Multi-scene limited space operation safety control method based on AI algorithm

Abstract

The invention relates to the technical field of safety management, and particularly discloses a multi-scene limited space operation safety control method based on an AI algorithm, and the method comprises the steps: carrying out the multi-modal collection of limited space physical environment information and operator information, carrying out the risk feature extraction of a limited space operation safety management original data package, and carrying out the risk feature extraction; the method comprises the following steps: constructing a limited space operation risk accident model, outputting a limited space operation risk probability, a real-time risk level and a dynamic early warning instruction, predicting a ventilation equipment load rate and a first-aid kit demand quantity based on an LSTM model, determining a limited space operation resource allocation scheme, and matching historical accident cases by using an Euclidean distance. A limited space operation rescue instruction is automatically generated, and the limited space operation rescue instruction is completed through multi-agent cooperative control. The problems that comprehensive and real-time monitoring is difficult to achieve in traditional limited space operation safety management, the efficiency is low, the flexibility is low, the risk missing report probability is high, and abnormal conditions cannot be tracked in time are solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of safety management, in particular to a multi-scene limited space operation safety control method based on an AI algorithm. BACKGROUND

[0002] Limited spaces have characteristics such as narrow space and poor ventilation, and are prone to accumulate toxic and harmful, flammable and explosive substances and insufficient oxygen content. With the continuous development of artificial intelligence technology, AI algorithms have shown great capabilities in image recognition, data analysis, risk early warning, etc. Using AI algorithms can analyze and process videos, images, environmental data, etc. of limited space operation sites in real time, timely detect unsafe behaviors of operation personnel, dangerous factors in the environment and abnormal states of equipment, etc., and provide new technical means and ideas for limited space operation safety management.

[0003] Nowadays, there are still some deficiencies in the research on limited space operation safety management, which are embodied in that the traditional limited space operation safety management mainly relies on manual supervision, but manual supervision has many limitations. Manual supervision is difficult to achieve comprehensive and real-time monitoring, and it is difficult to effectively monitor some hidden areas or complex environments in the operation site. Manual video monitoring has low efficiency, low flexibility, high risk omission probability, and delayed tracking of abnormal situations, etc., which cannot meet the high requirements of limited space operation safety management. SUMMARY

[0004] In view of the deficiencies of the prior art, the present application provides a multi-scene limited space operation safety control method based on an AI algorithm, which can effectively solve the problems involved in the background art.

[0005] Therefore, the present application provides a multi-scene limited space operation safety control method based on an AI algorithm, which comprises the following steps:

[0006] Multi-modal collection of limited space physical environment information and operation personnel information is performed, and a limited space operation safety management original data package is output;

[0007] Risk feature extraction is performed on the limited space operation safety management original data package, and a limited space operation risk feature parameter is output;

[0008] An a priori risk knowledge mapping atlas stored in a database is acquired, and a limited space operation risk accident model is constructed in combination with the limited space operation risk feature parameter, and a limited space operation risk probability is output;

[0009] Based on the limited space operation risk probability, a real-time risk level is output and a corresponding dynamic early warning instruction is set;

[0010] Based on the real-time risk level, the corresponding resource consumption and risk trend are obtained, and the real-time operation progress and external weather data are obtained. Based on the LSTM model, the ventilation equipment load rate and the first aid kit demand are predicted, and the limited space operation resource allocation scheme is determined.

[0011] Based on the risk characteristic parameters of limited space operation, the Euclidean distance is used to match historical accident cases, and the limited space operation rescue instruction is automatically generated. The multi-agent cooperates to rescue the limited space operation based on the limited space operation rescue instruction, and outputs the limited space operation resource allocation scheme, the real-time risk level and the corresponding dynamic warning instruction.

[0012] Optionally, the limited space physical environment information and operation personnel information are collected in multiple modes, and the limited space operation safety management original data package is output, including:

[0013] The structure diagram, equipment layout and ventilation port position of the limited space are collected. Key sensor clusters are deployed at the entrance and exit of the limited space, ventilation dead angle and equipment deployment position to collect the limited space Concentration, limited space Concentration, and time sequence of limited space temperature and humidity, wherein the key sensor cluster includes Concentration sensor, Concentration sensor, temperature sensor and humidity sensor;

[0014] The laser radar and binocular camera are carried by the inspection robot. The simultaneous localization and mapping technology is used to construct a 3D point cloud map with centimeter-level precision in real time, and the RGB image and thermal imaging video stream with spatial coordinates are collected. The thermal imaging temperature matrix of the limited space is extracted based on the thermal imaging video stream;

[0015] The smart bracelet worn by the operation personnel uploads the operation personnel heart rate and operation personnel position data based on Bluetooth protocol. The federated learning technology is used to complete the heart rate anomaly detection locally to obtain the operation personnel heart rate anomaly frequency. Based on the operation personnel position data and the dangerous area marked on the limited space structure diagram, it is judged whether the operation personnel is in the dangerous area: if the operation personnel is not in the dangerous area, the operation personnel position is marked as 0; if the operation personnel is in the dangerous area, the operation personnel position is marked as 1.

[0016] The limited space operation safety management original data package is output, including the limited space Concentration, limited space Concentration, time sequence of limited space temperature and humidity, RGB image with spatial coordinates, thermal imaging temperature matrix of limited space, operation personnel heart rate anomaly frequency and operation personnel position mark.

[0017] Optionally: the risk feature extraction is performed on the limited space operation safety management original data packet, and a limited space operation risk feature parameter is output, including:

[0018] The limited space is identified using a DBSCAN clustering algorithm The concentration and the limited space The abnormal cluster of the concentration, the duration of the abnormal cluster, and the peak concentration are extracted;

[0019] Based on the RGB image with space coordinates, a crack classification model is trained using MobileNetV3+SE attention module, and the number of cracks in the limited space is output, wherein the crack classification model is transfer learning+Fine-tuning;

[0020] Based on the limited space thermal imaging temperature matrix, the high-temperature dangerous area is marked by image registration algorithm and CFD simulation gas diffusion path superposition, and the area of the high-temperature dangerous area is extracted;

[0021] The limited space operation risk feature parameter is output, including the limited space environment risk feature parameter, the limited space visual risk feature parameter, and the limited space personnel risk feature parameter;

[0022] Among them:

[0023] The limited space environment risk feature parameter includes the limited space The concentration, the limited space The duration of the abnormal cluster of the concentration and the peak concentration;

[0024] The limited space visual risk feature parameter includes the number of cracks in the limited space and the area of the high-temperature dangerous area;

[0025] The limited space personnel risk feature parameter includes the abnormal frequency of the heart rate of the operator and the position mark of the operator.

[0026] Optionally: the prior risk knowledge mapping atlas stored in the database is acquired, and the limited space operation risk accident model is constructed by combining the limited space operation risk feature parameter, and the limited space operation risk probability is output, including:

[0027] The prior risk knowledge mapping atlas stored in the database is acquired;

[0028] Based on the limited space operation risk feature parameter, the risk probability corresponding to the limited space operation risk feature parameter is matched from the prior risk knowledge mapping atlas, specifically including the limited space The abnormal cluster duration risk probability of the concentration , the limited space The abnormal cluster duration risk probability of the concentration , the limited space Abnormal concentration cluster peak concentration risk probability , limited space Abnormal concentration cluster peak concentration risk probability , limited space crack number risk probability , high temperature dangerous area risk probability , abnormal heart rate frequency risk probability of operating personnel , and operating personnel position marking risk probability ;

[0029] A limited space operation risk accident model is constructed, and based on the risk probability corresponding to the risk characteristic parameter of the limited space operation, a limited space operation risk factor is obtained.

[0030] The limited space operation risk factor-limited space operation risk probability mapping set stored in the database is obtained, and based on the current limited space operation risk factor, the matched limited space operation risk probability is determined.

[0031] The limited space operation risk probability is output.

[0032] Optionally, the expression of the limited space operation risk accident model is:

[0033] ;

[0034] In the formula, is the limited space operation risk factor, and e is a natural constant.

[0035] Optionally, the real-time risk level and dynamic warning instruction are output based on the limited space operation risk probability, including:

[0036] If the limited space operation risk probability is less than the first probability threshold value stored in the database, the real-time risk level of the limited space is low risk, marked as level 1, and the dynamic warning instruction is to record a log and display a green safety mark on the LED screen.

[0037] If the limited space operation risk probability is not less than the first probability threshold value stored in the database and less than the second probability threshold value stored in the database, the real-time risk level of the limited space is medium risk, marked as level 2, and the dynamic warning instruction is to sound and light alarm, the buzzer is 85 dB, the red light flashes 3 times per second, and the smart bracelet worn by the operating personnel vibrates to prompt evacuation.

[0038] If the limited space operation risk probability is not less than the second probability threshold value stored in the database, the real-time risk level of the limited space is high risk, marked as level 3, and the dynamic warning instruction is to trigger an emergency plan, turn off the power and start the exhaust, and the smart bracelet worn by the operating personnel emits an alarm sound to prompt evacuation.

[0039] Optionally, the corresponding resource consumption and risk trend are obtained based on the real-time risk level, including:

[0040] Based on the real-time risk level, the historical resource consumption and risk trend corresponding to the same historical risk level as the real-time risk level stored in the database are obtained;

[0041] The average of the historical resource consumption and risk trend corresponding to the same historical risk level as the real-time risk level stored in the database is recorded as the resource consumption and risk trend corresponding to the real-time risk level.

[0042] Optionally, the real-time operation progress and external weather data are obtained, the ventilation equipment load rate and first aid kit demand are predicted based on the LSTM model, and the limited space operation resource allocation scheme is determined, including:

[0043] The real-time operation progress and external weather data are obtained, wherein the external weather data includes external environment temperature, external environment humidity and external environment wind speed;

[0044] The real-time risk level corresponding resource consumption, the real-time risk level corresponding risk trend, the real-time operation progress and the external weather data are input into the trained LSTM model, and the ventilation equipment load rate and the first aid kit demand are output;

[0045] The number of limited space operation personnel is obtained, the number of limited space operation personnel, the area of high temperature dangerous area, the ventilation equipment load rate and the first aid kit demand are stored as specified labels, the specified label-limited space operation resource allocation scheme mapping table stored in the database is obtained, and the matched limited space operation resource allocation scheme is determined based on the current specified label.

[0046] Optionally, the limited space operation rescue instruction is automatically generated based on the limited space operation risk feature parameter and the Euclidean distance matching historical accident case, including:

[0047] The limited space operation risk feature parameters of the historical accident case are obtained;

[0048] Based on the limited space operation risk feature parameters, the most similar historical accident case is matched using the Euclidean distance;

[0049] The limited space operation rescue instruction of the most similar historical accident case is obtained as the current limited space operation rescue instruction.

[0050] Optionally, the multi-agent cooperates to carry out limited space operation rescue based on the limited space operation rescue instruction, including:

[0051] The current limited space operation rescue instruction is divided into sub-tasks, including robot carrying rescue materials and unmanned aerial vehicle building communication relay;

[0052] determine the starting point and the ending point of the robot carrying the rescue materials in the current limited space operation rescue instruction;

[0053] Based on the limited space structure diagram, equipment layout, air vent position and operation personnel position data, the robot is path planned using A* algorithm, the optimal path of the robot from the current position to the starting point carrying the rescue materials to the ending point is planned, so that the path is the shortest and obstacles are avoided;

[0054] The unmanned aerial vehicle is used to build a communication relay, and the robot is controlled to carry the rescue materials based on the optimal path, so that the robot sends a just-in-place signal to the unmanned aerial vehicle when it reaches the starting point and the ending point of carrying the rescue materials, thereby achieving collaborative control of the limited space operation rescue instruction.

[0055] From the above technical solutions, the present application has the following advantages:

[0056] The present application provides a multi-scenario limited space operation safety management method based on AI algorithm, which can break through the limitations of single sensor and avoid the false reporting problem of traditional single-point monitoring through multi-modal data collection of physical environment and personnel information. Based on the extraction of risk features from raw data, combined with the prior knowledge graph to build an accident model, the risk probability calculation accuracy can be improved. The LSTM model is used to predict the ventilation equipment load rate and the demand for first aid kits, so as to realize the dynamic allocation of resources. By matching the historical accident cases through Euclidean distance, the rescue instruction can be automatically generated, and the standardization of the rescue process can be realized through multi-agent collaborative control. BRIEF DESCRIPTION OF DRAWINGS

[0057] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed in the embodiments or prior art description will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present application, and those skilled in the art can obtain other drawings according to these drawings without creative labor.

[0058] Figure 1 The flowchart of the multi-scenario limited space operation safety control method based on AI algorithm provided by the embodiment of the present application is shown.

[0059] Figure 2 The step flowchart of outputting the limited space operation safety management raw data packet provided by the embodiment of the present application is shown. DETAILED DESCRIPTION

[0060] In order to make the application purposes, features and advantages of the present application more obvious and easy to understand, the technical solutions in the embodiments of the present application will be described clearly and completely below in conjunction with the drawings of the embodiments of the present application. Obviously, the embodiments described below are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0061] Please refer to Figure 1 The AI algorithm-based multi-scene limited space operation safety control method provided in the embodiments of the present application comprises:

[0062] Step 101, multi-modal collection of limited space physical environment information and operation personnel information is performed, and a limited space operation safety management original data packet is output.

[0063] As Figure 2 shown, in one embodiment, step 101 comprises:

[0064] Step 1011, the structure diagram, equipment layout and ventilation port position of the limited space are collected, key sensor clusters are deployed at the entrances and exits of the limited space, ventilation dead angles and equipment deployment positions, and time series sequences of the limited space concentration, the limited space concentration, and the limited space temperature and humidity are collected, wherein the key sensor clusters comprise concentration sensors, concentration sensors, temperature sensors and humidity sensors.

[0065] Step 1012, a laser radar and a binocular camera are carried by a patrol robot, a simultaneous localization and mapping technology is used to construct a 3D point cloud map with centimeter-level precision in real time, and an RGB image and a thermal imaging video stream with spatial coordinates are collected, a limited space thermal imaging temperature matrix is extracted based on the thermal imaging video stream.

[0066] Step 1013, the heart rate of the operation personnel and the position data of the operation personnel are uploaded based on the smart bracelet worn by the operation personnel using a Bluetooth protocol, heart rate anomaly detection is completed locally using a federated learning technology, the frequency of heart rate anomalies of the operation personnel is obtained, and it is judged whether the operation personnel is in a dangerous area based on the position data of the operation personnel and the dangerous area marked on the structure diagram of the limited space: if the operation personnel is not in the dangerous area, the position of the operation personnel is marked as 0, and if the operation personnel is in the dangerous area, the position of the operation personnel is marked as 1.

[0067] Step 1014, a limited space operation safety management original data packet is output, comprising the limited space concentration, the limited space Concentration, limited space humidity time sequence, RGB image with space coordinates, limited space thermal imaging temperature matrix, abnormal frequency of heart rate of operating personnel, and position mark of operating personnel.

[0068] It can be understood that the concentration of hydrogen sulfide in the limited space is collected Concentration of hydrogen sulfide, Concentration of oxygen, humidity time sequence, can fully grasp the gas composition and humidity condition in the space. For example, in the reaction kettle and other limited spaces of chemical enterprises, Concentration is too high and may cause poisoning risk, Concentration anomaly is related to suffocation hazard, and humidity change will affect the stability of equipment and materials. Timely sensing of these information can effectively prevent accidents. Through the patrol robot carrying laser radar and binocular camera, a 3D point cloud map with centimeter level precision is constructed, and RGB image with space coordinates is obtained. This can accurately locate equipment, passageway and dangerous area in limited space with complex structure, such as underground pipe gallery and mine tunnel. Operating personnel wear smart bracelet to upload heart rate data and detect abnormal frequency. Personnel body abnormalities caused by physical exhaustion and environmental discomfort can be found in time. Whether the personnel are in dangerous area is judged by combining personnel position data and dangerous area mark. For example, in limited space with flammable and explosive substances, it can be known in real time whether the personnel are close to dangerous area, so as to avoid the danger caused by personnel mistakenly entering, and improve the operation safety.

[0069] Note: RGB image is a common color image representation. RGB represents red (Red), green (Green) and blue (Blue) three basic colors, which are called three primary colors. In RGB image, the color of each pixel point is represented by the combination of the values of red, green and blue three channels. Each channel is usually stored by one byte (8 bits), and the value range is 0-255. By adjusting different value combinations of the three channels, various colors can be mixed.

[0070] Step 102, risk feature extraction is performed on the limited space operation safety management original data packet, and limited space operation risk feature parameters are output.

[0071] In one embodiment, step 102 includes:

[0072] The DBSCAN clustering algorithm is used to identify the limited space Concentration and limited space Anomaly clusters of concentration are identified, and their duration and peak concentration are extracted. A crack classification model is trained using MobileNetV3+SE attention module based on RGB images with spatial coordinates, outputting the number of cracks in the confined space. The crack classification model employs transfer learning and fine-tuning. Based on the temperature matrix of confined space thermal imaging, high-temperature hazard areas are marked and their areas are extracted by overlaying the image registration algorithm with CFD-simulated gas diffusion paths. Confined space operation risk characteristic parameters are output, including confined space environmental risk characteristic parameters, confined space visual risk characteristic parameters, and confined space personnel risk characteristic parameters. The confined space environmental risk characteristic parameters include the confined space... Concentration, confined space The duration and peak concentration of abnormal clusters; visual risk characteristic parameters of confined spaces include the number of cracks in the confined space and the area of ​​high-temperature hazard zones; personnel risk characteristic parameters of confined spaces include the frequency of abnormal heart rates of workers and the location markers of workers.

[0073] Understandably, using the DBSCAN clustering algorithm to identify clusters of abnormal concentrations can accurately pinpoint the time periods and peak values ​​of gas concentration anomalies. For example, in anaerobic tanks of wastewater treatment plants, it can promptly detect abnormalities caused by microbial activity. During periods of sudden concentration increases, early warning of poisoning risks is provided, offering greater sensitivity and timeliness compared to conventional timed detection. A crack classification model trained with MobileNetV3+SE attention modules can accurately count the number of cracks in confined spaces. In spaces such as underground tunnels and mine shafts, an increase in cracks may indicate structural safety hazards; quantifying the number of cracks facilitates risk assessment and provides a basis for structural reinforcement and other maintenance measures. Continuously monitoring the frequency and location of abnormal heart rates among workers can reveal risks to their physical and behavioral state. In confined space work scenarios involving high temperatures and high physical exertion, an increased frequency of abnormal heart rates in personnel in dangerous positions can detect discomfort and potential dangers in advance, preventing accidents. Risk characteristic parameters are categorized into environmental, visual, and personnel risks, making risk classification clear. Different risk categories correspond to different management strategies; for example, environmental risks focus on gas monitoring and ventilation control, visual risks focus on structural maintenance, and personnel risks focus on health protection and behavioral norms, contributing to targeted and refined management.

[0074] Note:

[0075] MobileNetV3+SE is a model architecture that combines the lightweight convolutional neural network MobileNetV3 with the SE (Squeeze-and-Excitation) attention module.

[0076] The combination of "transfer learning + fine-tuning" first uses the idea of transfer learning to utilize the knowledge of the pre-trained model, and then through the method of fine-tuning, the model achieves better performance on the target task.

[0077] CFD, which stands for Computational Fluid Dynamics, can simulate the airflow distribution and pollutant diffusion in a confined space, and is used in the safety control of confined space operations.

[0078] DBSCAN clustering algorithm is a density-based spatial clustering algorithm that divides areas with sufficient density into clusters and discovers clusters of arbitrary shapes in noisy data. In the context of safety control in confined space operations, DBSCAN clustering algorithm can perform clustering analysis on various risk data collected, such as environmental parameters and personnel behavior data. By setting appropriate density thresholds and neighborhood radii, the algorithm can identify dense and sparse regions in the data and group similar risk data points into the same cluster. This helps quickly identify patterns of risk aggregation and outliers, such as clustering areas with similar environmental parameters and potential risks, providing a basis for precise risk identification and targeted management.

[0079] Step 103, acquire the prior risk knowledge mapping atlas stored in the database, combine the risk feature parameters of confined space operations, construct a risk accident model of confined space operations, and output the risk probability of confined space operations.

[0080] In one embodiment, step 103 includes:

[0081] Acquire the prior risk knowledge mapping atlas stored in the database; based on the risk feature parameters of confined space operations, match the corresponding risk probability corresponding to the risk feature parameters of confined space operations from the prior risk knowledge mapping atlas, specifically including the risk probability of the following risk feature parameters of confined space operations: Risk probability of duration of concentration anomaly cluster Risk probability of number of cracks in confined space Risk probability of duration of concentration anomaly cluster Risk probability of peak concentration of concentration anomaly cluster in confined space Risk probability of peak concentration of concentration anomaly cluster in confined space Risk probability of number of cracks in confined space Risk probability of area of high-temperature danger zone Risk probability of frequency of abnormal heart rate of operating personnel Risk probability of location marker of operating personnel Risk probability of location marker of operating personnel Risk probability of location marker of operating personnel ; a limited space operation risk accident model is constructed, a limited space operation risk factor is obtained based on a risk probability corresponding to a risk characteristic parameter of the limited space operation, a limited space operation risk factor-limited space operation risk probability mapping set stored in the database is acquired, a matched limited space operation risk probability is determined based on the current limited space operation risk factor, and the limited space operation risk probability is output.

[0082] The expression of the limited space operation risk accident model is as follows:

[0083] ;

[0084] In the formula, e is a natural constant.

[0085] It can be understood that the limited space operation risk is converted into a specific numerical value and a risk probability by means of a definite mathematical model and comprehensive risk probabilities corresponding to various risk characteristic parameters. The fuzziness of risk judgment by experience or qualitatively is changed, the risk degree is accurately quantified, and the risk is conveniently managed in stages. The risk is comprehensively evaluated by covering various risk factors such as environment (gas concentration), structure (crack number), and personnel (abnormal heart rate and position). The one-sidedness of single-factor evaluation is avoided, and the overall risk level is comprehensively obtained, which is more in line with the actual situation.

[0086] In step 104, a real-time risk level is output based on the limited space operation risk probability, and a corresponding dynamic warning instruction is set.

[0087] In one embodiment, step 104 includes:

[0088] If the limited space operation risk probability is less than a first probability threshold value stored in the database, the real-time risk level of the limited space is low risk, which is marked as level 1, and the dynamic warning instruction is to record a log, and the LED screen displays a green safety mark. If the limited space operation risk probability is not less than the first probability threshold value stored in the database and less than a second probability threshold value stored in the database, the real-time risk level of the limited space is medium risk, which is marked as level 2, and the dynamic warning instruction is to sound and light alarm, the buzzer is 85 dB, the red light flashes 3 times per second, and the smart bracelet worn by the operator vibrates to prompt evacuation. If the limited space operation risk probability is not less than the second probability threshold value stored in the database, the real-time risk level of the limited space is high risk, which is marked as level 3, and the dynamic warning instruction is to trigger an emergency plan, to shut down the power supply and start the exhaust, and to sound an alarm to prompt evacuation.

[0089] ​It can be understood that the risk is divided into three levels of low, medium and high, and is presented in a simple and clear numerical mark (1st, 2nd and 3rd). The operating personnel and the management personnel can quickly understand the degree of risk faced by the current limited space operation. According to the risk probability, the risk level and the early warning instruction are output in real time, so that the risk change can be captured in time. When the risk factors such as gas concentration in the limited space dynamically change, the system quickly reacts, such as immediately issuing an audible and visual alarm when the low risk is upgraded to a medium risk. Compared with the artificial regular inspection to find the risk, the system is more timely, and the risk window period is reduced. Different early warning instructions are set for different risk levels, so that the early warning is more targeted. The low risk only records the log and displays the safety mark, and does not cause excessive interference; the medium risk reminds evacuation through sound, light and bracelet vibration; the high risk triggers the emergency plan and forces power off and starts exhaust, reasonably arranges the response measures according to the risk degree, and avoids insufficient or excessive response.

[0090] Step 105, based on the real-time risk level, acquiring corresponding resource consumption and risk trend, and acquiring real-time operation progress and external weather data, predicting the ventilation equipment load rate and the first-aid kit demand amount based on the LSTM model, and determining the limited space operation resource allocation scheme.

[0091] In one embodiment, step 105 includes:

[0092] Based on the real-time risk level, the historical resource consumption and risk trend corresponding to the historical risk level of the same level as the real-time risk level stored in the database are acquired; the mean value of the historical resource consumption and risk trend corresponding to the historical risk level of the same level as the real-time risk level stored in the database is recorded as the resource consumption and risk trend corresponding to the real-time risk level.

[0093] Acquiring real-time operation progress and external weather data, wherein the external weather data includes external environment temperature, external environment humidity and external environment wind speed; inputting the real-time risk level corresponding resource consumption, the real-time risk level corresponding risk trend, the real-time operation progress and the external weather data into the trained LSTM model to output the ventilation equipment load rate and the first-aid kit demand amount; acquiring the number of limited space operation personnel, storing the number of limited space operation personnel, the area of high-temperature dangerous area, the ventilation equipment load rate and the first-aid kit demand amount as a specified label, acquiring the specified label-limited space operation resource allocation scheme mapping table stored in the database, and determining the matched limited space operation resource allocation scheme based on the current specified label.

[0094] It can be understood that by matching the historical resource consumption and trend mean of the same risk level, the system can quickly anchor the basic resource demand. For example, if the current risk level is "high", the system automatically retrieves the data of the use time of the ventilation equipment, the consumption speed of first aid materials and other data under the historical high-risk scene, avoiding insufficient or excessive resource allocation. The LSTM model can capture the nonlinear characteristics of risk changes by learning the time series patterns in historical data. For example, in high-temperature weather, the model can predict that the gas in a limited space will expand due to the increase of external temperature, so as to dynamically increase the load rate of the ventilation equipment, which is more in line with the actual demand than the traditional static configuration scheme. The work progress, weather data (temperature, humidity, wind speed) are included in the prediction to realize the whole-element collaborative decision-making. A "label-resource allocation scheme" mapping table is established to convert complex decision-making into a standardized process.

[0095] Note:

[0096] The LSTM model is a special type of recurrent neural network that can effectively handle long-term dependencies in historical data. For example, when analyzing the changes in gas concentration over time in a confined space, it can remember the concentration change patterns over a long period of time, even if there are some fluctuations in between, and accurately predict the future trend of gas concentration. Moreover, it can also integrate various dynamic factors in the work process, such as the running state of the equipment, the operation frequency of the workers, etc., for comprehensive analysis and prediction. When unexpected situations occur in the work environment, such as equipment failure or personnel operation error, the LSTM model can quickly adjust the risk assessment and resource allocation recommendations based on historical data and current conditions, providing more accurate and timely safety protection for confined space work.

[0097] Step 106, based on the risk characteristic parameters of confined space work, using Euclidean distance to match historical accident cases, automatically generating confined space work rescue instructions, so that multiple agents collaborate to rescue confined space work based on the confined space work rescue instructions, and output the confined space work resource allocation scheme, real-time risk level and corresponding dynamic warning instructions.

[0098] In one embodiment, step 106 includes:

[0099] Obtain the risk characteristic parameters of the historical accident cases; based on the risk characteristic parameters of the confined space work, use Euclidean distance to match the most similar historical accident cases; obtain the confined space work rescue instructions of the most similar historical accident cases as the current confined space work rescue instructions.

[0100] The current limited space operation rescue instruction is divided into sub-tasks, including robot carrying rescue materials and unmanned aerial vehicle building communication relay; the starting point and the ending point of the robot carrying rescue materials in the current limited space operation rescue instruction are determined; the A* algorithm is used to plan the path of the robot based on the limited space structure diagram, equipment layout, air vent position and operation personnel position data, the optimal path of the robot from the current position to the starting point carrying rescue materials to the ending point is planned, so that the path is the shortest and obstacles are avoided; the unmanned aerial vehicle builds communication relay, and controls the robot to carry rescue materials based on the optimal path, so that the robot sends a positioning signal to the unmanned aerial vehicle when it reaches the starting point and the ending point of carrying rescue materials, thereby cooperatively controlling the completion of the limited space operation rescue instruction.

[0101] And output the limited space operation resource allocation scheme, real-time risk level and corresponding dynamic early warning instruction.

[0102] It can be understood that the most similar historical accident case is matched by the Euclidean distance, and the system can quickly retrieve the verified effective rescue scheme. The automatically generated rescue instruction is based on historical successful experience, which avoids the subjectivity and uncertainty of manual decision. Through the A* algorithm to plan the optimal path, the robot can avoid obstacles (such as pipelines, accumulations), and quickly deliver first aid kits, respirators and other materials to the designated position. In the limited space (such as underground pipe gallery) with serious signal shielding, the unmanned aerial vehicle ensures real-time transmission of the rescue instruction by deploying relay nodes. The robot positioning signal triggers cooperative action, such as automatic loading when reaching the material point and sending a completion signal when reaching the ending point. If there is a new obstacle (such as collapse) in the path, the system can re-plan the path and instruct the unmanned aerial vehicle to adjust the relay position.

[0103] Note: A* algorithm is a kind of direct search method for solving the shortest path in static road network, which combines the advantages of best-first search and Dijkstra algorithm. In robot path planning, A* algorithm combines the characteristics of heuristic search and Dijkstra algorithm. It not only considers the actual cost from the starting point to the current node, but also estimates the cost from the current node to the target node. This estimated cost is usually calculated using a heuristic function, such as Euclidean distance or Manhattan distance. In this way, A* algorithm can purposefully move towards the target node during the search process, rather than blindly searching all possible paths like Dijkstra algorithm. In the limited space operation scene, this enables the robot to quickly and efficiently find the shortest path to the target location while effectively avoiding various obstacles, greatly improving the efficiency and safety of rescue material carrying.

[0104] The above merely illustrates and describes the structure of the present application, and those skilled in the art can make various modifications or supplements to the described specific embodiments or adopt similar ways to replace, as long as the modifications or supplements do not deviate from the structure of the present application or exceed the scope defined by the present claims, and should belong to the protection scope of the present application.

Claims

1. A method for safety control of multi-scene limited space operation based on an AI algorithm, characterized in that, The method comprises the following steps: carrying out multi-modal collection of physical environment information and worker information in a limited space, and outputting original data packets for limited space operation safety management; extracting risk characteristics from the original data packets for limited space operation safety management, and outputting limited space operation risk characteristic parameters; obtaining a priori risk knowledge mapping graph stored in a database, combining the limited space operation risk characteristic parameters, constructing a limited space operation risk accident model, and outputting a limited space operation risk probability; outputting a real-time risk level based on the limited space operation risk probability and setting a corresponding dynamic early warning instruction; based on the real-time risk level, obtaining corresponding resource consumption and risk trend, and obtaining real-time operation progress and external weather data, predicting ventilation equipment load rate and first aid kit demand based on an LSTM model, and determining a limited space operation resource allocation scheme; based on the limited space operation risk characteristic parameters, matching historical accident cases using the Euclidean distance, automatically generating a limited space operation rescue instruction, so that multiple intelligent agents cooperate to carry out limited space operation rescue based on the limited space operation rescue instruction, and outputting the limited space operation resource allocation scheme, the real-time risk level and the corresponding dynamic early warning instruction. 2.The AI algorithm-based multi-scene limited space operation safety control method according to claim 1, characterized in that, The multi-modal collection of physical environment information and worker information in a limited space, and outputting original data packets for limited space operation safety management, comprises: collecting a structure diagram of the limited space, a device layout, and a ventilation port position, deploying a key sensor cluster at an entrance and exit of the limited space, a ventilation dead angle, and a device deployment position of the limited space, collecting limited space concentration, limited space concentration, and a limited space temperature and humidity time sequence, wherein the key sensor cluster comprises a concentration sensor, a concentration sensor, a temperature sensor, and a humidity sensor; a laser radar and a binocular camera are carried on a patrol robot, a 3D point cloud map with centimeter-level precision is constructed in real time by using a simultaneous localization and mapping technology, and RGB images and thermal imaging video streams with spatial coordinates are collected, and a limited space thermal imaging temperature matrix is extracted based on the thermal imaging video stream; a smart bracelet worn by a worker is used to upload worker heart rate and worker position data by using a Bluetooth protocol, a federated learning technology is used to complete heart rate anomaly detection locally, worker heart rate anomaly frequency is obtained, and whether the worker is in a dangerous area is judged based on worker position data and a dangerous area marked on a limited space structure: if the worker is not in the dangerous area, the worker position is marked as 0, and if the worker is in the dangerous area, the worker position is marked as 1; Output limited space operation safety management raw data package, including limited space concentration, limited space concentration, limited space temperature and humidity time sequence, RGB image with space coordinates, limited space thermal imaging temperature matrix, operation personnel heart rate abnormal frequency, and operation personnel position mark. 3.The AI algorithm-based multi-scene limited space operation safety control method according to claim 2, characterized in that, The risk characteristic extraction from the original data packets for limited space operation safety management, and the output of the limited space operation risk characteristic parameters, comprises: Identifying limited spaces using DBSCAN clustering algorithm Concentration and limited space Abnormal clusters of concentration, extracting abnormal cluster duration and peak concentration; based on the RGB images with spatial coordinates, a crack classification model is trained by using a MobileNetV3+SE attention module, and the number of cracks in the limited space is output, wherein the crack classification model is transfer learning+Fine-tuning; based on the limited space thermal imaging temperature matrix, a high-temperature dangerous area is marked by superimposing the image registration algorithm and the CFD simulated gas diffusion path, and the area of the high-temperature dangerous area is extracted; the limited space operation risk characteristic parameters are output, including limited space environment risk characteristic parameters, limited space visual risk characteristic parameters and limited space personnel risk characteristic parameters; wherein: The limited space environment risk characteristic parameters include limited space concentration, limited space anomalous cluster duration and peak concentration of the limited space the limited space visual risk characteristic parameters include the number of cracks in the limited space and the area of the high-temperature dangerous area; the limited space personnel risk characteristic parameters include the worker heart rate anomaly frequency and the worker position mark. 4.The AI algorithm-based multi-scenario limited space operation safety control method according to claim 1, wherein, The prior risk knowledge mapping pattern stored in the database is acquired, a limited space operation risk accident model is constructed combining with the limited space operation risk characteristic parameters, and a limited space operation risk probability is output, including: Acquiring the prior risk knowledge mapping pattern stored in the database; Based on the limited space operation risk characteristic parameters, risk probabilities corresponding to the limited space operation risk characteristic parameters are matched from a prior risk knowledge mapping atlas, and specifically include limited space Concentration anomaly cluster duration risk probability , limited space Concentration anomaly cluster duration risk probability , limited space Concentration anomaly cluster peak concentration risk probability , limited space Concentration anomaly cluster peak concentration risk probability , limited space crack number risk probability High temperature dangerous area risk probability , operation personnel heart rate anomaly frequency risk probability , and operation personnel position marking risk probability ; Constructing a limited space operation risk accident model, and obtaining a limited space operation risk factor based on the risk probability corresponding to the limited space operation risk characteristic parameter; Acquiring the limited space operation risk factor-limited space operation risk probability mapping set stored in the database, and determining the matched limited space operation risk probability based on the current limited space operation risk factor; Outputting the limited space operation risk probability. 5.The AI algorithm-based multi-scene limited space operation safety control method according to claim 4, characterized in that, The expression of the limited space operation risk accident model is: ； wherein is the risk factor for finite space operations, and e is the natural constant. 6.The AI algorithm-based multi-scene limited space operation safety control method according to claim 1, wherein, The real-time risk level and the dynamic early warning instruction are output based on the limited space operation risk probability, including: If the limited space operation risk probability is less than the first probability threshold value stored in the database, the real-time risk level of the limited space is low risk, marked as level 1, and the dynamic early warning instruction is to record a log and display a green safety mark on the LED screen; If the limited space operation risk probability is not less than the first probability threshold value stored in the database and less than the second probability threshold value stored in the database, the real-time risk level of the limited space is medium risk, marked as level 2, and the dynamic early warning instruction is to sound and light alarm, the buzzer is 85 dB, the red light flashes 3 times per second, and the smart bracelet worn by the operation personnel vibrates to prompt evacuation; If the limited space operation risk probability is not less than the second probability threshold value stored in the database, the real-time risk level of the limited space is high risk, marked as level 3, and the dynamic early warning instruction is to trigger an emergency plan, turn off the power and start the exhaust, and the smart bracelet worn by the operation personnel emits an alarm sound to prompt evacuation. 7.The AI algorithm-based multi-scene limited space operation safety control method according to claim 1, wherein, Based on the real-time risk level, the corresponding resource consumption and risk trend are acquired, including: Based on the real-time risk level, the historical resource consumption and risk trend corresponding to the historical risk level identical to the real-time risk level stored in the database are acquired; The mean value of the historical resource consumption and risk trend corresponding to the historical risk level identical to the real-time risk level stored in the database is recorded as the resource consumption and risk trend corresponding to the real-time risk level. 8.The AI algorithm-based multi-scene limited space operation safety control method according to claim 7, characterized in that, Real-time operation progress and external weather data are acquired, the ventilation equipment load rate and the first aid kit demand are predicted based on the LSTM model, and the limited space operation resource allocation scheme is determined, including: Real-time operation progress and external weather data are acquired, wherein the external weather data includes external environment temperature, external environment humidity and external environment wind speed; The real-time risk level corresponding resource consumption, the real-time risk level corresponding risk trend, the real-time operation progress and the external weather data are input into the trained LSTM model, and the ventilation equipment load rate and the first aid kit demand are output; The number of limited space operation personnel is acquired, the number of limited space operation personnel, the high-temperature dangerous area area, the ventilation equipment load rate and the first aid kit demand are stored as specified tags, a specified tag-limited space operation resource allocation scheme mapping table stored in the database is acquired, and the matched limited space operation resource allocation scheme is determined based on the current specified tag. 9.The AI algorithm-based multi-scene limited space operation safety control method according to claim 1, wherein, The limited space operation risk feature parameter is used to match historical accident cases by using Euclidean distance to automatically generate a limited space operation rescue instruction, which includes: Obtaining the limited space operation risk feature parameter of the historical accident case; Based on the limited space operation risk feature parameter, the most similar historical accident case is matched by using Euclidean distance; Obtaining the limited space operation rescue instruction of the most similar historical accident case as the current limited space operation rescue instruction. 10.The AI algorithm-based multi-scene limited space operation safety control method according to claim 9, characterized in that, The multi-agent cooperates to carry out limited space operation rescue based on the limited space operation rescue instruction, which includes: Dividing the current limited space operation rescue instruction into sub-tasks, including robot carrying rescue materials and unmanned aerial vehicle building communication relay; Determining the starting point and the ending point of the robot carrying rescue materials in the current limited space operation rescue instruction; Based on the limited space structure diagram, the equipment layout, the ventilation opening position and the operation personnel position data, the A* algorithm is used to plan the path of the robot, to plan the optimal path of the robot from the current position to the starting point to carry the rescue materials to the ending point, so that the path is the shortest and avoids obstacles; Through the unmanned aerial vehicle building communication relay, and controlling the robot to carry the rescue materials based on the optimal path, the robot sends a just-in-place signal to the unmanned aerial vehicle when it reaches the starting point and the ending point of carrying the rescue materials, so as to cooperatively control the completion of the limited space operation rescue instruction.

Images