, cannot identify potential safety hazards in chemical industrial parks in advance and conduct safety early warning management, resulting in low accuracy and timeliness of safety early warning in chemical industrial parks and reducing the safety of chemical industrial parks. Summary of the Invention

[0005] The purpose of the present invention is to provide a safety early warning system for chemical industrial parks based on a multimodal large model, which can conduct real-time monitoring of chemical industrial parks, can identify potential safety hazards in chemical industrial parks in advance and conduct safety early warning management, effectively improving the accuracy and timeliness of safety early warning in chemical industrial parks and enhancing the safety of chemical industrial parks, and solving the problems raised in the above background art.

[0006] To achieve the above purpose, the present invention provides the following technical solutions:

[0007] A safety early warning system for chemical industrial parks based on a multimodal large model, including:

[0008] A park data acquisition module configured to collect multimodal data of the chemical industrial park;

[0009] A data processing and analysis module, configured to preprocess and analyze the multi-modal data collected from a chemical industrial park, and identify potential safety hazards in the chemical industrial park in advance;

[0010] A safety early warning management module, configured to perform safety early warning management on potential safety hazards in the chemical industrial park.

[0011] Preferably, the multi-modal data collected from the chemical industrial park includes:

[0012] Based on intelligent collection devices, the temperature, humidity, pressure and gas concentration in the chemical industrial park are monitored in real time to collect the environmental data of the chemical industrial park;

[0013] Based on intelligent collection devices, the vibration, current and voltage of the equipment in the chemical industrial park are monitored in real time to collect the equipment data of the chemical industrial park;

[0014] Based on intelligent collection devices, the personnel location and helmet-wearing situation in the chemical industrial park are monitored in real time to collect the personnel data of the chemical industrial park;

[0015] Based on intelligent collection devices, the video, image and sound in the chemical industrial park are monitored in real time to collect the visual and auditory data of the chemical industrial park;

[0016] According to the environmental data, equipment data, personnel data, visual and auditory data of the chemical industrial park, the multi-modal data of the chemical industrial park is determined.

[0017] Preferably, the preprocessing of the multi-modal data collected from the chemical industrial park includes:

[0018] Clean the multi-modal data of the chemical industrial park to remove the noise and duplicate data that are useless for the safety early warning of the chemical industrial park;

[0019] Check the multi-modal data of the chemical industrial park to determine whether there are missing values and outliers, and process the missing values and outliers existing in the multi-modal data of the chemical industrial park;

[0020] Judge the missing values and outliers existing in the multi-modal data of the chemical industrial park to determine whether the missing values and outliers existing in the multi-modal data of the chemical industrial park are useful for the safety early warning of the chemical industrial park;

[0021] If it is useful, fill in the missing values and replace the outliers in the multi-modal data of the chemical industrial park. If it is useless, remove the missing values and outliers existing in the multi-modal data of the chemical industrial park.

[0022] Preferably, the preprocessing of the multi-modal data collected from the chemical industrial park further includes:

[0023] Normalize the multi-modal data of the chemical industrial park to convert the multi-modal data of the chemical industrial park into a unified data format, remove the dimensional differences in the multi-modal data of the chemical industrial park, and determine the standardized multi-modal data of the chemical industrial park;

[0024] Integrate the multi-modal data of the chemical industrial park, integrate the multi-modal data of the chemical industrial park into a unified data view, and perform data integrity verification on the integrated multi-modal data of the chemical industrial park. After the data integrity verification is qualified, securely store the integrated multi-modal data of the chemical industrial park;

[0025] Perform modal alignment and feature extraction on the multi-modal data of the chemical industrial park, extract the feature vectors useful for the safety warning of the chemical industrial park from the multi-modal data of the chemical industrial park, and perform weighted fusion on the extracted feature vectors to determine the feature data of the chemical industrial park.

[0026] Preferably, extracting the feature vectors useful for the safety warning of the chemical industrial park from the multi-modal data of the chemical industrial park includes:

[0027] Perform regional data division on the obtained multi-modal data of the chemical industrial park to obtain the sub-region data corresponding to each chemical sub-region of the current chemical industrial park;

[0028] Adopt the operation status evaluation index adapted to the chemical sub-region, and perform operation status evaluation based on the sub-region data to obtain the real-time operation status of the chemical sub-region;

[0029] Obtain the historical risk record data of each chemical sub-region within the preset time period;

[0030] Extract the historical risk categories, the corresponding number of risk occurrences, and the risk severity levels of each chemical sub-region from the historical risk record data, and input them into the pre-established risk assessment model to obtain the estimated risk status of the corresponding chemical sub-region;

[0031] When the real-time operation status of a chemical sub-region is in a dangerous state, extract all the feature vectors adapted to the current chemical sub-region being in a dangerous state and output them as key feature vectors;

[0032] When the real-time operation status of a chemical sub-region is not in a dangerous state, label the chemical sub-regions where the state level of the estimated risk status is the same as the real-time operation status as the first region, label the chemical sub-regions where the state level of the estimated risk status is greater than the state level of the real-time operation status as the second region, and label the chemical sub-regions where the state level of the estimated risk status is less than the state level of the real-time operation status as the third region;

[0033] Output all the feature vectors adapted to the real-time operation status of the current first region as the first key feature vectors;

[0034] Obtain the number of the first state levels where the corresponding estimated risk state of the second area exceeds the real-time operation state;

[0035] Respectively use the risk occurrence frequencies and the number of the first state levels of various historical risk categories existing in the second area as matching conditions to determine the enhanced feature vectors corresponding to various historical risk categories;

[0036] From all the feature vectors adapted to the real-time operation state of the second area, screen out the feature vectors associated with various historical risk categories existing in the second area, and after replacing them with the corresponding enhanced feature vectors, output them together with all other non-replaced feature vectors as the key feature vectors;

[0037] Obtain the number of the second state levels where the corresponding estimated risk state of the third area is less than the real-time operation state;

[0038] Respectively use the risk occurrence frequencies and the number of the second state levels of various historical risk categories existing in the third area as matching conditions to determine the mitigation feature vectors corresponding to various historical risk categories;

[0039] From all the feature vectors adapted to the real-time operation state of the third area, screen out the feature vectors associated with various historical risk categories existing in the third area, and after replacing them with the corresponding mitigation feature vectors, output them together with all other non-replaced feature vectors as the key feature vectors;

[0040] Summarize all the key feature vectors and output them as the feature vectors useful for the safety warning of the chemical industrial park.

[0041] Preferably, it further includes:

[0042] Real-time monitor the changes in the operation state and risk state of each chemical sub-area in the chemical industrial park, and correspondingly obtain the operation change data and risk change data;

[0043] According to the operation change data, determine the number of the first states of the chemical sub-areas where the real-time operation state remains in the dangerous state per unit time within the set sliding window of the chemical industrial park, and the number of the first state times when the real-time operation state changes from the warning state to the dangerous state;

[0044] Use the number of the first states and the number of the first state times to construct an operation danger matrix;

[0045] Determine the number of the second states of the chemical sub-areas where the real-time operation state remains in the safe state per unit time within the set sliding window of the chemical industrial park, and the number of the second state times when the real-time operation state changes from the warning state to the safe state;

[0046] Construct a running safety matrix using the number of second states and the number of times of the second state;

[0047] By mapping the running hazard matrix and the running safety matrix using the first preset adjustment analysis strategy, determine the first adjustment direction and the first adjustment amount for the preset time period;

[0048] According to the risk change data, determine the first risk number of chemical sub-regions in the chemical industrial park where the estimated risk state remains at a high-risk state per unit time within the set sliding window, and the first risk number of times when the estimated risk state changes from a medium-risk state to a high-risk state;

[0049] Construct a high-risk change matrix using the first risk number and the first risk number of times;

[0050] Determine the second risk number of chemical sub-regions in the chemical industrial park where the estimated risk state remains at a low-risk state per unit time within the set sliding window, and the second risk number of times when the estimated risk state changes from a medium-risk state to a low-risk state;

[0051] Construct a low-risk change matrix using the second risk number and the second risk number of times;

[0052] By mapping the low-risk change matrix and the high-risk change matrix using the second preset adjustment analysis strategy, determine the second adjustment direction and the second adjustment amount for the preset time period;

[0053] By combining the first adjustment direction, the first adjustment amount, the second adjustment direction, and the second adjustment amount with the running impact weight and the risk impact weight to obtain a comprehensive adjustment amount, perform an adjustment process on the preset time period.

[0054] Preferably, perform data integrity verification on the integrated multi-modal data of the chemical industrial park, including:

[0055] Compare and analyze the integrated multi-modal data of the chemical industrial park with the multi-modal data of the chemical industrial park before integration, determine whether the integrated multi-modal data of the chemical industrial park is missing, and verify the data integrity of the integrated multi-modal data of the chemical industrial park;

[0056] When the integrated multi-modal data of the chemical industrial park is the same as the multi-modal data of the chemical industrial park before integration, the integrated multi-modal data of the chemical industrial park is not missing, and the data integrity verification is qualified;

[0057] When the multimodal data of the integrated chemical industrial park is different from that of the chemical industrial park before integration, the multimodal data of the integrated chemical industrial park is missing, and the data integrity verification fails. At this time, the missing data is checked and the missing data is supplemented to make the multimodal data of the integrated chemical industrial park the same as that of the chemical industrial park before integration and make the data integrity verification qualified.

[0058] Preferably, potential safety hazards in the chemical industrial park are identified in advance, including:

[0059] According to the safety early warning requirements of the chemical industrial park, a large chemical industrial park risk identification model is established;

[0060] The large chemical industrial park risk identification model is deployed, and the large chemical industrial park risk identification model is deployed in the actual chemical industrial park risk identification environment;

[0061] The chemical industrial park characteristic data is input into the large chemical industrial park risk identification model, the chemical industrial park characteristic data is analyzed according to the large chemical industrial park risk identification model, and the chemical industrial park risk is identified to identify potential safety hazards in the chemical industrial park in advance and determine the chemical industrial park risk identification result.

[0062] Preferably, establishing a large chemical industrial park risk identification model includes:

[0063] Collect the historical data of the chemical industrial park and divide the collected historical data of the chemical industrial park into a training set and a test set;

[0064] Based on deep learning technology, the training set is used to train the deep learning model, so that the deep learning model autonomously learns the chemical industrial park risk identification behavior from the training set and identifies potential safety hazards in the chemical industrial park in advance to determine the large chemical industrial park risk identification model based on deep learning;

[0065] The test set is used to test the large chemical industrial park risk identification model based on deep learning, evaluate the performance of the large chemical industrial park risk identification model based on deep learning, and judge whether the large chemical industrial park risk identification model based on deep learning can achieve the expected effect of identifying potential safety hazards in the chemical industrial park in advance;

[0066] When the large chemical industrial park risk identification model based on deep learning cannot achieve the expected effect of identifying potential safety hazards in the chemical industrial park in advance, the parameters of the large chemical industrial park risk identification model based on deep learning are adjusted and optimized until the large chemical industrial park risk identification model based on deep learning can achieve the expected effect of identifying potential safety hazards in the chemical industrial park in advance, and then the best large chemical industrial park risk identification model is determined.

[0067] Preferably, it further includes:

[0068] Using a test set to test the large chemical industrial park risk identification model based on deep learning to obtain the identification test results;

[0069] Based on the identification test results, perform performance evaluation on the large chemical industrial park risk identification model based on deep learning to obtain the model performance evaluation score;

[0070] According to the model performance evaluation score, judge whether the large chemical industrial park risk identification model based on deep learning can achieve the expected effect of identifying potential safety hazards in the chemical industrial park in advance.

[0071] Preferably, perform safety early warning management on potential safety hazards in the chemical industrial park, including:

[0072] According to the chemical industrial park risk identification results, conduct timely safety early warning on potential safety hazards in the chemical industrial park, remind the chemical industrial park management personnel to pay attention to safety hazards, and intelligently guide the chemical industrial park management personnel to eliminate safety hazards, forming a closed-loop management of real-time monitoring and safety early warning in the chemical industrial park.

[0073] Compared with the prior art, the beneficial effects of the present invention are:

[0074] The present invention collects environmental data, equipment data, personnel data, visual and auditory data of the chemical industrial park to determine the multi-modal data of the chemical industrial park. By preprocessing the collected multi-modal data of the chemical industrial park, the characteristic data of the chemical industrial park is determined. By establishing a large chemical industrial park risk identification model, analyzing the characteristic data of the chemical industrial park, and identifying the risks of the chemical industrial park, potential safety hazards in the chemical industrial park are identified in advance, and the chemical industrial park risk identification results are determined. Furthermore, timely safety early warning is conducted on potential safety hazards in the chemical industrial park, reminding the chemical industrial park management personnel to pay attention to safety hazards, and intelligently guiding the chemical industrial park management personnel to eliminate safety hazards, forming a closed-loop management of real-time monitoring and safety early warning in the chemical industrial park. It can conduct real-time monitoring of the chemical industrial park, can identify and manage potential safety hazards in the chemical industrial park in advance and conduct safety early warning, effectively improving the accuracy and timeliness of safety early warning in the chemical industrial park and enhancing the safety of the chemical industrial park. BRIEF DESCRIPTION OF THE DRAWINGS

[0075] Figure 1 It is a module diagram of the chemical industrial park safety early warning system based on the multi-modal large model of the present invention;

[0076] Figure 2 It is a flow chart of the chemical industrial park safety early warning system based on the multi-modal large model of the present invention;

[0077] Figure 3This is a flowchart for the present invention to verify the data integrity of multimodal data in a chemical industrial park. Specific embodiments

[0078] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0079] To solve the problems of the existing system that cannot conduct real-time monitoring of chemical industrial parks, cannot identify potential safety hazards in chemical industrial parks in advance, and cannot manage safety early warnings, resulting in low accuracy and timeliness of safety early warnings in chemical industrial parks and reducing the safety of chemical industrial parks, please refer to Figures 1-3 , the following technical solutions are provided in this embodiment:

[0080] A safety early warning system for chemical industrial parks based on a multimodal large model includes: a park data collection module, a data processing and analysis module, and a safety early warning management module.

[0081] Specifically, through the interactive communication between the park data collection module, the data processing and analysis module, and the safety early warning management module, real-time monitoring of chemical industrial parks can be carried out, potential safety hazards in chemical industrial parks can be identified in advance and safety early warning management can be carried out, effectively improving the accuracy and timeliness of safety early warnings in chemical industrial parks and enhancing the safety of chemical industrial parks.

[0082] Among them, the park data collection module is used to collect multimodal data of chemical industrial parks;

[0083] In this embodiment, collecting multimodal data of chemical industrial parks includes:

[0084] Based on intelligent collection devices, real-time monitoring of temperature, humidity, pressure, and gas concentration in chemical industrial parks is carried out to collect environmental data of chemical industrial parks;

[0085] Based on intelligent collection devices, real-time monitoring of equipment vibration, current, and voltage in chemical industrial parks is carried out to collect equipment data of chemical industrial parks;

[0086] Based on intelligent collection devices, real-time monitoring of personnel positions and helmet-wearing situations in chemical industrial parks is carried out to collect personnel data of chemical industrial parks;

[0087] Based on intelligent collection devices, real-time monitoring of videos, images, and sounds in chemical industrial parks is carried out to collect visual and auditory data of chemical industrial parks;

[0088] Determine the multi-modal data of the chemical industrial park based on the environmental data, equipment data, personnel data, visual and auditory data of the chemical industrial park.

[0089] It should be noted that by collecting the environmental data, equipment data, personnel data, visual and auditory data of the chemical industrial park, the multi-modal data of the chemical industrial park is determined, providing a data basis for the subsequent early identification of potential safety hazards in the chemical industrial park.

[0090] Among them, the data processing and analysis module is used to preprocess and analyze the collected multi-modal data of the chemical industrial park, and early identify potential safety hazards in the chemical industrial park;

[0091] In this embodiment, the preprocessing of the collected multi-modal data of the chemical industrial park includes:

[0092] Clean the multi-modal data of the chemical industrial park, and remove the noise and duplicate data that are useless for the safety warning of the chemical industrial park from the multi-modal data of the chemical industrial park;

[0093] Check the multi-modal data of the chemical industrial park, judge whether there are missing values and outliers in the multi-modal data of the chemical industrial park, and process the missing values and outliers existing in the multi-modal data of the chemical industrial park;

[0094] Judge the missing values and outliers existing in the multi-modal data of the chemical industrial park, and judge whether the missing values and outliers existing in the multi-modal data of the chemical industrial park are useful for the safety warning of the chemical industrial park;

[0095] If it is useful, fill in the missing values and replace the outliers in the multi-modal data of the chemical industrial park. If it is useless, remove the missing values and outliers existing in the multi-modal data of the chemical industrial park;

[0096] It should be noted that by cleaning the multi-modal data of the chemical industrial park, the quality of the multi-modal data of the chemical industrial park can be improved.

[0097] Normalize the multi-modal data of the chemical industrial park, convert the multi-modal data of the chemical industrial park into a unified data format, remove the dimensional differences in the multi-modal data of the chemical industrial park, and determine the standardized multi-modal data of the chemical industrial park;

[0098] Integrate the multi-modal data of the chemical industrial park, integrate the multi-modal data of the chemical industrial park into a unified data view, and perform data integrity verification on the integrated multi-modal data of the chemical industrial park. After the data integrity verification is qualified, securely store the integrated multi-modal data of the chemical industrial park;

[0099] Perform modal alignment and feature extraction on the multi-modal data of the chemical industrial park, extract the feature vectors useful for the safety warning of the chemical industrial park from the multi-modal data of the chemical industrial park, and perform weighted fusion on the extracted feature vectors to determine the characteristic data of the chemical industrial park.

[0100] It should be noted that by normalizing, integrating, performing modal alignment and feature extraction on the multi-modal data of the chemical industrial park, the characteristic data of the chemical industrial park is determined, which is convenient for better analyzing the potential safety hazards in the chemical industrial park.

[0101] In this embodiment, the feature vectors useful for the safety warning of the chemical industrial park are extracted from the multi-modal data of the chemical industrial park, including:

[0102] Perform regional data division on the obtained multi-modal data of the chemical industrial park to obtain the sub-region data corresponding to each chemical sub-region of the current chemical industrial park;

[0103] Adopt the operation status evaluation index adapted to the chemical sub-region, and perform operation status evaluation based on the sub-region data to obtain the real-time operation status of the chemical sub-region;

[0104] Obtain the historical risk record data of each chemical sub-region within the preset time period;

[0105] Extract the historical risk categories, the corresponding number of risk occurrences and the risk severity levels of each chemical sub-region from the historical risk record data, and input them into the pre-established risk assessment model to obtain the estimated risk status of the corresponding chemical sub-region;

[0106] When the real-time operation status of a chemical sub-region is in a dangerous state, extract all the feature vectors adapted to the current chemical sub-region being in a dangerous state and output them as key feature vectors;

[0107] When the real-time operation status of a chemical sub-region is not in a dangerous state, label the chemical sub-region with the same state level of the estimated risk status and the real-time operation status as the first region, label the chemical sub-region with the state level of the estimated risk status greater than the state level of the real-time operation status as the second region, and label the chemical sub-region with the state level of the estimated risk status less than the state level of the real-time operation status as the third region;

[0108] Output all the feature vectors adapted to the real-time operation status of the current first region as the first key feature vectors;

[0109] Obtain the number of the first state levels by which the corresponding estimated risk status of the second region exceeds the real-time operation status;

[0110] Using the risk occurrence frequencies of various historical risk categories existing in the second region and the number of first state levels as matching conditions, determine the enhanced feature vectors corresponding to various historical risk categories;

[0111] From all the feature vectors adapted to the real-time operating state of the second region, screen out the feature vectors associated with various historical risk categories existing in the second region, and after replacing them with the corresponding enhanced feature vectors, output them together with all the other non-replaced feature vectors as key feature vectors;

[0112] Obtain the number of second state levels corresponding to the estimated risk state of the third region that is less than the real-time operating state;

[0113] Using the risk occurrence frequencies of various historical risk categories existing in the third region and the number of second state levels as matching conditions, determine the mitigation feature vectors corresponding to various historical risk categories;

[0114] From all the feature vectors adapted to the real-time operating state of the third region, screen out the feature vectors associated with various historical risk categories existing in the third region, and after replacing them with the corresponding mitigation feature vectors, output them together with all the other non-replaced feature vectors as key feature vectors;

[0115] Summarize all the key feature vectors and output them as the feature vectors useful for the safety warning of the chemical industrial park.

[0116] In this embodiment, the chemical sub-region refers to a relatively independent regional unit pre-divided according to risk characteristics, such as flammable and explosive areas, personnel gathering areas, hazardous substance storage areas, hazardous chemical transportation channels and loading and unloading areas, etc.; sub-region data refers to the data set extracted from the multi-modal data of the chemical industrial park according to the division of the chemical sub-regions and related to specific sub-regions; the operating state evaluation index refers to a set of quantitative standards pre-established for evaluating the operating state of the chemical sub-region according to the characteristics and risk factors of the chemical sub-region, such as the operating state evaluation index of the flammable and explosive area includes temperature, pressure, gas concentration, and equipment operating parameters, etc.; the operating state evaluation index of the hazardous chemical transportation channel and the loading and unloading area includes the operating parameters of the transport vehicle, the operating parameters of the loading and unloading machinery, the transportation environmental conditions, and the tank sealing performance, etc.

[0117] In this embodiment, the real-time operating states include three types: safe state, early warning state, and dangerous state, and the corresponding state levels are represented as level 1, level 2, and level 3 respectively. The specific acquisition steps are as follows: First, extract the index data related to the corresponding operating state evaluation index from the sub-region data of the chemical industrial sub-region. Then, compare the index data with the corresponding threshold range of the operating state evaluation index. Score 0 for the operating state evaluation index whose index data exceeds the threshold range, and score 1 for the operating state evaluation index whose index data does not exceed the threshold range. Then, directly perform weighted average on the scores of all corresponding operating state evaluation indexes of the current chemical industrial sub-region and the weights of each operating state evaluation index determined by the analytic hierarchy process (the value range is all ), to obtain the comprehensive score of the sub-region. Finally, match the comprehensive score of the sub-region with the preset state level range (the value range is all ), and take the corresponding real-time operating state of the preset state level range to which the comprehensive score of the sub-region belongs as the real-time operating state of the current chemical industrial sub-region. Among them, the basis for using the weighted average method is that the operating state of the chemical industrial sub-region is affected by multiple evaluation indexes, and a single index is difficult to comprehensively and accurately reflect the situation. The weighted average method can integrate multi-index data, comprehensively consider various factors, and make the evaluation result more comprehensive.

[0118] Among them, the basis for using the analytic hierarchy process (the core steps include constructing a hierarchical structure model, constructing a judgment matrix based on the expert scoring results obtained by the 1-9 scale method, calculating weights for the judgment matrix, consistency test, and final weight synthesis) to determine the weights of each operating state evaluation index is that the evaluation of the operating state of the chemical industrial sub-region involves multiple evaluation indexes. By constructing a hierarchical structure model, the relative importance of each index can be clarified, providing a reasonable framework for the determination of weights. And when determining the weights of evaluation indexes, both the objective attributes of the indexes and subjective judgments need to be considered. The analytic hierarchy process quantifies the weights through a method combining qualitative and quantitative methods and consistency test, making the weight determination result more scientific and reasonable.

[0119] In this embodiment, for example, the operating state evaluation indexes of the chemical industrial sub-region 1 - flammable and explosive sub-region include temperature, pressure, gas concentration, and equipment operating parameters. According to the analytic hierarchy process, the temperature weight is (0.4), the pressure weight is (0.3), the gas concentration weight is (0.2), and the equipment operating parameter weight is (0.1);

[0120] Real-time monitoring data: Temperature: 75°C (threshold range: ≤80°C); Pressure: 0.8 MPa (threshold range: ≤1.0 MPa); Gas concentration: 15% LEL (threshold range: ≤10% LEL); Equipment operating parameter: Vibration frequency 25 Hz (threshold range: ≤20 Hz);

[0121] The index scores are as follows:

[0122] Temperature: 75°C 80°C, the score is 1;

[0123] Pressure: 0.8 MPa 1.0 MPa, the score is 1;

[0124] Gas concentration: 15% LEL 10% LEL, the score is 0;

[0125] Equipment operating parameter: 25 Hz 20 Hz, the score is 0;

[0126] At this time, the sub-region comprehensive score of Chemical Sub-region 1 = (Temperature score × Temperature weight) + (Pressure score × Pressure weight) + (Gas concentration score × Gas concentration weight) + (Equipment operating parameter score × Equipment operating parameter weight), that is, the sub-region comprehensive score of Chemical Sub-region 1 .

[0127] In this embodiment, the preset time period refers to a specific time period set when extracting historical risk record data; historical risk record data refers to the relevant data recording various safety risk events that occurred in the chemical sub-region in the past period of time, including the time, location (specific to the chemical sub-region), risk category (such as fire, explosion, leakage, etc.), cause of the risk occurrence, and risk severity level, etc.; historical risk category refers to the category of historical risk accidents that occurred in the chemical sub-region extracted from the historical risk record data during the preset time period, such as fire, explosion, toxic and harmful gas leakage, chemical substance leakage, equipment failure, etc.; risk occurrence frequency refers to the number of times a certain historical risk category occurs in a specific chemical sub-region during the preset time period; risk severity level is a quantitative index used to measure the possible harm degree of risk events occurring in the chemical sub-region, which is obtained by comprehensively evaluating and classifying the risk severity in advance according to the possible casualties, property losses, and environmental pollution degrees caused by the risk events, and is divided into four levels: minor, general, serious, and major.

[0128] In this embodiment, the risk assessment model is used to estimate the risk status of chemical industrial sub-regions. It is obtained by collecting historical risk record data of each chemical industrial sub-region in the chemical industrial park, including information such as the time and location of risk occurrence, risk category, number of risk occurrences, risk severity level, and losses caused, and cleaning, classifying, and labeling these data to construct a complete risk data set. Then, features related to regional risk assessment are extracted from the risk data set, such as risk occurrence frequency, risk severity level, risk type distribution, and regional features (such as equipment type, complexity of the process flow, and personnel density), and used as a training data set to train a neural network. The neural network adopts a multi-layer feed-forward structure, including an input layer, a hidden layer, and an output layer. The number of nodes in the input layer corresponds to the dimension of risk features, such as the number of risk occurrences, risk severity level, and risk type distribution. The activation function of the hidden layer selects ReLU (to alleviate the vanishing gradient), and the output layer uses the Softmax activation function to convert the output of the neural network into a risk level probability distribution. The estimated risk status includes three types: low risk, medium risk, and high risk, and the corresponding status levels are represented as level 1, level 2, and level 3, respectively.

[0129] In this embodiment, the first region refers to a chemical industrial sub-region where, when the real-time operating state of the chemical industrial sub-region is not in a dangerous state, the estimated risk status is consistent with the status level of the real-time operating state; the second region refers to a chemical industrial sub-region where, in the case that the real-time operating state of the chemical industrial sub-region is not in a dangerous state, the status level of the estimated risk status is higher than the status level of the real-time operating state; the third region refers to a chemical industrial sub-region where, when the real-time operating state of the chemical industrial sub-region is not in a dangerous state, the status level of the estimated risk status is lower than the status level of the real-time operating state. For example, there is a chemical industrial sub-region 2 with a real-time operating state of a safe state (status level represented as level 1) and an estimated risk status of medium risk (status level represented as level 2). The status level (level 2) of the estimated risk status of chemical industrial sub-region 1 is higher than the status level (level 1) of the real-time operating state. At this time, chemical industrial sub-region 2 is labeled as the second region.

[0130] In this embodiment, the number of the first state levels refers to the specific number of state levels where the predicted risk state of the second area exceeds its real-time operating state. For example, if the real-time operating state level is level 2 and the predicted risk state level is level 3, then the number of the first state levels is 1 level (3 - 2 = 1); the enhanced feature vector refers to a feature set determined for the second area (where the predicted risk state is higher than the real-time operating state) and used to replace the feature vector associated with the risk category, which is screened from the preset feature data list by using the risk occurrence frequency of various historical risk categories and the number of the first state levels as matching conditions; among them, the preset feature data list is a feature set formed by integrating multi-source data such as the historical safety data of chemical industrial parks, industry safety specifications, and expert experience in advance, covering the feature mapping relationships of different risk categories and state levels.

[0131] In this embodiment, the number of the second state levels refers to the specific number of state levels where the predicted risk state of the third area is less than its real-time operating state. For example, if the real-time operating state level is level 2 and the predicted risk state level is level 1, then the number of the second state levels is 1 level (2 - 1 = 1); the mitigation feature vector refers to a feature set determined for the third area (where the predicted risk state is lower than the real-time operating state) and used to replace the feature vector associated with the risk category, which is screened from the preset feature data list by using the risk occurrence frequency of various historical risk categories and the number of the second state levels as matching conditions.

[0132] In this embodiment, for example, the real-time operating state of chemical sub-region 3 is in a warning state (level 2), but the predicted risk state is in a dangerous state (level 3), and the number of the first state levels is 1 level;

[0133] The matching conditions include: risk category: leakage, risk occurrence frequency: 3 times / 6 months, number of the first state levels: 1 level;

[0134] The following enhanced feature vectors are matched from the preset feature data list:

[0135] Feature 1: Response time of leakage detector (historical average 0.5 seconds, current 1.2 seconds);

[0136] Feature 2: Simulated risk radius of leakage diffusion (standard 50 meters, current 80 meters);

[0137] Replacement operation:

[0138] Replace "leakage detector status" in all feature vectors determined according to the warning state of chemical sub-region 3 with "response time 1.2 seconds", and "simulated risk radius of leakage diffusion" with "80 meters".

[0139] The working principle of the above technical solution is as follows: First, the multi-modal data of the chemical industrial park is divided according to chemical sub-regions to form sub-region data; then, the operation status evaluation indicators adapted to each sub-region are used to evaluate the real-time operation status based on the sub-region data; at the same time, the historical risk record data of each sub-region within a preset period is obtained, the risk categories, occurrence times and severity levels are extracted, and the predicted risk status is obtained by inputting into the risk assessment model, and potential risks are predicted from the historical dimension; then, differential processing is carried out according to the real-time operation status: if the sub-region is in a dangerous state, directly extract the feature vector adapted to this state as the key feature vector; if it is not in a dangerous state, the sub-region is divided into the first, second, and third regions according to the relationship between the predicted risk status and the real-time operation status level; for the first region, directly extract the feature vector adapted to its state; for the second region, determine the enhanced feature vector with the occurrence frequency of the historical risk category and the number of predicted risk levels exceeding the real-time status level as the matching conditions, and output after replacing the associated feature vector; for the third region, output after replacing the associated feature vector with the mitigation feature vector according to a similar logic. Finally, all the key feature vectors are aggregated and output as the feature vectors useful for the safety warning of the chemical industrial park.

[0140] The beneficial effects of the above technical solution are as follows: By dividing the multi-modal data according to sub-regions and using the adapted indicators to evaluate the real-time operation status, and combining the historical risk record data to predict the risk status, the safety situation of each sub-region is analyzed multi-dimensionally and comprehensively, and the feature vectors useful for the safety warning of the chemical industrial park are screened out, which helps to avoid the misselection of feature vectors caused by a single data source or a simple evaluation method, and provides a solid and reliable basis for subsequent safety warnings.

[0141] In this embodiment, it further includes:

[0142] Real-time monitor the changes in the operation status and risk status of each chemical sub-region in the chemical industrial park, and correspondingly obtain the operation change data and risk change data;

[0143] According to the operation change data, determine the number of the first status of the chemical sub-regions whose real-time operation status remains in the dangerous state per unit time within the set sliding window in the chemical industrial park, and the number of the first status times when the real-time operation status changes from the warning state to the dangerous state;

[0144] Use the number of the first status and the number of the first status times to construct an operation danger matrix;

[0145] Determine the number of the second status of the chemical sub-regions whose real-time operation status remains in the safe state per unit time within the set sliding window in the chemical industrial park, and the number of the second status times when the real-time operation status changes from the warning state to the safe state;

[0146] Use the number of the second status and the number of the second status times to construct an operation safety matrix;

[0147] By mapping the operation risk matrix and the operation safety matrix using the first preset adjustment analysis strategy, determine the first adjustment direction and the first adjustment amount for a preset time period;

[0148] According to the risk change data, determine, for each unit of time within the set sliding window of the chemical industrial park, the first number of risks of the chemical sub-regions whose estimated risk status remains at a high-risk status, and the first number of risk transitions from a medium-risk status to a high-risk status;

[0149] Using the first number of risks and the first number of risk transitions, construct a high-risk change matrix;

[0150] Determine, for each unit of time within the set sliding window of the chemical industrial park, the second number of risks of the chemical sub-regions whose estimated risk status remains at a low-risk status, and the second number of risk transitions from a medium-risk status to a low-risk status;

[0151] Using the second number of risks and the second number of risk transitions, construct a low-risk change matrix;

[0152] By mapping the low-risk change matrix and the high-risk change matrix using the second preset adjustment analysis strategy, determine the second adjustment direction and the second adjustment amount for a preset time period;

[0153] By combining the first adjustment direction, the first adjustment amount, the second adjustment direction, and the second adjustment amount with the operation impact weight and the risk impact weight to obtain a comprehensive adjustment amount, perform adjustment processing on the preset time period.

[0154] In this embodiment, the operation change data refers to the real-time operation status of each chemical sub-region within the chemical industrial park through real-time monitoring; the risk change data refers to the estimated risk status of each chemical sub-region within the chemical industrial park through real-time monitoring.

[0155] In this embodiment, the set sliding window is used to perform segmented analysis on the operation change data and the risk change data of the chemical industrial park. The fixed length is pre-determined, such as 24 hours; the first number of statuses refers to the number of chemical sub-regions whose real-time operation status remains at a dangerous status within each unit of time of the set sliding window. For example, if the set sliding window is 12 hours and each unit of time is 0.5 hours, the first number of status transitions refers to the number of times the real-time operation status of the chemical sub-region changes from a warning status to a dangerous status within each unit of time of the set sliding window.

[0156] In this embodiment, the operating risk matrix refers to a matrix constructed using the number of first states and the number of times of the first states, where the rows represent time intervals and the columns represent the number of first states and the number of times of the first states of chemical industrial sub - regions; the number of second states refers to the number of chemical industrial sub - regions whose real - time operating state remains in a safe state per unit time within a set sliding window; the number of second times refers to the number of times that the real - time operating state of a chemical industrial sub - region changes from a warning state to a safe state per unit time within a set sliding window; the operating safety matrix refers to a matrix constructed using the number of second states and the number of times of the second states, where the rows represent time intervals and the columns represent the number of second states and the number of times of the second states of chemical industrial sub - regions.

[0157] In this embodiment, the specific operation steps of the first preset adjustment analysis strategy are as follows:

[0158] 1. Normalize the ratio obtained by directly dividing the sum of all the number of times of the first states in the operating risk matrix by the sum of all the number of first states to obtain the risk state diffusion rate;

[0159] 2. Normalize the ratio obtained by directly dividing the sum of all the number of times of the second states in the operating safety matrix by the sum of all the number of second states to obtain the safety state recovery rate;

[0160] 3. Calculate the risk difference obtained by subtracting the preset risk threshold from the risk state diffusion rate, and the safety difference obtained by subtracting the preset safety threshold from the safety state recovery rate;

[0161] 4. Obtain the first adjustment amount by directly weighted - summing the risk difference and the safety difference;

[0162] Among them, the first adjustment amount is expressed as ; in the formula, represents the first adjustment amount; represents the risk difference; represents the preset risk threshold; represents the contribution weight of the risk difference to calculating the first adjustment amount; represents the contribution weight of the safety difference to calculating the first adjustment amount; represents the safety difference; represents the preset safety threshold;

[0163] Among them, the weights assigned to the risk difference and the safety difference are obtained by solving the matrix constructed through pairwise comparison and scoring using the analytic hierarchy process, and the value ranges are both ; the preset risk threshold or the preset safety threshold refers to the value obtained by first sorting out the historical data of a chemical industrial park in the past period (such as 6 months) from the current moment, and then summarizing and averaging all the risk state diffusion rates or safety state recovery rates in the historical data.

[0164] 5. When the first adjustment amount is positive, the first adjustment direction is extension; when the first adjustment amount is negative, the first adjustment direction is shortening.

[0165] Among them, the specific steps of assigning weights to the risk difference and safety difference using the analytic hierarchy process refer to: First, construct a hierarchical structure model consisting of an objective layer (focusing on determining the weights of the risk difference and safety difference in the evaluation of the operation status of chemical sub-regions), a criterion layer (key factors affecting the weight distribution of the risk difference and safety difference), and a scheme layer (the risk difference and safety difference to be weighted); then, according to the 1 - 9 scale method, conduct pairwise comparison and scoring of each criterion in the criterion layer and the elements in the scheme layer for each criterion by experts, and construct a judgment matrix based on the expert scoring results; then, for the judgment matrix, calculate the weights using the geometric mean method or the arithmetic mean method and conduct a consistency test; finally, if there are multiple criteria in the criterion layer, synthesize their weights with the weights of the elements in the scheme layer corresponding to the criteria to obtain the final weights of the risk difference and safety difference, and the values of both are between 0 - 1 and their sum is 1.

[0166] In this embodiment, the first adjustment direction refers to the direction of adjusting the preset time period determined after analyzing the operation risk matrix and the operation safety matrix according to the first preset adjustment analysis strategy, including two directions: shortening and extension; the first adjustment amount refers to the specific ratio of adjusting the preset time period further determined on the basis of determining the first adjustment direction.

[0167] In this embodiment, the first risk number refers to the number of chemical sub-regions whose predicted risk status remains at a high-risk status per unit time within the set sliding window. For example, taking a 12-hour set sliding window and 0.5 hours as the unit time as an example, the number of chemical sub-regions with a predicted high-risk status per 0.5 hours within 12 hours is counted and output as the first risk number; the first risk frequency refers to the number of times the predicted risk status of a chemical sub-region changes from a medium-risk status to a high-risk status per unit time within the set sliding window. For example, taking a 12-hour set sliding window and 0.5 hours as the unit time as an example, the number of times the predicted risk status changes from a medium-risk status to a high-risk status per 0.5 hours within 12 hours is counted and output as the first risk frequency.

[0168] In this embodiment, the high-risk change matrix refers to a matrix constructed using the first risk number and the first risk frequency, where the rows represent time intervals and the columns represent the first risk number and the first risk frequency of chemical sub-regions.

[0169] In this embodiment, the number of second risks refers to the number of chemical sub - regions whose predicted risk status remains at a low risk per unit time within a set sliding window; the number of second risk occurrences refers to the number of times the predicted risk status of a chemical sub - region changes from a medium - risk state to a low - risk state per unit time within a set sliding window; the low - risk change matrix refers to the number of times the predicted risk status of a chemical sub - region changes from a medium - risk state to a low - risk state per unit time within a set sliding window, where the rows represent time intervals and the columns represent the number of second risks and the number of second risk occurrences of chemical sub - regions.

[0170] In this embodiment, the specific operation steps of the second preset adjustment analysis strategy are as follows:

[0171] 1. Normalize the ratio obtained by directly dividing the sum of all first risk occurrences in the high - risk change matrix by the sum of all first risk numbers to obtain the high - risk upgrade rate.

[0172] 2. Normalize the ratio obtained by directly dividing the sum of all second risk occurrences in the low - risk change matrix by the sum of all second risk numbers to obtain the low - risk downgrade rate.

[0173] 3. Calculate the upgrade difference by subtracting the preset upgrade threshold from the high - risk upgrade rate, and the downgrade difference by subtracting the preset downgrade threshold from the low - risk downgrade rate.

[0174] 4. Obtain the second adjustment amount by directly weighted - summing the upgrade difference and the downgrade difference.

[0175] Among them, the second adjustment amount is expressed as ; in the formula, represents the second adjustment amount; represents the upgrade difference; represents the preset upgrade threshold; represents the contribution weight of the upgrade rate to calculating the second adjustment amount; represents the contribution weight of the downgrade rate to calculating the second adjustment amount; represents the downgrade difference; represents the preset downgrade threshold;

[0176] Among them, the weights assigned to the upgrade rate and the downgrade rate are obtained by solving a matrix constructed through pairwise comparison and scoring using the analytic hierarchy process (the core steps include constructing a hierarchical structure model, constructing a judgment matrix based on the expert scoring results obtained by the 1 - 9 scale method, calculating weights for the judgment matrix, consistency test, and final weight synthesis), and the value ranges are both ; The preset upgrade threshold or preset downgrade threshold refers to the value obtained by first sorting out the historical data of the chemical industrial park in the past period (such as 6 months) from the current moment, and then summarizing and averaging all the upgrade rates or downgrade rates in the historical data.

[0177] 5. When the second adjustment amount is positive, the second adjustment direction is extension; when the second adjustment amount is negative, the second adjustment direction is shortening.

[0178] In this embodiment, the second adjustment direction refers to the direction of adjusting the preset time period determined after analyzing the low-risk change matrix and the high-risk change matrix according to the second preset adjustment analysis strategy, including two types: shortening and extension; the second adjustment amount refers to the specific ratio of adjusting the preset time period further clarified on the basis of determining the second adjustment direction.

[0179] In this embodiment, the comprehensive adjustment amount refers to the final adjustment ratio of the preset time period obtained by combining the first adjustment direction, the first adjustment amount, the second adjustment direction and the second adjustment amount with the operation influence weight and the risk influence weight. Among them, the operation influence weight reflects the influence degree of the change of the operation state of the chemical industrial park on the time period adjustment (the value range is ), and the risk influence weight reflects the influence degree of the change of the estimated risk state on the time period adjustment (the value range is ); the operation influence weight and the risk influence weight are obtained by solving the matrix constructed by pairwise comparison and scoring using the analytic hierarchy process.

[0180] In this embodiment, for example, there is a preset time period of Chemical Industrial Park 1 ( , expressed in minutes). The first adjustment direction is shortening, and the first adjustment amount is 0.3; the second adjustment direction is shortening, and the second adjustment amount is 0.8; the operation influence weight is 0.4, and the risk influence weight is 0.6;

[0181] At this time, the comprehensive adjustment amount of Chemical Industrial Park 1 , the adjustment direction is shortening, and the adjusted preset time period is min, that is, 42 minutes.

[0182] In this embodiment, for example, there is a preset time period of Chemical Industrial Park 2 ( expressed in minutes). The first adjustment direction is extension, and the first adjustment amount is 1.2; the second adjustment direction is shortening, and the second adjustment amount is 0.3; the operation influence weight is 0.3, and the risk influence weight is 0.6;

[0183] At this time, the comprehensive adjustment amount of Chemical Industrial Park 2 , adjust the direction to extension, and the adjusted preset time period is min, that is, 103.5 minutes.

[0184] The working principle of the above technical solution is as follows: First, continuously collect the real-time operation status and estimated risk status of each chemical sub-region in the chemical industrial park, and continuously obtain operation change data and risk change data; then, take the set sliding window as the time analysis unit, analyze the operation change data and risk change data, and respectively obtain the adjustment direction and adjustment amount related to operation and risk; finally, comprehensively consider the adjustment direction and adjustment amount of operation and risk, combine the operation influence weight and risk influence weight, calculate the comprehensive adjustment amount, and accurately adjust the preset time period accordingly.

[0185] The beneficial effects of the above technical solution are: By real-time monitoring the changes in the operation status and risk status of chemical sub-regions, constructing corresponding matrices, it is possible to accurately analyze the operation situation of the park, determine the adjustment direction and adjustment amount related to operation and risk respectively; then, analyze and comprehensively calculate the obtained adjustment direction and adjustment amount related to operation and risk, combine the corresponding influence weights to obtain the comprehensive adjustment amount, determine the comprehensive adjustment amount, and comprehensively adjust the preset time period, so as to realize the accurate adjustment of the frequency of data collection and analysis, avoid missing the best warning opportunity due to too long time interval, and greatly improve the accuracy and timeliness of safety warning.

[0186] In this embodiment, data integrity verification is performed on the integrated multi-modal data of the chemical industrial park, including:

[0187] Compare and analyze the integrated multi-modal data of the chemical industrial park with the multi-modal data of the chemical industrial park before integration, judge whether there is any omission in the integrated multi-modal data of the chemical industrial park, and verify the data integrity of the integrated multi-modal data of the chemical industrial park;

[0188] When the integrated multi-modal data of the chemical industrial park is the same as the multi-modal data of the chemical industrial park before integration, there is no omission in the integrated multi-modal data of the chemical industrial park, and the data integrity verification is qualified;

[0189] When the integrated multi-modal data of the chemical industrial park is different from the multi-modal data of the chemical industrial park before integration, there is an omission in the integrated multi-modal data of the chemical industrial park, and the data integrity verification is unqualified. At this time, check the missing data and supplement the missing data to make the integrated multi-modal data of the chemical industrial park the same as the multi-modal data of the chemical industrial park before integration, and make the data integrity verification qualified.

[0190] In this embodiment, potential safety hazards in the chemical industrial park are identified in advance, including:

[0191] According to the safety warning requirements of chemical industrial parks, collect the historical data of chemical industrial parks, and divide the collected historical data of chemical industrial parks into a training set and a test set;

[0192] Based on deep learning technology, use the training set to train the deep learning model, enabling the deep learning model to autonomously learn the risk identification behavior of chemical industrial parks from the training set, and to identify potential safety hazards in chemical industrial parks in advance, thereby determining a large-scale chemical industrial park risk identification model based on deep learning;

[0193] Use the test set to test the large-scale chemical industrial park risk identification model based on deep learning, evaluate the performance of the large-scale chemical industrial park risk identification model based on deep learning, and determine whether the large-scale chemical industrial park risk identification model based on deep learning can achieve the expected effect of identifying potential safety hazards in chemical industrial parks in advance;

[0194] When the large-scale chemical industrial park risk identification model based on deep learning cannot achieve the expected effect of identifying potential safety hazards in chemical industrial parks in advance, adjust and optimize the parameters of the large-scale chemical industrial park risk identification model based on deep learning until the large-scale chemical industrial park risk identification model based on deep learning can achieve the expected effect of identifying potential safety hazards in chemical industrial parks in advance, and then determine the best large-scale chemical industrial park risk identification model;

[0195] Deploy the large-scale chemical industrial park risk identification model, and deploy the large-scale chemical industrial park risk identification model in the actual chemical industrial park risk identification environment;

[0196] Input the chemical industrial park feature data into the large-scale chemical industrial park risk identification model, analyze the chemical industrial park feature data according to the large-scale chemical industrial park risk identification model, identify the chemical industrial park risks, and identify potential safety hazards in chemical industrial parks in advance to determine the chemical industrial park risk identification results.

[0197] Among them, the safety warning management module is used to conduct safety warning management on potential safety hazards in chemical industrial parks.

[0198] In this embodiment, it further includes:

[0199] Use the test set to test the large-scale chemical industrial park risk identification model based on deep learning to obtain the identification test results;

[0200] Based on the identification test results, conduct performance evaluation on the large-scale chemical industrial park risk identification model based on deep learning to obtain the model performance evaluation score;

[0201] According to the model performance evaluation score, it is judged whether the deep learning-based large model for chemical industrial park risk identification can achieve the expected effect of identifying potential safety hazards in chemical industrial parks in advance.

[0202] In this embodiment, the identification test result refers to the specific result output after verifying the deep learning risk identification large model based on the test set data; the model performance evaluation score is obtained by evaluating the response ability and identification ability of the model through the identification test result obtained by testing the deep learning-based chemical industrial park risk identification large model with the test set, and is used to reflect the identification performance of the model.

[0203] Among them, the calculation formula of the model performance evaluation score is as follows:

[0204] ;

[0205] In the formula, represents the model performance evaluation score of the deep learning-based chemical industrial park risk identification large model; represents the average response time for generating the identification test result of the deep learning-based chemical industrial park risk identification large model; e represents the base of the natural logarithm, with a value of 2.7; represents the i-th identification ability evaluation index, where i = 1, 2, 3; represents the influence weight of the i-th set performance index on evaluating the model performance;

[0206] Among them, the identification ability evaluation index refers to accuracy, precision, and recall rate; the identification ability evaluation index is determined by comparing the identification test result with the pre-determined true identification result; the weights assigned to the set performance indicators are obtained by solving the matrix constructed through pairwise comparison and scoring using the analytic hierarchy process, and the value ranges are all .

[0207] In this embodiment, the design logic of the model performance evaluation score formula is developed by comprehensively considering two key factors: the model response speed and identification accuracy. Among them, the part of the formula is used to measure the response speed of the model. By combining the response time of the model with the decay characteristics of the exponential function, it means that the shorter the response time, the larger the value, the better the model performs in terms of response speed, and the greater the positive contribution to the evaluation score; the part of the formula is used to measure the identification accuracy of the model. By combining the comprehensive performance index involving multiple factors related to identification accuracy with the decay characteristics of the exponential function, then the larger, the smaller, The larger it is, that is, the better the comprehensive performance index ( the larger it is), the greater the positive contribution to the evaluation score; by combining the response speed and recognition accuracy, a comprehensive model performance evaluation score is formed to jointly reflect the overall performance of the model;

[0208] Among them, the reason for using the exponential function in the model performance evaluation score formula to smoothly represent the weights of response time and recognition ability is that when evaluating the model performance, the changes in response time and comprehensive performance index are often continuous, and the exponential function can well capture this continuous change, making the evaluation score change smoothly with the changes in response time and comprehensive performance index; through the form of the exponential function, the weights of response time and recognition accuracy in the evaluation score can be conveniently adjusted; and the improvement of model performance is often not linear but has a certain marginal effect, and the exponential function can better simulate this marginal effect, that is, as the response time shortens or the comprehensive performance index increases, the improvement amplitude of model performance gradually decreases.

[0209] In this embodiment, when the model performance evaluation score exceeds the preset performance threshold, it is determined that the large model for chemical industrial park risk identification based on deep learning can achieve the expected effect of pre-identifying potential safety hazards in the chemical industrial park; when the model performance evaluation score does not exceed the preset performance threshold, it is determined that the large model for chemical industrial park risk identification based on deep learning cannot achieve the expected effect of pre-identifying potential safety hazards in the chemical industrial park; among them, the preset performance threshold is determined in advance based on the comprehensive consideration of the safety requirements and industry norms of the chemical industry and combined with expert scoring.

[0210] The beneficial effects of the above technical solution are: by using the test set to test the large model, accurate recognition test results are obtained; based on the recognition test results, the model performance is evaluated to obtain the model performance evaluation score, which can intuitively reflect the model's ability to identify potential safety hazards in the chemical industrial park; judging whether the model can achieve the expected effect of pre-identifying safety hazards according to the model performance evaluation score helps to screen out models with excellent performance and eliminate models that do not meet the requirements, so as to ensure that the models put into use have reliable risk identification capabilities.

[0211] In this embodiment, safety early warning management for potential safety hazards in the chemical industrial park includes:

[0212] According to the chemical industrial park risk identification results, timely safety early warning is carried out for potential safety hazards in the chemical industrial park, reminding the chemical industrial park management personnel to pay attention to safety hazards and intelligently guiding the chemical industrial park management personnel to eliminate safety hazards, forming a closed-loop management of real-time monitoring and safety early warning in the chemical industrial park.

[0213] In summary, by collecting environmental data, equipment data, personnel data, visual and auditory data of a chemical industrial park, multimodal data of the chemical industrial park is determined. By preprocessing the collected multimodal data of the chemical industrial park, characteristic data of the chemical industrial park is determined. By establishing a large model for risk identification in the chemical industrial park, the characteristic data of the chemical industrial park is analyzed, and the risks in the chemical industrial park are identified. Potential safety hazards in the chemical industrial park are identified in advance, and the risk identification results of the chemical industrial park are determined. Furthermore, timely safety warnings are issued for the potential safety hazards in the chemical industrial park, reminding the management personnel of the chemical industrial park to pay attention to the safety hazards and intelligently guiding the management personnel of the chemical industrial park to eliminate the safety hazards, forming a closed-loop management of real-time monitoring and safety warning in the chemical industrial park. The chemical industrial park can be monitored in real time, and potential safety hazards in the chemical industrial park can be identified and managed for safety warning in advance, effectively improving the accuracy and timeliness of safety warning in the chemical industrial park and enhancing the safety of the chemical industrial park.

[0214] It should be noted that, in this document, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or sequence between these entities or operations. Moreover, the term "comprising", "including" or any other variation thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device comprising a series of elements not only includes those elements but also includes other elements not expressly listed, or further includes elements inherent to such process, method, article or device.

[0215] Although the embodiments of the present invention have been shown and described, it will be understood by those of ordinary skill in the art that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A chemical industrial park safety early warning system based on a multimodal large model, characterized in that, Including: A park data collection module configured to collect multi-modal data of a chemical industrial park; A data processing and analysis module configured to preprocess and analyze the collected multi-modal data of the chemical industrial park to identify potential safety hazards in the chemical industrial park in advance; A safety early warning management module configured to conduct safety early warning management on potential safety hazards in the chemical industrial park; Extract feature vectors useful for safety early warning of the chemical industrial park from the multi-modal data of the chemical industrial park, including: Conduct regional data division on the obtained multi-modal data of the chemical industrial park to obtain sub-region data corresponding to each chemical sub-region of the current chemical industrial park; Adopt operation status evaluation indicators adapted to the chemical sub-region, and conduct operation status evaluation based on the sub-region data to obtain the real-time operation status of the chemical sub-region; Obtain the historical risk record data of each chemical sub-region within a preset time period; Extract the historical risk categories, the corresponding number of risk occurrences, and the risk severity levels of each chemical sub-region from the historical risk record data, and input them into a pre-established risk assessment model to obtain the estimated risk status of the corresponding chemical sub-region; When there is a chemical sub-region with a real-time operation status of a dangerous state, extract all feature vectors adapted to the current chemical sub-region being in a dangerous state as the key feature vectors for output; When there is no chemical sub-region with a real-time operation status of a dangerous state, label the chemical sub-regions with the same state level of the estimated risk status and the real-time operation status as the first region, label the chemical sub-regions with the state level of the estimated risk status greater than the state level of the real-time operation status as the second region, and label the chemical sub-regions with the state level of the estimated risk status less than the state level of the real-time operation status as the third region; Output all feature vectors adapted to the real-time operation status of the current first region as the first key feature vectors; Obtain the number of the first state levels by which the corresponding estimated risk status of the second region exceeds the real-time operation status; Determine the enhanced feature vectors corresponding to various historical risk categories respectively with the risk occurrence frequencies of various historical risk categories existing in the second region and the number of the first state levels as matching conditions; Screen out the feature vectors associated with various historical risk categories existing in the second region from all feature vectors adapted to the real-time operation status of the second region, and after replacing them with the corresponding enhanced feature vectors, output them together with all other non-replaced feature vectors as the key feature vectors; Obtain the number of the second state levels by which the corresponding estimated risk status of the third region is less than the real-time operation status; Determine the mitigation feature vectors corresponding to various historical risk categories respectively with the risk occurrence frequencies of various historical risk categories existing in the third region and the number of the second state levels as matching conditions; Screen out the feature vectors associated with various historical risk categories existing in the third region from all feature vectors adapted to the real-time operation status of the third region, and after replacing them with the corresponding mitigation feature vectors, output them together with all other non-replaced feature vectors as the key feature vectors; Summarize all key feature vectors and output them as feature vectors useful for the safety warning of chemical industrial parks.

2. The chemical industrial park safety early warning system based on the multi-modal large model according to claim 1, characterized in that, Collect multi-modal data of chemical industrial parks, including: Based on intelligent acquisition devices, monitor the temperature, humidity, pressure and gas concentration in the chemical industrial park in real time, and collect the environmental data of the chemical industrial park; Based on intelligent acquisition devices, monitor the equipment vibration, current and voltage in the chemical industrial park in real time, and collect the equipment data of the chemical industrial park; Based on intelligent acquisition devices, monitor the personnel location and helmet wearing situation in the chemical industrial park in real time, and collect the personnel data of the chemical industrial park; Based on intelligent acquisition devices, monitor the video, image and sound in the chemical industrial park in real time, and collect the visual and auditory data of the chemical industrial park; Determine the multi-modal data of the chemical industrial park according to the environmental data, equipment data, personnel data, visual and auditory data of the chemical industrial park.

3. The chemical industrial park safety early warning system based on the multi-modal large model according to claim 1, characterized in that, Preprocess the collected multi-modal data of the chemical industrial park, including: Clean the multi-modal data of the chemical industrial park, and remove the noise and duplicate data that are useless for the safety warning of the chemical industrial park in the multi-modal data of the chemical industrial park; Check the multi-modal data of the chemical industrial park, judge whether there are missing values and outliers in the multi-modal data of the chemical industrial park, and process the missing values and outliers in the multi-modal data of the chemical industrial park; Judge the missing values and outliers in the multi-modal data of the chemical industrial park, and judge whether the missing values and outliers in the multi-modal data of the chemical industrial park are useful for the safety warning of the chemical industrial park; If it is useful, fill in the missing values and replace the outliers in the multi-modal data of the chemical industrial park. If it is useless, remove the missing values and outliers in the multi-modal data of the chemical industrial park.

4. The chemical industrial park safety early warning system based on a multimodal large model according to claim 1, characterized in that Preprocessing the collected multi-modal data of the chemical industrial park also includes: Normalize the multi-modal data of the chemical industrial park, convert the multi-modal data of the chemical industrial park into a unified data format, remove the dimensional differences in the multi-modal data of the chemical industrial park, and determine the standardized multi-modal data of the chemical industrial park; Integrate the multi-modal data of the chemical industrial park, integrate the multi-modal data of the chemical industrial park into a unified data view, and perform data integrity verification on the integrated multi-modal data of the chemical industrial park. After the data integrity verification is qualified, safely store the integrated multi-modal data of the chemical industrial park; Perform modal alignment and feature extraction on the multi-modal data of the chemical industrial park, extract the feature vectors useful for the safety warning of the chemical industrial park from the multi-modal data of the chemical industrial park, and perform weighted fusion on the extracted feature vectors to determine the characteristic data of the chemical industrial park.

5. The chemical industrial park safety warning system based on the multi-modal large model according to claim 4, characterized in that, Also include: Real-time monitor the changes in the operating status and risk status of each chemical sub-region in the chemical industrial park, and correspondingly obtain the operating change data and risk change data; According to the operating change data, determine the number of the first status of the chemical sub-regions whose real-time operating status remains in the dangerous state per unit time within the set sliding window of the chemical industrial park, and the number of the first status times when the real-time operating status changes from the warning state to the dangerous state; Use the number of the first status and the number of the first status times to construct an operating danger matrix; Determine the number of second states of chemical sub-regions in the chemical industrial park where the real-time operating state remains in a safe state per unit time within the set sliding window, and the number of times of the second state when the real-time operating state changes from the warning state to the safe state; Construct an operation safety matrix using the number of second states and the number of times of the second state; By mapping the operation risk matrix and the operation safety matrix using the first preset adjustment analysis strategy, determine the first adjustment direction and the first adjustment amount for the preset time period; According to the risk change data, determine the number of first risks of chemical sub-regions in the chemical industrial park where the estimated risk state remains in a high-risk state per unit time within the set sliding window, and the number of first risks when the estimated risk state changes from a medium-risk state to a high-risk state; Construct a high-risk change matrix using the number of first risks and the number of times of the first risk; Determine the number of second risks of chemical sub-regions in the chemical industrial park where the estimated risk state remains in a low-risk state per unit time within the set sliding window, and the number of second risks when the estimated risk state changes from a medium-risk state to a low-risk state; Construct a low-risk change matrix using the number of second risks and the number of times of the second risk; By mapping the low-risk change matrix and the high-risk change matrix using the second preset adjustment analysis strategy, determine the second adjustment direction and the second adjustment amount for the preset time period; By combining the first adjustment direction, the first adjustment amount, the second adjustment direction, and the second adjustment amount with the operation influence weight and the risk influence weight to obtain a comprehensive adjustment amount, perform an adjustment process on the preset time period.

6. The chemical industrial park safety early warning system based on the multi-modal large model according to claim 4, characterized in that Perform data integrity verification on the integrated multi-modal data of the chemical industrial park, including: Conduct a comparative analysis of the integrated multi-modal data of the chemical industrial park and the multi-modal data of the chemical industrial park before integration, determine whether there are omissions in the integrated multi-modal data of the chemical industrial park, and verify the data integrity of the integrated multi-modal data of the chemical industrial park; When the integrated multi-modal data of the chemical industrial park is the same as the multi-modal data of the chemical industrial park before integration, there are no omissions in the integrated multi-modal data of the chemical industrial park, and the data integrity verification is qualified; When the integrated multi-modal data of the chemical industrial park is different from the multi-modal data of the chemical industrial park before integration, there are omissions in the integrated multi-modal data of the chemical industrial park, and the data integrity verification is unqualified. At this time, check the missing data and make supplementary recordings of the missing data to make the integrated multi-modal data of the chemical industrial park the same as the multi-modal data of the chemical industrial park before integration and make the data integrity verification qualified.

7. The chemical industrial park safety early warning system based on the multi-modal large model according to claim 4, characterized in that, Identify potential safety hazards in the chemical industrial park in advance, including: Establish a large chemical industrial park risk identification model according to the safety warning requirements of the chemical industrial park; Deploy the large chemical industrial park risk identification model and deploy the large chemical industrial park risk identification model in the actual chemical industrial park risk identification environment; Input the chemical industrial park feature data into the large chemical industrial park risk identification model, analyze the chemical industrial park feature data according to the large chemical industrial park risk identification model, identify the chemical industrial park risks, identify potential safety hazards in the chemical industrial park in advance, and determine the chemical industrial park risk identification results; Establish a large chemical industrial park risk identification model, including: Collect the historical data of the chemical industrial park, and divide the collected historical data of the chemical industrial park into a training set and a test set; Based on deep learning technology, use the training set to train the deep learning model, so that the deep learning model can autonomously learn the risk identification behavior of the chemical industrial park from the training set, and identify potential safety hazards in the chemical industrial park in advance, and determine the large model for risk identification of the chemical industrial park based on deep learning; Use the test set to test the large model for risk identification of the chemical industrial park based on deep learning, evaluate the performance of the large model for risk identification of the chemical industrial park based on deep learning, and judge whether the large model for risk identification of the chemical industrial park based on deep learning can achieve the expected effect of identifying potential safety hazards in the chemical industrial park in advance; When the large model for risk identification of the chemical industrial park based on deep learning cannot achieve the expected effect of identifying potential safety hazards in the chemical industrial park in advance, adjust and optimize the parameters of the large model for risk identification of the chemical industrial park based on deep learning until the large model for risk identification of the chemical industrial park based on deep learning can achieve the expected effect of identifying potential safety hazards in the chemical industrial park in advance, and then determine the best large model for risk identification of the chemical industrial park.

8. The chemical industrial park safety warning system based on the multi-modal large model according to claim 7, characterized in that, It also includes: Use the test set to test the large model for risk identification of the chemical industrial park based on deep learning to obtain the identification test results; Based on the identification test results, evaluate the performance of the large model for risk identification of the chemical industrial park based on deep learning to obtain the model performance evaluation score; According to the model performance evaluation score, judge whether the large model for risk identification of the chemical industrial park based on deep learning can achieve the expected effect of identifying potential safety hazards in the chemical industrial park in advance.

9. The chemical industrial park safety warning system based on the multi-modal large model according to claim 8, characterized in that, Conduct safety early warning management for potential safety hazards in the chemical industrial park, including: According to the risk identification results of the chemical industrial park, conduct timely safety early warning for potential safety hazards in the chemical industrial park, remind the management personnel of the chemical industrial park to pay attention to safety hazards, and intelligently guide the management personnel of the chemical industrial park to eliminate safety hazards, so as to form a closed-loop management of real-time monitoring and safety early warning in the chemical industrial park.

Images