A self-adaptive routing decision method for a fusion communication network based on distributed agents

Abstract

The application discloses a kind of based on distributed Agent's fusion communication network adaptive routing decision method, comprising: constructing terminal Agent layer and routing node Agent layer, terminal Agent receives the power grid terminal initiated power grid service, and generates QoS constraint according to service category;Routing node Agent real-time acquisition link state, and generate network topology graph by interacting with adjacent routing node Agent;Based on multiple network topology graphs and QoS constraint generates candidate path set, and filters optimal path from candidate path set, carries out the data transmission of power grid service.The application constructs layered distributed Agent architecture, and routing node Agent is based on local topology and service QoS constraint independently completes path decision, without central controller, eliminates single point failure, reduces signaling overhead, while being able to realize differentiated service accurate guarantee;And the local topology of routing node Agent can real-time perceive link change, to adapt to complex power grid environment.

Description

Technical Field

[0001] This invention belongs to the field of complex power grid converged communication networks, specifically relating to an adaptive routing decision method for converged communication networks based on distributed agents. Background Technology

[0002] With the rapid advancement of the construction of new power systems, the massive access of power grid terminal equipment and the continuous diversification of business scenarios, and the fact that transmission lines cross complex terrains such as deserts, mountains, and densely populated urban areas, a single communication network cannot meet the needs of full-scenario coverage and differentiated business requirements. Heterogeneous converged communication networks composed of technologies such as power fiber optics, self-organizing networks, and satellite communication have become the core development direction of power communication systems.

[0003] Routing decisions are a core technology in converged communication networks, directly determining the reliability, latency, bandwidth, and security of data transmission, thereby affecting the safe and stable operation of core services such as online monitoring of transmission lines and collection of electricity consumption information. Complex power grid environments impose stringent and specific requirements on routing decisions.

[0004] Currently, routing solutions for converged communication networks fall into three categories: traditional static / semi-dynamic routing protocols (such as RIP and OSPF) cannot detect dynamic changes in heterogeneous links in real time, have slow fault convergence (on the order of seconds), cannot adapt to differentiated QoS, and are prone to bandwidth constraints for critical services; centralized routing decisions based on SDN suffer from high signaling overhead and decision latency in wide-area, massive terminal scenarios, and single-point failures of the controller can lead to network-wide paralysis, making it difficult to achieve cross-heterogeneous network collaboration; heuristic routing optimization based on intelligent algorithms is computationally complex, difficult to deploy on low-computing-power nodes, often involves local optimization on a single node, lacks distributed collaboration, is prone to global load imbalance, relies on offline training, and is insufficiently adaptable to sudden electromagnetic interference and link failures, with weak self-healing capabilities. All three types of solutions are ill-suited to the complex environment and differentiated service requirements of new power systems.

[0005] Therefore, there is an urgent need to design a routing decision method for complex power grid scenarios, based on a distributed architecture, and with adaptive capabilities for converged communication networks, to solve the core problems of poor adaptability, slow convergence, insufficient reliability, and difficulty in heterogeneous network collaboration in existing technologies. Summary of the Invention

[0006] To address the aforementioned problems in existing technologies, this invention provides an adaptive routing decision method for converged communication networks based on distributed agents. The technical problem to be solved by this invention is achieved through the following technical solution: An adaptive routing decision method for converged communication networks based on distributed agents includes: Construct the terminal agent layer and the routing node agent layer; The terminal agent layer includes multiple terminal agents, which are used to receive power grid services initiated by power grid terminals and generate a first constraint set based on the power grid services. The first constraint set includes thresholds corresponding to multiple QoS indicators. The routing node agent layer includes multiple routing node agents deployed on communication nodes. Each routing node agent is used to collect link status data of multiple links with a communication node as the endpoint in real time. The link status data includes multiple QoS indicators. The routing node agent generates a network topology map based on first link status data, which includes the link status data collected in real time by the routing node agent and its adjacent multiple routing node agents. When the terminal Agent sends a routing request, it generates a candidate path set based on multiple network topology maps and the first constraint set, and filters the first path from the candidate path set to obtain the first path. The candidate path set includes multiple routing paths corresponding to the routing request. Data transmission of the power grid services is performed based on the first path.

[0007] In one embodiment of the present invention, after constructing the terminal agent layer and the routing node agent layer, the method further includes: A regional control agent layer is constructed, which includes multiple regional control agents deployed on communication control nodes. One regional control agent is used to receive the network topology map transmitted by multiple first routing node agents and construct a subnet network topology map to obtain the subnet network topology map. The first routing node agents are the routing node agents in a first preset area. One first preset area corresponds to one regional control agent and multiple routing node agents. The first preset area is divided based on regional subnets within the entire network. The regional management agent performs route coordination optimization, network resource scheduling, load balancing management, and fault event aggregation and reporting within the first preset region based on the network topology map of its corresponding subnet. The area management agents corresponding to adjacent subnets coordinate cross-subnet routing by exchanging boundary routing information.

[0008] In one embodiment of the present invention, after constructing the region control agent layer, the method further includes: A global decision agent layer is constructed, including a global decision agent deployed at the central node of the communication network, which is used to receive the subnet network topology map reported by each of the regional control agents, and construct a unified network topology map based on multiple subnet network topology maps to obtain the unified network topology map; The global decision agent performs network-wide routing strategy formulation, global network resource coordination, cross-regional routing collaborative optimization, global network anomaly handling, and routing algorithm model iterative optimization based on the unified network topology map.

[0009] In one embodiment of the present invention, generating the first constraint set based on the power grid service includes: The terminal agent classifies the power grid services to obtain the categories of the power grid services, and quantifies the thresholds of multiple QoS indicators of the power grid services based on the categories of the power grid services to obtain the first constraint set. The categories include production control, operation monitoring and information collection.

[0010] In one embodiment of the present invention, generating a candidate path set based on a plurality of network topology graphs and the first constraint set includes: A first link set is generated based on the first constraint set. The first link set includes multiple first links. Each QoS indicator of the first link satisfies the threshold corresponding to the QoS indicator in the first constraint set. Multiple routing paths are generated in the first link set based on the depth-first search algorithm and the pruning algorithm to obtain the candidate path set. The routing path is a loop-free reachable path that starts from the source node of the routing request, ends at the destination node, and is located in the topology formed by the first link set.

[0011] In one embodiment of the present invention, the step of filtering a first path from the candidate path set includes: Obtain a set of routing path metrics, which includes multiple standardized QoS metrics for each second path, wherein the second path is the routing path in the candidate routing path set; Obtain a first weight set, which includes multiple first weights, and one first weight corresponds to a QoS indicator of the second path. A standardized decision matrix is ​​constructed based on the routing path indicator set, wherein the column vectors of the standardized decision matrix consist of multiple QoS indicators of a second path; A weighted standardized decision matrix is ​​generated based on the standardized decision matrix and the first weight vector, and the first weight vector is generated based on the first weight set. Calculate the Euclidean distance between each column vector of the weighted standardized decision matrix and a preset first vector to obtain a first distance vector. Calculate the Euclidean distance between each column vector of the weighted standardized decision matrix and a preset second vector to obtain a second distance vector. Both the first and second vectors are vectors composed of multiple preset QoS indicators. One of the Euclidean distances in the first and second distance vectors corresponds to one of the column vectors. Multiple proximity scores are calculated based on the first distance vector and the second distance vector, and each proximity score corresponds to a second path. The first path is obtained by taking the second path with the highest proximity as the first path.

[0012] In one embodiment of the present invention, the calculation of multiple proximity scores based on the first distance vector and the second distance vector can be expressed as follows: ; in, The proximity of the j-th second path, and These are the Euclidean distances between the column vector corresponding to the j-th second path and the first and second vectors, respectively.

[0013] In one embodiment of the present invention, obtaining the first weight set includes: Obtain a service priority set, which includes the priorities of multiple QoS indicators of the power grid service; Based on the service priority set and the hierarchical analysis method, multiple demand weights are calculated to obtain multiple demand weights, and each demand weight corresponds to a QoS indicator. Based on each standardized QoS metric of each second path, multiple entropy weights are calculated to obtain multiple entropy weights, and each entropy weight corresponds to a QoS metric. The first weight is calculated based on the entropy weight and the demand weight corresponding to each QoS indicator, resulting in multiple first weights.

[0014] In one embodiment of the present invention, the calculation of multiple entropy weights based on each standardized QoS metric of each second path includes: Calculate the first weight of the first QoS indicator to obtain the first weight. The first weight is the ratio of the normalized value of the first QoS indicator to the sum of the normalized values ​​of all second QoS indicators. The first QoS indicator is one of the two QoS indicators. The second QoS indicator is the QoS indicator of the second path. The normalization of the second QoS indicator is calculated based on the QoS indicators of all second paths in the candidate route path set. The entropy value of the first QoS indicator is calculated based on the first weight of the first QoS indicator and the preset number of alternative links, and the entropy value is obtained. The difference coefficient is calculated based on the entropy value of the first QoS indicator to obtain the difference coefficient of the first QoS indicator; Calculate the difference coefficients of all the second QoS indicators and sum them to obtain the difference coefficient sum; The entropy weight of the first QoS indicator is obtained by calculating the difference coefficient of the first QoS indicator and the difference coefficient.

[0015] In one embodiment of the present invention, after obtaining the first path, the method further includes: Based on the proximity of each second path, multiple second paths are selected from the candidate routing path set as backup paths, and the backup paths are used to replace the first path for data transmission of the power grid service.

[0016] The beneficial effects of this invention are: This solution constructs a hierarchical distributed agent architecture. Routing node agents collect link status data in real time and interact with neighboring routing node agents, enabling rapid detection of dynamic changes in heterogeneous links and updates to the local topology. This eliminates the need for a central controller to generate candidate paths and select the optimal path, mitigating the single-point-of-failure risk of centralized solutions and significantly reducing network-wide signaling overhead. Simultaneously, terminal agents carry constraints containing multiple QoS metrics when initiating routing requests. Routing node agents use these constraints as the core basis for link filtering and path matching, ensuring differentiated quality of service for different power grid services. Furthermore, routing node agents generate candidate paths and make decisions based on real-time topology and constraint sets, enabling rapid response and adaptive route adjustments to link fluctuations in complex power grid environments.

[0017] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0018] Figure 1This is a flowchart illustrating an adaptive routing decision method for converged communication networks based on distributed agents, provided in an embodiment of the present invention. Figure 2 This is a schematic diagram of the layered distributed agent architecture provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of the process of the agent-based adaptive routing decision method provided in the embodiments of the present invention. Detailed Implementation

[0019] The present invention will be further described in detail below with reference to specific embodiments, but the implementation of the present invention is not limited thereto.

[0020] Example 1 It should be noted that routing decision-making is a core technology of converged communication networks, directly determining the reliability, latency, bandwidth, and security of data transmission, thereby affecting the safe and stable operation of core services such as online monitoring of transmission lines and collection of electricity consumption information. Complex power grid environments impose stringent specific requirements on routing decision-making, specifically manifested in the following ways: 1. The complex terrain and environment, with sand dunes, mountains and buildings blocking signal propagation, cause violent dynamic fluctuations in link status and high failure rate, requiring routing decisions to have extremely strong dynamic adaptability and resilience. 2. The power grid has a strong electromagnetic environment. Electromagnetic interference generated by the operation of high-voltage transmission lines and substation equipment can easily cause the error rate of communication links to soar, requiring routing decisions to have strong anti-interference and rapid avoidance capabilities. 3. Power grid services are highly differentiated. Production control services require millisecond-level low latency and high reliability, operation monitoring services require high bandwidth and high stability, and information collection services require large connections and low power consumption. Traditional single-rule routing cannot meet the differentiated QoS requirements. 4. Converged communication networks are highly heterogeneous, with significant differences in link characteristics, coverage, and transmission capabilities among different communication technologies. Cross-network routing coordination is challenging, requiring routing decisions to achieve unified management and resource coordination across multiple network types. 5. The power grid terminals are widely distributed and numerous, with some terminals located in unattended areas. The network topology changes frequently, requiring routing decisions to have the characteristics of being distributed, scalable, and having low operation and maintenance costs.

[0021] Currently, the routing decision-making technologies used in complex power grid converged communication networks are ill-suited to the complex environment and differentiated business needs of new power systems. Specifically: 1. Poor environmental adaptability, unable to cope with complex dynamic link environments in power grids. Traditional static / semi-dynamic routing protocol schemes are the mainstream application schemes in current power grid communication networks. They are based on traditional routing protocols such as RIP (Routing Information Protocol), OSPF (Open Shortest Path First), and BGP (Border Gateway Protocol), and perform route calculations based on fixed metrics (hop count, bandwidth, link cost), using static configuration or semi-dynamic update modes. This scheme has a simple architecture and strong compatibility, and is widely used in stable communication scenarios such as power grid fiber optic backbone networks and substation LANs. However, in complex wide-area converged communication scenarios in power grids, it cannot perceive the dynamic changes of heterogeneous links in real time. When a link fails, the route convergence speed is slow (seconds), which easily leads to data packet loss and excessive latency. Furthermore, it cannot adapt to the QoS requirements of differentiated services in the power grid. Different services share the same routing rules, which can easily lead to the problem of bandwidth being squeezed for critical services.

[0022] Furthermore, traditional static / semi-dynamic routing protocols are based on fixed metrics and static configurations, which cannot detect dynamic changes in heterogeneous network links (electromagnetic interference, terrain obstruction, link failure) in real time. When the link state changes abruptly, the route convergence is slow, and data packet loss and latency exceed the standard are likely to occur, which cannot meet the high reliability requirements of power grid production control business. At the same time, the adaptability to different scenarios of complex power grids is weak, and the routing strategy cannot be adjusted for the characteristics of scenarios such as desert, mountainous areas, and areas with strong electromagnetic interference.

[0023] 2. Centralized decision-making schemes suffer from poor scalability and insufficient reliability. SDN (Software-Defined Networking)-based centralized routing decision-making schemes rely on the SDN architecture, using a central controller to globally collect network status information, centrally complete route calculations and policy distribution, and achieve unified management and control of the entire network's routing. Some schemes combine network slicing technology to achieve differentiated carrying of power grid services. This scheme solves the problem of high difficulty in distributed management and control of traditional protocols and has been piloted in regional scenarios such as urban distribution networks and industrial park power grids. However, in scenarios with complex power grid wide-area coverage and massive terminal access, it has significant drawbacks: the signaling interaction overhead between the central controller and nodes is high, and the decision-making latency increases significantly as the network scales up; the controller is at risk of single-point failure, and once the controller fails, the entire network routing is paralyzed, failing to meet the high security and high availability requirements of power grid communication; at the same time, it is difficult to achieve distributed routing coordination across heterogeneous networks, and the response to dynamic changes in links is lagging.

[0024] Furthermore, SDN-based centralized routing decision-making schemes rely entirely on the global management of a central controller. In power grid scenarios with wide coverage and massive terminal access, the signaling interaction overhead is large, the decision latency is high, and the controller is at risk of single point of failure. Once the central node fails, the entire network routing will be paralyzed, which cannot meet the high security and high availability requirements of power grid communication. At the same time, it is difficult to achieve distributed routing coordination across heterogeneous networks, and the utilization rate of redundant resources in multiple networks is low.

[0025] 3. Intelligent routing solutions suffer from poor feasibility and insufficient collaborative capabilities. Heuristic routing optimization schemes based on intelligent algorithms incorporate genetic algorithms, ant colony algorithms, and machine learning, using link state parameters as input and the optimal route path as output to achieve dynamic route optimization. Some schemes have undergone lightweight algorithm modifications for power grid scenarios. While these schemes possess some adaptive optimization capabilities and can achieve route optimization under multiple constraints, they exhibit significant shortcomings in practical power grid implementation: high algorithm computational complexity, high computational requirements for routing nodes, and difficulty in deployment on numerous low-power, low-computing-power terminals and routing nodes within the power grid; many schemes involve single-node local optimization without building a distributed collaborative architecture, easily leading to local route optimization but global network load imbalance; the algorithm's decision-making logic relies on offline training, resulting in insufficient adaptability to scenarios such as sudden electromagnetic interference and link failures in the power grid, and weak anti-interference and fault self-healing capabilities.

[0026] Furthermore, existing routing optimization schemes based on intelligent algorithms have high computational complexity and high requirements for node computing power and storage, making them difficult to deploy on a large number of low-power terminals and edge routing nodes in the power grid. At the same time, most of these schemes are single-node optimizations and do not introduce a distributed multi-agent collaborative architecture, which makes it impossible to achieve global collaborative optimization of routing across nodes and heterogeneous networks. This can easily lead to local optima and global load imbalance, and the ability to respond to sudden failures and interference is insufficient.

[0027] 4. Weak ability to guarantee differentiated services and high difficulty in heterogeneous network coordination. Existing solutions mostly employ routing decision logic with fixed weights and fixed rules, which cannot be adapted to the differentiated QoS requirements of different power grid services. The low latency and high reliability requirements of critical production control services cannot be rigidly guaranteed. At the same time, existing solutions are mostly designed for single-type communication networks and have not been deeply optimized for the heterogeneous characteristics of the power grid's converged communication networks. They cannot achieve unified routing decisions and dynamic switching across multiple network types, routing information is not shared between different networks, and redundant resources of multiple networks cannot be fully utilized.

[0028] Therefore, this embodiment designs an adaptive routing decision method for converged communication networks based on distributed agents. The core of the method is to construct a hierarchical distributed multi-agent architecture to achieve distributed perception of network status, hierarchical collaborative decision-making of routing strategies, and rapid self-healing of link faults. This solves the core problems of poor adaptability, slow convergence, insufficient reliability, and difficulty in heterogeneous network collaboration in existing technologies.

[0029] For details, see Figure 1 An adaptive routing decision method for converged communication networks based on distributed agents, comprising: Construct the terminal agent layer and the routing node agent layer; The terminal agent layer includes multiple terminal agents, which are used to receive power grid services initiated by power grid terminals and generate a first constraint set based on the power grid services. The first constraint set includes thresholds for multiple QoS indicators. The routing node agent layer includes multiple routing node agents deployed on communication nodes. Each routing node agent is used to collect link status data of multiple links with a communication node as the endpoint in real time. The link status data includes multiple QoS indicators. The routing node agent generates a network topology map based on first link status data, which includes the link status data collected in real time by the routing node agent and its adjacent multiple routing node agents. When the terminal Agent sends a routing request, it generates a candidate path set based on multiple network topology maps and the first constraint set, and filters the first path from the candidate path set to obtain the first path. The candidate path set includes multiple routing paths corresponding to the routing request. Data transmission of the power grid services is performed based on the first path.

[0030] Furthermore, after constructing the terminal agent layer and the routing node agent layer, the method further includes: A regional control agent layer is constructed, which includes multiple regional control agents deployed on communication control nodes. One regional control agent is used to receive the network topology map transmitted by multiple first routing node agents and construct a subnet network topology map to obtain the subnet network topology map. The first routing node agents are the routing node agents in a first preset area. One first preset area corresponds to one regional control agent and multiple routing node agents. The first preset area is divided based on regional subnets within the entire network. The regional management agent performs route coordination optimization, network resource scheduling, load balancing management, and fault event aggregation and reporting within the first preset region based on the network topology map of its corresponding subnet. The area management agents corresponding to adjacent subnets coordinate cross-subnet routing by exchanging boundary routing information.

[0031] Furthermore, after constructing the regional control agent layer, it also includes: A global decision agent layer is constructed, including a global decision agent deployed at the central node of the communication network, which is used to receive the subnet network topology map reported by each of the regional control agents, and construct a unified network topology map based on multiple subnet network topology maps to obtain the unified network topology map; The global decision agent performs network-wide routing strategy formulation, global network resource coordination, cross-regional routing collaborative optimization, global network anomaly handling, and routing algorithm model iterative optimization based on the unified network topology map.

[0032] Understandably, in combination Figure 2 As shown, this embodiment constructs a four-layer architecture: a terminal agent layer, a routing node agent layer, a regional control agent layer, and a global decision agent layer. Each agent layer operates independently and autonomously, while simultaneously achieving vertical and horizontal collaboration, adapting to the harsh environment and business requirements of complex power grid converged communication networks. Furthermore, the agent-based adaptive routing decision method in this embodiment can accurately match business needs with network resources, generate the optimal routing path, and achieve dynamic closed-loop optimization. The specific design of the layered distributed agent architecture is as follows: The terminal agent layer includes multiple terminal agents deployed on various terminal devices in the power grid. Its core functions are identifying service QoS requirements, collecting terminal status data, initiating routing requests, and providing feedback on path execution.

[0033] Each terminal agent autonomously performs the following functions: The system categorizes and assigns QoS labels to power grid services initiated by terminals. First, it distinguishes between three major service categories: production control, operation monitoring, and information collection. Then, it uniformly quantifies QoS indicators such as latency, packet loss rate, bandwidth, and reliability according to each category. Each service category corresponds to a set of QoS indicator constraints, including thresholds for multiple QoS indicators. The system collects the terminal's computing power, power consumption, and communication module status. It then initiates routing requests to the corresponding routing node (Agent), reporting service QoS requirements and terminal status. Finally, it executes the assigned routing path, monitors data transmission status in real time, and provides feedback on transmission performance to the routing node (Agent).

[0034] Furthermore, the generation of the first constraint set based on the power grid service includes: The terminal agent classifies the power grid services to obtain the categories of the power grid services, and quantifies the thresholds of multiple QoS indicators of the power grid services based on the categories of the power grid services to obtain the first constraint set. The categories include production control, operation monitoring and information collection.

[0035] Understandably, production control applications, such as remote circuit breaker control, stability control system commands, and relay protection signals, have QoS quantification characteristics, i.e., threshold constraints, requiring extremely low packet loss rate and extremely high reliability. Bandwidth is typically low, but guaranteed. For operation monitoring applications, such as transmission line status monitoring, substation video surveillance, and drone inspection feedback, bandwidth requirements are high (e.g., ≥2Mbps to 10Mbps), latency requirements are moderate (e.g., ≤200ms to 1s), packet loss rate tolerance is moderate (e.g., ≤1%), and reliability requirements are high but not extreme. For information collection applications, such as user electricity consumption information collection, distribution automation data reporting, and environmental sensor data, the number of connections is massive, power consumption is low, latency is not sensitive (seconds or even minutes are sufficient), bandwidth is generally narrowband (e.g., a few kbps), and there is a certain tolerance for packet loss, which can be compensated for by retransmission.

[0036] Classifying the QoS indicator characteristics of three types of power grid services enables precise matching of service requirements and network resources. Since different services have vastly different network performance requirements, classification quantifies the abstract "service requirements" into specific QoS thresholds, forming a set of QoS constraints. This allows routing decisions to move away from a "one-size-fits-all" best-effort approach and instead allocate link resources on demand. Simultaneously, it provides input for differentiated routing algorithms. The classified QoS constraints directly serve as hard filtering conditions during subsequent candidate path generation (eliminating links that do not meet requirements for latency, packet loss rate, etc.), and also as the basis for subjective weighting in multi-attribute decision-making (higher weights are given to latency and reliability for production control, and higher weights for bandwidth for operation monitoring). Furthermore, it enables layered service quality assurance. In converged communication networks, different types of services can be mapped to different transmission priorities, queue scheduling strategies, and even different underlying physical networks (e.g., production control services prioritize fiber optics or dedicated 5G slices, while information collection services can use low-power self-organizing networks), achieving refined operation of network resources.

[0037] The routing node agent layer includes multiple routing node agents, which are deployed on the communication nodes of the converged communication network. It is the core execution layer for routing decisions, and its core functions are real-time perception of heterogeneous link status, information exchange between adjacent nodes, rapid local route calculation, self-healing of link faults, and monitoring of route execution.

[0038] Each routing node Agent autonomously performs the following functions: The system collects five types of parameters in real time: operational status, transmission quality, resource load, channel quality, and link reliability of heterogeneous communication links under its jurisdiction. Specifically, these parameters include: end-to-end latency, available bandwidth, packet loss rate, bit error rate, link load rate, signal-to-noise ratio, signal received strength, channel occupancy, link connectivity, link stability, interference level, remaining link capacity, transmission power, modulation scheme, and link fault alarm status. It interacts with adjacent routing node agents in real time to exchange link status information and construct a local network topology view. For routing requests from terminal agents, it quickly generates candidate routing paths based on local link status. When the primary link fails or its performance deteriorates, it autonomously triggers local route self-healing and switches to a backup link. It monitors the transmission effect of routing paths in real time and reports link status, routing execution status, and fault information to the regional control agents.

[0039] Furthermore, the regional control agent layer includes multiple regional control agents, which are deployed on communication control nodes and divided into regional subnets within the entire network. Their core functions are regional routing collaborative optimization, regional network resource scheduling, regional load balancing control, inter-regional routing information exchange, and fault event aggregation and reporting.

[0040] Each regional control agent manages all routing node agents within its subnet, performing the following tasks: summarizing link status, network topology, and service transmission data within the subnet to construct a subnet-level global network view; conducting subnet-level route collaborative optimization to address routing conflicts and load imbalances within the subnet, dynamically adjusting routing policies to achieve load balancing within the subnet; exchanging boundary routing information with control agents in adjacent subnets to achieve cross-subnet route collaboration; summarizing network faults and service anomalies within the subnet and reporting them to the global decision agent; and receiving optimization policies from the global decision agent, forwarding them to its subordinate routing node agents, and executing them.

[0041] The Global Decision Agent layer includes the Global Decision Agent, which is deployed at the central node of the communication network. It is the highest control layer for routing decisions, and its core functions are to formulate the whole network routing strategy, coordinate global network resources, optimize cross-regional routing, handle network anomalies globally, and iteratively optimize the routing algorithm model.

[0042] The global decision agent autonomously performs the following functions: aggregates network status and business operation data reported by regional control agents across the entire network to build a unified network view across the entire network; and formulates unified routing decision rules and security control strategies across the entire network to adapt to the overall business plan.

[0043] It is understood that this embodiment designs a four-layered distributed architecture consisting of a terminal agent, a routing node agent, a regional control agent, and a global decision agent. This architecture enables distributed network status awareness, layered collaborative decision-making of routing strategies, and graded self-healing of faults. It solves the problems of high single-point failure risk, poor scalability, and large decision-making delay in traditional centralized routing schemes, and is suitable for scenarios with complex power grid wide-area coverage and massive terminal access.

[0044] Meanwhile, this embodiment designs a distributed heterogeneous link status real-time perception mechanism for routing node agents. Combined with a three-level topology view collaborative construction method of local-regional-global, it realizes low-latency and high-precision perception of heterogeneous network link status and real-time updating of network topology, solving the problems of incomplete heterogeneous network link perception and lagging topology updates in traditional solutions.

[0045] Furthermore, the step of generating a candidate path set based on multiple network topology graphs and the first constraint set includes: A first link set is generated based on the first constraint set. The first link set includes multiple first links. Each QoS indicator of the first link satisfies the threshold corresponding to the QoS indicator in the first constraint set. Multiple routing paths are generated in the first link set based on the depth-first search algorithm and the pruning algorithm to obtain the candidate path set. The routing path is a loop-free reachable path that starts from the source node of the routing request, ends at the destination node, and is located in the topology formed by the first link set.

[0046] Furthermore, the step of filtering the first path from the candidate path set includes: Obtain a set of routing path metrics, which includes multiple standardized QoS metrics for each second path, wherein the second path is the routing path in the candidate routing path set; Obtain a first weight set, which includes multiple first weights, and one first weight corresponds to a QoS indicator of the second path. A standardized decision matrix is ​​constructed based on the routing path indicator set, wherein the column vectors of the standardized decision matrix consist of multiple QoS indicators of a second path; A weighted standardized decision matrix is ​​generated based on the standardized decision matrix and the first weight vector, and the first weight vector is generated based on the first weight set. Calculate the Euclidean distance between each column vector of the weighted standardized decision matrix and a preset first vector to obtain a first distance vector. Calculate the Euclidean distance between each column vector of the weighted standardized decision matrix and a preset second vector to obtain a second distance vector. Both the first and second vectors are vectors composed of multiple preset QoS indicators. One of the Euclidean distances in the first and second distance vectors corresponds to one of the column vectors. Multiple proximity scores are calculated based on the first distance vector and the second distance vector, and each proximity score corresponds to a second path. The first path is obtained by taking the second path with the highest proximity as the first path.

[0047] Furthermore, the calculation of multiple proximity scores based on the first distance vector and the second distance vector can be expressed as: ; in, The proximity of the j-th second path, and These are the Euclidean distances between the column vector corresponding to the j-th second path and the first and second vectors, respectively.

[0048] Furthermore, obtaining the first weight set includes: Obtain a service priority set, which includes the priorities of multiple QoS indicators of the power grid service; Based on the service priority set and the hierarchical analysis method, multiple demand weights are calculated to obtain multiple demand weights, and each demand weight corresponds to a QoS indicator. Based on each standardized QoS metric of each second path, multiple entropy weights are calculated to obtain multiple entropy weights, and each entropy weight corresponds to a QoS metric. The first weight is calculated based on the entropy weight and the demand weight corresponding to each QoS indicator, resulting in multiple first weights.

[0049] Furthermore, the calculation of multiple entropy weights based on each standardized QoS metric for each of the second paths includes: Calculate the first weight of the first QoS indicator to obtain the first weight. The first weight is the ratio of the normalized value of the first QoS indicator to the sum of the normalized values ​​of all second QoS indicators. The first QoS indicator is one of the two QoS indicators. The second QoS indicator is the QoS indicator of the second path. The normalization of the second QoS indicator is calculated based on the QoS indicators of all second paths in the candidate route path set. The entropy value of the first QoS indicator is calculated based on the first weight of the first QoS indicator and the preset number of alternative links, and the entropy value is obtained. The difference coefficient is calculated based on the entropy value of the first QoS indicator to obtain the difference coefficient of the first QoS indicator; Calculate the difference coefficients of all the second QoS indicators and sum them to obtain the difference coefficient sum; The entropy weight of the first QoS indicator is obtained by calculating the difference coefficient of the first QoS indicator and the difference coefficient.

[0050] Specifically, in combination Figure 2 and Figure 3 The core routing decision method of this embodiment is described below. This embodiment adopts a full-process logic of service parsing, state awareness, topology construction, candidate route generation, optimal path selection, route execution, closed-loop optimization, and fault self-healing. The specific steps are as follows: Part 1: Analysis and Identification of Business QoS Requirements The terminal agent receives power grid services initiated by the terminal, identifies the service type, and quantifies and analyzes QoS requirements. Services are categorized according to level, and for different service types, corresponding QoS constraints are quantified, including maximum allowable latency, maximum allowable packet loss rate, minimum available bandwidth, service priority, and link reliability requirements. Simultaneously, a unique service identifier and priority label are assigned to each service. This service identifier, priority label, and corresponding QoS constraints are reported to the associated routing node agent along with the routing request, providing core constraints for subsequent routing decisions.

[0051] 2. Distributed real-time perception of heterogeneous link status: The routing node agent monitors the links and performs real-time, periodic status awareness on all heterogeneous communication links under its jurisdiction. The awareness period can be adaptively adjusted according to the link stability. It collects core parameters such as latency, throughput, packet loss rate, signal reception strength, and signal-to-noise ratio. The routing node agent preprocesses the collected raw link status data, including outlier removal, data standardization, and noise smoothing, to eliminate data collection errors. The preprocessed data is used for local route calculation on the one hand, and synchronized in real time to neighboring routing node agents and the regional management agent on the other hand.

[0052] The steps for standardizing data are as follows: For positive indicators: ; For negative indicators: ; in, The standardized index value, These are the original collected values. This is the historical maximum value of the indicator. This is the historical minimum value of this indicator.

[0053] 3. Agent-based collaborative network topology construction: This embodiment employs a hierarchical distributed topology construction mechanism to avoid the problems of high signaling overhead and untimely updates associated with centralized topology collection. Routing node agents interact with each other based on the link status information of neighboring nodes to construct a topology view of the adjacent network within a certain range, and update it in real time. The update cycle is synchronized with the link awareness cycle for local rapid route calculation and fault self-healing. The regional control agent aggregates the link status and topology information reported by all routing node agents within its jurisdiction to construct a complete network topology view within the region, including information on all nodes, links, and heterogeneous network resources within the region, for regional route collaborative optimization. The global decision agent aggregates the regional topology and cross-regional link information reported by all regional control agents to construct a unified network topology view for the entire network, for cross-regional global route optimization and policy formulation.

[0054] 4. Candidate route generation based on multi-attribute decision-making: After receiving a routing request from the terminal agent, the routing node agent uses the service QoS requirements as the core constraint and the preprocessed link state parameters as the basis. It then employs a depth-first search combined with a pruning algorithm to generate a set of candidate routing paths that satisfy the hard constraints of the service QoS. Specifically: 1. Hard constraint filtering: Eliminate links that do not meet the hard constraints of the business, narrow the range of route search, and reduce computational complexity.

[0055] 2. Candidate Path Generation: Starting from the source node of the routing request and ending at the destination node, search for all loop-free reachable paths in the filtered link topology to generate a candidate route path set.

[0056] 3. Path Attribute Quantification: For each candidate route path, the comprehensive attribute value of the path is quantitatively calculated to obtain the QoS indicators of each route path in the candidate route path set, including end-to-end total latency, total packet loss rate, total available bandwidth, path load rate, and path reliability score, providing data support for subsequent optimal path selection.

[0057] 5. Adaptive weighted multi-attribute decision-making for optimal route selection: For the generated set of candidate routes, the TOPSIS multi-attribute decision algorithm with adaptive weights is adopted. Combining business QoS requirements, the weights of each decision indicator are adaptively adjusted to select the optimal route, solving the problem that traditional fixed-weight routing cannot adapt to differentiated business needs. The specific steps are as follows: 1. Adaptive Weight Allocation of Decision Indicators: Based on the core QoS requirements of the service quantified by the terminal agent, the weight coefficients of each indicator are adaptively adjusted. Weight allocation employs a combination of the analytic hierarchy process (AHP) and entropy weighting, taking into account both subjective business needs and objective link conditions. The weight calculation formula is as follows: ; in, The comprehensive weight of the i-th indicator is... Subjective weighting based on business needs, The objective entropy weight is based on the link state, and α is the weight balancing coefficient, which is adjusted according to the business priority. The higher the priority, the larger α is.

[0058] Furthermore, for the objective entropy weight of the i-th indicator... The calculation steps include: The QoS metrics for each path in the candidate route path set are standardized. The QoS metrics for each path include latency, bandwidth, packet loss rate, signal-to-noise ratio, etc. The weight of each QoS indicator is calculated. For the weight of the i-th QoS indicator, the specific calculation can be expressed as follows: ; in, This represents the normalized value of the i-th QoS indicator. This represents the summation of the normalized values ​​for each QoS indicator.

[0059] Calculate the entropy value of the i-th QoS indicator based on its weight. , can be represented as: ; in, The entropy value of the i-th QoS metric, where m is the number of alternative links. The weight of the i-th QoS indicator; Furthermore, calculate the difference coefficient of the i-th QoS indicator. , can be represented as: ; in, and These are the difference coefficient and entropy value of the i-th QoS indicator, respectively.

[0060] Calculate the objective entropy weight of the i-th indicator , can be represented as: ; in, This represents the sum of the difference coefficients for all QoS metrics.

[0061] For the subjective weight of the i-th indicator The subjective weights are calculated based on the QoS importance of each specific power grid service and using the Analytic Hierarchy Process (AHP).

[0062] It is understandable that traditional routing schemes (such as OSPF) use fixed metrics (such as hop count and static cost), which cannot adapt to the dynamic differences in QoS requirements of power grid services. This embodiment considers the following factors when designing weight allocation: 1) Consider subjective business needs: Different power grid services have drastically different sensitivities to different QoS indicators. For example, production control services are extremely sensitive to latency and reliability, but relatively insensitive to bandwidth; operation monitoring video services are the opposite. Therefore, the weighting must reflect the subjective priorities of the service initiator.

[0063] 2) Consider the objective state of the links: Within the same candidate path set, the same QoS metric (such as remaining bandwidth) may vary significantly across different links or paths. If a metric shows little difference across paths, its contribution to distinguishing path quality is low, and it should not be assigned excessive weight; conversely, metrics with large differences should receive higher weights to highlight the discriminative power of the decision. This is the objective basis of the entropy weight method.

[0064] 3) Considering the balance between subjective and objective factors: Relying entirely on subjective weights may lead to path selection deviating from the actual network state (for example, overemphasizing latency while ignoring bandwidth even when all path latency meets requirements); relying entirely on objective entropy weights may ignore the specific needs of the service. Therefore, this embodiment introduces a weight balancing coefficient α, where the fusion ratio of subjective and objective weights is determined by the service priority, achieving a dynamic balance.

[0065] In this way, by injecting business preferences into the decision-making process through subjective weights, the selected optimal path truly meets the QoS requirements of the service, rather than adopting a uniform rule, thus achieving precise adaptation between the service and the network. Furthermore, by automatically amplifying significantly different indicators through objective entropy weights, the decision-making algorithm can more clearly distinguish between superior and inferior paths when faced with multiple candidate paths, avoiding random selection due to similar indicator values, and improving the discriminative power and robustness of the decision. Simultaneously, when the network state changes (such as a sudden increase in latency on a link), the entropy weight of that indicator will increase accordingly, thus having a greater impact on path evaluation, enabling the optimal path selection to promptly "avoid" degraded links, achieving self-adaptation and responding to dynamic network changes.

[0066] 2. Construction of standardized decision matrix: For each indicator value of the candidate path, process it according to the standardization formula in step 2 to construct a standardized decision matrix.

[0067] 3. Calculation of weighted standardized decision matrix: Multiply the standardized decision matrix by the comprehensive weight of each indicator to obtain the weighted standardized decision matrix.

[0068] 4. Determination of positive and negative ideal solution vectors: Determine the positive ideal solution (optimal value) and negative ideal solution (worst value) for each index, thereby constructing the positive ideal solution vector and the negative ideal solution vector.

[0069] Understandably, for a given QoS metric, the setting of the positive ideal solution considers the optimal value of that positive metric, such as the theoretical maximum value of bandwidth, signal-to-noise ratio, or reliability, or the theoretical minimum value of latency, packet loss rate, or bit error rate; while the setting of the negative ideal solution considers the worst value of that metric, such as the theoretical minimum value of a positive metric or the theoretical maximum value of a negative metric.

[0070] Understandably, in the positive ideal solution, for each QoS indicator, its "best possible value" is set based on the service type and the theoretical extreme value of the network. For positive indicators, the higher the better, such as available bandwidth, signal-to-noise ratio, and reliability score, the positive ideal solution takes the theoretical maximum value of the indicator or the upper limit that the service can actually enjoy. For negative indicators, the lower the better, such as latency, packet loss rate, and bit error rate, the positive ideal solution takes the theoretical minimum value of the indicator. For example, the positive ideal solution for latency in production control services can be set to 0 ms, although this is practically impossible to achieve, it serves as a benchmark for comparison.

[0071] In the negative ideal solution, a worst-case acceptable value or theoretical lower limit is set for each indicator. The negative ideal solution for a positive indicator is the theoretical minimum value (e.g., bandwidth of 0). The negative ideal solution for a negative indicator is the upper limit of the threshold in the business constraint set (e.g., the maximum allowable latency for this type of business) or the theoretical maximum value.

[0072] The value of the ideal solution should be tied to the business category. For example, for production control business, the positive ideal solution for latency should be much stricter than that for information collection business, in order to reflect differentiated protection.

[0073] 5. Proximity Calculation: Calculate the Euclidean distance between each column vector in the weighted standardized decision matrix and the positive ideal solution vector and the negative ideal solution vector, and then calculate the overall proximity of each path. The formula is as follows: ; in, Let be the overall proximity of the j-th candidate path. Let Euclidean distance be the column vector corresponding to the path and the negative ideal solution vector. This is the Euclidean distance between the column vector corresponding to the path and the positive ideal solution vector. The value ranges from 0 to 1. The larger the value, the better the overall path performance.

[0074] The further away from the positive ideal solution vector, i.e. The larger the value, the closer it is to the positive ideal solution vector, i.e. The smaller the value, the better the path; at this point, the proximity is... The closer the connection is to 1, the higher the proximity value, and the better the overall path performance.

[0075] Understandably, after the weighted standardized decision matrix is ​​formed, each candidate path corresponds to a point in a multidimensional space, i.e., a column vector. The decision objective is to find the point that is closest to the positive ideal solution and furthest from the negative ideal solution. Euclidean distance is a classic method for measuring the similarity between points in a multidimensional space. In the weighted space composed of QoS indicators, Euclidean distance reflects the comprehensive degree of difference between the path and the ideal solution, including the joint influence of all indicators, rather than an independent comparison of a single indicator.

[0076] Furthermore, in the proximity formula, the two objectives of approaching the positive ideal solution and moving away from the negative ideal solution are merged into a single scalar. When the path infinitely approaches the positive ideal solution, Approaching 0, closeness The path approaches 1; when the path is infinitely close to the negative ideal solution, Approaching 0, closeness The similarity value tends to 0. The similarity value is between 0 and 1, which facilitates sorting and comparison.

[0077] At the same time, it avoids the dominance of a single indicator. If a simple weighted sum is used, a situation may arise where a particularly good indicator masks multiple poor indicators. Euclidean distance, by using the square root of the sum of squares, is more sensitive to the deviations of each indicator and can more evenly evaluate the overall performance of the path.

[0078] 6. Optimal Path Selection: Select the candidate path with the highest overall proximity as the primary routing path for this service, and select the two paths with the next highest proximity as backup routing paths. Send the results to the terminal Agent and the relevant routing node Agent for execution.

[0079] Six: Real-time monitoring and closed-loop optimization of link status: Construct a closed-loop optimization mechanism for the entire process to achieve dynamic adaptive adjustment of routing strategies, specifically including: Real-time monitoring: The routing node agent continuously monitors the status of each link and the service transmission effect of the primary routing path. The terminal agent provides real-time feedback on indicators such as end-to-end service transmission latency and packet loss rate. If link performance degradation (not meeting service QoS requirements), link failure, network congestion, etc. are detected, route recalculation is triggered immediately.

[0080] Local rapid optimization: For single link failures or performance degradation, the routing node Agent directly activates the backup routing path to complete local self-healing, while reporting the fault information to the regional management and control Agent. The self-healing switching latency does not exceed 50ms, meeting the requirements of power grid production control services.

[0081] Regional collaborative optimization: When the regional management agent detects network load imbalance, simultaneous degradation of multiple links, or large-scale electromagnetic interference within the region, it triggers regional routing collaborative optimization, recalculates the routing paths of relevant services within the region, adjusts the routing strategies of each node, achieves load balancing within the region, and avoids interference areas.

[0082] Global Iterative Optimization: Based on network-wide operational data, the global decision agent iteratively optimizes the weight model and link prediction model of the routing decision algorithm on a regular (every 24 hours) basis. At the same time, it adjusts the network-wide routing strategy in advance to address seasonal and regional changes in the communication environment (such as the sandstorm season in desert areas and the rainy season in mountainous areas), thereby improving the network's adaptability to the environment.

[0083] It is understood that this embodiment is based on an adaptive routing decision method for QoS differentiation of power grid services: for the differentiated QoS requirements of power grid services, an adaptive weighted multi-attribute decision routing algorithm is designed, which combines the subjective requirements of services with the objective status of links to adaptively allocate index weights, accurately match the optimal routing path for different types of power grid services, and solves the problem that traditional fixed-weight routing cannot adapt to the differentiated service requirements of power grids.

[0084] In summary, it is understandable that this embodiment significantly improves the environmental adaptability and reliability of routing decisions, solving the problem that traditional solutions cannot adapt to complex dynamic power grid link environments. Through a distributed multi-agent architecture, real-time perception of heterogeneous link status and dynamic adaptive adjustment of routing strategies are achieved, enabling millisecond-level local self-healing in the event of link failures. Simultaneously, it can adaptively avoid link degradation areas caused by electromagnetic interference and terrain obstruction, greatly improving the reliability and stability of data transmission in complex power grid environments.

[0085] Furthermore, it achieves precise QoS assurance for differentiated power grid services, solving the problem that traditional solutions cannot adapt to the diverse needs of power grid services. Through a routing decision algorithm with differentiated service QoS identifiers and adaptive weights, it can match exclusive optimal routing paths for different types of power grid services, while taking into account the high bandwidth requirements of video monitoring services and the low power consumption requirements of data acquisition services. Compared with traditional fixed-rule routing schemes, it achieves differentiated carrying of all power grid services and meets the stringent requirements of power safety production.

[0086] Furthermore, the scalability and security of routing decisions are significantly improved, resolving the issues of high single-point failure risk and poor scalability in traditional centralized solutions. Adopting a layered distributed architecture, the routing decision-making function is decentralized to the routing node Agent, eliminating the need for a central controller. Even if the global decision-making Agent is offline, regional and local routing decisions and data forwarding can still be completed normally, completely avoiding single-point failure risks. Simultaneously, the distributed architecture can smoothly adapt to the growth in the number of power grid terminals, supporting massive terminal access. Compared to centralized SDN routing solutions, network signaling overhead is reduced by more than 60%, and decision-making efficiency is significantly improved.

[0087] The above description, in conjunction with specific preferred embodiments, provides a further detailed explanation of the present invention. It should not be construed that the specific implementation of the present invention is limited to these descriptions. For those skilled in the art, various simple deductions or substitutions can be made without departing from the concept of the present invention, and all such modifications and substitutions should be considered within the scope of protection of the present invention.

Claims

1. An adaptive routing decision method for converged communication networks based on distributed agents, characterized in that, include: Construct the terminal agent layer and the routing node agent layer; The terminal agent layer includes multiple terminal agents, which are used to receive power grid services initiated by power grid terminals and generate a first constraint set based on the power grid services. The first constraint set includes thresholds corresponding to multiple QoS indicators. The routing node agent layer includes multiple routing node agents deployed on communication nodes. Each routing node agent is used to collect link status data of multiple links with a communication node as the endpoint in real time. The link status data includes multiple QoS indicators. The routing node agent generates a network topology map based on first link status data, which includes the link status data collected in real time by the routing node agent and its adjacent multiple routing node agents. When the terminal Agent sends a routing request, it generates a candidate path set based on multiple network topology maps and the first constraint set, and filters the first path from the candidate path set to obtain the first path. The candidate path set includes multiple routing paths corresponding to the routing request. Data transmission of the power grid services is performed based on the first path.

2. The adaptive routing decision method for converged communication networks based on distributed agents according to claim 1, characterized in that, After constructing the terminal agent layer and the routing node agent layer, the following is also included: A regional control agent layer is constructed, which includes multiple regional control agents deployed on communication control nodes. One regional control agent is used to receive the network topology map transmitted by multiple first routing node agents and construct a subnet network topology map to obtain the subnet network topology map. The first routing node agents are the routing node agents in a first preset area. One first preset area corresponds to one regional control agent and multiple routing node agents. The first preset area is divided based on regional subnets within the entire network. The regional management agent performs route coordination optimization, network resource scheduling, load balancing management, and fault event aggregation and reporting within the first preset region based on the network topology map of its corresponding subnet. The area management agents corresponding to adjacent subnets coordinate cross-subnet routing by exchanging boundary routing information.

3. The adaptive routing decision method for converged communication networks based on distributed agents according to claim 2, characterized in that, After constructing the regional control agent layer, the following is also included: A global decision agent layer is constructed, including a global decision agent deployed at the central node of the communication network, which is used to receive the subnet network topology map reported by each of the regional control agents, and construct a unified network topology map based on multiple subnet network topology maps to obtain the unified network topology map; The global decision agent performs network-wide routing strategy formulation, global network resource coordination, cross-regional routing collaborative optimization, global network anomaly handling, and routing algorithm model iterative optimization based on the unified network topology map.

4. The adaptive routing decision method for converged communication networks based on distributed agents according to claim 1, characterized in that, The generation of the first constraint set based on the power grid service includes: The terminal agent classifies the power grid services to obtain the categories of the power grid services, and quantifies the thresholds of multiple QoS indicators of the power grid services based on the categories of the power grid services to obtain the first constraint set. The categories include production control, operation monitoring and information collection.

5. The adaptive routing decision method for converged communication networks based on distributed agents according to claim 1, characterized in that, The step of generating a candidate path set based on multiple network topology graphs and the first constraint set includes: A first link set is generated based on the first constraint set. The first link set includes multiple first links. Each QoS indicator of the first link satisfies the threshold corresponding to the QoS indicator in the first constraint set. Multiple routing paths are generated in the first link set based on the depth-first search algorithm and the pruning algorithm to obtain the candidate path set. The routing path is a loop-free reachable path that starts from the source node of the routing request, ends at the destination node, and is located in the topology formed by the first link set.

6. The adaptive routing decision method for converged communication networks based on distributed agents according to claim 1, characterized in that, The step of filtering the first path from the candidate path set includes: Obtain a set of routing path metrics, which includes multiple standardized QoS metrics for each second path, wherein the second path is the routing path in the candidate routing path set; Obtain a first weight set, which includes multiple first weights, and one first weight corresponds to a QoS indicator of the second path. A standardized decision matrix is ​​constructed based on the routing path indicator set, wherein the column vectors of the standardized decision matrix consist of multiple QoS indicators of a second path; A weighted standardized decision matrix is ​​generated based on the standardized decision matrix and the first weight vector, and the first weight vector is generated based on the first weight set. Calculate the Euclidean distance between each column vector of the weighted standardized decision matrix and a preset first vector to obtain a first distance vector. Calculate the Euclidean distance between each column vector of the weighted standardized decision matrix and a preset second vector to obtain a second distance vector. Both the first and second vectors are vectors composed of multiple preset QoS indicators. One of the Euclidean distances in the first and second distance vectors corresponds to one of the column vectors. Multiple proximity scores are calculated based on the first distance vector and the second distance vector, and each proximity score corresponds to a second path. The first path is obtained by taking the second path with the highest proximity as the first path.

7. The adaptive routing decision method for converged communication networks based on distributed agents according to claim 6, characterized in that, The calculation of multiple proximity scores based on the first distance vector and the second distance vector can be expressed as follows: ； in, The proximity of the j-th second path, and These are the Euclidean distances between the column vector corresponding to the j-th second path and the first and second vectors, respectively.

8. The adaptive routing decision method for converged communication networks based on distributed agents according to claim 6, characterized in that, The process of obtaining the first weight set includes: Obtain a service priority set, which includes the priorities of multiple QoS indicators of the power grid service; Based on the service priority set and the hierarchical analysis method, multiple demand weights are calculated to obtain multiple demand weights, and each demand weight corresponds to a QoS indicator. Based on each standardized QoS metric of each second path, multiple entropy weights are calculated to obtain multiple entropy weights, and each entropy weight corresponds to a QoS metric. The first weight is calculated based on the entropy weight and the demand weight corresponding to each QoS indicator, resulting in multiple first weights.

9. The adaptive routing decision method for converged communication networks based on distributed agents according to claim 8, characterized in that, The calculation of multiple entropy weights based on each standardized QoS metric for each of the second paths includes: Calculate the first weight of the first QoS indicator to obtain the first weight. The first weight is the ratio of the normalized value of the first QoS indicator to the sum of the normalized values ​​of all second QoS indicators. The first QoS indicator is one of the two QoS indicators. The second QoS indicator is the QoS indicator of the second path. The normalization of the second QoS indicator is calculated based on the QoS indicators of all second paths in the candidate route path set. The entropy value of the first QoS indicator is calculated based on the first weight of the first QoS indicator and the preset number of alternative links, and the entropy value is obtained. The difference coefficient is calculated based on the entropy value of the first QoS indicator to obtain the difference coefficient of the first QoS indicator; Calculate the difference coefficients of all the second QoS indicators and sum them to obtain the difference coefficient sum; The entropy weight of the first QoS indicator is obtained by calculating the entropy weight of the first QoS indicator based on the difference coefficient of the first QoS indicator and the difference coefficient.

10. The adaptive routing decision method for converged communication networks based on distributed agents according to claim 6, characterized in that, After obtaining the first path, the process further includes: Based on the proximity of each second path, multiple second paths are selected from the candidate routing path set as backup paths, and the backup paths are used to replace the first path for data transmission of the power grid service.

Images