A method and system for safety monitoring in a paint shop

Abstract

The application discloses a kind of production paint workshop safety monitoring method and system, it is related to safety monitoring technical field.A kind of production paint workshop safety monitoring system, including have: data acquisition module, real-time monitoring module, prevention module, control module, early warning module and recording module.The application is multidimensional by sensor data, sound data and image data acquisition, in combination with pre-trained identification model, realize the comprehensive safety monitoring of worker, equipment, environment and production quality in workshop;Through pre-trained identification model and the real-time monitoring model and prevention model of joint training, improve the accuracy of abnormal identification, can identify the abnormal situation under different security categories, and through optimization training enhance the accuracy of anomaly detection;Through regional control early warning strategy library and association control early warning strategy library, according to monitoring result and prediction result realizes the automatic control and early warning of workshop, improves processing efficiency and accuracy.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of safety monitoring, in particular to a safety monitoring method and system for a paint production workshop. BACKGROUND

[0002] Paint plays an important role in industrial production and is widely used in automobile manufacturing, furniture production, home appliance manufacturing and other industries. However, there are many potential hazards in the workshop for producing paint, mainly including volatile organic compound (VOC) release, accumulation of flammable and explosive gases, high-temperature operation of production equipment, and safety hazards in worker operation, etc. Especially under the background of efficient coating operation, how to ensure the safety of workers, equipment, environment and production quality is particularly critical.

[0003] In the prior art, the monitoring means for the safety of the paint workshop are mainly concentrated in one aspect, such as simple environmental monitoring or equipment fault detection, and a complete safety management system cannot be formed; only the generated data is judged in real time, and the prevention ability of the workshop safety is lacking; the safety events are recorded in logs, but intelligent analysis of the records is lacking, and the records and safety monitoring cannot be connected and optimized; the quality in the paint production process cannot be intelligently identified; the analysis of safety data does not combine the connection between data for identification and optimization, and the analysis result is not accurate. SUMMARY

[0004] In view of the deficiencies of the prior art, the present application proposes a comprehensive safety monitoring method and system for a paint workshop, which ensures the all-round safety monitoring of workers, equipment, environment and production quality through multi-level, multi-dimensional data acquisition, real-time detection, prevention control, early warning recording and other modules, and organically integrates worker safety, equipment safety, environmental safety and production quality safety. Through intelligent monitoring means, the safety and efficient production of the paint workshop are ensured.

[0005] A safety monitoring method for a paint production workshop, comprising the following steps:

[0006] Step S1: obtaining sensor data, sound data and image data of each area of the paint workshop, sending the sound data and image data into a pre-trained identification model for abnormal identification to obtain sound monitoring results and image monitoring results, taking the sensor data, sound monitoring results and image monitoring results as safety data, and preprocessing the safety data to add safety monitoring marks to the safety data; the safety monitoring marks include safety categories, data acquisition areas, acquisition times and data types, and the safety categories include worker safety, equipment safety, environmental safety and production quality safety;

[0007] Step S2: the real-time safety data obtained is sent into the real-time monitoring model one by one for calculation to obtain preliminary real-time monitoring results, the preliminary real-time monitoring results are verified according to the preliminary prediction results obtained by the prevention model to obtain real-time monitoring results; the real-time monitoring model and the prevention model are optimized through joint training, the results of the models are verified between the models, and the identification accuracy of the models is improved;

[0008] Step S3: the time sequence safety data obtained is sent into the prevention model for calculation to obtain preliminary prediction results, the preliminary prediction results are optimized according to the real-time monitoring results to obtain prediction results, and prediction control results are obtained by matching the prediction results; emergency treatment is performed on the emergency cases in the real-time monitoring results and the prediction control results; the processing results of the safety data are saved, and a safety training report is generated;

[0009] Step S4: the real-time monitoring results and the prediction control results are divided according to the data acquisition area and the safety category; the real-time monitoring results and the prediction control results in the same area are matched with the area control strategy and the area early warning strategy in the area control early warning strategy library, and the workshop is automatically controlled and warned;

[0010] Step S5: the real-time monitoring results and the prediction control results of the same safety category are matched with the associated control early warning strategy in the associated control early warning strategy library, and automatic control and warning are performed;

[0011] Step S6: the safety data and the processing results in the workshop are recorded, and the records are used to optimize the real-time monitoring model, the prevention model, the area control early warning strategy library and the associated control early warning strategy library.

[0012] As a preferred technical solution of the present application, the pre-trained identification model comprises:

[0013] Each sound data and image data acquisition source is trained by an independent identification model, and each identification model is only used for identifying corresponding data; the identification model is constructed based on a support vector machine, is trained and optimized through a plurality of labeled historical data, and finally outputs the monitoring results of the sound data and the image data.

[0014] As a preferred technical solution of the present application, the real-time monitoring model and the prevention model comprise:

[0015] The real-time monitoring model comprises:

[0016] The sensor data processing layer is used for processing sensor data, mainly gas concentration sensor data, temperature sensor data and humidity sensor data; the data of each sensor is first subjected to threshold judgment to obtain an abnormality weight alpha, then an abnormality distinguishing value P is obtained according to the change value of the sensor data, the abnormality value K of the sensor is calculated through the abnormality distinguishing value and the abnormality weight, and the sensor data processing result is obtained according to the interval where the abnormality value is located; the calculation formula of the abnormality value is:

[0017]

[0018] In the formula, alpha1 represents the abnormality weight of the first sensor data, and P1 represents the abnormality distinguishing value of the first sensor data.

[0019] The real-time monitoring layer is used for classifying according to the sensor data processing result, the sound monitoring result and the image monitoring result to obtain a preliminary real-time monitoring result.

[0020] The result optimization layer is used for optimizing the preliminary real-time monitoring result according to the preliminary prediction result to obtain a real-time monitoring result.

[0021] The prevention model comprises:

[0022] The data merging layer is used for reconstructing the sensor data processing result, the sound monitoring result and the image monitoring result to obtain time-series safety prediction data.

[0023] The safety prediction layer is used for calculating according to the safety prediction data to obtain a preliminary prediction result.

[0024] The result optimization layer is used for optimizing the preliminary prediction result according to the real-time monitoring result to obtain a prediction result.

[0025] As a preferred technical scheme of the present application, the joint training of the real-time monitoring model and the prevention model comprises:

[0026] The initial real-time monitoring model and the initial prevention model are constructed, the initial real-time monitoring model is constructed based on a deep convolutional neural network, the initial prevention model is constructed based on a long short-time neural network, the initial real-time monitoring model comprises 1 input layer, 1 data calculation layer, 16 hidden layers and 2 output layers, the initial prevention model comprises 1 input layer, 36 hidden layers and 2 output layers, and the last 4 hidden layers of the two models are shared for optimizing the result.

[0027] A plurality of safety data are obtained, and real-time abnormality results and prediction results are labeled, the labeled data are combined into a safety data set, and the safety data set is divided into a verification set, a test set and a training set.

[0028] The initial real-time monitoring model and the initial prevention model are trained by selecting a certain proportion of data in the safe data training set and the validation set respectively, and the optimized real-time monitoring model and the optimized prevention model are obtained by taking the real-time abnormal result and the prediction result as the target respectively.

[0029] The optimized real-time monitoring model and the optimized prevention model are trained by selecting a certain proportion of data in the safe data test set respectively, and the final output layer output and the actual correct proportion are taken as the accuracy value, and the two models are adjusted according to the accuracy value, so that the real-time monitoring model and the prevention model with the final accuracy rate reaching the standard are obtained.

[0030] As a preferred technical solution of the present application, the regional control early warning strategy library and the associated control early warning strategy library include:

[0031] The regional control early warning strategy library stores specific control and early warning strategies for real-time monitoring results and prevention control results in the region; the control and early warning strategies in the regional control early warning strategy library are more targeted, and the abnormal situation of the current region is controlled and warned in the first time;

[0032] The associated control early warning strategy library stores specific control and early warning strategies for all real-time monitoring results and prevention control results of the same safety category; compared with the regional control early warning strategy library, the associated control early warning strategy library is associated with safety data analysis results between different regions, and potential abnormal situation processing strategies are obtained through the association between regions.

[0033] As a preferred technical solution of the present application, the verification of the real-time monitoring result according to the prediction result includes:

[0034] After the real-time monitoring model outputs the result, the corresponding relationship between the prediction result and the real-time monitoring result in the real-time prediction association table is verified; the real-time prediction association table stores the direct and indirect corresponding relationship between the prediction result and the real-time monitoring result; the direct corresponding relationship indicates that there is a strong correlation between the prediction result and the real-time monitoring result, and the indirect corresponding relationship indicates that there is a weak correlation between the prediction result and the real-time monitoring result; if the verification result is a strong correlation, the verification is passed; if the verification result is a weak correlation, the verification is passed only if a plurality of verifications are weak correlations; if the frequency of the verification result without correlation exceeds a threshold value, it is recorded, and the record will be used to update the real-time prediction association table.

[0035] As a preferred technical solution of the present application, the generation of the safety training report includes:

[0036] The safety data processing results in different time periods are acquired and processed, including sensor data, sound data, image data, and corresponding abnormality detection results and emergency processing records; the safety data is subjected to cluster analysis, similar safety events are classified by grouping according to the cluster results, and a cluster result of each type of event is generated; key words are extracted from key features in the cluster result, statistical analysis is performed according to the occurrence frequency of the key words, and report contents related to each type of safety hazard are generated; through a time series analysis method, based on historical data and time sequence characteristics of safety events, future possible safety risks are predicted, and the prediction results are included in the safety training report; the safety training report includes graphical display of the safety data, and intuitively presents safety risks and data distribution in different regions.

[0037] A safety monitoring system for a paint shop, comprising the following modules:

[0038] A data acquisition module comprising a data acquisition unit and a data processing unit, the data acquisition unit being configured to acquire sensor data, sound data and image data of each region of the paint shop; the data processing unit is configured to input the sound data and the image data into a pre-trained recognition model for abnormality recognition to obtain sound monitoring results and image monitoring results, and to input the sensor data, the sound monitoring results and the image monitoring results as safety data, and to pre-process the safety data and add safety monitoring marks to the safety data;

[0039] A real-time monitoring module comprising a real-time monitoring unit and a model optimization unit, the real-time monitoring unit comprising a real-time monitoring model configured to calculate the real-time acquired safety data to obtain real-time monitoring results; the model optimization unit is configured to optimize the real-time monitoring model and the prevention model, and to associate the real-time monitoring model and the prevention model by joint training and record optimization model training set;

[0040] A prevention module comprising a prediction unit, a prediction control unit, an emergency handling unit and a safety training unit, the prediction unit being configured to input the acquired safety data into the prevention model for calculation, and to optimize the prediction results according to the real-time monitoring results to obtain the prediction results; the prediction control unit is configured to match the prediction control results according to the prediction results; the emergency handling unit is configured to handle the emergency in the real-time monitoring results and the prevention control results; the safety training unit is configured to save the processing results of the safety data and generate a safety training report;

[0041] The early warning module and the control module are integrated in a control early warning module, the control early warning module comprises a result division unit and a matching unit in addition to the early warning module and the control module, the result division unit is used for dividing real-time monitoring results and prediction control results according to data acquisition areas and safety categories; the matching unit is used for matching regional control strategies and regional early warning strategies in a regional control early warning strategy library according to real-time monitoring results and prediction control results in the same area, and matching associated control strategies and associated early warning strategies in an associated control early warning strategy library according to real-time monitoring results and prediction control results of the same safety category;

[0042] The control module is used for automatically controlling the workshop according to the matched regional control strategies and associated control strategies;

[0043] The early warning module is used for automatically early warning the workshop according to the matched regional early warning strategies and associated early warning strategies;

[0044] The recording module is used for recording safety data and processing results in the workshop, and the recording will be used for optimizing real-time monitoring models, prevention models, a regional control early warning strategy library and an associated control early warning strategy library.

[0045] The present application has the following advantages:

[0046] The present application realizes comprehensive safety monitoring of workers, equipment, environment and production quality in the workshop through multi-dimensional acquisition of sensor data, sound data and image data combined with a pre-trained recognition model; the pre-trained recognition model and the jointly trained real-time monitoring model and prevention model improve the accuracy of abnormal identification, can identify abnormal conditions under different safety categories (worker safety, equipment safety, environmental safety, production quality safety), and enhance the accuracy of abnormal detection through an optimization algorithm; the regional control early warning strategy library and the associated control early warning strategy library realize automatic control and early warning of the workshop according to monitoring results and prediction results, and improve the processing efficiency and accuracy.

[0047] The present application predicts potential safety risks through time series data, and performs emergency handling according to the risk level, thereby improving the prevention ability of safety accidents.

[0048] The present application records all safety data and processing results, and uses them for subsequent model training and optimization, thereby continuously improving the performance of real-time monitoring models and prevention models. BRIEF DESCRIPTION OF DRAWINGS

[0049] Figure 1 A structure schematic view of a safety monitoring system of a production paint workshop adopted by an embodiment of the present application;

[0050] Figure 2A structural diagram of a prevention module of a safety monitoring system of a paint workshop is provided in an embodiment of the present application. DETAILED DESCRIPTION

[0051] In order to enable personnel in the technical field to better understand the technical solutions in the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below in combination with the drawings in the embodiments of the present application.

[0052] Embodiment 1, a safety monitoring method of a paint workshop, comprising the following steps:

[0053] Step S1: obtaining sensor data, sound data and image data of each area of the paint workshop, sending the sound data and the image data into a pre-trained recognition model for abnormality recognition to obtain sound monitoring results and image monitoring results, taking the sensor data, the sound monitoring results and the image monitoring results as safety data, pre-processing the safety data, and adding safety monitoring marks to the safety data; the safety monitoring marks include a safety category, a data collection area, a collection time and a data type, and the safety category includes worker safety, equipment safety, environmental safety and production quality safety;

[0054] Obtaining structure information of the workshop, and dividing the paint workshop into areas according to the structure information of the workshop, and marking the source of data acquisition, the corresponding data category and the safety category to which the data belongs in the area; in the specific implementation process, the structure of the workshop is constructed using 2D or 3D according to the complexity and actual situation of the workshop, and the structure diagram of the workshop is used for intuitive display of the workshop and association between areas of the workshop, and the association is obtained from the safety category to which the data belongs;

[0055] The pre-trained recognition model includes:

[0056] Each source of sound data and image data is trained by an independent recognition model, and each recognition model is only for the identification of the corresponding data; the recognition model is constructed based on a support vector machine, is trained and optimized through a plurality of labeled historical data, and finally outputs the monitoring results of the sound data and the image data; in the specific implementation process, the sound data and the image data are used for identification of equipment safety and production quality safety of the equipment in the area, and identification of workers to ensure the safety of the workers, so the amount of picture data of the workers in the data set is increased for the area where the workers may appear;

[0057] Step S2: The real-time acquired safety data is sent into the real-time monitoring model one by one for calculation to obtain a preliminary real-time monitoring result, the preliminary real-time monitoring result is verified according to the preliminary prediction result obtained by the prevention model to obtain a real-time monitoring result; the real-time monitoring model and the prevention model are optimized through joint training, the results of the models are verified between the models to improve the recognition accuracy of the models;

[0058] Step S3: The acquired time sequence safety data is sent into the prevention model for calculation to obtain a preliminary prediction result, the preliminary prediction result is optimized according to the real-time monitoring result to obtain a prediction result, and a prediction control result is obtained by matching according to the prediction result; emergency treatment is performed on the emergency cases in the real-time monitoring result and the prediction control result; the processing result of the safety data is saved to generate a safety training report;

[0059] The real-time monitoring model comprises:

[0060] A sensor data processing layer is configured to process sensor data, mainly gas concentration sensor data, temperature sensor data and humidity sensor data; the data of each sensor is first subjected to threshold determination to obtain an abnormality weight α, then an abnormality distinguishing value P is obtained according to the change value of the sensor data, the abnormality value K of the sensor is calculated through the abnormality distinguishing value and the abnormality weight, and the sensor data processing result is obtained according to the interval in which the abnormality value is located; the calculation formula of the abnormality value is:

[0061]

[0062] In the formula, α1 represents the abnormality weight of the data at the first moment, and P1 represents the change value of the sensor at the first moment;

[0063] A real-time monitoring layer is configured to classify according to the sensor data processing result, the sound monitoring result and the image monitoring result to obtain a preliminary real-time monitoring result.

[0064] A result optimization layer is configured to optimize the preliminary real-time monitoring result according to the preliminary prediction result to obtain a real-time monitoring result.

[0065] The prevention model comprises:

[0066] A data merging layer is configured to reconstruct the sensor data processing result, the sound monitoring result and the image monitoring result to obtain time sequence safety prediction data.

[0067] A safety prediction layer is configured to calculate according to the safety prediction data to obtain a preliminary prediction result.

[0068] A result optimization layer is configured to optimize the preliminary prediction result according to the real-time monitoring result to obtain a prediction result.

[0069] The joint training of the real-time monitoring model and the prevention model comprises:

[0070] An initial real-time monitoring model and an initial prevention model are constructed, the initial real-time monitoring model is constructed based on a deep convolutional neural network, the initial prevention model is constructed based on a long short-term neural network, the initial real-time monitoring model comprises one input layer, one data calculation layer, 16 hidden layers, and two output layers, the initial prevention model comprises one input layer, 36 hidden layers, and two output layers, the last four hidden layers of the two models are shared, and are used for optimizing the results;

[0071] A plurality of safety data are acquired, real-time abnormal results and prediction results are labeled, the labeled data are combined into a safety data set, and the safety data set is divided into a verification set, a test set, and a training set;

[0072] A certain proportion of data in the safety data training set and the safety data verification set are selected to train the initial real-time monitoring model and the initial prevention model, the real-time abnormal results and the prediction results are taken as targets, and an optimized real-time monitoring model and an optimized prevention model are obtained;

[0073] A certain proportion of data in the safety data test set is selected to train the optimized real-time monitoring model and the optimized prevention model, the output of the final output layer is taken as an accurate value, the two models are adjusted according to the accurate value, and a real-time monitoring model and a prevention model with a final accuracy that meets a standard are obtained;

[0074] The real-time monitoring result is verified according to the prediction result, and the verification comprises:

[0075] After the real-time monitoring model outputs a result, the result is verified according to a corresponding relationship between the prediction result and the real-time monitoring result in a real-time prediction association table, the real-time prediction association table stores a direct corresponding relationship and an indirect corresponding relationship between the prediction result and the real-time monitoring result, the direct corresponding relationship indicates that there is a strong correlation between the prediction result and the real-time monitoring result, the indirect corresponding relationship indicates that there is a weak correlation between the prediction result and the real-time monitoring result, if the verification result is a strong correlation, the verification is passed, if the verification result is a weak correlation, the verification is passed only if a plurality of verifications are weak correlations, and if the frequency of the verification result being no correlation exceeds a threshold value, the result is recorded, and the record is used to update the real-time prediction association table.

[0076] The safety training report is generated, and the generation comprises:

[0077] The safety data processing results in different time periods are acquired and processed, including sensor data, sound data, image data, and corresponding abnormality detection results and emergency handling records; the safety data is subjected to cluster analysis, similar safety events are classified by grouping according to the cluster results, and a cluster result of each type of event is generated; key words are extracted from key features in the cluster result, statistical analysis is performed according to the occurrence frequency of the key words, and report contents related to each type of safety hazard are generated; through a time series analysis method, based on historical data and time sequence characteristics of safety events, future possible safety risks are predicted, and the prediction results are included in the safety training report; according to the latest safety policy library and monitoring data, the report contents are dynamically adjusted, and individualized safety training plans for workshops in different regions are generated, including operation specifications and preventive measures for high-frequency abnormal situations in specific regions; the safety training report includes graphical display of safety data, and visually presents safety risks and data distribution in different regions; the safety training report can be generated periodically, and is optimized and adjusted according to the accumulation of historical data and the update of real-time monitoring data, to provide safety training suggestions that meet the actual situation of the workshop.

[0078] Step S4: The real-time monitoring results and the prediction control results are divided according to the data acquisition region and the safety category; according to the real-time monitoring results and the prediction control results in the same region, the regional control strategy and the regional early warning strategy are matched in the regional control and early warning strategy library, and the workshop is automatically controlled and warned;

[0079] Step S5: According to the real-time monitoring results and the prediction control results of the same safety category, the associated control strategy and the associated early warning strategy are matched in the associated control and early warning strategy library, and automatic control and early warning are performed;

[0080] The control and the warning are independent operations and have certain correlation, but can be independently performed, and the specific correlation is obtained through the strategy library;

[0081] The regional control and early warning strategy library and the associated control and early warning strategy library include:

[0082] The regional control and early warning strategy library stores specific control and early warning strategies for the real-time monitoring results and the prediction control results in the region; the control and early warning strategies in the regional control and early warning strategy library are more targeted, and can control and warn the abnormal situation in the current region in the first time;

[0083] The associated control and early warning strategy library stores specific control and early warning strategies for all real-time monitoring results and prediction control results of the same safety category; compared with the regional control and early warning strategy library, the associated control and early warning strategy library associates safety data analysis results between different regions, and obtains potential abnormal situation handling strategies through the association between regions.

[0084] Step S6: record the safety data and processing results in the workshop, which will be used to optimize the real-time monitoring model, prevention model, regional control warning strategy library and associated control warning strategy library.

[0085] The recorded safety data includes original safety data, real-time monitoring results, prediction control results, control strategies and warning strategies. Incorrect handling of abnormal situations in the saved records will be marked. Records exceeding the rated number of marks will be added to the regional control warning strategy library and the associated control warning strategy library. Correct records will be used as training data for the real-time monitoring model and the prevention model to optimize the models.

[0086] Embodiment 2, a safety monitoring system for a paint workshop, as shown in Figure 1 and Figure 2 , comprising the following modules:

[0087] The data acquisition module includes a data acquisition unit and a data processing unit. The data acquisition unit is used to acquire sensor data, sound data and image data of each area in the paint workshop. The data processing unit is used to input the sound data and image data into a pre-trained recognition model for abnormal identification to obtain sound monitoring results and image monitoring results. The sensor data, sound monitoring results and image monitoring results are used as safety data. The safety data is preprocessed and safety monitoring marks are added to the safety data;

[0088] The real-time monitoring module includes a real-time monitoring unit and a model optimization unit. The real-time monitoring unit includes a real-time monitoring model for calculating the real-time acquired safety data to obtain real-time monitoring results. The model optimization unit is used to optimize the real-time monitoring model and the prevention model. The real-time monitoring model and the prevention model are associated through joint training and record optimization, and the training set of the record optimization model is used;

[0089] The prevention module includes a prediction unit, a prediction control unit, an emergency handling unit and a safety training unit. The prediction unit is used to input the acquired safety data into the prevention model for calculation, and optimize the prediction results according to the real-time monitoring results to obtain the prediction results. The prediction control unit is used to match the prediction control results according to the prediction results. The emergency handling unit is used to handle the emergency in the real-time monitoring results and the prevention control results. The safety training unit is used to save the processing results of the safety data and generate a safety training report.

[0090] The early warning module and the control module are integrated in a control early warning module, the control early warning module further comprises a result dividing unit and a matching unit in addition to the early warning module and the control module, the result dividing unit is used for dividing the real-time monitoring result and the prediction control result according to the data acquisition area and the safety category; the matching unit is used for matching the regional control strategy and the regional early warning strategy in the regional control early warning strategy library according to the real-time monitoring result and the prediction control result in the same area, and matching the associated control strategy and the associated early warning strategy in the associated control early warning strategy library according to the real-time monitoring result and the prediction control result of the same safety category;

[0091] The control module is used for automatically controlling the workshop according to the matched regional control strategy and the associated control strategy;

[0092] The early warning module is used for automatically early warning the workshop according to the matched regional early warning strategy and the associated early warning strategy;

[0093] The recording module is used for recording the safety data and the processing result in the workshop, and the recording will be used for optimizing the real-time monitoring model, the prevention model, the regional control early warning strategy library and the associated control early warning strategy library.

[0094] It should be understood that, for those skilled in the art, improvements or changes can be made according to the above description, and all these improvements and changes shall belong to the protection scope of the appended claims of the present application. The parts not described in detail in the specification belong to the prior art known to those skilled in the art.

Claims

1. A method for safety monitoring of a production paint shop, characterized in that Comprise the following steps: Step S1: Obtain the structure information of the workshop, and divide the paint workshop into areas according to the structure information of the workshop, obtain the sensor data, sound data and image data of each area of the paint workshop, send the sound data and image data into the pre-trained identification model for abnormal identification, obtain the sound monitoring result and image monitoring result, take the sensor data, sound monitoring result and image monitoring result as safety data, preprocess the safety data, add safety monitoring marks to the safety data; The safety monitoring marks include safety categories, data collection areas, collection times and data types, and the safety categories include worker safety, equipment safety, environmental safety and production quality safety; Mark the source of data acquisition, the corresponding data category and the safety category to which the data belongs for the workshop area; Step S2: The real-time safety data obtained is sent into the real-time monitoring model one by one for calculation to obtain the preliminary real-time monitoring result, the preliminary real-time monitoring result is verified according to the preliminary prediction result obtained by the prevention model to obtain the real-time monitoring result; The real-time monitoring model and the prevention model are optimized through joint training, so that the results of the models are verified between each other, and the identification accuracy of the models is improved; Step S3: The time sequence safety data obtained is sent into the prevention model for calculation to obtain the preliminary prediction result, the preliminary prediction result is optimized according to the real-time monitoring result to obtain the prediction result, and the prediction control result is obtained by matching the prediction result; Emergency treatment is performed on the emergency cases in the real-time monitoring result and the prediction control result; Save the processing result of the safety data, and generate a safety training report; Step S4: The real-time monitoring result and the prediction control result are divided according to the data collection area and the safety category; According to the real-time monitoring result and the prediction control result in the same area, the area control strategy and the area early warning strategy are matched in the area control early warning strategy library, and the workshop is automatically controlled and warned; Step S5: According to the real-time monitoring result and the prediction control result of the same safety category, the associated control early warning strategy library is matched with the associated control strategy and the associated early warning strategy, and automatic control and early warning are performed; Step S6: Record the safety data and the processing result in the workshop, and the record will be used to optimize the real-time monitoring model, the prevention model, the area control early warning strategy library and the associated control early warning strategy library; The recorded safety data includes original safety data, real-time monitoring result, prediction control result, control strategy and early warning strategy, incorrect processing of the saved record is marked, records exceeding the rated number of marks are added to the area control early warning strategy library and the associated control early warning strategy library, and correct records are used as training data of the real-time monitoring model and the prevention model to optimize the model.

2. A method of safety monitoring of a production paint shop according to claim 1, characterized in that, The pre-trained identification model comprises: An independent identification model is trained for each source of sound data and image data, and each identification model is only used for identification of corresponding data; The identification model is constructed based on support vector machine, is trained and optimized through a plurality of labeled historical data, and the final output of the pre-trained identification model is the monitoring result of the sound data and the image data.

3. A method of safety monitoring of a production paint shop according to claim 1, characterized in that, The real-time monitoring model and the prevention model comprise: The real-time monitoring model comprises: A sensing data processing layer for processing sensor data, mainly gas concentration sensor data, temperature sensor data and humidity sensor data; the data of each sensor is first subjected to threshold determination to obtain an abnormality weight α, and then subjected to determination of an abnormality distinguishing value P according to the change value of the sensor data, and the abnormality value K of the sensor is calculated through the abnormality distinguishing value and the abnormality weight, and the sensing data processing result is obtained according to the interval in which the abnormality value is located; the calculation formula of the abnormality value is: In the formula, α1 represents the abnormality weight of the first sensor data, and P1 represents the abnormality distinguishing value of the first sensor data; A real-time monitoring layer for classifying according to the sensing data processing result, the sound monitoring result and the image monitoring result to obtain a preliminary real-time monitoring result; A result optimization layer for optimizing the preliminary real-time monitoring result according to the preliminary prediction result obtained in the prevention model to obtain a real-time monitoring result; The prevention model comprises: A data merging layer for reconstructing the sensing data processing result, the sound monitoring result and the image monitoring result to obtain time-series safety prediction data; A safety prediction layer for calculating according to the safety prediction data to obtain a preliminary prediction result; A result optimization layer for optimizing the preliminary prediction result according to the real-time monitoring result to obtain a prediction result.

4. The method for safety monitoring of a production spray booth according to claim 1, characterized in that, The joint training of the real-time monitoring model and the prevention model comprises: An initial real-time monitoring model and an initial prevention model are constructed, the initial real-time monitoring model is constructed based on a deep convolutional neural network, the initial prevention model is constructed based on a long short-term neural network, the initial real-time monitoring model comprises 1 input layer, 1 data calculation layer, 16 hidden layers and 2 output layers, the initial prevention model comprises 1 input layer, 36 hidden layers and 2 output layers, and the last 4 hidden layers of the two models are shared for optimizing the result; A plurality of safety data are obtained, and real-time abnormality results and prediction results are labeled, the labeled data are combined into a safety data set, and the safety data set is divided into a verification set, a test set and a training set; A certain proportion of data in the safety data training set and the safety data verification set are selected to train the initial real-time monitoring model and the initial prevention model, and the optimized real-time monitoring model and the optimized prevention model are obtained by taking the real-time abnormality result and the prediction result as the target respectively; A certain proportion of data in the safety data test set is selected to train the optimized real-time monitoring model and the optimized prevention model, and the output of the final output layer is taken as the correct proportion of actual labeling as the accuracy value, and the two models are adjusted in terms of hyperparameters according to the accuracy value, so that the real-time monitoring model and the prevention model with a final accuracy value reaching a standard are obtained.

5. The method for safety monitoring of a production spray booth according to claim 1, characterized in that, The regional control early warning strategy library and the association control early warning strategy library comprise: The regional control early warning strategy library stores specific control and early warning strategies for the real-time monitoring result and the prevention control result in a region; the control and early warning strategies in the regional control early warning strategy library are relatively targeted, and can control and warn the abnormal situation in the current region in the first time. The correlation control early warning strategy library stores specific control and early warning strategies for all real-time monitoring results and prevention control results of the same safety category. Compared with the regional control early warning strategy library, the correlation control early warning strategy library is associated with safety data analysis results between different regions, and potential abnormal situation processing strategies are obtained through the correlation between regions.

6. The method for safety monitoring of a production spray booth according to claim 1, characterized in that The verification of the real-time monitoring result according to the prediction result includes: After the real-time monitoring model outputs the result, the corresponding relationship between the prediction result and the real-time monitoring result in the real-time prediction correlation table is verified. The real-time prediction correlation table stores the direct and indirect corresponding relationships between the prediction result and the real-time monitoring result. The direct corresponding relationship indicates that there is a strong correlation between the prediction result and the real-time monitoring result, and the indirect corresponding relationship indicates that there is a weak correlation between the prediction result and the real-time monitoring result. If the verification result is a strong correlation, it is verified. If the verification result is a weak correlation, it needs to be verified for multiple times. If the frequency of the verification result without correlation is more than a threshold, it is recorded. The record is used to update the real-time prediction correlation table.

7. The method for safety monitoring of a production spray booth according to claim 1, characterized in that, Generating a safety training report includes: Obtaining and processing safety data processing results in different time periods, including sensor data, sound data, image data, and corresponding abnormal detection results and emergency handling records; clustering analysis is performed on the safety data, similar safety events are classified according to the clustering results, and clustering results of each type of event are generated; key words are extracted from the key features in the clustering results, and statistical analysis is performed according to the frequency of the key words to generate report content related to each type of safety hazard; through time series analysis method, based on historical data and time sequence characteristics of safety events, future possible safety risks are predicted, and the prediction results are included in the safety training report; the safety training report includes graphical display of safety data, and intuitively presents safety risks and data distribution in different regions.

8. A safety monitoring system for a paint shop, characterized in that The system applies the safety monitoring method of a paint shop according to any one of claims 1 to 7, including the following modules: The data acquisition module includes a data acquisition unit and a data processing unit. The data acquisition unit is used to obtain sensor data, sound data and image data of each region of the paint shop. The data processing unit is used to input the sound data and image data into the pre-trained recognition model for abnormal identification to obtain sound monitoring results and image monitoring results. The sensor data, sound monitoring results and image monitoring results are used as safety data, and the safety data is preprocessed by adding safety monitoring marks to the safety data. The real-time monitoring module includes a real-time monitoring unit and a model optimization unit. The real-time monitoring unit includes a real-time monitoring model for calculating the real-time acquired safety data to obtain real-time monitoring results. The model optimization unit is used to optimize the real-time monitoring model and the prevention model. The real-time monitoring model and the prevention model are associated through joint training and record optimization, and the training set of the record optimization model is used. The prevention module comprises a prediction unit, a prediction control unit, an emergency treatment unit and a safety training unit, the prediction unit is used for sending the obtained safety data into a prevention model for calculation, and optimizing the prediction result according to the real-time monitoring result to obtain a prediction result; the prediction control unit is used for matching a prediction control result according to the prediction result; the emergency treatment unit is used for treating the emergency in the real-time monitoring result and the prevention control result; The safety training unit is used for saving the processing result of the safety data, and generating a safety training report; The early warning module and the control module are integrated in a control early warning module, the control early warning module further comprises a result division unit and a matching unit in addition to the early warning module and the control module, the result division unit is used for dividing the real-time monitoring result and the prediction control result according to the data acquisition area and the safety category; the matching unit is used for matching the regional control strategy and the regional early warning strategy in the regional control early warning strategy library according to the real-time monitoring result and the prediction control result in the same area, and matching the associated control strategy and the associated early warning strategy in the associated control early warning strategy library according to the real-time monitoring result and the prediction control result of the same safety category; The control module is used for automatically controlling the workshop according to the matched regional control strategy and the associated control strategy; The early warning module is used for automatically warning the workshop according to the matched regional early warning strategy and the associated early warning strategy; The recording module is used for recording the safety data and the processing result in the workshop, and the recording will be used for optimizing the real-time monitoring model, the prevention model, the regional control early warning strategy library and the associated control early warning strategy library.

Images