Heterogeneous computing network task allocation and path optimization system based on neural network

Abstract

The application relates to the technical field of network task allocation, and discloses a heterogeneous computing power network task allocation and path optimization system based on a neural network, which comprises a graph construction module, a node task allocation module and a path optimization module.The graph construction module is used for constructing a traffic dispersion graph, modeling network topology and node data through the traffic dispersion graph, and modeling nodes and edges in combination with graph indexing and graph displacement, and extracting network communication traffic and node features.The node task allocation module is used for inputting graph indexes of all nodes into an improved graph neural network model to intelligently predict and allocate node tasks, and outputting node task allocation prediction results.The path optimization module is used for obtaining path optimization results by minimizing transmission delay and balancing node computing power in combination with node task prediction results.The system optimization module is used for optimizing the overall performance of the system based on the path optimization results and the task allocation prediction results.The application improves the task allocation and path optimization efficiency of the heterogeneous computing power network.

Description

Technical Field

[0001] This invention relates to the field of network task allocation technology, and in particular to a heterogeneous computing power network task allocation and path optimization system based on neural networks. Background Technology

[0002] With the rapid development of information technology, computing networks play a crucial role in many fields, such as cloud computing, edge computing, and big data processing. In heterogeneous computing networks, different types of computing nodes (such as CPUs, GPUs, and FPGAs) have their own unique performance characteristics, and the network topology is also complex and diverse.

[0003] Currently, task allocation and path optimization face numerous challenges in heterogeneous computing networks. Traditional task allocation and path planning methods often fail to fully consider real-time performance changes of network nodes and complex topologies, leading to problems such as uneven task distribution, high transmission latency, and low resource utilization. For example, in large-scale data processing tasks, unreasonable task allocation may overload some nodes while leaving other nodes idle, affecting overall processing efficiency. In applications with high real-time requirements (such as autonomous driving and telemedicine), high transmission latency may cause untimely system response, resulting in serious consequences.

[0004] Neural networks excel at handling complex data relationships, but traditional neural networks also have limitations when applied to task allocation and path optimization in heterogeneous computing networks. These limitations include insufficient processing capabilities for graph-structured data and a tendency for oversmoothing as depth increases, making it difficult to effectively address dynamic changes and complex topologies in heterogeneous computing networks. Therefore, a new technical solution is urgently needed to address these issues and improve the efficiency of task allocation and path optimization in heterogeneous computing networks. Summary of the Invention

[0005] This invention aims to solve the technical problems of uneven task allocation, high transmission latency, and low resource utilization in heterogeneous computing networks by traditional task allocation and path optimization, and proposes a task allocation and path optimization system for heterogeneous computing networks based on neural networks.

[0006] The objective of this invention can be achieved through the following technical solutions:

[0007] This invention provides a heterogeneous computing power network task allocation and path optimization system based on neural networks, comprising:

[0008] The graph construction module is used to construct a traffic dispersion graph from captured network data streams, model the network topology and node data using the traffic dispersion graph, and model nodes and edges by combining graph indexes and graph displacements, and extract network communication traffic and node features.

[0009] The node task allocation module is used to input the graph index set of each node into the improved graph neural network model, intelligently predict and allocate node tasks, and output the task allocation prediction results of the nodes.

[0010] The path optimization module combines the node task prediction results to select the optimal data transmission path in order to minimize transmission delay and balance node computing power, thus obtaining the path optimization result.

[0011] The system optimization module optimizes the overall system performance based on the path optimization results and the task allocation prediction results.

[0012] In some preferred embodiments, the graph construction module includes:

[0013] The network data stream capture unit is used to collect performance data of heterogeneous computing power nodes in real time, including communication traffic, throughput and path information;

[0014] Individual flow graph construction unit is used to construct the graph structure of network nodes and communication paths, where nodes represent hosts or computing power nodes in the network, edges represent communication flow paths between nodes, and edge weights represent data flow throughput between nodes.

[0015] The traffic dispersion map generation unit is used to merge multiple individual flow graphs to generate a traffic dispersion map representing the overall network topology, including: using an adjacency matrix to represent the connection relationship between nodes; and identifying abnormal nodes based on the traffic threshold of throughput nodes.

[0016] In some preferred embodiments, the graph construction module further includes:

[0017] Graph index and graph displacement construction units are used to construct graph indexes and graph displacements; where...

[0018] The graph index includes computing power, load level, and throughput, which are used to represent the attribute information of nodes;

[0019] The graph displacement is used to realize the translation operation of the graph index on the graph. The weighted matrix Laplace is selected as the graph displacement to weight and combine the node signal with its neighboring nodes, extract the structural features of the graph, and finally realize the local smoothing of the graph index.

[0020] In some preferred embodiments, the graph construction module further includes:

[0021] The graph Fourier transform unit is used to perform graph Laplace eigenvalue decomposition on the weighted matrix Laplace, extract the frequency domain features of the graph index, and then project the graph index onto the frequency domain of the graph through graph Fourier transform to analyze the global structure and transmission characteristics between nodes.

[0022] In some preferred embodiments, the improved graph neural network model introduces identity mapping and residual connections, and updates node features through multiple iterations to finally obtain the task assignment prediction results for the nodes.

[0023] In some preferred embodiments, the node feature H (l) The update formula is as follows:

[0024]

[0025] Among them, H (l) : The node feature matrix of the l-th layer; H (0) =X: Initial node feature matrix of the input; μ: Identity mapping coefficient, used to balance the initial features H (0) The influence of; ρ: residual connectivity coefficient, used to introduce the feature H of the previous layer. (l-1) The impact; The normalized adjacency matrix of the graph.

[0026]

[0027] Where A: original adjacency matrix; I: identity matrix; D: degree matrix of the graph;

[0028] σ: Activation function, expressed as:

[0029]

[0030] Where x' is an input variable; θ1∈(0,∞) and

[0031] In some preferred embodiments, the node task allocation module further includes a unit that defines a node task allocation objective function to measure node load balancing and resource utilization, with objectives including maximizing overall computing efficiency, load balancing, and minimizing network communication overhead.

[0032] In some preferred embodiments, the path optimization module is used to set a path optimization objective function, which consists of a transmission delay term from any node to other nodes and a load balancing weight term for any node. The load balancing weight term for any node includes the task prediction result of any node, the current task load of any node, and the computing power information of any node. The objective function aims to minimize transmission delay and balance node computing power load, ultimately selecting the optimal data transmission path.

[0033] In some preferred embodiments, the path optimization module further performs path optimization by incorporating graph Fourier transform domain frequency characteristics to find the path with the best load balancing performance, including:

[0034] Input node load balancing weights and transmission delays;

[0035] Perform a graph Fourier transform on the load balancing weights of the nodes;

[0036] Filter out high-frequency components and retain low-frequency components;

[0037] The inverse transformation is performed back to the node domain to obtain the smoothed node load balancing weights.

[0038] The optimal path is determined based on the updated node load balancing weights and transmission delays.

[0039] In some preferred embodiments, the system optimization module constructs a loss function for task scheduling and system optimization, which includes a transmission delay loss term and a load balancing loss term, to minimize the sum of delays on the task transmission path and the variance of node load, respectively.

[0040] The system optimization module also includes: using gradient descent to iteratively update the weight parameters in order to optimize the overall system performance.

[0041] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0042] (1) This invention, through an improved graph neural network model, can more accurately predict and allocate node tasks, avoiding the problem of uneven task allocation in traditional methods; by combining physical information and graph index, the system can analyze the performance data of heterogeneous computing power nodes in real time, thereby making more reasonable task allocation decisions and improving task allocation efficiency.

[0043] (2) Through the path optimization module, this invention can select the optimal data transmission path and minimize transmission latency, especially in applications with high real-time requirements, such as autonomous driving and telemedicine, effectively reducing transmission latency. The application of graph Fourier transform enables the system to analyze the global structure and transmission characteristics between nodes, further optimizing path selection.

[0044] (3) By balancing node load, this invention avoids situations where some nodes are overloaded while other nodes are idle, thereby improving the overall resource utilization of the computing network. By constructing a loss function for task scheduling and system optimization, the variance of transmission delay and node load can be minimized, further improving resource utilization.

[0045] (4) By introducing identity mapping and residual connection, this invention solves the oversmoothing problem that occurs when the depth of traditional graph neural networks increases, thereby enhancing the robustness and stability of the system.

[0046] In summary, this invention significantly improves the efficiency of task allocation and path optimization in heterogeneous computing networks, reduces transmission latency, enhances resource utilization, and strengthens the robustness and stability of the system. Attached Figure Description

[0047] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the description of the embodiments taken in conjunction with the following drawings, in which:

[0048] Figure 1 This is a schematic diagram of a task allocation and path optimization system based on a neural network provided in an embodiment of the present invention.

[0049] Figure 2 This is a schematic diagram of the unit of the graph construction module 101 provided in an embodiment of the present invention;

[0050] Figure 3 This is a flowchart illustrating a method for task allocation and path optimization based on a neural network in a heterogeneous computing power network, as provided in an embodiment of the present invention. Detailed Implementation

[0051] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0052] In the description of this invention, it should be understood that the orientation descriptions, such as up, down, front, back, left, right, etc., are based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limiting this invention.

[0053] In the description of this invention, "several" means one or more, "more than" means two or more, "greater than," "less than," and "exceeding" are understood to exclude the stated number, while "above," "below," and "within" are understood to include the stated number. The use of "first" and "second" in the description is merely for distinguishing technical features and should not be construed as indicating or implying relative importance, or implicitly indicating the number of indicated technical features, or implicitly indicating the order of the indicated technical features.

[0054] Figure 1 This is a schematic diagram of a heterogeneous computing power network task allocation and path optimization system based on neural networks, provided in an embodiment of the present invention. Figure 3 This is a flowchart illustrating a method for task allocation and path optimization based on a neural network in heterogeneous computing power, as provided in an embodiment of the present invention. Please refer to... Figure 1 and Figure 3 As shown, the present invention provides a heterogeneous computing power network task allocation and path optimization system 100 based on neural networks, comprising:

[0055] Graph construction module 101 is used to construct a traffic dispersion map through captured network data streams, model the network topology and node data through the traffic dispersion map, and model nodes and edges by combining graph index and graph displacement, and extract network communication traffic and node features.

[0056] This is the first step of the system: modeling the network topology and node data using a traffic dispersion map, and modeling nodes and edges using graph indexes and graph displacements to extract network communication traffic and node features.

[0057] (1) Graph construction

[0058] The graph construction module 101 includes:

[0059] The network data stream capture unit 1011 is used to collect performance data of heterogeneous computing power nodes in real time, including but not limited to communication traffic, throughput and path information.

[0060] By monitoring network data packets, the following information can be extracted:

[0061] Source IP address: The IP address of the node that sent the data packet sTc .

[0062] Destination IP address: IP address of the node receiving the data packet dst .

[0063] Path information: Path P of data packet transmission src→dst .

[0064] Throughput T: The amount of data transmitted per unit of time.

[0065] Individual flow graph construction unit 1012 is used to construct a graph structure of network nodes and communication paths, where nodes represent hosts or computing power nodes in the network, edges represent communication flow paths between nodes, and edge weights represent data flow throughput between nodes.

[0066] Individual flow graph construction: Construct a graph structure G = (V, E, W) of network nodes and communication paths.

[0067] The communication traffic of data packets is modeled as a graph structure, where:

[0068] Node: A host or computing node in a network.

[0069] Edge: The communication traffic path between nodes.

[0070] Edge weight: the data flow throughput T between nodes ij .

[0071] Specifically represented as Figure G:

[0072] G = (V, E, W)

[0073] V: Node set (all host or computing power nodes); E: Edge set (communication paths between nodes); W: Edge weight set, w ij =T ij This represents the throughput between node i and node j.

[0074] The traffic dispersion map generation unit 1013 is used to merge multiple individual flow maps to generate a traffic dispersion map representing the overall network topology, including: using an adjacency matrix to represent the connection relationship between nodes, and identifying abnormal nodes based on the flow threshold of throughput nodes.

[0075] Traffic Dispersion Map Generation: Merge multiple individual flow maps generated by individual flow map building unit 1012 to generate a traffic dispersion map, which is used to represent the overall network topology.

[0076] Adjacency matrix A: Represents the connection relationships between nodes. If there is a data flow between nodes i and j, then:

[0077]

[0078] Abnormal node detection: based on throughput T ij and the flow threshold T of the node th Identify anomalous nodes. The anomalousness of node i is defined as:

[0079]

[0080] in, The set of neighboring nodes of node i; T ij : Actual throughput of nodes i and j; T th Normal traffic threshold.

[0081] If D(i)>∈(the set abnormal threshold), then node i is determined to be abnormal.

[0082] The above-mentioned abnormal node detection mechanism enables the system to identify and handle abnormal situations in the network in a timely manner, ensuring the normal operation of the system.

[0083] (2) Constructing graph indexes and graph shifts

[0084] In some preferred embodiments, the graph construction module 101 further includes:

[0085] Graph index and graph displacement construction unit 1014 is used to construct graph indexes and graph displacements; wherein...

[0086] A graph index is a vector set of indices from graph nodes, representing the attribute information of the nodes. In this system, the graph index x of node i is... i =[C i ,L i ,T i ],include:

[0087] Computational ability C i : The computing resources (CPU / GPU performance) of node i.

[0088] Load level L i : The workload of the task currently being executed by node i.

[0089] Throughput T i : Data transmission throughput of node i (in bps).

[0090] Graph displacement is used to implement the translation operation of graph index on the graph. The weighted matrix Laplace is selected as the graph displacement to perform weighted combination of node signals with their neighboring nodes, extract the structural features of the graph, and finally achieve local smoothing of the graph index.

[0091] Defined as:

[0092] Laplace=DA

[0093] in:

[0094] D: The degree matrix of the graph, which is a diagonal matrix. ii =∑ j A ij .

[0095] A: The adjacency matrix of the graph.

[0096] When the graph displacement Laplace is applied to the graph index x, it is represented as:

[0097] Laplacex=(DA)x

[0098] This operation achieves local smoothing of the graph index by weighting the node signal with its neighboring nodes to extract the structural features of the graph.

[0099] In some preferred embodiments, the graph construction module 101 further includes:

[0100] The graph Fourier transform unit 1015 is used to perform graph Laplace eigenvalue decomposition on the weighted matrix Laplace, extract the frequency domain features of the graph index, and then project the graph index onto the frequency domain of the graph through graph Fourier transform to analyze the global structure and transmission characteristics between nodes.

[0101] (3) Graph Fourier Transform

[0102] Perform graph Laplace eigenvalue decomposition on the weighted matrix Laplace to extract the frequency domain features of the graph index:

[0103] Laplace=UΛU T

[0104] in:

[0105] U: The eigenvector matrix of the weighted matrix Laplace.

[0106] Λ: The diagonal matrix of the eigenvalues ​​of the weighted matrix Laplace.

[0107] The graph Fourier transform projects the graph index x onto the frequency domain of the graph to analyze the global structure and transmission characteristics between nodes. The graph Fourier transform of the graph index x is expressed as:

[0108]

[0109] The representation of the graph index x in the graph Fourier domain (frequency component), i.e., the frequency characteristics of the graph Fourier transform domain.

[0110] U T : The transpose of the eigenvectors of the graph Laplacian matrix.

[0111] The graph construction module 101 models the network topology and node data using a traffic dispersion map, and models nodes and edges using graph indexes and graph displacements, enabling comprehensive extraction of network communication traffic and node characteristics. An abnormal node detection mechanism allows the system to promptly identify and handle anomalies in the network, ensuring the accuracy and reliability of the data.

[0112] The node task allocation module 102 is used to input the graph index set of each node into the improved graph neural network model, perform intelligent prediction and allocation of node tasks, and output the task allocation prediction results of the nodes.

[0113] (1) Improved graph neural network model

[0114] To address the oversmoothing problem that occurs as the depth of traditional graph neural networks increases: 1) The improved graph neural network model first introduces an identity mapping, which preserves the initial features H of the nodes through the coefficient μ. (0)Even with a deeper network, initial information can still propagate effectively, preventing excessive feature smoothing. 2) The improved graph neural network model also introduces residual connections, which introduce features H from the previous layer through coefficient ρ. (l-1) This allows information to propagate in leaps between layers, mitigating the gradient vanishing problem.

[0115] An improved graph neural network model: By introducing identity mapping and residual connections, and updating node features through multi-layer iterations, the task assignment prediction results of the nodes are finally obtained.

[0116] Node features H are updated through multi-level iterations. (l) Finally, the task allocation prediction results for the nodes are obtained. The core objective of task allocation is to minimize the unevenness of node load through network features H, while making full use of the computing resources of the nodes.

[0117] Among them, node features H (l) The update formula is as follows:

[0118]

[0119] Among them, H (l) : The node feature matrix of the l-th layer; H (0) =X: Initial node feature matrix of the input; μ: Identity mapping coefficient, used to balance the initial features H (0) The influence of; ρ: residual connectivity coefficient, used to introduce the feature H of the previous layer. (l-1) The impact; The normalized adjacency matrix of the graph, through the normalized adjacency matrix For node features H (l-1) Aggregation is performed to effectively integrate information about nodes and their neighbors. Defined as:

[0120]

[0121] Where A: original adjacency matrix; I: identity matrix (with self-loops to ensure the participation of node self-information); D: degree matrix of the graph;

[0122] σ: Activation function, expressed as:

[0123]

[0124] Where x' is an input variable; θ1∈(0,∞) and

[0125] (2) Define the objective function for node task allocation

[0126] The node task allocation module 102 also includes a unit that defines the objective function for node task allocation, which is used to measure node load balancing and resource utilization. The objectives include maximizing overall computing efficiency, load balancing, and minimizing network communication overhead.

[0127] The objective function J is defined as follows:

[0128]

[0129] Where J is the comprehensive cost function (which needs to be minimized).

[0130] N: Total number of nodes.

[0131] C i The computing power of node i (such as the computing power of CPU or GPU).

[0132] L i : The current task load of node i.

[0133] T ij : Communication throughput between node i and node j.

[0134] d ij The communication overhead (such as latency) between nodes i and j is the transmission delay from node i to node j.

[0135] The average load across all nodes in the network.

[0136] ΔL i : Predict the new task load to be assigned to node i.

[0137] α, β, γ, δ: Weighting coefficients used to balance the importance of computational efficiency, communication overhead, load balancing, and resource utilization.

[0138] ∈0: A very small positive number to avoid a denominator of zero.

[0139] By training an improved graph neural network model and minimizing J, the node task assignment prediction result ΔL is obtained. i This refers to the new task load on each node. Ensure that node tasks are allocated reasonably to avoid overload and resource waste.

[0140] The node task allocation module 102 inputs a graph index set, composed of the graph indices of each node, into an improved graph neural network model, enabling intelligent prediction and allocation of node tasks. A node task allocation objective function is defined to measure node load balancing and resource utilization, ensuring the rationality and effectiveness of task allocation.

[0141] The path optimization module 103, in conjunction with the node task prediction results obtained by the node task allocation module 102, selects the optimal data transmission path to minimize transmission delay and balance node computing power, thereby obtaining the path optimization result.

[0142] Specifically, in the path optimization phase, the obtained node task prediction results ΔL are combined. i The goal is to select the optimal data transmission path, minimize transmission latency, balance node computing power, and achieve efficient data scheduling and path planning.

[0143] The path optimization module 103 is used to set a path optimization objective function, which consists of a transmission delay term from any node to other nodes and a load balancing weight term for any node. The load balancing weight term for any node includes the task prediction result of any node, the current task load of any node, and the computing power information of any node. The objective function minimizes the transmission delay and balances the computing power load of nodes, and finally selects the optimal data transmission path.

[0144] (1) Set path optimization goals

[0145] Task load and node computing power information together determine the availability and load of each node in the network. Path optimization aims to minimize transmission latency while considering node computing power load balancing. Through path selection, the objective function is constructed by minimizing transmission latency and load imbalance. as follows:

[0146]

[0147] λ: Weighting parameter, which adjusts the importance of transmission delay and load balancing.

[0148] P: The set of nodes along the data transmission path.

[0149] W i The load balancing weight of node i is expressed as:

[0150]

[0151] (2) Path optimization based on the frequency characteristics of the Fourier transform domain of the graph.

[0152] In the graph Fourier transform domain, low-frequency components reflect smooth changes in node load balancing weights, indicating good overall load balance. High-frequency components reflect drastic changes in node load balancing weights, indicating that some nodes in the network are overloaded or have severe transmission bottlenecks. Therefore, it is necessary to filter high-frequency components and retain low-frequency components to find the path with the best load balancing. Based on the graph frequency domain information, high-load nodes or transmission bottleneck nodes are eliminated.

[0153] The steps for path optimization based on the frequency characteristics of the graph Fourier transform domain include:

[0154] 2.1 Input node load balancing weight W and transmission delay d ij The node load balancing weight vector W represents the ratio of task load to computing power of each node in the network. The higher the W, the busier or more resource-limited the node is. In path selection, we should try to avoid selecting too many of these nodes.

[0155] 2.2 Perform a graph Fourier transform on the load balancing weights of the nodes:

[0156]

[0157] Here, U is the eigenvector matrix of the graph Laplacian matrix, used to project the node weights W into the graph Fourier frequency domain.

[0158] 2.3 Retain the low-frequency components after the Fourier transform of the graph and filter out the high-frequency components, i.e., remove the high-load bottleneck point. The signal... In the frequency domain, it is represented as Each of them This is the component corresponding to a certain frequency k. We choose a filter function H(k) to suppress high-frequency components in the frequency domain, and calculate the filtered frequency signal through element-wise multiplication:

[0159]

[0160] in, It is a parameter that controls the filtering strength.

[0161] 2.4 Perform an inverse transformation back to the node domain to obtain the smoothed node load balancing weight W. ' :

[0162]

[0163] 2.5 Based on the updated node load balancing weight W ' and transmission delay d ij Solve for the optimal path P * :

[0164]

[0165] Where, arg min P Its function is to find path P * To minimize the objective function, dynamic programming can be used to obtain the optimal output path P by recursively calculating the weights of the segmented paths. * It can ensure low transmission delay while also taking into account the balance of node weight distribution.

[0166] The path optimization module 103 combines the node task prediction results to select the optimal data transmission path to minimize transmission delay and balance node computing power. By analyzing the global structure and transmission characteristics between nodes through graph Fourier transform, the path selection is further optimized, thereby improving data transmission efficiency.

[0167] The system optimization module 104 optimizes the overall system performance based on the path optimization results and task allocation prediction results.

[0168] (1) Constructing the loss function for task scheduling and system optimization

[0169] The system optimization module 104 constructs a loss function for task scheduling and system optimization, aiming to minimize transmission latency and balance the computing load of nodes. This loss function includes a transmission latency loss term and a load balancing loss term. The specific loss function... This can be expressed as a weighted sum of the two parts:

[0170]

[0171] in: Overall loss function; Transmission delay loss term; Load balancing loss term; γ1 and γ2: weighting coefficients in the loss function, used to balance the effects of the two parts.

[0172] Among them, transmission delay loss item This is the time required for the task to be transmitted from the source node to the target node. Assume the task is transmitted via path P, and the delay from node i to node j is d. ij Transmission delay loss term Defined as:

[0173]

[0174] P: The optimal path for task transmission.

[0175] d ij : Transmission delay between node i and node j.

[0176] The goal of this loss term is to minimize the total latency along the task transmission path.

[0177] Load balancing loss items The goal is to distribute the task load more evenly across all nodes, avoiding overload on any single node. Load balancing loss term. Defined as:

[0178]

[0179] W i: Load balancing weight of node i.

[0180] The average node weight, representing the overall average load level of the system, is defined as:

[0181]

[0182] N: The set of all nodes in the network.

[0183] The goal of this loss term is to minimize the variance of node load, so that the load is evenly distributed throughout the network.

[0184] (2) Weight parameter λ j Adjustments:

[0185] In path optimization and task scheduling, the weight parameter λ j Adjusting these parameters is crucial for improving system accuracy. Gradient descent is used to iteratively update the weight parameters, with the following update formula:

[0186]

[0187] in:

[0188] The weight parameters for the (t+1)th iteration.

[0189] η: Learning rate, which controls the update step size.

[0190] The partial derivative of the loss function with respect to the weight parameters.

[0191] The system optimization module, through the aforementioned loss function for task scheduling and system optimization, minimizes transmission latency and node load variance, further improving overall system performance. Gradient descent is used to iteratively update weight parameters to optimize overall system performance and ensure system stability and reliability in various application scenarios.

[0192] The above is an introduction to the system embodiments. The following method embodiments will further illustrate the solution of the present invention.

[0193] Another embodiment of the present invention provides a dynamic demand computing network fusion scheduling optimization method based on an improved large model, comprising:

[0194] A traffic dispersion map is constructed by capturing network data streams. The network topology and node data are modeled using the traffic dispersion map. Nodes and edges are modeled by combining graph indexes and graph displacements, and network communication traffic and node features are extracted.

[0195] The graph indexes of each node are combined into a graph index set and input into an improved graph neural network model to intelligently predict and allocate tasks to nodes, and output the task allocation prediction results of nodes.

[0196] By combining the node task prediction results, the optimal data transmission path is selected to minimize transmission latency and balance node computing power, thus obtaining the path optimization result.

[0197] Based on the path optimization results and the task allocation prediction results, the overall system performance is optimized.

[0198] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the described method steps can be referred to the corresponding process in the foregoing system embodiments, and will not be repeated here.

[0199] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the claims of this invention should still fall within the scope of the claims.

[0200] It should be understood that the various forms of processes shown above can be used to reorder, add, or delete steps. For example, the steps described in this invention can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution disclosed in this invention can be achieved, and this is not limited herein.

[0201] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention.

Claims

1. A neural network based heterogeneous computing power network task allocation and path optimization system, characterized in that, Comprise: A graph construction module for constructing a traffic dispersion graph from captured network data streams, modeling network topology and node data through the traffic dispersion graph, and modeling nodes and edges in combination with graph index and graph displacement, extracting network communication traffic and node features; A node task allocation module for inputting the graph index of each node into an improved graph neural network model to intelligently predict and allocate node tasks, and outputting node task allocation prediction results; the improved graph neural network model updates node features through multiple iterations by introducing identity mapping and residual connection, and finally obtains node task allocation prediction results; A path optimization module that selects the optimal data transmission path based on the node task prediction results to minimize transmission delay and balance node computing power, and obtains path optimization results; A system optimization module that optimizes the overall performance of the system based on the path optimization results and the task allocation prediction results.

2. The neural network based heterogeneous computational power network task allocation and path optimization system of claim 1, wherein, Wherein, The graph construction module comprises: A network data stream capture unit for collecting performance data of heterogeneous computing power nodes in real time, including communication traffic, throughput, and path information; An individual flow graph construction unit for constructing a graph structure of network nodes and communication paths, wherein nodes represent hosts or computing power nodes in the network, edges represent communication traffic paths between nodes, and edge weights represent data flow throughput between nodes; A traffic dispersion graph generation unit for merging multiple individual flow graphs to generate a traffic dispersion graph representing the overall topology of the network, including: using an adjacency matrix to represent the connection relationship between nodes; identifying abnormal nodes based on the traffic threshold of the throughput node.

3. The improved neural network based heterogeneous computing power network task allocation and path optimization system according to claim 1, characterized in that, Wherein, The graph construction module further comprises: A graph index and graph displacement construction unit for constructing a graph index and a graph displacement; wherein, The graph index includes computing power, load level, and throughput, and is used to represent the attribute information of the node; The graph displacement is used to realize the translation operation of the graph index on the graph, and a weighted matrix is selected As the graph displacement, the node signal is combined with its neighbor nodes by weighting, the structural features of the graph are extracted, and finally the local smoothing of the graph index is realized.

4. The neural network based heterogeneous computational power network task allocation and path optimization system of claim 3, wherein, Wherein, The graph construction module further comprises: a graph fourier transform unit for applying a weighting matrix graph laplacian eigen decomposition is performed to extract the frequency domain features of the graph index, and then the graph index is projected into the frequency domain of the graph through the graph fourier transform to analyze the global structure and transmission characteristics between nodes.

5. The neural network-based heterogeneous compute network task allocation and path optimization system of claim 1, wherein, Wherein, The improved graph neural network model updates node features through multiple iterations by introducing identity mapping and residual connection, and finally obtains node task allocation prediction results.

6. The neural network based heterogeneous computational effort network task allocation and path optimization system of claim 5, wherein, Wherein, Node characteristics The update formula for is as follows: ； wherein, : the first : the node feature matrix of the layer; : the input initial node feature matrix; : identity mapping coefficients for balancing the influence of the initial features ; : residual connection coefficients for introducing the influence of the features of the previous layer; : the normalized adjacency matrix of the graph, ； wherein, : original adjacency matrix; : identity matrix; : degree matrix of a graph; : activation function, expressed as: ； wherein is an input variable; and .

7. The neural network-based heterogeneous compute network task allocation and path optimization system of claim 6, wherein, Wherein, The node task allocation module further comprises a unit for defining a node task allocation objective function for measuring node load balancing and resource utilization, including maximizing overall computing efficiency, load balancing, and minimizing network communication overhead.

8. The neural network-based heterogeneous computational effort network task allocation and path optimization system of claim 1, wherein, Wherein, The path optimization module is used to set a path optimization objective function composed of transmission delay items from any node to other nodes and load balancing weight items of the any node, wherein the load balancing weight items of the any node include the any node task prediction results, the any node current task load, and the any node computing power information; the path optimization module minimizes transmission delay and balances node computing power load to finally select the optimal data transmission path.

9. The neural network-based heterogeneous computational effort network task allocation and path optimization system of claim 8, wherein, Wherein, The path optimization module further optimizes the path in combination with the graph Fourier transform domain frequency characteristics to find the path with the best load balancing, including: Inputting node load balancing weight and transmission delay; Performing graph Fourier transform on the node load balancing weight; Filtering out high-frequency components and retaining low-frequency components; inverse transform back to the node domain, to get the smoothed node load balancing weight; According to the updated node load balancing weight and the transmission delay, the optimal path is solved.

10. The neural network-based heterogeneous computational effort network task allocation and path optimization system of claim 9, wherein, The system optimization module builds a loss function of task scheduling and system optimization, which includes a transmission delay loss term and a load balancing loss term, for minimizing the total delay on the task transmission path and the variance of node load, respectively. The system optimization module further comprises: using gradient descent method to iteratively update the weight parameters to optimize the overall performance of the system.

Images