Graph-based complex network optimization scheduling system

Abstract

The invention relates to the technical field of optimization scheduling, in particular to a graph-based complex network optimization scheduling system, which comprises a dynamic multi-layer graph generation module for generating a multi-layer graph structure according to node states, edge states and association relationships among nodes in a complex network, reflecting dynamic change characteristics of the complex network in real time in the multi-layer graph structure, and generating a dynamic multi-layer graph structure; the multi-layer graph structure comprises a physical network layer, a logic network layer and a load layer; the non-linear relation capturing module is used for analyzing non-linear relations between nodes and edges of all layers in the multi-layer graph structure; and the cross-hierarchy network scheduling module is used for carrying out cross-hierarchy scheduling optimization based on the nonlinear analysis result of the multilayer graph structure provided by the nonlinear relation capturing module. According to the invention, the scheduling strategy can comprehensively consider different optimization targets of each level, thereby improving the utilization efficiency of the whole network resource, reducing the data transmission delay, and improving the load balance of the system.

Description

Technical Field

[0001] The present invention relates to the technical field of optimization scheduling, and in particular to a complex network optimization scheduling system based on a graph. Background Art

[0002] In a complex network environment, cross-level resource scheduling and optimization is one of the key challenges to improving network performance. Modern network structures usually include multiple layers such as the physical network layer, logical network layer, and load layer. Each layer has different resources and performance indicators. For example, the physical network layer focuses on bandwidth, latency, physical link reliability, etc. The logical network layer mainly deals with virtual paths and data flow transmission efficiency. The load layer involves node computing resource utilization and load balancing. However, in existing technologies, the scheduling and optimization of these three layers are usually independent of each other, which makes it difficult for the system to coordinate the performance goals of multiple layers globally and cannot achieve optimal resource allocation in a complex network environment.

[0003] Existing network optimization methods mostly rely on single-level optimization, such as optimizing bandwidth utilization at the physical network layer, or optimizing data transmission paths at the logical layer, but lack a cross-level comprehensive scheduling mechanism. When faced with a dynamic and complex network environment, these methods often fail to fully consider the interdependence and nonlinear relationships of each layer, resulting in a decline in overall performance, especially when dealing with large-scale networks or high-load scenarios. In addition, traditional scheduling mechanisms are mostly based on linear relationship assumptions and cannot capture the nonlinear interactive relationships hidden in the network, resulting in low scheduling efficiency and difficulty in solving bottleneck problems in a network environment with complex dependencies between nodes. Summary of the invention

[0004] The present invention provides a complex network optimization scheduling system based on graph.

[0005] Graph-based complex network optimization scheduling system, including:

[0006] Dynamic multi-layer graph generation module: Generate a multi-layer graph structure according to the node status, edge status and the relationship between nodes in the complex network, and reflect the dynamic change characteristics of the complex network in real time in the multi-layer graph structure. The multi-layer graph structure includes a physical network layer, a logical network layer and a load layer. The multi-layer graph structure uses the mathematical representation of the graph. Each layer of the graph structure is a graph G = (V, E), where V is a node set and E is an edge set. At the same time, in the multi-layer graph structure, each layer represents different attributes and characteristics in the complex network, using a multi-layer graph: G multi = {G phy ,G log ,G load} represents the set of all levels, where G phy is the graph representation of the physical network layer, G logis the graph representation of the logical network layer, G load It is a graph representation of the load layer, each layer contains its own set of nodes and edges; the layers are connected through the association matrix to achieve dynamic feature mapping in complex networks from a global perspective;

[0007] Nonlinear relationship capture module: used to analyze the nonlinear relationship between nodes and edges in each layer in a multi-layer graph structure. The nonlinear relationship capture module is based on an adaptive nonlinear mapping algorithm to dynamically capture the implicit interaction relationship between nodes and nodes, nodes and edges, and edges and edges in a complex network, and analyze the state transfer path of nodes at different levels in a multi-layer graph;

[0008] Cross-level network scheduling module: performs cross-level scheduling optimization based on the nonlinear analysis results of the multi-layer graph structure provided by the nonlinear relationship capture module.

[0009] Optionally, the dynamic multi-layer graph generation module specifically includes:

[0010] Node state acquisition submodule: real-time acquisition of multi-dimensional state characteristics S(v) of each node in a complex network i ), the multi-dimensional status characteristics include the operation status, load status and network resource occupancy of the physical device, wherein the node status of the physical network layer reflects the current utilization rate of the hardware resources of the physical device; the node status of the logical network layer reflects the activity level and data flow status of each virtual node or logical connection; the node status of the load layer describes the load level of each node, including the allocation and usage of computing resources;

[0011] Edge status monitoring submodule: collects edge status information in each layer of the network by monitoring the link status in the network. The status information includes the bandwidth, delay, and packet loss rate data of the link in the physical network layer. The edge status in the logical network layer reflects the usage frequency and traffic size of the virtual path. The edge status in the load layer is measured based on the load transfer cost between nodes.

[0012] Association calculation submodule: Based on the status data of nodes and edges, the association relationship between nodes is calculated, and the association strength between nodes is calculated using an algorithm based on graph theory. The association strength includes the communication frequency between devices in the physical network layer, the dependency of data flows in the logical network layer, and the cost of resource sharing or load migration between nodes in the load layer.

[0013] Multi-layer graph structure generation submodule: Dynamically generate a multi-layer graph structure based on node status, edge status and their associations. The graph generated by the physical network layer reflects the connection and data transmission path between actual physical devices; the graph generated by the logical network layer reflects the topological structure of the virtual network or application layer connection, including the logical relationship between each node; the load layer generates a graph structure based on resource usage, describing the load status of each node and the load balancing between nodes.

[0014] Optionally, the multi-layer graph structure is dynamically adjusted when the state information of network nodes and edges is updated in real time, reflecting the state change characteristics of the complex network in real time. When the state of a node or link changes, the multi-layer graph structure will update the original edge weight W(e ij ), and get the updated edge weight W′(e ij ), where e ij Represents node v i and node v j The edge between.

[0015] Optionally, the nonlinear relationship capture module is based on the multidimensional state feature S(v i ), an adaptive nonlinear mapping algorithm is used to construct adaptive nonlinear mapping models between nodes and nodes, nodes and edges, and edges and edges in complex networks. The implicit interaction relationship between nodes and edges is captured through the adaptive nonlinear mapping model; based on the implicit interaction relationship, the transmission path of node status in different levels is analyzed, and how the change of node status in the multi-layer graph affects other nodes through the edges is quantified.

[0016] Optionally, the adaptive nonlinear mapping model captures the implicit interaction relationship between nodes and nodes, nodes and edges, and edges and edges in the multi-layer graph through a graph convolutional network and an attention mechanism, and uses the graph convolutional network to monitor the nodes and their neighboring nodes v i Perform nonlinear aggregation to generate a new feature representation H(v i ), the attention weight in the attention mechanism is used to weight the influence of different neighbor nodes, so that the neighbor nodes with high correlation with the target node have a greater influence on it.

[0017] Optionally, the graph convolutional network performs a i The features are nonlinearly aggregated to generate a new feature representation H(v i ), expressed as:

[0018] Among them, H(v i ) is the node v i The new feature representation captures the relationship between the node and its neighbor node v j The nonlinear interaction relationship is node vi The set of neighbor nodes of v i There are connected nodes, A ij is the adjacency matrix, representing the node v i and v j The strength of the association between them (Relation), W is the weight matrix that needs to be learned, which is used to weight the features of different nodes, S(v j ) is the node v j The original state features, σ is a nonlinear activation function.

[0019] Optionally, the nonlinear relationship capture module obtains the node v based on the results of the graph convolution network and the attention mechanism analysis. i With neighbor node v j The implicit interaction relationship R(v i ,v j ), expressed as:

[0020] R(v i ,v j )=α ij ·H(v j ), where α ij is node v i and neighbor node v j The attention weight between j ) is the neighbor node v j The new feature representation of the node implicit interaction relationship is used for state transfer path analysis, specifically:

[0021] Among them, T(v i ,v j ) is the node v i To node v j The state transfer path of is node v i and node v j The set of all possible paths between ik It is also the attention weight, which represents the interaction strength between node vi and the intermediate node vk on the path. k ) is the node v k New feature representation.

[0022] Optionally, the cross-level network scheduling module obtains the cross-level transfer path of the node state between the physical network layer, the logical network layer and the load layer based on the state transfer path between nodes at different levels, identifies the key nodes and edges that have a great impact on the global network performance, and also includes:

[0023] Cross-layer weight mapping and scheduling submodule: Based on the results of state transfer path identification, a weight mapping mechanism is used to assign different scheduling weights to each node and edge, and dynamically assign scheduling priorities to nodes at each level;

[0024] Multi-objective scheduling optimization submodule: Design a multi-objective scheduling optimization function, comprehensively consider the optimization objectives of each level, generate the optimal cross-level scheduling plan, and ensure the optimal global performance.

[0025] Optionally, the cross-layer weight mapping and scheduling submodule includes calculating the cross-layer scheduling weight based on the identification result of the state transfer path: the calculation is expressed as: W(v i )=β phy ·T phy (v i )+β log ·T log (v i )+β load ·T load (v i ), where W(v i ) represents node v i The cross-level scheduling weight of β indicates the priority of the node in scheduling. phy ,β log ,β load They represent the scheduling weight coefficients of the physical network layer, logical network layer, and load layer, respectively, and are used to adjust the influence of each layer in the scheduling weight. phy (v i ) represents node v i The state transfer strength in the physical network layer, T log (v i ) represents node v i State transfer strength in the logical network layer, T load (v i ) represents node v i State transfer strength in the load layer.

[0026] Optionally, the optimization objectives of each level include: maximizing resource utilization of the physical network layer, optimizing data transmission efficiency of the logical network layer, and minimizing load imbalance of the load layer. The multi-objective scheduling optimization function is expressed as: Where L represents the comprehensive optimization loss function, represents the global scheduling optimization target, λ1, λ2, λ3 represent the weight coefficients of different optimization targets, which are used to adjust the weights of physical layer resource utilization, logical layer data transmission efficiency and load layer load balancing. i ) get, U phyIt represents the resource utilization of the physical network layer, which is defined as the ratio of the used bandwidth or computing resources to the total available resources. The goal is to maximize the resource utilization. log It represents the data transmission efficiency in the logical network layer and the transmission level of the logical connection in the network. The goal is to maximize the efficiency. load It represents the load imbalance degree of the load layer and the imbalance of the load distribution of nodes in the network. The goal is to minimize the load imbalance. This comprehensive optimization objective function achieves global cross-level scheduling optimization by adjusting the scheduling objectives of different levels.

[0027] Beneficial effects of the present invention:

[0028] The present invention, through a dynamic multi-layer graph generation module, reflects the state changes of different levels (physical network layer, logical network layer, load layer) in a complex network in real time to form a multi-layer graph structure. The system organically combines physical resources such as bandwidth and latency of the physical network layer, data transmission efficiency of the logical network layer, and node resource allocation and load conditions of the load layer, thereby realizing cross-level global state mapping. This multi-layer graph structure can timely capture key nodes and bottleneck links in the network, ensuring that the scheduling strategy can comprehensively consider the different optimization goals of each level, thereby improving the overall network resource utilization efficiency, reducing data transmission delay, and improving the system load balance.

[0029] The present invention introduces an adaptive nonlinear mapping algorithm, which can dynamically capture the nonlinear interaction relationships at different levels in complex networks through graph convolutional networks and multiple attention mechanisms, identify the implicit interaction relationships between nodes, nodes and edges, and edges and edges, and analyze the state transfer paths in multi-layer graphs. Through this nonlinear relationship capture mechanism, it can automatically identify the implicit critical paths and resource bottlenecks in the network, optimize cross-level scheduling strategies, and enable scheduling to allocate resources and adjust paths more accurately, avoiding complex network dependency problems that are difficult for traditional linear scheduling models to deal with.

[0030] In the present invention, the cross-level scheduling module is based on the weight calculation of the state transfer strength of each layer (physical layer, logical layer and load layer), and adopts a multi-objective optimization algorithm. It can dynamically adjust the scheduling strategy according to the real-time state of the system and the global performance goals. By combining the state transfer strength of each layer and the cross-level weight, the system can adaptively adjust the utilization of physical resources, the delay of data transmission and the balance of load, to ensure the optimal scheduling effect in different application scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings in the following description are only for the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0032] Figure 1 A schematic diagram of the functional modules of the scheduling system according to an embodiment of the present invention;

[0033] Figure 2 Schematic diagram of multi-objective scheduling optimization according to an embodiment of the present invention. DETAILED DESCRIPTION

[0034] The present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments. At the same time, it is explained here that in order to make the embodiments more detailed, the following embodiments are the best and preferred embodiments, and those skilled in the art may also adopt other alternatives to implement some known technologies; and the accompanying drawings are only for more specific description of the embodiments, and are not intended to specifically limit the present invention.

[0035] It should be noted that the references to "one embodiment", "an embodiment", "an exemplary embodiment", "some embodiments" and the like in the specification indicate that the embodiments described may include specific features, structures or characteristics, but not every embodiment may include the specific features, structures or characteristics. In addition, when a specific feature, structure or characteristic is described in conjunction with an embodiment, it should be within the knowledge of a person skilled in the art to implement such feature, structure or characteristic in conjunction with other embodiments (whether or not explicitly described).

[0036] In general, a term can be understood, at least in part, from its use in context. For example, depending, at least in part, on the context, the term "one or more" as used herein can be used to describe any feature, structure, or characteristic in the singular sense, or can be used to describe a combination of features, structures, or characteristics in the plural sense. Additionally, the term "based on" can be understood as not necessarily intended to convey an exclusive set of factors, but can instead, depending, at least in part, on the context, allow for the presence of other factors that are not necessarily explicitly described.

[0037] like Figure 1-Figure 2 As shown in the figure, the complex network optimization scheduling system based on graph includes:

[0038] Dynamic multi-layer graph generation module: Generate a multi-layer graph structure based on the node status, edge status and the relationship between nodes in the complex network, and reflect the dynamic change characteristics of the complex network in real time in the multi-layer graph structure. The multi-layer graph structure includes the physical network layer, the logical network layer and the load layer. The multi-layer graph structure uses the mathematical representation of the graph. Each layer of the graph structure is a graph G = (V, E), where V is the node set and E is the edge set. At the same time, in the multi-layer graph structure, each layer represents different attributes and characteristics in the complex network, using a multi-layer graph: G multl = {G phy ,G log ,G load} represents the set of all levels, where G phy is the graph representation of the physical network layer, G log is the graph representation of the logical network layer, G load is the graph representation of the load layer, each layer contains its own set of nodes and edges;

[0039] in;

[0040] Physical network layer: used to describe the connection relationship between each node and physical link in the network. The nodes in the physical network layer represent physical devices (servers, switches), and the edge weights reflect the bandwidth and latency properties of the physical links.

[0041] Logical network layer: describes the logical connection or virtual path structure based on the physical network. The edge weights in the logical network layer are adjusted based on the usage frequency, data transmission requirements, and load conditions of the logical connection.

[0042] Load layer: represents the load status of each node and link. The node weights in the load layer are adjusted based on resource utilization and node performance, and the edge weights take into account the overhead of load transfer between nodes.

[0043] Each layer is connected through the association matrix to achieve dynamic feature mapping in complex networks from a global perspective;

[0044] Nonlinear relationship capture module: used to analyze the nonlinear relationships between nodes and edges in each layer in a multi-layer graph structure. The nonlinear relationship capture module is based on an adaptive nonlinear mapping algorithm (such as a graph convolutional network algorithm and a multiple attention mechanism) to dynamically capture the implicit interaction relationships between nodes and nodes, nodes and edges, and edges and edges in a complex network, and analyze the state transfer paths of nodes at different levels in a multi-layer graph.

[0045] Cross-level network scheduling module: performs cross-level scheduling optimization based on the nonlinear analysis results of the multi-layer graph structure provided by the nonlinear relationship capture module.

[0046] In a multi-layer graph structure:

[0047] 1. Diagram representation of the physical network layer: Physical network layer G phy =(V phy ,E phy ) represents the connection relationship of physical devices, where V phy It is a node set of physical devices, such as servers, switches, etc. i ∈V phy , E phy is the edge set of physical links, representing the physical connection between devices. ij ∈E phy Connecting physical devices i and v j , whose weight W phy (e ij ) is defined as: W phy (e ij )=w bandwidth (e ij )+w latency (e ij )+w loss (e ij ), which takes into account the physical properties of bandwidth, latency, and packet loss rate.

[0048] 2. Diagram representation of the logical network layer: Logical network layer G log =(V log ,E log ) describes a virtual connection or a logical connection, indicating the communication relationship between the application layers. log It is a set of logical nodes, representing virtual devices or application layer entities. i ∈V log , E log It is a set of logically connected edges, representing the data flow transmission between virtual paths or application layers. ij ∈E log Connect to virtual node v i and v j , whose weight W log (e ij ) is defined as: W log (e ij )=w frequency (e ij )+w data (e ij )+w load (e ij ), where w frequency Indicates the frequency of use of logical connections, w data Indicates the amount of data transmitted, w load Indicates load conditions.

[0049] 3. Graphical representation of the load layer: Load layer Gload =(V load ,E load ) reflects the load status of network nodes and their resource allocation, where V load is a set of nodes representing the load of physical devices, node v i ∈V load , E load is the edge set of load migration, indicating the load transfer or resource allocation between nodes. ij ∈E load Connect node v i and v j , whose weight W load (e ij ) is defined as: W load (e ij )=W transfer (e ij )+w cost (e ij ), where w transfer represents the load migration amount, w cost Represents the migration cost.

[0050] In a multi-layer graph structure, the association between nodes is not only reflected in the graph of the same layer, but also in the association matrix between different layers. The cross-layer association matrix A is defined as cross , indicating the node association relationship between the physical network layer, logical network layer and load layer:

[0051]

[0052] Relation represents the strength of association or degree of relationship between two nodes. The overall graph across levels is represented as: G multi =({G phy ,G log ,G load},A cross ).

[0053] The dynamic multi-layer graph generation module specifically includes:

[0054] Node state acquisition submodule: real-time acquisition of multi-dimensional state characteristics S(v) of each node in a complex network i ), the multi-dimensional status characteristics include the running status, load status and network resource usage of physical devices. Among them, the node status of the physical network layer reflects the current usage rate of the hardware resources (such as CPU, memory, bandwidth) of the physical device; the node status of the logical network layer reflects the activity level and data flow status of each virtual node or logical connection; the node status of the load layer describes the load level of each node, including the allocation and usage of computing resources;

[0055] Edge status monitoring submodule: collects edge status information in each layer of the network by monitoring the link status in the network. The status information includes the bandwidth, delay, and packet loss rate data of the link in the physical network layer. The edge status in the logical network layer reflects the usage frequency and traffic size of the virtual path. The edge status in the load layer is measured based on the load transfer cost between nodes.

[0056] Association calculation submodule: Based on the status data of nodes and edges, the association between nodes is calculated, and the association strength between nodes is calculated using an algorithm based on graph theory. The association strength includes the communication frequency between devices in the physical network layer, the dependency of data flows in the logical network layer, and the cost of resource sharing or load migration between nodes in the load layer.

[0057] Multi-layer graph structure generation submodule: Dynamically generate a multi-layer graph structure based on node status, edge status and their associations. The graph generated by the physical network layer reflects the connection and data transmission path between actual physical devices; the graph generated by the logical network layer reflects the topological structure of the virtual network or application layer connection, including the logical relationship between each node; the load layer generates a graph structure based on resource usage, describing the load status of each node and the load balancing between nodes.

[0058] More specific:

[0059] 1. Node status collection: For each node v in the network i and its associated properties, using the state vector S(v i ) represents the state of the node. The node state vector contains the attributes of the physical network layer, logical network layer and load layer, which are specifically defined as: in:

[0060] s phy (v i ) is the node v in the physical network layer i The state indicates the resource utilization rate of the physical device, with a value range of [0,1] and is defined as: Among them, UsedResources(v i ) indicates the currently used resources of the physical device, TotalResources(v i ) represents the total available resources of the physical device;

[0061] s log (v i ) is the node v in the logical network layer i The state indicates the activity of the logical connection and the data flow status, which is defined as: Among them, ActiveConnections(v i) indicates the number of active logical connections of the current node, TotalConnections(v i ) represents the number of all logical connections of the node;

[0062] s load (v i ) is the node v in the load layer i The state indicates the load condition of the node and is defined as: Among them, UsedCapacity(v i ) indicates the used capacity of the current node, TotalCapacity(v i ) represents the total load capacity of the node.

[0063] 2. Edge status monitoring: For node v i and node v j The edge between ij , its state W(e ij ) represents the weight of a physical link or virtual connection and is defined as:

[0064] in:

[0065] w bandwidth (e ij ) represents node v i and v j The bandwidth utilization between is defined as:

[0066] Among them, UsedBandwidth(e ij ) indicates the bandwidth used by the current link, TotalBandwidth(e ij ) represents the total bandwidth of the link;

[0067] w latency (e ij ) represents node v i and v j The time delay between is defined as:

[0068] Among them, CurrentLatency(e ij ) represents the delay of the current link, MaxLatency(e ij ) represents the maximum allowed delay of the link;

[0069] w loss (e ij ) represents node v i and v j The packet loss rate between is defined as:

[0070] Among them, PacketLoss(e ij ) indicates the number of packet losses on the current link, TotalPackets(e ij ) represents the total number of packets passing through the link.

[0071] 3. Association relationship calculation: For node v i and v j The association relationship between them is calculated based on the weighted distance of the state vector, which is defined as:

[0072] Relation(v i ,v j )=α·d phy (v i ,v j )+β·d log (v i ,v j )+γ·d load (v i ,v j ),in:

[0073] d phy (v i ,v j ) represents the node v in the physical network layer i and v j The physical distance between them, representing the difference in node status, is defined as:

[0074] d phy (v i ,v j )=|s phy (v i )-s phy (v j )|;

[0075] d log (v i ,v j ) represents the node v in the logical network layer i and v j The state difference between is defined as:

[0076] d log (v i ,v j )=|s log (v i )-s log (v j )|;

[0077] d load (v i ,v j ) represents the node v in the load layeri and v j The load difference is defined as:

[0078] d load (v i ,v j )=|s load (v i )-s load (v j )|;

[0079] α, β, and γ are weighted coefficients of the association relationship at each layer, indicating the weight of the influence of the physical, logical, and load layers on the association relationship. The specific values ​​can be adjusted according to the actual situation.

[0080] The multi-layer graph structure is dynamically adjusted when the state information of network nodes and edges is updated in real time, reflecting the state change characteristics of complex networks in real time. When the state of a node or link changes (such as bandwidth bottleneck, node overload, etc.), the multi-layer graph structure will update the original edge weight W(e ij ), and get the updated edge weight W′(e ij ), where e ij Represents node v i and node v j The edges between them ensure that the scheduler obtains the latest network status data for resource optimization and scheduling adjustment.

[0081] When the state of a node or link in the network changes, the weights in the association relationship and graph structure will be updated in real time. ij The update weight calculation formula is: W′(e ij )=W(e ij )+δW(e ij ), where W′(e ij ) represents the updated edge weight, δW(e ij ) represents the edge weight change caused by state change, which is defined as: η is the weight update factor, ΔS(v i ) and ΔS(v j ) represent the nodes v i and v j The amount of change in state, specifically:

[0082]

[0083] In summary, through the state collection, association relationship calculation and weight dynamic update formulas at each level mentioned above, it is possible to achieve the generation of multi-layer graph structures and dynamic change reflection of complex networks.

[0084] The nonlinear relationship capture module is based on the multidimensional state characteristics S(vi ), an adaptive nonlinear mapping algorithm is used to construct adaptive nonlinear mapping models between nodes and nodes, nodes and edges, and edges and edges in complex networks. The implicit interaction relationship between nodes and edges is captured through the adaptive nonlinear mapping model; based on the implicit interaction relationship, the transmission path of node status in different levels is analyzed, and how the change of node status in the multi-layer graph affects other nodes through the edges is quantified.

[0085] The adaptive nonlinear mapping model uses graph convolutional networks and attention mechanisms to capture the implicit interactions between nodes and nodes, nodes and edges, and edges and edges in multi-layer graphs. The graph convolutional network is used to analyze the nodes and their neighbor nodes v i Perform nonlinear aggregation to generate a new feature representation H(v i ), the attention weights in the attention mechanism are used to weight the influence of different neighbor nodes, so that the neighbor nodes with high correlation with the target node have a greater influence on it. This process dynamically adjusts the weights of neighbor nodes according to the correlation between node features, thereby accurately capturing the nonlinear interaction relationship between nodes.

[0086] In this way, the complex implicit interaction relationships between nodes and nodes, nodes and edges, and edges and edges in multi-layer graph structures can be identified and captured. These implicit interaction relationships can be further used to analyze the transmission paths of node states in different levels, help identify the impact of cross-level state transmission on the overall network, and thus provide data support for global scheduling optimization.

[0087] The graph convolutional network is used to calculate the node v i The features are nonlinearly aggregated to generate a new feature representation H(v i ), expressed as:

[0088] Among them, H(v i ) is the node v i The new feature representation captures the relationship between the node and its neighbor node v j The nonlinear interaction relationship is node v i The set of neighbor nodes of v i There are connected nodes, A ij is the adjacency matrix, representing the node v i and v j The strength of the association between them (Relation), W is the weight matrix that needs to be learned, which is used to weight the features of different nodes, S(v j ) is the node v jis the original state feature, σ is a nonlinear activation function, defined as: σ(x) = max(0,x). By taking the weighted sum of the feature vectors of neighboring nodes and applying a nonlinear transformation, a new feature representation of each node can be obtained, reflecting the implicit interaction relationship between the nodes.

[0089] In order to improve the effect of nonlinear mapping, a multi-attention mechanism is used to weight the influence of different neighbor nodes. The formula is as follows:

[0090] Among them, α ij is node v i and neighbor node v j The attention weight between them reflects the importance of neighbor nodes to the target node. a is the weight vector in the attention mechanism, which is learned through training. W is the weight matrix, which is consistent with that in the graph convolutional network. || represents the feature concatenation operation. i and v j The feature vectors are concatenated, LeakyReLU(x) is the LeakyReLU activation function, defined as: Through the attention mechanism, the influence weights of different neighbor nodes on the target node can be dynamically adjusted to capture more accurate nonlinear interaction relationships between nodes.

[0091] The nonlinear relationship capture module is based on the analysis results of the graph convolutional network and the attention mechanism to obtain the node v i With neighbor node v j The implicit interaction relationship R(v i ,v j ), expressed as:

[0092] R(v i ,v j )=α ij ·H(v j ), where α ij is node v i and neighbor node v j The attention weight between j ) is the neighbor node v j The new feature representation of the node implicit interaction relationship is used for state transfer path analysis, specifically:

[0093] Among them, T(v i ,v j ) is the node v i To node v j The state transfer path of is node v i and node v j The set of all possible paths betweenik It is also the attention weight, which represents the interaction strength between node vi and the intermediate node vk on the path. k ) is the node v k New feature representation.

[0094] The cross-level network scheduling module obtains the cross-level transfer paths of node states between the physical network layer, logical network layer, and load layer based on the state transfer paths between nodes at different levels, identifies key nodes and edges that have a great impact on global network performance, and also includes:

[0095] Cross-layer weight mapping and scheduling submodule: Based on the results of state transfer path identification, a weight mapping mechanism is used to assign different scheduling weights to each node and edge, and dynamically assign scheduling priorities to nodes at each level. The resource constraints of the physical network layer, the data transmission requirements of the logical network layer, and the resource usage of the load layer will jointly determine the calculation of scheduling weights to ensure that the scheduling requirements between layers are reasonably balanced;

[0096] Multi-objective scheduling optimization submodule: Design a multi-objective scheduling optimization function, comprehensively consider the optimization objectives of each level, generate the optimal cross-level scheduling plan, and ensure the optimal global performance.

[0097] The cross-layer weight mapping and scheduling submodule includes the calculation of cross-layer scheduling weights based on the recognition results of the state transfer path: the calculation is expressed as: W(v i )=β phy ·T phy (v i )+β log ·T log (v i )+β log ·T log (v i ), where W(v i ) represents node v i The cross-level scheduling weight of β indicates the priority of the node in scheduling. phy ,β log ,β load Represent the scheduling weight coefficients of the physical network layer, logical network layer and load layer respectively. Although T phy (v i ), T log (v i ), T load (v i) already have different transmission strengths. In actual networks, sometimes different layers have different importances. For example, in some scenarios, the physical layer (resource utilization) may be more important than the logical layer (transmission delay); the load layer (load balancing) may require a higher priority in some cases. In these cases, β can be added according to specific network performance requirements. phy ,β log ,β load Set weights, set by yourself, to adjust the influence of each layer in the scheduling weight, T phy (v i ) represents node v i The state transfer strength in the physical network layer (the nonlinear interaction relationship and state transfer strength between node vi and its neighboring nodes and edges in the physical network layer) indicates the criticality of the node in the physical network and is calculated by the state transfer path formula, T log (v i ) represents node v i The state transfer strength in the logical network layer (the nonlinear interaction relationship and state transfer strength between node vi and its neighboring nodes and edges in the logical network layer) indicates the transmission demand and activity of the node in the logical layer, T load (v i ) represents node v i The state transfer strength in the load layer (the strength of the nonlinear interaction and state transfer between node vi and its neighboring nodes and edges in the load layer) indicates the load of the node and its impact on the overall load balance, T phy (v i ), T log (v i ), T load (v i ) are all through T(v i ,v j ) calculation formula. The state transfer strength of each node at different levels reflects the importance of the node in the network at that level. The global scheduling priority of node vi is obtained by weighted summing of these state transfer strengths T(vi).

[0098] The optimization objectives at each level include: maximizing resource utilization at the physical network layer, optimizing data transmission efficiency at the logical network layer, and minimizing load imbalance at the load layer. The multi-objective scheduling optimization function is expressed as: Where L represents the comprehensive optimization loss function, represents the global scheduling optimization target, λ1, λ2, λ3 represent the weight coefficients of different optimization targets, which are used to adjust the weights of physical layer resource utilization, logical layer data transmission efficiency and load layer load balancing. i ) get, Uphy It represents the resource utilization of the physical network layer, which is defined as the ratio of the used bandwidth or computing resources to the total available resources. The goal is to maximize the resource utilization. log It represents the data transmission efficiency in the logical network layer and the transmission level of the logical connection in the network. The goal is to maximize the efficiency. load It represents the load imbalance degree of the load layer and the imbalance of the load distribution of nodes in the network. The goal is to minimize the load imbalance. This comprehensive optimization objective function achieves global cross-level scheduling optimization by adjusting the scheduling objectives of different levels.

[0099] λ1,λ2,λ3 are based on W(v i ) is obtained, W(v i ) is the cross-level scheduling weight, which indicates the comprehensive importance of a node vi in ​​multiple levels. λ1,λ2,λ3 represent the priorities of different levels (physical network, logical network, load) in the global optimization objective. W(v i ) is the comprehensive weight of each node in each layer. By summarizing multiple nodes, the global priority of a certain layer (physical layer, logical layer, load layer) can be determined, and then the weight coefficients λ1, λ2, λ3 of each optimization target are determined by the weighted summary results of these nodes.

[0100] By solving this comprehensive optimization function L, we can obtain an optimal scheduling strategy that balances these three objectives. The specific performance of this strategy is as follows:

[0101] The resource allocation of the physical network layer will be based on the optimization results of Uphy to maximize the utilization of physical layer resources such as bandwidth and computing resources;

[0102] The data transmission path optimization in the logical network layer will select the transmission path with the least delay based on the minimization of Dlog. This strategy can dynamically adjust the connection or virtual path of the logical network layer to reduce transmission delay and ensure fast data delivery.

[0103] The optimization of the load layer minimizes Bload and balances the load of different nodes. Based on this strategy, tasks are reasonably allocated to prevent some nodes from being overloaded or idle, ensuring that the load in the network is evenly distributed, thereby improving the overall performance of the system.

[0104] The present invention covers any substitution, modification, equivalent method and scheme made on the essence and scope of the present invention. In order to make the public have a thorough understanding of the present invention, specific details are described in detail in the following preferred embodiments of the present invention, but those skilled in the art can fully understand the present invention without the description of these details. In addition, in order to avoid unnecessary confusion about the essence of the present invention, well-known methods, processes, procedures, components and circuits are not described in detail.

[0105] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A complex network optimization scheduling system based on graph, characterized by: include: Dynamic multi-layer graph generation module: Generate a multi-layer graph structure according to the node status, edge status and the relationship between nodes in the complex network, and reflect the dynamic change characteristics of the complex network in real time in the multi-layer graph structure. The multi-layer graph structure includes a physical network layer, a logical network layer and a load layer. Each layer of the graph structure in the multi-layer graph structure is represented as a graph G = (V, E), where V is a node set and E is an edge set. At the same time, in the multi-layer graph structure, each layer represents different attributes and characteristics in the complex network, using a multi-layer graph: G multi = {G phy ,G log ,G load } represents the set of all levels, where G phy is the graph representation of the physical network layer, G log is the graph representation of the logical network layer, G load It is a graph representation of the load layer, each layer contains its own set of nodes and edges; the layers are connected through the association matrix to achieve dynamic feature mapping in complex networks from a global perspective; Nonlinear relationship capture module: used to analyze the nonlinear relationship between nodes and edges in each layer in a multi-layer graph structure. The nonlinear relationship capture module is based on an adaptive nonlinear mapping algorithm to dynamically capture the implicit interaction relationship between nodes and nodes, nodes and edges, and edges and edges in a complex network, and analyze the state transfer path of nodes at different levels in a multi-layer graph; Cross-level network scheduling module: performs cross-level scheduling optimization based on the nonlinear analysis results of the multi-layer graph structure provided by the nonlinear relationship capture module.

2. The complex network optimization scheduling system based on graph according to claim 1, characterized in that: The dynamic multi-layer graph generation module specifically includes: Node state acquisition submodule: real-time acquisition of multi-dimensional state characteristics S(v) of each node in a complex network i ), the multi-dimensional status characteristics include the operation status, load status and network resource occupancy of the physical device, wherein the node status of the physical network layer reflects the current utilization rate of the hardware resources of the physical device; the node status of the logical network layer reflects the activity level and data flow status of each virtual node or logical connection; the node status of the load layer describes the load level of each node, including the allocation and usage of computing resources; Edge status monitoring submodule: collects edge status information in each layer of the network by monitoring the link status in the network. The status information includes the bandwidth, delay, and packet loss rate data of the link in the physical network layer. The edge status in the logical network layer reflects the usage frequency and traffic size of the virtual path. The edge status in the load layer is measured based on the load transfer cost between nodes. Association calculation submodule: Based on the status data of nodes and edges, the association relationship between nodes is calculated, and the association strength between nodes is calculated using an algorithm based on graph theory. The association strength includes the communication frequency between devices in the physical network layer, the dependency of data flows in the logical network layer, and the cost of resource sharing or load migration between nodes in the load layer. Multi-layer graph structure generation submodule: Dynamically generate a multi-layer graph structure based on node status, edge status and their associations. The graph generated by the physical network layer reflects the connection and data transmission path between actual physical devices; the graph generated by the logical network layer reflects the topological structure of the virtual network or application layer connection, including the logical relationship between each node; the load layer generates a graph structure based on resource usage, describing the load status of each node and the load balancing between nodes.

3. The complex network optimization scheduling system based on graph according to claim 2 is characterized in that: The multi-layer graph structure is dynamically adjusted when the state information of network nodes and edges is updated in real time, reflecting the state change characteristics of complex networks in real time. When the state of a node or link changes, the multi-layer graph structure will update the original edge weight W(e ij ), and get the updated edge weight W′(e ij ), where e ij Represents node v i and node v j The edge between.

4. The complex network optimization scheduling system based on graph according to claim 3 is characterized in that: The nonlinear relationship capture module is based on the multidimensional state characteristics S(v i ), an adaptive nonlinear mapping algorithm is used to construct adaptive nonlinear mapping models between nodes and nodes, nodes and edges, and edges and edges in complex networks. The implicit interaction relationship between nodes and edges is captured through the adaptive nonlinear mapping model; based on the implicit interaction relationship, the transmission path of node status in different levels is analyzed, and how the change of node status in the multi-layer graph affects other nodes through the edges is quantified.

5. The complex network optimization scheduling system based on graph according to claim 4 is characterized in that: The adaptive nonlinear mapping model captures the implicit interaction between nodes and nodes, nodes and edges, and edges and edges in the multi-layer graph through the graph convolution network and the attention mechanism. i Perform nonlinear aggregation to generate a new feature representation H(v i ), the attention weight in the attention mechanism is used to weight the influence of different neighbor nodes, so that the neighbor nodes with high correlation with the target node have a greater influence on it.

6. The complex network optimization scheduling system based on graph according to claim 5 is characterized in that: The graph convolutional network performs nonlinear aggregation on the features of node vi to generate a new feature representation H(v i ), expressed as: Among them, H(v i ) is the node v i The new feature representation captures the relationship between the node and its neighbor node v j The nonlinear interaction relationship is node v i The set of neighbor nodes of v i There are connected nodes, A ij is the adjacency matrix, representing the node v i and v j The strength of association between them, W is the weight matrix that needs to be learned, which is used to weight the features of different nodes, S(v j ) is the node v j The original state features, σ is a nonlinear activation function.

7. The complex network optimization scheduling system based on graph according to claim 6 is characterized in that: The nonlinear relationship capture module obtains the node v based on the results of graph convolutional network and attention mechanism analysis. i With neighbor node v j The implicit interaction relationship R(v i ,v j ), expressed as: R(v i ,v j )=α ij ·H(v j ), where α ij is node v i and neighbor node v j The attention weight between j ) is the neighbor node v j The new feature representation of the node implicit interaction relationship is used for state transfer path analysis, specifically: Among them, T(v i ,v j ) is the node v i To node v j The state transfer path of is node v i and node v j The set of all paths between ik It is also the attention weight, which represents the interaction strength between node vi and the intermediate node vk on the path. k ) is the node v k New feature representation.

8. The complex network optimization scheduling system based on graph according to claim 7 is characterized in that: The cross-level network scheduling module obtains the cross-level transfer path of the node state between the physical network layer, the logical network layer and the load layer based on the state transfer path between nodes at different levels, identifies the key nodes and edges that have a great impact on the global network performance, and also includes: Cross-layer weight mapping and scheduling submodule: Based on the results of state transfer path identification, a weight mapping mechanism is used to assign different scheduling weights to each node and edge, and dynamically assign scheduling priorities to nodes at each level; Multi-objective scheduling optimization submodule: Design a multi-objective scheduling optimization function, comprehensively consider the optimization objectives of each level, generate the optimal cross-level scheduling plan, and ensure the optimal global performance.

9. The complex network optimization scheduling system based on graph according to claim 8, characterized in that: The cross-layer weight mapping and scheduling submodule includes calculating the cross-layer scheduling weight based on the recognition result of the state transfer path: the calculation is expressed as: W(v i )=β phy ·T phy (v i )+β log ·T log (v i )+β load ·T load (v i ), where W(v i ) represents node v i The cross-level scheduling weight of β indicates the priority of the node in scheduling. phy ,β log ,β load They represent the scheduling weight coefficients of the physical network layer, logical network layer, and load layer, respectively, and are used to adjust the influence of each layer in the scheduling weight. phy (v i ) represents node v i The state transfer strength in the physical network layer, T log (v i ) represents node v i State transfer strength in the logical network layer, T load (v i ) represents node v i State transfer strength in the load layer.

10. The complex network optimization scheduling system based on graph according to claim 8, characterized in that: The optimization objectives of each level include: maximizing resource utilization at the physical network layer, optimizing data transmission efficiency at the logical network layer, and minimizing load imbalance at the load layer. The multi-objective scheduling optimization function is expressed as: Where L represents the comprehensive optimization loss function, represents the global scheduling optimization target, λ1, λ2, λ3 represent the weight coefficients of different optimization targets, which are used to adjust the weights of physical layer resource utilization, logical layer data transmission efficiency and load layer load balancing. i ) get, U phy It represents the resource utilization of the physical network layer, which is defined as the ratio of the used bandwidth or computing resources to the total available resources. The goal is to maximize the resource utilization. log It represents the data transmission efficiency in the logical network layer and the transmission level of the logical connection in the network. The goal is to maximize the efficiency. load It represents the load imbalance degree of the load layer, which indicates the imbalance of the load distribution of nodes in the network. The goal is to minimize the load imbalance.

Images