A network access node selection method and system based on mobile phone terminal AI computing power

Abstract

The application discloses a network access node selection method and system based on AI operation capacity of a mobile phone terminal, and the method comprises the following steps: acquiring a node list, a calculation mode preference, and chipset information and a driver version of the mobile phone terminal; when a trigger condition is met, an algorithm mode is selected according to the calculation mode preference; node load information of each node in the node list under the algorithm mode is determined; a plurality of candidate nodes are selected according to the node load information; an adaptive speed measurement request packet is generated according to the chipset information and the driver version, and each candidate node is subjected to speed measurement; algorithm resources required for operation are determined based on the algorithm mode; the resource occupation state of the mobile phone terminal is checked according to the algorithm resources, and a corresponding AI model is loaded; the speed measurement result is input into the AI model for calculation to output a probability value of node selection; and the optimal access node is selected according to the probability value. The accuracy and efficiency of node selection are fundamentally improved, and the system energy consumption and dependence on cloud computing are significantly reduced.

Description

Technical Field

[0001] This invention relates to the field of network acceleration technology, and in particular to a method and system for selecting network access nodes based on the AI ​​computing power of mobile devices. Background Technology

[0002] With the rapid development of mobile internet, applications such as online games, real-time audio and video communication, and ultra-high-definition streaming media have placed extreme demands on network quality (such as low latency and high stability). To optimize the network experience, network accelerators and intelligent node selection technologies have emerged. The core of these technologies lies in their ability to intelligently select an optimal node from numerous globally distributed access nodes for data relay or direct connection when a user device (such as a smartphone) connects to the internet, thereby avoiding network congestion and improving transmission quality.

[0003] Currently, most mainstream node selection schemes rely on simple network speed testing mechanisms. The typical process is: the client obtains a list of nodes from the server, then performs network latency (ping) or bandwidth tests on each node in the list, ultimately selecting the node with the lowest latency or highest bandwidth as the access point. However, this decision-making method based on a single or few dimensions is increasingly revealing its inherent limitations, making it difficult to meet the needs of modern complex network environments and diverse application scenarios. The shortcomings of existing technologies are specifically reflected in the following aspects: They lack a fine-grained perception and dynamic adaptation mechanism for the terminal's current resource status, failing to intelligently adjust algorithm modes or select computing resources based on the terminal's current health, potentially negatively impacting device battery life and system smoothness while optimizing network experience. Furthermore, this method ignores the impact of load and cost factors on nodes, resulting in the selected node not being the most cost-effective choice, reducing its practicality. Summary of the Invention

[0004] To address the aforementioned problems, this invention provides a network access node selection method and system based on mobile AI computing capabilities. This addresses the lack of a fine-grained perception and dynamic adaptation mechanism for the terminal's current resource status, as mentioned in the background art. This results in the inability to intelligently adjust algorithm modes or select computing resources based on the terminal's current health, potentially negatively impacting device battery life and system smoothness while optimizing network experience. Furthermore, it ignores the influence of load and cost factors on nodes, leading to the selection of nodes that are not the most cost-effective, thus reducing practicality.

[0005] A method for selecting network access nodes based on mobile AI computing power includes the following steps: The configuration information is obtained through the network interface of the mobile device, and the hardware information of the mobile device is also obtained. The node list and computing mode preference are obtained based on the configuration information, and the chipset information and driver version of the mobile device are obtained based on the hardware information. When the triggering condition is met, the algorithm mode is selected according to the calculation mode preference, the node load information of each node in the node list is determined in the algorithm mode, and multiple candidate nodes are selected based on the node load information. Based on the chipset information and driver version, an adapted speed test request packet is generated, and the speed test request packet is used to perform speed tests on each candidate node to obtain the speed test results. The algorithm resources required for the operation are determined based on the algorithm pattern. The resource usage status of the mobile device is checked based on the algorithm resources. The corresponding AI model is then loaded based on the resource usage status. The speed measurement results of each candidate node are input into the AI ​​model to calculate the probability value of node selection, and the optimal access node is selected based on the probability value.

[0006] Preferably, the step of obtaining configuration information through the mobile phone's network interface, simultaneously obtaining the mobile phone's hardware information, obtaining the node list and computing mode preferences based on the configuration information, and obtaining the mobile phone's chipset information and driver version based on the hardware information includes: The system configuration information package includes a node list and computing mode preference settings parameters, which are initiated by the network interface component of the mobile operating system to send a secure connection request to a predefined configuration server. Call the hardware abstraction layer interface of the mobile operating system to obtain chipset identification information, computing power level parameters and current resource usage status, including CPU, GPU and NPU; The driver version numbers of the GPU and NPU are queried through the hardware abstraction layer interface and compared with the preset compatibility list to determine whether AI model inference tasks are supported. If so, output the chipset information and driver version of the mobile device, as well as the node list and computing mode preference settings.

[0007] Preferably, when the triggering condition is met, an algorithm mode is selected according to the computation mode preference, the node load information of each node in the node list under the algorithm mode is determined, and multiple candidate nodes are selected based on the node load information, including: Real-time monitoring of mobile system status and user behavior; when preset trigger conditions are met, the node filtering process is initiated. Based on pre-configured computing mode preferences, a target algorithm mode suitable for the current scenario is selected from multiple algorithm modes; For each access node in the node list, determine the corresponding node load information evaluation dimension according to the target algorithm mode, and obtain multi-dimensional load data; Based on multi-dimensional load data, multiple candidate nodes that meet the requirements are selected from the node list through preset filtering rules and threshold conditions.

[0008] Preferably, the step of generating an adapted speed test request packet based on chipset information and driver version, and using the speed test request packet to perform speed tests on each candidate node and obtain speed test results includes: Based on chipset information and driver version analysis, the mobile device supports network protocol stacks, data processing capabilities, and hardware acceleration features. Based on the network protocol stack, data processing capabilities, and hardware acceleration features, an adapted speed test request packet is dynamically generated. The speed test request packet includes determining the speed test protocol type, data packet size, timestamp accuracy, and verification mechanism. The speed test request packet is used to concurrently initiate speed test requests to multiple candidate nodes, and the request sending time and node identification information are recorded. Receive response data packets from each candidate node, extract key speed measurement indicators, and generate a standardized speed measurement result dataset.

[0009] Preferably, the step of receiving response data packets from each candidate node, extracting key speed measurement indicators, and generating a standardized speed measurement result dataset includes: Response packets from each candidate node are received concurrently through multiple network sockets, and the arrival timestamp of each response packet is recorded. The response data packet is parsed according to the predefined speed measurement protocol format, the embedded sending timestamp, sequence number and node status information are extracted, and the integrity and validity of the response data packet are verified. One-way delay is calculated based on the sending timestamp and arrival timestamp, jitter is calculated based on the delay sequence of multiple consecutive data packets, packet loss rate is calculated based on the continuity of sequence numbers, and the proportion of high-latency data packets is statistically analyzed. The extracted raw speed measurement indicators are normalized and their units are unified. Node status information is then integrated to generate a standardized speed measurement result dataset containing temporal features and node attributes.

[0010] Preferably, the step of determining the algorithm resources required for execution based on the algorithm pattern, checking the resource usage status of the mobile device based on the algorithm resources, and loading the corresponding AI model based on the resource usage status includes: Based on algorithmic patterns, a target model is determined from multiple AI model variants, and the computing resource specifications required to execute the target model are mapped. The computing resource specifications include computing unit type, memory size, computing precision, and expected power consumption. The current resource usage status of the mobile terminal computing unit type is checked in real time through the runtime interface of the operating system. The current resource usage status includes the current utilization rate, available memory, temperature and remaining battery power. The system compares the computing resource specifications with the current resource occupancy status. If resources are sufficient, they are allocated. If there is a resource conflict, a replacement computing unit or a simplified AI model is selected according to the preset degradation strategy. Based on resource allocation or arbitration results, the corresponding AI model is loaded from local storage or network to the designated computing unit to complete model initialization.

[0011] Preferably, the step of determining the target model from multiple AI model variants based on algorithmic patterns and mapping the computational resource specifications required to execute the target model includes: Multiple candidate model variants are obtained by querying and matching from a pre-configured model variant library based on algorithm patterns; Determine the computational resource specifications required to execute each candidate model variant, including dimensions such as compute unit type, peak memory usage, expected inference latency, computational accuracy, and energy consumption level; A matching degree assessment is performed based on the computing resource specifications and the current static capability description of the mobile device to determine the final target model and its complete computing resource specifications from the candidate model variants.

[0012] Preferably, the step of inputting the speed measurement results of each candidate node into the AI ​​model to calculate the probability value of node selection, and selecting the optimal access node based on the probability value, includes: Based on the speed test results of each candidate node, multi-dimensional test indicators are obtained, and the static attributes of each candidate node are also obtained. The multi-dimensional test indicators and the static attributes of the nodes are then fused to construct a multi-dimensional feature vector. The multidimensional feature vector is input into the loaded AI model to perform forward inference calculation and obtain the comprehensive quality probability value of each candidate node. The overall quality probability value output by the AI ​​model is calibrated and smoothed to generate the final selection probability value; Compare the final selection probability values ​​of all candidate nodes, and select the target candidate node with the highest probability value as the optimal access node.

[0013] Preferably, the step of obtaining multi-dimensional test indicators based on the speed measurement results of each candidate node, simultaneously obtaining the node static attributes of each candidate node, and fusing the multi-dimensional test indicators and node static attributes to construct a multi-dimensional feature vector includes: Extract a set of dynamic test metrics reflecting real-time network quality from the speed test results of each candidate node; Query and obtain the inherent static attributes of each candidate node from the local node information database or the remote configuration server; Data cleaning, outlier handling, and data standardization are performed on dynamic test metrics and node static attributes, respectively. The standardized dynamic test metrics and node static attributes are spliced ​​and combined in a predetermined order to generate a unified, fixed-dimensional multidimensional feature vector.

[0014] A network access node selection system based on mobile AI computing power, the system comprising: The acquisition module is used to obtain configuration information through the network interface of the mobile phone, and at the same time obtain the hardware information of the mobile phone. Based on the configuration information, it obtains the node list and computing mode preference, and based on the hardware information, it obtains the chipset information and driver version of the mobile phone. The first selection module is used to select an algorithm mode according to the calculation mode preference when the triggering condition is met, determine the node load information of each node in the node list under the algorithm mode, and select multiple candidate nodes based on the node load information. The speed test module is used to generate an appropriate speed test request packet based on the chipset information and driver version, and use the speed test request packet to test the speed of each candidate node and obtain the speed test results. The loading module is used to determine the algorithm resources required for execution based on the algorithm pattern, check the resource usage status of the mobile device based on the algorithm resources, and load the corresponding AI model based on the resource usage status. The second selection module is used to input the speed measurement results of each candidate node into the AI ​​model to calculate the probability value of node selection, and select the optimal access node based on the probability value.

[0015] Other features and advantages of the invention will be set forth in the following description, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in the written description and the accompanying drawings.

[0016] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0017] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the embodiments of the invention to explain the invention and do not constitute a limitation thereof.

[0018] Figure 1 A flowchart illustrating the workflow of a network access node selection method based on mobile AI computing power provided by this invention; Figure 2 Another flowchart of a network access node selection method based on mobile AI computing power provided by the present invention; Figure 3 This is another flowchart of a network access node selection method based on mobile AI computing power provided by the present invention. Figure 4 This is a schematic diagram of the structure of a network access node selection system based on mobile AI computing power provided by the present invention. Detailed Implementation

[0019] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numerals in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this disclosure. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this disclosure as detailed in the appended claims.

[0020] A method for selecting network access nodes based on mobile AI computing power, such as Figure 1 As shown, it includes the following steps: Step S101: Obtain configuration information through the network interface of the mobile phone, and at the same time obtain the hardware information of the mobile phone. Obtain the node list and computing mode preference according to the configuration information, and obtain the chipset information and driver version of the mobile phone according to the hardware information. Step S102: When the triggering condition is met, select the algorithm mode according to the calculation mode preference, determine the node load information of each node in the node list under the algorithm mode, and select multiple candidate nodes according to the node load information. Step S103: Generate an adapted speed test request packet based on the chipset information and driver version, and use the speed test request packet to perform speed tests on each candidate node and obtain the speed test results. Step S104: Determine the algorithm resources required for the operation based on the algorithm mode, check the resource usage status of the mobile device based on the algorithm resources, and load the corresponding AI model based on the resource usage status. Step S105: Input the speed measurement results of each candidate node into the AI ​​model to calculate the probability value of node selection, and select the optimal access node based on the probability value.

[0021] In this embodiment, "mobile terminal" refers to a portable smart terminal device with mobile communication capabilities and an artificial intelligence computing unit, including but not limited to smartphones, tablets, etc., and its operating system may include Android, iOS, HarmonyOS, etc. In this embodiment, computing mode preference is represented by a user actively selecting through the application settings interface, or recommended by the system based on the user's historical behavior patterns, or a preference configuration for computing resource usage preset by the system according to the default strategy. It is used to guide the system to select between modes such as "minimum computing power", "maximum computing power" and "balanced computing power". In this embodiment, node load information is used as a comprehensive indicator to characterize the resource usage status of access nodes and the degree of network congestion, including: node computing load, node line load, node computing cost, and node line cost. In this embodiment, the algorithm modes include minimum computing power mode, maximum computing power mode, and balanced computing power mode; In this embodiment, the speed test results include latency, jitter, packet loss rate, high latency percentage, node computing load, node line load, node computing cost, node line cost, user's historical gameplay performance, current time and date attributes; In this embodiment, the AI ​​model is represented as a machine learning model that performs a comprehensive quality score on candidate nodes, and its basic architecture is a multilayer perceptron. In this embodiment, the resource occupancy status is represented by a set of runtime status information about the mobile phone's CPU, GPU, NPU and other computing units, which are obtained in real time through the operating system interface. Its core indicators include, but are not limited to: current utilization rate (average utilization rate in the past short time window), available physical memory, chip core temperature, and system remaining battery percentage.

[0022] The working principle of the above data scheme is as follows: Configuration information and hardware information of the mobile device are obtained through the network interface. Based on the configuration information, a node list and computing mode preferences are obtained; based on the hardware information, the chipset information and driver version of the mobile device are obtained. When the triggering condition is met, an algorithm mode is selected based on the computing mode preference. The node load information of each node in the node list under the algorithm mode is determined, and multiple candidate nodes are selected based on the node load information. An adapted speed test request packet is generated based on the chipset information and driver version, and the speed test request packet is used to test the speed of each candidate node, obtaining the speed test results. Based on the algorithm mode, the algorithm resources required for execution are determined. The resource occupancy status of the mobile device is checked based on the algorithm resources, and the corresponding AI model is loaded based on the resource occupancy status. The speed test results of each candidate node are input into the AI ​​model to calculate and output the probability value of node selection, and the optimal access node is selected based on the probability value.

[0023] The beneficial effects of the above technical solution are as follows: By combining the node load selection algorithm mode and calculating the selection probability of candidate nodes based on the speed test results of the AI ​​model, a deep integration of mobile AI computing power and network node selection is achieved. Through systematic steps, hardware perception, intelligent triggering, dynamic speed measurement, resource adaptation and AI decision-making are organically combined, which fundamentally improves the accuracy and efficiency of node selection, while significantly reducing system energy consumption and dependence on cloud computing. It solves the problem mentioned in the existing technology of lacking a fine perception and dynamic adaptation mechanism for the current resource status of the terminal itself, and being unable to intelligently adjust the algorithm mode or select computing resources according to the current health of the terminal. This may negatively affect the device's battery life and system smoothness while optimizing the network experience. At the same time, it ignores the impact of load factors and cost factors on nodes, resulting in the selection of nodes that are not the most cost-effective choice, reducing practicality.

[0024] In one embodiment, obtaining configuration information through the mobile phone's network interface, simultaneously obtaining the mobile phone's hardware information, obtaining a node list and computing mode preferences based on the configuration information, and obtaining the mobile phone's chipset information and driver version based on the hardware information, includes: The system configuration information package includes a node list and computing mode preference settings parameters, which are initiated by the network interface component of the mobile operating system to send a secure connection request to a predefined configuration server. Call the hardware abstraction layer interface of the mobile operating system to obtain chipset identification information, computing power level parameters and current resource usage status, including CPU, GPU and NPU; The driver version numbers of the GPU and NPU are queried through the hardware abstraction layer interface and compared with the preset compatibility list to determine whether AI model inference tasks are supported. If so, output the chipset information and driver version of the mobile device, as well as the node list and computing mode preference settings.

[0025] The beneficial effects of the above technical solution are as follows: by constructing a complete trusted chain from secure configuration retrieval and in-depth hardware capability probing to driver compatibility verification, a reliable execution environment foundation is provided for the stable and efficient execution of subsequent highly complex AI tasks. It not only ensures the availability of system functions, but also avoids system crashes or performance degradation caused by hardware driver incompatibility through proactive compatibility judgment, thereby improving the robustness and security of the system.

[0026] In one embodiment, such as Figure 2 As shown, when the triggering condition is met, an algorithm mode is selected according to the computation mode preference, the node load information of each node in the node list under the algorithm mode is determined, and multiple candidate nodes are selected based on the node load information, including: Step S201: Monitor the mobile system status and user behavior in real time. When the preset trigger conditions are met, start the node filtering process. Step S202: Based on the pre-configured computing mode preference, select the target algorithm mode suitable for the current scenario from multiple algorithm modes; Step S203: For each access node in the node list, determine the corresponding node load information evaluation dimension according to the target algorithm mode, and obtain multi-dimensional load data; Step S204: Based on multi-dimensional load data, select multiple candidate nodes that meet the requirements from the node list through preset filtering rules and threshold conditions.

[0027] In this embodiment, the preset triggering conditions include: timed triggering conditions: periodically triggering node filtering based on preset time intervals; event-driven triggering conditions: triggering when a user starts a specific application, the network connection status changes, or the system performance index exceeds a threshold; manual triggering conditions: receiving a forced refresh command initiated by the user through interface interaction; and predictive triggering conditions: triggering in advance based on historical usage patterns to predict when entering a network-sensitive scenario. In this embodiment, the algorithm mode selection step includes: Read user-preset or system-default calculation mode preferences; The computing mode preference is adaptively adjusted based on the mobile device's status, including battery level, temperature, and performance requirements. When selecting the minimum computing power mode, the computing path with the highest energy efficiency ratio should be given priority. When selecting the highest computing power mode, the computing path with the strongest computing performance should be given priority. When the computing power balancing mode is selected, the optimal balance between performance and energy consumption is sought based on a multi-objective optimization algorithm.

[0028] The beneficial effects of the above technical solution are as follows: through multi-dimensional load assessment and dynamic threshold mechanism, a large number of unsuitable nodes are efficiently filtered out before the AI ​​model intervenes, which greatly reduces the overhead of subsequent speed testing and AI calculation and improves the overall response speed of the system.

[0029] In one embodiment, the step of generating an adapted speed test request packet based on chipset information and driver version, and using the speed test request packet to perform speed tests on each candidate node and obtain the speed test results includes: Based on chipset information and driver version analysis, the mobile device supports network protocol stacks, data processing capabilities, and hardware acceleration features. Based on the network protocol stack, data processing capabilities, and hardware acceleration features, an adapted speed test request packet is dynamically generated. The speed test request packet includes determining the speed test protocol type, data packet size, timestamp accuracy, and verification mechanism. The speed test request packet is used to concurrently initiate speed test requests to multiple candidate nodes, and the request sending time and node identification information are recorded. Receive response data packets from each candidate node, extract key speed measurement indicators, and generate a standardized speed measurement result dataset.

[0030] The beneficial effects of the above technical solution are as follows: by dynamically generating the most suitable speed measurement package based on the terminal chip's capabilities and driving characteristics, and using hardware acceleration for concurrent speed measurement, the speed measurement efficiency is greatly improved while ensuring speed measurement accuracy, and the occupation of system resources and power consumption during the speed measurement process are reduced.

[0031] In one embodiment, such as Figure 3 As shown, the process of receiving response data packets from each candidate node, extracting key speed measurement indicators, and generating a standardized speed measurement result dataset includes: Step S301: Receive response data packets from each candidate node concurrently through multiple network sockets, and record the arrival timestamp of each response data packet; Step S302: Parse the response data packet according to the predefined speed measurement protocol format, extract the embedded sending timestamp, sequence number and node status information, and verify the integrity and validity of the response data packet; Step S303: Calculate the one-way delay based on the sending timestamp and the arrival timestamp, calculate the jitter based on the delay sequence of multiple consecutive data packets, calculate the packet loss rate based on the continuity of the sequence number, and count the proportion of high-latency data packets. Step S304: Normalize the extracted original speed measurement indicators and unify the units, and integrate the node status information to generate a standardized speed measurement result dataset containing time-series features and node attributes.

[0032] The beneficial effects of the above technical solution are as follows: by using high-precision timestamps, multi-indicator comprehensive calculation and data standardization, high-quality and well-organized model input data is generated, laying a solid data foundation for the accurate prediction of AI models.

[0033] In one embodiment, determining the algorithm resources required for execution based on the algorithm pattern, checking the resource usage status of the mobile device based on the algorithm resources, and loading the corresponding AI model based on the resource usage status include: Based on algorithmic patterns, a target model is determined from multiple AI model variants, and the computing resource specifications required to execute the target model are mapped. The computing resource specifications include computing unit type, memory size, computing precision, and expected power consumption. The current resource usage status of the mobile terminal computing unit type is checked in real time through the runtime interface of the operating system. The current resource usage status includes the current utilization rate, available memory, temperature and remaining battery power. The system compares the computing resource specifications with the current resource occupancy status. If resources are sufficient, they are allocated. If there is a resource conflict, a replacement computing unit or a simplified AI model is selected according to the preset degradation strategy. Based on resource allocation or arbitration results, the corresponding AI model is loaded from local storage or network to the designated computing unit to complete model initialization.

[0034] The beneficial effects of the above technical solution are as follows: through specification-state comparison and intelligent arbitration mechanism, it ensures that the AI ​​model can run on the computing unit in the most appropriate form, while meeting performance requirements, saving energy to the maximum extent, and ensuring system stability. It effectively solves the challenge of AI task deployment in a dynamic and ever-changing environment of mobile resources. Through intelligent degradation and resource reservation strategies, it ensures the smoothness of the critical AI service experience and achieves the best balance between performance and power consumption.

[0035] In one embodiment, determining the target model from multiple AI model variants based on algorithmic patterns and mapping the computational resource specifications required to execute the target model includes: Multiple candidate model variants are obtained by querying and matching from a pre-configured model variant library based on algorithm patterns; Determine the computational resource specifications required to execute each candidate model variant, including dimensions such as compute unit type, peak memory usage, expected inference latency, computational accuracy, and energy consumption level; A matching degree assessment is performed based on the computing resource specifications and the current static capability description of the mobile device to determine the final target model and its complete computing resource specifications from the candidate model variants.

[0036] The beneficial effects of the above technical solution are as follows: through multi-dimensional evaluation and matching degree calculation, it can automatically select the most suitable AI model and its execution carrier based on the constraints of the target scenario (such as performance, energy consumption, and cost), which reflects the high intelligence and adaptability of the system.

[0037] In one embodiment, the step of inputting the speed measurement results of each candidate node into an AI model to calculate the probability value of node selection, and selecting the optimal access node based on the probability value, includes: Based on the speed test results of each candidate node, multi-dimensional test indicators are obtained, and the static attributes of each candidate node are also obtained. The multi-dimensional test indicators and the static attributes of the nodes are then fused to construct a multi-dimensional feature vector. The multidimensional feature vector is input into the loaded AI model to perform forward inference calculation and obtain the comprehensive quality probability value of each candidate node. The overall quality probability value output by the AI ​​model is calibrated and smoothed to generate the final selection probability value; Compare the final selection probability values ​​of all candidate nodes, and select the target candidate node with the highest probability value as the optimal access node.

[0038] In this embodiment, the multidimensional feature vector represents a numerical feature representation constructed for each candidate node and used as input to the AI ​​model. Its dimensions are fixed and predefined, and it is composed of a "dynamic speed measurement index segment," a "node static attribute segment," and an optional "context feature segment" concatenated sequentially. The order of features within each segment is determined by a configuration file during system initialization.

[0039] The beneficial effects of the above technical solution are as follows: through feature fusion, probability post-processing, and comparison selection based on probability values, the node selection decision is not only based on the instantaneous prediction of the model, but also integrates historical data and business logic, thus making it more accurate, stable and reliable. It effectively overcomes the volatility and short-sightedness problems that may exist in pure model decision-making, and improves the decision quality and reliability of the system in complex real-world environments.

[0040] In one embodiment, the step of obtaining multi-dimensional test metrics based on the speed measurement results of each candidate node, simultaneously obtaining the node static attributes of each candidate node, and fusing the multi-dimensional test metrics and node static attributes to construct a multi-dimensional feature vector includes: Extract a set of dynamic test metrics reflecting real-time network quality from the speed test results of each candidate node; Query and obtain the inherent static attributes of each candidate node from the local node information database or the remote configuration server; Data cleaning, outlier handling, and data standardization are performed on dynamic test metrics and node static attributes, respectively. The standardized dynamic test metrics and node static attributes are spliced ​​and combined in a predetermined order to generate a unified, fixed-dimensional multidimensional feature vector.

[0041] In this embodiment, the dynamic test metrics include: basic connectivity metrics: including minimum latency, average latency, and maximum latency; network stability metrics: including latency jitter, stability coefficient calculated based on the standard deviation of the latency sequence, and packet loss rate; and advanced traffic characteristic metrics: including the proportion of high-latency packets, effective bandwidth estimate, and coefficient of variation of packet arrival time interval. In this embodiment, the static attributes of a node include: infrastructure attributes: including the node's physical geographical location, data center, network operator, and access bandwidth capacity; service and cost attributes: including the node's service level agreement type, computing resource unit price, network bandwidth unit price, and historical average availability; and topology and relationship attributes: including the autonomous system path length between the node and the user's current access point, and whether the node and the user belong to the same operator. In this embodiment, the standardized dynamic test metrics and node static attributes are concatenated and combined in a predetermined order, including: The standard structure of a feature vector is defined with the following dimensional order: [Dynamic speed measurement index segment, node static attribute segment, context feature segment]; The processed dynamic test metrics and node static attributes are concatenated according to a standard structure.

[0042] The beneficial effects of the above technical solution are as follows: by combining real-time dynamic indicators with static attributes and standardizing them, comprehensive, consistent and high-quality input information is provided for the AI ​​model, effectively improving the feature expression ability of the AI ​​model and providing an information theory basis for high-precision node selection.

[0043] In one embodiment, this embodiment also discloses a network access node selection system based on the AI ​​computing power of a mobile device, such as... Figure 4 As shown, the system includes: The acquisition module 401 is used to acquire configuration information through the network interface of the mobile terminal, and at the same time acquire the hardware information of the mobile terminal. Based on the configuration information, it acquires the node list and computing mode preference, and based on the hardware information, it acquires the chipset information and driver version of the mobile terminal. The first selection module 402 is used to select an algorithm mode according to the calculation mode preference when the triggering condition is met, determine the node load information of each node in the node list under the algorithm mode, and select multiple candidate nodes according to the node load information. The speed test module 403 is used to generate an adapted speed test request packet based on the chipset information and driver version, and use the speed test request packet to perform speed tests on each candidate node and obtain the speed test results. Loading module 404 is used to determine the algorithm resources required to perform the operation based on the algorithm mode, check the resource usage status of the mobile device based on the algorithm resources, and load the corresponding AI model based on the resource usage status. The second selection module 405 is used to input the speed measurement results of each candidate node into the AI ​​model to calculate the probability value of node selection, and select the optimal access node based on the probability value.

[0044] The working principle and beneficial effects of the above technical solution have been explained in the method embodiments, and will not be repeated here.

[0045] Those skilled in the art should understand that the "first" and "second" in this invention simply refer to different application stages.

[0046] Other embodiments of this disclosure will readily occur to those skilled in the art upon consideration of the specification and practice of the disclosure herein. This application is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of this disclosure are indicated by the following claims.

[0047] It should be understood that this disclosure is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this disclosure is limited only by the appended claims.

Claims

1. A method for selecting network access nodes based on mobile AI computing power, characterized in that, Includes the following steps: The configuration information is obtained through the network interface of the mobile device, and the hardware information of the mobile device is also obtained. The node list and computing mode preference are obtained based on the configuration information, and the chipset information and driver version of the mobile device are obtained based on the hardware information. When the triggering condition is met, the algorithm mode is selected according to the calculation mode preference, the node load information of each node in the node list is determined in the algorithm mode, and multiple candidate nodes are selected based on the node load information. Based on the chipset information and driver version, an adapted speed test request packet is generated, and the speed test request packet is used to perform speed tests on each candidate node to obtain the speed test results. The algorithm resources required for the operation are determined based on the algorithm pattern. The resource usage status of the mobile device is checked based on the algorithm resources. The corresponding AI model is then loaded based on the resource usage status. The speed measurement results of each candidate node are input into the AI ​​model to calculate the probability value of node selection, and the optimal access node is selected based on the probability value.

2. The network access node selection method based on mobile AI computing power according to claim 1, characterized in that, The process of obtaining configuration information via the mobile phone's network interface, simultaneously obtaining the mobile phone's hardware information, obtaining a node list and computing mode preferences based on the configuration information, and obtaining the mobile phone's chipset information and driver version based on the hardware information includes: The system configuration information package includes a node list and computing mode preference settings parameters, which are initiated by the network interface component of the mobile operating system to send a secure connection request to a predefined configuration server. Call the hardware abstraction layer interface of the mobile operating system to obtain chipset identification information, computing power level parameters and current resource usage status, including CPU, GPU and NPU; The driver version numbers of the GPU and NPU are queried through the hardware abstraction layer interface and compared with the preset compatibility list to determine whether AI model inference tasks are supported. If so, output the chipset information and driver version of the mobile device, as well as the node list and computing mode preference settings.

3. The network access node selection method based on mobile AI computing power according to claim 1, characterized in that, When the triggering condition is met, an algorithm mode is selected according to the computation mode preference, the node load information of each node in the node list under the algorithm mode is determined, and multiple candidate nodes are selected based on the node load information, including: Real-time monitoring of mobile system status and user behavior; when preset trigger conditions are met, the node filtering process is initiated. Based on pre-configured computing mode preferences, a target algorithm mode suitable for the current scenario is selected from multiple algorithm modes; For each access node in the node list, determine the corresponding node load information evaluation dimension according to the target algorithm mode, and obtain multi-dimensional load data; Based on multi-dimensional load data, multiple candidate nodes that meet the requirements are selected from the node list through preset filtering rules and threshold conditions.

4. The network access node selection method based on mobile AI computing power according to claim 1, characterized in that, The process of generating an adapted speed test request packet based on chipset information and driver version, and using the speed test request packet to perform speed tests on each candidate node to obtain the speed test results includes: Based on chipset information and driver version analysis, the mobile device supports network protocol stacks, data processing capabilities, and hardware acceleration features. Based on the network protocol stack, data processing capabilities, and hardware acceleration features, an adapted speed test request packet is dynamically generated. The speed test request packet includes determining the speed test protocol type, data packet size, timestamp accuracy, and verification mechanism. The speed test request packet is used to concurrently initiate speed test requests to multiple candidate nodes, and the request sending time and node identification information are recorded. Receive response data packets from each candidate node, extract key speed measurement indicators, and generate a standardized speed measurement result dataset.

5. The network access node selection method based on mobile AI computing power according to claim 4, characterized in that, The process of receiving response data packets from each candidate node, extracting key speed measurement indicators, and generating a standardized speed measurement result dataset includes: Response packets from each candidate node are received concurrently through multiple network sockets, and the arrival timestamp of each response packet is recorded. The response data packet is parsed according to the predefined speed measurement protocol format, the embedded sending timestamp, sequence number and node status information are extracted, and the integrity and validity of the response data packet are verified. One-way delay is calculated based on the sending timestamp and arrival timestamp, jitter is calculated based on the delay sequence of multiple consecutive data packets, packet loss rate is calculated based on the continuity of sequence numbers, and the proportion of high-latency data packets is statistically analyzed. The extracted raw speed measurement indicators are normalized and their units are unified. Node status information is then integrated to generate a standardized speed measurement result dataset containing temporal features and node attributes.

6. The network access node selection method based on mobile AI computing power according to claim 1, characterized in that, The process of determining the algorithm resources required for execution based on the algorithm pattern, checking the resource usage status of the mobile device based on the algorithm resources, and loading the corresponding AI model based on the resource usage status includes: Based on algorithmic patterns, a target model is determined from multiple AI model variants, and the computing resource specifications required to execute the target model are mapped. The computing resource specifications include computing unit type, memory size, computing precision, and expected power consumption. The current resource usage status of the mobile terminal computing unit type is checked in real time through the runtime interface of the operating system. The current resource usage status includes the current utilization rate, available memory, temperature and remaining battery power. The system compares the computing resource specifications with the current resource occupancy status. If resources are sufficient, they are allocated. If there is a resource conflict, a replacement computing unit or a simplified AI model is selected according to the preset degradation strategy. Based on resource allocation or arbitration results, the corresponding AI model is loaded from local storage or network to the designated computing unit to complete model initialization.

7. The network access node selection method based on mobile AI computing power according to claim 6, characterized in that, The method of determining the target model from multiple AI model variants based on algorithmic patterns and mapping the computational resource specifications required to execute the target model includes: Multiple candidate model variants are obtained by querying and matching from a pre-configured model variant library based on algorithm patterns; Determine the computational resource specifications required to execute each candidate model variant, including dimensions such as compute unit type, peak memory usage, expected inference latency, computational accuracy, and energy consumption level; A matching degree assessment is performed based on the computing resource specifications and the current static capability description of the mobile device to determine the final target model and its complete computing resource specifications from the candidate model variants.

8. The network access node selection method based on mobile AI computing power according to claim 1, characterized in that, The step of inputting the speed measurement results of each candidate node into the AI ​​model to calculate the probability value of node selection, and selecting the optimal access node based on the probability value, includes: Based on the speed test results of each candidate node, multi-dimensional test indicators are obtained, and the static attributes of each candidate node are also obtained. The multi-dimensional test indicators and the static attributes of the nodes are then fused to construct a multi-dimensional feature vector. The multidimensional feature vector is input into the loaded AI model to perform forward inference calculation and obtain the comprehensive quality probability value of each candidate node. The overall quality probability value output by the AI ​​model is calibrated and smoothed to generate the final selection probability value; Compare the final selection probability values ​​of all candidate nodes, and select the target candidate node with the highest probability value as the optimal access node.

9. The network access node selection method based on mobile AI computing power according to claim 8, characterized in that, The process of obtaining multi-dimensional test metrics based on the speed measurement results of each candidate node, simultaneously obtaining the node static attributes of each candidate node, and fusing the multi-dimensional test metrics and node static attributes to construct a multi-dimensional feature vector includes: Extract a set of dynamic test metrics reflecting real-time network quality from the speed test results of each candidate node; Query and obtain the inherent static attributes of each candidate node from the local node information database or the remote configuration server; Data cleaning, outlier handling, and data standardization are performed on dynamic test metrics and node static attributes, respectively. The standardized dynamic test metrics and node static attributes are concatenated and combined in a predetermined order to generate a unified, fixed-dimensional multidimensional feature vector.

10. A network access node selection system based on mobile AI computing power, characterized in that, The system includes: The acquisition module is used to obtain configuration information through the network interface of the mobile phone, and at the same time obtain the hardware information of the mobile phone. Based on the configuration information, it obtains the node list and computing mode preference, and based on the hardware information, it obtains the chipset information and driver version of the mobile phone. The first selection module is used to select an algorithm mode according to the calculation mode preference when the triggering condition is met, determine the node load information of each node in the node list under the algorithm mode, and select multiple candidate nodes based on the node load information. The speed test module is used to generate an appropriate speed test request packet based on the chipset information and driver version, and to use the speed test request packet to test the speed of each candidate node and obtain the speed test results. The loading module is used to determine the algorithm resources required to perform the operation based on the algorithm pattern, check the resource usage status of the mobile device based on the algorithm resources, and load the corresponding AI model based on the resource usage status. The second selection module is used to input the speed measurement results of each candidate node into the AI ​​model to calculate the probability value of node selection, and select the optimal access node based on the probability value.

Images