Big data collaborative management method and system based on PTD Cloud

By using a visual association processing network generated by semi-supervised debugging cost indicators and collaborative business data learning annotations in the PTD Cloud environment, the challenges of big data management and display are solved, and the efficient management and application of PTD Cloud business big data is achieved, improving the The accuracy and efficiency of data mining and presentation are achieved.

CN118035336BActive Publication Date: 2024-12-03BEIJING UNITED MEDIA TECH CO LTD
View PDF 7 Cites 0 Cited by

Patent Information

Application Number
CN202410032765.7
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-01-09
Publication Date
2024-12-03
Estimated Expiration
2044-01-09

AI Technical Summary

Technical Problem

In the PTD Cloud environment, how to effectively manage and utilize a large amount of business big data, mine valuable information from it and display it in an intuitive way has become an urgent problem.

Method used

Using a visual association processing network generated based on semi-supervised debugging cost indicators and collaborative business data learning annotation debugging, feature mining of PTD Cloud service big data to be performed for visual association processing, and through deep residual connection branches, hollow convolution processing branches, and decision-making Tree branches and visual association output branches perform visual optimization feature mining and image involvement description refinement to generate visual association display data.

Benefits of technology

Improve the performance and accuracy of the visual association processing network when mining the features involved in visual optimization elements, ensure that the involved features of visual optimization elements are fully mined from the overall data set, and realize orderly summary and targeted display of chaotic business big data. , improves the calling efficiency and accuracy of specific data sets in the later stage.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118035336B_ABST
    Figure CN118035336B_ABST
Patent Text Reader

Abstract

The big data collaborative management method and system based on PTD Cloud provided by the embodiment of the present invention adopts a method based on semi-supervised debugging cost indicators and collaborative business data learning annotations to debug and mine features of the visual association processing network for PTD Cloud business big data. Among them, the semi-supervised debugging cost indicator takes into account the fluctuations between simple and complex training examples and between the network training cost indicator of image involvement description extraction and the network training cost indicator of visual optimization elements; it can effectively improve the performance when mining the involved features of visual optimization elements and improve the mining accuracy. The introduction of collaborative business data learning annotations ensures that the involved features of visual optimization elements are fully mined from the overall data set, and further improves the efficiency and accuracy of later calls for specific data sets. The method achieves efficient management and application of business big data in the PTD Cloud environment by systematically mining, summarizing and displaying business big data.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of big data technology, and in particular to a big data collaborative management method and system based on PTD Cloud. Background Art

[0002] PTD Cloud is a big data collaborative management technology based on platform, technology and data, which plays an important role in industrial Internet scenarios.

[0003] With the development of the industrial Internet, big data is increasingly used in production and management. However, due to the huge amount of data and scattered distribution, the effective management and utilization of these business big data has become an important but challenging task. Especially in the PTD Cloud environment, how to mine valuable information from a large amount of business data and display it in an intuitive way is an urgent problem to be solved. Summary of the invention

[0004] In order to improve the above problems, the present invention provides a big data collaborative management method and system based on PTD Cloud.

[0005] In a first aspect, a big data collaborative management method based on PTD Cloud is provided, which is applied to a big data collaborative management system, and the method comprises:

[0006] Obtain PTD Cloud business big data to be visualized and associated;

[0007] The PTD Cloud business big data to be visually associated is transmitted to a visual association processing network, and the visual association processing network is used to perform visual optimization element mining and image-related description refinement on the PTD Cloud business big data to be visually associated, so as to obtain visual association display data in the PTD Cloud business big data to be visually associated;

[0008] Among them, the visual association processing network is generated by debugging the visual association processing network to be debugged according to the semi-supervised debugging cost indicator and the collaborative business data learning annotation. The semi-supervised debugging cost indicator improves the quantity offset of learning examples and performs attention processing on the image-involved description extraction project.

[0009] In some embodiments, the visual association processing network includes a cascaded deep residual connection branch, a hole convolution processing branch, a decision tree branch, and a visual association output branch; the visual association processing network is used to perform visual optimization element mining and image involvement description refinement on the PTD Cloud business big data to be visually associated, so as to obtain visual association display data in the PTD Cloud business big data to be visually associated, including:

[0010] The PTD Cloud business big data to be processed for visual association is passed into the deep residual connection branch, and the business data image block in the PTD Cloud business big data to be processed for visual association is reversibly processed through the deep residual connection branch to obtain a reversible processing result corresponding to the business data image block;

[0011] The reversible processing result is transmitted to the dilated convolution processing branch, and the dilated convolution processing branch is used to perform dilated convolution on the reversible processing result to obtain dilated convolution features;

[0012] The hole convolution feature is passed into the decision tree branch, and visual optimization factor mining is performed according to the hole convolution feature through the decision tree branch to obtain a visual optimization factor vector corresponding to the PTD Cloud business big data to be visualized and associated;

[0013] The visual optimization element vector is transmitted to the visual association output branch, and image involvement description is refined based on the visual optimization element vector through the visual association output branch to obtain the visual association display data.

[0014] In some solutions, before transferring the PTD Cloud business big data to be subjected to visual association processing to the visual association processing network, the method further includes:

[0015] Obtaining a business big data learning example, transferring the business big data learning example into the proposed debugging visualization association processing network, and performing visual optimization element mining and image involvement description refinement on the business big data learning example through the proposed debugging visualization association processing network to obtain a visualization association display prediction result;

[0016] The visualization association processing network to be debugged is improved according to the prior visual optimization element vector and the prior visualization association display data corresponding to the business big data learning example, and the visualization association display prediction result to obtain the visualization association processing network.

[0017] In some embodiments, the proposed debugging visualization association processing network includes a deep residual connection branch, a proposed debugging hole convolution processing branch, a proposed debugging decision tree branch, and a proposed debugging visual association output branch; the visual optimization element mining and image involvement description refinement of the business big data learning example through the proposed debugging visualization association processing network to obtain a visualization association display prediction result includes:

[0018] The business big data learning example is passed into the deep residual connection branch, and the business data image block in the business big data learning example is reversibly processed through the deep residual connection branch to obtain a reversible processing result corresponding to the business data image block;

[0019] The reversible processing result is transmitted to the proposed debugging dilated convolution processing branch, and the dilated convolution is performed on the reversible processing result through the proposed debugging dilated convolution processing branch to obtain a dilated convolution feature;

[0020] The hole convolution feature is passed into the decision tree branch to be debugged, and visual optimization factor mining is performed according to the hole convolution feature through the decision tree branch to be debugged to obtain a visual optimization factor vector set corresponding to the business big data learning example;

[0021] The visual optimization element vector set is passed into the proposed debugging visual association output branch, and image involvement description is refined based on the visual optimization element vector set through the proposed debugging visual association output branch to obtain the visual association display prediction result.

[0022] In some schemes, the deep residual connection branch is a pre-debugging component; the step of passing the business big data learning example into the deep residual connection branch, and performing reversible processing on the business data image block in the business big data learning example through the deep residual connection branch to obtain a reversible processing result corresponding to the business data image block includes:

[0023] Performing consistency analysis on the business data image block in the business big data learning example and the business data image block in the image feature pool to obtain an image descriptor corresponding to the business data image block in the business big data learning example;

[0024] The image descriptors corresponding to each of the business data image blocks are subjected to local feature focusing processing through a plurality of local feature enhancement subnetworks, so as to obtain the reversible processing results corresponding to each of the business data image blocks.

[0025] In some embodiments, the step of refining the image-related description based on the visual optimization element vector set through the visual association output branch to be debugged to obtain the visual association display prediction result includes:

[0026] Determine multiple visual optimization elements in the business big data learning example based on the visual optimization element vector set, and determine the business data image block feature corresponding to each of the visual optimization elements based on the visual optimization element vector set and the hole convolution feature;

[0027] Taking any two visual optimization elements among the plurality of visual optimization elements as target visual optimization elements, and acquiring business data image block features corresponding to set business data image blocks corresponding to the target visual optimization elements;

[0028] Aggregating and downsampling the business data image block features corresponding to the set business data image block to obtain image linear description features;

[0029] The image linear description features are processed using an image feature decoding core to obtain a visualization association display prediction result corresponding to the target visual optimization element.

[0030] In some schemes, the improving the visualization association processing network to be debugged according to the prior visual optimization element vector and the prior visualization association display data corresponding to the business big data learning example, and the visualization association display prediction result to obtain the visualization association processing network includes:

[0031] Determine a first network training cost index corresponding to the hole convolution processing branch to be debugged and the decision tree branch to be debugged based on the prior visual optimization element vector;

[0032] Determine a second network training cost indicator corresponding to the visual association output branch to be debugged based on the prior visual association display data and the visual association display prediction result;

[0033] The semi-supervised debugging cost indicator is determined based on the first network training cost indicator and the second network training cost indicator, and network variables of the visual association processing network to be debugged are improved based on the semi-supervised debugging cost indicator to obtain the visual association processing network.

[0034] In some embodiments, determining the first network training cost index corresponding to the hole convolution processing branch to be debugged and the decision tree branch to be debugged based on the prior visual optimization element vector includes:

[0035] Determine a plurality of visual matching strategies based on the business data image blocks and the visual optimization element vector set contained in the business big data learning example, and determine a priori matching strategies based on the priori visual optimization element vector;

[0036] Determining a first visual optimization error and a first feature mapping error corresponding to each of the visual matching strategies, and a second visual optimization error and a second feature mapping error corresponding to the priori matching strategy based on the atrous convolution feature and the mapping feature relationship network;

[0037] Determine a visual matching strategy error corresponding to each of the visual matching strategies based on the first visual optimization error and the first feature mapping error, and determine a priori matching strategy error corresponding to the priori matching strategy based on the second visual optimization error and the second feature mapping error;

[0038] Acquire a global matching error variable according to each of the visual matching strategy errors, and acquire a priori matching error variable based on the priori matching strategy errors;

[0039] The first network training cost indicator is determined by combining the prior matching error variable and the global matching error variable.

[0040] In some embodiments, the determining, based on the prior visual association display data and the visual association display prediction result, a second network training cost indicator corresponding to the visual association output branch to be debugged includes:

[0041] Obtaining a learning example confidence and an explicit example discrimination weight, and improving the visualization association display prediction result based on the learning example confidence and the explicit example discrimination weight to obtain a visualization association display improvement label;

[0042] A discriminant training cost indicator is determined based on the prior visual association display data and the visual association display improved label, and the discriminant training cost indicator is used as the second network training cost indicator.

[0043] In some embodiments, determining the semi-supervised debugging cost indicator based on the first network training cost indicator and the second network training cost indicator includes:

[0044] Obtain a first strengthening factor and a second strengthening factor, wherein the first strengthening factor is less than the second strengthening factor;

[0045] Weighting the first reinforcement factor and the first network training cost index to obtain a first semi-supervised debugging cost index, and weighting the second reinforcement factor and the second network training cost index to obtain a second semi-supervised debugging cost index;

[0046] The first semi-supervised debugging cost indicator is summed with the second semi-supervised debugging cost indicator to obtain the semi-supervised debugging cost indicator.

[0047] In some schemes, when the business big data learning example is passed into the visualization association processing network to be debugged, the method also includes: passing the collaborative business data learning annotation corresponding to the business big data learning example into the visualization association processing network to be debugged for debugging.

[0048] In some schemes, the collaborative business data learning annotation is feature involvement indication information corresponding to the business big data learning example, and the feature involvement indication information includes a first indication and a second indication, wherein the first indication reflects that the relationship between the visual optimization elements in the business big data learning example is a first feature involvement category, and the second indication reflects that the relationship between the visual optimization elements in the business big data learning example is a second feature involvement category, and the first feature involvement category is different from the second feature involvement category.

[0049] In a second aspect, a big data collaborative management system is provided, comprising a processor and a memory communicating with each other, wherein the processor is used to retrieve a computer program from the memory and implement the method described in the first aspect by running the computer program.

[0050] According to a third aspect, a computer-readable storage medium is provided, on which a computer program is stored, and the computer program implements the method described in the first aspect when running.

[0051] The big data collaborative management method and system based on PTD Cloud provided by the embodiment of the present invention uses a visualization association processing network obtained by debugging based on semi-supervised debugging cost indicators and collaborative business data learning annotations to perform feature mining on the PTD Cloud business big data to be subjected to visualization association processing. The semi-supervised debugging cost indicators introduce fluctuations between simple training examples and complex training examples and between the network training cost indicators of image involvement description extraction and the network training cost indicators of visual optimization elements. Therefore, the visualization association processing network to be debugged is debugged according to the semi-supervised debugging cost indicators and collaborative business data learning annotations, which can improve the performance of the visualization association processing network in mining the involved features of visual optimization elements, and improve the accuracy of visual optimization element mining and the mining of involved features of visual optimization elements. In addition, when debugging the visualization association processing network to be debugged, collaborative business data learning annotations are introduced to ensure the performance of mining the involved features of visual optimization elements from the overall data set, thereby ensuring the comprehensiveness of the involved features of the visual optimization elements obtained by mining. On this basis, visualization association processing is performed, and the orderly induction and targeted display of the messy business big data can be realized through the visualization association display data, thereby improving the efficiency and accuracy of the later call for specific data sets.

[0052] In summary, a method based on semi-supervised debugging cost index and collaborative business data learning annotation is used to debug and mine features of visual association processing network for PTDCloud business big data. Among them, the semi-supervised debugging cost index takes into account the fluctuations between simple and complex training examples and between the network training cost index of image involvement description extraction and the network training cost index of visual optimization elements. This strategy can effectively improve the performance in mining the involved features of visual optimization elements and improve the mining accuracy. The introduction of collaborative business data learning annotation ensures that the involved features of visual optimization elements are fully mined from the overall data set, further improving the efficiency and accuracy of later calls for specific data sets. In general, this method realizes the efficient management and application of big data in the PTD Cloud environment by systematically mining, summarizing and displaying messy business big data. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments are briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without creative work.

[0054] Figure 1 A flowchart of a big data collaborative management method based on PTD Cloud provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0055] In order to better understand the above technical scheme, the technical scheme of the present invention is described in detail below through the accompanying drawings and specific embodiments. It should be understood that the embodiments of the present invention and the specific features in the embodiments are detailed descriptions of the technical scheme of the present invention, rather than limitations on the technical scheme of the present invention. In the absence of conflict, the embodiments of the present invention and the technical features in the embodiments can be combined with each other.

[0056] Figure 1 A big data collaborative management method based on PTD Cloud is shown, which is applied to a big data collaborative management system. The method includes the following steps 110 to 120.

[0057] Step 110: Obtain PTD Cloud business big data to be visualized and associated.

[0058] Among them, the PTD Cloud business big data to be visualized and associated mainly involves how to effectively manage and visualize a large amount of business data in the industrial Internet by using platform, technology and data.

[0059] In this step, the "platform" serves as a carrier for data collection, storage and processing. It can support various types of data formats and provide flexible interfaces so that various business data can enter the system smoothly.

[0060] "Technology" here mainly refers to data processing and visualization technology. Using advanced data mining and analysis algorithms, valuable information can be extracted from a large amount of complex business data. At the same time, with the help of modern data visualization technology, this information can be displayed intuitively so that corporate decision makers can better understand the data and make decisions accordingly.

[0061] "Data" is the core of this step, including the collected raw business data and processed data results. In PTD Cloud, all data will undergo strict quality control and security protection to ensure the accuracy and reliability of the data.

[0062] Using PTD Cloud business big data to be visualized and correlated as raw material, it can achieve effective management and visualized correlation processing of business big data, helping enterprises to better understand and use data, thereby optimizing decision-making and improving efficiency.

[0063] In the application scenarios of the Industrial Internet, PTD Cloud business big data may include real-time operation data, historical operation data, maintenance records, fault records, etc. collected from various industrial equipment and systems. These data cover all stages of the equipment life cycle and can reflect the equipment's operating status, performance indicators, health status and other information.

[0064] For example, a factory uses sensors and IoT technology to install various monitoring devices on each machine, which can collect real-time data such as machine temperature, vibration, noise, and current. At the same time, each maintenance situation is also recorded, including maintenance time, replaced parts, maintenance personnel, etc. All of this data will be uploaded to the PTD Cloud platform and become business big data to be visualized and associated.

[0065] The value of these data lies in that they can help factory managers gain insight into the operating status of equipment, predict possible problems, and perform maintenance in advance, thereby avoiding losses caused by downtime. At the same time, through in-depth analysis of these data, it is also possible to find the direction of optimizing equipment performance and improve production efficiency. Therefore, these business big data play a very important role in the Industrial Internet.

[0066] Step 120: The PTD Cloud business big data to be visually associated is transmitted to a visual association processing network, and the visual association processing network is used to perform visual optimization element mining and image involvement description refinement on the PTD Cloud business big data to be visually associated, so as to obtain visual association display data in the PTD Cloud business big data to be visually associated.

[0067] Among them, the visual association processing network is generated by debugging the visual association processing network to be debugged according to the semi-supervised debugging cost indicator and the collaborative business data learning annotation. The semi-supervised debugging cost indicator improves the quantity offset of learning examples and performs attention processing on the image-involved description extraction project.

[0068] Further, a visual association processing network is a network model built using machine learning or deep learning technology, which is specifically used to process and analyze big data, and then visualize the results. Visual optimization factor mining is a process of extracting valuable information from a large amount of data and visualizing it by using data mining technology. Image implicated description extraction is an image processing technology that can extract important features or information from an image. Visual association display data is a data display that can be intuitively understood by users, such as charts, images, etc., generated by the above process. Semi-supervised debugging cost index is an evaluation criterion used to measure model performance in a semi-supervised learning environment, taking into account the influence of labeled data and unlabeled data. Collaborative business data learning annotation may refer to the use of business knowledge and manually annotated data to jointly learn the model during the model training process. The visual association processing network to be debugged refers to the visual association processing network that is about to be optimized or adjusted. The number of learning examples is a situation in machine learning where the distribution of the training data set and the actual data is inconsistent, which may affect the performance of the model. Image implicated description extraction project (task) refers to a specific task or process for extracting features or information from an image. Attention processing is a mechanism within a model that is used to focus on the most important parts and ignore the unimportant parts when processing information.

[0069] In the application scenario of the Industrial Internet, the Visual Relational Processing Network is a deep learning model that is trained to parse and analyze big data collected from various industrial equipment. For example, the network can predict whether a machine is about to fail by learning the patterns of data such as machine temperature, vibration, and noise.

[0070] Visual optimization factor mining refers to finding key factors that can improve equipment performance or production efficiency through data analysis and visualizing them. For example, through the analysis of historical operation data, it may be found that a certain temperature or vibration frequency is highly correlated with equipment failure. After this information is visualized, it can help engineers understand the operating status of the equipment more intuitively and adjust equipment parameters or formulate maintenance plans accordingly.

[0071] Image-related description extraction includes extracting information from equipment monitoring videos, or analyzing X-ray or infrared scan images of equipment. For example, by analyzing monitoring videos, abnormal behavior of equipment (such as overheating, abnormal noise, etc.) can be identified. Or, by analyzing X-ray images, possible faults inside the equipment can be detected.

[0072] All of the above processing steps will generate visual correlation display data, such as line charts, heat maps or 3D models, so that enterprise decision makers and engineers can intuitively understand the operating status of the equipment and make decisions accordingly.

[0073] In the industrial Internet, visual correlation display data is the result of processing and analyzing the above-mentioned various business big data and displaying them in the form of graphics, charts, etc.

[0074] For example, for real-time operation data, a dashboard can be designed that contains real-time readings of indicators such as temperature, vibration, noise, and current, and these readings will change in real time as the data is updated. In this way, operators can monitor the operating status of all equipment on one interface, and if there are any abnormalities, they can be discovered and handled immediately.

[0075] For historical operation data and maintenance records, you can use time series graphs, bar graphs, etc. to display the operation trend and failure history of the equipment. For example, you can plot the average temperature of a machine every day for a year to see if there is an upward trend; you can also count the number of failures per month to see if there is a specific season or month with a higher failure rate.

[0076] In addition, more complex data analysis methods, such as association rules and cluster analysis, can be used to find the correlation between different data, and then display them in the form of scatter plots, network diagrams, etc. For example, it may be found that when a certain indicator (such as temperature) exceeds a certain threshold, the failure rate of the machine will increase significantly; or it may be found that certain types of machines are more prone to certain failures.

[0077] By displaying data in this visual and correlated way, factory managers can not only intuitively see the operating status and problems of the equipment, but also gain insight into potential patterns and trends, thereby making better decisions and management.

[0078] In some complete examples, we can use an application scenario of the intelligent manufacturing industry to specifically introduce the PTD Cloud business big data to be visualized and associated.

[0079] In the smart manufacturing industry, factories generate massive amounts of data every day, including information on machine operating status, product quality, production efficiency, and other aspects. If this information can be effectively used, it can provide support for improving production efficiency, ensuring product quality, and even innovating business models. PTD Cloud, as a big data collaborative management technology based on platform, technology, and data, can help factories achieve this goal.

[0080] For example, step 110 first needs to obtain the PTD Cloud business big data to be visually associated. These data may include raw data collected from various devices and systems on the production line, such as the machine's operating time, downtime, number of failures, production quantity, etc. Next, in step 120, these business big data will be sent to the visual association processing network. This network will analyze and mine the data through a series of algorithms, such as machine learning and deep learning. It may find out the time period when certain equipment often fails and which products have higher production efficiency. Finally, this related information will be presented in the form of charts or images, so that factory managers can see the operating status of the production line at a glance, find out possible problems, and formulate corresponding optimization strategies. Therefore, through the visual association processing of PTD Cloud business big data, factories can better utilize their own data resources, improve production efficiency and product quality, reduce operating costs, and achieve the goal of intelligent manufacturing.

[0081] By using the above embodiment, the visual association processing network obtained by debugging based on the semi-supervised debugging cost index and the collaborative business data learning annotation is used to perform feature mining on the PTD Cloud business big data to be subjected to visual association processing. The semi-supervised debugging cost index introduces the fluctuation between simple training examples and complex training examples and between the network training cost index of image involvement description extraction and the network training cost index of visual optimization elements. Therefore, the visual association processing network to be debugged is debugged according to the semi-supervised debugging cost index and the collaborative business data learning annotation, which can improve the performance of the visual association processing network in mining the involved features of visual optimization elements, and improve the accuracy of visual optimization element mining and the involved features of visual optimization elements. In addition, when debugging the visual association processing network to be debugged, the collaborative business data learning annotation is introduced, thereby ensuring the performance of mining the involved features of visual optimization elements from the overall data set, and then ensuring the comprehensiveness of the involved features of the visual optimization elements obtained by mining. On this basis, visual association processing is performed, which can realize the orderly induction and targeted display of messy business big data through visual association display data, and improve the efficiency and accuracy of later calls for specific data sets.

[0082] In detail, by introducing the semi-supervised debugging cost indicator, the network can better balance the learning of simple training examples and complex training examples, while taking into account the cost of image description extraction and visual optimization factor mining. This can not only improve the performance of the model in mining visual optimization factors, but also improve the accuracy of feature mining.

[0083] By learning annotations with collaborative business data, the network can mine comprehensive relevant features from the overall data set. This means that even if the data set contains a lot of noise or irrelevant information, the network can find the key features and generate more accurate and insightful visual association display data.

[0084] Since the generated visual correlation display data can orderly summarize and specifically display complex business big data, enterprise decision makers can understand the data more quickly and make decisions accordingly, which will undoubtedly greatly improve decision-making efficiency.

[0085] Through visual correlation processing, enterprises can better understand and optimize their business processes. For example, in industrial production, visualized equipment operation data and performance indicators can help engineers discover and solve potential problems of equipment, thereby improving equipment operation efficiency and production quality.

[0086] Semi-supervised learning can effectively utilize unlabeled data, reduce reliance on manually labeled data, and save manpower and time costs. At the same time, accurate predictions and decisions can also avoid ineffective operations and waste, saving corporate resources.

[0087] In some possible examples, the visual association processing network includes a cascaded deep residual connection branch, a hole convolution processing branch, a decision tree branch, and a visual association output branch. Based on this, step 120 performs visual optimization element mining and image involvement description refinement on the PTD Cloud business big data to be visually associated through the visual association processing network to obtain visual association display data in the PTD Cloud business big data to be visually associated, including steps 121-124.

[0088] Step 121: pass the PTD Cloud business big data to be visually associated into the deep residual connection branch, and perform reversible processing on the business data image block in the PTD Cloud business big data to be visually associated through the deep residual connection branch to obtain a reversible processing result corresponding to the business data image block.

[0089] Step 122: passing the reversible processing result to the dilated convolution processing branch, and performing dilated convolution on the reversible processing result through the dilated convolution processing branch to obtain dilated convolution features.

[0090] Step 123: pass the hole convolution feature into the decision tree branch, and perform visual optimization factor mining based on the hole convolution feature through the decision tree branch to obtain a visual optimization factor vector corresponding to the PTDCloud business big data to be visualized and associated.

[0091] Step 124: transfer the visual optimization element vector to the visual association output branch, and perform image involvement description refinement based on the visual optimization element vector through the visual association output branch to obtain the visual association display data.

[0092] In the above embodiment, the deep residual connection branch is a special neural network structure that can help the network better learn the mapping relationship between input and output and solve the common gradient vanishing and explosion problems in deep neural networks. For example, in step 121, the business data image block is first sent to the deep residual connection branch for processing to generate a reversible processing result.

[0093] Dilated convolution processing branch: Dilated convolution is a variant of convolution operation. It introduces "holes" in the convolution kernel to expand its receptive field, and can capture a wider range of information without increasing parameters and computational complexity. In step 122, the reversible processing result is sent to the dilated convolution processing branch to obtain the dilated convolution feature.

[0094] Decision tree branch: Decision tree is a commonly used machine learning model that makes predictions or classifications through a series of rule judgments. In step 123, the dilated convolution feature is sent to the decision tree branch, through which visual optimization elements are mined to obtain a visual optimization element vector corresponding to the original data.

[0095] The visual association output branch is the last branch of the network, which is responsible for generating the final visual association display data based on the visual optimization element vector obtained in the previous step. Specific operations may include image involvement description extraction, that is, extracting useful information from the element vector and converting it into an easy-to-understand image or chart.

[0096] Through the above steps, valuable information can be extracted from the operating data of industrial equipment, such as the health status of the equipment, predicted future failures, etc., and presented in an intuitive way. This method can help factory managers better understand the status of equipment, discover and deal with problems in a timely manner, thereby improving production efficiency and reducing maintenance costs.

[0097] In the field of industrial Internet, this technical solution is introduced by taking a manufacturing company called "SmartFactory" as an example.

[0098] "SmartFactory" uses IoT devices to collect a large amount of PTD Cloud business big data, including real-time machine operation data (such as temperature, vibration frequency, etc.), historical maintenance records, etc. It is hoped that by analyzing this data, the production process can be optimized and equipment failures can be predicted in advance, thereby improving production efficiency and product quality.

[0099] In step 121, the PTD Cloud service big data to be processed is passed to the deep residual connection branch for reversible processing. For example, a thermal imaging image with 1024x1024 pixels is cut into multiple image blocks of 64x64 pixels, and then each image block is feature extracted through a deep residual network, and the result is a 512-dimensional vector, which is the so-called reversible processing result.

[0100] Then, in step 122, the reversible processing result is passed to the dilated convolution processing branch for dilated convolution to obtain dilated convolution features. For example, each 512-dimensional vector is processed by a dilated convolution layer with different receptive fields, and finally a new 256-dimensional vector is generated, which is the dilated convolution feature.

[0101] Then, in step 123, the hole convolution feature is passed into the decision tree branch for visual optimization factor mining. For example, a 256-dimensional vector is input into a pre-trained decision tree model, and the model mines the features that best reflect the device status by judging the importance of each feature. These features constitute the visual optimization factor vector.

[0102] Then, in step 124, the visual optimization factor vector is passed to the visual association output branch for image involvement description extraction to obtain visual association display data. For example, the visual optimization factor vector obtained in the previous step is converted into a color-coded heat map, which is the final visual association display data, which can clearly show the parts of the equipment that may have problems.

[0103] Through the above steps, "SmartFactory" can quickly and accurately identify the key factors that may affect production efficiency and product quality from massive business big data, and display this information in an intuitive way to provide strong support for decision makers.

[0104] In the “SmartFactory” example, the following is a further explanation of the reversible processing results, the dilated convolution features, and the visual optimization feature vector:

[0105] The reversible processing result is obtained by processing the business data image block through the deep residual connection branch. For example, there is a 64x64 pixel image block, which may represent the temperature distribution of a certain part of the machine. After being processed by the deep residual network, a 512-dimensional vector may be obtained, and each dimension represents a feature of the image, such as [0.1, 0.2, ..., 0.5]. These features may include information such as color, texture, shape, etc. of the image.

[0106] The dilated convolution feature is the result of passing the above reversible processing result into the dilated convolution processing branch. The dilated convolution can capture a wider range of contextual information, so a new 256-dimensional vector can be obtained, such as [0.3, 0.4, ..., 0.6]. This vector contains higher-level features of the original image, such as temperature change trends in different regions.

[0107] The visual optimization feature vector is the result of passing the dilated convolution feature into the decision tree branch. The decision tree can help find the most important features, so you may get a smaller vector, such as [0.4, 0.5, ..., 0.6]. This vector may contain only a few dozen elements, but each element is the key factor that has the greatest impact on the device status.

[0108] Through the above steps, not only can valuable information be extracted from a large amount of business data, but this information can also be presented in an efficient and intuitive way, providing strong support for decision makers.

[0109] In some alternative embodiments, before the PTD Cloud business big data to be subjected to visual association processing is transmitted to the visual association processing network, the method further includes steps 210 to 220.

[0110] Step 210: Obtain a business big data learning example, and pass the business big data learning example into the proposed debugging visualization association processing network. Through the proposed debugging visualization association processing network, the business big data learning example is subjected to visual optimization element mining and image involvement description refinement to obtain a visualization association display prediction result.

[0111] Step 220: improving the visualization association processing network to be debugged according to the prior visual optimization element vector and the prior visualization association display data corresponding to the business big data learning example, and the visualization association display prediction result, so as to obtain the visualization association processing network.

[0112] In step 210, for example, a company wants to optimize its services by analyzing its PTD Cloud business big data. Its business big data is passed as a learning example to the visual association processing network to be debugged. These data may contain various information such as customer usage habits, traffic information, failure rate, etc. Through this network, visual optimization factor mining and image involvement description extraction can be performed on these big data, such as identifying which factors affect service quality, or which time periods have the highest traffic. Then, a visual association display prediction result can be obtained, which will display this information in a more intuitive way.

[0113] Next, in step 220, based on the prior visual optimization factor vector and the prior visual association display data corresponding to the business big data learning example, as well as the visual association display prediction results, the visual association processing network to be debugged can be improved. For example, if the prediction results show that the traffic is too high in certain time periods, resulting in a decrease in service quality, the company can adjust its network so that it can better handle large traffic in these time periods. In this way, the visual association processing network can be continuously improved according to actual conditions to optimize its services.

[0114] In summary, this solution can optimize the mining of elements and refine the image-related descriptions of big data through the visual association processing network, so that the business data can be better understood. Based on the visual prediction results, decisions can be made more intuitively and quickly. Through feedback adjustment, the visual association processing network is continuously improved to meet business needs more accurately.

[0115] In some examples, the proposed debugging visualization association processing network includes a deep residual connection branch, a proposed debugging hole convolution processing branch, a proposed debugging decision tree branch, and a proposed debugging visual association output branch. Based on this, step 210 performs visual optimization element mining and image involvement description refinement on the business big data learning example through the proposed debugging visualization association processing network to obtain a visualization association display prediction result, including steps 211-214.

[0116] Step 211: pass the business big data learning example into the deep residual connection branch, and perform reversible processing on the business data image block in the business big data learning example through the deep residual connection branch to obtain a reversible processing result corresponding to the business data image block.

[0117] Step 212: passing the reversible processing result to the proposed debugging dilated convolution processing branch, and performing dilated convolution on the reversible processing result through the proposed debugging dilated convolution processing branch to obtain dilated convolution features.

[0118] Step 213: pass the hole convolution feature into the decision tree branch to be debugged, and perform visual optimization factor mining based on the hole convolution feature through the decision tree branch to be debugged to obtain a visual optimization factor vector set corresponding to the business big data learning example.

[0119] Step 214: passing the visual optimization element vector set to the proposed debugging visual association output branch, and performing image involvement description refinement based on the visual optimization element vector set through the proposed debugging visual association output branch to obtain the visual association display prediction result.

[0120] In the application scenario of the Industrial Internet, taking "SmartFactory" as an example, the detailed explanation of steps 211 to 214 is as follows:

[0121] Suppose there is a set of business big data learning examples about the operating status of machines on a manufacturing line, including real-time images of temperature, vibration frequency and other information. First, these business big data are passed to the deep residual connection branch, and the deep residual network is used to extract features for each image block. For example, a 1024x1024 pixel thermal imaging image can be cut into multiple 64x64 pixel image blocks, and then the deep residual network is used to extract the features of each image block to obtain a 512-dimensional vector corresponding to the original image block, which is the inverse processing result.

[0122] Then, these reversible processing results are passed to the dilated convolution processing branch to be debugged. This branch uses dilated convolution layers of different sizes to process each 512-dimensional vector, thereby capturing more contextual information and obtaining a new 256-dimensional vector, which is the dilated convolution feature.

[0123] Next, these hole convolution features are passed to the decision tree branch to be debugged. The decision tree branch will evaluate the importance of each feature based on the hole convolution features and mine the key features that best reflect the device status. For example, the decision tree model may find that certain features have higher weights when predicting device failures. These features constitute a set of visual optimization feature vectors.

[0124] Finally, these visual optimization feature vector sets are passed to the visual association output branch to be debugged. This branch generates an image description based on the visual optimization feature vector set, such as a color-coded heat map, which clearly and intuitively shows the parts where problems may exist, and obtains the final visual association display prediction results.

[0125] Through the above steps, the key factors that may affect production efficiency and product quality can be accurately found from a large amount of business data, and displayed in an intuitive way, providing effective decision-making basis for decision makers. At the same time, pre-discovery and handling of potential equipment failures can avoid production interruptions, save maintenance costs, and further improve production efficiency and product quality.

[0126] In some examples, the deep residual connection branch is a pre-debugging component. Based on this, in step 211, the business big data learning example is passed to the deep residual connection branch, and the business data image block in the business big data learning example is reversibly processed by the deep residual connection branch to obtain a reversible processing result corresponding to the business data image block, including steps 2111-2112.

[0127] Step 2111: Perform consistency analysis on the business data image blocks in the business big data learning example and the business data image blocks in the image feature pool to obtain image descriptors corresponding to the business data image blocks in the business big data learning example.

[0128] Step 2112: Perform local feature focusing processing on the image descriptors corresponding to each of the business data image blocks through multiple local feature enhancement subnetworks to obtain the reversible processing results corresponding to each of the business data image blocks.

[0129] In the application scenario of the Industrial Internet, taking "SmartFactory" as an example, step 2111 and step 2112 can be explained in more detail.

[0130] For example, SmartFactory collects a large number of machine vibration frequency image blocks as business big data learning examples. These image blocks will be analyzed for consistency with the image blocks in the image feature pool. Specifically, the consistency analysis may compare the feature vectors of the image blocks by calculating cosine similarity and other methods. If the feature vectors of the image blocks are defined as v and u, then their cosine similarity can be expressed as cos(θ) = (v·u) / (||v||||u||), where θ is the angle between v and u, and ||v|| and ||u|| are the lengths (or norms) of v and u, respectively. Through this step, the image descriptor corresponding to the image block in the business big data learning example can be obtained.

[0131] Next, the image descriptors corresponding to each business data image block are processed by local feature focusing through multiple local feature enhancement subnets. For example, if the image descriptor indicates that the vibration frequency of a specific area is significantly higher than that of other areas, the local feature enhancement subnet will highlight this information. The specific implementation method may be to perform weighted processing on the image descriptors through some techniques, such as attention mechanism. In this way, a reversible processing result corresponding to each business data image block can be obtained.

[0132] With this design, the key features related to the business can be extracted more accurately by performing consistency analysis with the image blocks in the image feature pool. Focusing the image descriptor through the local feature enhancement subnet can highlight key information and facilitate subsequent decision-making. The pre-debugged deep residual connection branch can quickly and effectively extract valuable information from a large amount of business data, thereby improving the efficiency and accuracy of prediction.

[0133] In some examples, step 214 performs image involvement description refinement based on the visual optimization element vector set through the proposed debugging visual association output branch to obtain the visual association display prediction result, including steps 2141 to 2144.

[0134] Step 2141: determine multiple visual optimization elements in the business big data learning example based on the visual optimization element vector set, and determine the business data image block features corresponding to each of the visual optimization elements based on the visual optimization element vector set and the hole convolution features.

[0135] Step 2142: Take any two visual optimization elements among the multiple visual optimization elements as target visual optimization elements, and obtain business data image block features corresponding to the set business data image blocks corresponding to the target visual optimization elements.

[0136] Step 2143: Aggregate and downsample the business data image block features corresponding to the set business data image block to obtain image linear description features.

[0137] Step 2144: Use an image feature decoding core to process the image linear description features to obtain a visualization association display prediction result corresponding to the target visual optimization element.

[0138] In the application scenario of the industrial Internet, taking "smart grid" as an example, steps 2141 to 2144 are explained in detail as follows.

[0139] Based on the visual optimization element vector set, multiple visual optimization elements in the business big data learning example are determined. For example, in the power grid monitoring data, factors such as current, voltage, and temperature may be of interest. These are all visual optimization elements. Then, the void convolution features and the visual optimization element vector set can be used to determine the business data image block features corresponding to each visual optimization element.

[0140] Select two elements (such as current and temperature) as target visual optimization elements, and obtain the business data image block features corresponding to the set business data image blocks corresponding to the two target elements. For example, there may be a series of image blocks representing current and temperature changes and their corresponding features.

[0141] Next, the business data image block features corresponding to the set business data image block are aggregated and downsampled. Assuming there are m set business data image blocks and n features, these features can be reduced from mn dimensions to 1n dimensions by using methods such as average pooling or maximum pooling to obtain image linear description features.

[0142] Finally, a decoder is used to process the linear description features of the image to generate a visual correlation display prediction result corresponding to the target visual optimization element. For example, this 1*n-dimensional feature vector can be mapped back to the original image space to generate a heat map representing the correlation between current and temperature.

[0143] Through the above steps, the relationship between factors such as current and temperature can be clearly seen, and based on this, equipment management and optimized operation strategies can be carried out to reduce equipment failure rates and improve grid stability and efficiency. In addition, this solution can also greatly save human resources, because the complex data processing and analysis process is completed by machines, and the staff only need to focus on the final visualization results.

[0144] In some preferred embodiments, step 220 improves the visualization association processing network to be debugged according to the prior visual optimization element vector and prior visualization association display data corresponding to the business big data learning example, and the visualization association display prediction result to obtain the visualization association processing network, including steps 221-223.

[0145] Step 221: Determine a first network training cost indicator corresponding to the proposed debugging hole convolution processing branch and the proposed debugging decision tree branch based on the prior visual optimization element vector.

[0146] Step 222: Determine a second network training cost indicator corresponding to the visual association output branch to be debugged based on the prior visual association display data and the visual association display prediction result.

[0147] Step 223: determine the semi-supervised debugging cost indicator based on the first network training cost indicator and the second network training cost indicator, and perform network variable improvement on the visual association processing network to be debugged based on the semi-supervised debugging cost indicator to obtain the visual association processing network.

[0148] In the application scenario of the Industrial Internet, taking "SmartFactory" as an example, steps 221 to 223 can be explained in detail.

[0149] Based on the prior visual optimization factor vector, determine the first network training cost index corresponding to the hole convolution processing branch to be debugged and the decision tree branch to be debugged. For example, using the mean square error (MSE) as the training cost index, the formula is as follows:

[0150]

[0151] Where n is the number of samples, y_i is the true value, _i is the predicted value. The greater the deviation between the predicted result and the true result, the higher the MSE value.

[0152] Based on the prior visual association display data and the visual association display prediction results, the second network training cost index corresponding to the visual association output branch to be debugged is determined. MSE or other indicators suitable for evaluating image similarity, such as Structural Similarity Index (SSIM), etc., can also be used here.

[0153] Based on the first network training cost indicator and the second network training cost indicator, the semi-supervised debugging cost indicator is determined. The semi-supervised debugging cost indicator may be a weighted average of the two, that is, Total_Cost = w1*Cost1+w2*Cost2;

[0154] Among them, w1 and w2 are weights, Cost1 and Cost2 are the first network training cost index and the second network training cost index respectively. Then, based on the semi-supervised debugging cost index, the network variables of the visual association processing network to be debugged are improved, such as updating the network parameters through optimization algorithms such as gradient descent, so as to obtain the visual association processing network.

[0155] In this way, the network can be personalized tuned based on prior knowledge, making the network more adaptable to specific business needs, thereby improving prediction accuracy. By defining cost indicators with practical business significance, network performance can be evaluated more intuitively and network debugging can be accelerated. Automatically adjusting network parameters based on semi-supervised debugging cost indicators can save human resources and reduce operation and maintenance costs.

[0156] Under some exemplary design ideas, step 221 determines the first network training cost indicator corresponding to the proposed debugging hole convolution processing branch and the proposed debugging decision tree branch based on the prior visual optimization element vector, including steps 2211-2215.

[0157] Step 2211: determine multiple visual matching strategies based on the business data image blocks and visual optimization element vector sets contained in the business big data learning example, and determine a priori matching strategies based on the a priori visual optimization element vectors.

[0158] Step 2212: Determine a first visual optimization error and a first feature mapping error corresponding to each of the visual matching strategies, and a second visual optimization error and a second feature mapping error corresponding to the prior matching strategy based on the atrous convolution feature and the mapping feature relationship network.

[0159] Step 2213: determine the visual matching strategy errors corresponding to the respective visual matching strategies based on the first visual optimization error and the first feature mapping error, and determine the a priori matching strategy errors corresponding to the a priori matching strategies based on the second visual optimization error and the second feature mapping error.

[0160] Step 2214: Obtain a global matching error variable according to each of the visual matching strategy errors, and obtain a priori matching error variable based on the priori matching strategy errors.

[0161] Step 2215: Determine the first network training cost indicator by combining the prior matching error variable and the global matching error variable.

[0162] In the application scenario of the industrial Internet, taking the "intelligent logistics system" as an example, steps 2211 to 2215 can be explained in more detail.

[0163] For example, a series of big data learning examples of warehouse management were collected, which included image patches such as inventory levels and cargo flow speeds and corresponding visual optimization feature vector sets. Based on this information, multiple visual matching strategies were determined. At the same time, there is prior knowledge, such as the cargo flow speed may increase at certain times (such as holidays), which forms a priori matching strategy.

[0164] Then, the first visual optimization error and the first feature mapping error, as well as the second visual optimization error and the second feature mapping error corresponding to each visual matching strategy and the prior matching strategy are determined using the dilated convolutional features and the mapping feature relationship network. These errors may be calculated using a loss function (such as mean square error MSE or cross entropy loss CE).

[0165] Next, the visual matching strategy error corresponding to each visual matching strategy is determined according to the first visual optimization error and the first feature mapping error. Similarly, the a priori matching strategy error corresponding to the a priori matching strategy is determined according to the second visual optimization error and the second feature mapping error.

[0166] Then, a global matching error variable is calculated based on all visible matching strategy errors. For example, the average value of all visible matching strategy errors can be taken as the global matching error variable. Similarly, a priori matching error variable can also be obtained based on a priori matching strategy errors.

[0167] Finally, by combining the prior matching error variable and the global matching error variable, the first network training cost index can be determined. For example, this cost index may be a weighted sum of the two, such as L = Q1*global matching error variable + Q2*prior matching error variable, where Q1 and Q2 are weight coefficients.

[0168] Through the above steps, we can more accurately understand the big data of warehouse management, including key factors such as inventory levels and cargo flow speed, and formulate effective strategies accordingly. During the training process, the system will also take into account prior knowledge to avoid over-reliance on data and improve the robustness of the model. Ultimately, the system can manage warehouses more effectively, improve logistics efficiency, and reduce operating costs.

[0169] In some preferred embodiments, determining the second network training cost indicator corresponding to the visual association output branch to be debugged based on the prior visual association display data and the visual association display prediction result in step 222 includes steps 2221 and 2222.

[0170] Step 2221: Obtain the learning example confidence and the explicit example discrimination weight, and improve the visualization association display prediction result based on the learning example confidence and the explicit example discrimination weight to obtain a visualization association display improvement label.

[0171] Step 2222: determine a discriminant training cost indicator based on the prior visual association display data and the visual association display improved label, and use the discriminant training cost indicator as the second network training cost indicator.

[0172] In the application scenario of the industrial Internet, taking the "intelligent traffic management system" as an example, steps 2221 and 2222 can be explained in detail.

[0173] First, obtain the learning example confidence and explicit example discrimination weight. For example, in the case of predicting traffic flow, if a prediction result comes from a learning example with high confidence (such as rich historical data and good stability), and the prediction result has a significant impact on the decision (that is, it has a high explicit example discrimination weight), then this prediction result will be given a higher weight. Assuming there are n prediction results, each with a confidence of c_i and a discrimination weight of w_i, the visual association display improvement label can be obtained by the following formula:

[0174] Improved_Label=∑(c_i*w_i*Predicted_Result_i) / ∑(c_i*w_i);

[0175] Among them, Predicted_Result_i is the i-th prediction result.

[0176] Then, based on the prior visual association display data and the visual association display improved label, the discriminative training cost index is determined. This cost index can use the mean square error (MSE) or other suitable loss functions. This cost index is used as the second network training cost index.

[0177] In this way, by learning the confidence of the example and the explicit example discrimination weight to improve the prediction results, the accuracy of the prediction results can be improved. Using the discrimination training cost index as the second network training cost index helps the model maintain good prediction performance when facing different types of data. This method can adaptively adjust the weight of the prediction results according to the actual situation, making the model more intelligent.

[0178] In some examples, determining the semi-supervised debugging cost indicator based on the first network training cost indicator and the second network training cost indicator in step 223 includes steps 2231-2233.

[0179] Step 2231: Obtain a first enhancement factor and a second enhancement factor, wherein the first enhancement factor is smaller than the second enhancement factor.

[0180] Step 2232: Weight the first reinforcement factor and the first network training cost indicator to obtain a first semi-supervised debugging cost indicator, and weight the second reinforcement factor and the second network training cost indicator to obtain a second semi-supervised debugging cost indicator.

[0181] Step 2233: sum the first semi-supervised debugging cost indicator and the second semi-supervised debugging cost indicator to obtain the semi-supervised debugging cost indicator.

[0182] In the application scenarios of the industrial Internet, taking "intelligent manufacturing" as an example, steps 2231-2233 can be explained in detail.

[0183] First, two reinforcement factors need to be obtained, the first reinforcement factor and the second reinforcement factor, where the first reinforcement factor is smaller than the second reinforcement factor. These reinforcement factors may be determined based on actual business needs or experience. For example, if it is believed that the second network training cost indicator has a greater impact on the model, then a larger reinforcement factor will be assigned to it.

[0184] Next, the first reinforcement factor is weighted with the first network training cost index to obtain the first semi-supervised debugging cost index. Similarly, the second reinforcement factor is weighted with the second network training cost index to obtain the second semi-supervised debugging cost index. Specifically, assuming that the first reinforcement factor is w1, the second reinforcement factor is w2, the first network training cost index is Cost1, and the second network training cost index is Cost2, then:

[0185] The first semi-supervised debugging cost index = w1*Cost1;

[0186] The second semi-supervised debugging cost indicator = w2*Cost2.

[0187] Finally, the first semi-supervised debugging cost index is summed with the second semi-supervised debugging cost index to obtain the semi-supervised debugging cost index.

[0188] Specifically: semi-supervised debugging cost index = first semi-supervised debugging cost index + second semi-supervised debugging cost index = w1*Cost1+w2*Cost2.

[0189] Through the above steps, we can combine the two network training cost indicators to obtain a more comprehensive semi-supervised debugging cost indicator, so as to better evaluate and optimize the model performance. In addition, by using the reinforcement factor, we can also adjust the weight of each network training cost indicator in the final cost indicator according to the actual business needs, so that the model is more in line with the business goals. This will help improve the prediction accuracy and decision-making efficiency of the intelligent manufacturing system, thereby improving production efficiency and reducing production costs.

[0190] In other independent embodiments, when the business big data learning example is passed into the visualization association processing network to be debugged, the method also includes: passing the collaborative business data learning annotation corresponding to the business big data learning example into the visualization association processing network to be debugged for debugging.

[0191] In other examples, the collaborative business data learning annotation is feature involvement indication information corresponding to the business big data learning example, and the feature involvement indication information includes a first indication and a second indication, wherein the first indication reflects that the relationship between the visual optimization elements in the business big data learning example is a first feature involvement category, and the second indication reflects that the relationship between the visual optimization elements in the business big data learning example is a second feature involvement category, and the first feature involvement category is different from the second feature involvement category.

[0192] In the application scenarios of the Industrial Internet, taking "intelligent manufacturing" as an example, this solution can be explained in detail.

[0193] For example, we are studying a business big data learning example of an automobile production line, which may include various visual optimization factors, such as production efficiency, equipment failure rate, product quality, etc. For these factors, there may be two different feature involvement categories: the first may be a positive correlation (such as production efficiency and product quality), and the second may be a negative correlation (such as equipment failure rate and product quality).

[0194] At this time, the collaborative business data learning annotation is the feature-related indication information corresponding to these business big data learning examples. For example, the first indication may indicate that production efficiency is positively correlated with product quality, that is, when production efficiency increases, product quality will also increase; the second indication may indicate that equipment failure rate is negatively correlated with product quality, that is, when equipment failure rate increases, product quality will decrease.

[0195] When business big data learning examples and collaborative business data learning annotations are passed into the visual association processing network to be debugged for debugging, the network can understand the relationship between these elements more accurately and optimize accordingly.

[0196] With this design, by introducing collaborative business data learning annotations, the network can more accurately understand the relationship between various elements, thereby improving the accuracy of prediction and optimization. By simultaneously passing in learning examples and corresponding annotations, the network can obtain more information during the learning process and improve learning efficiency. Clear feature involvement indication information can guide the network to optimize, making the optimization results more in line with actual needs.

[0197] Based on the above, a big data collaborative management system is provided, including a processor and a memory that communicate with each other, the processor is used to call a computer program from the memory, and implement the above method by running the computer program.

[0198] Based on the above, a computer-readable storage medium is provided, on which a computer program is stored, and the computer program implements the above method when running.

[0199] The above are only embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent substitution, improvement, etc. made within the spirit and principle of the present invention should be included in the scope of the claims of the present invention.

Claims

1. A big data collaborative management method based on PTD cloud, characterized in that: Applied to a big data collaborative management system, the method includes: Obtain the PTD cloud business big data to be visualized and associated; The PTD cloud business big data to be visually associated is transmitted to a visual association processing network, and the visual association processing network is used to perform visual optimization element mining and image-related description refinement on the PTD cloud business big data to be visually associated, so as to obtain visual association display data in the PTD cloud business big data to be visually associated; The visual association processing network is generated by debugging the visual association processing network to be debugged according to the semi-supervised debugging cost index and the collaborative business data learning annotation, the semi-supervised debugging cost index improves the number of learning examples and performs attention processing on the image involvement description refinement project; the visual association processing network includes a cascaded deep residual connection branch, a hole convolution processing branch, a decision tree branch and a visual association output branch; the visual optimization element mining and image involvement description refinement of the PTD cloud business big data to be visually associated with the visual association processing network are performed to obtain the visual association display data in the PTD cloud business big data to be visually associated with the visual association processing, including: The PTD cloud business big data to be processed for visual association is passed into the deep residual connection branch, and the business data image block in the PTD cloud business big data to be processed for visual association is reversibly processed through the deep residual connection branch to obtain a reversible processing result corresponding to the business data image block; The reversible processing result is transmitted to the dilated convolution processing branch, and the dilated convolution processing branch is used to perform dilated convolution on the reversible processing result to obtain dilated convolution features; The hole convolution feature is passed into the decision tree branch, and visual optimization factor mining is performed according to the hole convolution feature through the decision tree branch to obtain a visual optimization factor vector corresponding to the PTD cloud business big data to be visualized and associated; The visual optimization element vector is transmitted to the visual association output branch, and image involvement description is refined based on the visual optimization element vector through the visual association output branch to obtain the visual association display data.

2. The method according to claim 1, characterized in that Before transferring the PTDcloud business big data to be subjected to visual association processing to the visual association processing network, the method further includes: Acquire a business big data learning example, transfer the business big data learning example to the proposed debugging visualization association processing network, and perform visual optimization element mining and image involvement description refinement on the business big data learning example through the proposed debugging visualization association processing network to obtain a visualization association display prediction result; The visualization association processing network to be debugged is improved according to the prior visual optimization element vector and the prior visualization association display data corresponding to the business big data learning example, and the visualization association display prediction result to obtain the visualization association processing network.

3. The method according to claim 2, characterized in that The proposed debugging visualization association processing network includes a deep residual connection branch, a proposed debugging hole convolution processing branch, a proposed debugging decision tree branch and a proposed debugging visual association output branch; the proposed debugging visualization association processing network is used to perform visual optimization element mining and image involvement description refinement on the business big data learning example to obtain a visualization association display prediction result, including: The business big data learning example is passed into the deep residual connection branch, and the business data image block in the business big data learning example is reversibly processed through the deep residual connection branch to obtain a reversible processing result corresponding to the business data image block; The reversible processing result is transmitted to the proposed debugging dilated convolution processing branch, and the dilated convolution is performed on the reversible processing result through the proposed debugging dilated convolution processing branch to obtain a dilated convolution feature; The hole convolution feature is passed into the decision tree branch to be debugged, and visual optimization factor mining is performed according to the hole convolution feature through the decision tree branch to be debugged to obtain a visual optimization factor vector set corresponding to the business big data learning example; The visual optimization element vector set is passed into the proposed debugging visual association output branch, and image involvement description is refined based on the visual optimization element vector set through the proposed debugging visual association output branch to obtain the visual association display prediction result.

4. The method according to claim 3, characterized in that The deep residual connection branch is a pre-debugging component; the business big data learning example is passed into the deep residual connection branch, and the business data image block in the business big data learning example is reversibly processed by the deep residual connection branch to obtain a reversible processing result corresponding to the business data image block, including: Performing consistency analysis on the business data image block in the business big data learning example and the business data image block in the image feature pool to obtain an image descriptor corresponding to the business data image block in the business big data learning example; The image descriptors corresponding to each of the business data image blocks are subjected to local feature focusing processing through a plurality of local feature enhancement subnetworks, so as to obtain the reversible processing results corresponding to each of the business data image blocks.

5. The method according to claim 3, characterized in that: The step of refining the image involvement description based on the visual optimization element vector set through the visual association output branch to be debugged to obtain the visual association display prediction result includes: Determine multiple visual optimization elements in the business big data learning example based on the visual optimization element vector set, and determine the business data image block feature corresponding to each of the visual optimization elements based on the visual optimization element vector set and the hole convolution feature; Taking any two visual optimization elements among the plurality of visual optimization elements as target visual optimization elements, and acquiring business data image block features corresponding to set business data image blocks corresponding to the target visual optimization elements; Aggregating and downsampling the business data image block features corresponding to the set business data image block to obtain image linear description features; The image linear description features are processed using an image feature decoding core to obtain a visualization association display prediction result corresponding to the target visual optimization element.

6. The method according to claim 2, characterized in that The step of improving the visualization association processing network to be debugged according to the prior visual optimization element vector and the prior visualization association display data corresponding to the business big data learning example, and the visualization association display prediction result to obtain the visualization association processing network includes: Determine a first network training cost index corresponding to a hole convolution processing branch to be debugged and a decision tree branch to be debugged based on the prior visual optimization element vector; Determine a second network training cost indicator corresponding to the visual association output branch to be debugged based on the prior visual association display data and the visual association display prediction result; The semi-supervised debugging cost indicator is determined based on the first network training cost indicator and the second network training cost indicator, and network variables of the visual association processing network to be debugged are improved based on the semi-supervised debugging cost indicator to obtain the visual association processing network.

7. The method according to claim 6, characterized in that The determining, based on the prior visual optimization element vector, a first network training cost index corresponding to the hole convolution processing branch to be debugged and the decision tree branch to be debugged includes: Determine a plurality of visual matching strategies based on the business data image blocks and the visual optimization element vector set contained in the business big data learning example, and determine a priori matching strategies based on the priori visual optimization element vector; Determining a first visual optimization error and a first feature mapping error corresponding to each of the visual matching strategies, and a second visual optimization error and a second feature mapping error corresponding to the prior matching strategy based on the dilated convolution feature and the mapping feature relationship network; Determine a visual matching strategy error corresponding to each of the visual matching strategies based on the first visual optimization error and the first feature mapping error, and determine a priori matching strategy error corresponding to the priori matching strategy based on the second visual optimization error and the second feature mapping error; Acquire a global matching error variable according to each of the visual matching strategy errors, and acquire a priori matching error variable based on the priori matching strategy errors; Determine the first network training cost indicator by combining the prior matching error variable and the global matching error variable; The determining of the second network training cost index corresponding to the visual association output branch to be debugged based on the prior visual association display data and the visual association display prediction result includes: Obtaining a learning example confidence and an explicit example discrimination weight, and improving the visualization association display prediction result based on the learning example confidence and the explicit example discrimination weight to obtain a visualization association display improvement label; Determine a discriminative training cost indicator based on the prior visual association display data and the visual association display improved label, and use the discriminative training cost indicator as the second network training cost indicator; Wherein, determining the semi-supervised debugging cost indicator based on the first network training cost indicator and the second network training cost indicator includes: Obtain a first strengthening factor and a second strengthening factor, wherein the first strengthening factor is less than the second strengthening factor; Weighting the first reinforcement factor and the first network training cost index to obtain a first semi-supervised debugging cost index, and weighting the second reinforcement factor and the second network training cost index to obtain a second semi-supervised debugging cost index; The first semi-supervised debugging cost indicator is summed with the second semi-supervised debugging cost indicator to obtain the semi-supervised debugging cost indicator.

8. The method according to claim 2, characterized in that: When the business big data learning example is transferred into the visualization association processing network to be debugged, the method further includes: transferring the collaborative business data learning annotation corresponding to the business big data learning example into the visualization association processing network to be debugged for debugging; Among them, the collaborative business data learning annotation is feature involvement indication information corresponding to the business big data learning example, and the feature involvement indication information includes a first indication and a second indication, wherein the first indication reflects that the relationship between the visual optimization elements in the business big data learning example is a first feature involvement category, and the second indication reflects that the relationship between the visual optimization elements in the business big data learning example is a second feature involvement category, and the first feature involvement category is different from the second feature involvement category.

9. A big data collaborative management system, characterized in that: The big data collaborative management system includes a processor and a memory that communicate with each other, and the processor is used to call a computer program from the memory and implement the method described in any one of claims 1-8 by running the computer program.

Citation Information

Patent Citations

  • Mobile intelligent patrol inspection management system for new energy power generation equipment and patrol inspection method

    CN106127879A

  • Clinical data mining analysis and aided decision-making method based on Internet integrated medical platform

    CN109830303A

  • Big data mining method and system based on cloud edge collaboration

    CN113761028A

  • Intelligent information mining method and system based on digital service

    CN115858602A

  • Remote sensing image road extraction method based on asymmetric codec

    CN116246170A