Data routing method for non-direct connection

Abstract

The invention discloses a non-direct connection-oriented data routing method, which comprises the following steps of: S1, acquiring topological structures, link bandwidths, time delays, loads and energy consumption states of nodes in a network in real time, and constructing a time sequence dynamic matrix of the nodes and links; s2, on the basis of the dynamic matrix, adopting a neural network adaptive enhanced ant colony optimization method to generate alternative paths; s3, generating a grey wolf optimization algorithm initial population by using the alternative paths, and constructing a multi-dimensional composite fitness function; s4, according to the fitness function, driving the grey wolf optimization algorithm to perform multi-scale iteration to update the path; s5, topology and node state prediction is carried out based on the dominant path, and the prediction path is optimized in advance; and S6, issuing the optimal path and the alternative path at the same time, carrying out data parallel forwarding, and driving a neural enhanced ant colony optimization algorithm to update online. The method improves the network path selection efficiency and the resource utilization rate, and is suitable for data routing in a complex network environment.

Description

Technical Field

[0001] This invention relates to the field of data communication and network optimization technology, and in particular to a data routing method for non-direct connections. Background Technology

[0002] With the rapid development of information technology and network scale, data routing technology has gradually become one of the important foundational technologies in the field of communication networks. Especially in network environments with complex topologies, efficient and reliable data routing directly determines the effective utilization of network resources and the improvement of overall performance. Currently, conventional data routing technologies mainly include static routing methods based on shortest path algorithms and dynamic routing methods based on heuristic search optimization algorithms. Static routing methods based on shortest path algorithms, such as Dijkstra's algorithm or Bellman-Ford algorithm, usually rely on predefined network topology and relatively fixed path weights. Although easy to implement, they cannot effectively adapt to rapid dynamic changes in network conditions. On the other hand, heuristic search optimization algorithms, such as traditional ant colony optimization or particle swarm optimization algorithms, achieve a certain degree of adaptability and dynamic optimization performance in routing selection through information interaction and path evaluation mechanisms among individuals, and have been widely used in large-scale, dynamically changing network environments.

[0003] However, the aforementioned traditional dynamic routing technologies often exhibit significant shortcomings when there are no direct paths between nodes: First, the traditional ACO algorithm relies on simple pheromone accumulation and path search mechanisms, making it difficult to efficiently handle complex multidimensional state information in non-directly connected environments, resulting in convergence speed and path quality that cannot meet the demands of high real-time network environments; Second, although some studies have attempted to introduce more complex intelligent optimization algorithms, such as the standard Grey Wolf Optimization algorithm, its initial population path generation method lacks in-depth mining of multidimensional network state information, leading to low convergence efficiency in the early stages; In addition, existing algorithms generally lack mechanisms for proactively predicting and pre-optimizing path performance, making it difficult to respond quickly to sudden changes in network topology or node states, resulting in degraded transmission performance and reduced overall network resource utilization.

[0004] Therefore, how to provide a data routing method for non-direct connections is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0005] One objective of this invention is to propose a data routing method for non-direct connections, which achieves efficient path selection and optimized resource allocation by combining neural-enhanced ant colony optimization with gray wolf optimization. This invention has the beneficial effects of strong routing adaptability, high path quality, and significantly improved network resource utilization.

[0006] A data routing method for non-direct connections according to an embodiment of the present invention includes the following steps:

[0007] S1. Real-time collection of topology, link bandwidth, latency, load and energy consumption status of each node in the network, and construction of a time-series dynamic matrix of dual node and link status;

[0008] S2. Based on the time-series dynamic matrix, an ant colony optimization method with neural network adaptive enhancement is adopted. By having individual ants perform asynchronous multi-dimensional state information perception search within the network, a set of alternative paths is automatically generated.

[0009] S3. Generate the initial population of the Grey Wolf optimization algorithm using alternative paths, and construct a multi-dimensional composite fitness function based on the node congestion index, link quality coefficient and network load balance degree on each path.

[0010] S4. The multi-scale iterative update of the path is carried out based on the multi-dimensional composite fitness function-driven gray wolf optimization algorithm. In each iteration, the matching degree between each path solution and the network dynamic matrix is ​​checked in real time, and the advantageous path is continuously screened and iteratively retained.

[0011] S5. An active prediction mechanism based on the dominant path is introduced, which introduces network topology and node status. The neural reinforcement ant colony optimization algorithm is used to generate a set of predicted paths in advance, and the gray wolf optimization algorithm is used to pre-optimize the paths in the set of predicted paths in advance to obtain the optimal path.

[0012] S6. Simultaneously send the optimized optimal path and at least two alternative paths to the network control plane for parallel data forwarding, and use the actual performance differences of the paths as a basis to drive the neural reinforcement ant colony optimization algorithm for online continuous adaptive updates.

[0013] Optionally, S1 specifically includes:

[0014] S11. Perform dual-dimensional joint encoding of node role and position for all nodes in the network. Based on the actual functional positioning and topological relative position of the node in the network, assign a unique and dynamically updated multi-dimensional feature identifier to each node to form a node topological role matrix.

[0015] S12. Utilize a traffic-aware link bandwidth adaptive measurement mechanism to automatically increase the measurement frequency when the link load change rate exceeds a preset threshold. Collect data in real time and quantify bandwidth characteristics in three dimensions, including the effective transmission rate of the link, the link utilization rate, and the proportion of idle bandwidth, to construct a link bandwidth status matrix.

[0016] S13. By pre-embedding clock synchronization units in each node, nodes alternately send and receive delay calibration beacon data frames within each preset interval. Combine the timestamp information of the sent and received beacon data frames to perform real-time bidirectional delay difference calculation and automatic compensation, and construct a link delay matrix after dynamic self-calibration.

[0017] S14. Integrate data cache monitoring sensor probes into the data processing unit of each node to detect and record the cache queue data length, average data packet waiting time and node data throughput in real time and time period, and construct a multi-dimensional quantitative matrix of node real-time load status.

[0018] S15. Integrate a high-sensitivity digital current and voltage real-time sensing unit inside the node chip to continuously measure the current operating power consumption of each node and record the transient power fluctuations in time sequence, thereby constructing a dynamic high-resolution digital matrix of node energy consumption status.

[0019] S16. Based on the timestamp marking synchronization mechanism, and in accordance with a unified time series benchmark, the node topology role matrix, link bandwidth status matrix, link delay matrix, node real-time load status multidimensional quantization matrix, and node energy consumption status dynamic high-resolution digital matrix are fused element by element to form a time series dynamic matrix of dual node and link status.

[0020] S17. Utilizing the network real-time state difference adaptive feedback mechanism, based on the continuous monitoring and prediction of the real-time change amplitude and trend of each element in the time-series dynamic matrix, the real-time update frequency of the time-series dynamic matrix is ​​adjusted, and when the change amplitude of matrix elements exceeds the preset threshold, a real-time event-type update of the data in the time-series dynamic matrix is ​​triggered.

[0021] Optionally, S2 specifically includes:

[0022] S21. Based on the time-series dynamic matrix, the network is dynamically divided into multiple network-aware regions according to the real-time load differences and topological relationships of the nodes, and ant colony individuals with region-specific identifiers are deployed in each region to perform path search.

[0023] S22. A real-time path status sensing probe is built into each node to measure the link bandwidth change rate, latency dynamic fluctuation, and node load difference between the node and all adjacent nodes on a cycle-by-cycle basis. Combined with a feedforward neural network model, a jump node priority sequence is generated in real time.

[0024] S23. Individual ant colonies in each region periodically exchange current path exploration data, using an asynchronous interactive data sharing mechanism. Based on the real-time path exploration status data received from other regions, individual ant colonies dynamically adjust their next path exploration direction.

[0025] S24. When making node hopping decisions, individual ants in the colony dynamically adjust the probability distribution of the next hop node selection by calculating the link bandwidth change trend, link latency stability and node load fluctuation between the current node and the candidate node, and eliminate paths with unstable link status or excessive node load.

[0026] S25. Using a node and link search history state marking matrix, a cumulative memory weight matrix of path search history states is formed by recording and accumulating the characteristic information of node and link state changes in the past path search process of individual ants.

[0027] S26. After each ant colony individual completes the multi-hop path search, a composite evaluation index is established based on the network node load balance, the persistence of link status within the path, and the global network path redundancy. A set of global alternative paths that can simultaneously take into account path stability and network load balance is automatically generated.

[0028] Optionally, S3 specifically includes:

[0029] S31. For the global candidate paths, extract the feature data of node congestion index, link quality coefficient and network load balance degree of each path, and calculate the feature Euclidean distance between paths. Based on this distance, dynamically aggregate them into several path subgroups and assign a unique path clustering identifier to each subgroup.

[0030] S32. Integrate a real-time congestion monitoring unit within each node of the alternative path, collect the current data buffer queue length, data packet processing rate, and node buffer saturation level of each node, and calculate the real-time congestion index of the node by combining the above values ​​to form a node congestion index matrix.

[0031] S33. Perform continuous dynamic sampling of data streams for each link of the candidate path, and calculate and form a link quality coefficient matrix by measuring the instantaneous effective transmission rate, link delay fluctuation amplitude and link interruption frequency in real time.

[0032] S34. Based on the node congestion index matrix, calculate the degree of spatial distribution difference of the node congestion index in each path among the path nodes, and obtain the network load balance by analyzing the spatial distribution entropy of the congestion index among the path nodes.

[0033] S35. Based on the above node congestion index matrix, link quality coefficient matrix and network load balance, construct a three-dimensional feature tensor of the path state space, and use high-dimensional space projection mapping to map the feature tensor to the state feature vector space so as to facilitate the comparison of feature values ​​of the state vectors.

[0034] S36. Construct a multidimensional composite fitness function based on the node congestion index matrix, link quality coefficient matrix, and network load balancing degree.

[0035] S37. Sort the paths of each path subgroup according to the multidimensional composite fitness function value, and select the top-ranked paths from the sorting results to form the initial population of the Grey Wolf Optimization Algorithm.

[0036] Optionally, S4 specifically includes:

[0037] S41. Based on the initial population of the Grey Wolf Optimization Algorithm, and combined with the real-time changing trends of node congestion index, link quality coefficient and load balance in the network state matrix, a multi-scale state space is divided, and a unique scale state identifier is assigned to the path in each scale state space.

[0038] S42. In each scale state space, based on the real-time trends of node load, link quality and congestion state changes in the network state matrix, an adaptive scale perturbation method for path position vectors is used to generate a scale-adaptive updated path set that matches the real-time network state.

[0039] S43. After each scale-adaptive update of the path set, collect the current node congestion index matrix, link quality coefficient matrix and network load balance data in real time, and calculate the fitness value of the path at the current scale for each path based on the multidimensional composite fitness function.

[0040] S44. Establish a feature matching degree verification mechanism for the path and network dynamic matrix in real time. By calculating the matching degree values ​​of the path location vector and the network dynamic matrix in the three-dimensional space of node congestion index feature difference, link quality feature difference and load balancing feature difference, the matching degree of the real-time status of the path and the network is determined.

[0041] S45. Construct a two-dimensional joint evaluation index using path fitness values ​​and path real-time feature matching values. Update the path set at the current scale according to the joint evaluation index, mark the paths with the highest joint evaluation index as advantageous paths and retain them, and eliminate the paths with the lowest ranking in real time.

[0042] S46. After each iteration update, the state feature data of the retained advantageous path is integrated into the historical state feature memory of the network state matrix in real time. The dynamic accumulation of historical state feature memory data actively guides the update direction of the path position vector at the next scale.

[0043] S47. Repeat steps S41 to S46 until the number of iterations of the Grey Wolf optimization algorithm reaches the preset number or the path fitness value reaches the convergence threshold, and finally obtain a set of advantageous paths.

[0044] Optionally, S42 specifically includes:

[0045] S421. Based on the multidimensional trend change characteristics of node load, link quality and congestion status in the network state matrix, the path position vector dimension space is divided into several independent scale subspaces, and the path position vector dimension in each scale subspace is assigned a scale subspace identifier that is dynamically updated in real time.

[0046] S422. For each dimension of the path position vector in each scale subspace, calculate the instantaneous change gradient of the feature data of the corresponding dimension of the network state matrix in real time, and obtain the initial perturbation amplitude of each dimension by calculating the spatial difference of the feature change gradient values.

[0047] S423. Establish an interactive perturbation amplitude propagation mechanism for feature changes between the dimensional scale subspaces of the path position vector, dynamically calculate and execute the interactive coupling of perturbation amplitudes between different dimensions in real time based on the initial perturbation amplitude, and realize the dynamic joint perturbation of the path position vector in multiple dimensions.

[0048] S424. After completing the joint perturbation update, calculate the feature difference between the feature change of each dimension in the path position vector and the corresponding dimension in the real-time network state matrix, and form the path position vector feature update matching degree matrix dimension by dimension.

[0049] S425. Update the real-time feature matching error of each dimension in the matching degree matrix based on the path position vector features, determine the correction direction and magnitude of each path position vector dimension in real time, and dynamically perform multi-dimensional scale backtracking correction of each dimension of the path position vector.

[0050] S426. Based on the overall feature matching between the real-time corrected path position vector and the current network state matrix, generate a set of scale-adaptive updated paths that are highly matched with the real-time dynamic state of the network.

[0051] Optionally, S46 specifically includes:

[0052] S461. After each iteration of the Grey Wolf optimization algorithm, the node congestion index, link quality coefficient, and load balance status data of the retained advantageous paths are jointly encoded in terms of time and space dimensions to generate a real-time spatiotemporal joint feature identifier matrix of path status features.

[0053] S462. Based on the real-time spatiotemporal joint feature identifier matrix of the path state features, calculate the feature space cosine similarity between each path and the corresponding feature data in the historical state feature memory to obtain the real-time matching degree value between the path features and the historical state data.

[0054] S463. Construct a feature fusion weight matrix between path state feature data and historical state feature memory based on real-time matching degree value, dynamically fuse path state feature data into historical state feature memory through weighted average method, and update historical state feature memory in real time.

[0055] S464. Introduce a dynamic time decay mechanism for feature data in the historical state feature memory bank. Assign a time decay coefficient to the real-time dwell time of each state feature data in the memory bank and automatically update the weight of historical state feature data by gradually decreasing the feature weight as time increases.

[0056] S465. Using the historical state feature memory after time decay, extract the cumulative change trend of three dimensions of feature data: node congestion index, link quality coefficient and load balance in real time, and generate multi-dimensional trend guidance vectors for the next scale path location vector update respectively.

[0057] S466. Based on multi-dimensional trend guidance vectors, feedforward active pre-regulation is implemented on the update direction of the path position vector at the next scale. Through real-time prediction and active adjustment of feature trends, the path position vector is guided to actively adapt to the dynamic change trend of the network's real-time state.

[0058] Optionally, S5 specifically includes:

[0059] S51. Real-time collection of congestion index, link quality coefficient and network load balance status of each node in the advantageous path, construction of a multi-dimensional dynamic feature space of path status, real-time extraction of the changing trends of path status features in spatial and temporal dimensions, and establishment of a path status trend feature prediction space.

[0060] S52. Predict the space using path state trend features, and generate path topology change sensitivity matrix and node state change sensitivity matrix through neural augmented ant colony optimization algorithm, which respectively represent the sensitivity of the path to future network topology and node performance changes.

[0061] S53. Based on the path topology change sensitivity matrix and the node state change sensitivity matrix, calculate the topology prediction weight factor and the node state prediction weight factor for each path, and adjust the topology prediction scale and node state prediction granularity of the neural augmented ant colony optimization algorithm prediction path set in real time accordingly.

[0062] S54. Real-time path prediction confidence assessment is performed on the predicted path set generated in advance by the neural reinforcement ant colony optimization algorithm. Specifically, the real-time prediction confidence of each path in the predicted path set is obtained by calculating the feature space deviation between the predicted path state vector and the network state trend prediction benchmark.

[0063] S55. Input the path position vectors in the predicted path set and the real-time prediction confidence value into the Grey Wolf optimization algorithm, and dynamically adjust the preset optimization scale and optimization direction of the path position vectors in the Grey Wolf optimization algorithm in real time according to the path prediction confidence, and perform preset optimization of the path position vectors.

[0064] S56. A multi-scale real-time feedforward correction mechanism for path position vectors is adopted. The feedforward correction magnitude of each dimension of the path position vector is calculated in advance based on the real-time predicted topology change trend and node state change trend, and the predicted path position vector is actively optimized by feedforward.

[0065] S57. After multiple rounds of active feedforward optimization and adjustment of the path position vector, the optimal path is finally obtained by ranking the path position vector in real time according to the multidimensional matching degree between the path position vector and the network state trend prediction benchmark.

[0066] Optionally, S52 specifically includes:

[0067] S521. Real-time spatiotemporal joint feature identification data of node congestion index, link quality coefficient and network load balance in the prediction space based on path state trend characteristics, construct the coupling sensitivity prediction matrix of path topology state and node state, and dynamically represent the bidirectional interactive coupling degree of path to network topology changes and node state changes.

[0068] S522. Using the coupling sensitivity prediction matrix of path topology state and node state, calculate in real time the feedforward influence coefficient of path topology state change trend on node state prediction, and the feedback influence coefficient of node state change trend on path topology state prediction.

[0069] S523. Using the neural reinforcement ant colony optimization algorithm, the feedforward influence coefficient of topology state on node state and the feedback influence coefficient of node state on topology state are used as real-time dynamic weighting factors. The sensitivity feature values ​​of topology structure change are dynamically weighted and fused for each path to generate a path topology change sensitivity matrix that can characterize the degree of proactive response of the path to future network topology changes.

[0070] S524. Using the neural reinforcement ant colony optimization algorithm, the cross-fusion analysis of the state change trend of the path node itself and the topological state change trend of the neighbor node is used to dynamically calculate the neighbor topological cross sensitivity of the node state for each node, and form a node state change sensitivity matrix in real time that can characterize the degree of active response of the node to the future node state change trend.

[0071] S525. Real-time construction of an interactive feature verification mechanism between the path topology change sensitivity matrix and the node state change sensitivity matrix. By verifying the spatial feature difference between the sensitivity features of the path topology change sensitivity matrix and the node state change sensitivity matrix, the feature values ​​of the path topology change sensitivity matrix and the node state change sensitivity matrix are dynamically and in real time corrected.

[0072] S526. Based on the real-time corrected path topology change sensitivity matrix and node state change sensitivity matrix, a joint sensitivity evaluation vector of the path to future network topology and node state changes is dynamically generated through the fusion operation of the feature matrix.

[0073] Optionally, S6 specifically includes:

[0074] S61. The optimized optimal path and at least two alternative paths are converted into multi-dimensional path forwarding description data in real time, carrying path status prediction and early warning values, real-time topology stability indicators and node performance dynamic indicators, and are simultaneously distributed to all forwarding nodes in the network control plane.

[0075] S62. Establish an adaptive dynamic parallel allocation mechanism for data packet transmission paths in real time. Based on the state prediction warning value and topology stability identifier in the forwarding description data of each path, calculate the forwarding weight of data packets to the optimal path and alternative paths in real time, and dynamically adjust the forwarding ratio of data packets between each path.

[0076] S63. Real-time acquisition of instantaneous transmission performance characteristics of data packets on the optimal path and alternative paths, including link delay change gradient, spatial distribution probability of data packet loss, and instantaneous utilization saturation rate of path bandwidth.

[0077] S64. Construct a spatiotemporal multi-scale difference analysis model for the actual performance characteristics of the path in real time. By comparing the real-time changing trends of the multi-dimensional performance characteristics of the optimal path and the alternative paths, a multi-scale spatiotemporal joint evaluation matrix of path performance differences is dynamically generated.

[0078] S65. The multi-scale spatiotemporal joint evaluation matrix of path performance differences is used in real time to perform dynamic weighted fusion operation on the path search history memory weight matrix in the neural augmented ant colony optimization algorithm, and the path search history memory weight matrix is ​​updated through real-time feature fusion.

[0079] S66. The adaptive adjustment factor of the path selection probability space is calculated in real time based on the dynamically updated path search history memory weight matrix. This dynamically drives the neural enhanced ant colony optimization algorithm to adjust the path search direction and the probability distribution of path generation in real time, thus completing the algorithm's online continuous adaptive update.

[0080] The beneficial effects of this invention are:

[0081] (1) This invention constructs a time-series dynamic matrix of nodes and links in real time and generates alternative paths based on the neural reinforcement ant colony optimization algorithm, thereby realizing efficient perception of multi-dimensional network status information and accurate routing decision-making. This effectively improves the path search efficiency and algorithm convergence speed, and enhances the real-time performance and adaptability of routing methods in complex network environments.

[0082] (2) By constructing a multidimensional composite fitness function and combining it with the gray wolf optimization algorithm for multi-scale iterative updates of paths, this invention can achieve dynamic and precise scheduling and load balancing of network resources, significantly improve path stability and link quality during data transmission, and exhibit better adaptability and performance stability in complex network environments with frequent topology changes.

[0083] (3) In terms of network state prediction and path optimization, this invention effectively solves the shortcomings of existing technologies in responding to network state changes in a timely manner by actively predicting topology and node states and pre-setting the predicted path set in advance. It breaks through the real-time response bottleneck of existing routing methods in dynamic network environments, achieves a significant improvement in network performance, and thus effectively improves the reliability and efficiency of data transmission in complex network environments.

[0084] (4) This invention constructs a multi-scale spatiotemporal joint evaluation matrix of path performance differences, and uses the evaluation matrix to drive the online adaptive update of the neural reinforcement ant colony optimization algorithm in real time. This effectively realizes the real-time optimization and dynamic adaptation of the path selection strategy, significantly reduces network resource redundancy and data forwarding delay, and improves the overall network resource utilization and overall data routing performance. Attached Figure Description

[0085] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0086] Figure 1 This is a schematic diagram of the overall technical process of a non-directly connected data routing method proposed in this invention;

[0087] Figure 2 This is a schematic diagram illustrating the network state time-series dynamic matrix construction process of a non-directly connected data routing method proposed in this invention;

[0088] Figure 3 This is a schematic diagram of the path active prediction and adaptive optimization process of a data routing method for non-direct connections proposed in this invention. Detailed Implementation

[0089] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0090] refer to Figures 1-3 A data routing method for non-directly connected networks includes the following steps:

[0091] S1. Real-time collection of topology, link bandwidth, latency, load and energy consumption status of each node in the network, and construction of a time-series dynamic matrix of dual node and link status;

[0092] S2. Based on the time-series dynamic matrix, an ant colony optimization method with neural network adaptive enhancement is adopted. By having individual ants perform asynchronous multi-dimensional state information perception search within the network, a set of alternative paths is automatically generated.

[0093] S3. Generate the initial population of the Grey Wolf optimization algorithm using alternative paths, and construct a multi-dimensional composite fitness function based on the node congestion index, link quality coefficient and network load balance degree on each path.

[0094] S4. The multi-scale iterative update of the path is carried out based on the multi-dimensional composite fitness function-driven gray wolf optimization algorithm. In each iteration, the matching degree between each path solution and the network dynamic matrix is ​​checked in real time, and the advantageous path is continuously screened and iteratively retained.

[0095] S5. An active prediction mechanism based on the dominant path is introduced, which introduces network topology and node status. The neural reinforcement ant colony optimization algorithm is used to generate a set of predicted paths in advance, and the gray wolf optimization algorithm is used to pre-optimize the paths in the set of predicted paths in advance to obtain the optimal path.

[0096] S6. Simultaneously send the optimized optimal path and at least two alternative paths to the network control plane for parallel data forwarding, and use the actual performance differences of the paths as a basis to drive the neural reinforcement ant colony optimization algorithm for online continuous adaptive updates.

[0097] By constructing a multi-dimensional time-series dynamic matrix of nodes and links in real time, and adopting a path generation and optimization method that combines ant colony optimization and gray wolf optimization algorithms enhanced by neural networks, the autonomous selection of network paths and resource allocation can be achieved efficiently. At the same time, the introduction of an active prediction mechanism to optimize paths in advance further improves the stability and adaptability of paths, effectively enhances the overall resource utilization of the network and the real-time response capability of data routing, and improves the problems of difficult path selection and low resource utilization in non-directly connected network scenarios of traditional methods.

[0098] In this embodiment, S1 specifically includes:

[0099] S11. Perform dual-dimensional joint encoding of node role and position for all nodes in the network. Based on the actual functional positioning and topological relative position of the node in the network, assign a unique and dynamically updated multi-dimensional feature identifier to each node to form a node topological role matrix.

[0100] S12. Utilize a traffic-aware link bandwidth adaptive measurement mechanism to automatically increase the measurement frequency when the link load change rate exceeds a preset threshold. Collect data in real time and quantify bandwidth characteristics in three dimensions, including the effective transmission rate of the link, the link utilization rate, and the proportion of idle bandwidth, to construct a link bandwidth status matrix.

[0101] S13. By pre-embedding clock synchronization units in each node, nodes alternately send and receive delay calibration beacon data frames within each preset interval. Combine the timestamp information of the sent and received beacon data frames to perform real-time bidirectional delay difference calculation and automatic compensation, and construct a link delay matrix after dynamic self-calibration.

[0102] S14. Integrate data cache monitoring sensor probes into the data processing unit of each node to detect and record the cache queue data length, average data packet waiting time and node data throughput in real time and time period, and construct a multi-dimensional quantitative matrix of node real-time load status.

[0103] S15. Integrate a high-sensitivity digital current and voltage real-time sensing unit inside the node chip to continuously measure the current operating power consumption of each node and record the transient power fluctuations in time sequence, thereby constructing a dynamic high-resolution digital matrix of node energy consumption status.

[0104] S16. Based on the timestamp marking synchronization mechanism, and in accordance with a unified time series benchmark, the node topology role matrix, link bandwidth status matrix, link delay matrix, node real-time load status multidimensional quantization matrix, and node energy consumption status dynamic high-resolution digital matrix are fused element by element to form a time series dynamic matrix of dual node and link status.

[0105] S17. Utilizing the network real-time state difference adaptive feedback mechanism, based on the continuous monitoring and prediction of the real-time change amplitude and trend of each element in the time-series dynamic matrix, the real-time update frequency of the time-series dynamic matrix is ​​adjusted, and when the change amplitude of matrix elements exceeds the preset threshold, a real-time event-type update of the data in the time-series dynamic matrix is ​​triggered.

[0106] By implementing joint coding of node roles and locations, adaptive measurement of link bandwidth, dynamic calibration of node latency, multidimensional quantification of real-time load status, and high-resolution digital monitoring of node energy consumption, and fusing them into a time-series dynamic matrix of dual node and link status based on a unified time-series benchmark, real-time, high-precision, and dynamic acquisition and processing of network status data is achieved. This effectively improves the accuracy and timeliness of status information and enhances the perception and adaptability of data routing methods to changes in network status.

[0107] In this embodiment, S2 specifically includes:

[0108] S21. Based on the time-series dynamic matrix, the network is dynamically divided into multiple network-aware regions according to the real-time load differences and topological relationships of the nodes, and ant colony individuals with region-specific identifiers are deployed in each region to perform path search.

[0109] S22. A real-time path status sensing probe is built into each node to measure the link bandwidth change rate, latency dynamic fluctuation, and node load difference between the node and all adjacent nodes on a cycle-by-cycle basis. Combined with a feedforward neural network model, a jump node priority sequence is generated in real time.

[0110] S23. Individual ant colonies in each region periodically exchange current path exploration data, using an asynchronous interactive data sharing mechanism. Based on the real-time path exploration status data received from other regions, individual ant colonies dynamically adjust their next path exploration direction.

[0111] S24. When making node hopping decisions, individual ants in the colony dynamically adjust the probability distribution of the next hop node selection by calculating the link bandwidth change trend, link latency stability and node load fluctuation between the current node and the candidate node, and eliminate paths with unstable link status or excessive node load.

[0112] S25. Using a node and link search history state marking matrix, a cumulative memory weight matrix of path search history states is formed by recording and accumulating the characteristic information of node and link state changes in the past path search process of individual ants.

[0113] S26. After each ant colony individual completes the multi-hop path search, a composite evaluation index is established based on the network node load balance, the persistence of link status within the path, and the global network path redundancy. A set of global alternative paths that can simultaneously take into account path stability and network load balance is automatically generated.

[0114] By dividing the network into multiple sensing regions and deploying dedicated ant colonies for each region, real-time sensing of node path status and asynchronous interactive data sharing among ant colonies are achieved. The probability of node jump decisions is dynamically adjusted, and a cumulative memory matrix of historical path search status is constructed. This enables regional collaboration and dynamic adaptive adjustment of ant colony-optimized path search, effectively improving path search accuracy and efficiency, and enhancing the overall performance of network path stability and network load balance.

[0115] In this embodiment, S3 specifically includes:

[0116] S31. For the global candidate paths, extract the feature data of node congestion index, link quality coefficient and network load balance degree of each path, and calculate the feature Euclidean distance between paths. Based on this distance, dynamically aggregate them into several path subgroups and assign a unique path clustering identifier to each subgroup.

[0117] S32. Integrate a real-time congestion monitoring unit within each node of the alternative path, collect the current data buffer queue length, data packet processing rate, and node buffer saturation level of each node, and calculate the real-time congestion index of the node by combining the above values ​​to form a node congestion index matrix.

[0118] S33. Perform continuous dynamic sampling of data streams for each link of the candidate path, and calculate and form a link quality coefficient matrix by measuring the instantaneous effective transmission rate, link delay fluctuation amplitude and link interruption frequency in real time.

[0119] S34. Based on the node congestion index matrix, calculate the degree of spatial distribution difference of the node congestion index in each path among the path nodes, and obtain the network load balance by analyzing the spatial distribution entropy of the congestion index among the path nodes.

[0120] S35. Based on the above node congestion index matrix, link quality coefficient matrix and network load balance, construct a three-dimensional feature tensor of the path state space, and use high-dimensional space projection mapping to map the feature tensor to the state feature vector space so as to facilitate the comparison of feature values ​​of the state vectors.

[0121] S36. Construct a multi-dimensional composite fitness function based on the node congestion index matrix, link quality coefficient matrix, and network load balancing degree:

[0122]

[0123] Among them, F p The multidimensional composite fitness function value represents the path, where n represents the total number of nodes in the path, m represents the total number of links in the path, and α represents the path value. i The adaptive weighting coefficients of the congestion index of the i-th node satisfy the following condition: Q i T represents the length of the data cache queue for the i-th node. i R represents the packet processing rate of the i-th node. i β represents the cache saturation level of the i-th node. j Let the adaptive weight coefficient of the quality coefficient of the j-th link satisfy the following condition: B j D represents the instantaneous effective transmission rate of the j-th link. j V represents the average delay of the j-th link. j M represents the variance of the delay fluctuation of the j-th link. j p represents the interruption frequency of the j-th link, γ represents the weighting coefficient of network load balancing, and its value ranges from (0,1]. i This represents the proportion of the congestion index of the i-th node to the sum of the congestion indices of all nodes in the path, satisfying the following condition:

[0124] The proposed multidimensional composite fitness function integrates feature data from three dimensions: node congestion index, link quality coefficient, and network load balancing. It constructs node-dimensional and link-dimensional fitness sub-functions based on node congestion status and link quality characteristics, respectively, and introduces network load balancing entropy as a third-dimensional adaptive adjustment factor to achieve a comprehensive evaluation of path quality. The formula uses adaptive weighting coefficients to weight and fuse the state parameters of each node and link, effectively reflecting the real-time state characteristics differences of nodes and links in the path. Furthermore, the introduction of entropy enhances the constraint effect of load balancing on path fitness, thereby improving the rationality and accuracy of path selection.

[0125] S37. Sort the paths of each path subgroup according to the multidimensional composite fitness function value, and select the top-ranked paths from the sorting results to form the initial population of the Grey Wolf Optimization Algorithm.

[0126] By extracting the node congestion index, link quality coefficient, and network load balancing degree of the path, a multidimensional composite fitness function is constructed. The path feature dynamic clustering method is used to divide the path into subgroups and calculate the node congestion status and link quality characteristics in the path in real time. At the same time, a three-dimensional feature tensor is established and high-dimensional space mapping is implemented, which effectively realizes the accuracy of path fitness evaluation and dynamic optimization of path selection, and significantly improves the path congestion control capability, link transmission quality, and overall network load balancing effect.

[0127] In this embodiment, S4 specifically includes:

[0128] S41. Based on the initial population of the Grey Wolf Optimization Algorithm, and combined with the real-time changing trends of node congestion index, link quality coefficient and load balance in the network state matrix, a multi-scale state space is divided, and a unique scale state identifier is assigned to the path in each scale state space.

[0129] S42. In each scale state space, based on the real-time trends of node load, link quality and congestion state changes in the network state matrix, an adaptive scale perturbation method for path position vectors is used to generate a scale-adaptive updated path set that matches the real-time network state.

[0130] S43. After each scale-adaptive update of the path set, collect the current node congestion index matrix, link quality coefficient matrix and network load balance data in real time, and calculate the fitness value of the path at the current scale for each path based on the multidimensional composite fitness function.

[0131] S44. Establish a feature matching degree verification mechanism for the path and network dynamic matrix in real time. By calculating the matching degree values ​​of the path location vector and the network dynamic matrix in the three-dimensional space of node congestion index feature difference, link quality feature difference and load balancing feature difference, the matching degree of the real-time status of the path and the network is determined.

[0132] S45. Construct a two-dimensional joint evaluation index using path fitness values ​​and path real-time feature matching values. Update the path set at the current scale according to the joint evaluation index, mark the paths with the highest joint evaluation index as advantageous paths and retain them, and eliminate the paths with the lowest ranking in real time.

[0133] S46. After each iteration update, the state feature data of the retained advantageous path is integrated into the historical state feature memory of the network state matrix in real time. The dynamic accumulation of historical state feature memory data actively guides the update direction of the path position vector at the next scale.

[0134] S47. Repeat steps S41 to S46 until the number of iterations of the Grey Wolf optimization algorithm reaches the preset number or the path fitness value reaches the convergence threshold, and finally obtain a set of advantageous paths.

[0135] By constructing a multi-scale state space for paths, implementing adaptive scale dynamic perturbation of path position vectors, and calculating path fitness and feature matching degree in real time, a two-dimensional joint evaluation index is constructed to achieve real-time path screening and retention of advantageous paths. At the same time, a historical memory mechanism for path state features is established to dynamically guide path optimization and updates, effectively realizing the efficiency and real-time performance of path iterative optimization, enhancing the algorithm's rapid convergence and accurate matching ability of network states, and significantly improving the quality of path optimization and the overall performance of the network.

[0136] In this embodiment, S42 specifically includes:

[0137] S421. Based on the multidimensional trend change characteristics of node load, link quality and congestion status in the network state matrix, the path position vector dimension space is divided into several independent scale subspaces, and the path position vector dimension in each scale subspace is assigned a scale subspace identifier that is dynamically updated in real time.

[0138] S422. For each dimension of the path position vector in each scale subspace, calculate the instantaneous change gradient of the feature data of the corresponding dimension of the network state matrix in real time, and obtain the initial perturbation amplitude of each dimension by calculating the spatial difference of the feature change gradient values.

[0139] S423. Establish an interactive perturbation amplitude propagation mechanism for feature changes between the dimensional scale subspaces of the path position vector, dynamically calculate and execute the interactive coupling of perturbation amplitudes between different dimensions in real time based on the initial perturbation amplitude, and realize the dynamic joint perturbation of the path position vector in multiple dimensions.

[0140] S424. After completing the joint perturbation update, calculate the feature difference between the feature change of each dimension in the path position vector and the corresponding dimension in the real-time network state matrix, and form the path position vector feature update matching degree matrix dimension by dimension.

[0141] S425. Update the real-time feature matching error of each dimension in the matching degree matrix based on the path position vector features, determine the correction direction and magnitude of each path position vector dimension in real time, and dynamically perform multi-dimensional scale backtracking correction of each dimension of the path position vector.

[0142] S426. Based on the overall feature matching between the real-time corrected path position vector and the current network state matrix, generate a set of scale-adaptive updated paths that are highly matched with the real-time dynamic state of the network.

[0143] By establishing a scale subspace for path position vectors and implementing dynamic interactive coupling of dimensional feature gradient difference calculation and perturbation amplitude, adaptive multi-dimensional scale joint perturbation of path position vectors is achieved. At the same time, the path position vectors are corrected in real time through dimensional feature matching error, forming a path set that is highly matched with the real-time dynamic state of the network. This effectively improves the accuracy and dynamic adaptability of path optimization and enhances the algorithm's response and adaptability to real-time changes in network state.

[0144] In this embodiment, S46 specifically includes:

[0145] S461. After each iteration of the Grey Wolf optimization algorithm, the node congestion index, link quality coefficient, and load balance status data of the retained advantageous paths are jointly encoded in terms of time and space dimensions to generate a real-time spatiotemporal joint feature identifier matrix of path status features.

[0146] S462. Based on the real-time spatiotemporal joint feature identifier matrix of the path state features, calculate the feature space cosine similarity between each path and the corresponding feature data in the historical state feature memory to obtain the real-time matching degree value between the path features and the historical state data.

[0147] S463. Construct a feature fusion weight matrix between path state feature data and historical state feature memory based on real-time matching degree value, dynamically fuse path state feature data into historical state feature memory through weighted average method, and update historical state feature memory in real time.

[0148] S464. Introduce a dynamic time decay mechanism for feature data in the historical state feature memory bank. Assign a time decay coefficient to the real-time dwell time of each state feature data in the memory bank and automatically update the weight of historical state feature data by gradually decreasing the feature weight as time increases.

[0149] S465. Using the historical state feature memory after time decay, extract the cumulative change trend of three dimensions of feature data: node congestion index, link quality coefficient and load balance in real time, and generate multi-dimensional trend guidance vectors for the next scale path location vector update respectively.

[0150] S466. Based on multi-dimensional trend guidance vectors, feedforward active pre-regulation is implemented on the update direction of the path position vector at the next scale. Through real-time prediction and active adjustment of feature trends, the path position vector is guided to actively adapt to the dynamic change trend of the network's real-time state.

[0151] By constructing a real-time spatiotemporal joint encoding of path state features and a historical state feature memory, real-time path matching degree calculation and weighted fusion update are implemented. A dynamic time decay mechanism is introduced to adjust the weight of historical data in real time, so as to realize the active guidance and dynamic pre-control of the path update direction by historical state data. This effectively improves the prediction accuracy and real-time response capability of the algorithm, and significantly enhances the adaptability and stability of network path selection and optimization to complex dynamic environments.

[0152] In this embodiment, S5 specifically includes:

[0153] S51. Real-time collection of congestion index, link quality coefficient and network load balance status of each node in the advantageous path, construction of a multi-dimensional dynamic feature space of path status, real-time extraction of the changing trends of path status features in spatial and temporal dimensions, and establishment of a path status trend feature prediction space.

[0154] S52. Predict the space using path state trend features, and generate path topology change sensitivity matrix and node state change sensitivity matrix through neural augmented ant colony optimization algorithm, which respectively represent the sensitivity of the path to future network topology and node performance changes.

[0155] S53. Based on the path topology change sensitivity matrix and the node state change sensitivity matrix, calculate the topology prediction weight factor and the node state prediction weight factor for each path, and adjust the topology prediction scale and node state prediction granularity of the neural augmented ant colony optimization algorithm prediction path set in real time accordingly.

[0156] S54. Real-time path prediction confidence assessment is performed on the predicted path set generated in advance by the neural reinforcement ant colony optimization algorithm. Specifically, the real-time prediction confidence of each path in the predicted path set is obtained by calculating the feature space deviation between the predicted path state vector and the network state trend prediction benchmark.

[0157] S55. Input the path position vectors in the predicted path set and the real-time prediction confidence value into the Grey Wolf optimization algorithm, and dynamically adjust the preset optimization scale and optimization direction of the path position vectors in the Grey Wolf optimization algorithm in real time according to the path prediction confidence, and perform preset optimization of the path position vectors.

[0158] S56. A multi-scale real-time feedforward correction mechanism for path position vectors is adopted. The feedforward correction magnitude of each dimension of the path position vector is calculated in advance based on the real-time predicted topology change trend and node state change trend, and the predicted path position vector is actively optimized by feedforward.

[0159] S57. After multiple rounds of active feedforward optimization and adjustment of the path position vector, the optimal path is finally obtained by ranking the path position vector in real time according to the multidimensional matching degree between the path position vector and the network state trend prediction benchmark.

[0160] By constructing a multidimensional dynamic feature space of path states and generating path topology and node state sensitivity matrices in real time, the prediction scale of path topology and the granularity of node state are dynamically adjusted. At the same time, the confidence of the predicted path set is evaluated in real time, and dynamic pre-optimization and multi-scale feedforward correction of the Grey Wolf Optimization Algorithm are implemented to finally obtain the optimal path. This effectively improves the prediction accuracy of network topology and node state changes and the effect of path pre-optimization, and significantly enhances the forward-looking response capability of data routing methods to changes in the dynamic network environment.

[0161] In this embodiment, S52 specifically includes:

[0162] S521. Real-time spatiotemporal joint feature identification data of node congestion index, link quality coefficient and network load balance in the prediction space based on path state trend characteristics, construct the coupling sensitivity prediction matrix of path topology state and node state, and dynamically represent the bidirectional interactive coupling degree of path to network topology changes and node state changes.

[0163] S522. Using the coupling sensitivity prediction matrix of path topology state and node state, calculate in real time the feedforward influence coefficient of path topology state change trend on node state prediction, and the feedback influence coefficient of node state change trend on path topology state prediction.

[0164] S523. Using the neural reinforcement ant colony optimization algorithm, the feedforward influence coefficient of topology state on node state and the feedback influence coefficient of node state on topology state are used as real-time dynamic weighting factors. The sensitivity feature values ​​of topology structure change are dynamically weighted and fused for each path to generate a path topology change sensitivity matrix that can characterize the degree of proactive response of the path to future network topology changes.

[0165] S524. Using the neural reinforcement ant colony optimization algorithm, the cross-fusion analysis of the state change trend of the path node itself and the topological state change trend of the neighbor node is used to dynamically calculate the neighbor topological cross sensitivity of the node state for each node, and form a node state change sensitivity matrix in real time that can characterize the degree of active response of the node to the future node state change trend.

[0166] S525. Real-time construction of an interactive feature verification mechanism between the path topology change sensitivity matrix and the node state change sensitivity matrix. By verifying the spatial feature difference between the sensitivity features of the path topology change sensitivity matrix and the node state change sensitivity matrix, the feature values ​​of the path topology change sensitivity matrix and the node state change sensitivity matrix are dynamically and in real time corrected.

[0167] S526. Based on the real-time corrected path topology change sensitivity matrix and node state change sensitivity matrix, a joint sensitivity evaluation vector of the path to future network topology and node state changes is dynamically generated through the fusion operation of the feature matrix.

[0168] By constructing a coupled sensitivity prediction matrix of path topology state and node state, the feedforward and feedback influence coefficients between the trends of topology state and node state changes are calculated in real time. Dynamic weights are used to fuse and generate path topology change sensitivity matrix and node state change sensitivity matrix. At the same time, the spatial feature difference between the two is verified, the sensitivity matrix is ​​dynamically corrected and a joint sensitivity evaluation vector is formed. This effectively realizes accurate prediction and proactive response to the trends of network path and node state changes, and significantly improves the adaptability and accuracy of data routing to complex network state changes.

[0169] In this embodiment, S6 specifically includes:

[0170] S61. The optimized optimal path and at least two alternative paths are converted into multi-dimensional path forwarding description data in real time, carrying path status prediction and early warning values, real-time topology stability indicators and node performance dynamic indicators, and are simultaneously distributed to all forwarding nodes in the network control plane.

[0171] S62. Establish an adaptive dynamic parallel allocation mechanism for data packet transmission paths in real time. Based on the state prediction warning value and topology stability identifier in the forwarding description data of each path, calculate the forwarding weight of data packets to the optimal path and alternative paths in real time, and dynamically adjust the forwarding ratio of data packets between each path.

[0172] S63. Real-time acquisition of instantaneous transmission performance characteristics of data packets on the optimal path and alternative paths, including link delay change gradient, spatial distribution probability of data packet loss, and instantaneous utilization saturation rate of path bandwidth.

[0173] S64. Construct a spatiotemporal multi-scale difference analysis model for the actual performance characteristics of the path in real time. By comparing the real-time changing trends of the multi-dimensional performance characteristics of the optimal path and the alternative paths, a multi-scale spatiotemporal joint evaluation matrix of path performance differences is dynamically generated.

[0174] S65. The multi-scale spatiotemporal joint evaluation matrix of path performance differences is used in real time to perform dynamic weighted fusion operation on the path search history memory weight matrix in the neural augmented ant colony optimization algorithm, and the path search history memory weight matrix is ​​updated through real-time feature fusion.

[0175] S66. The adaptive adjustment factor of the path selection probability space is calculated in real time based on the dynamically updated path search history memory weight matrix. This dynamically drives the neural enhanced ant colony optimization algorithm to adjust the path search direction and the probability distribution of path generation in real time, thus completing the algorithm's online continuous adaptive update.

[0176] By constructing multi-dimensional path forwarding description data in real time, dynamically calculating the path forwarding ratio of data packets, monitoring path transmission performance in real time, and generating a multi-scale spatiotemporal joint evaluation matrix, and then merging and updating the path search historical memory weight matrix in real time, the algorithm's path search direction and generation probability are adjusted online adaptively. This effectively improves the real-time response capability and accuracy of network path selection and optimization, and significantly enhances the adaptability of data routing to real-time network state changes and its dynamic performance optimization capability.

[0177] Example 1:

[0178] To verify the feasibility of this invention in practice, it was applied to a data transmission optimization scenario of a large-scale communication network operator, particularly for data routing tasks in network environments where there are few direct connections between network nodes and the topology and traffic load change frequently. In actual network operation, traditional static routing or general heuristic optimization algorithms show significant shortcomings, including slow convergence speed, poor routing adaptability, and low network resource utilization. For example, in complex topologies where nodes are not directly connected, traditional ant colony optimization methods typically require more than 50 iterations to converge, with average latency fluctuations exceeding 15% and average packet loss rates exceeding 3.2%, making it difficult to meet the real-time and stability requirements of the network.

[0179] To address the aforementioned issues, this embodiment specifically applies the non-directly connected data routing method proposed in this invention. First, the topology, link bandwidth, latency, node load, and energy consumption status of all nodes in the network are collected in real time. The network operating status is reflected in real-time through a time-series dynamic matrix of both node and link states. Then, the neural-enhanced ant colony optimization algorithm of this invention is used to quickly and efficiently generate a set of alternative paths adapted to the current network state. Specifically, this algorithm employs a fully connected neural network model with three hidden layers and 128, 64, and 32 nodes per layer, respectively. The heuristic weights of the ant colony path search are automatically adjusted using time-series matrix information, and convergence in the path generation stage is achieved after 10 iterations.

[0180] Next, this embodiment imports the aforementioned candidate paths into the Gray Wolf optimization algorithm. A multi-dimensional composite fitness function is constructed based on three indicators: node congestion index, link quality coefficient, and network load balancing. The path is then iteratively updated and dynamically adjusted at multiple scales. In this embodiment, the Gray Wolf optimization algorithm employs a multi-scale update mechanism during its iterative process, with a total of 20 main iterations, each containing 5 scale update steps, completing a total of 100 path optimization iterations. In actual optimization, the convergence speed of the path fitness function is improved by approximately 60% compared to traditional methods, and the time to determine the optimal path is reduced to about 40% of the original time.

[0181] Regarding path optimization and network state prediction, this embodiment employs the proactive prediction mechanism designed in this invention. By predicting network topology and node performance change trends in advance, it proactively optimizes the path set. In specific implementation, the prediction model constructs a path topology change sensitivity matrix and a node state sensitivity matrix using historical change data of node state and link quality. This allows for accurate prediction and advance optimization of network topology changes and node states within the next 30 minutes. In a real network operation environment, through this mechanism, the average path response delay time after a sudden network state change is reduced to 25% of its original value, effectively mitigating the impact of sudden topology changes on network performance.

[0182] To further improve the performance and stability of data routing, this embodiment also simultaneously distributes the optimized optimal path and two additional alternative paths to the network control plane, enabling dynamic parallel forwarding of data packets. The forwarding ratio of data packets is adjusted in real time based on path status prediction warning values ​​and topology stability to ensure data transmission stability and efficiency. Simultaneously, by real-time monitoring of data packet transmission latency, data packet loss probability, and path bandwidth saturation, a multi-scale spatiotemporal joint evaluation matrix for path performance differences is constructed. This evaluation matrix is ​​then used to drive the adaptive online update of the neural reinforcement ant colony optimization algorithm in real time, enabling the algorithm to continuously optimize its path selection strategy.

[0183] During implementation, the network operation was monitored for 24 consecutive hours and compared with the traditional ant colony routing optimization method. The results are shown in Table 1.

[0184] Table 1 Comparison of routing performance between the present invention and traditional methods

[0185] Performance indicators Traditional ant colony optimization algorithm Method of the present invention Improvement effect Average path convergence iterations 52 times 16 times 69.2％↓ Average route selection time 8.3 seconds 3.1 seconds 62.7％↓ Average packet loss rate 3.21％ 0.74％ 76.9％↓ Link average latency fluctuation 15.4％ 3.6％ 76.6％↓ Bandwidth utilization efficiency 68.2％ 87.5％ 19.3％↑

[0186] As shown in the detailed data analysis in Table 1 above, the method of this invention significantly reduces the number of convergence iterations and the path selection time. Especially in terms of path convergence speed, the average number of iterations is reduced by approximately 69.2%, significantly improving the efficiency of network path decision-making. Simultaneously, the actual execution time of path selection is significantly shortened from 8.3 seconds to 3.1 seconds, an improvement of up to 62.7%, greatly meeting the real-time response requirements of network operation.

[0187] Regarding the stability and reliability of data transmission, the data packet loss rate of this invention is reduced to 0.74%, a significant reduction of 76.9% compared to the 3.21% of the traditional method, effectively ensuring the integrity of data transmission. Furthermore, the average volatility of link latency is reduced to 3.6%, a reduction of approximately 76.6% compared to the traditional method, improving the stability of link quality and ensuring the reliability of the data transmission process.

[0188] Regarding network resource utilization efficiency, the method of this invention achieves an average network bandwidth utilization efficiency of 87.5%, which is nearly 20 percentage points higher than the 68.2% of the traditional method. This improvement in bandwidth utilization efficiency further optimizes the overall utilization of network resources and effectively alleviates the problem of network resource redundancy.

[0189] Based on the real data from the above embodiments, it is evident that the data routing method for non-directly connected networks proposed in this invention demonstrates significant and objective advantages over existing methods in several key network performance indicators, including path optimization efficiency, data transmission stability, and resource utilization. Specifically, this manifests as faster path selection response speed, lower packet loss rate, more stable network latency, and more efficient bandwidth utilization, providing an effective and efficient technical solution for data transmission in complex network environments.

[0190] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A data routing method for non-directly connected networks, characterized in that, Includes the following steps: S1. Real-time collection of topology, link bandwidth, latency, load and energy consumption status of each node in the network, and construction of a time-series dynamic matrix of dual node and link status; S2. Based on the time-series dynamic matrix, an ant colony optimization method with neural network adaptive enhancement is adopted. By having individual ants perform asynchronous multi-dimensional state information perception search within the network, a set of alternative paths is automatically generated. S3. Generate the initial population of the Grey Wolf optimization algorithm using alternative paths, and construct a multi-dimensional composite fitness function based on the node congestion index, link quality coefficient and network load balance degree on each path. S4. The multi-scale iterative update of the path is carried out based on the multi-dimensional composite fitness function-driven gray wolf optimization algorithm. In each iteration, the matching degree between each path solution and the network dynamic matrix is ​​checked in real time, and the advantageous path is continuously screened and iteratively retained. S5. An active prediction mechanism based on the dominant path is introduced, which introduces network topology and node status. The neural reinforcement ant colony optimization algorithm is used to generate a set of predicted paths in advance, and the gray wolf optimization algorithm is used to pre-optimize the paths in the set of predicted paths in advance to obtain the optimal path. S6. Simultaneously send the optimized optimal path and at least two alternative paths to the network control plane for parallel data forwarding, and use the actual performance differences of the paths as a basis to drive the neural reinforcement ant colony optimization algorithm for online continuous adaptive updates.

2. The data routing method for non-direct connections according to claim 1, characterized in that, S1 specifically includes: S11. Perform dual-dimensional joint encoding of node role and position for all nodes in the network. Based on the actual functional positioning and topological relative position of the node in the network, assign a unique and dynamically updated multi-dimensional feature identifier to each node to form a node topological role matrix. S12. Utilize a traffic-aware link bandwidth adaptive measurement mechanism to automatically increase the measurement frequency when the link load change rate exceeds a preset threshold. Collect data in real time and quantify bandwidth characteristics in three dimensions, including the effective transmission rate of the link, the link utilization rate, and the proportion of idle bandwidth, to construct a link bandwidth status matrix. S13. By pre-embedding clock synchronization units in each node, nodes alternately send and receive delay calibration beacon data frames within each preset interval. Combine the timestamp information of the sent and received beacon data frames to perform real-time bidirectional delay difference calculation and automatic compensation, and construct a link delay matrix after dynamic self-calibration. S14. Integrate data cache monitoring sensor probes into the data processing unit of each node to detect and record the cache queue data length, average data packet waiting time and node data throughput in real time and time period, and construct a multi-dimensional quantitative matrix of node real-time load status. S15. Integrate a high-sensitivity digital current and voltage real-time sensing unit inside the node chip to continuously measure the current operating power consumption of each node and record the transient power fluctuations in time sequence, thereby constructing a dynamic high-resolution digital matrix of node energy consumption status. S16. Based on the timestamp marking synchronization mechanism, and in accordance with a unified time series benchmark, the node topology role matrix, link bandwidth status matrix, link delay matrix, node real-time load status multidimensional quantization matrix, and node energy consumption status dynamic high-resolution digital matrix are fused element by element to form a time series dynamic matrix of dual node and link status. S17. Utilizing the network real-time state difference adaptive feedback mechanism, based on the continuous monitoring and prediction of the real-time change amplitude and trend of each element in the time-series dynamic matrix, the real-time update frequency of the time-series dynamic matrix is ​​adjusted, and when the change amplitude of matrix elements exceeds the preset threshold, a real-time event-type update of the data in the time-series dynamic matrix is ​​triggered.

3. The data routing method for non-directly connected networks according to claim 1, characterized in that, S2 specifically includes: S21. Based on the time-series dynamic matrix, the network is dynamically divided into multiple network-aware regions according to the real-time load differences and topological relationships of the nodes, and ant colony individuals with region-specific identifiers are deployed in each region to perform path search. S22. A real-time path status sensing probe is built into each node to measure the link bandwidth change rate, latency dynamic fluctuation, and node load difference between the node and all adjacent nodes on a cycle-by-cycle basis. Combined with a feedforward neural network model, a jump node priority sequence is generated in real time. S23. Individual ant colonies in each region periodically exchange current path exploration data, using an asynchronous interactive data sharing mechanism. Based on the real-time path exploration status data received from other regions, individual ant colonies dynamically adjust their next path exploration direction. S24. When making node hopping decisions, individual ants in the colony dynamically adjust the probability distribution of the next hop node selection by calculating the link bandwidth change trend, link latency stability and node load fluctuation between the current node and the candidate node, and eliminate paths with unstable link status or excessive node load. S25. Using a node and link search history state marking matrix, a cumulative memory weight matrix of path search history states is formed by recording and accumulating the characteristic information of node and link state changes in the past path search process of individual ants. S26. After each ant colony individual completes the multi-hop path search, a composite evaluation index is established based on the network node load balance, the persistence of link status within the path, and the global network path redundancy. A set of global alternative paths that can simultaneously take into account path stability and network load balance is automatically generated.

4. The data routing method for non-directly connected networks according to claim 1, characterized in that, S3 specifically includes: S31. For the global candidate paths, extract the feature data of node congestion index, link quality coefficient and network load balance degree of each path, and calculate the feature Euclidean distance between paths. Based on this distance, dynamically aggregate them into several path subgroups and assign a unique path clustering identifier to each subgroup. S32. Integrate a real-time congestion monitoring unit within each node of the alternative path, collect the current data buffer queue length, data packet processing rate, and node buffer saturation level of each node, and calculate the real-time congestion index of the node by combining the above values ​​to form a node congestion index matrix. S33. Perform continuous dynamic sampling of data streams for each link of the candidate path, and calculate and form a link quality coefficient matrix by measuring the instantaneous effective transmission rate, link delay fluctuation amplitude and link interruption frequency in real time. S34. Based on the node congestion index matrix, calculate the degree of spatial distribution difference of the node congestion index in each path among the path nodes, and obtain the network load balance by analyzing the spatial distribution entropy of the congestion index among the path nodes. S35. Based on the above node congestion index matrix, link quality coefficient matrix and network load balance, construct a three-dimensional feature tensor of the path state space, and use high-dimensional space projection mapping to map the feature tensor to the state feature vector space so as to facilitate the comparison of feature values ​​of the state vectors. S36. Construct a multidimensional composite fitness function based on the node congestion index matrix, link quality coefficient matrix, and network load balancing degree. S37. Sort the paths of each path subgroup according to the multidimensional composite fitness function value, and select the top-ranked paths from the sorting results to form the initial population of the Grey Wolf Optimization Algorithm.

5. A data routing method for non-directly connected networks according to claim 1, characterized in that, S4 specifically includes: S41. Based on the initial population of the gray wolf optimization algorithm, and combined with the real-time changing trends of node congestion index, link quality coefficient and load balance in the network state matrix, multi-scale state space is divided, and a unique scale state identifier is assigned to the path in each scale state space. S42. In each scale state space, based on the real-time trends of node load, link quality and congestion state changes in the network state matrix, an adaptive scale perturbation method for path position vectors is used to generate a scale-adaptive updated path set that matches the real-time network state. S43. After each scale-adaptive update of the path set, collect the current node congestion index matrix, link quality coefficient matrix and network load balance data in real time, and calculate the fitness value of the path at the current scale for each path based on the multidimensional composite fitness function. S44. Establish a feature matching degree verification mechanism for the path and network dynamic matrix in real time. By calculating the matching degree values ​​of the path location vector and the network dynamic matrix in the three-dimensional space of node congestion index feature difference, link quality feature difference and load balancing feature difference, the matching degree of the real-time status of the path and the network is determined. S45. Construct a two-dimensional joint evaluation index using path fitness values ​​and real-time path feature matching values. Update the path set at the current scale according to the joint evaluation index, mark the paths with the highest joint evaluation index as advantageous paths and retain them, and eliminate the paths with the lowest ranking in real time. S46. After each iteration update, the state feature data of the retained advantageous path is integrated into the historical state feature memory of the network state matrix in real time. The dynamic accumulation of historical state feature memory data actively guides the update direction of the path position vector at the next scale. S47. Repeat steps S41 to S46 until the number of iterations of the Grey Wolf optimization algorithm reaches the preset number or the path fitness value reaches the convergence threshold, and finally obtain a set of advantageous paths.

6. A data routing method for non-directly connected networks according to claim 5, characterized in that, S42 specifically includes: S421. Based on the multidimensional trend change characteristics of node load, link quality and congestion status in the network state matrix, the path position vector dimension space is divided into several independent scale subspaces, and the path position vector dimension in each scale subspace is assigned a scale subspace identifier that is dynamically updated in real time. S422. For each dimension of the path position vector in each scale subspace, calculate the instantaneous change gradient of the feature data of the corresponding dimension of the network state matrix in real time, and obtain the initial perturbation amplitude of each dimension by calculating the spatial difference of the feature change gradient values. S423. Establish an interactive perturbation amplitude propagation mechanism for feature changes between the dimensional scale subspaces of the path position vector, dynamically calculate and execute the interactive coupling of perturbation amplitudes between different dimensions in real time based on the initial perturbation amplitude, and realize the dynamic joint perturbation of the path position vector in multiple dimensions. S424. After completing the joint perturbation update, calculate the feature difference between the feature change of each dimension in the path position vector and the corresponding dimension in the real-time network state matrix, and form the path position vector feature update matching degree matrix dimension by dimension. S425. Update the real-time feature matching error of each dimension in the matching degree matrix based on the path position vector features, determine the correction direction and magnitude of each path position vector dimension in real time, and dynamically perform multi-dimensional scale backtracking correction of each dimension of the path position vector. S426. Based on the overall feature matching between the real-time corrected path position vector and the current network state matrix, generate a set of scale-adaptive updated paths that are highly matched with the real-time dynamic state of the network.

7. A data routing method for non-directly connected networks according to claim 5, characterized in that, Specifically, S46 includes: S461. After each iteration of the Grey Wolf optimization algorithm, the node congestion index, link quality coefficient, and load balance status data of the retained advantageous paths are jointly encoded in terms of time and space dimensions to generate a real-time spatiotemporal joint feature identifier matrix of path status features. S462. Based on the real-time spatiotemporal joint feature identifier matrix of the path state features, calculate the feature space cosine similarity between each path and the corresponding feature data in the historical state feature memory to obtain the real-time matching degree value between the path features and the historical state data. S463. Construct a feature fusion weight matrix between path state feature data and historical state feature memory based on real-time matching degree value, dynamically fuse path state feature data into historical state feature memory through weighted average method, and update historical state feature memory in real time. S464. Introduce a dynamic time decay mechanism for feature data in the historical state feature memory bank. Assign a time decay coefficient to the real-time dwell time of each state feature data in the memory bank and automatically update the weight of historical state feature data by gradually decreasing the feature weight as time increases. S465. Using the historical state feature memory after time decay, extract the cumulative change trend of three dimensions of feature data: node congestion index, link quality coefficient and load balance in real time, and generate multi-dimensional trend guidance vectors for the next scale path location vector update respectively. S466. Based on multi-dimensional trend guidance vectors, feedforward active pre-regulation is implemented on the update direction of the path position vector at the next scale. Through real-time prediction and active adjustment of feature trends, the path position vector is guided to actively adapt to the dynamic change trend of the network's real-time state.

8. A data routing method for non-directly connected networks according to claim 1, characterized in that, S5 specifically includes: S51. Real-time collection of congestion index, link quality coefficient and network load balance status of each node in the advantageous path, construction of a multi-dimensional dynamic feature space of path status, real-time extraction of the changing trends of path status features in spatial and temporal dimensions, and establishment of a path status trend feature prediction space. S52. Predict the space using path state trend features, and generate path topology change sensitivity matrix and node state change sensitivity matrix through neural augmented ant colony optimization algorithm, which respectively represent the sensitivity of the path to future network topology and node performance changes. S53. Based on the path topology change sensitivity matrix and the node state change sensitivity matrix, calculate the topology prediction weight factor and the node state prediction weight factor for each path, and adjust the topology prediction scale and node state prediction granularity of the neural augmented ant colony optimization algorithm prediction path set in real time accordingly. S54. Real-time path prediction confidence assessment is performed on the predicted path set generated in advance by the neural reinforcement ant colony optimization algorithm. Specifically, the real-time prediction confidence of each path in the predicted path set is obtained by calculating the feature space deviation between the predicted path state vector and the network state trend prediction benchmark. S55. Input the path position vectors in the predicted path set and the real-time prediction confidence value into the Grey Wolf optimization algorithm, and dynamically adjust the preset optimization scale and optimization direction of the path position vectors in the Grey Wolf optimization algorithm in real time according to the path prediction confidence, and perform preset optimization of the path position vectors. S56. A multi-scale real-time feedforward correction mechanism for path position vectors is adopted. The feedforward correction magnitude of each dimension of the path position vector is calculated in advance based on the real-time predicted topology change trend and node state change trend, and the predicted path position vector is actively optimized by feedforward. S57. After multiple rounds of active feedforward optimization and adjustment of the path position vector, the optimal path is finally obtained by ranking the path position vector in real time according to the multidimensional matching degree between the path position vector and the network state trend prediction benchmark.

9. A data routing method for non-directly connected networks according to claim 8, characterized in that, Specifically, S52 includes: S521. Real-time spatiotemporal joint feature identification data of node congestion index, link quality coefficient and network load balance in the prediction space based on path state trend characteristics, construct the coupling sensitivity prediction matrix of path topology state and node state, and dynamically represent the bidirectional interactive coupling degree of path to network topology changes and node state changes. S522. Using the coupling sensitivity prediction matrix of path topology state and node state, calculate in real time the feedforward influence coefficient of path topology state change trend on node state prediction, and the feedback influence coefficient of node state change trend on path topology state prediction. S523. Using the neural reinforcement ant colony optimization algorithm, the feedforward influence coefficient of topology state on node state and the feedback influence coefficient of node state on topology state are used as real-time dynamic weighting factors. The sensitivity feature values ​​of topology structure change are dynamically weighted and fused for each path to generate a path topology change sensitivity matrix that can characterize the degree of proactive response of the path to future network topology changes. S524. Using the neural reinforcement ant colony optimization algorithm, the cross-fusion analysis of the state change trend of the path node itself and the topological state change trend of the neighbor node is used to dynamically calculate the neighbor topological cross sensitivity of the node state for each node, and form a node state change sensitivity matrix in real time that can characterize the degree of active response of the node to the future node state change trend. S525. Real-time construction of an interactive feature verification mechanism between the path topology change sensitivity matrix and the node state change sensitivity matrix. By verifying the spatial feature difference between the sensitivity features of the path topology change sensitivity matrix and the node state change sensitivity matrix, the feature values ​​of the path topology change sensitivity matrix and the node state change sensitivity matrix are dynamically and in real time corrected. S526. Based on the real-time corrected path topology change sensitivity matrix and node state change sensitivity matrix, a joint sensitivity evaluation vector of the path to future network topology and node state changes is dynamically generated through the fusion operation of the feature matrix.

10. A data routing method for non-directly connected networks according to claim 1, characterized in that, S6 specifically includes: S61. The optimized optimal path and at least two alternative paths are converted into multi-dimensional path forwarding description data in real time, carrying path status prediction and early warning values, real-time topology stability indicators and node performance dynamic indicators, and are simultaneously distributed to all forwarding nodes in the network control plane. S62. Establish an adaptive dynamic parallel allocation mechanism for data packet transmission paths in real time. Based on the state prediction warning value and topology stability identifier in the forwarding description data of each path, calculate the forwarding weight of data packets to the optimal path and alternative paths in real time, and dynamically adjust the forwarding ratio of data packets between each path. S63. Real-time acquisition of instantaneous transmission performance characteristics of data packets on the optimal path and alternative paths, including link delay change gradient, spatial distribution probability of data packet loss, and instantaneous utilization saturation rate of path bandwidth. S64. Construct a spatiotemporal multi-scale difference analysis model for the actual performance characteristics of the path in real time. By comparing the real-time changing trends of the multi-dimensional performance characteristics of the optimal path and the alternative paths, a multi-scale spatiotemporal joint evaluation matrix of path performance differences is dynamically generated. S65. The multi-scale spatiotemporal joint evaluation matrix of path performance differences is used in real time to perform dynamic weighted fusion operation on the path search history memory weight matrix in the neural augmented ant colony optimization algorithm, and the path search history memory weight matrix is ​​updated through real-time feature fusion. S66. The adaptive adjustment factor of the path selection probability space is calculated in real time based on the dynamically updated path search history memory weight matrix. This dynamically drives the neural enhanced ant colony optimization algorithm to adjust the path search direction and the probability distribution of path generation in real time, thus completing the algorithm's online continuous adaptive update.

Images