Monitoring method and system for safety production and storage of fireworks and crackers and storage medium

Abstract

The invention discloses a monitoring method and system for safety production and storage of fireworks and crackers and a storage medium. The method comprises the steps that production equipment operation state data, operator action standard data and warehouse image data are acquired; inputting the environment image information and the action image information of the operator into an image recognition model for danger recognition processing to obtain a dangerous state of the production equipment and a dangerous action of the operator; calculating the total equivalent of the warehouse based on the warehouse image data and the cargo warehouse-in and warehouse-out data; inputting the equipment state information, the dangerous state of the production equipment, the dangerous action of the operator, the position information and the total equivalent of the warehouse into a trained large safety regulation knowledge model, and performing safety assessment processing to obtain an early warning level and a dangerous source position; performing visualization processing on the early warning level and the dangerous source position, and feeding back to an operator in real time; the fireworks and crackers safety production and storage process can be monitored, hazard source discovery and early warning are achieved, and safety accidents in the fireworks and crackers production and storage process are effectively prevented and reduced.

Description

Technical Field

[0001] The invention relates to a method, system and storage medium for monitoring safe production and storage of fireworks and firecrackers, belonging to the field of safe production and storage monitoring. Background Art

[0002] Fireworks and firecrackers production involves a variety of high-risk chemicals and delicate operations, so safety in production has always been a top priority for the industry. Common safety issues include uneven mixing of raw materials, improper operation, and accidental ignition caused by static electricity. These problems may lead to explosions, fires, and serious casualties.

[0003] Just as important as the production process is the storage safety of fireworks and firecrackers. There are also a large number of potential safety hazards in the storage of fireworks and firecrackers, especially due to improper storage environment, irregular stacking or aging management system, which can easily cause serious safety problems. Common storage safety issues include: out-of-control warehouse temperature and humidity: excessive temperature or humidity may cause the chemical composition of fireworks and firecrackers to deteriorate or spontaneously combust; poor warehouse ventilation: fireworks and firecrackers stored for a long time without good ventilation may accumulate volatile chemicals and increase the risk of explosion; overstorage: storage capacity exceeding the warehouse's design capacity will increase the risk of fire and explosion, and make rescue work more complicated when an accident occurs.

[0004] Historically, the industry has experienced many serious accidents, resulting in huge losses of life and property, which are often caused by inadequate monitoring and early warning systems.

[0005] At present, the monitoring technologies for the safety of fireworks and firecrackers production mainly include traditional video monitoring systems, temperature and humidity monitoring, and rule-based automatic shutdown systems. Although these systems have improved the safety level to a certain extent, they still have obvious limitations. For example, traditional video monitoring systems rely on manual observation and are susceptible to human fatigue and distraction; although environmental monitoring systems can monitor temperature and humidity in real time, they lack the ability to respond to and warn of complex operations in real time; automatic shutdown systems are often slow to respond and cannot handle sudden and changing situations. Therefore, it is necessary to develop a fireworks and firecrackers safety production and storage monitoring system that integrates generative artificial intelligence with LLM big language. Summary of the invention

[0006] The purpose of the present invention is to overcome the shortcomings of the prior art and to provide a method, system and storage medium for monitoring the safe production and storage of fireworks and firecrackers, which can monitor the safe production and storage process of fireworks and firecrackers, realize hazard source discovery and early warning, and effectively prevent and reduce safety accidents in the production and storage process of fireworks and firecrackers.

[0007] To achieve the above object, the present invention is implemented by adopting the following technical solutions:

[0008] In a first aspect, the present invention provides a method for monitoring the safe production and storage of fireworks and firecrackers, comprising:

[0009] Acquire production equipment operation status data, operator action specification data and warehouse image data; the production equipment operation status data includes environmental image information and equipment status information; the operator action specification data includes operator position information and action image information;

[0010] Input the environmental image information and the operator's action image information into the image recognition model for hazard recognition processing to obtain the dangerous state of the production equipment and the operator's dangerous actions;

[0011] Calculate the total equivalent of the warehouse based on warehouse image data and goods in and out data;

[0012] The equipment status information, dangerous status of production equipment, dangerous actions of operators, location information and total equivalent of warehouse are input into the trained safety knowledge model for safety assessment and processing to obtain the warning level and location of dangerous sources of fireworks and firecrackers production and storage;

[0013] The warning levels and hazard locations of fireworks production and storage are visualized and fed back to operators in real time.

[0014] Optionally, the step of inputting the environment image information and the operator's action image information into an image recognition model for hazard recognition processing to obtain the dangerous state of the production equipment and the operator's dangerous action includes:

[0015] Inputting environmental image information into the dangerous state recognition sub-model of the image recognition model to perform image preprocessing, feature extraction mapping processing and dangerous state recognition respectively, to obtain the dangerous state of the production equipment; the dangerous state of the production equipment includes: equipment overheating, gunpowder leakage and equipment failure;

[0016] The operator's action image information is input into the dangerous action recognition sub-model of the image recognition model to perform image human shadow denoising, trajectory analysis and dangerous action recognition to obtain the operator's dangerous actions; the operator's dangerous actions include: illegal operation, entry into dangerous areas, and improper use of tools.

[0017] Optionally, the step of inputting the environmental image information into the dangerous state recognition sub-model of the image recognition model to perform image preprocessing, feature extraction and mapping processing, and dangerous state recognition respectively to obtain the dangerous state of the production equipment includes:

[0018] Performing image preprocessing on the environmental image information to obtain the environmental image information after image preprocessing; the image preprocessing includes: image denoising, contrast enhancement and normalization processing;

[0019] Use deep convolutional neural network and graph convolutional network to perform feature extraction and feature mapping on the environmental image information after image preprocessing to obtain the fused feature vector;

[0020] The fused feature vector is input into the Transformers unit for classification, and the classification results are analyzed using a multi-layer perceptron to obtain the dangerous state of the production equipment.

[0021] Optionally, the step of inputting the operator's action image information into a dangerous action recognition sub-model of an image recognition model to perform image human shadow denoising, trajectory analysis and dangerous action recognition respectively to obtain the operator's dangerous action includes:

[0022] Performing human shadow denoising on the action image information to obtain denoised action image information;

[0023] Extract action features from the denoised action image information, and use image segmentation technology to perform trajectory analysis on the extracted action features to obtain data after trajectory analysis;

[0024] Conduct risk assessment on the data after trajectory analysis to identify high-risk actions;

[0025] High-risk actions are classified and identified to obtain dangerous actions of operators.

[0026] Optionally, the total equivalent of the warehouse is calculated based on the warehouse image data and the goods in and out data, including:

[0027] Identify and classify the warehouse image data to obtain category data of the goods, where the categories include product A, product B, and product C;

[0028] Performing point cloud processing on the warehouse image data to obtain inventory volume data, wherein the inventory volume data includes: volume information of each item in the warehouse;

[0029] Obtain the equivalent coefficients corresponding to different types of goods;

[0030] Based on the equivalent coefficients corresponding to different types of goods, the category data of goods and the inventory volume data, the total equivalent of the initial warehouse is obtained;

[0031] The total equivalent of the warehouse is obtained based on the total equivalent of the initial warehouse and the data of goods entering and leaving the warehouse.

[0032] Optionally, the equipment status information, dangerous status of production equipment, dangerous actions of operators, location information and total warehouse equivalent are input into the trained safety knowledge model to perform safety assessment processing to obtain the warning level and hazard source location of fireworks and firecrackers production and storage, including:

[0033] Input the equipment status information, dangerous status of production equipment, dangerous actions of operators, location information and total warehouse equivalent into the preprocessing module of the trained safety knowledge model for data preprocessing to obtain a preprocessed data set;

[0034] Extract key features from the preprocessed data set to obtain key feature data sets in the production and storage process of fireworks and firecrackers; the key features include: abnormal operation of equipment, abnormal behavior of operators, dangerous conditions at key locations, and high equivalent areas of dangerous goods in the warehouse;

[0035] The multi-factor decision-making model is used to identify the warning levels under single key features and different key feature combinations, and the location of the hazard source is obtained by combining the location information.

[0036] Optionally, the method for acquiring the trained safety knowledge model includes:

[0037] Obtain a fireworks and firecrackers safety regulation dataset, which includes: standards, regulations, operation guidelines, historical safety accident reports, and case study documents related to the safe production and storage of fireworks and firecrackers;

[0038] Structural processing and data annotation are performed on the fireworks and firecrackers safety regulation data set as labels for the safety regulation knowledge model, wherein the data annotation includes: safety hazard identification, safety operation process and safety level classification;

[0039] The fireworks and firecrackers safety regulations dataset is processed by text cleaning, word segmentation and tagging, and data enhancement in turn to obtain a training sample set;

[0040] Based on the optimal parameters of the model, a well-trained safety knowledge model is obtained;

[0041] In a second aspect, the present invention provides a monitoring system for safe production and storage of fireworks and firecrackers, comprising:

[0042] A data acquisition module, used to obtain production equipment operation status data, operator action specification data and warehouse image data; the production equipment operation status data includes environmental image information and equipment status information; the operator action specification data includes operator position information and action image information;

[0043] The recognition module is used to input the environmental image information and the operator's action image information into the image recognition model for hazard recognition processing to obtain the dangerous state of the production equipment and the operator's dangerous action;

[0044] An equivalent calculation module is used to calculate the total equivalent of the warehouse based on the warehouse image data and the goods in and out data;

[0045] The early warning module is used to input the equipment status information, dangerous status of production equipment, dangerous actions of operators, location information and total warehouse equivalent into the trained safety knowledge model to conduct safety assessment and obtain the early warning level and location of dangerous sources of fireworks and firecrackers production and storage;

[0046] The display module is used to visualize the warning levels and hazard source locations of fireworks and firecrackers production and storage, and provide real-time feedback to operators.

[0047] Optionally, the recognition module includes: a dangerous state recognition submodule and a dangerous action recognition submodule; the dangerous state recognition submodule is used to input the environmental image information into the dangerous state recognition submodel of the image recognition model to perform image preprocessing, feature extraction and mapping processing and dangerous state recognition, so as to obtain the dangerous state of the production equipment; the dangerous state of the production equipment includes: equipment overheating, gunpowder leakage and equipment failure; the dangerous action recognition submodule is used to input the operator's action image information into the dangerous action recognition submodel of the image recognition model to perform image human shadow denoising, trajectory analysis and dangerous action recognition, so as to obtain the operator's dangerous actions; the operator's dangerous actions include: illegal operation, entry into dangerous areas and improper use of tools.

[0048] In a third aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method for monitoring the safe production and storage of fireworks and firecrackers as described in the first aspect.

[0049] Compared with the prior art, the present invention has the following beneficial effects:

[0050] The method for monitoring the safe production and storage of fireworks and firecrackers proposed in the present invention uses a large model of safety knowledge to monitor the safe production and storage process of fireworks and firecrackers, realizes the discovery and early warning of dangerous sources, effectively prevents and reduces safety accidents in the production and storage process of fireworks and firecrackers, improves the level of production safety, reduces accident risks, protects the life safety and property safety of production personnel, and reduces the adverse effects of production interruptions on enterprises and society. At the same time, the present invention also improves production efficiency and production quality, and makes positive contributions to the sustainable development of the fireworks and firecrackers industry.

[0051] The present invention monitors the equivalent of dangerous goods during the storage of fireworks and firecrackers, including real-time monitoring of warehousing and inventory. When the total equivalent of dangerous goods does not meet the standards set by the safety model, an early warning is issued to ensure the safe storage and management of dangerous goods, thereby minimizing safety risks.

[0052] The present invention monitors equipment hazards and operation hazards existing in the safe production process of fireworks and firecrackers, and realizes accurate identification of dangerous behaviors and equipment abnormalities in the production process of fireworks and firecrackers. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative labor, among which:

[0054] Figure 1 The flowchart of the method for monitoring the safe production and storage of fireworks and firecrackers in the embodiment of the present invention is shown;

[0055] Figure 2 The schematic diagram of the structure of the monitoring system for safe production and storage of fireworks and firecrackers in the embodiment of the present invention is shown;

[0056] Figure 3 A schematic diagram of equivalent management in an embodiment of the present invention is shown. DETAILED DESCRIPTION

[0057] The present invention will be further described below in conjunction with the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and cannot be used to limit the protection scope of the present invention.

[0058] The application principle of the present invention is described in detail below in conjunction with the accompanying drawings.

[0059] Example 1

[0060] like Figure 1 As shown, an embodiment of the present invention provides a method for monitoring the safe production and storage of fireworks and firecrackers, comprising the following steps:

[0061] S01: Acquire production equipment operation status data, operator action specification data and warehouse image data; the production equipment operation status data includes environmental image information and equipment status information; the operator action specification data includes operator position information and action image information;

[0062] S02: Inputting the environment image information and the operator's action image information into the image recognition model for hazard recognition processing to obtain the dangerous state of the production equipment and the operator's dangerous action;

[0063] S03: Calculate the total equivalent of the warehouse based on the warehouse image data and the goods in and out data;

[0064] S04: Input the equipment status information, dangerous status of production equipment, dangerous actions of operators, location information and total equivalent of warehouse into the trained safety knowledge model, conduct safety assessment and obtain the warning level and location of dangerous sources of fireworks and firecrackers production and storage;

[0065] S05: Visualize the warning levels and hazard locations of fireworks production and storage, and provide real-time feedback to operators.

[0066] The present invention can utilize a large safety knowledge model to monitor the safe production and storage processes of fireworks and firecrackers, realize hazard source discovery and early warning, effectively prevent and reduce safety accidents in the production and storage processes of fireworks and firecrackers, improve production safety levels, reduce accident risks, protect the lives and property safety of production personnel, and reduce the adverse effects of production interruptions on enterprises and society.

[0067] In this embodiment, in step S01, the data on the operation status of the production equipment is collected by a fixed camera on the equipment end and temperature, pressure, and rotation speed sensors; the fixed camera on the equipment end collects environmental image information, and the temperature, pressure, and rotation speed sensors collect equipment temperature, pressure, mechanical operation speed and other parameters, namely, equipment status information; the data on the operator's action specification is collected by a portable monitoring camera and a position sensor, and the portable monitoring camera collects action image information, and the position sensor collects the operator's position information; warehouse image data is collected by a binocular vision camera.

[0068] In this embodiment, in step S02, the environment image information and the operator's action image information are input into the image recognition model for danger recognition processing to obtain the dangerous state of the production equipment and the operator's dangerous action. The specific steps are as follows:

[0069] (1) Inputting environmental image information into the dangerous state recognition sub-model of the image recognition model to perform image preprocessing, feature extraction and mapping processing, and dangerous state recognition, respectively, to obtain the dangerous state of the production equipment; the dangerous state of the production equipment includes: equipment overheating, gunpowder leakage, and equipment failure;

[0070] Specifically, image preprocessing is performed on the environmental image information to obtain the environmental image information after image preprocessing; the image preprocessing includes: image denoising, contrast enhancement and normalization processing to optimize the image quality and remove unnecessary background interference;

[0071] Use deep convolutional neural networks (Deep CNN) to extract high-level features related to the device state, such as the shape, edges, texture, etc. of the device. This process may use multiple convolution and pooling operations to gradually extract important features in the image.

[0072] Use the Graph Neural Network (GNN) to fuse and associate multiple extracted features, identify the spatial relationship between features and potential equipment failure modes, and obtain the fused feature vector;

[0073] The fused feature vector is input into the Transformers unit for classification, and the preliminary classification results of the dangerous state of the equipment are identified through the Self-Attention Mechanism; this process can effectively capture the complex patterns and changes of the equipment status.

[0074] The preliminary classification results were analyzed using a multilayer perceptron (MLP) to obtain the dangerous status of the production equipment;

[0075] The step-by-step processing steps use the current advanced image processing and deep learning technologies, such as CNN, GNN and Transformers, to achieve accurate identification and evaluation of the status of production equipment. This layered processing method ensures the robustness and efficiency of the system, and can timely detect and warn of various equipment risks that may occur during the production process.

[0076] (2) The operator's action image information is input into the dangerous action recognition sub-model of the image recognition model to perform image human shadow denoising, trajectory analysis and dangerous action recognition, and obtain the operator's dangerous actions; the operator's dangerous actions include: illegal operation, entering a dangerous area, and improper use of tools.

[0077] Specifically, the action image information is subjected to human shadow denoising processing (such as non-local mean denoising algorithm) to remove background noise and irrelevant information, while enhancing the action details of the image to improve the accuracy of subsequent processing, thereby obtaining denoised action image information;

[0078] Through the deep convolutional neural network (Deep CNN), the action features are extracted from the denoised action image information, focusing on extracting key features related to the action trajectory and action posture, such as hand position, limb angle and movement direction;

[0079] Image segmentation technology (such as region growing or edge detection) and target detection algorithms (such as YOLO or FasterR-CNN) are used to perform trajectory analysis on the extracted motion features to obtain data after trajectory analysis; the analysis process includes path tracking, speed measurement, trajectory deviation detection, etc., to identify motion behaviors that may not comply with safety operating procedures.

[0080] Through the integrated risk assessment model (such as risk assessment based on Bayesian network), the data after trajectory analysis is evaluated to identify high-risk actions; during the evaluation process, the operator's real-time action data is combined with the pre-set safety operation specifications to determine whether there are potential dangerous actions. The model can automatically quantify the risk level and identify high-risk actions that may lead to safety accidents.

[0081] A classification algorithm (such as a support vector machine SVM or a neural network classifier) ​​is used to classify and identify high-risk actions to obtain operator dangerous actions; the operator dangerous actions are dangerous action types.

[0082] This method achieves accurate identification of dangerous actions of operators through multi-level image processing and motion analysis. In particular, in the motion trajectory analysis layer, image segmentation and target detection technology are combined to effectively identify behaviors that are inconsistent with safe operating procedures, which has significant advantages in improving production safety. The risk assessment layer integrates advanced risk assessment models, which can not only identify high-risk actions, but also quantify the risk level, providing strong support for real-time warning and safety decision-making. Dangerous action classification can greatly improve the intelligence and response speed of the system by identifying dangerous actions.

[0083] In this embodiment, in step S03, the total equivalent of the warehouse is calculated based on the warehouse image data and the goods in and out data, specifically including:

[0084] Identify and classify the warehouse image data to obtain category data of the goods, where the categories include product A, product B, and product C;

[0085] Performing point cloud processing on the warehouse image data to obtain inventory volume data, wherein the inventory volume data includes: volume information of each item in the warehouse;

[0086] Obtain the equivalent coefficients corresponding to different types of goods;

[0087] Based on the equivalent coefficients corresponding to different types of goods, the category data of goods and the inventory volume data, the total equivalent of the initial warehouse is obtained;

[0088] The total equivalent of the warehouse is obtained based on the total equivalent of the initial warehouse and the data of goods entering and leaving the warehouse.

[0089] Specifically,

[0090] (1) Cargo identification

[0091] The image classification model in the deep learning algorithm is used to identify and classify warehouse image data. The classification model may be based on a convolutional neural network (Mob-CNN) or other existing lightweight image classification algorithms to classify each product into a specific product type, such as product A, product B, and product C. The specific recognition process includes image preprocessing, feature extraction, classifier training and inference. Output: Category information of M goods , these category information include product A, product B and product C, etc.

[0092] (2) Obtaining inventory volume data

[0093] Using binocular vision technology, the three-dimensional volume of each cargo is calculated through depth information. The process first performs stereo matching on the images collected by the binocular camera to generate a depth map, and then extracts the spatial contour and volume information of the cargo through point cloud processing technology. The final output is: the inventory volume data V of various cargoes, including the volume information of M cargoes in the warehouse .

[0094] (3) Calculation of the initial total equivalent of goods

[0095] Obtain the class information Class of each cargo and the corresponding inventory volume data Cargo;

[0096] According to the pre-set equivalent coefficient of each category of goods, the volume data of the goods is multiplied by the equivalent coefficient to calculate the equivalent of each category of goods; the equivalent of each category of goods is accumulated to obtain the total equivalent of the initial warehouse. The formula is as follows:

[0097]

[0098] In the formula, is the total equivalent of the initial warehouse, is the equivalent coefficient of the i-th category of goods, The corresponding inventory volume, m or M is the number of types.

[0099] (4) Obtaining the inbound and outbound data of goods

[0100] The data of goods in and out of warehouse include the time of each in and out operation, the quantity and category of goods. The quantity and category of goods can be obtained through the above-mentioned goods category identification, inventory volume acquisition, and goods equivalent calculation functions, or automatically obtained through barcode scanning or RFID tags. The in and out time is automatically recorded by the warehouse management system, and the system combines these data to generate a complete in and out record.

[0101] (5) Calculation of total dynamic warehouse equivalent

[0102] The total equivalent of the warehouse is dynamically updated according to each cargo entry and exit operation. The specific calculation formula is as follows:

[0103]

[0104] In the formula, represents the total equivalent in the warehouse at time t, Represents the total equivalent of various types of goods entering the warehouse at time t, Represents the total equivalent of various types of goods shipped out of the warehouse at time t.

[0105] In summary, the core functions of dangerous goods management include the following three points: The combination of classification model and volume calculation: The classification model is mainly used to identify the category of goods, while binocular vision technology and depth information processing are used to calculate the volume of goods. The combination of the two can obtain the category and volume data of the goods respectively, which are then used for equivalent calculation. Acquisition of inbound and outbound data: The inbound and outbound time is automatically recorded by the system, the quantity of goods is obtained through the goods equivalent identification system or automated scanning (such as barcodes or RFID tags), and the category information corresponds to the records in the inventory management system. Dynamic equivalent calculation formula: The dynamically updated total warehouse equivalent takes into account the inbound and outbound operations of goods, ensuring that the data in the inventory management system is always consistent with the actual situation.

[0106] In this embodiment, in step S04, the equipment status information, the dangerous status of the production equipment, the dangerous actions of the operator, the location information and the total equivalent of the warehouse are input into the trained safety knowledge model to perform safety assessment processing to obtain the warning level and the location of the dangerous source of fireworks and firecrackers production and storage. The specific steps are as follows:

[0107] The equipment status information, dangerous status of production equipment, dangerous actions of operators, location information and total warehouse equivalent are input into the preprocessing module of the trained safety knowledge model for data preprocessing to obtain the preprocessed data set; the preprocessing module performs data cleaning, noise removal, incomplete data repair, outlier detection and data standardization on these data to ensure the consistency and accuracy of all input data;

[0108] The key features of the preprocessed data set are extracted using a feature extraction algorithm (such as principal component analysis PCA or a feature extraction layer in a deep learning network) to obtain a key feature data set in the production and storage process of fireworks and firecrackers; the key features include: abnormal operation of equipment, abnormal behavior of operators, dangerous conditions at key locations, and high equivalent areas of dangerous goods in the warehouse;

[0109] Using multi-factor decision-making models , identify the warning level under single key features and different key feature combinations, and combine the location information to obtain the location of the hazard source. Specifically, based on the current key feature data and location information, the model accurately locates the potential hazard source in the warehouse or production site. This step is done through spatial analysis algorithms (such as spatial matching algorithms in the production plant geographic information system GIS) or by extracting the location information associated with the fault and the hazard source, and finally generating a spatial risk distribution map.

[0110] The warning levels are shown in the following table:

[0111]

[0112] Specifically, the method for obtaining the trained safety knowledge model includes:

[0113] Obtain the fireworks and firecrackers safety regulation dataset, which includes: standards, regulations, operating guidelines and case study documents related to the safe production and storage of fireworks and firecrackers; specifically including national and international safety production standards and regulations; generally applicable operating guidelines and process specifications in the industry; safety accident case study documents related to the production and storage of fireworks and firecrackers; historical safety accident reports and their handling methods, etc.

[0114] Performing structured processing and data annotation on the fireworks and firecrackers safety regulations dataset as labels for the safety regulations knowledge model;

[0115] Specifically, structured processing: remove redundant information and classify it into different subject modules, such as equipment safety, operator behavior standards, environmental control, etc., to ensure data coverage and accuracy. Data annotation: annotate the data according to actual needs, especially focusing on safety hazards, operating standards and accident prevention measures, so as to facilitate the subsequent model fine-tuning process.

[0116] The fireworks and firecrackers safety regulations dataset is processed by text cleaning, word segmentation and tagging, and data enhancement in turn to obtain a training sample set;

[0117] Specifically, text cleaning: cleaning the collected text data, including removing irrelevant information, standardizing terms and symbols, eliminating ambiguity, etc., to ensure the clarity and consistency of the text.

[0118] Word segmentation and annotation: Perform word segmentation on text data and annotate features based on domain knowledge. The annotated features include safety hazard identification, safety operation procedures, and safety level classification. Safety hazard identification: Annotate descriptions of equipment failures, operator misoperation, and environmental anomalies; Safety operation procedures: Annotate standard operating steps, emergency response methods, and risk prevention measures; Safety level classification: Classify different types of security threats and determine their corresponding warning levels.

[0119] Data enhancement: In order to improve the generalization ability of the model, the data can be enhanced, such as synonym replacement, sentence structure change, etc., to generate more training samples.

[0120] Input the training sample set into the initial safety knowledge model for training to obtain the optimal parameters of the model; the initial safety knowledge model is a Llama model;

[0121] Based on the optimal parameters of the model, a trained safety knowledge model is obtained. The model can generate intelligent safety management suggestions based on the input equipment status information, dangerous status of production equipment, operator actions, location information, etc.; and provide users with instant safety warnings related to the production and storage of fireworks and firecrackers, including hazard levels and operational suggestions.

[0122] Specifically, during the training process,

[0123] Model architecture adjustment: According to the characteristics of fireworks and firecrackers safety production, the architecture of the Llama model is appropriately adjusted, such as adjusting the attention mechanism to enhance the model's understanding of vocabulary and concepts in specific fields.

[0124] Loss function design: Design a loss function suitable for safety management tasks, such as adding weights for safety hazard identification and classification to ensure the model's high sensitivity to key risk points.

[0125] Training parameter setting: Set appropriate parameters such as learning rate, batch size, and number of training rounds to ensure that the model can fully learn domain knowledge while avoiding overfitting.

[0126] Training process: Use the training sample set to train the Llama model so that it can generate accurate recommendations and risk assessments related to the safe production of fireworks and firecrackers.

[0127] Specifically, model verification and optimization:

[0128] Validation set test: Run the validation set on the trained model to test the model’s performance in fireworks and firecrackers safety production scenarios to ensure the accuracy and practicality of its output results.

[0129] Optimization and adjustment: Based on the validation results, further optimization and adjustment of the model is performed, which may include readjusting model parameters, improving the data preprocessing process, or introducing more training data.

[0130] Testing and Evaluation: The fine-tuned model is thoroughly evaluated through a series of test cases to ensure that it can provide security management recommendations stably and reliably in practical applications.

[0131] Through the above detailed training methods, the fireworks and firecrackers safety production LLM model can effectively integrate various standards and specifications for fireworks and firecrackers safety management and provide intelligent safety management support. This not only improves the practical application value of the model, but also ensures its reliability in complex safety environments.

[0132] The fireworks and firecrackers safety regulations dataset is structured and labeled as a label for the safety regulations knowledge model. The labeling process includes:

[0133] Example 2

[0134] The embodiment of the present invention provides a monitoring system for safe production and storage of fireworks and firecrackers, including:

[0135] A data acquisition module, used to obtain production equipment operation status data, operator action specification data and warehouse image data; the production equipment operation status data includes environmental image information and equipment status information; the operator action specification data includes operator position information and action image information;

[0136] The recognition module is used to input the environmental image information and the operator's action image information into the image recognition model for hazard recognition processing to obtain the dangerous state of the production equipment and the operator's dangerous action;

[0137] An equivalent calculation module is used to calculate the total equivalent of the warehouse based on the warehouse image data and the goods in and out data;

[0138] The early warning module is used to input the equipment status information, dangerous status of production equipment, dangerous actions of operators, location information and total warehouse equivalent into the trained safety knowledge model to conduct safety assessment and obtain the early warning level and location of dangerous sources of fireworks and firecrackers production and storage;

[0139] The display module is used to visualize the warning levels and hazard source locations of fireworks and firecrackers production and storage, and provide real-time feedback to operators.

[0140] Specifically, the recognition module includes: a dangerous state recognition submodule and a dangerous action recognition submodule; the dangerous state recognition submodule is used to input the environmental image information into the dangerous state recognition submodel of the image recognition model to perform image preprocessing, feature extraction mapping processing and dangerous state recognition, so as to obtain the dangerous state of the production equipment; the dangerous state of the production equipment includes: equipment overheating, gunpowder leakage and equipment failure; the dangerous action recognition submodule is used to input the operator's action image information into the dangerous action recognition submodel of the image recognition model to perform image human shadow denoising, trajectory analysis processing and dangerous action recognition, so as to obtain the operator's dangerous action; the operator's dangerous action includes: illegal operation, entry into a dangerous area, and improper use of tools.

[0141] Example 3

[0142] An embodiment of the present invention provides a computer-readable storage medium.

[0143] A computer-readable storage medium stores a program thereon, which, when executed by a processor, implements the steps in a method for monitoring the safe production and storage of fireworks and firecrackers as described in Example 1 of the present disclosure.

[0144] The detailed steps are the same as those of the method for monitoring the safe production and storage of fireworks and firecrackers provided in Example 1, and will not be repeated here.

[0145] Example 4

[0146] A fourth embodiment of the present disclosure provides an electronic device.

[0147] An electronic device comprises a memory, a processor and a program stored in the memory and executable on the processor, wherein when the processor executes the program, the steps in a method for monitoring the safe production and storage of fireworks and firecrackers as described in Example 1 of the present disclosure are implemented.

[0148] The detailed steps are the same as those of the method for monitoring the safe production and storage of fireworks and firecrackers provided in Example 1, and will not be repeated here.

[0149] The embodiments of the present invention are described above in conjunction with the accompanying drawings, but the present invention is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the enlightenment of the present invention, ordinary technicians in this field can also make many forms without departing from the scope of protection of the purpose of the present invention and the claims, which all fall within the protection of the present invention.

Claims

1. A method for monitoring the safe production and storage of fireworks and firecrackers, characterized in that: include: Acquire production equipment operation status data, operator action specification data and warehouse image data; the production equipment operation status data includes environmental image information and equipment status information; the operator action specification data includes operator position information and action image information; Input the environmental image information and the operator's action image information into the image recognition model for hazard recognition processing to obtain the dangerous state of the production equipment and the operator's dangerous actions; Calculate the total equivalent of the warehouse based on warehouse image data and goods in and out data; The equipment status information, dangerous status of production equipment, dangerous actions of operators, location information and total equivalent of warehouse are input into the trained safety knowledge model for safety assessment and processing to obtain the warning level and location of dangerous sources of fireworks and firecrackers production and storage; The warning levels and hazard locations of fireworks production and storage are visualized and fed back to operators in real time.

2. The method for monitoring the safe production and storage of fireworks and firecrackers according to claim 1, characterized in that: The step of inputting the environment image information and the operator's action image information into the image recognition model for hazard recognition processing to obtain the dangerous state of the production equipment and the operator's dangerous action includes: Inputting environmental image information into the dangerous state recognition sub-model of the image recognition model to perform image preprocessing, feature extraction mapping processing and dangerous state recognition respectively, to obtain the dangerous state of the production equipment; the dangerous state of the production equipment includes: equipment overheating, gunpowder leakage and equipment failure; The operator's action image information is input into the dangerous action recognition sub-model of the image recognition model to perform image human shadow denoising, trajectory analysis and dangerous action recognition to obtain the operator's dangerous actions; the operator's dangerous actions include: illegal operation, entry into dangerous areas, and improper use of tools.

3. The method for monitoring the safe production and storage of fireworks and firecrackers according to claim 2, characterized in that: The step of inputting the environmental image information into the dangerous state recognition sub-model of the image recognition model to perform image preprocessing, feature extraction and mapping processing, and dangerous state recognition respectively to obtain the dangerous state of the production equipment includes: Performing image preprocessing on the environmental image information to obtain the environmental image information after image preprocessing; the image preprocessing includes: image denoising, contrast enhancement and normalization processing; Use deep convolutional neural network and graph convolutional network to perform feature extraction and feature mapping on the environmental image information after image preprocessing to obtain the fused feature vector; The fused feature vector is input into the Transformers unit for classification, and the classification results are analyzed using a multi-layer perceptron to obtain the dangerous state of the production equipment.

4. The method for monitoring the safe production and storage of fireworks and firecrackers according to claim 3, characterized in that: The operator's action image information is input into the dangerous action recognition sub-model of the image recognition model to perform image human shadow denoising, trajectory analysis and dangerous action recognition respectively to obtain the operator's dangerous action, including: Performing human shadow denoising on the action image information to obtain denoised action image information; Extract action features from the denoised action image information, and use image segmentation technology to perform trajectory analysis on the extracted action features to obtain data after trajectory analysis; Conduct risk assessment on the data after trajectory analysis to identify high-risk actions; High-risk actions are classified and identified to obtain dangerous actions of operators.

5. The method for monitoring the safe production and storage of fireworks and firecrackers according to claim 4, characterized in that: The total equivalent of the warehouse is calculated based on the warehouse image data and the goods in and out data, including: Identify and classify the warehouse image data to obtain category data of the goods, where the categories include product A, product B, and product C; Performing point cloud processing on the warehouse image data to obtain inventory volume data, wherein the inventory volume data includes: volume information of each item in the warehouse; Obtain the equivalent coefficients corresponding to different types of goods; Based on the equivalent coefficients corresponding to different types of goods, the category data of goods and the inventory volume data, the total equivalent of the initial warehouse is obtained; The total equivalent of the warehouse is obtained based on the total equivalent of the initial warehouse and the data of goods entering and leaving the warehouse.

6. The method for monitoring the safe production and storage of fireworks and firecrackers according to claim 5, characterized in that: The equipment status information, dangerous status of production equipment, dangerous actions of operators, location information and total warehouse equivalent are input into the trained safety knowledge model to perform safety assessment processing to obtain the warning level and dangerous source location of fireworks and firecrackers production and storage, including: Input the equipment status information, dangerous status of production equipment, dangerous actions of operators, location information and total warehouse equivalent into the preprocessing module of the trained safety knowledge model for data preprocessing to obtain a preprocessed data set; Extract key features from the preprocessed data set to obtain key feature data sets in the production and storage process of fireworks and firecrackers; the key features include: abnormal operation of equipment, abnormal behavior of operators, dangerous conditions at key locations, and high equivalent areas of dangerous goods in the warehouse; The multi-factor decision-making model is used to identify the warning levels under single key features and different key feature combinations, and the location of the hazard source is obtained by combining the location information.

7. The method for monitoring the safe production and storage of fireworks and firecrackers according to claim 6, characterized in that: The method for acquiring the trained safety knowledge model includes: Obtain a fireworks and firecrackers safety regulation dataset, which includes: standards, regulations, operation guidelines, historical safety accident reports, and case study documents related to the safe production and storage of fireworks and firecrackers; Structural processing and data annotation are performed on the fireworks and firecrackers safety regulation data set as labels for the safety regulation knowledge model, wherein the data annotation includes: safety hazard identification, safety operation process and safety level classification; The fireworks and firecrackers safety regulations dataset is processed by text cleaning, word segmentation and tagging, and data enhancement in turn to obtain a training sample set; Based on the optimal parameters of the model, a well-trained large model of safety knowledge is obtained.

8. A monitoring system for safe production and storage of fireworks and firecrackers, characterized in that: include: Data acquisition module, used to obtain production equipment operation status data, operator action specification data and warehouse image data; The production equipment operation status data includes environmental image information and equipment status information; the operator action specification data includes operator position information and action image information; The recognition module is used to input the environmental image information and the operator's action image information into the image recognition model for hazard recognition processing to obtain the dangerous state of the production equipment and the operator's dangerous action; An equivalent calculation module is used to calculate the total equivalent of the warehouse based on the warehouse image data and the goods in and out data; The early warning module is used to input the equipment status information, dangerous status of production equipment, dangerous actions of operators, location information and total warehouse equivalent into the trained safety knowledge model to conduct safety assessment and obtain the early warning level and location of dangerous sources of fireworks and firecrackers production and storage; The display module is used to visualize the warning levels and hazard source locations of fireworks and firecrackers production and storage, and provide real-time feedback to operators.

9. The monitoring system for safe production and storage of fireworks and firecrackers according to claim 8, characterized in that: The recognition module includes: a dangerous state recognition submodule and a dangerous action recognition submodule; the dangerous state recognition submodule is used to input environmental image information into the dangerous state recognition submodel of the image recognition model to perform image preprocessing, feature extraction and mapping processing and dangerous state recognition respectively, so as to obtain the dangerous state of the production equipment; the dangerous state of the production equipment includes: equipment overheating, gunpowder leakage and equipment failure; the dangerous action recognition submodule is used to input the operator's action image information into the dangerous action recognition submodel of the image recognition model to perform image human shadow denoising processing, trajectory analysis processing and dangerous action recognition respectively, so as to obtain the operator's dangerous action; the operator's dangerous action includes: illegal operation, entry into a dangerous area and improper use of tools.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method for monitoring the safe production and storage of fireworks and firecrackers as described in any one of claims 1-7 are implemented.

Images