Data processing method and system applied to climbing early warning analysis

Abstract

The embodiment of the invention relates to the technical field of data processing, in particular to a data processing method and system applied to climbing early warning analysis, and aims at obtaining a plurality of personnel action images and corresponding climbing structure stress state data under the same power production scene, and performing trend early warning detection by using a deep learning technology. And a climbing early warning decision network is obtained through debugging and optimization. And finally, the deep learning network and the climbing early warning decision network are debugged and improved cooperatively, so that the performance and the accuracy of the whole early warning system are improved. According to the method, more reliable and efficient safety early warning can be provided for climbing operation.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and more specifically, to a data processing method and system for application in high-altitude warning analysis. Background Art

[0002] In the power production and other industries that require high-altitude operations, ensuring the safety of operators is of utmost importance. With the development of technology, various sensors and monitoring devices are widely used at the work site to monitor the actions of high-altitude operators and the stress states of high-altitude structures in real time. These data provide rich information for predicting potential safety risks.

[0003] In recent years, deep learning technology has made remarkable progress in the fields of image processing and data analysis. Deep learning models, especially convolutional neural networks (CNNs), have been widely used in tasks such as image recognition, classification, and object detection. These models can learn and extract useful features from large amounts of data to make accurate predictions and classifications.

[0004] In high-altitude operation warning systems, traditional methods usually rely on preset rules and thresholds for safety assessment. However, this method often performs poorly when dealing with complex and variable actual scenarios. Summary of the Invention

[0005] To improve the technical problems existing in the related art, the present invention provides a data processing method and system for application in high-altitude warning analysis.

[0006] In a first aspect, an embodiment of the present invention provides a data processing method for application in high-altitude warning analysis, which is applied to an AI data processing system. The method includes: Obtaining X personnel action images corresponding to high-altitude warning monitoring information samples and a set of high-altitude structure stress state data for each personnel action image. The power production scenarios of the X personnel action images are the same, and each set of high-altitude structure stress state data includes at least one high-altitude structure stress state data, where X is a positive integer; Using a first deep learning network to perform trend warning detection on each personnel action image and the set of high-altitude structure stress state data of the personnel action image, to obtain a set of trend warning detection results corresponding to the set of high-altitude structure stress state data of the X personnel action images. Each set of trend warning detection results includes at least two trend warning detection results; Debugging a neural network to be debugged through the set of high-altitude structure stress state data of the X personnel action images and the set of trend warning detection results corresponding to the set of high-altitude structure stress state data of the X personnel action images, to obtain a high-altitude warning decision network, where the high-altitude warning decision network is used to identify the confidence weights of the trend warning detection results; Cooperatively debug and improve the first deep learning network and the high-altitude warning decision-making network to obtain an improved first deep learning network and an improved high-altitude warning decision-making network, which are used to perform high-altitude warning analysis on the high-altitude warning monitoring information set to be analyzed.

[0007] Preferably, each set of high-altitude structure stress state data includes first high-altitude structure stress state data and second high-altitude structure stress state data; using the first deep learning network to perform trend warning detection on each personnel action image and the set of high-altitude structure stress state data of the personnel action image, the obtained set of trend warning detection results corresponding to the set of high-altitude structure stress state data of the X personnel action images includes: Using the first deep learning network to perform trend warning detection on the u-th personnel action image and the first high-altitude structure stress state data of the u-th personnel action image to obtain a first trend warning detection result, where u is a positive integer not greater than X; Using the first deep learning network to perform trend warning detection on the u-th personnel action image and the second high-altitude structure stress state data of the u-th personnel action image to obtain a second trend warning detection result; Add the first trend warning detection result and the second trend warning detection result to the set of trend warning detection results corresponding to the set of high-altitude structure stress state data of the u-th personnel action image.

[0008] Preferably, each set of high-altitude structure stress state data includes third high-altitude structure stress state data; using the first deep learning network to perform trend warning detection on each personnel action image and the set of high-altitude structure stress state data of the personnel action image, the obtained set of trend warning detection results corresponding to the set of high-altitude structure stress state data of the X personnel action images includes: Using the first deep learning network to perform trend warning detection on the u-th personnel action image and the third high-altitude structure stress state data of the u-th personnel action image to obtain a third trend warning detection result, where u is a positive integer not greater than X; Add the third trend warning detection result to the set of trend warning detection results corresponding to the set of high-altitude structure stress state data of the u-th personnel action image; Among them, the set of trend warning detection results corresponding to the set of high-altitude structure stress state data of the u-th personnel action image further includes a fourth trend warning detection result, which is obtained by using the second deep learning network to perform trend warning detection on the u-th personnel action image and the third high-altitude structure stress state data of the u-th personnel action image.

[0009] Preferably, the collaborative debugging and improvement of the first deep learning network and the altitude warning decision network to obtain an improved first deep learning network and an improved altitude warning decision network includes: Obtain a first long short-term memory network, a second long short-term memory network and a decision tree network, wherein the first long short-term memory network and the second long short-term memory network are updated based on the first deep learning network, and the decision tree network is updated based on the altitude warning decision network; The linkage monitoring information is processed by the first long short-term memory network, the second long short-term memory network, the decision tree network and the climbing warning decision network to obtain a first network quality evaluation and a second network quality evaluation; the linkage monitoring information is obtained by integrating the target person action image, the climbing structure stress state data set of the target person action image, and the trend warning detection result set corresponding to the climbing structure stress state data set of the target person action image, and the target person action image is sampled from the X person action images; The first long short-term memory network and the decision tree network are collaboratively debugged according to the first network quality evaluation and the second network quality evaluation to obtain an improved first deep learning network and an improved altitude warning decision network.

[0010] Preferably, the linkage monitoring information is processed by the first long short-term memory network, the second long short-term memory network, the decision tree network and the altitude warning decision network to obtain a first network quality evaluation and a second network quality evaluation, including: Obtaining a first viewpoint output weight, a second viewpoint output weight, a first credible factor and a first decision coefficient, wherein the first viewpoint output weight is obtained by using the first long short-term memory network to perform first feature mining and identification on the linkage monitoring information, the second viewpoint output weight is obtained by using the second long short-term memory network to perform second feature mining and identification on the linkage monitoring information, the first credible factor is obtained by using the altitude warning decision network to perform confidence weight identification on the linkage monitoring information, and the first decision coefficient is obtained by using the decision tree network to perform warning decision identification on the linkage monitoring information; Determining a consistency comparison result through the first viewpoint output weight and the second viewpoint output weight, wherein the consistency comparison result is used to characterize the difference between the first viewpoint output weight and the second viewpoint output weight; Adjusting the first trust factor according to the consistency comparison result to obtain a second network quality evaluation; Determine the first network quality evaluation based on the second network quality evaluation and the first decision coefficient.

[0011] Preferably, the collaborative debugging of the first long short-term memory network and the decision tree network according to the first network quality evaluation and the second network quality evaluation to obtain the improved first deep learning network and the improved high-altitude warning decision network includes: Combine the first view output weight and the second view output weight, and the first network quality evaluation to generate a first training error variable; Obtain a second decision coefficient, and generate a second training error variable through the second decision coefficient and the second network quality evaluation. The second decision coefficient is obtained by the decision tree network after loop debugging to perform warning decision recognition on the training subset, and the training subset is obtained by disassembling the linkage monitoring information; Generate a third training error variable based on the first training error variable and the second training error variable; Perform collaborative debugging on the first long short-term memory network and the decision tree network through the third training error variable to obtain the improved first deep learning network and the improved high-altitude warning decision network.

[0012] Preferably, the obtaining of X human action images corresponding to the high-altitude warning monitoring information sample includes: Obtain the high-altitude warning monitoring information sample; If the power production scenario of the high-altitude warning monitoring information sample is different from the set power production scenario, perform a power production scenario mapping on the high-altitude warning monitoring information sample to obtain the high-altitude warning monitoring information sample under the set power production scenario; Perform a block processing on the high-altitude warning monitoring information sample under the set power production scenario using the information block rule to obtain Y monitoring sample sub-blocks, where Y is a positive integer not greater than X; Generate the X human action images corresponding to the high-altitude warning monitoring information sample based on the Y monitoring sample sub-blocks.

[0013] Preferably, each monitoring sample sub-block is associated with an annotation information; the generating of the X human action images corresponding to the high-altitude warning monitoring information sample based on the Y monitoring sample sub-blocks includes: Obtain the action noise feature threshold; If there is a monitoring sample sub-block among the Y monitoring sample sub-blocks whose action feature value is not greater than the action noise feature threshold, determine this monitoring sample sub-block as the human action image; If there is a monitoring sample sub-block among the Y monitoring sample sub-blocks whose action feature value is greater than the action noise feature threshold, then disassemble this monitoring sample sub-block to obtain at least two human action images, and add the annotation information associated with this monitoring sample sub-block to the at least two human action images, and the action feature value of each human action image is not greater than the action noise feature threshold.

[0014] Preferably, the method for debugging the neural network to be debugged by using the set of high-altitude structure stress state data of the X human action images and the set of trend warning detection results corresponding to the set of high-altitude structure stress state data of the X human action images to obtain a high-altitude warning decision network includes: Arbitrarily select a first trend warning detection result and a second trend warning detection result from the set of trend warning detection results corresponding to the high-altitude structure stress state data set of the u-th human action image, the first trend warning detection result and the second trend warning detection result are different, and u is a positive integer not greater than X; Use the neural network to be debugged to respectively identify the confidence weights of the first trend warning detection result and the second trend warning detection result based on the high-altitude structure stress state data set of the u-th human action image, and obtain a second credibility factor and a third credibility factor; Generate a fourth training error variable based on the difference between the second credibility factor and the third credibility factor; Debug the neural network to be debugged according to the fourth training error variable to obtain a high-altitude warning decision network.

[0015] Preferably, the method further includes: Obtain Z pieces of high-altitude warning monitoring information to be analyzed and the high-altitude structure stress state data of each piece of high-altitude warning monitoring information to be analyzed. The power production scenario of the Z pieces of high-altitude warning monitoring information to be analyzed is a set power production scenario, and Z is a positive integer; Use the improved first deep learning network to perform trend warning detection on each piece of high-altitude warning monitoring information to be analyzed and the high-altitude structure stress state data of this piece of high-altitude warning monitoring information to be analyzed, and obtain the trend warning detection result corresponding to each high-altitude structure stress state data; Select according to the improved high-altitude warning decision network the trend warning detection results corresponding to the high-altitude structure stress state data of the Z pieces of high-altitude warning monitoring information to be analyzed, and obtain v trend warning detection results that meet the set requirements, where v is a positive integer not greater than Z; Jointly import the v trend warning detection results that meet the set requirements, as well as the high-altitude structure stress state data and the high-altitude warning monitoring information to be analyzed corresponding to each trend warning detection result that meets the set requirements, into the cloud service space.

[0016] Preferably, the obtaining of Z to-be-analyzed high-altitude warning monitoring information and the force state data of the high-altitude structure for each to-be-analyzed high-altitude warning monitoring information includes: Obtaining Z to-be-analyzed high-altitude warning monitoring information; If there is target to-be-analyzed high-altitude warning monitoring information in the Z to-be-analyzed high-altitude warning monitoring information whose power production scenario is different from the set power production scenario, then perform power production scenario mapping on the target to-be-analyzed high-altitude warning monitoring information to obtain Z to-be-analyzed high-altitude warning monitoring information under the set power production scenario; According to the elements of the to-be-analyzed high-altitude warning monitoring information, associate the force state data of the high-altitude structure with each to-be-analyzed high-altitude warning monitoring information under the set power production scenario.

[0017] Preferably, the selecting of v trend warning detection results that meet the set requirements from the trend warning detection results corresponding to the force state data of the high-altitude structure of the Z to-be-analyzed high-altitude warning monitoring information according to the improved high-altitude warning decision network includes: Obtaining a decision threshold; Using the improved high-altitude warning decision network to perform warning decision identification on each trend warning detection result corresponding to the force state data of the high-altitude structure of each to-be-analyzed high-altitude warning monitoring information to obtain a third decision coefficient for each trend warning detection result; If the third decision coefficient of any trend warning detection result is higher than the decision threshold, then determine the trend warning detection result as a trend warning detection result that meets the set requirements.

[0018] In a second aspect, the present invention also provides an AI data processing system, including: a memory for storing program instructions and data; a processor for being coupled to the memory and executing the instructions in the memory to implement the method as described above.

[0019] In a third aspect, the present invention also provides a computer storage medium containing instructions, which when executed on a processor, implement the method as described above.

[0020] In the actual application process, by training a deep learning model to identify unsafe behaviors and predict potential risks, the accuracy and reliability of the warning system can be significantly improved. However, a single deep learning model may not be able to fully utilize all available information. For example, when identifying unsafe behaviors in high-altitude operations, the model may need to consider both the actions of personnel and the force state of the high-altitude structure. In addition, different deep learning models may have different advantages and limitations.

[0021] Based on this, embodiments of the present invention obtain multiple human action images and their corresponding force states of the climbing structure in the same power production scenario, use deep learning technology for trend warning detection, and obtain a climbing warning decision network through debugging and optimization. Finally, the deep learning network and the climbing warning decision network are co-debugged and improved to improve the performance and accuracy of the entire warning system. This method can provide a more reliable and efficient safety warning for climbing operations. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 FIG. is a schematic flowchart of a data processing method for climbing warning analysis provided by an embodiment of the present invention.

[0023] Figure 2 FIG. is a block diagram of the structure of an AI data processing system 300 provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0024] The following will describe the technical solutions in the present invention in conjunction with the accompanying drawings.

[0025] The method embodiments provided by the embodiments of the present invention can be executed in an AI data processing system, a computer device, or a similar computing device. Taking running on an AI data processing system as an example, the AI data processing system may include one or more processors (the processor may include, but is not limited to, a processing device such as a microprocessor MCU or a programmable logic device FPGA) and a memory for storing data. Optionally, the above AI data processing system may further include a transmission device for communication functions. Those of ordinary skill in the art can understand that the above structure is only illustrative and does not limit the structure of the above AI data processing system. For example, the AI data processing system may further include more or fewer components than those shown above, or have a different configuration from those shown above.

[0026] The memory can be used to store computer programs, for example, software programs and modules of application software, such as the computer program corresponding to a data processing method for climbing warning analysis in an embodiment of the present invention. The processor executes various functional applications and data processing by running the computer program stored in the memory, that is, implements the above method. The memory may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memories, or other non-volatile solid-state memories. In some instances, the memory may further include a memory remotely provided relative to the processor, and these remote memories can be connected to the AI data processing system through a network. Examples of the above network include, but are not limited to, the Internet, an enterprise intranet, a local area network, a mobile communication network, and combinations thereof.

[0027] The transmission device is used to receive or send data via a network. Specific examples of the above-mentioned network may include a wireless network provided by a communication provider of an AI data processing system. In one example, the transmission device includes a network adapter (abbreviated as NIC), which can be connected to other network devices through a base station so as to communicate with the Internet. In one example, the transmission device can be a radio frequency (RF) module, which is used to communicate with the Internet wirelessly.

[0028] Based on this, please refer to Figure 1 , Figure 1 which is a schematic flowchart of a data processing method for altitude warning analysis provided by an embodiment of the present invention. This method is applied to an AI data processing system and may further include steps 110 - 140.

[0029] Step 110: Obtain X personnel action images corresponding to altitude warning monitoring information samples and a set of altitude structure stress state data for each personnel action image. The X personnel action images have the same power production scenario, and each set of altitude structure stress state data includes at least one altitude structure stress state data, where X is a positive integer.

[0030] In the altitude operation area of a power production enterprise, the AI data processing system first collects data through high-definition cameras and various sensors. These data include real-time action images of X staff members performing altitude operations in the same power production scenario, as well as real-time stress state data of the altitude structure during the altitude operation. The action images capture every action detail of the staff members during the altitude process, such as climbing, standing, operating, etc. The stress state data is obtained through sensors installed on the altitude structure and includes data in multiple dimensions such as stress, strain, and vibration. These data reflect the real-time stress conditions of the altitude structure under different actions. The collected data is organized into individual monitoring information samples, and each sample contains a personnel action image and a corresponding set of altitude structure stress state data.

[0031] Step 120: Use the first deep learning network to perform trend warning detection on each personnel action image and the set of altitude structure stress state data of this personnel action image, and obtain a set of trend warning detection results corresponding to the set of altitude structure stress state data of the X personnel action images. Each set of trend warning detection results includes at least two trend warning detection results.

[0032] Next, the AI data processing system uses the first deep learning network to perform trend warning detection on these monitoring information samples. This deep learning network is pre-trained and specifically used to identify potential safety hazards in images and data. The network first analyzes the personnel action images to identify the specific actions of the staff and potential risk behaviors. At the same time, the network also combines the stress state data to determine whether the stress on the climbing structure is abnormal under different actions, whether there is structural damage or the risk of approaching the stress critical point. After in-depth analysis, the first deep learning network generates a set of trend warning detection results for each monitoring information sample. This result set contains multiple warning detection results, each corresponding to a potential safety hazard and its possible development trend.

[0033] Step 130: Debug the neural network to be debugged through the set of stress state data of the climbing structure of the X personnel action images and the set of trend warning detection results corresponding to the set of stress state data of the climbing structure of the X personnel action images, to obtain a climbing warning decision network, which is used to identify the confidence weights of the trend warning detection results.

[0034] After obtaining a large number of trend warning detection results, the AI data processing system starts to train a new neural network - the climbing warning decision network. The purpose of this network is to identify the confidence weights of different warning detection results, that is, to judge which warning results are reliable and which may be false alarms. During the training process, the system uses a large number of monitoring information samples and corresponding warning detection results as input data. By continuously adjusting the network parameters and optimizing the algorithm, the climbing warning decision network gradually learns how to accurately evaluate the confidence of different warning results. After training, this network will be able to automatically assign a confidence weight to each warning result, helping the enterprise to more accurately judge which warnings need to be focused on.

[0035] Step 140: Coordinate and debug and improve the first deep learning network and the climbing warning decision network to obtain an improved first deep learning network and an improved climbing warning decision network, which are used to perform climbing warning analysis on the climbing warning monitoring information set to be analyzed.

[0036] The last step is to coordinate and debug and improve the first deep learning network and the climbing warning decision network. The purpose of this step is to further improve the performance of the two networks so that they can more accurately identify potential safety hazards and give more reliable warning results. During the debugging process, the system will consider the output results of the two networks at the same time and adjust their parameters and algorithms according to the actual situation. Through continuous iteration and optimization, the performance of the two networks will gradually improve, thus providing the enterprise with more accurate and timely climbing warning analysis services.

[0037] In this way, through the combined application of deep learning networks and neural networks, a comprehensive early warning analysis of high-altitude operations in the power production scenario is achieved. It can not only identify the action risks of workers, but also monitor the stress state of high-altitude structures in real time, providing comprehensive safety guarantees for enterprises.

[0038] To further understand the entire technical solution, the following provides corresponding noun explanations and detailed introductions for each of the above steps 110 - 140.

[0039] Regarding step 110, the high-altitude early warning monitoring information sample is a comprehensive data unit that contains all relevant information collected during high-altitude operations at a specific time and location. This information is crucial for subsequent safety analysis and early warning. A typical high-altitude early warning monitoring information sample may include the action images of workers at a certain time point, the environmental conditions at that time, the detailed data of the high-altitude equipment used, and any other safety or operation records related to this high-altitude operation. These information samples are the basis for the AI data processing system to conduct risk analysis and early warning.

[0040] The personnel action image refers to the real-time action pictures of workers captured by a high-definition camera during high-altitude operations. These images not only record the specific actions of workers, such as climbing and operating equipment, but also reflect their body postures, action frequencies, and possible fatigue levels. Personnel action images are an important basis for evaluating the safety and standardization of workers' behaviors. Through the analysis of deep learning networks, unsafe behaviors or actions can be identified and early warnings can be issued in a timely manner.

[0041] The high-altitude structure stress state data set is a data set composed of multiple data points, reflecting the real-time stress conditions of the high-altitude structure during workers' operations. These data are collected through various sensors installed on the high-altitude structure, including but not limited to stress sensors, strain sensors, and vibration sensors, etc. This data set provides key information on the stability and safety of the high-altitude structure under different actions and loads. By analyzing this data, the AI system can predict possible problems with the structure, such as overload, stress concentration, or fatigue damage, etc., and issue early warnings in a timely manner.

[0042] The power production scenario refers to a specific working environment in power production enterprises, where working at heights is one of the common activities. The power production scenario usually includes facilities such as power stations, substations, and power transmission and distribution lines. These facilities need to be maintained and repaired regularly, so working at heights is indispensable. This scenario has its unique characteristics and challenges, such as complex equipment layouts, the presence of high-voltage electricity, and variable weather conditions, etc. When working at heights in such an environment, it is necessary to strictly comply with safety regulations and ensure that all equipment and structures are in good working condition to prevent accidents from occurring.

[0043] It can be seen that the terms such as the sample of working-at-heights warning monitoring information, the images of personnel actions, the data set of the stress state of the working-at-heights structure, and the power production scenario are all closely related, and they together constitute the core elements of the working-at-heights safety warning system in power production enterprises.

[0044] When performing step 110, the primary task is to obtain key monitoring information to provide a data basis for subsequent warning analysis.

[0045] Specifically, the AI data processing system will first connect to the enterprise's data center or database and retrieve and extract samples of monitoring information related to working-at-heights warnings from it. These samples are pre-stored and contain X images of personnel actions (X is a positive integer representing the number of images) taken in the same power production scenario. These images capture various actions of the staff when performing working-at-heights operations, such as climbing, standing, operating tools or equipment, etc. Due to the particularity of the power production scenario, such as complex equipment layouts, the possible high-voltage electrical environment, and variable weather conditions, these action images are crucial for evaluating the behavioral safety and standardization of the staff.

[0046] In addition to the images of personnel actions, the AI data processing system will also obtain the data set of the stress state of the working-at-heights structure corresponding to each image of personnel actions. These data are collected in real time by various sensors installed on the working-at-heights structure, including but not limited to stress sensors, strain sensors, and vibration sensors. The sensors will record the real-time stress conditions of the working-at-heights structure under different actions and loads, such as stress distribution, deformation, vibration frequency, etc., thus forming a data set of stress states containing multiple data points. These data are the key basis for evaluating the stability and safety of the working-at-heights structure.

[0047] Importantly, all these images of personnel actions and the data set of the stress state of the working-at-heights structure are collected in the same power production scenario. This means that they reflect similar environmental conditions and operation requirements, making subsequent data analysis and comparison more accurate and meaningful.

[0048] After obtaining this data, the AI data processing system will perform preliminary data cleaning and preprocessing tasks to ensure the accuracy and consistency of the data. This may include steps such as removing noise, filling in missing values, and performing data normalization. The preprocessed data will be used for subsequent deep learning network analysis and training of the early warning decision-making network.

[0049] Thus, step 110 is the basis for the AI data processing system to conduct altitude warning analysis. It ensures that the system can obtain accurate and comprehensive monitoring information, providing solid data support for subsequent warning detection, network training, and collaborative debugging and improvement.

[0050] Regarding step 120, the first deep learning network is a neural network model based on deep learning. It is specifically used to process and analyze a large amount of image and data information to identify potential safety hazards. This network model may adopt architectures such as convolutional neural networks (CNNs) or recurrent neural networks (RNNs), and has powerful feature extraction and pattern recognition capabilities. In the altitude warning system, the first deep learning network is trained to identify unsafe behaviors in personnel movement images and detect abnormal patterns from the data of the stress state of the altitude structure. For example, the first deep learning network can process the movement images of workers during altitude work. By learning the features in the images, such as the postures of the workers, movement frequencies, and tools held, it can identify potential safety risks. At the same time, it can also analyze the stress data of the altitude structure and detect abnormal stress distributions or deformation patterns, which may be precursors to structural damage or failures.

[0051] Performing trend warning detection is a process of using the first deep learning network to deeply analyze the collected data to predict and identify potential safety problems and their development trends. This is not just an assessment of the current state, but more importantly, a prediction and warning of what may happen in the future. In the altitude work warning system, trend warning detection means analyzing historical data and real-time data to predict potential safety hazards during altitude work and issuing warnings in a timely manner. For example, if the first deep learning network finds that the operations of workers are not standardized or there are signs of fatigue when processing personnel movement images, the system will perform trend warning detection, analyze the possible consequences if this behavior continues, and issue a warning in advance. Similarly, if the stress state data shows that the stress of a certain part of the altitude structure continues to increase, the system will also perform trend analysis to predict whether there may be structural damage and take timely measures to prevent accidents.

[0052] The set of trend warning detection results is the sum of a series of warning information obtained after analysis by the first deep learning network. This set contains multiple detection results, and each result corresponds to a potential safety hazard and its possible development trend. These results may include, but are not limited to: warnings of unsafe behaviors of staff, warnings of abnormal forces on climbing structures, warnings of abnormal changes in environmental temperature or humidity, etc. For example, during a climbing operation, the first deep learning network may detect multiple potential problems, such as staff not wearing safety belts, climbing ladders shaking, and abnormal structural stress at a certain location. These detection results will be integrated into a set of trend warning detection results for managers to refer to and make decisions. Through such a set, enterprises can comprehensively understand the current safety status of climbing operations and take timely measures to prevent accidents.

[0053] In step 120, the AI data processing system will use the first deep learning network to deeply analyze each personnel action image and its corresponding set of climbing structure force state data obtained in step 110 to perform trend warning detection. This step is a key link to ensure the safety of climbing operations, and it aims to identify and predict potential safety hazards through deep learning technology.

[0054] First, the AI data processing system will take each personnel action image and the corresponding set of climbing structure force state data as inputs and pass them to the already trained first deep learning network. This network model has powerful feature extraction and pattern recognition capabilities and can effectively mine safety-related information from images and data.

[0055] When processing personnel action images, the first deep learning network will carefully analyze the actions, postures, operation methods, etc. of the staff to identify whether there are unsafe behaviors or operations. For example, it can detect whether the staff is wearing necessary safety equipment, such as safety belts, helmets, etc., and can also judge whether their actions are standardized and whether there are situations of excessive fatigue or operation errors.

[0056] At the same time, the network will also analyze in combination with the force state data of the climbing structure. These data reflect the real-time force conditions of the climbing structure under different actions and loads and are crucial for evaluating the safety and stability of the structure. The first deep learning network will carefully study these data to find abnormal stress distributions, deformation patterns, or vibration frequencies, etc., which are signs of structural damage or approaching the force critical point.

[0057] After in-depth analysis and processing, the first deep learning network generates a set of trend warning detection results for each personnel action image and its corresponding data set of the force-bearing state of the climbing structure. This set contains multiple trend warning detection results, each corresponding to a potential safety hazard and its possible development trend. For example, one result may indicate that the staff member is not wearing a safety belt during operation, which increases the risk of falling; another result may warn of abnormal stress in a certain climbing structure, which may lead to structural damage if not dealt with in time.

[0058] These trend warning detection results not only provide insights into the current safety situation but also predict potential future problems. This enables enterprises to take timely measures, such as strengthening training, replacing equipment, or carrying out structural reinforcement, to prevent accidents from occurring.

[0059] Based on this, in step 120, by using the first deep learning network to deeply analyze the personnel action images and the data of the force-bearing state of the climbing structure, a set of trend warning detection results is generated. This set provides valuable safety warning information for enterprises, helps to detect and handle potential safety hazards in a timely manner, and ensures the safe conduct of climbing operations.

[0060] Regarding step 130, the neural network to be debugged refers to a neural network model that needs to be debugged and optimized to ensure its performance and accuracy before being put into actual use. This model may be an initial version or a version that needs to be improved after problems are found in actual use. In the climbing operation warning system, the neural network to be debugged may be a model for identifying unsafe behaviors of staff members or abnormal states of climbing structures. Through debugging, the parameters and structure of the network can be adjusted to improve its recognition accuracy of unsafe behaviors and abnormal states. For example, in the power production scenario, the neural network to be debugged may be trained to identify unsafe behaviors such as staff members not wearing safety belts and irregular operations, or to detect abnormal stress and deformation of climbing structures. During the debugging process, real power production scenario data can be used to verify the performance of the network, and corresponding adjustments and optimizations can be made according to the verification results.

[0061] The high-altitude warning decision-making network is a neural network model specifically designed to process warning information related to high-altitude operations. This network model is designed to automatically make warning decisions based on input data (such as personnel movement images, high-altitude structure stress state data, etc.). It may be a deep learning model with powerful data processing and pattern recognition capabilities. In high-altitude operations in power production, the high-altitude warning decision-making network will receive the data analysis results from the first deep learning network, etc., and make decisions on whether to issue a warning based on this data. For example, if the network detects unsafe behaviors of the staff or abnormal states of the high-altitude structure, it will automatically trigger the warning system to notify relevant personnel to take necessary measures to prevent accidents.

[0062] The confidence weight of the trend warning detection result is a numerical index representing the reliability of the detection result. When the neural network conducts trend warning detection, each detection result will be assigned a confidence weight to indicate the reliability of the result. This weight may be calculated based on multiple factors, including data accuracy, model confidence, statistical distribution of historical data, etc. In the high-altitude operation warning system in power production, the confidence weight of the trend warning detection result is very important for decision-makers. A detection result with a high confidence weight means that the result is more reliable and requires more attention and corresponding preventive measures. For example, if the confidence weight of a certain trend warning detection result is very high, indicating that there may be serious unsafe behaviors of the staff, then the decision-maker needs to take immediate action to correct this behavior to ensure the safety of the staff.

[0063] It can be seen that the neural network to be debugged, the high-altitude warning decision-making network, and the confidence weight of the trend warning detection result are all very important concepts and components in the high-altitude operation warning system in power production. They jointly ensure the effectiveness and reliability of the system, providing strong protection for the safety of the staff.

[0064] In step 130, the AI data processing system will use the set of high-altitude structure stress state data of X personnel movement images obtained in the previous steps, as well as the corresponding set of trend warning detection results of these data sets, to debug and optimize the neural network to be debugged, so as to obtain the high-altitude warning decision-making network. This decision-making network will be used to identify the confidence weight of the trend warning detection result, providing an important basis for subsequent warning decisions.

[0065] First of all, the AI data processing system will integrate the data obtained in step 110 and step 120, including the set of high-altitude structure stress state data of X personnel movement images and the corresponding set of trend warning detection results. These data sets will be used as the basic data for debugging the neural network.

[0066] Next, the system will initiate the debugging process of the neural network to be debugged. This process includes optimizing and adjusting the network parameters, structure, and learning algorithm to improve the recognition accuracy of the neural network for trend warning detection results and the judgment ability of confidence weights. During the debugging process, the AI data processing system will continuously input the basic data into the neural network to be debugged, and gradually adjust parameters such as the weights and biases of the network by comparing the differences between the network output and the actual trend warning detection results.

[0067] During the debugging process, the AI data processing system will also apply various optimization algorithms, such as the gradient descent method, backpropagation algorithm, etc., to minimize the error between the network output and the actual results. At the same time, the system will also make appropriate adjustments to the network structure, such as adding or reducing hidden layers, adjusting the connection methods of neurons, etc., to further improve the performance of the network.

[0068] After repeated debugging and optimization, the neural network to be debugged will gradually adapt to and accurately identify the data of the force state of the climbing structure and the trend warning detection results. When the performance of the network reaches the predetermined standard, the debugging process ends, and the neural network at this time is named the climbing warning decision-making network.

[0069] The climbing warning decision-making network not only has the ability to recognize trend warning detection results but also can assign a confidence weight to each detection result. This confidence weight reflects the degree of trust of the network in the detection result, and the higher the weight, the higher the degree of trust of the network in this result. The introduction of the confidence weight makes the subsequent warning decisions more accurate and reliable.

[0070] It can be understood that in step 130, by using the basic data to debug and optimize the neural network to be debugged, the climbing warning decision-making network is obtained. This network not only improves the recognition accuracy of trend warning detection results but also can assign a confidence weight to each result, providing strong support for subsequent warning decisions.

[0071] Regarding step 140, collaborative debugging improvement refers to the process of synchronously debugging and optimizing among multiple systems or modules to improve the overall performance and accuracy. In the context of the power production and high-altitude operation warning system, collaborative debugging improvement specifically refers to jointly debugging the first deep learning network and the high-altitude warning decision-making network to ensure seamless connection and efficient cooperation between the two. This improvement method involves meticulous adjustment of the network's parameters, structure, and algorithms to achieve higher precision and response speed in identifying unsafe behaviors and predicting potential risks. For example, during the collaborative debugging improvement process, it may be found that the first deep learning network has errors in identifying certain specific types of unsafe behaviors, so network parameters need to be adjusted or the structure optimized accordingly. At the same time, it is also necessary to ensure that the high-altitude warning decision-making network can accurately make timely and effective warning decisions based on the output of the first deep learning network.

[0072] The improved first deep learning network refers to a deep learning model with higher performance and more accurate recognition ability after collaborative debugging improvement. This network has been optimized through a large amount of training data and a refined debugging process, enabling it to exhibit higher precision and stability in identifying unsafe behaviors in high-altitude operations and predicting potential risks. For example, the improved first deep learning network may be able to more accurately identify subtle movement changes of workers during high-altitude operations, such as not wearing a safety belt, operating errors, etc., and promptly transmit this information to the high-altitude warning decision-making network. In addition, it may also have stronger generalization ability and be able to maintain stable recognition performance in different high-altitude operation scenarios.

[0073] The improved high-altitude warning decision-making network refers to a neural network model with more efficient and accurate warning decision-making ability after collaborative debugging improvement. This network can receive data from the improved first deep learning network and quickly and accurately make a judgment on whether to issue a warning. Specifically, the improved high-altitude warning decision-making network may have higher processing speed and lower false alarm rate. It can quickly analyze and judge the safety status of the current high-altitude operation after receiving the data transmitted from the first deep learning network, and then decide whether to trigger the warning system according to the preset warning threshold and confidence weight. This improved network can greatly enhance the overall performance and reliability of the high-altitude operation warning system.

[0074] In step 140, the AI data processing system will perform a key task, that is, to conduct collaborative debugging improvement on the first deep learning network and the high-altitude warning decision-making network. This process aims to improve the performance and accuracy of the two networks, so as to be able to conduct more accurate high-altitude warning analysis on the set of high-altitude warning monitoring information to be analyzed.

[0075] At the beginning of the collaborative debugging and improvement, the AI data processing system will first evaluate the performance of the first deep learning network and the height warning decision network in the previous steps. This includes analyzing the accuracy, speed and stability of the network when processing the images of personnel movements and the stress state data of the height structure. The system will also check the network's ability to identify unsafe behaviors and predict potential risks, as well as whether there are false positives or negatives.

[0076] Next, the AI data processing system will undergo a series of debugging operations. For the first deep learning network, this may include adjusting the number of network layers, the number of neurons, the learning rate and other parameters, as well as optimizing the network's activation function and loss function. These adjustments are designed to improve the network's ability to identify unsafe behaviors and potential risks, while reducing the possibility of false positives and false negatives.

[0077] For the height warning decision network, debugging may focus on improving decision logic, optimizing the calculation method of confidence weights, and increasing the network's response speed to trend warning detection results. These improvements help ensure that the network can make warning decisions faster and more accurately, and promptly notify relevant personnel to take necessary preventive measures.

[0078] During the collaborative debugging and improvement process, the AI data processing system will continue to input data from actual scenarios into the two networks, and further fine-tune the parameters and structure of the networks by comparing the differences between the network outputs and the actual results. This process will continue until the performance of both networks reaches the predetermined standards.

[0079] When the collaborative debugging and improvement is completed, the AI data processing system will get the improved first deep learning network and the improved height warning decision network. These two networks will have higher accuracy and stability in identifying unsafe behaviors and predicting potential risks, providing strong support for subsequent height warning analysis.

[0080] It can be seen that step 140 is a key link to improve the performance and accuracy of the height warning system. By coordinating and improving the two core networks, the AI data processing system can more effectively analyze the height warning monitoring information set to be analyzed, thereby timely and accurately discovering potential safety hazards and issuing warnings.

[0081] Based on the above, embodiments of the present invention can construct a rich and representative dataset by obtaining X personnel action images and their corresponding force state data sets of the climbing structure in the same power production scenario. This helps the deep learning network to more accurately learn and identify unsafe behaviors and abnormal structural forces during climbing operations, thereby improving the accuracy of the warning system. Using the first deep learning network to perform trend warning detection on each personnel action image and its corresponding force state data set of the climbing structure, multiple trend warning detection results can be obtained. This multiple detection result method increases the reliability and stability of the warning system and reduces the possibility of false alarms and missed alarms. By using a large amount of action image data and the corresponding trend warning detection results to debug the neural network to be debugged, the obtained climbing warning decision network can more accurately identify the confidence weights of the trend warning detection results. This means that the system can more accurately evaluate the reliability of different detection results and thus make more reasonable warning decisions. By jointly debugging and improving the first deep learning network and the climbing warning decision network, not only can the performance of a single network be improved, but also the cooperation efficiency between the two networks can be optimized. This enables the entire warning system to more quickly and accurately complete climbing warning analysis when processing the climbing warning monitoring information set to be analyzed, further improving the practicality and efficiency of the system.

[0082] In some alternative embodiments, each force state data set of the climbing structure includes first force state data of the climbing structure and second force state data of the climbing structure; the using of the first deep learning network to perform trend warning detection on each personnel action image and the force state data set of the climbing structure of this personnel action image to obtain the trend warning detection result set corresponding to the force state data set of the climbing structure of the X personnel action images includes: using the first deep learning network to perform trend warning detection on the u-th personnel action image and the first force state data of the climbing structure of the u-th personnel action image to obtain a first trend warning detection result, where u is a positive integer not greater than X; using the first deep learning network to perform trend warning detection on the u-th personnel action image and the second force state data of the climbing structure of the u-th personnel action image to obtain a second trend warning detection result; and adding the first trend warning detection result and the second trend warning detection result to the trend warning detection result set corresponding to the force state data set of the climbing structure of the u-th personnel action image.

[0083] For example, the system makes a more detailed division of each set of data on the stress state of the climbing structure, specifically including the first set of data on the stress state of the climbing structure and the second set of data on the stress state of the climbing structure. These two types of data may reflect the stress conditions of the climbing structure in different aspects. For example, the first set of data on the stress state of the climbing structure may focus on the horizontal stress of the structure, while the second set of data on the stress state of the climbing structure may focus on the vertical stress of the structure. Such a division helps to more comprehensively evaluate the safety of climbing operations.

[0084] When the system performs trend warning detection, it will perform multiple detections on each personnel action image and its corresponding set of data on the stress state of the climbing structure. Taking the u-th personnel action image as an example, the system will first use the first deep learning network to detect this action image and the corresponding first set of data on the stress state of the climbing structure, and obtain the first trend warning detection result. This result mainly reflects whether there are safety risks in the personnel action under a specific horizontal stress state.

[0085] Immediately afterwards, the system will use the same first deep learning network to perform another detection on the u-th personnel action image and the second set of data on the stress state of the climbing structure, and obtain the second trend warning detection result. This time, the detection result focuses on evaluating the safety of the personnel action under a specific vertical stress state.

[0086] After completing the above two detections, the system will merge the first trend warning detection result and the second trend warning detection result, and incorporate them into the set of trend warning detection results corresponding to the set of data on the stress state of the climbing structure of the u-th personnel action image. In this way, the set of trend warning detection results for each action image contains safety evaluations under different stress states, providing more comprehensive and accurate data support for subsequent climbing warning decisions.

[0087] Through the above technical solution, the system can achieve comprehensive monitoring and warning of personnel actions and the stress state of the climbing structure during climbing operations. Such a detailed detection method not only improves the accuracy and reliability of the warning system, but also helps operators to timely identify and correct unsafe behaviors, thereby effectively reducing the safety risks in climbing operations.

[0088] Under some optional design concepts, each set of data on the force state of the climbing structure includes the data on the force state of the third climbing structure; when using the first deep learning network to perform trend warning detection on each personnel action image and the set of data on the force state of the climbing structure corresponding to the personnel action image, the resulting set of trend warning detection results corresponding to the sets of data on the force state of the climbing structure of the X personnel action images includes: using the first deep learning network to perform trend warning detection on the u-th personnel action image and the data on the force state of the third climbing structure of the u-th personnel action image to obtain a third trend warning detection result, where u is a positive integer not greater than X; adding the third trend warning detection result to the set of trend warning detection results corresponding to the set of data on the force state of the climbing structure of the u-th personnel action image; where the set of trend warning detection results corresponding to the set of data on the force state of the climbing structure of the u-th personnel action image further includes a fourth trend warning detection result, and the fourth trend warning detection result is obtained by using the second deep learning network to perform trend warning detection on the u-th personnel action image and the data on the force state of the third climbing structure of the u-th personnel action image.

[0089] For another example, the system further expands the types of data on the force state of the climbing structure. In particular, the data on the force state of the third climbing structure is added. This data may represent a specific force condition. For example, in the power production scenario, the data on the force state of the third climbing structure may focus on the impact of the vibration of power equipment on the climbing structure, or the structural stress changes caused by environmental factors such as wind and temperature.

[0090] When performing trend warning detection, in addition to considering the data on the force state of the first and second climbing structures, the system will specifically analyze the data on the force state of the third climbing structure. Specifically, the system will use the first deep learning network to detect the u-th personnel action image and the corresponding data on the force state of the third climbing structure, so as to obtain a third trend warning detection result. This result reflects the safety assessment of the personnel action under a specific third force state.

[0091] After obtaining the third trend warning detection result, the system will add it to the set of trend warning detection results corresponding to the set of data on the force state of the climbing structure of the u-th personnel action image. In this way, the result set contains the warning detection results under multiple different force states, providing more abundant information for subsequent decision-making.

[0092] It should be noted that the system also introduces a second deep learning network to conduct another independent detection on the stress state data of the same third climbing structure, and obtains a fourth trend warning detection result. This design concept of dual networks can increase the diversity and robustness of warning detection because different networks may focus on different features, thus providing a more comprehensive safety assessment.

[0093] Finally, the fourth trend warning detection result will also be incorporated into the trend warning detection result set corresponding to the stress state data set of the climbing structure of the u-th personnel action image. In this way, the system constructs a comprehensive warning system that includes multiple detection results and can reflect the safety under various stress states.

[0094] Through the implementation of the above technical solutions, the system can achieve comprehensive monitoring and warning of various complex stress states during the climbing operation in the power production scenario. This design not only improves the accuracy of the warning system but also enhances its adaptability to different scenarios. Especially in a complex and changeable environment such as power production, the multiple detection mechanism and dual network design of the system can more effectively identify and prevent potential safety risks, ensuring the safety of operators. At the same time, this technical solution also demonstrates the broad application prospects of deep learning in complex industrial safety monitoring.

[0095] In some other preferred embodiments, the collaborative debugging and improvement of the first deep learning network and the climbing warning decision network to obtain the improved first deep learning network and the improved climbing warning decision network includes: obtaining a first long short-term memory network, a second long short-term memory network, and a decision tree network, where the first long short-term memory network and the second long short-term memory network are updated based on the first deep learning network, and the decision tree network is updated based on the climbing warning decision network; processing the linkage monitoring information through the first long short-term memory network, the second long short-term memory network, the decision tree network, and the climbing warning decision network to obtain a first network quality evaluation and a second network quality evaluation; the linkage monitoring information is obtained by integrating the target personnel action image, the stress state data set of the climbing structure of the target personnel action image, and the trend warning detection result set corresponding to the stress state data set of the climbing structure of the target personnel action image, and the target personnel action image is sampled from the X personnel action images; and performing collaborative debugging on the first long short-term memory network and the decision tree network according to the first network quality evaluation and the second network quality evaluation to obtain the improved first deep learning network and the improved climbing warning decision network.

[0096] For example, the system adopts a more advanced strategy to co - debug and improve the first deep - learning network and the high - altitude warning decision - making network. This process involves the integration and optimization of multiple neural network models.

[0097] First, the system obtains the first long short - term memory network (LSTM1), the second long short - term memory network (LSTM2), and the decision tree network. These two long short - term memory networks (LSTMs) are updated based on the first deep - learning network, while the decision tree network is updated based on the high - altitude warning decision - making network. The long short - term memory network is a special type of recurrent neural network that can learn and remember long - term dependencies and is very suitable for processing time - series data, such as continuous high - altitude structure stress state data. The decision tree network is good at dealing with classification and decision - making problems.

[0098] Next, the system uses these networks to process the linkage monitoring information. The linkage monitoring information is obtained by integrating the target personnel action images, the set of high - altitude structure stress state data of this image, and the corresponding set of trend warning detection results. In this process, the target personnel action images are sampled from the original X personnel action images, ensuring the representativeness and effectiveness of the data.

[0099] After processing the linkage monitoring information, the system will obtain two network quality evaluations: the first network quality evaluation and the second network quality evaluation. These two evaluations reflect the performance of the long short - term memory network and the decision tree network when processing the linkage monitoring information.

[0100] Finally, based on these two network quality evaluations, the system will co - debug the first long short - term memory network and the decision tree network. This may include adjusting network parameters, optimizing network structures, improving learning algorithms, etc. Through this series of debugging and optimization steps, the system can finally obtain the improved first deep - learning network and the improved high - altitude warning decision - making network.

[0101] This method of co - debugging and improvement has significant beneficial effects. First, it improves the accuracy and reliability of the warning system because through the collaborative work of multiple networks, the system can more comprehensively capture and analyze the features in the data. Second, this method enhances the robustness of the system, enabling the warning system to maintain stable performance in the face of complex and changing power production scenarios. Finally, by optimizing the combination method of the deep - learning network and the decision tree network, the system can make more effective warning decisions, thus improving the safety of power production.

[0102] In the next step, the linkage monitoring information is processed by the first long short-term memory network, the second long short-term memory network, the decision tree network, and the altitude warning decision network to obtain a first network quality evaluation and a second network quality evaluation, including: obtaining a first view output weight, a second view output weight, a first credibility factor, and a first decision coefficient. The first view output weight is obtained by using the first long short-term memory network to perform first feature mining and identification on the linkage monitoring information. The second view output weight is obtained by using the second long short-term memory network to perform second feature mining and identification on the linkage monitoring information. The first credibility factor is obtained by using the altitude warning decision network to perform confidence weight identification on the linkage monitoring information. The first decision coefficient is obtained by using the decision tree network to perform warning decision identification on the linkage monitoring information; determining a consistency comparison result through the first view output weight and the second view output weight, where the consistency comparison result is used to represent the difference between the first view output weight and the second view output weight; adjusting the first credibility factor according to the consistency comparison result to obtain the second network quality evaluation; determining the first network quality evaluation through the second network quality evaluation and the first decision coefficient.

[0103] For example, through a series of complex processing procedures, the system uses multiple network models to comprehensively evaluate the linkage monitoring information to obtain a first network quality evaluation and a second network quality evaluation.

[0104] First, the system obtains several key parameters. The first view output weight is obtained by the first long short-term memory network (LSTM1) performing first feature mining and identification on the linkage monitoring information. This can be understood as the LSTM1 network giving a kind of "viewpoint" or weight assignment to the features in the linkage monitoring information after analyzing it, reflecting the importance the network attaches to these features. Similarly, the second view output weight is obtained by the second long short-term memory network (LSTM2) performing second feature mining and identification on the linkage monitoring information, representing the "viewpoint" of the LSTM2 network.

[0105] In addition, the system also uses the altitude warning decision network to perform confidence weight identification on the linkage monitoring information to obtain a first credibility factor. This credibility factor can be understood as the confidence level of the warning decision network for its judgment result. At the same time, the decision tree network also performs warning decision identification on the linkage monitoring information and generates a first decision coefficient, which reflects the decision tree network's judgment on whether a warning should be issued.

[0106] Next, the system determines a consistency comparison result by comparing the first view output weight and the second view output weight. This consistency comparison result actually measures whether the two LSTM networks agree on the importance of features in the linkage monitoring information. If the two views are close, it indicates that the two networks have a relatively unified interpretation of the data; if the views differ greatly, it indicates that the data may have some complex or contradictory features.

[0107] Based on this consistency comparison result, the system adjusts the first credibility factor to obtain the second network quality evaluation. This adjustment process actually corrects the confidence level of the early warning decision-making network according to the consistency of the two LSTM networks. If the views of the two LSTM networks are highly consistent, the confidence level of the early warning decision-making network may be increased; otherwise, it may be decreased.

[0108] Finally, the system combines the second network quality evaluation and the first decision coefficient to determine the first network quality evaluation. This evaluation comprehensively considers the feature mining ability of the LSTM network, the confidence level of the early warning decision-making network, and the early warning decision-making ability of the decision tree network, so it can more comprehensively reflect the performance of the entire early warning system.

[0109] Through this series of complex processing and analysis processes, the system can not only more accurately evaluate the risk situation in the linkage monitoring information, but also optimize the accuracy of early warning decisions based on the comprehensive judgment of multiple network models. This way of multi-model collaborative work significantly improves the reliability and accuracy of the high-altitude operation early warning system in the power production scenario, providing strong technical support for ensuring the safety of operators.

[0110] Further, the collaborative debugging of the first long short-term memory network and the decision tree network according to the first network quality evaluation and the second network quality evaluation to obtain the improved first deep learning network and the improved high-altitude warning decision-making network includes: combining the first view output weight and the second view output weight, and the first network quality evaluation to generate a first training error variable; obtaining a second decision coefficient, and generating a second training error variable through the second decision coefficient and the second network quality evaluation, where the second decision coefficient is obtained by the decision tree network after cyclic debugging to perform early warning decision recognition on the training subset, and the training subset is obtained by disassembling the linkage monitoring information; generating a third training error variable based on the first training error variable and the second training error variable; and performing collaborative debugging on the first long short-term memory network and the decision tree network through the third training error variable to obtain the improved first deep learning network and the improved high-altitude warning decision-making network.

[0111] For example, based on the previously obtained first network quality evaluation and second network quality evaluation, the system performed collaborative debugging on the first long short-term memory network and the decision tree network to optimize the network performance.

[0112] First, the system combined the first view output weight and the second view output weight, and considered the first network quality evaluation to generate a first training error variable. This variable reflects the error situation of the long short-term memory network when processing linkage monitoring information, as well as the impact of the network quality evaluation on the error. Through this step, the system can more accurately locate the deficiencies of the long short-term memory network when processing data.

[0113] Next, the system obtained a second decision coefficient, which was obtained by the decision tree network after cyclic debugging to identify early warning decisions for the training subset. The training subset was obtained by disassembling the linkage monitoring information, which can better target specific parts of the data for training and optimization. Using the second decision coefficient and the second network quality evaluation, the system generated a second training error variable. This variable represents the error of the decision tree network when processing the training subset and takes into account the factors of the network quality evaluation.

[0114] Then, based on the first training error variable and the second training error variable, the system generated a third training error variable. This variable synthesizes the error situations of the long short-term memory network and the decision tree network when processing data, providing comprehensive error feedback for subsequent collaborative debugging.

[0115] Finally, through the third training error variable, the system performed collaborative debugging on the first long short-term memory network and the decision tree network. During this process, the system adjusts the parameters and structure of the network according to the magnitude and direction of the error variable to reduce the error and improve the performance of the network. After such a debugging process, the system finally obtained an improved first deep learning network and an improved high-altitude warning decision network.

[0116] This method of collaborative debugging not only improves the accuracy of the deep learning network and the warning decision network, but also enhances the collaborative working ability between the two. By comprehensively considering the views and decisions of multiple networks, the system can analyze data more comprehensively and make more accurate early warning decisions. In addition, by disassembling the linkage monitoring information into training subsets and conducting specialized training for these subsets, the system can more effectively handle complex and variable data situations. Generally speaking, this method improves the overall performance of the high-altitude operation warning system in the power production scenario, providing stronger support for ensuring the safety of operators.

[0117] In an alternative embodiment, obtaining the X personnel action images corresponding to the sample of the high-altitude warning monitoring information includes: obtaining the sample of the high-altitude warning monitoring information; if the power production scenario of the sample of the high-altitude warning monitoring information is different from the set power production scenario, then performing a power production scenario mapping on the sample of the high-altitude warning monitoring information to obtain a sample of the high-altitude warning monitoring information under the set power production scenario; performing a chunking process on the sample of the high-altitude warning monitoring information under the set power production scenario by using an information chunking rule to obtain Y monitoring sample sub-chunks, where Y is a positive integer not greater than X; and generating the X personnel action images corresponding to the sample of the high-altitude warning monitoring information based on the Y monitoring sample sub-chunks.

[0118] First, the system obtains a sample of the high-altitude warning monitoring information. These information samples may come from different power production scenarios, such as wind farms, thermal power plants, hydropower plants, etc. The high-altitude operation environments and risk factors in each scenario may be different, so these information samples need to be further processed.

[0119] Next, the system checks whether the power production scenario of the obtained sample of the high-altitude warning monitoring information is the same as the set power production scenario. If not, the system performs a power production scenario mapping on these information samples. This mapping process actually converts the data under different scenarios into equivalent data under the set power production scenario to ensure the accuracy and effectiveness of subsequent processing. For example, if the original information sample comes from a wind farm and the set scenario is a thermal power plant, the system will adjust and map the data according to factors such as environmental differences and operation process differences between the two scenarios.

[0120] After completing the scenario mapping, the system obtains a sample of the high-altitude warning monitoring information under the set power production scenario. To analyze these data more precisely, the system performs a chunking process on these information samples by using an information chunking rule. This process splits the overall monitoring information into multiple small parts, namely monitoring sample sub-chunks. The advantage of doing this is that each small part can be analyzed and processed separately, so as to more accurately identify potential risk points. For example, the system can divide the data into multiple sample sub-chunks according to different actions of the operating personnel or different time periods.

[0121] Finally, based on these monitoring sample sub-chunks, the system generates the X personnel action images corresponding to the sample of the high-altitude warning monitoring information. These images visually show the specific actions and states of the operating personnel under the set power production scenario, providing an important visual basis for subsequent risk assessment and early warning.

[0122] In this way, through the mapping of the power production scenario, the system can process the monitoring information from different scenarios, improving the generality and applicability of the data. Secondly, the information chunking rule enables the system to conduct a more refined analysis of the data, contributing to the discovery of potential safety hazards. Finally, the generated images of personnel actions provide intuitive monitoring means for the management personnel, facilitating their timely understanding and grasp of the on-site operation situation, so as to make effective early warnings and decisions. Generally speaking, this processing method enhances the accuracy and reliability of the high-altitude warning system and plays an important role in ensuring the safety of power production.

[0123] In some alternative embodiments, each monitoring sample sub-block is associated with an annotation information; generating the X personnel action images corresponding to the high-altitude warning monitoring information sample according to the Y monitoring sample sub-blocks includes: obtaining an action noise feature threshold; if there is a monitoring sample sub-block in the Y monitoring sample sub-blocks whose action feature value is not greater than the action noise feature threshold, then determining this monitoring sample sub-block as a personnel action image; If there is a monitoring sample sub-block in the Y monitoring sample sub-blocks whose action feature value is greater than the action noise feature threshold, then disassembling this monitoring sample sub-block to obtain at least two personnel action images, and adding the annotation information associated with this monitoring sample sub-block to the at least two personnel action images, and the action feature value of each personnel action image is not greater than the action noise feature threshold.

[0124] First of all, the system will obtain an action noise feature threshold. This threshold is a preset value used to distinguish whether the action features in the monitoring sample sub-blocks are significant. In short, it helps the system determine which actions are effective and worthy of attention, and which are minor and negligible noises.

[0125] Next, the system will check each of the Y monitoring sample sub-blocks one by one. For each sample sub-block, the system will extract its action feature value and compare it with the action noise feature threshold.

[0126] If the action feature value of a certain monitoring sample sub-block is not greater than the action noise feature threshold, it means that the action in this sample sub-block is relatively weak and may be caused by noise or other non-critical actions. In this case, the system will directly determine this monitoring sample sub-block as a personnel action image because it is already simple and clear enough and does not need to be further disassembled.

[0127] However, if the action feature value of a certain monitoring sample sub-block is greater than the action noise feature threshold, it means that the action in this sample sub-block is relatively complex or significant. In this case, the system will disassemble this monitoring sample sub-block. The purpose of disassembling is to decompose this complex action into several simpler parts for more accurate analysis and identification.

[0128] After disassembly, the system will obtain at least two human action images. These images respectively represent different action stages or components in the original monitoring sample sub-block. To ensure the integrity and comprehensibility of these images, the system will add the annotation information associated with this monitoring sample sub-block to these human action images. In this way, when managers view these images, they can better understand the meaning and context of each action through the annotation information.

[0129] Finally, the system will ensure that the action feature value of each human action image is not greater than the action noise feature threshold. This is to ensure that each image is simple and clear enough for subsequent analysis and recognition.

[0130] In this way, complex actions can be disassembled into simpler parts, and additional context can be provided through annotation information. This not only improves the accuracy of action recognition but also enables managers to more easily understand and analyze on-site operation situations. In addition, by setting the action noise feature threshold, the system can effectively filter out noise and non-critical actions, thus focusing on truly important actions and events. Generally speaking, this method improves the performance and usability of the high-altitude warning system and provides stronger support for ensuring power production safety.

[0131] Under some preferred design ideas, debugging the neural network to be debugged by using the set of high-altitude structure stress state data of the X human action images and the set of trend warning detection results corresponding to the set of high-altitude structure stress state data of the X human action images to obtain a high-altitude warning decision network includes: arbitrarily selecting a first trend warning detection result and a second trend warning detection result from the set of trend warning detection results corresponding to the set of high-altitude structure stress state data of the u-th human action image, where the first trend warning detection result and the second trend warning detection result are different, and u is a positive integer not greater than X; using the neural network to be debugged to respectively perform confidence weight recognition on the first trend warning detection result and the second trend warning detection result according to the set of high-altitude structure stress state data of the u-th human action image to obtain a second credibility factor and a third credibility factor; generating a fourth training error variable through the difference between the second credibility factor and the third credibility factor; and debugging the neural network to be debugged according to the fourth training error variable to obtain a high-altitude warning decision network.

[0132] For example, the system finely debugs the neural network to be debugged by analyzing in detail the high-altitude structure stress state corresponding to the human action image and the trend warning detection results of these states to construct an efficient high-altitude warning decision network.

[0133] First, the system randomly selects two different trend warning detection results from the set of trend warning detection results corresponding to the force state data set of the climbing structure in the u-th personnel action image, namely the first trend warning detection result and the second trend warning detection result. These two results reflect the changing trends of the force state of the climbing structure at different times or under different conditions.

[0134] Next, the system uses the neural network to be debugged to identify the confidence weights of these two trend warning detection results respectively according to the force state data set of the climbing structure in the u-th personnel action image. This process is actually to evaluate the trust degree of the neural network for these two different trend warning results, so as to obtain the second confidence factor and the third confidence factor. These confidence factors represent the confidence level of the neural network for different warning results.

[0135] Then, the system generates a fourth training error variable by comparing the differences between the second confidence factor and the third confidence factor. This error variable reflects the judgment accuracy of the neural network when facing different trend warning detection results, as well as the ability of the network to distinguish different results.

[0136] Finally, the system debugs the neural network to be debugged based on this fourth training error variable. The purpose of debugging is to optimize the parameters and structure of the network so that it can make more accurate and timely warning decisions when facing different changing trends of the force state of the climbing structure. After such a debugging process, the system finally obtains an efficient and accurate climbing warning decision-making network.

[0137] In this way, the recognition ability and accuracy of the neural network for different trend warning detection results are improved. By comparing the confidence factors under different results, the system can more precisely adjust the parameters and structure of the network, so that it can make more reliable warning decisions when facing the complex and changeable force state of the climbing structure. This not only improves the safety of climbing operations in the power production scenario, but also provides more accurate and timely warning information for relevant management personnel, helping them to take more effective safety prevention and control measures.

[0138] It should be noted that generating the fourth training error variable through the difference between the second confidence factor and the third confidence factor is a key step in the neural network training process. The purpose of this step is to adjust and optimize the parameters of the neural network to improve its recognition accuracy for different trend warning detection results.

[0139] Specifically, the second confidence factor and the third confidence factor are the confidence evaluations of the neural network for two different trend warning detection results. The difference between these two factors reflects the stability and consistency of the neural network's judgment results when facing similar but not exactly the same data inputs.

[0140] If the difference between the second credibility factor and the third credibility factor is large, it indicates that the neural network has significant uncertainty or fluctuations in the judgment of these two similar inputs. This may mean that some parameters or structures of the neural network need to be adjusted to improve its recognition accuracy for such inputs.

[0141] Based on this difference, the system generates a fourth training error variable. This error variable is actually a quantization index used to measure the judgment error of the neural network when facing such similar inputs. By analyzing and optimizing this error variable, the problems existing in the neural network can be found and corresponding adjustments can be made.

[0142] During the training process of the neural network, this fourth training error variable is used as the error signal in the backpropagation algorithm to guide the update of the network weights. In this way, the neural network can gradually learn how to more accurately identify and process such similar input data.

[0143] In summary, generating the fourth training error variable by comparing the differences between the second credibility factor and the third credibility factor is an important part of the neural network training and optimization process. It helps to discover and correct the judgment bias of the neural network when processing similar inputs, thereby improving the accuracy and stability of the network.

[0144] Under another design concept, the method further includes: obtaining Z pieces of climbing warning monitoring information to be analyzed and the stress state data of the climbing structure for each piece of climbing warning monitoring information to be analyzed, where the power production scenario of the Z pieces of climbing warning monitoring information to be analyzed is a set power production scenario, and Z is a positive integer; using the improved first deep learning network to perform trend warning detection on each piece of climbing warning monitoring information to be analyzed and the stress state data of the climbing structure for this piece of climbing warning monitoring information to be analyzed, and obtaining the trend warning detection result corresponding to each stress state data of the climbing structure; selecting according to the improved climbing warning decision network the trend warning detection results corresponding to the stress state data of the Z pieces of climbing warning monitoring information to be analyzed, and obtaining v trend warning detection results that meet the set requirements, where v is a positive integer not greater than Z; jointly importing the v trend warning detection results that meet the set requirements, as well as the stress state data of the climbing structure and the climbing warning monitoring information to be analyzed corresponding to each trend warning detection result that meets the set requirements, into the cloud service space.

[0145] In the power production scenario, the system not only needs to perform real-time warning monitoring on climbing operations, but also needs to deeply analyze and process this monitoring information. The following is a more detailed implementation process of the technical solution.

[0146] First, the system will obtain Z high-altitude warning monitoring information to be analyzed, as well as the data of the force state of the high-altitude structure corresponding to each monitoring information. These data all come from the set power production scenario, where Z is a positive integer representing the number of monitoring information to be analyzed.

[0147] Next, the system will use the improved first deep learning network to perform trend warning detection on each high-altitude warning monitoring information to be analyzed and its corresponding high-altitude structure force state data. This deep learning network has been specifically optimized and improved to more accurately identify the change trend of the force on the high-altitude structure and potential risks. Through this step, the system will generate a corresponding trend warning detection result for each high-altitude structure force state data.

[0148] Then, the system will further select these trend warning detection results according to the improved high-altitude warning decision network. This decision network can, based on a large amount of historical data and empirical knowledge, intelligently judge which warning results meet the set requirements and which may be false alarms or unnecessary. After this round of screening, the system will obtain v trend warning detection results that meet the set requirements, where v is a positive integer not greater than Z.

[0149] Finally, the system will jointly import these v trend warning detection results that meet the requirements, as well as the data of the force state of the high-altitude structure corresponding to each result and the original high-altitude warning monitoring information, into a cloud service space. This cloud service space not only provides a large amount of data storage capacity but also supports a variety of data analysis and visualization tools, enabling management personnel to access and analyze these data anytime and anywhere, so as to better understand the safety status of the power production site and make timely responses and decisions.

[0150] In this way, first, through the accurate detection of the deep learning network, the system can more accurately identify the force changes and potential risks of the high-altitude structure. Second, through the intelligent screening of the high-altitude warning decision network, the system can filter out most false alarms and unnecessary warnings, improving the accuracy and effectiveness of warning information. Finally, by importing the data into the cloud service space, the system realizes the centralized storage and remote access of data, greatly improving the efficiency and convenience of data management. Generally speaking, this technical solution provides a strong technical guarantee for the safety of high-altitude operations in the power production scenario.

[0151] In other preferred embodiments, the obtaining of Z heightening warning monitoring information to be analyzed and the stress state data of the heightening structure for each piece of heightening warning monitoring information to be analyzed includes: obtaining Z heightening warning monitoring information to be analyzed; if there is target heightening warning monitoring information in the Z heightening warning monitoring information to be analyzed whose power production scenario is different from the set power production scenario, then performing power production scenario mapping on the target heightening warning monitoring information to be analyzed to obtain Z heightening warning monitoring information to be analyzed under the set power production scenario; and associating stress state data of the heightening structure with each piece of heightening warning monitoring information to be analyzed under the set power production scenario according to the elements of the heightening warning monitoring information to be analyzed.

[0152] First, the system obtains Z heightening warning monitoring information to be analyzed. This information may come from different power production scenarios and contains various data related to the safety of heightening operations.

[0153] Next, the system checks the power production scenarios in the Z heightening warning monitoring information to be analyzed. If it is found that there is target heightening warning monitoring information in them whose power production scenario is different from the set power production scenario, the system will perform power production scenario mapping. This mapping process is to convert this information to the set power production scenario for unified analysis and processing. Through mapping, the system can ensure that all monitoring information is based on the same scenario standard, thereby improving the accuracy and effectiveness of analysis.

[0154] After completing the scenario mapping, the system will obtain Z heightening warning monitoring information to be analyzed under the set power production scenario. This information is now standardized and can be directly used for subsequent analysis and warning work.

[0155] Next, the system will associate stress state data of the heightening structure with each piece of heightening warning monitoring information to be analyzed under the set power production scenario according to the elements of the heightening warning monitoring information to be analyzed. These data reflect the stress conditions of the heightening structure in different scenarios and are key indicators for evaluating the safety of heightening operations. By associating these data, the system can more comprehensively understand the actual situation of heightening operations and thus make more accurate warning decisions.

[0156] Generally speaking, this technical solution provides comprehensive data support for the safety of climbing operations in the power production scenario by acquiring, mapping, and correlating a series of climbing warning monitoring information to be analyzed and the corresponding stress state data of the climbing structure. This not only improves the accuracy of early warnings but also enables managers to make more scientific and reasonable safety decisions based on actual data. It can be seen that by uniformly processing and analyzing the climbing warning monitoring information in different power production scenarios, the accuracy and consistency of early warning decisions are ensured. At the same time, by correlating the stress state data of the climbing structure, the system can more deeply understand the actual risk situation of climbing operations, thereby taking necessary preventive measures in advance to ensure the safety of operators. This comprehensive data analysis and early warning mechanism helps to improve the safety of power production and reduce the likelihood of accidents.

[0157] In other preferred embodiments, the method of selecting the trend warning detection results corresponding to the stress state data of the climbing structure of the Z climbing warning monitoring information to be analyzed according to the improved climbing warning decision network, to obtain v trend warning detection results that meet the set requirements, includes: obtaining a decision threshold; using the improved climbing warning decision network to perform warning decision recognition on each trend warning detection result corresponding to the stress state data of each climbing warning monitoring information to be analyzed, to obtain the third decision coefficient of each trend warning detection result; if the third decision coefficient of any trend warning detection result is higher than the decision threshold, then determine that trend warning detection result as a trend warning detection result that meets the set requirements.

[0158] In the power production scenario, the system uses the improved climbing warning decision network to select the trend warning detection results of the climbing warning monitoring information to be analyzed through a series of precise operations. The following is a detailed example to explain this process.

[0159] First, the system obtains a decision threshold. This decision threshold is a preset standard value used to evaluate the reliability or importance of the trend warning detection results. Only when the decision coefficient of a certain trend warning detection result is higher than this threshold will it be considered to meet the set requirements.

[0160] Next, the system uses the improved climbing warning decision network to perform warning decision recognition on each trend warning detection result corresponding to the stress state data of each climbing warning monitoring information to be analyzed. This decision network is specially optimized and improved to more accurately identify the importance and reliability of the trend warning detection results. Through this step, the system generates a third decision coefficient for each trend warning detection result.

[0161] Then, the system will check these third decision coefficients one by one. If the third decision coefficient of any trend warning detection result is higher than the previously obtained decision threshold, then the system will determine that this trend warning detection result meets the set requirements. This means that this trend warning detection result is considered to be relatively important or reliable, and may correspond to a security risk that requires attention or response.

[0162] Through this process, the system can accurately select the trend warning detection results that meet the set requirements, helping managers to more accurately understand the safety status of the power production site and make responses and decisions in a timely manner.

[0163] In this way, the accuracy and efficiency of the warning monitoring are improved. By setting the decision threshold and using the improved warning decision network for height climbing to identify warning decisions, the system can automatically screen out important and reliable trend warning detection results, reducing the workload of manual analysis and judgment, and improving the timeliness and accuracy of the warning. At the same time, this also enables managers to more quickly grasp the safety status of the power production site, make more scientific and reasonable safety decisions, thus ensuring the safety and stability of power production.

[0164] In an independent embodiment, a fourth training error variable is generated through the difference between the second credibility factor and the third credibility factor, including: calculating the difference between the second credibility factor and the third credibility factor to obtain an initial difference value; normalizing the initial difference value to eliminate the influence of different dimensions and orders of magnitude on the result, obtaining a normalized difference value; setting a difference threshold for judging whether the difference between the second credibility factor and the third credibility factor is significant. If the normalized difference value exceeds this difference threshold, proceed to the next step, otherwise return to re-obtain the second credibility factor and the third credibility factor; based on the normalized difference value, calculate the fourth training error variable through a preset error generation function; use the fourth training error variable as a feedback signal and input it into the deep learning network for backpropagation to adjust the network weights and optimize the performance of the height climbing warning system.

[0165] Among them, the second credibility factor represents the confidence level of the deep learning network for the first trend warning detection result of the force state data of a certain height climbing structure, and the third credibility factor represents the confidence level of the deep learning network for the second trend warning detection result of the force state data of the same height climbing structure. The fourth training error variable reflects the stability and consistency of the judgment results of the deep learning network when facing similar but not exactly the same input data.

[0166] For example, the system calculates the difference between the second trust factor and the third trust factor to obtain an initial difference value. Here, the second trust factor represents the confidence level of the deep learning network for the first trend warning detection result of the force state data of a certain elevated structure, while the third trust factor represents the confidence level of the deep learning network for the second trend warning detection result of the force state data of the same elevated structure. By calculating the difference between these two trust factors, the system can initially understand the stability of the network's judgment results when facing similar input data. Next, to eliminate the influence of different dimensions and orders of magnitude on the results, the system normalizes the initial difference value to obtain a normalized difference value. This step is to ensure the accuracy and fairness of subsequent processing. Then, the system sets a difference threshold to determine whether the difference between the second trust factor and the third trust factor is significant. If the normalized difference value exceeds this set difference threshold, it indicates that the difference in the network's judgment results when facing similar input data is relatively large and further processing is required. Otherwise, the system will return to re-obtain the second trust factor and the third trust factor to ensure the accuracy and effectiveness of the data. When the normalized difference value exceeds the difference threshold, the system calculates a fourth training error variable based on this normalized difference value through a preset error generation function. This error variable reflects the stability and consistency of the deep learning network's judgment results when facing similar but not exactly the same input data. Finally, the system uses the fourth training error variable as a feedback signal and inputs it into the deep learning network for backpropagation. By adjusting the weights of the network, the system can optimize the performance of the elevated warning system and improve its judgment stability and consistency when facing similar input data.

[0167] In this way, it is possible to generate targeted training error variables by analyzing the differences in the judgment results of the deep learning network when facing similar input data, thereby optimizing the performance of the network. This not only improves the accuracy and reliability of the elevated warning system but also enables the system to make more stable and consistent warning judgments when facing complex and changing power production scenarios, thus ensuring the safety and stability of power production.

[0168] Figure 2 The structural block diagram of the AI data processing system 300 is shown, including: a memory 310 for storing program instructions and data; a processor 320 for being coupled to the memory 310 and executing the instructions in the memory 310 to implement the above method.

[0169] Furthermore, a computer storage medium is provided, containing instructions that, when executed on a processor, implement the above method.

[0170] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, various modifications and variations can be made to the present invention. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A data processing method applied to altitude warning analysis, characterized in that, Applied to an AI data processing system, the method includes: Obtaining X personnel action images corresponding to the sample of high-altitude warning monitoring information and the set of high-altitude structure stress state data for each personnel action image, where the power production scenarios of the X personnel action images are the same, each set of high-altitude structure stress state data includes at least one high-altitude structure stress state data, and X is a positive integer; Using a first deep learning network to perform trend warning detection on each personnel action image and the set of high-altitude structure stress state data for that personnel action image, to obtain a set of trend warning detection results corresponding to the set of high-altitude structure stress state data for the X personnel action images, and each set of trend warning detection results includes at least two trend warning detection results; Debugging the neural network to be debugged through the set of high-altitude structure stress state data for the X personnel action images and the set of trend warning detection results corresponding to the set of high-altitude structure stress state data for the X personnel action images, to obtain a high-altitude warning decision network, and the high-altitude warning decision network is used to identify the confidence weights of the trend warning detection results; Performing collaborative debugging and improvement on the first deep learning network and the high-altitude warning decision network to obtain an improved first deep learning network and an improved high-altitude warning decision network, and the improved first deep learning network and the improved high-altitude warning decision network are used to perform high-altitude warning analysis on the set of high-altitude warning monitoring information to be analyzed.

2. The method according to claim 1, characterized in that, Each set of high-altitude structure stress state data includes first high-altitude structure stress state data and second high-altitude structure stress state data; the using of the first deep learning network to perform trend warning detection on each personnel action image and the set of high-altitude structure stress state data for that personnel action image, to obtain the set of trend warning detection results corresponding to the set of high-altitude structure stress state data for the X personnel action images, includes: Using the first deep learning network to perform trend warning detection on the u-th personnel action image and the first high-altitude structure stress state data of the u-th personnel action image, to obtain a first trend warning detection result, where u is a positive integer not greater than X; Using the first deep learning network to perform trend warning detection on the u-th personnel action image and the second high-altitude structure stress state data of the u-th personnel action image, to obtain a second trend warning detection result; Adding the first trend warning detection result and the second trend warning detection result to the set of trend warning detection results corresponding to the set of high-altitude structure stress state data of the u-th personnel action image.

3. The method according to claim 1, wherein Each set of high-altitude structure stress state data includes third high-altitude structure stress state data; the using of the first deep learning network to perform trend warning detection on each personnel action image and the set of high-altitude structure stress state data for that personnel action image, to obtain the set of trend warning detection results corresponding to the set of high-altitude structure stress state data for the X personnel action images, includes: Using the first deep learning network to perform trend warning detection on the u-th personnel action image and the third high-altitude structure stress state data of the u-th personnel action image, obtaining a third trend warning detection result, where u is a positive integer not greater than X; Adding and allocating the third trend warning detection result to the trend warning detection result set corresponding to the high-altitude structure stress state data set of the u-th personnel action image; Among them, the trend warning detection result set corresponding to the high-altitude structure stress state data set of the u-th personnel action image further includes a fourth trend warning detection result, and the fourth trend warning detection result is obtained by using the second deep learning network to perform trend warning detection on the u-th personnel action image and the third high-altitude structure stress state data of the u-th personnel action image.

4. The method according to claim 1, wherein The collaborative debugging and improvement of the first deep learning network and the high-altitude warning decision-making network to obtain an improved first deep learning network and an improved high-altitude warning decision-making network includes: Obtaining a first long short-term memory network, a second long short-term memory network, and a decision tree network. The first long short-term memory network and the second long short-term memory network are updated based on the first deep learning network, and the decision tree network is updated based on the high-altitude warning decision-making network; Processing the linkage monitoring information through the first long short-term memory network, the second long short-term memory network, the decision tree network, and the high-altitude warning decision-making network to obtain a first network quality evaluation and a second network quality evaluation; the linkage monitoring information is obtained by integrating the target personnel action image, the high-altitude structure stress state data set of the target personnel action image, and the trend warning detection result set corresponding to the high-altitude structure stress state data set of the target personnel action image, and the target personnel action image is sampled from the X personnel action images; Collaboratively debugging the first long short-term memory network and the decision tree network according to the first network quality evaluation and the second network quality evaluation to obtain an improved first deep learning network and an improved high-altitude warning decision-making network.

5. The method according to claim 4, characterized in that, The processing of the linkage monitoring information through the first long short-term memory network, the second long short-term memory network, the decision tree network, and the high-altitude warning decision-making network to obtain a first network quality evaluation and a second network quality evaluation includes: Obtaining a first view output weight, a second view output weight, a first credibility factor, and a first decision coefficient. The first view output weight is obtained by using the first long short-term memory network to perform first feature mining and recognition on the linkage monitoring information, the second view output weight is obtained by using the second long short-term memory network to perform second feature mining and recognition on the linkage monitoring information, the first credibility factor is obtained by using the high-altitude warning decision-making network to perform confidence weight recognition on the linkage monitoring information, and the first decision coefficient is obtained by using the decision tree network to perform warning decision recognition on the linkage monitoring information; Determine a consistency comparison result through the first view output weight and the second view output weight, where the consistency comparison result is used to characterize the difference between the first view output weight and the second view output weight; Adjust the first credibility factor according to the consistency comparison result to obtain a second network quality evaluation; Determine a first network quality evaluation through the second network quality evaluation and the first decision coefficient; Among them, the collaborative debugging of the first long short-term memory network and the decision tree network according to the first network quality evaluation and the second network quality evaluation to obtain an improved first deep learning network and an improved high-altitude warning decision network includes: Combine the first view output weight and the second view output weight, and the first network quality evaluation to generate a first training error variable; Obtain a second decision coefficient, and generate a second training error variable through the second decision coefficient and the second network quality evaluation. The second decision coefficient is obtained by the decision tree network after cyclic debugging to perform early warning decision recognition on the training subset, and the training subset is obtained by disassembling the linkage monitoring information; Generate a third training error variable according to the first training error variable and the second training error variable; Perform collaborative debugging on the first long short-term memory network and the decision tree network through the third training error variable to obtain an improved first deep learning network and an improved high-altitude warning decision network.

6. The method according to claim 1, characterized in that, The obtaining of X personnel action images corresponding to the high-altitude warning monitoring information sample includes: Obtain a high-altitude warning monitoring information sample; If the power production scenario of the high-altitude warning monitoring information sample is different from the set power production scenario, perform a power production scenario mapping on the high-altitude warning monitoring information sample to obtain a high-altitude warning monitoring information sample under the set power production scenario; Perform a block processing on the high-altitude warning monitoring information sample under the set power production scenario by using an information block rule to obtain Y monitoring sample sub-blocks, where Y is a positive integer not greater than X; Generate X personnel action images corresponding to the high-altitude warning monitoring information sample according to the Y monitoring sample sub-blocks; Among them, each monitoring sample sub-block is associated with an annotation information; the generating of X personnel action images corresponding to the high-altitude warning monitoring information sample according to the Y monitoring sample sub-blocks includes: Obtain an action noise feature threshold; If there is a monitoring sample sub-block in the Y monitoring sample sub-blocks whose action feature value is not greater than the action noise feature threshold, determine the monitoring sample sub-block as a personnel action image; If there is a monitoring sample sub-block in the Y monitoring sample sub-blocks whose action feature value is greater than the action noise feature threshold, disassemble the monitoring sample sub-block to obtain at least two personnel action images, and add the annotation information associated with the monitoring sample sub-block to the at least two personnel action images, and the action feature value of each personnel action image is not greater than the action noise feature threshold.

7. The method according to claim 1, wherein Debugging the neural network to be debugged by using the set of data on the force-bearing state of the climbing structure in the X personnel action images and the set of trend warning detection results corresponding to the set of data on the force-bearing state of the climbing structure in the X personnel action images to obtain a climbing warning decision network, including: Arbitrarily select a first trend warning detection result and a second trend warning detection result from the set of trend warning detection results corresponding to the set of data on the force-bearing state of the climbing structure in the u-th personnel action image, where the first trend warning detection result and the second trend warning detection result are different, and u is a positive integer not greater than X; Use the neural network to be debugged to respectively identify the confidence weights of the first trend warning detection result and the second trend warning detection result based on the set of data on the force-bearing state of the climbing structure in the u-th personnel action image, and obtain a second credibility factor and a third credibility factor; Generate a fourth training error variable based on the difference between the second credibility factor and the third credibility factor; Debug the neural network to be debugged according to the fourth training error variable to obtain a climbing warning decision network.

8. The method according to claim 1, wherein The method further includes: Obtain Z pieces of climbing warning monitoring information to be analyzed and the data on the force-bearing state of the climbing structure for each piece of climbing warning monitoring information to be analyzed, where the power production scenario of the Z pieces of climbing warning monitoring information to be analyzed is a set power production scenario, and Z is a positive integer; Use the improved first deep learning network to perform trend warning detection on each piece of climbing warning monitoring information to be analyzed and the data on the force-bearing state of the climbing structure for this piece of climbing warning monitoring information to be analyzed, and obtain the trend warning detection result corresponding to each piece of data on the force-bearing state of the climbing structure; Select from the trend warning detection results corresponding to the data on the force-bearing state of the climbing structure in the Z pieces of climbing warning monitoring information to be analyzed according to the improved climbing warning decision network, and obtain v trend warning detection results that meet the set requirements, where v is a positive integer not greater than Z; Jointly import the v trend warning detection results that meet the set requirements, and the data on the force-bearing state of the climbing structure and the climbing warning monitoring information to be analyzed corresponding to each trend warning detection result that meets the set requirements into the cloud service space; Among them, the obtaining of Z pieces of climbing warning monitoring information to be analyzed and the data on the force-bearing state of the climbing structure for each piece of climbing warning monitoring information to be analyzed includes: Obtain Z pieces of climbing warning monitoring information to be analyzed; If there is a target piece of climbing warning monitoring information to be analyzed with a power production scenario different from the set power production scenario among the Z pieces of climbing warning monitoring information to be analyzed, then perform power production scenario mapping on the target piece of climbing warning monitoring information to be analyzed to obtain Z pieces of climbing warning monitoring information to be analyzed under the set power production scenario; Associate the data on the force-bearing state of the climbing structure for each piece of climbing warning monitoring information to be analyzed under the set power production scenario according to the elements of the climbing warning monitoring information to be analyzed; Among them, selecting the trend warning detection results corresponding to the high-altitude structure stress state data of the Z to-be-analyzed high-altitude warning monitoring information according to the improved high-altitude warning decision network, and obtaining v trend warning detection results meeting the set requirements, including: Obtaining a decision threshold; Using the improved high-altitude warning decision network to perform warning decision recognition on each trend warning detection result corresponding to the high-altitude structure stress state data of each to-be-analyzed high-altitude warning monitoring information, and obtaining a third decision coefficient of each trend warning detection result; If the third decision coefficient of any trend warning detection result is higher than the decision threshold, then determine this trend warning detection result as a trend warning detection result meeting the set requirements.

9. An AI data processing system, characterized in that, Including: A memory for storing program instructions and data; A processor for being coupled with the memory and executing the instructions in the memory to implement the method according to any one of claims 1-8.

10. A computer storage medium, characterized in that, Containing instructions, when the instructions are executed on the processor, implementing the method according to any one of claims 1-8.

Images