A method and system for AI function expansion based on deep learning technology

Abstract

The present invention relates to the field of artificial intelligence technology, and in particular to an AI function expansion method and system based on deep learning technology, the method comprising: obtaining node information of multiple network nodes in the same network, extracting corresponding node function features from the node information, and constructing a node function table; performing cluster analysis on the functions of all artificial intelligence terminals in the network nodes, and obtaining the node functions of the network nodes; constructing a function vector matrix matching the node function table, and obtaining AI model data matching the node function features through the function vector matrix; importing a preset network node connection strategy, initializing the network node connection, and adjusting and updating the network node connection strategy through a preset deep learning model, and predicting the operating status of each network node; and expanding the AI ​​function of one or more network nodes according to the connection status of the network nodes.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to an AI function expansion method and system based on deep learning technology. Background Art

[0002] Artificial intelligence (AI) technology is a technology that simulates human intelligence through computer programs. It covers multiple fields such as machine learning, deep learning, and natural language processing. With the advancement of computer technology and algorithms, especially the development of machine learning and deep learning, artificial intelligence has begun to usher in explosive growth. Among them, machine learning allows computers to learn from data and continuously optimize their own performance; while deep learning based on neural network architecture gives machines more advanced abstract thinking capabilities such as image recognition and speech understanding.

[0003] Typically, artificial intelligence devices rely on built-in AI models to complete specific tasks such as pattern recognition, content creation, and data analysis. These models are often fully trained on large-scale data sets and then transplanted to each independent smart device to perform the corresponding operations. When the computing resources of a device are close to saturation, its work efficiency may decline. Although this problem can be alleviated by upgrading to more powerful hardware, it is often accompanied by higher economic investment.

[0004] To this end, we propose a technical solution that can expand the AI ​​function of a single terminal and improve its processing power without increasing hardware costs. Summary of the invention

[0005] The present invention improves the data communication efficiency between different network nodes by adjusting the connection mode of each network node in the network, effectively utilizes the functions of the artificial intelligence terminals in other network nodes, and avoids the situation where the processing efficiency of a single network node decreases.

[0006] The technical solution proposed by the present invention is: an AI function expansion method based on deep learning technology, the method comprising:

[0007] Obtain node information of multiple network nodes in the same network, extract corresponding node functional characteristics from the node information, and construct a node function table; the functional characteristics of the node include the number of artificial intelligence terminals in the node, the function of each artificial intelligence terminal, and the maximum computing power of each artificial intelligence terminal;

[0008] Perform cluster analysis on the functions of all artificial intelligence terminals in the network nodes to obtain the node functions of the network nodes;

[0009] Constructing a function vector matrix matching the node function table, and obtaining AI model data matching the node function characteristics through the function vector matrix;

[0010] Import the preset network node connection strategy, initialize the network node connection, and adjust and update the network node connection strategy through the preset deep learning model to predict the operating status of each network node;

[0011] Expand the AI ​​function of one or more network nodes based on the connection status of the network nodes;

[0012] The node function table includes a function storage area and an extension area matching each network node, and the extension area is used to store extended function characteristics; the network node operation status includes normal load and abnormal load.

[0013] Preferably, extracting corresponding node function features from the node information and constructing a node function table comprises the following steps:

[0014] Acquire the functions of an artificial intelligence terminal of each network node, where the functions of the artificial intelligence terminal include one or more of human-computer interaction, speech recognition, content generation, and data analysis;

[0015] Digitize the functions of each AI terminal and generate a function identification code that uniquely corresponds to the AI ​​terminal function;

[0016] Add node identification tags to each network node;

[0017] A node function table is formed by using the node identification tag of each network node and a plurality of function identification codes of the corresponding network node.

[0018] Preferably, the constructing of a function vector matrix matching the node function table and obtaining AI model data matching the node function features through the function vector matrix comprises the following steps:

[0019] Generate a function identification matrix based on the node function table ;in, Indicates The network node The function identification code of each function; ; , ;

[0020] Obtaining an AI model stored in each artificial intelligence terminal, one or more of the AI ​​models matching a function of the artificial intelligence terminal;

[0021] Obtain the storage address of each AI model in the corresponding AI terminal, add the identification tag of the network node where the AI ​​terminal is located and the function identification tag of the AI ​​terminal, and form an AI model address vector;

[0022] Then the function vector matrix ;in, Indicates The network node has The address vector of the AI ​​model of each node function.

[0023] Preferably, the importing of a preset network node connection strategy to initialize the network node connection includes:

[0024] Establishing data connection between nodes through a preset network node connection strategy, wherein the preset network node connection strategy includes a serial connection mode and a bus connection mode;

[0025] Add a connection identification label 1 for the serial connection mode, and add a connection identification label 2 for the bus connection mode;

[0026] The serial connection method is that multiple network nodes are connected in serial mode; the artificial intelligence terminal of each node sends a data request to the intelligent terminal of the next node and obtains feedback data;

[0027] The bus connection mode is that multiple network nodes are connected via a bus, and data communication is performed between each network node via a bus protocol.

[0028] Preferably, it also includes: collecting the CPU usage and GPU usage of each node's artificial intelligence terminal according to a preset collection frequency to form a state vector of each intelligent terminal ;in, Indicates The network node The state vector of an AI terminal, Respectively Moment The network node CPU usage and GPU usage of each AI terminal;

[0029] The state vectors of multiple artificial intelligence terminals with the same functions form a functional feature dataset ,in, express The number of AI terminals with different functions, Indicates The network nodes have The state vector of the functional artificial intelligence terminal, Indicates the function identification code;

[0030] From the node function table, obtain the function information of each node, and determine the operation status of the network node according to the state vector of each intelligent terminal;

[0031] According to the operating status and combined with the existing network node connection strategy, adjust the connection method between each node.

[0032] Preferably, the step of acquiring the function information of each node from the node function table and judging the operation status of the network node according to the state vector of each intelligent terminal comprises the following steps:

[0033] Obtain the CPU usage and GPU usage of multiple artificial intelligence terminals with the same function identification code in the corresponding node from the node function table;

[0034] Calculate the load corresponding to the function identification code ;

[0035] ,in Represents CPU weight and GPU weight respectively;

[0036] if , judge the load abnormality of the corresponding node function in the network node, and output the load abnormality alarm information.

[0037] Preferably, the adjusting and updating of the network node connection strategy through a preset deep learning model includes:

[0038] Obtain connection adjustment logs, CPU usage and GPU usage of the artificial intelligence terminals of network nodes, and load anomaly judgment results to build a historical sample set;

[0039] Extract multiple key features from the historical sample set and normalize them to form a key feature set;

[0040] The key features include the maximum CPU usage rate of the network node, the maximum GPU usage rate, the judgment result of load abnormality, the connection mode of the network node before adjustment, and the connection mode of the network node after adjustment;

[0041] Input the elements in the key feature set into the deep learning model to predict the probability of adjusting the network node connection strategy ;

[0042] if , then it is determined that the corresponding network node needs to adjust the connection mode at a future moment, where represents the adjustment probability threshold, ;

[0043] Get the current network connection strategy and adjust the network connection mode according to the prediction results of the deep learning model, including the following steps:

[0044] Obtain the function identification code of the node function with abnormal load;

[0045] Obtain the current network node connection strategy and load abnormality alarm information of the network. If it is a series connection mode, adjust the network node connection strategy to establish a data connection between the network node with abnormal load and the network node that is not directly connected in the network; if it is a bus connection mode, do not adjust the network node connection strategy;

[0046] The adjusted network connection strategy is sent to the host computer to update the current network node connection strategy in the host computer.

[0047] Preferably, the step of expanding the AI ​​function of one or more network nodes comprises the following steps:

[0048] Determine whether there is an expandable node in the network; if there is an expandable network node, call the AI ​​model in the expandable network node; the expandable node is a network node in the network that is connected to the network node with abnormal load and has normal load;

[0049] The AI ​​model in the scalable network node is used to undertake the calculation of the network node in the next moment when the network node is in an abnormal load state, so as to ensure the node function of the network node with abnormal load;

[0050] If there is no scalable network node or the scalable network node does not have an AI model with a corresponding node function;

[0051] Then, from the network node with abnormal load, obtain the AI ​​model that performs the corresponding node function, and copy the obtained AI model to the expansion area of ​​any network node;

[0052] In the extended area, using the knowledge distillation technology, the acquired AI model is used as the teacher model to train a student model with the same functions as the teacher model as the extended AI model;

[0053] By expanding the AI ​​model, the computing task of the network node in the abnormal load state at the next moment is undertaken to ensure the node function of the corresponding network node.

[0054] The present invention also provides an AI function expansion system based on deep learning technology, which is used to execute the AI ​​function expansion method based on deep learning technology.

[0055] The present invention also provides a computer-readable storage medium storing a computer program, wherein the computer program is executed by a processor to implement the AI ​​function expansion method based on deep learning technology.

[0056] Beneficial effects of the present invention:

[0057] 1. The present invention, by identifying the operating status of each network node and adjusting the connection mode of each network node according to the existing network node connection strategy. In this way, each network node can more conveniently obtain the required data from other network nodes, such as the returned calculation results and AI models with corresponding functions. In addition, by utilizing the computing power of other network nodes, the computing pressure of network nodes with abnormal load conditions can be alleviated, thereby indirectly improving the processing capacity of the network node. By utilizing the AI ​​models obtained from other network nodes, the node function of the network node is expanded so that it can handle tasks that were originally unable to be handled, thereby expanding the AI ​​function of the network node.

[0058] 2. The present invention uses a deep learning model to learn the relationship between the operating status of network nodes and the connection mode that needs to be adjusted, and predicts when the connection mode needs to be adjusted. If it is determined that the connection mode needs to be adjusted, the pre-stored network node connection strategy is used to adjust the connection mode between network nodes. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] Figure 1 This is a flow chart of an AI function expansion method based on deep learning technology of the present invention;

[0060] Figure 2 Schematic diagram of network node connection strategy in an embodiment of the present invention Figure 1 ;

[0061] Figure 3 Schematic diagram of network node connection strategy in an embodiment of the present invention Figure 2 . DETAILED DESCRIPTION

[0062] The following description is used to disclose the present invention so that those skilled in the art can implement the present invention. The preferred embodiments described below are only examples, and those skilled in the art can think of other obvious variations. The basic principles of the present invention defined in the following description can be applied to other embodiments, variations, improvements, equivalents, and other technical solutions that do not deviate from the spirit and scope of the present invention.

[0063] It is to be understood that the term "one" should be understood as "at least one" or "one or more", that is, in one embodiment, the number of an element may be one, while in another embodiment, the number of the element may be multiple, and the term "one" should not be understood as a limitation on the quantity.

[0064] refer to Figure 1 The technical solution provided by the present invention is: an AI function expansion method based on deep learning technology, comprising the following steps:

[0065] Step 1: Obtain node information of multiple network nodes in the same network, extract corresponding node function characteristics from the node information, and construct a node function table; the function characteristics of the node include the number of artificial intelligence terminals in the node, the function of each artificial intelligence terminal and the maximum computing power of each artificial intelligence terminal; the node function table includes a function storage area and an extension area matching each network node, and the extension area is used to store extended function characteristics.

[0066] The method for constructing a node function table comprises the following steps:

[0067] Acquire the functions of an artificial intelligence terminal of each network node, where the functions of the artificial intelligence terminal include one or more of human-computer interaction, speech recognition, content generation, and data analysis;

[0068] Digitize the functions of each AI terminal and generate a function identification code that uniquely corresponds to the AI ​​terminal function;

[0069] A node identification tag is added to each network node, and a node function table is formed using the node identification tag of each network node and a plurality of function identification codes of the corresponding network node.

[0070] For example, if there are multiple network nodes in the network, the node functions of the first network node include speech recognition (identification code is I), content generation (identification code is II) and data processing (identification code is III), the node functions of the second network node include content generation and data processing, the node functions of the third network node include data processing, and the node functions of the fourth network node include speech recognition, then the node function table is shown in Table 1:

[0071] Table 1

[0072]

[0073] Step 2: Perform cluster analysis on the functions of all artificial intelligence terminals in the network nodes to obtain the node functions of the network nodes; as shown in the above table, it can be seen that the first network node includes at least 3 artificial intelligence terminals, and each artificial intelligence terminal can perform at least 1 function (speech recognition, content generation and data processing). Finally, classify the functions of all artificial intelligence terminals to obtain the classification of the node functions of the network.

[0074] Step 3: construct a function vector matrix matching the node function table, and obtain AI model data matching the node function characteristics through the function vector matrix; including the following steps:

[0075] Obtaining AI model data matching the node function features through the function vector matrix includes the following steps:

[0076] Generate a function identification matrix based on the node function table ;in, Indicates The network node The function identification code of each function; ; , ;

[0077] Obtaining AI models stored in each AI terminal, where the AI ​​terminal performs its functions through one or more AI models;

[0078] If the network node does not have the corresponding node function, the value of the corresponding element in the function identification matrix is ​​zero. For example, if the second network node does not have the speech recognition function, then is zero.

[0079] In this embodiment, the function identification matrix corresponding to Table 1 is: ;

[0080] Since the second network node has two functions identical to those of the first network node, , ; Since the third network node has the same function as the first network node, that is ; Since the fourth network node has the same function as the first network node, that is ;

[0081] The function identification matrix is ​​finally .

[0082] Obtain the storage address of each AI model in the corresponding AI terminal, add the identification tag of the network node where the AI ​​terminal is located and the function identification tag of the AI ​​terminal, and form an AI model address vector;

[0083] Then the function vector matrix ;in, Indicates The network node has The address vector of the AI ​​model of each node function.

[0084] In this embodiment , , and the function vector matrix corresponding to Table 1 can be expressed as .

[0085] After obtaining the AI ​​model for realizing the speech recognition function from the first network node and obtaining the extended AI model that can realize the speech recognition function in the extended area of ​​the second network node through the knowledge distillation technology, a storage address is allocated for the extended AI model (the extended area includes multiple storage address segments, and the corresponding address can be allocated to the AI ​​model), and the artificial intelligence terminal with the lowest CPU utilization is selected, and its identification label and the corresponding function identification label are combined with the address of the allocated extended AI model to form the extended AI model address, and stored in the corresponding position of the function vector matrix.

[0086] In this embodiment, after expansion, ,in A function identification code indicating a speech recognition function; A function identifier representing the content generation function, The function identification code representing the data processing function is expanded to form a new node function table:

[0087] Table 2

[0088]

[0089] Step 4: Import the preset network node connection strategy, initialize the network node connection, and adjust and update the network node connection strategy through the preset deep learning model to predict the operating status of each network node; the network node operating status includes normal load and abnormal load.

[0090] The steps for initial network node connection are as follows:

[0091] Establishing data connection between nodes through a preset network node connection strategy, wherein the preset network node connection strategy includes a serial connection mode and a bus connection mode;

[0092] Add a connection identification label 1 for the serial connection mode, and add a connection identification label 2 for the bus connection mode;

[0093] The serial connection method is that multiple network nodes are connected in serial mode; the artificial intelligence terminal of each node sends a data request to the intelligent terminal of the next node and obtains feedback data;

[0094] The bus connection mode is that multiple network nodes are connected via a bus, and data communication is performed between each network node via a bus protocol.

[0095] Adjusting and updating the network node connection strategy through the preset deep learning model includes the following steps:

[0096] First, get the running status of the network node, specifically:

[0097] The CPU usage and GPU usage of each node’s AI terminal are collected at the preset collection frequency to form the state vector of each intelligent terminal. ;in, Indicates The network node The state vector of an AI terminal, Respectively Moment The network node CPU usage and GPU usage of each AI terminal;

[0098] The state vectors of multiple artificial intelligence terminals with the same functions form a functional feature dataset ,in, express The number of AI terminals with different functions, Indicates The network nodes have The state vector of the functional artificial intelligence terminal, Indicates the function identification code;

[0099] Then, the function information of each node is obtained from the node function table, and the operation status of the network node is determined according to the state vector of each intelligent terminal, specifically:

[0100] Obtain the CPU usage and GPU usage of multiple artificial intelligence terminals with the same function identification code in the corresponding node from the node function table;

[0101] Calculate the load corresponding to the function identification code ;

[0102] ,in Represents CPU weight and GPU weight respectively;

[0103] if , judge the load abnormality of the corresponding node function in the network node, and output the load abnormality alarm information.

[0104] Finally, according to the running status and the existing network node connection strategy, adjust the connection mode between each node, specifically:

[0105] Obtain connection adjustment logs, CPU usage and GPU usage of the artificial intelligence terminals of network nodes, and load anomaly judgment results to build a historical sample set;

[0106] Extract multiple key features from the historical sample set and normalize them to form a key feature set;

[0107] The key features include the maximum CPU usage rate of the network node, the maximum GPU usage rate, the judgment result of load abnormality, the connection mode of the network node before adjustment, and the connection mode of the network node after adjustment;

[0108] Input the elements in the key feature set into the deep learning model to predict the probability of adjusting the network node connection strategy ;

[0109] if , then it is determined that the corresponding network node needs to adjust the connection mode at a future moment, where represents the adjustment probability threshold, .

[0110] The deep learning model in this embodiment may adopt a random forest model or a linear regression model.

[0111] Obtain the function identification code of the node function with abnormal load; obtain the current network node connection strategy and load abnormality alarm information of the network. If it is a series connection method, adjust the network node connection strategy to establish a data connection between the network node with abnormal load and the network nodes that are not directly connected in the network; if it is a bus connection method, do not adjust the network node connection strategy.

[0112] The serial connection method is suitable for scenarios where the task flow has a clear sequence, such as speech recognition, data processing, and content generation. Each network node processes one or two task flows in turn, making full use of AI terminal resources and reducing system costs.

[0113] The bus connection method is suitable for scenarios where multiple computing tasks are performed in parallel. All nodes share a communication bus, which is convenient for quickly adding or reducing devices, but attention should be paid to signal interference and security issues. It is suitable for small network construction.

[0114] Of course, there are other connection methods, such as the star connection method, which connects each sub-node through a central node (such as a switch), which is suitable for networks that require high reliability and easy management. Each node operates independently, and fault isolation is convenient, but a central node failure will cause the entire network to fail.

[0115] In the ring connection mode, each node is connected to two adjacent nodes to form a closed loop. It is suitable for high-speed transmission media such as optical fiber. Data transmission is unidirectional with low bit error rate, but any node failure will affect the entire network.

[0116] The hybrid connection mode combines the advantages of multiple topologies, such as the star-ring hybrid topology. It is suitable for complex network environments, provides high flexibility and scalability, and has high performance and reliability. It is a common choice for large networks.

[0117] In this embodiment, a series connection mode and a bus connection mode are taken as examples.

[0118] If a function requires too much computation and exceeds the processing capacity of the AI ​​model, it is necessary to connect to other network nodes and use the AI ​​models of other network nodes for processing.

[0119] like Figure 2 At this time, if the network nodes are connected in series, the communication distance between the first network node and the third network node is very long, and the second network node is required for transfer, which reduces the data processing efficiency. Therefore, it is necessary to adjust the connection strategy.

[0120] like Figure 3 In this embodiment, the connection strategy is changed to enable each network node to be directly connected to other network nodes without the need for transfer, thereby increasing the speed of retrieving AI models from other network nodes.

[0121] If the network adopts a bus connection mode, each network node on the bus can obtain data from other network nodes through the bus protocol. In this embodiment, the connection strategy adopted is not to adjust the connection mode.

[0122] The adjusted network connection strategy is sent to the host computer to update the current network node connection strategy in the host computer.

[0123] Step 5: Expand the AI ​​function of one or more network nodes according to the connection status of the network nodes, including the following steps:

[0124] Determine whether there is an expandable node in the network; if there is an expandable network node, call the AI ​​model in the expandable network node; the expandable node is a network node in the network that is connected to the network node with abnormal load and has normal load;

[0125] The AI ​​model in the scalable network node is used to undertake the calculation of the network node in the next moment when the network node is in an abnormal load state, so as to ensure the node function of the network node with abnormal load;

[0126] If there is no scalable network node or the scalable network node does not have an AI model with a corresponding node function;

[0127] Then, from the network node with abnormal load, obtain the AI ​​model that performs the corresponding node function, and copy the obtained AI model to the expansion area of ​​any network node;

[0128] In the extended area, using the knowledge distillation technology, the acquired AI model is used as the teacher model to train a student model with the same functions as the teacher model as the extended AI model;

[0129] By expanding the AI ​​model, the computing task of the network node in the abnormal load state at the next moment is undertaken to ensure the node function of the corresponding network node.

[0130] The specific process of the so-called knowledge distillation technology includes training the teacher model, extracting soft labels, training the student model, and evaluating and optimizing. Knowledge distillation uses a large teacher model to teach a smaller student model, so that the student model can learn similar functions but more efficiently. In this embodiment, the model training process is not described in detail.

[0131] The present invention also provides an AI function expansion system based on deep learning technology, which is used to execute the AI ​​function expansion method based on deep learning technology.

[0132] The present invention also provides a computer-readable storage medium, which stores a computer program, and the computer program is executed by a processor to implement the AI ​​function expansion method based on deep learning technology.

[0133] In the embodiments disclosed in the present invention, the process described above with reference to the flowchart can be implemented as a computer software program. The embodiments disclosed in the present invention include a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program contains a program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from the network through the communication part, and / or installed from a removable medium. When the computer program is executed by the central processing unit (CPU), the above functions defined in the method of the present application are executed. It should be noted that the computer-readable medium mentioned above in the present application can be a computer-readable signal medium or a computer-readable storage medium or any combination of the above two. The computer-readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared or semiconductor system, device or device, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to, an electrical connection with one or more wire segments, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present application, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In the present application, a computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, which carries a computer-readable program code. Such propagated data signals may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. A computer-readable signal medium may also be any computer-readable medium other than a computer-readable storage medium, which may send, propagate, or transmit a program for use by or in conjunction with an instruction execution system, apparatus, or device. The program code embodied on the computer readable medium may be transmitted using any appropriate medium, including but not limited to wireless, electrical wire, optical cable, RF, etc., or any suitable combination of the foregoing.

[0134] The flowcharts and block diagrams in the accompanying drawings illustrate the possible architecture, functions and operations of the systems, methods and computer program products according to various embodiments of the present invention. In this regard, each box in the flowchart or block diagram can represent a module, a program segment or a part of a code, and the module, a program segment or a part of a code contains one or more executable instructions for implementing the specified logical functions. It should also be noted that in some alternative implementations, the functions marked in the boxes can also occur in a different order than the order marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flowchart, and the combination of boxes in the block diagram and / or flowchart, can be implemented with a dedicated hardware-based system that performs the specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0135] Those skilled in the art should understand that the embodiments of the present invention described above and shown in the accompanying drawings are only examples and do not limit the present invention. The objectives of the present invention have been fully and effectively achieved. The functions and structural principles of the present invention have been demonstrated and explained in the embodiments. Without departing from the principles, the implementation methods of the present invention may be subject to any changes or modifications.

Claims

1. A method for expanding AI functions based on deep learning technology, characterized in that: The method comprises: Obtain node information of multiple network nodes in the same network, extract corresponding node functional characteristics from the node information, and construct a node function table; the functional characteristics of the node include the number of artificial intelligence terminals in the node, the function of each artificial intelligence terminal, and the maximum computing power of each artificial intelligence terminal; Perform cluster analysis on the functions of all artificial intelligence terminals in the network nodes to obtain the node functions of the network nodes; Constructing a function vector matrix matching the node function table, and obtaining AI model data matching the node function characteristics through the function vector matrix; Import the preset network node connection strategy, initialize the network node connection, and adjust and update the network node connection strategy through the preset deep learning model to predict the operating status of each network node; Expand the AI ​​function of one or more network nodes based on the connection status of the network nodes; The node function table includes a function storage area and an extension area matching each network node, wherein the extension area is used to store extended function characteristics; the network node operation status includes normal load and abnormal load; The step of extracting corresponding node function features from the node information and constructing a node function table includes the following steps: Acquire the functions of an artificial intelligence terminal of each network node, where the functions of the artificial intelligence terminal include one or more of human-computer interaction, speech recognition, content generation, and data analysis; Digitize the functions of each AI terminal and generate a function identification code that uniquely corresponds to the AI ​​terminal function; Add node identification tags to each network node; Using the node identification tag of each network node and a plurality of function identification codes of the corresponding network node, a node function table is formed; The step of constructing a function vector matrix matching the node function table and obtaining AI model data matching the node function features through the function vector matrix includes the following steps: Generate a function identification matrix based on the node function table ;in, Indicates The network node The function identification code of each function; ; , ; Obtaining an AI model stored in each artificial intelligence terminal, one or more of the AI ​​models matching a function of the artificial intelligence terminal; Obtain the storage address of each AI model in the corresponding AI terminal, add the identification tag of the network node where the AI ​​terminal is located and the function identification tag of the AI ​​terminal, and form an AI model address vector; Then the function vector matrix ;in, Indicates The network node has The address vector of the AI ​​model of each node function.

2. According to claim 1, the AI ​​function expansion method based on deep learning technology is characterized in that: The step of importing a preset network node connection strategy and initializing the network node connection includes: Establishing data connection between nodes through a preset network node connection strategy, wherein the preset network node connection strategy includes a serial connection mode and a bus connection mode; Add a connection identification label 1 for the serial connection mode, and add a connection identification label 2 for the bus connection mode; The serial connection method is that multiple network nodes are connected in serial mode; the artificial intelligence terminal of each node sends a data request to the intelligent terminal of the next node and obtains feedback data; The bus connection mode is that multiple network nodes are connected via a bus, and data communication is performed between each network node via a bus protocol.

3. The AI ​​function expansion method based on deep learning technology according to claim 2, characterized in that: Also includes: The CPU usage and GPU usage of each node’s AI terminal are collected at the preset collection frequency to form the state vector of each intelligent terminal. ;in, Indicates The network node The state vector of an AI terminal, Respectively Moment The network node CPU usage and GPU usage of each AI terminal; The state vectors of multiple artificial intelligence terminals with the same functions form a functional feature dataset ,in, express The number of AI terminals with different functions, Indicates The network nodes have The state vector of the functional artificial intelligence terminal, Indicates the function identification code; From the node function table, obtain the function information of each node, and determine the operation status of the network node according to the state vector of each intelligent terminal; According to the operating status and combined with the existing network node connection strategy, adjust the connection method between each node.

4. The AI ​​function expansion method based on deep learning technology according to claim 3 is characterized in that: The function information of each node is obtained from the node function table, and the operation status of the network node is determined according to the state vector of each intelligent terminal, including the following steps: Obtain the CPU usage and GPU usage of multiple artificial intelligence terminals with the same function identification code in the corresponding node from the node function table; Calculate the load corresponding to the function identification code ; ,in Represents CPU weight and GPU weight respectively; if , judge the load abnormality of the corresponding node function in the network node, and output the load abnormality alarm information.

5. The AI ​​function expansion method based on deep learning technology according to claim 4 is characterized in that: The adjusting and updating of the network node connection strategy through the preset deep learning model includes: Obtain connection adjustment logs, CPU usage and GPU usage of the artificial intelligence terminals of network nodes, and load anomaly judgment results to build a historical sample set; Extract multiple key features from the historical sample set and normalize them to form a key feature set; The key features include the maximum CPU usage rate of the network node, the maximum GPU usage rate, the judgment result of load abnormality, the connection mode of the network node before adjustment, and the connection mode of the network node after adjustment; Input the elements in the key feature set into the deep learning model to predict the probability of adjusting the network node connection strategy ; if , then it is determined that the corresponding network node needs to adjust the connection mode at a future moment, where represents the adjustment probability threshold, ; Get the current network connection strategy and adjust the network connection mode according to the prediction results of the deep learning model, including the following steps: Obtain the function identification code of the node function with abnormal load; Obtain the current network node connection strategy and load abnormality alarm information of the network. If it is a series connection mode, adjust the network node connection strategy to establish a data connection between the network node with abnormal load and the network node that is not directly connected in the network; if it is a bus connection mode, do not adjust the network node connection strategy; The adjusted network connection strategy is sent to the host computer to update the current network node connection strategy in the host computer.

6. The AI ​​function expansion method based on deep learning technology according to claim 5, characterized in that: The step of expanding the AI ​​function of one or more network nodes includes the following steps: Determine whether there is an expandable node in the network; if there is an expandable network node, call the AI ​​model in the expandable network node; the expandable node is a network node in the network that is connected to the network node with abnormal load and has normal load; The AI ​​model in the scalable network node is used to undertake the calculation of the network node in the next moment when the network node is in an abnormal load state, so as to ensure the node function of the network node with abnormal load; If there is no scalable network node or the scalable network node does not have an AI model with a corresponding node function; Then, from the network node with abnormal load, obtain the AI ​​model that performs the corresponding node function, and copy the obtained AI model to the expansion area of ​​any network node; In the extended area, using the knowledge distillation technology, the acquired AI model is used as the teacher model to train a student model with the same functions as the teacher model as the extended AI model; By expanding the AI ​​model, the computing task of the network node in the abnormal load state at the next moment is undertaken to ensure the node function of the corresponding network node.

7. An AI function expansion system based on deep learning technology, characterized in that: The system is used to execute an AI function expansion method based on deep learning technology as described in any one of claims 1 to 6 above.

8. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and the computer program is executed by a processor to implement an AI function expansion method based on deep learning technology as described in any one of claims 1 to 6.

Images