A hyper-converged AI performance optimization system based on adaptive network topology algorithm

Abstract

The present invention discloses a hyper-converged extended AI performance optimization system based on an adaptive network topology algorithm, which relates to cloud computing software, including a user interface end, a system management end, and a performance optimization end. The user interface end is used to provide a tool for visualizing system status display and system configuration, and realizes user interaction and optimization report generation functions. The system management end is used to manage data transmission and storage, and realizes user identity access control and system security maintenance. The performance optimization end is used to integrate network resources, adaptively and dynamically adjust network structure, and use AI to optimize system performance. The present invention discloses a hyper-converged extended AI performance optimization system based on an adaptive network topology algorithm, proposes an adaptive network topology algorithm based on a neural network to monitor and dynamically adjust network traffic, delay, and node status in real time, and proposes a task scheduling performance optimization algorithm based on swarm intelligence to intelligently schedule system computing tasks to maximize resource utilization and minimize computing time.

Description

Technical Field

[0001] The present invention relates to the field of cloud computing software, in particular to the fields of neural networks, network topology and swarm intelligence optimization, and specifically to a hyper-converged extended AI performance optimization system based on an adaptive network topology algorithm. Background Art

[0002] Neural network and network topology technology is a technology that dynamically adjusts and optimizes network topology through neural network algorithms. It aims to solve the problem that traditional network topology lacks adaptability when facing dynamic changes and complex environments. Neural network and network topology technology can monitor network traffic, latency and node status in real time, and use the self-learning and adaptive capabilities of neural networks to dynamically adjust network topology, improve network performance and resource utilization efficiency, thereby optimizing data transmission paths and reducing network latency. The combination of neural network and network topology technology is widely used in various scenarios that require efficient and flexible network management.

[0003] Swarm intelligence optimization technology is an optimization algorithm that imitates the behavior of groups in nature. It aims to solve the problems of low efficiency and uneven resource utilization of traditional task scheduling methods when dealing with large-scale complex tasks. By simulating the intelligent behavior of biological groups, swarm intelligence optimization technology can realize intelligent scheduling and resource allocation of computing tasks. According to the real-time resource status and task requirements of the system, swarm intelligence optimization technology can dynamically adjust the scheduling strategy of computing tasks to maximize resource utilization and minimize computing time, thereby significantly improving the overall performance and efficiency of system computing. In systems that require efficient resource management and intelligent task scheduling, swarm intelligence optimization technology has important application value for system performance optimization. .

[0004] However, the network topology structure of the existing hyper-converged expansion AI performance optimization system based on adaptive network topology algorithm lacks adaptive capabilities and cannot be dynamically adjusted according to the real-time network status, resulting in low network resource utilization and poor data transmission efficiency. When faced with large-scale and complex system computing tasks, it is unable to fully utilize system resources, resulting in long computing time and waste of resources. The improvement of the overall system performance depends on static configuration and manual adjustment, and lacks intelligent and automated optimization methods. Summary of the invention

[0005] The purpose of the present invention is to provide a cloud computing software, a hyper-converged expansion AI performance optimization system based on an adaptive network topology algorithm, so as to solve the problems that the system network topology structure of the existing hyper-converged expansion AI performance optimization system based on an adaptive network topology algorithm proposed in the above background technology lacks adaptability and cannot be dynamically adjusted according to the real-time network status, resulting in low network resource utilization and poor data transmission efficiency. When faced with large-scale and complex system computing tasks, system resources cannot be fully utilized, resulting in long computing time and waste of resources, and the improvement of the overall system performance depends on static configuration and manual adjustment, and lacks intelligent and automated optimization means.

[0006] To achieve the above-mentioned purpose, the present invention provides the following technical solutions: a hyper-converged extended AI performance optimization system based on an adaptive network topology algorithm, comprising a user interface end, a system management end, and a performance optimization end, characterized in that: the user interface end comprises a control panel and configuration tool module and a user interaction and report generation module, the control panel and configuration tool module are used to build a visual control and management interface, the control panel displays the real-time performance indicators and operating status of the system, the configuration tool module allows users to customize and adjust system parameters, the user interaction and report generation module is used for real-time interaction between the user and the system, and is responsible for generating reports on system performance and resource utilization, providing data support and decision-making reference for users, the system management end comprises a data transmission and storage module and an access control and security maintenance module, the data transmission and storage module is used to efficiently process and manage data transmission and storage, and ensure that data is in the system The access control and security maintenance module is used to implement identity authentication and access control mechanisms, manage user access rights to system resources, and maintain the safe operation of the system to prevent unauthorized access and potential security threats. The performance optimization end includes a hyper-converged infrastructure module, an adaptive network topology management module, and an AI performance optimization module. The hyper-converged infrastructure module is used to support the rapid configuration and expansion of network resources to achieve efficient utilization of system resources. The adaptive network topology management module proposes an adaptive network topology algorithm based on a neural network to dynamically adjust the network topology structure to optimize the data transmission path and network resource utilization. The AI ​​performance optimization module proposes a task scheduling performance optimization algorithm based on swarm intelligence to intelligently schedule system computing tasks to maximize network resource utilization and minimize computing time to improve the overall performance of the system.

[0007] Preferably, the user interface end includes a control panel and a configuration tool module, which display various indicators and performance data of the system through a visual interface, so that users can intuitively understand the system operation status and resource utilization, and ensure that users can easily adjust system parameters and function settings to meet the requirements of different needs and scenarios.

[0008] Preferably, the user interface end includes a user interaction and report generation module, which enables users to interact with the system conveniently by providing a user-friendly interface, ensuring that users can understand the operation status and performance of the system in real time, thereby managing and optimizing the system, and collecting the operation data and performance indicators of various parts of the system to generate detailed performance reports, thereby helping users make appropriate adjustments and improvements.

[0009] Preferably, the system management end includes a data transmission and storage module, which optimizes the data transmission mechanism and provides an efficient data storage solution to meet the needs of efficient flow and secure storage of data within the system, ensure that the system can quickly respond to data read and write requests, and protect the integrity and security of the data.

[0010] Preferably, the system management end includes an access control and security maintenance module, which implements strict identity authentication and access control mechanisms so that only authorized users can access system resources and information, ensuring that the system can detect and respond to security incidents in a timely manner and protect the security of the system and data.

[0011] Preferably, the performance optimization end includes a hyper-converged infrastructure module, which abstracts physical resources into virtual resources by managing and scheduling network resources within the system, and provides a unified resource pool for multiple network environments to achieve resource sharing and dynamic allocation, ensuring that the system can provide an efficient, flexible and scalable computing platform to meet the needs of various application scenarios.

[0012] Preferably, the performance optimization end includes an adaptive network topology management module, which proposes an adaptive network topology algorithm based on a neural network, improves system performance and efficiency by monitoring the network status in real time and making dynamic adjustments, and ensures that the system can adjust the network topology according to the real-time network status to adapt to different workloads and environmental changes.

[0013] Specifically, the neural network-based adaptive network topology algorithm is as follows: First, the parameters of the neural network, including the weight matrix W and the bias vector b, are initialized by introducing a heuristic initialization strategy that considers the network structure and the expected learning task characteristics. The specific formula of the initialization method is expressed as:

[0014]

[0015] in, It is represented as the initial value of the weight from the i-th input layer neuron to the j-th hidden layer neuron, i represents the neuron index of the input layer, j represents the neuron index of the hidden layer, U(δ lower ,δ upper ) is expressed as lower ,δ upper ] is a random variable uniformly distributed in the range, δ lower Represents the lower limit of weight initialization, δ upper Represents the upper limit of weight initialization, Represents the initial value of the bias of the jth hidden layer neuron, ∈ represents a constant coefficient, and uses uniform distribution to generate the initial weights within a certain initialization range, and initializes the bias term to a small constant coefficient to ensure that the neuron does not bias toward any specific activation state at the beginning of training. Then, the initialized weights and biases are applied to the forward propagation process of the input data, and the input data is propagated through the network structure to generate the output of the network. The specific formula is expressed as:

[0016]

[0017] in, It is represented as the linear combination of the weighted input and bias of the jth neuron in the hidden layer during the tth iteration, where t is the index of the number of iterations of the neural network. Represented as the input feature of the i-th input layer neuron in the t-th iteration process, It is represented as the activation output of the jth neuron in the hidden layer during the tth iteration, and g is represented as the activation function. It is represented as the predicted output of the kth neuron in the output layer during the tth iteration, k represents the neuron index of the output layer, and W' kj It is represented as the weight from the jth neuron in the hidden layer to the kth neuron in the output layer, b' k It is expressed as the bias term of the kth neuron in the output layer. Secondly, according to the error between the output and the expected output, the network weight is adjusted to optimize the network performance. The specific formula is expressed as:

[0018]

[0019] Among them, E represents the error function of the neural network, N represents the total number of samples, which is used to normalize the error to ensure that the errors under different sample numbers are comparable, and T represents the total number of iterations of the neural network. It is expressed as the expected target output in the tth iteration process, Represented as the predicted output during the tth iteration, it is the prediction made by the neural network based on the current weights and biases, Expressed as the square of the Euclidean norm, it is used to calculate the prediction error of the data sample during the tth iteration. It is represented as the updated weight from the i-th neuron in the input layer to the j-th neuron in the hidden layer during the t+1th iteration. It is represented by the weight value from the i-th neuron in the input layer to the j-th neuron in the hidden layer before updating in the t-th iteration, η is the learning rate, and w ij It is represented as the weight from the i-th input layer neuron to the j-th hidden layer neuron, Expressed as a partial derivative operation, It is represented as the updated bias term of the jth neuron in the corresponding hidden layer during the t+1th iteration. It is represented as the bias term before updating the jth neuron in the corresponding hidden layer in the tth iteration, b j It is expressed as the bias term of the jth hidden layer neuron. The difference between the predicted output of the neural network and the actual target value is evaluated by calculating the error and updating the weights. The weights and biases of the network are adjusted accordingly to minimize the difference and continuously optimize the performance of the neural network. Finally, the network structure is adaptively and dynamically adjusted according to the network performance and the requirements of the learning task, including adding or reducing neurons and adjusting the connection mode. The specific formula for the adaptive network topology adjustment decision is expressed as:

[0020]

[0021] Among them, I j is represented as the importance index of the jth neuron, which is used to evaluate the role and importance of the neuron in the entire network. |·| is represented as the absolute value function, Θ j Represented as the decision variable of the jth neuron, according to the importance index I j With the preset threshold I thresh The comparison results determine the degree of neuron enhancement and weakening. thresh Represented as a preset importance threshold, used to determine the critical point of neuron importance, Represented as the updated connection weight from the input layer i to the jth neuron, It is represented as the connection weight from the input layer i to the jth neuron before updating, and R is represented as a random perturbation term, which is used to introduce randomness when updating the weight. It is represented as the updated bias term corresponding to the jth neuron, It is represented as the bias term before updating corresponding to the jth neuron, r is represented as the random bias perturbation term, It is represented as the connection state from the i-th neuron to the j-th neuron, whether it is activated or disconnected after the update, W thresh The connection threshold represented as a weight is used to determine the state of the weight remaining connected. By evaluating the importance of neurons and making decisions on adding or reducing neurons accordingly, the neural network-based adaptive network topology algorithm can adaptively optimize the network structure, monitor network traffic, latency, and node status in real time, and make dynamic adjustments to cope with different learning tasks and performance requirements.

[0022] Preferably, the performance optimization end includes an AI performance optimization module, which proposes a task scheduling performance optimization algorithm based on swarm intelligence, and reasonably allocates computing resources according to task priority and system resource availability to maximize resource utilization and minimize computing time, ensuring that the system can intelligently schedule and optimize system computing tasks according to real-time task requirements and system resource conditions, thereby improving the overall performance and efficiency of the system.

[0023] Specifically, the task scheduling performance optimization algorithm based on swarm intelligence is as follows: First, an initial task scheduling solution group X is established. In the initialization stage, a group of task scheduling solutions is randomly generated, and each solution represents an individual in the group. The specific formula is expressed as follows:

[0024]

[0025] Among them, X (0) Represented as the initial population, which contains all individuals in the population, It is represented as the initial task scheduling plan of the i-th individual in the group, i represents the index of the individual in the group, and n represents the total number of individuals in the group. During the initialization process, the scheduling plan of each individual will be determined based on the nature of the task and available resources. It is represented as the initial task scheduling scheme of n individuals in the group. Then, the fitness value is calculated for each individual in the initialization group to quantitatively evaluate the performance of each individual in the task scheduling problem. The fitness value is calculated based on the satisfaction degree of the individual task scheduling scheme with the predetermined performance index. The specific formula is expressed as:

[0026] F(x i )=w1·C makespan (xi )+w2·C resource (x i )

[0027] Among them, F(x i ) is expressed as the fitness value of the i-th individual, x i It is represented as the task scheduling plan of the i-th individual in the group, w1 is the weight coefficient related to the total time required to complete all tasks, and C makespan (x i ) is represented by individual x i The completion time cost of the task scheduling scheme is expressed as w2, which is the weight coefficient related to the resource utilization rate when executing the task, C resource (x i ) is represented by individual x i The resource utilization cost of the task scheduling scheme is calculated. Secondly, by simulating group behavior in nature, the individuals in the group are updated to explore the solution space and approach the optimal solution. The specific formula is expressed as:

[0028]

[0029] in, It represents the updated scheduling plan of individual i in the t+1th iteration, t represents the index of the number of algorithm iterations, It is represented as the current scheduling plan of individual i in the tth iteration process, It is represented as the optimal scheduling scheme for the entire group in the tth iteration process, φ is represented as the learning rate, which is used to control the step size of learning to the optimal solution, δ is represented as the perturbation factor, rand(0,1) is represented as a random number in the interval [0,1], which is used to introduce randomness in the update process to maintain group diversity and prevent the algorithm from converging to the local optimal solution too early, rand(0,1) is represented as a random number uniformly distributed in the interval [0,1], which is used to realize random perturbations in the update process. In the group update process, each individual will be evaluated according to its own fitness value, and corresponding update operations will be performed according to the evaluation results, including information exchange between individuals, mutation operations, and combination with other individuals to generate a new generation of individuals. Then, through selection and elite strategies, excellent individuals are retained and poor individuals are eliminated to promote the evolution of the group to the optimal solution. The specific formula is expressed as follows:

[0030]

[0031] Among them, X (t+1) It is represented as the new group set in the t+1th iteration after selection, SelectTop k It means that k individuals with the highest fitness values ​​will be selected from the current population, and k represents the number of elite individuals selected to be retained. It is expressed as the fitness value of individual i in the tth iteration process, It is expressed as existence in mathematical operations. At the end of each iteration, a selection operation is performed based on the fitness value of the individual. The selection process will determine the excellent individuals to enter the next iteration according to the fitness value. The optimal solution in the group is retained through the elite strategy, and the diversity of the group is maintained through the selection mechanism, thereby effectively balancing the relationship between discovering new solutions and improving the current solutions. Finally, at the end of each iteration, the preset termination condition is used to evaluate whether to continue the algorithm. The specific formula is expressed as:

[0032] If(t≥MaxIter)or(F(x best )≥Threshold)then stop

[0033] Among them, MaxIter represents the preset maximum number of iterations, F(x best ) represents the fitness value of the best individual in the current group, Threshold represents the preset fitness threshold, which is used to evaluate the satisfaction of the solution, and stop means that the algorithm will stop iterating and output the best task scheduling solution currently found.

[0034] Compared with the prior art, the present invention has the following beneficial effects:

[0035] 1. Adaptive network topology management module proposes an adaptive network topology algorithm based on neural network. This algorithm optimizes the data transmission path and network resource utilization by real-time monitoring of network status and dynamic adjustment, thereby improving system performance and efficiency. Specifically, this algorithm uses the learning and adaptability of neural networks to adaptively adjust the network topology structure according to the real-time network status to ensure that the system can flexibly respond to different workloads and environmental changes. By building a neural network model, the system can adjust network nodes and connection methods in real time to make data transmission more efficient and reduce transmission bottlenecks caused by fixed topology structures. This dynamic adjustment capability not only improves the speed and reliability of data transmission, but also reduces the overall delay of the system. The system can flexibly increase and decrease neural network nodes according to the current network load situation, effectively responding to The fluctuation of network traffic avoids the performance degradation caused by network congestion and resource waste. The adaptive network topology algorithm based on neural network can maintain efficient operation in complex and changeable environments. Through continuous learning and adjustment, the neural network system can adapt to different network environments and task requirements, ensuring that the system can maintain optimal performance under various circumstances. The adaptive ability of the constructed neural network model enables the system to quickly reconfigure the network and restore normal operation when nodes fail and topology changes, reducing system downtime and maintenance costs. In general, the adaptive network topology algorithm based on neural network provides strong technical support for the hyper-converged expansion AI performance optimization system, significantly improving the performance, efficiency and reliability of the system, and can flexibly respond to complex and changeable network environments and task requirements, bringing users a better user experience.

[0036] 2. The AI ​​performance optimization module proposes a task scheduling performance optimization algorithm based on swarm intelligence. This algorithm optimizes the scheduling of complex data tasks in the hyper-converged extended AI performance optimization system based on the adaptive network topology algorithm by simulating the intelligence of swarm behavior in nature, thereby improving the efficiency and accuracy of data integration and analysis. The algorithm creates a diverse set of task scheduling schemes by initializing the swarm step, which provides a broad starting point for the algorithm's search process and adaptability to the diverse data environments in the hyper-converged extended AI performance optimization system based on the adaptive network topology algorithm. The fitness value calculation step in the algorithm comprehensively considers the two performance indicators of task completion time and resource utilization. Through the flexible adjustment of the weight coefficient, it can customize the optimization target for different data integration needs, ensuring the efficiency and flexibility of the algorithm in metaverse data integration and analysis. The swarm update step continuously optimizes the scheduling scheme in the swarm by simulating natural selection and genetic mechanisms. , which enables the algorithm to quickly adapt to changes in the data environment in the hyper-converged expansion AI performance optimization system based on the adaptive network topology algorithm, and maintains the advancement and effectiveness of the scheduling scheme. In addition, the selection and elite strategy steps ensure the inheritance of excellent scheduling schemes. By selecting individuals with the highest fitness values ​​to enter the next generation, the global search capability of the algorithm is enhanced, while also maintaining the diversity of the group. Finally, the termination condition judgment step enables the algorithm to intelligently stop after reaching the predetermined optimization goal or number of iterations, effectively saving computing resources and improving the practicality of the algorithm. In summary, the task scheduling performance optimization algorithm based on swarm intelligence not only improves the performance of data task scheduling, but also enhances the system's adaptability and flexibility to complex data environments. Through the algorithm's performance optimization of the system, the system's computing tasks are intelligently scheduled to maximize network resource utilization and minimize computing time to improve the overall performance of the system. It has important practical value and broad application prospects. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] The invention is further illustrated by the accompanying drawings, but the embodiments in the accompanying drawings do not constitute any limitation to the invention. A person skilled in the art can obtain other drawings based on the following drawings without creative work.

[0038] Figure 1 It is a structural schematic diagram of the present invention. DETAILED DESCRIPTION

[0039] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0040] See also Figure 1 The present invention provides a hyper-converged extended AI performance optimization system based on an adaptive network topology algorithm, including a user interface end, a system management end, and a performance optimization end, characterized in that: the user interface end includes a control panel and configuration tool module and a user interaction and report generation module, the control panel and configuration tool module are used to build a visual control and management interface, the control panel displays the real-time performance indicators and operating status of the system, the configuration tool module allows users to customize and adjust system parameters, the user interaction and report generation module is used for real-time interaction between users and the system, and is responsible for generating reports on system performance and resource utilization, providing data support and decision-making reference for users, the system management end includes a data transmission and storage module and an access control and security maintenance module, the data transmission and storage module is used to efficiently process and manage data transmission and storage, and ensure the security of data within the system The access control and security maintenance module is used to implement identity authentication and access control mechanisms, manage user access rights to system resources, and maintain the safe operation of the system to prevent unauthorized access and potential security threats. The performance optimization end includes a hyper-converged infrastructure module, an adaptive network topology management module, and an AI performance optimization module. The hyper-converged infrastructure module is used to support the rapid configuration and expansion of network resources to achieve efficient utilization of system resources. The adaptive network topology management module proposes an adaptive network topology algorithm based on a neural network to dynamically adjust the network topology structure to optimize the data transmission path and network resource utilization. The AI ​​performance optimization module proposes a task scheduling performance optimization algorithm based on swarm intelligence to intelligently schedule system computing tasks to maximize network resource utilization and minimize computing time to improve the overall performance of the system.

[0041] See also Figure 1 Furthermore, the user interface includes a control panel and a configuration tool module, which displays various indicators and performance data of the system through a visual interface, so that users can intuitively understand the system operation status and resource utilization, and ensure that users can easily adjust system parameters and function settings to meet the requirements of different needs and scenarios.

[0042] See also Figure 1 Furthermore, the user interface end includes a user interaction and report generation module, which enables users to interact with the system conveniently by providing a user-friendly interface, ensuring that users can understand the operation status and performance of the system in real time, thereby managing and optimizing the system, and collecting the operation data and performance indicators of each part of the system to generate a detailed performance report, thereby helping users make appropriate adjustments and improvements.

[0043] See also Figure 1 Furthermore, the system management end includes a data transmission and storage module, which optimizes the data transmission mechanism and provides an efficient data storage solution to meet the needs of efficient flow and secure storage of data within the system, ensure that the system can quickly respond to data read and write requests, and ensure the integrity and security of the data.

[0044] See also Figure 1 Furthermore, the system management end includes an access control and security maintenance module, which implements strict identity authentication and access control mechanisms so that only authorized users can access system resources and information, ensuring that the system can detect and respond to security incidents in a timely manner and protect the security of the system and data.

[0045] See also Figure 1 ,Furthermore, the performance optimization end includes a hyper-converged infrastructure module, which ,abstracts physical resources into virtual resources by managing and scheduling network resources within the ,system, and provides a unified resource pool for multiple network environments, ,implementing resource sharing and dynamic allocation, ensuring that the system can provide an ,efficient, flexible and scalable computing platform to meet the needs of ,various application scenarios.

[0046] See also Figure 1 ,Furthermore, the performance optimization end includes an adaptive network topology management module, ,which proposes an adaptive network topology algorithm based on ,neural network, which improves system performance and efficiency by real-time monitoring of ,network status and making dynamic adjustments, ensuring that the system can adjust the ,network topology according to the real-time network status to adapt to different ,workloads and environmental changes.

[0047] See also Figure 1 ,Furthermore, the adaptive network topology algorithm based on neural network is as follows: First, initialize the parameters of the neural network, including the weight matrix W and the bias vector b, and initialize the neural network parameters by introducing a heuristic initialization strategy that considers the network structure and the expected learning task characteristics. The specific formula of the initialization method is expressed as:

[0048]

[0049] in, It is represented as the initial value of the weight from the i-th input layer neuron to the j-th hidden layer neuron, i represents the neuron index of the input layer, j represents the neuron index of the hidden layer, U(δ lower ,δ upper ) is expressed as lower ,δ upper ] is a random variable uniformly distributed in the range, δ lowerRepresents the lower limit of weight initialization, δ upper Represents the upper limit of weight initialization, Represents the initial value of the bias of the jth hidden layer neuron, ∈ represents a constant coefficient, and uses uniform distribution to generate the initial weights within a certain initialization range, and initializes the bias term to a small constant coefficient to ensure that the neuron does not bias toward any specific activation state at the beginning of training. Then, the initialized weights and biases are applied to the forward propagation process of the input data, and the input data is propagated through the network structure to generate the output of the network. The specific formula is expressed as:

[0050]

[0051] in, It is represented as the linear combination of the weighted input and bias of the jth neuron in the hidden layer during the tth iteration, where t is the index of the number of iterations of the neural network. Represented as the input feature of the i-th input layer neuron in the t-th iteration process, It is represented as the activation output of the jth neuron in the hidden layer during the tth iteration, and g is represented as the activation function. It is represented as the predicted output of the kth neuron in the output layer during the tth iteration, k represents the neuron index of the output layer, and W' kj It is represented as the weight from the jth neuron in the hidden layer to the kth neuron in the output layer, b' k It is expressed as the bias term of the kth neuron in the output layer. Secondly, according to the error between the output and the expected output, the network weight is adjusted to optimize the network performance. The specific formula is expressed as:

[0052]

[0053] Among them, E represents the error function of the neural network, N represents the total number of samples, which is used to normalize the error to ensure that the errors under different sample numbers are comparable, and T represents the total number of iterations of the neural network. It is expressed as the expected target output in the tth iteration process, Represented as the predicted output during the tth iteration, it is the prediction made by the neural network based on the current weights and biases, Expressed as the square of the Euclidean norm, it is used to calculate the prediction error of the data sample during the tth iteration. It is represented as the updated weight from the i-th neuron in the input layer to the j-th neuron in the hidden layer during the t+1th iteration. It is represented by the weight value from the i-th neuron in the input layer to the j-th neuron in the hidden layer before updating in the t-th iteration, η is the learning rate, and w ijIt is represented as the weight from the i-th input layer neuron to the j-th hidden layer neuron, Expressed as a partial derivative operation, It is represented as the updated bias term of the jth neuron in the corresponding hidden layer during the t+1th iteration. It is represented as the bias term before updating the jth neuron in the corresponding hidden layer in the tth iteration, b j It is expressed as the bias term of the jth hidden layer neuron. The difference between the predicted output of the neural network and the actual target value is evaluated by calculating the error and updating the weights. The weights and biases of the network are adjusted accordingly to minimize the difference and continuously optimize the performance of the neural network. Finally, the network structure is adaptively and dynamically adjusted according to the network performance and the requirements of the learning task, including adding or reducing neurons and adjusting the connection mode. The specific formula for the adaptive network topology adjustment decision is expressed as:

[0054]

[0055] Among them, I j is represented as the importance index of the jth neuron, which is used to evaluate the role and importance of the neuron in the entire network. |·| is represented as the absolute value function, Θ j Represented as the decision variable of the jth neuron, according to the importance index I j With the preset threshold I thresh The comparison results determine the degree of neuron enhancement and weakening. thresh Represented as a preset importance threshold, used to determine the critical point of neuron importance, Represented as the updated connection weight from the input layer i to the jth neuron, It is represented as the connection weight from the input layer i to the jth neuron before updating, and R is represented as a random perturbation term, which is used to introduce randomness when updating the weight. It is represented as the updated bias term corresponding to the jth neuron, It is represented as the bias term before updating corresponding to the jth neuron, r is represented as the random bias perturbation term, It is represented as the connection state from the i-th neuron to the j-th neuron, whether it is activated or disconnected after the update, W thresh The connection threshold represented as a weight is used to determine the state of the weight remaining connected. By evaluating the importance of neurons and making decisions on adding or reducing neurons accordingly, the neural network-based adaptive network topology algorithm can adaptively optimize the network structure, monitor network traffic, latency, and node status in real time, and make dynamic adjustments to cope with different learning tasks and performance requirements.

[0056] See also Figure 1,Furthermore, the performance optimization end includes an AI performance optimization module, which proposes a task scheduling ,performance optimization algorithm based on swarm intelligence, and reasonably allocates ,computing resources according to the priority of the task and the availability of ,system resources to maximize resource utilization and minimize computing time, ensuring that the system can ,intelligently schedule and optimize system computing tasks according to real-time ,task requirements and system resource conditions, thereby improving the ,overall performance and efficiency of the system.

[0057] See also Figure 1 ,Furthermore, the task scheduling performance optimization algorithm based on swarm intelligence is as follows: ,First, an initial task scheduling solution group X is established. ,In the initialization stage, a set of task scheduling schemes are randomly generated, ,each scheme represents an individual in the group. ,The specific formula is expressed as:

[0058]

[0059] Among them, X (0) Represented as the initial population, which contains all individuals in the population, It is represented as the initial task scheduling plan of the i-th individual in the group, i represents the index of the individual in the group, and n represents the total number of individuals in the group. During the initialization process, the scheduling plan of each individual will be determined based on the nature of the task and available resources. It is represented as the initial task scheduling scheme of n individuals in the group. Then, the fitness value is calculated for each individual in the initialization group to quantitatively evaluate the performance of each individual in the task scheduling problem. The fitness value is calculated based on the satisfaction degree of the individual task scheduling scheme with the predetermined performance index. The specific formula is expressed as:

[0060] F(x i )=w1·C makespan (x i )+w2·C resource (x i )

[0061] Among them, F(x i ) is expressed as the fitness value of the i-th individual, x i It is represented as the task scheduling plan of the i-th individual in the group, w1 is the weight coefficient related to the total time required to complete all tasks, and C makespan (x i ) is represented by individual x i The completion time cost of the task scheduling scheme is expressed as w2, which is the weight coefficient related to the resource utilization rate when executing the task, C resource (x i ) is represented by individual x iThe resource utilization cost of the task scheduling scheme is calculated. Secondly, by simulating group behavior in nature, the individuals in the group are updated to explore the solution space and approach the optimal solution. The specific formula is expressed as:

[0062]

[0063] in, It represents the updated scheduling plan of individual i in the t+1th iteration, t represents the index of the number of algorithm iterations, It is represented as the current scheduling plan of individual i in the tth iteration process, It is represented as the optimal scheduling scheme for the entire group in the tth iteration process, φ is represented as the learning rate, which is used to control the step size of learning to the optimal solution, δ is represented as the perturbation factor, rand(0,1) is represented as a random number in the interval [0,1], which is used to introduce randomness in the update process to maintain group diversity and prevent the algorithm from converging to the local optimal solution too early, rand(0,1) is represented as a random number uniformly distributed in the interval [0,1], which is used to realize random perturbations in the update process. In the group update process, each individual will be evaluated according to its own fitness value, and corresponding update operations will be performed according to the evaluation results, including information exchange between individuals, mutation operations, and combination with other individuals to generate a new generation of individuals. Then, through selection and elite strategies, excellent individuals are retained and poor individuals are eliminated to promote the evolution of the group to the optimal solution. The specific formula is expressed as follows:

[0064]

[0065] Among them, X (t+1) It is represented as the new group set in the t+1th iteration after selection, SelectTop k It means that k individuals with the highest fitness values ​​will be selected from the current population, and k represents the number of elite individuals selected to be retained. It is expressed as the fitness value of individual i in the tth iteration process, It is expressed as existence in mathematical operations. At the end of each iteration, a selection operation is performed based on the fitness value of the individual. The selection process will determine the excellent individuals to enter the next iteration according to the fitness value. The optimal solution in the group is retained through the elite strategy, and the diversity of the group is maintained through the selection mechanism, thereby effectively balancing the relationship between discovering new solutions and improving the current solutions. Finally, at the end of each iteration, the preset termination condition is used to evaluate whether to continue the algorithm. The specific formula is expressed as:

[0066] If(t≥MaxIter)or(F(x best )≥Threshold)then stop

[0067] Among them, MaxIter represents the preset maximum number of iterations, F(x best ) represents the fitness value of the best individual in the current group, Threshold represents the preset fitness threshold, which is used to evaluate the satisfaction of the solution, and stop means that the algorithm will stop iterating and output the best task scheduling solution currently found.

[0068] When used specifically, first, the user interface end includes a control panel and configuration tool module and a user interaction and report generation module. The control panel and configuration tool module construct a visual control and management interface. The control panel displays the real-time performance indicators and operating status of the system. The configuration tool module allows users to customize and adjust system parameters. The user interaction and report generation module is used for real-time interaction between users and the system, and is responsible for generating reports on system performance and resource utilization to provide users with data support and decision-making references. Then, the system management end includes a data transmission and storage module and an access control and security maintenance module. The data transmission and storage module efficiently processes and manages data transmission and storage to ensure the security and reliability of data within the system. The access control and security maintenance module is used to implement identity authentication and access. Control mechanism, manage user access rights to system resources, and maintain the safe operation of the system to prevent unauthorized access and potential security threats. Finally, the performance optimization end includes a hyper-converged infrastructure module, an adaptive network topology management module, and an AI performance optimization module. The hyper-converged infrastructure module supports the rapid configuration and expansion of network resources to achieve efficient utilization of system resources. The adaptive network topology management module proposes an adaptive network topology algorithm based on a neural network to dynamically adjust the network topology structure to optimize the data transmission path and network resource utilization. The AI ​​performance optimization module proposes a task scheduling performance optimization algorithm based on swarm intelligence to intelligently schedule system computing tasks to maximize network resource utilization and minimize computing time to improve the overall performance of the system.

[0069] Although the present invention has been described in detail with reference to the aforementioned embodiments, it is still possible for those skilled in the art to modify the technical solutions described in the aforementioned embodiments, or to make equivalent substitutions for some of the technical features therein. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A hyper-converged AI performance optimization system based on an adaptive network topology algorithm, comprising a user interface end, a system management end, and a performance optimization end, characterized in that: The user interface end includes a control panel and configuration tool module and a user interaction and report generation module. The control panel and configuration tool module are used to build a visual control and management interface. The control panel displays the real-time performance indicators and operating status of the system. The configuration tool module allows users to customize and adjust system parameters. The user interaction and report generation module is used for real-time interaction between users and the system, and is responsible for generating reports on system performance and resource utilization to provide users with data support and decision-making references. The system management end includes a data transmission and storage module and an access control and security maintenance module. The data transmission and storage module is used to process and manage data transmission and storage to ensure the security and reliability of data within the system. The access control and security maintenance module is used to implement identity authentication and access control The system comprises a mechanism for managing user access rights to system resources and maintaining the safe operation of the system to prevent unauthorized access and potential security threats. The performance optimization end comprises a hyper-converged infrastructure module, an adaptive network topology management module and an AI performance optimization module. The hyper-converged infrastructure module is used to support the rapid configuration and expansion of network resources to achieve efficient utilization of system resources. The adaptive network topology management module proposes an adaptive network topology algorithm based on a neural network to dynamically adjust the network topology structure to optimize the data transmission path and network resource utilization. The AI ​​performance optimization module proposes a task scheduling performance optimization algorithm based on swarm intelligence to intelligently schedule system computing tasks to maximize network resource utilization and minimize computing time to improve the overall performance of the system. The performance optimization end includes an adaptive network topology management module, which proposes an adaptive network topology algorithm based on a neural network, monitors the network status in real time and makes dynamic adjustments to improve system performance and efficiency, and ensures that the system can adjust the network topology according to the real-time network status to adapt to different workloads and environmental changes; First, initialize the parameters of the neural network. The specific formula is: in, It is represented as the initial value of the weight from the i-th input layer neuron to the j-th hidden layer neuron, i represents the neuron index of the input layer, j represents the neuron index of the hidden layer, U(δ lower ,δ upper ) is expressed as lower ,δ upper ] is a random variable uniformly distributed in the range, δ lower Represents the lower limit of weight initialization, δ upper Represents the upper limit of weight initialization, It is represented as the initial value of the bias of the jth hidden layer neuron, ∈ is represented as a constant coefficient, and the initialized weights and biases are applied to the forward propagation process of the input data. The specific formula is expressed as follows: in, It is represented as the linear combination of the weighted input and bias of the jth neuron in the hidden layer during the tth iteration, where t is the index of the number of iterations of the neural network. Represented as the input feature of the i-th input layer neuron in the t-th iteration process, It is represented as the activation output of the jth neuron in the hidden layer during the tth iteration, and g is represented as the activation function. It is represented as the predicted output of the kth neuron in the output layer during the tth iteration, k represents the neuron index of the output layer, and W' kj It is represented as the weight from the jth neuron in the hidden layer to the kth neuron in the output layer, b' k It is expressed as the bias term of the kth neuron in the output layer. According to the error between the output and the expected output, the network weight is adjusted to optimize the network performance. The specific formula is expressed as: Among them, E represents the error function of the neural network, N represents the total number of samples, which is used to normalize the error to ensure that the errors under different sample numbers are comparable, and T represents the total number of iterations of the neural network. It is expressed as the expected target output in the tth iteration process, Represented as the predicted output during the tth iteration, it is the prediction made by the neural network based on the current weights and biases, Expressed as the square of the Euclidean norm, it is used to calculate the prediction error of the data sample during the tth iteration. It is represented as the updated weight from the i-th neuron in the input layer to the j-th neuron in the hidden layer during the t+1th iteration. It is represented by the weight value from the i-th neuron in the input layer to the j-th neuron in the hidden layer before updating in the t-th iteration, η is the learning rate, and w ij It is represented as the weight from the i-th input layer neuron to the j-th hidden layer neuron, Expressed as a partial derivative operation, It is represented as the updated bias term of the jth neuron in the corresponding hidden layer during the t+1th iteration. It is represented as the bias term before updating the jth neuron in the corresponding hidden layer in the tth iteration, b j It is expressed as the bias term of the jth hidden layer neuron. The network structure is adaptively and dynamically adjusted according to the network performance and the requirements of the learning task. The specific formula is expressed as: Among them, I j is represented as the importance index of the jth neuron, which is used to evaluate the role and importance of the neuron in the entire network. |·| is represented as the absolute value function, Θ j Represented as the decision variable of the jth neuron, according to the importance index I j With the preset threshold I thresh The comparison results determine the degree of neuron enhancement and weakening. thresh Represented as a preset importance threshold, Represented as the updated connection weight from the input layer i to the jth neuron, It is represented as the connection weight from the input layer i to the jth neuron before updating, R is represented as the random perturbation term, It is represented as the updated bias term corresponding to the jth neuron, It is represented as the bias term before updating corresponding to the jth neuron, r is represented as the random bias perturbation term, W represents the connection state from the i-th neuron to the j-th neuron, whether it is activated or disconnected after the update. thresh The connection threshold is expressed as a weight. By evaluating the importance of neurons and making decisions on adding or reducing neurons accordingly, the adaptive network topology algorithm based on neural networks can adaptively optimize the network structure, monitor network traffic, latency, and node status in real time, and make dynamic adjustments to cope with different learning tasks and performance requirements. The performance optimization end includes an AI performance optimization module, which proposes a task scheduling performance optimization algorithm based on swarm intelligence, and reasonably allocates computing resources according to the priority of the task and the availability of system resources to maximize resource utilization and minimize computing time, thereby ensuring that the system can intelligently schedule and optimize system computing tasks according to real-time task requirements and system resource conditions, thereby improving the overall performance and efficiency of the system; First, an initial task scheduling solution group X is established. In the initialization stage, the specific formula is expressed as: Among them, X (0) Represented as the initial population, which contains all individuals in the population, It is represented as the initial task scheduling plan of the i-th individual in the group, i is represented as the index of the individual in the group, and n is represented as the total number of individuals in the group. It is represented as the initial task scheduling plan of n individuals in the group. Then, the fitness value is calculated by the degree of satisfaction of the predetermined performance indicators based on the task scheduling plan of the individual. The specific formula is expressed as: F(x i )=w1·C makespan (x i )+w2·C resource (x i ) Among them, F(x i ) is expressed as the fitness value of the i-th individual, x i It is represented as the task scheduling plan of the i-th individual in the group, w1 is the weight coefficient related to the total time required to complete all tasks, and C makesoan (x i ) is represented by individual x i The completion time cost of the task scheduling scheme is expressed as w2, which is the weight coefficient related to the resource utilization rate when executing the task, C resource (x i ) is represented by individual x i The resource utilization cost of the task scheduling scheme is calculated. Secondly, by simulating the group behavior in nature, the individuals in the group are updated. The specific formula is expressed as: in, It represents the updated scheduling plan of individual i in the t+1th iteration, t represents the index of the number of algorithm iterations, It is represented as the current scheduling plan of individual i in the tth iteration process, It is represented as the optimal scheduling scheme for the entire group in the tth iteration process, φ is represented as the learning rate, which is used to control the step size of learning to the optimal solution, δ is represented as the perturbation factor, and rand(0,1) is represented as a random number in the interval [0,1]. Then, through selection and elite strategies, excellent individuals are retained and poor individuals are eliminated. The specific formula is expressed as follows: Among them, X (t+1) It is represented as the new group set in the t+1th iteration after selection, SelectTop k It means that k individuals with the highest fitness values ​​will be selected from the current population, and k represents the number of elite individuals selected to be retained. It is expressed as the fitness value of individual i in the tth iteration process, It is expressed as existence in mathematical operations. Finally, at the end of each generation of iteration, the preset termination condition is used to evaluate whether to continue executing the algorithm. The specific formula is expressed as: If(t≥MaxIter)or(F(x best )≥Threshold)then stop Among them, MaxIter represents the preset maximum number of iterations, F(x best ) represents the fitness value of the best individual in the current group, Threshold represents the preset fitness threshold, which is used to evaluate the satisfaction of the solution, and stop means that the algorithm will stop iterating and output the best solution currently found.

2. According to claim 1, a hyper-converged extended AI performance optimization system based on an adaptive network topology algorithm is characterized in that: The user interface end includes a control panel and a configuration tool module. The control panel and the configuration tool module display various indicators and performance data of the system through a visual interface, so that users can intuitively understand the system operation status and resource utilization, and ensure that users can easily adjust system parameters and function settings to meet the requirements of different needs and scenarios.

3. According to claim 1, a hyper-converged extended AI performance optimization system based on an adaptive network topology algorithm is characterized in that: The user interface end includes a user interaction and report generation module. The user interaction and report generation module enables users to interact with the system conveniently by providing a user-friendly interface, ensuring that users can understand the operation status and performance of the system in real time, thereby managing and optimizing the system, and helps users make adjustments and improvements by collecting operation data and performance indicators of various parts of the system to generate performance reports.

4. According to claim 1, a hyper-converged extended AI performance optimization system based on an adaptive network topology algorithm is characterized in that: The system management end includes a data transmission and storage module, which optimizes the data transmission mechanism and provides a data storage solution to meet the needs of efficient flow and secure storage of data within the system, ensure that the system can quickly respond to data read and write requests, and protect the integrity and security of the data.

5. According to claim 1, a hyper-converged extended AI performance optimization system based on an adaptive network topology algorithm is characterized in that: The system management end includes an access control and security maintenance module, which implements identity authentication and access control mechanisms so that only authorized users can access system resources and information, ensuring that the system can detect and respond to security incidents in a timely manner and protecting the security of the system and data.

6. According to claim 1, a hyper-converged extended AI performance optimization system based on an adaptive network topology algorithm is characterized in that: The performance optimization end includes a hyper-converged infrastructure module, which abstracts physical resources into virtual resources by managing and scheduling network resources within the system, and provides a unified resource pool for multiple network environments to achieve resource sharing and dynamic allocation, ensuring that the system can provide an efficient, flexible and scalable computing platform to meet the needs of various application scenarios.

Images