AI self-adaptive sub-domain management method and system for mass equipment terminals of articulated naturality web

Abstract

The invention relates to an articulated naturality web mass device terminal AI self-adaptive domain division management method and system, and the method comprises the steps: obtaining multi-source data from each device terminal, and carrying out the preprocessing of the multi-source data; and carrying out standardization processing on the preprocessed multi-source data through TensorFlow to obtain standardized data, and carrying out normalization processing on the standardized data by writing PyTorch codes. And extracting key features from the text type data of each device by using a Hive function library and a self-defined function, and creating device operation trend features based on the key features through combination of Spark and a deep learning framework. And automatically adjusting the fusion proportion of the visual features and the time sequence features in the equipment operation trend features according to the current network load condition. And constructing a relation graph of each equipment terminal, and analyzing the relation graph through a graph clustering algorithm to divide the equipment of which the correlation degree exceeds a first threshold value into the same region, and generating an initial domain division scheme.

Description

Technical Field

[0001] The present invention relates to the technical field of visual networking and big data management, and in particular to a method and system for AI adaptive domain management of a large number of device terminals in a visual networking. Background Art

[0002] With the rapid development of information technology, as a new network architecture integrating technologies such as high-definition video communication, Internet of Things, and big data, visual networking has been widely used globally. In the early stage, visual networking was mainly applied to the field of video conferencing to meet the needs of remote communication and collaboration. Nowadays, its application scenarios have expanded to many important fields such as smart city construction, intelligent security monitoring, distance education, and telemedicine.

[0003] In smart city construction, visual networking integrates terminal devices such as video surveillance devices and sensors in the city to achieve a comprehensive perception and real-time monitoring of the city's operation status, providing strong data support for urban management decision-making. In intelligent security monitoring, visual networking can collect and analyze a large amount of surveillance camera data to achieve rapid early warning and accurate identification of abnormal events, improving the city's security prevention capabilities.

[0004] Currently, the existing visual networking domain management methods usually divide domains according to a single dimension such as the geographical location and business type of devices, and then set up regional management centers to manage the devices within each region. The regional management center is responsible for tasks such as device registration and authentication, status monitoring, and configuration management. Data interaction and sharing between regions are carried out through fixed communication protocols. In terms of data security, conventional encryption algorithms are used to encrypt data transmission. However, this traditional domain division method is only based on a single dimension, making it difficult to adapt to the diverse changes in device terminal types and business requirements in visual networking. When the number of devices and business types change dynamically, it is difficult to adjust the domains in a timely and reasonable manner, resulting in low management efficiency. Secondly, the existing visual networking domain management models rely more on manual experience and preset rules, and have weak capabilities in real-time monitoring of device status, fault prediction, and dynamic allocation of resources. When facing complex data generated by a large number of devices, it is difficult to make accurate decisions quickly. Thirdly, its conventional encryption algorithms have certain limitations in ensuring data security, making it difficult to meet the extremely high requirements for data security and privacy in industries such as finance and healthcare. Moreover, the fixed data interaction protocol is difficult to be optimized according to different data types and business requirements, resulting in slow data transmission speed.

[0005] In summary, the existing visual networking domain management methods have low management efficiency, poor data security, and the flexibility of domain management needs to be further improved during the process of visual networking domain management. Summary of the Invention

[0006] Based on this, it is necessary to provide a method and system for AI adaptive domain management of a large number of visual network device terminals with high management efficiency, good flexibility, and the ability to ensure data security for the above-mentioned technical problems.

[0007] The present invention provides a method for AI adaptive domain management of a large number of visual network device terminals, and the method includes: Obtain multi-source data from each device terminal, and preprocess the multi-source data, where the preprocessing includes outlier detection and distributed data deduplication; Perform standardization processing on the preprocessed multi-source data through TensorFlow to obtain standardized data, and perform normalization processing on the standardized data by writing PyTorch code; Use the function library and custom functions of Hive to extract key features from the text data of each device, and create device operation trend features based on the key features through Spark combined with a deep learning framework; Automatically adjust the fusion ratio of visual features and temporal features in the device operation trend features according to the current network load condition; Construct a relationship graph of each device terminal, and analyze the relationship graph through a graph clustering algorithm to divide devices with an association degree exceeding a first threshold into the same area, generating an initial domain division scheme; Among them, the multi-source data includes geographical location information, functional type information, business requirement information, and historical operation data and real-time status data of each device. The historical operation data includes the CPU usage rate of the device per hour, the daily operation duration, and the number of faults per month. The real-time status data includes the current CPU usage rate, memory occupancy rate, and network connection status of the device; the nodes in the relationship graph represent each device, and the edges represent the association weights between devices, which are determined by the communication frequency and geographical distance between devices.

[0008] In one embodiment, the obtaining of multi-source data from each device terminal and the preprocessing of the multi-source data include: Calculate the quartiles and interquartile ranges of each feature column in the multi-source data by writing HiveQL statements, and set the outlier range based on the quartiles and interquartile ranges to perform outlier detection on the multi-source data; Call the elastic distributed dataset or deduplication function of Spark to perform deduplication processing on the multi-source data.

[0009] In one embodiment, the performing of standardization processing on the preprocessed multi-source data through TensorFlow to obtain standardized data, and the performing of normalization processing on the standardized data by writing PyTorch code includes: Calculate the mean and standard deviation of each numerical data in the multi-source data by constructing a TensorFlow computational graph, and call the Z-score normalization algorithm to normalize each data point based on the mean and standard deviation, so that the data of the same type of performance indicators of different devices are at the same order of magnitude; Normalize the data with a fixed range of each device to a preset interval by using the normalization function of PyTorch, and normalize the data with a fixed range by writing PyTorch code; Among them, the numerical data includes the CPU usage rate and memory occupancy rate of each device, and the data with a fixed range includes video quality scores and device signal strengths.

[0010] In one embodiment, extracting key features from the text data of each device by using the function library and custom functions of Hive, and creating device operation trend features based on the key features by combining Spark with a deep learning framework, including: Extract keyword information of each device from the text data by writing UDF functions and using regular expressions, where the keyword information includes the brand, model, function, and device business type of the device; Use the recurrent neural network and long short-term memory network in the deep learning framework, and adopt the distributed training mechanism of Spark to learn the preprocessed multi-source data to create the device operation trend features; Among them, the device operation trend features include the change trend of device performance over time and the traffic fluctuation range.

[0011] In one embodiment, extracting key features from the text data of each device by using the function library and custom functions of Hive, and creating device operation trend features based on the key features by combining Spark with a deep learning framework, further includes: Process the image data of the video capture device through the multi-layer convolutional layer and pooling layer of the convolutional neural network to extract the key features for judging the device function type in the image data, including the object contour, color feature, and texture information in the image data; Input the historical operation data into the recurrent neural network in chronological order, and use the recurrent structure inside the recurrent neural network to capture the dependency relationship between historical operation data, so as to predict the traffic change trend in the first future time period by learning the historical operation data.

[0012] In one embodiment, constructing a relationship graph of each device terminal, and analyzing the relationship graph through a graph clustering algorithm to divide devices with an association degree exceeding a first threshold into the same area to generate an initial domain division scheme, and then including: Based on the initial domain division scheme, a deep learning model that combines a convolutional neural network and a recurrent neural network outputs a predicted traffic heat map within the first future time period to display the traffic change trends of different domains at different time points; According to the traffic change trends of different domains and the actual operating status of the devices, cross-domain transfer of device terminals is performed, and the broadband quota is dynamically adjusted.

[0013] In one embodiment, the method further includes: Evaluating the effectiveness of the initial domain division scheme, and when the device latency reduction after implementing the initial domain division scheme is not less than the second threshold, quantifying the implementation effect of the initial domain division scheme to obtain a policy feedback; Obtaining the latest device operation data and policy feedback, and optimizing and training the combined deep learning model based on the latest device operation data and policy feedback.

[0014] The present invention also provides a vision network massive device terminal AI adaptive domain division management system, and the system includes: A data preprocessing module, configured to obtain multi-source data from each device terminal and preprocess the multi-source data, where the preprocessing includes outlier detection and distributed data deduplication; A standardization and normalization module, configured to perform standardization processing on the preprocessed multi-source data through TensorFlow to obtain standardized data, and perform normalization processing on the standardized data by writing PyTorch code; A feature extraction module, configured to extract key features from the text data of each device by using the function library and custom functions of Hive, and create device operation trend features based on the key features through Spark in combination with a deep learning framework; A dynamic adjustment module, configured to automatically adjust the fusion ratio of visual features and temporal features in the device operation trend features according to the current network load condition; A domain division policy generation module, configured to construct a relationship graph of each device terminal, and analyze the relationship graph through a graph clustering algorithm to divide devices with an association degree exceeding the first threshold into the same area to generate an initial domain division scheme; Wherein, the multi-source data includes geographical location information, function type information, service requirement information, and historical operation data and real-time status data of each device, the historical operation data includes the CPU usage rate of the device per hour, the daily operation duration, and the monthly failure times, and the real-time status data includes the current CPU usage rate, memory occupancy rate, and network connection status of the device; the nodes in the relationship graph represent each device, and the edges represent the association weights between the devices, which are determined by the communication frequency and geographical distance between the devices.

[0015] The present invention also provides an electronic device, including a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, the method for AI adaptive domain management of a large number of device terminals in a visual networking as described in any one of the above is implemented.

[0016] The present invention also provides a computer storage medium storing a computer program, and when the computer program is executed by a processor, the method for AI adaptive domain management of a large number of device terminals in a visual networking as described in any one of the above is implemented.

[0017] The above method and system for AI adaptive domain management of a large number of device terminals in a visual networking create device operation trend features based on key features through Spark combined with a deep learning framework for multi-dimensional adaptive domain division, and adjust the domain division strategy in real time according to the dynamic changes of device terminals and service requirements. On this basis, an intelligent distributed area management center is established, effectively improving the anti-attack ability and resource utilization efficiency of the domain management system. The method also designs a hierarchical AI collaborative management architecture based on the deep learning framework, improving the collaborative efficiency and system maintainability of domain management, optimizing the data interaction and sharing mechanism, ensuring the high efficiency and security of data transmission, and meeting the compliance requirements of different industries. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0019] Figure 1 One of the flow diagrams of the method for AI adaptive domain management of a large number of device terminals in a visual networking provided by the present invention; Figure 2 The schematic diagram of the visual networking domain management architecture of the method for AI adaptive domain management of a large number of device terminals in a visual networking in a specific embodiment provided by the present invention; Figure 3 Another flow diagram of the method for AI adaptive domain management of a large number of device terminals in a visual networking provided by the present invention; Figure 4 Another flow diagram of the method for AI adaptive domain management of a large number of device terminals in a visual networking provided by the present invention; Figure 5 Another flow diagram of the method for AI adaptive domain management of a large number of device terminals in a visual networking provided by the present invention; Figure 6It is the fifth flowchart of the AI adaptive domain management method for a large number of device terminals in the visual networking provided by the present invention; Figure 7 It is the sixth flowchart of the AI adaptive domain management method for a large number of device terminals in the visual networking provided by the present invention; Figure 8 It is the seventh flowchart of the AI adaptive domain management method for a large number of device terminals in the visual networking provided by the present invention; Figure 9 It is the structural diagram of the AI adaptive domain management system for a large number of device terminals in the visual networking provided by the present invention; Figure 10 It is the internal structure diagram of the electronic device provided by the present invention. Detailed implementation manners

[0020] To make the objectives, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Apparently, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0021] Next, in combination with Figures 1 to 10 Describe the AI adaptive domain management method and system for a large number of device terminals in the visual networking of the present invention.

[0022] As Figure 1 shown, in one embodiment, an AI adaptive domain management method for a large number of device terminals in the visual networking includes the following steps: Step S110, obtain multi-source data from each device terminal, and perform preprocessing on the multi-source data, where the preprocessing includes outlier detection and distributed data deduplication.

[0023] Among them, the multi-source data includes geographical location information, function type information, service requirement information, and historical operation data and real-time status data of each device. The historical operation data includes the CPU usage rate of the device per hour, the daily operation duration, and the monthly failure times. The real-time status data includes the current CPU usage rate, memory occupancy rate, and network connection status of the device.

[0024] Specifically, the domain management center obtains multi-source data from each device terminal, and performs preprocessing of outlier detection and distributed data deduplication on the obtained multi-source data to obtain the preprocessed multi-source data.

[0025] In combination with Figure 2As shown, in a specific embodiment, the AI adaptive domain management method for a large number of device terminals in the vision Internet of Things provided by the present invention, with the help of the powerful data collection ability of the vision Internet of Things, comprehensively collects various types of information of device terminals through multiple channels such as sensors, device management system interfaces, and network traffic monitoring tools. The geographical location information is accurate to longitude and latitude, which can not only locate the city and region where the device is located, but also be refined to the specific street and building floor, so as to perform more accurate domain division based on the geographical location. The function type information details whether the device is a high-definition camera, an infrared sensor, or a data forwarding device, etc., and clarifies its functional role in the vision Internet of Things. The business requirement information deeply understands the business scenarios served by the device, such as government emergency command, financial transaction monitoring, medical remote consultation, etc., and grasps the special requirements of different businesses for device performance, data security, and real-time performance. At the same time, continuously collect the historical operation data of the device in the past year or even longer, including the hourly data traffic fluctuation, daily operation duration, monthly failure times, etc., as well as the real-time status data, such as the current CPU usage rate, memory occupancy rate, network connection status, etc. Use AI big data analysis tools, such as Hive and Spark in the Hadoop ecosystem, combined with deep learning frameworks TensorFlow and PyTorch, to preprocess this large amount of data. Through data cleaning algorithms, such as the outlier detection method based on statistics, remove the noise data generated due to device failures, network interference, etc.; use normalization algorithms, such as min-max normalization and Z-score normalization, to unify different types of data into the numerical range of 0-1 or -1 to 1, making the data comparable and laying a foundation for subsequent in-depth analysis.

[0026] In this embodiment, data cleaning includes outlier detection based on Hive and distributed data deduplication of Spark. Among them, outlier detection based on Hive uses the query language HiveQL of Hive combined with statistical methods, such as the box plot method. By writing HiveQL statements, calculate the quartiles (Q1, Q3) and interquartile range (IQR = Q3 - Q1) of each feature column in the dataset. Set the outlier range as the data points less than Q1 - 1.5 * IQR or greater than Q3 + 1.5 * IQR. For example, for the device traffic data column, through the aggregation function and conditional filtering statements of Hive, the abnormal traffic values that may be caused by device failures, network fluctuations, etc. can be efficiently filtered out. For the image resolution data of video acquisition devices, if there are values that do not conform to common resolution specifications (such as much less than 1080p or much greater than 4K), they can also be identified as outliers through a similar method.

[0027] Distributed data deduplication in Spark leverages the distributed computing power of Spark and uses its deduplication functions for RDD (Resilient Distributed Dataset) or DataFrame. For example, in the historical operation records of devices, there may be duplicate records. By using the drop_Duplicates function in Spark, deduplication is performed on records containing key information such as device ID, timestamp, and operation status. Since Spark can process data in parallel in a cluster environment, for a large amount of device operation data, the deduplication operation can be completed in a short time, improving data quality.

[0028] Step S120: Standardize the preprocessed multi-source data through TensorFlow to obtain standardized data, and normalize the standardized data by writing PyTorch code.

[0029] Specifically, the domain management center standardizes the preprocessed multi-source data through TensorFlow and normalizes the standardized data by writing PyTorch code, so that different types of data can converge better in subsequent deep learning model training and provide more effective inputs for model learning.

[0030] It should be noted that TensorFlow is an open-source machine learning framework that supports multiple programming languages and platforms, can efficiently process large-scale datasets, and implement complex model training.

[0031] Combined Figure 2 As shown, in a specific embodiment, for the AI adaptive domain management method for a large number of device terminals in the visual networking provided by the present invention, in TensorFlow, for numerical data, such as continuous data like the CPU usage rate and memory occupancy rate of devices, the Z - score standardization method is adopted. By constructing a TensorFlow computational graph, the mean and standard deviation of the data are calculated, and then each data point is standardized, that is, (x - μ) / σ, where x is the original data point, μ is the mean, and σ is the standard deviation. After such processing, the mean of the data is 0 and the standard deviation is 1, enabling the data of the same type of performance indicators of different devices to be at the same order of magnitude and comparable. For example, for video capture devices of different brands and models, the original value ranges of their CPU performance parameters vary greatly. After standardization, they can be compared and analyzed on the same dimension.

[0032] In addition, for some data with fixed ranges in the device, such as video quality scores (0 - 10 points), signal strength of the device (0 - 100%), etc., the normalization function of PyTorch is used to normalize them to the range of 0 - 1. By writing PyTorch code, the data is divided by its maximum value to achieve data normalization. This normalization method enables different types of data to converge better in subsequent deep learning model training and provides more effective input for model learning.

[0033] Step S130: Use the function library of Hive and custom functions to extract key features from the text data of each device, and create device operation trend features based on the key features through Spark combined with a deep learning framework.

[0034] Specifically, the domain management center uses the function library of Hive and custom functions to extract key features from the text data of each device, and creates device operation trend features based on the key features through Spark combined with a deep learning framework.

[0035] Combined Figure 2 As shown, in a specific embodiment, the AI self-adaptive domain management method for a large number of device terminals in the vision network provided by the present invention uses the function library of Hive and custom functions (UDF) to extract key features from the text data of the device. For example, for the description information field of the device, by writing a UDF function, information such as the brand, model, and function keywords of the device is extracted using regular expressions. For the business requirement description of the device, keywords related to the business type, such as "government affairs", "finance", "medical care", etc., are extracted and converted into classification features for subsequent analysis.

[0036] In the process of creating features by Spark combined with a deep learning framework, Spark is used in combination with TensorFlow or PyTorch to create new features. For example, for the time series of historical operation data of the device, using recurrent neural networks (RNN) or long short-term memory networks (LSTM) in the deep learning framework, and through the distributed training mechanism of Spark, the data is learned to extract device operation trend features, such as the change trend of device performance over time, traffic fluctuation patterns, etc. These newly created features can more deeply reflect the operating state and potential laws of the device, and provide richer information for subsequent multi-dimensional intelligent domain division and dynamic adjustment.

[0037] In this embodiment, the multi-dimensional data collected in real time by the terminal device includes: Device status: battery power, storage, connection stability.

[0038] Environmental data: geographical location, task priority, security level.

[0039] Dynamic events: device failure alarms, network congestion, burst task requests.

[0040] Network bandwidth: refers to the ability of the network to transmit data, usually measured in Mbps. During the live broadcast, it is necessary to ensure sufficient upload bandwidth according to the video and audio bitrates to ensure that the live broadcast data can be stably transmitted to the server. At the same time, the viewer side also needs to have sufficient download bandwidth to watch the live broadcast smoothly, otherwise problems such as freezing and slow loading may occur.

[0041] Latency: refers to the time difference between when the camera captures the video and audio data and when the viewer side receives it. For live broadcasts, low latency is very important, especially in live broadcast scenarios with strong interactivity, such as real-time Q&A and online teaching. Generally, a live broadcast latency of within 1 second is considered a good level, and more than 3 seconds may affect the viewer's viewing experience and interactive effect.

[0042] Packet loss rate: refers to the proportion of data packets lost during network transmission to the total number of packets sent. A too high packet loss rate will cause problems such as freezing, screen distortion, and audio interruption in the video, seriously affecting the live broadcast quality. Generally, the packet loss rate should be controlled below 1% to ensure the smoothness of the live broadcast.

[0043] Jitter: refers to the fluctuation of the arrival time of network data packets. Network jitter will cause the video and audio to play smoothly, resulting in problems such as freezing and intermittent sound. During the live broadcast, it is necessary to use technologies such as buffering to deal with network jitter to ensure the stability of the playback.

[0044] In this embodiment, a complex deep learning model that integrates a Convolutional Neural Network (CNN) and a Recurrent Neural Network (RNN) is constructed. The CNN part processes the image data of the video acquisition device through multiple convolutional layers and pooling layers to extract key features in the images, such as object contours, color features, texture information, etc., so as to judge the functional type of the device, such as whether it has the function of intelligent behavior analysis. The RNN part inputs the historical operation data of the device in chronological order, and uses its internal recurrent structure to capture the time-dependent relationships in the data. By learning the data such as device traffic and load in the past period of time, it predicts the peak and trough periods of traffic in the next week, accurate to specific dates and time periods. The preprocessed data is input into this trained model, and the model realizes intelligent domain division in multiple dimensions such as geographical location, functional type, and business requirements according to the features and rules learned by the device in different dimensions. For example, devices located in the core area of the same city, with the function of high-definition video surveillance and serving the urban traffic management business, are divided into one area. This domain division method fully considers various characteristics of the devices, making the domain division result more in line with the actual business requirements.

[0045] It should be noted that a dynamic adjustment model is constructed based on the reinforcement learning algorithm. The agent in the model is responsible for generating different domain adjustment strategies, and the environment is the real-time changing device status, business requirements, and network conditions in the Visual Internet of Things. The agent interacts with the environment by continuously trying different strategies, such as transferring some devices in a certain area to other areas, merging or splitting existing areas, etc. The environment feedbacks a reward value according to the implementation effects of these strategies. The setting of the reward value is based on indicators such as the improvement of management efficiency, the load balancing of devices, and the shortening of business response time. For example, if the domain adjustment after adjustment shortens the device fault response time by 30% and improves the resource utilization rate by 20%, a higher reward will be given. The agent continuously learns and optimizes according to the reward value. When there are major changes such as a 50% surge in the number of devices in a certain area within a short period of time, the business type changes from ordinary monitoring to financial transaction monitoring, or abnormal situations such as serious network congestion and packet loss rate exceeding 10% occur, the model can quickly respond within a few minutes, select the optimal domain adjustment strategy, and reallocate device terminals to ensure that the domain division always remains efficient and reasonable and adapts to the dynamic changes of the Visual Internet of Things.

[0046] In this embodiment, the reward function R of the dynamic adjustment model based on deep reinforcement learning is defined as: ; In the formula, represents (adjusted management efficiency value - benchmark value) / benchmark value, represents (adjusted broadband utilization rate - pre-adjustment broadband utilization rate), Denote (original response time - new response time) / original response time, and and are adaptive adjustment coefficients. When detecting a DDoS attack, automatically increase the weight to 0.6, and increase it to 0.5 during the business peak period. Experimental data shows that this mechanism reduces the domain-adjusted response time from 4.2 s of the traditional solution to 0.8 s (based on the 100-node stress test).

[0047] Step S140, automatically adjust the fusion ratio of the visual feature and the temporal feature in the device operation trend feature according to the current network load condition.

[0048] Specifically, the domain management center automatically adjusts the fusion ratio of the visual feature and the temporal feature according to the current network load condition in combination with the predicted device operation trend feature in step S130.

[0049] Combined with Figure 2 As shown, in a specific embodiment, for the AI adaptive domain management method for a large number of device terminals in the vision network provided by the present invention, on the visual channel, a convolutional neural network (CNN) is used to extract device function features, such as camera resolution, intelligent analysis ability, etc. Through multi-layer convolution and pooling operations, key features of the device at the visual level are captured, providing a basis for judging the device function type. On the temporal channel, a Transformer model is used to analyze historical traffic patterns. By learning time series data, the traffic peak periods in the next 7 days are predicted, and the traffic change law of the device is accurately grasped. In addition, according to the current network load condition, the fusion ratio of the visual feature and the temporal feature is automatically adjusted. For example, when the network load is light, appropriately increase the weight of the visual feature and pay more attention to the device function characteristics; when the network load is heavy, reduce the weight of the visual feature and focus on traffic pattern analysis to improve the accuracy and adaptability of domain division.

[0050] Step S150, construct a relationship graph of each device terminal, and analyze the relationship graph through a graph clustering algorithm to divide devices with an association degree exceeding a first threshold into the same area, generating an initial domain division scheme.

[0051] Among them, the nodes in the relationship graph represent each device, and the edges represent the association weights between devices, which are determined by the communication frequency and geographical distance between devices.

[0052] Specifically, the domain management center constructs a relationship graph of each device and uses a graph clustering algorithm to analyze the relationship graph to divide devices with an association degree exceeding a first threshold into the same area, generating an initial domain division scheme.

[0053] Combined with Figure 2As shown, in a specific embodiment, the AI self-adaptive domain partitioning management method for a large number of devices in the Vision Internet of Things provided by the present invention constructs a device relationship graph, where nodes represent devices, and the weights of the edges are jointly determined by communication frequency and geographical distance to comprehensively reflect the physical connections and business association relationships between devices. Subsequently, a graph clustering algorithm is used to analyze the device relationship graph to generate an initial domain partitioning scheme, and devices with close connections are partitioned into the same area to optimize network communication and management efficiency.

[0054] In this embodiment, the real-time domain partitioning status and environmental parameters, such as sudden task requests, network congestion warnings, etc. are used as inputs to comprehensively reflect the current operating status of the Vision Internet of Things. According to the actual situation, some terminal devices are transferred across domains to balance the load and resource requirements of different regions, dynamically adjust the bandwidth quota, and prioritize ensuring network resources for critical services and high-demand regions. In case of emergencies such as DDoS attacks, emergency measures such as traffic cleaning are quickly enabled to ensure the safe and stable operation of the system. Through this hierarchical reward mechanism (basic reward = improvement in bandwidth utilization + reduction in response time; security reward = reduction in risk exposure surface × attack detection confidence; penalty term = overload penalty + QoS degradation penalty), the reinforcement learning model is incentivized to make better decisions in aspects such as improving resource utilization, ensuring security, and optimizing service quality. The effectiveness of the implemented strategy is evaluated. For example, after the implementation of the new strategy, the delay is reduced by 35%, quantifying the strategy effect and providing a reference for subsequent decisions. The model is updated online, with incremental training performed every hour. According to the latest operation data and strategy feedback, the reinforcement learning model is continuously optimized to improve the accuracy and timeliness of decision-making.

[0055] In this embodiment, the intelligent domain partitioning management center further includes a real-time monitoring dashboard for displaying the domain partitioning health status, visually presenting the operating status of each area through red / yellow / green three-color warnings, facilitating managers to quickly understand the overall situation of the system. Key metrics such as delay, packet loss rate, device load, etc. are tracked to monitor the system performance in real time and promptly discover potential problems. Domain partitioning adjustment suggestions are generated every 15 minutes. Based on real-time data and model analysis, continuous improvement directions are provided for system optimization. In case of sudden anomalies, emergency strategies are triggered within seconds to quickly respond to and handle emergencies, ensuring the stable operation of the Vision Internet of Things.

[0056] The above-mentioned AI adaptive domain management method for a large number of device terminals in the Visual Internet realizes intelligent and adaptive domain management of a large number of device terminals in the Visual Internet by using AI technology, and solves the problems of low management efficiency, poor security, poor flexibility, and difficulty in meeting personalized management requirements in the prior art. This method realizes multi-dimensional adaptive domain division, and adjusts the domain strategy in real time according to the dynamic changes of device terminals and business requirements; constructs an intelligent distributed area management center, improves the anti-attack ability and resource utilization efficiency of the system; designs a hierarchical AI collaborative management architecture, improves the collaborative efficiency of management and the maintainability of the system; optimizes the data interaction and sharing mechanism, and can ensure the high efficiency and security of data transmission to meet the compliance requirements of different industries.

[0057] As Figure 3 shown, in one embodiment, the AI adaptive domain management method for a large number of device terminals in the Visual Internet provided by the present invention, step S110 specifically includes the following steps: Step S111, calculate the quartiles and interquartile ranges of each feature column in the multi-source data by writing HiveQL statements, and set the outlier range based on the quartiles and interquartile ranges to detect outliers in the multi-source data.

[0058] Step S112, call the resilient distributed dataset or deduplication function of Spark to perform deduplication processing on the multi-source data.

[0059] As Figure 4 shown, in one embodiment, the AI adaptive domain management method for a large number of device terminals in the Visual Internet provided by the present invention, step S120 specifically includes the following steps: Step S121, calculate the mean and standard deviation of each numerical data in the multi-source data by constructing a TensorFlow computational graph, and call the Z-score normalization algorithm to perform normalization processing on each data point based on the mean and standard deviation, so that the data of the same type of performance indicators of different devices are at the same order of magnitude.

[0060] Step S122, use the normalization function of PyTorch to normalize the data with a fixed range of each device to a preset interval, and write PyTorch code to normalize the data with a fixed range.

[0061] Among them, the numerical data includes the CPU usage rate and memory occupancy rate of each device, and the data with a fixed range includes video quality scores and device signal strengths.

[0062] As Figure 5 shown, in one embodiment, the AI adaptive domain management method for a large number of device terminals in the Visual Internet provided by the present invention, step S130 specifically includes the following steps: Step S131: Extract the keyword information of each device from the text data by writing UDF functions and using regular expressions. The keyword information includes the brand, model, function, and device business type of the device.

[0063] Step S132: Use the recurrent neural network and long short-term memory network in the deep learning framework, and adopt the distributed training mechanism of Spark to learn the preprocessed multi-source data to create device operation trend features.

[0064] Among them, the device operation trend features include the change trend of device performance over time and the traffic fluctuation range.

[0065] As Figure 6 shown, in one embodiment, the AI self-adaptive domain division management method for a large number of device terminals in the visual networking provided by the present invention, step S130 specifically includes the following steps: Step S133: Process the image data of the video acquisition device through the multi-layer convolutional layer and pooling layer of the convolutional neural network to extract the key features for judging the device function type in the image data, including the object contour, color features, and texture information in the image data.

[0066] Step S134: Input the historical operation data into the recurrent neural network in chronological order, and use the recurrent structure inside the recurrent neural network to capture the dependency relationship between the historical operation data, so as to predict the traffic change trend in the first future time period through learning the historical operation data.

[0067] As Figure 7 shown, in one embodiment, after step S150 of the AI self-adaptive domain division management method for a large number of device terminals in the visual networking provided by the present invention, the following steps are included: Step S710: Based on the initial domain division scheme, combine the deep learning model fused with the convolutional neural network and the recurrent neural network to output the predicted traffic heat map in the first future time period to display the traffic change trend of different domains at different time points.

[0068] Step S720: Transfer the device terminals across domains according to the traffic change trend of different domains in combination with the actual operation state of the device, and dynamically adjust the broadband quota.

[0069] As Figure 8 shown, in one embodiment, the AI self-adaptive domain division management method for a large number of device terminals in the visual networking provided by the present invention further includes the following steps: Step S810: Evaluate the effectiveness of the initial domain division scheme, and when the device delay is reduced by no less than the second threshold after the implementation of the initial domain division scheme, quantify the implementation effect of the initial domain division scheme to obtain a policy feedback.

[0070] Step S820: Obtain the latest device operation data and policy feedback, and optimize and train the fused deep learning model based on the latest device operation data and policy feedback.

[0071] The following describes the Visual Internet of Things (VIoT) massive device terminal AI adaptive domain management system provided by the present invention. The VIoT massive device terminal AI adaptive domain management system described below can be correspondingly referred to the VIoT massive device terminal AI adaptive domain management method described above.

[0072] As Figure 9 shown, in one embodiment, a VIoT massive device terminal AI adaptive domain management system includes a data preprocessing module 910, a standardization and normalization module 920, a feature extraction module 930, a dynamic adjustment module 940, and a domain policy generation module 950.

[0073] The data preprocessing module 910 is used to obtain multi-source data from each device terminal and preprocess the multi-source data. The preprocessing includes outlier detection and distributed data deduplication.

[0074] The standardization and normalization module 920 is used to perform standardization processing on the preprocessed multi-source data through TensorFlow to obtain standardized data, and perform normalization processing on the standardized data by writing PyTorch code.

[0075] The feature extraction module 930 is used to extract key features from the text data of each device by using the function library and custom functions of Hive, and create device operation trend features based on the key features through Spark combined with a deep learning framework.

[0076] The dynamic adjustment module 940 is used to automatically adjust the fusion ratio of visual features and temporal features in the device operation trend features according to the current network load condition.

[0077] The domain policy generation module 950 is used to construct a relationship graph of each device terminal, and analyze the relationship graph through a graph clustering algorithm to divide devices with an association degree exceeding the first threshold into the same region, generating an initial domain division scheme.

[0078] Among them, the multi-source data includes geographical location information, functional type information, business requirement information, and historical operation data and real-time status data of each device. The historical operation data includes the CPU usage rate of the device per hour, the daily operation duration, and the monthly failure times. The real-time status data includes the current CPU usage rate, memory occupancy rate, and network connection status of the device; the nodes in the relationship graph represent each device, and the edges represent the association weights between devices, which are determined by the communication frequency and geographical distance between devices.

[0079] In this embodiment, for the AI adaptive domain management system for a vast number of device terminals in the vision network provided by the present invention, the data preprocessing module 910 is specifically configured to: Calculate the quartiles and interquartile ranges of each feature column in the multi-source data by writing HiveQL statements, and set the outlier range based on the quartiles and interquartile ranges to detect outliers in the multi-source data.

[0080] Call the resilient distributed dataset or the deduplication function of Spark to perform deduplication processing on the multi-source data.

[0081] In this embodiment, for the AI adaptive domain management system for a vast number of device terminals in the vision network provided by the present invention, the standardization and normalization module 920 is specifically configured to: Calculate the mean and standard deviation of each numerical data in the multi-source data by constructing a TensorFlow computational graph, and call the Z-score standardization algorithm to perform standardization processing on each data point based on the mean and standard deviation, so that the data of the same type of performance indicators of different devices are at the same order of magnitude.

[0082] Use the normalization function of PyTorch to normalize the data with a fixed range of each device to a preset interval, and write PyTorch code to normalize the data with a fixed range.

[0083] Among them, the numerical data includes the CPU usage rate and memory occupancy rate of each device, and the data with a fixed range includes the video quality score and the device signal strength.

[0084] In this embodiment, for the AI adaptive domain management system for a vast number of device terminals in the vision network provided by the present invention, the feature extraction module 930 is specifically configured to: Extract the keyword information of each device from the text data by writing UDF functions and using regular expressions. The keyword information includes the brand, model, function, and device service type of the device.

[0085] Use the recurrent neural network and long short-term memory network in the deep learning framework, and adopt the distributed training mechanism of Spark to learn the preprocessed multi-source data to create device operation trend features.

[0086] Among them, the device operation trend features include the change trend of device performance over time and the traffic fluctuation range.

[0087] In this embodiment, for the AI adaptive domain management system for a vast number of device terminals in the vision network provided by the present invention, the feature extraction module 930 is specifically further configured to: Process the image data of the video acquisition device through the multi-layer convolutional layer and pooling layer of the convolutional neural network to extract the key features in the image data for judging the device function type, including the object contour, color feature, and texture information in the image data.

[0088] Input the historical operation data into the recurrent neural network in chronological order, and use the recurrent structure inside the recurrent neural network to capture the dependency relationship between the historical operation data, so as to predict the traffic change trend in the first future time period through learning the historical operation data.

[0089] In this embodiment, the AI self-adaptive domain management system for massive video network devices provided by the present invention further includes a dynamic allocation module for: Based on the initial domain division scheme, combine the deep learning model fused by the convolutional neural network and the recurrent neural network to output the predicted traffic heat map in the first future time period, so as to display the traffic change trend of different domains at different time points.

[0090] Perform cross-domain transfer on the device terminals according to the traffic change trend of different domains in combination with the actual operation status of the devices, and dynamically adjust the broadband quota.

[0091] In this embodiment, the AI self-adaptive domain management system for massive video network devices provided by the present invention further includes a policy evaluation and model optimization module for: Evaluate the effectiveness of the initial domain division scheme, and when the device delay reduction after the implementation of the initial domain division scheme is not less than the second threshold, quantify the implementation effect of the initial domain division scheme to obtain policy feedback.

[0092] Obtain the latest device operation data and policy feedback, and optimize and train the fused deep learning model based on the latest device operation data and policy feedback.

[0093] Figure 10 Illustrates a schematic diagram of the physical structure of an electronic device. The electronic device can be a smart terminal, and its internal structure diagram can be as Figure 10 shown. The electronic device includes a processor, an internal memory, and a network interface connected through a system bus. Among them, the processor of the electronic device is used to provide computing and control capabilities. The memory of the electronic device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the electronic device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, it realizes the AI self-adaptive domain management method for massive video network devices, and the method includes: Obtain multi-source data from each device terminal and preprocess the multi-source data. The preprocessing includes outlier detection and distributed data deduplication; Perform standardization processing on the preprocessed multi-source data through TensorFlow to obtain standardized data, and perform normalization processing on the standardized data by writing PyTorch code; Use the function library and custom functions of Hive to extract key features from the text data of each device, and create device operation trend features based on the key features through Spark combined with a deep learning framework; Automatically adjust the fusion ratio of visual features and temporal features in the device operation trend features according to the current network load condition; Construct a relationship graph of each device terminal, and analyze the relationship graph through a graph clustering algorithm to divide devices with an association degree exceeding the first threshold into the same area, generating an initial domain division scheme; Among them, the multi-source data includes geographical location information, functional type information, business requirement information, and the historical operation data and real-time status data of each device. The historical operation data includes the CPU usage rate of the device per hour, the daily operation duration, and the number of faults per month. The real-time status data includes the current CPU usage rate, memory occupancy rate, and network connection status of the device; The nodes in the relationship graph represent each device, and the edges represent the association weights between devices, which are determined by the communication frequency and geographical distance between devices.

[0094] Those skilled in the art can understand that Figure 10 The structure shown in

[0095] is only a block diagram of some structures related to the solution of the present invention, and does not constitute a limitation on the electronic device to which the solution of the present invention is applied. The specific electronic device may include more or fewer components than those shown in the figure, or combine some components, or have different component arrangements. Obtain multi-source data from each device terminal and preprocess the multi-source data. The preprocessing includes outlier detection and distributed data deduplication; Perform standardization processing on the preprocessed multi-source data through TensorFlow to obtain standardized data, and perform normalization processing on the standardized data by writing PyTorch code; Use the function library and custom functions of Hive to extract key features from the text data of each device, and create device operation trend features based on the key features through Spark combined with a deep learning framework; Automatically adjust the fusion ratio of visual features and temporal features in the device operation trend features according to the current network load condition; Construct a relationship graph for each device terminal, and analyze the relationship graph through a graph clustering algorithm to divide devices with an association degree exceeding the first threshold into the same area, generating an initial domain division scheme; Among them, the multi-source data includes geographical location information, functional type information, business requirement information, as well as the historical operation data and real-time status data of each device. The historical operation data includes the CPU usage rate of the device per hour, the daily operation duration, and the number of faults per month. The real-time status data includes the current CPU usage rate, memory occupancy rate, and network connection status of the device; The nodes in the relationship graph represent each device, and the edges represent the association weights between devices, which are determined by the communication frequency and geographical distance between devices.

[0096] On the other hand, a computer program product or a computer program is provided. The computer program product or the computer program includes computer instructions, and the computer instructions are stored in a computer-readable storage medium. The processor of the electronic device reads the computer instructions from the computer-readable storage medium. When the processor executes the computer instructions, it implements the AI adaptive domain division management method for a large number of device terminals in the visual networking, and the method includes: Obtain multi-source data from each device terminal, and perform preprocessing on the multi-source data. The preprocessing includes outlier detection and distributed data deduplication; Perform standardization processing on the preprocessed multi-source data through TensorFlow to obtain standardized data, and perform normalization processing on the standardized data by writing PyTorch code; Use the function library and custom functions of Hive to extract key features from the text data of each device, and create device operation trend features based on the key features through Spark combined with a deep learning framework; Automatically adjust the fusion ratio of visual features and temporal features in the device operation trend features according to the current network load condition; Construct a relationship graph for each device terminal, and analyze the relationship graph through a graph clustering algorithm to divide devices with an association degree exceeding the first threshold into the same area, generating an initial domain division scheme; Among them, the multi-source data includes geographical location information, functional type information, business requirement information, as well as the historical operation data and real-time status data of each device. The historical operation data includes the CPU usage rate of the device per hour, the daily operation duration, and the number of faults per month. The real-time status data includes the current CPU usage rate, memory occupancy rate, and network connection status of the device; The nodes in the relationship graph represent each device, and the edges represent the association weights between devices, which are determined by the communication frequency and geographical distance between devices.

[0097] Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above methods. Among them, any reference to a memory, storage, database, or other medium used in the embodiments provided by the present invention can include non-volatile and / or volatile memories. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory.

[0098] By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and Rambus dynamic RAM (RDRAM), etc.

[0099] The technical features of the above embodiments can be combined arbitrarily. For the sake of brevity of description, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered as the scope described in this specification.

[0100] The above-described embodiments merely represent several implementation manners of the present invention. Their descriptions are relatively specific and detailed, but they should not be construed as limiting the scope of the patent of the present invention. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present invention, several modifications and improvements can still be made, and these all belong to the protection scope of the present invention. Therefore, the protection scope of the patent of the present invention should be subject to the appended claims.

Claims

1. An AI adaptive domain management method for a large number of device terminals in a visual networking, characterized in that, The method includes: Obtaining multi-source data from each device terminal and preprocessing the multi-source data, where the preprocessing includes outlier detection and distributed data deduplication; Performing standardization processing on the preprocessed multi-source data through TensorFlow to obtain standardized data, and performing normalization processing on the standardized data by writing PyTorch code; Using the function library and custom functions of Hive to extract key features from the text data of each device, and creating device operation trend features based on the key features through Spark combined with a deep learning framework; Automatically adjusting the fusion ratio of visual features and temporal features in the device operation trend features according to the current network load condition; Constructing a relationship graph of each device terminal, and analyzing the relationship graph through a graph clustering algorithm to divide devices with an association degree exceeding a first threshold into the same area, generating an initial domain division scheme; Among them, the multi-source data includes geographical location information, functional type information, business requirement information, and historical operation data and real-time status data of each device. The historical operation data includes the CPU usage rate of the device per hour, the daily operation duration, and the number of faults per month. The real-time status data includes the current CPU usage rate, memory occupancy rate, and network connection status of the device; the nodes in the relationship graph represent each device, and the edges represent the association weights between devices, which are determined by the communication frequency and geographical distance between devices.

2. The AI adaptive domain division management method for a large number of device terminals in the vision Internet according to claim 1, wherein, The obtaining of multi-source data from each device terminal and the preprocessing of the multi-source data include: Calculating the quartiles and interquartile ranges of each feature column in the multi-source data by writing HiveQL statements, and setting an outlier range based on the quartiles and interquartile ranges to perform outlier detection on the multi-source data; Invoking the elastic distributed dataset or deduplication function of Spark to perform deduplication processing on the multi-source data.

3. The AI adaptive domain division management method for a large number of device terminals in the vision network according to claim 1, wherein, The performing of standardization processing on the preprocessed multi-source data through TensorFlow to obtain standardized data, and the performing of normalization processing on the standardized data by writing PyTorch code includes: Calculating the mean and standard deviation of each numerical data in the multi-source data by constructing a TensorFlow computation graph, and invoking the Z-score standardization algorithm to perform standardization processing on each data point based on the mean and standard deviation, so that the data of the same type of performance indicators of different devices are at the same order of magnitude; Using the normalization function of PyTorch to normalize the data with a fixed range of each device to a preset interval, and performing normalization on the data with a fixed range by writing PyTorch code; Among them, the numerical data includes the CPU usage rate and memory occupancy rate of each device, and the data with a fixed range includes video quality scores and device signal strengths.

4. The AI adaptive domain division management method for a large number of device terminals in the vision Internet according to claim 1, wherein, The using of the function library and custom functions of Hive to extract key features from the text data of each device, and the creating of device operation trend features based on the key features through Spark combined with a deep learning framework includes: By writing UDF functions and using regular expressions to extract keyword information of each device from the text data, the keyword information includes the brand, model, function, and device business type of the device; Using the recurrent neural network and long short-term memory network in the deep learning framework, and adopting the distributed training mechanism of Spark to learn the preprocessed multi-source data to create the device operation trend characteristics; Among them, the device operation trend characteristics include the change trend of device performance over time and the traffic fluctuation range.

5. The method for AI adaptive domain division management of a large number of device terminals in a visual networking according to claim 4, wherein The extracting of key features from the text data of each device by using the function library and custom functions of Hive, and creating device operation trend characteristics based on the key features through Spark combined with the deep learning framework also includes: Processing the image data of the video acquisition device through the multi-layer convolutional layer and pooling layer of the convolutional neural network to extract the key features for judging the device function type in the image data, including the object contour, color feature, and texture information in the image data; Inputting the historical operation data into the recurrent neural network in chronological order, and using the recurrent structure inside the recurrent neural network to capture the dependency relationship between historical operation data, so as to predict the traffic change trend in the first future time period through learning the historical operation data.

6. The AI adaptive domain division management method for a large number of visual network devices terminals according to claim 5, wherein, Building a relationship graph of each device terminal, and analyzing the relationship graph through a graph clustering algorithm to divide devices with an association degree exceeding the first threshold into the same area to generate an initial domain division scheme, and then including: Based on the initial domain division scheme, combining the deep learning model fused by the convolutional neural network and the recurrent neural network to output a predicted traffic heat map in the first future time period to show the traffic change trend of different domains at different time points; According to the traffic change trend of different domains and the actual operation status of the device, perform cross-domain transfer on the device terminal and dynamically adjust the broadband quota.

7. The AI adaptive domain division management method for a large number of visual network devices terminals according to claim 6, wherein The method also includes: Evaluating the effectiveness of the initial domain division scheme, and when the device delay is reduced by no less than the second threshold after the implementation of the initial domain division scheme, quantifying the implementation effect of the initial domain division scheme to obtain a policy feedback; Obtaining the latest device operation data and policy feedback, and optimizing and training the fused deep learning model based on the latest device operation data and policy feedback.

8. An AI adaptive domain division management system for a large number of device terminals in the visual networking, characterized in that, The system includes: A data preprocessing module, used to obtain multi-source data from each device terminal and preprocess the multi-source data, and the preprocessing includes outlier detection and distributed data deduplication; A standardization and normalization module, used to perform standardization processing on the preprocessed multi-source data through TensorFlow to obtain standardized data, and perform normalization processing on the standardized data by writing PyTorch code; A feature extraction module, used to extract key features from the text data of each device by using the function library and custom functions of Hive, and create device operation trend characteristics based on the key features through Spark combined with the deep learning framework; A dynamic adjustment module, configured to automatically adjust the fusion ratio of the visual feature and the timing feature in the device operation trend feature according to the current network load condition; A domain division policy generation module, configured to construct a relationship graph of each device terminal, and analyze the relationship graph through a graph clustering algorithm, so as to divide devices with an association degree exceeding a first threshold into the same area and generate an initial domain division scheme; Wherein, the multi-source data includes geographical location information, function type information, service requirement information, and historical operation data and real-time status data of each device. The historical operation data includes the CPU usage rate of the device per hour, the daily operation duration, and the monthly failure times. The real-time status data includes the current CPU usage rate, memory occupancy rate, and network connection status of the device; the nodes in the relationship graph represent each device, and the edges represent the association weights between devices, which are determined by the communication frequency and geographical distance between devices.

9. An electronic device, comprising a memory and a processor, the memory storing a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the AI adaptive domain division management method for visual network massive device terminals according to any one of claims 1 to 7.

10. A computer storage medium stores a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of the AI adaptive domain division management method for visual network massive device terminals according to any one of claims 1 to 7.

Images