Machine learning based energy and spectrum co-scheduling method for cable tunnel digital optical cable

Abstract

The application discloses a machine learning-based digital optical cable scheduling method for energy and signal co-transmission cable tunnels, relates to the technical field of cable tunnel scheduling, and comprises the following steps: collecting an electromagnetic light signal coupling situation flow of the energy and signal co-transmission tunnel, generating a coupling situation hyper-dimensional embedding representation body, performing topological manifold folding reconstruction, and forming a scheduling self-organizing decision unit set; performing path conflict hidden phase folding adjustment and topological resonance elimination on the scheduling self-organizing decision unit set, generating a fusion resource situation tensor scheduling atlas domain; based on the fusion resource situation tensor scheduling body, performing resource occupation trajectory representation learning and nonlinear phase recoding fusion, and forming a dynamic priority scheduling decision primitive set; according to the dynamic priority scheduling decision primitive set, triggering the scheduling engine to perform mirror reverse isomorphism evolution and tunnel multi-domain resource collaborative self-adaptive adjustment; the application solves the problem of difficult dynamic self-adaptive adjustment of traditional cable tunnel resource conflicts through a machine learning-driven multi-domain collaborative scheduling mechanism.

Description

Technical Field

[0001] This invention relates to the field of cable tunnel scheduling technology, and more specifically, to a machine learning-based digital optical cable scheduling method for energy and information co-transmission cable tunnels. Background Technology

[0002] As the types of services carried by digital optical cables in shared-transmission cable tunnels continue to increase, the electromagnetic disturbances, optical transmission status, and service flow behavior within the tunnel exhibit characteristics such as multi-source heterogeneity, dynamic coupling, and cross-link propagation. This makes it difficult to achieve a unified and correlated expression of the resource status of different links in both the temporal and spatial topological dimensions. Simultaneously, the digital optical cable scheduling process is susceptible to the combined effects of changes in link occupancy, electromagnetic interference propagation, and concurrent switching of multiple services. Its resource competition relationship exhibits dynamic evolution and nonlinear interweaving characteristics, making it difficult for traditional scheduling methods based on fixed rules or static path allocation to accurately identify resource conflict states under complex propagation relationships. Furthermore, during the collaborative operation of multi-domain services, there are situations such as cross-propagation paths, resource occupancy coupling, and mutual influence of scheduling behaviors between different links. This makes the link status during scheduling prone to dynamic drift and cross-domain conflicts, further increasing the difficulty of collaborative adjustment and real-time optimization of digital optical cable resources. Therefore, how to achieve unified modeling of multi-domain resource status in shared-transmission tunnels, dynamic learning of complex propagation relationships, and adaptive optimization of the multi-link collaborative scheduling process have become urgent technical problems to be solved in the field of intelligent digital optical cable scheduling.

[0003] The above-disclosed technical solutions have at least the following technical problems: Traditional optical cable scheduling methods struggle to uniformly model multi-source coupling situations and lack adaptive scheduling capabilities for complex tunnel resource conflicts and dynamic propagation relationships.

[0004] To address the above problems, this invention proposes a solution. Summary of the Invention

[0005] To overcome the aforementioned deficiencies of the prior art, embodiments of the present invention provide a machine learning-based digital optical cable scheduling method for power and communication co-transmission cable tunnels. This method solves the problem of dynamic adaptive adjustment of resource conflicts in traditional cable tunnels through a machine learning-driven multi-domain collaborative scheduling mechanism.

[0006] To achieve the above objectives, the present invention provides the following technical solution: A machine learning-based digital optical cable scheduling method for energy-information co-transmission cable tunnels includes: acquiring the electromagnetic-optical-information coupling situation flow of the energy-information co-transmission tunnel, generating a hyperdimensional embedded representation of the coupling situation, and performing topological manifold folding reconstruction to form a set of self-organizing scheduling decision units; generating a fused resource situation tensor scheduling map domain by performing path conflict hidden phase folding adjustment and topological resonance resolution on the self-organizing scheduling decision unit set; performing resource occupancy trajectory representation learning and nonlinear phase recoding fusion based on the fused resource situation tensor scheduling volume to form a set of dynamic priority scheduling decision primitives; and triggering the scheduling engine to perform mirror reverse isomorphic evolution and multi-domain resource collaborative self-adaptive adjustment based on the dynamic priority scheduling decision primitives.

[0007] In a preferred technical solution, the acquisition of the electromagnetic-optical-information coupled situational flow of the energy-information co-transmission tunnel, the generation of a coupled situational hyper-embedded representation, and the execution of topological manifold folding and reconstruction to form a set of scheduling self-organizing decision units are specifically as follows: The electromagnetic-optical-information coupled situational flow of the energy-information co-transmission tunnel is acquired, unified temporal alignment is performed, and an association mapping is established according to the link topology to form the electromagnetic-optical-information coupled situational flow; a multimodal temporal learning model is introduced to preprocess the electromagnetic-optical-information coupled situational flow with cross-source feature encoding and link semantic alignment, and self-supervised training is performed to learn the joint temporal mapping relationship to form an electromagnetic-optical-information coupled situational association sequence; based on the electromagnetic-optical-information coupled situational association sequence, high-dimensional situational embedding reprojection and nonlinear interactive unwrapping are performed to generate a coupled situational hyper-embedded representation. A multi-dimensional embedded representation is constructed. Topological manifold phase migration constraint folding and disturbance propagation path evolution control are applied to the coupled situation superdimensional embedded representation to construct a multi-scale manifold potential skeleton structure. Based on the multi-scale manifold potential skeleton structure, link behavior trajectory temporal decoupling and recalibration and situation evolution path semantic alignment are performed to form a link behavior evolution sequence. Dynamic adjacency relationship reconstruction and implicit topological self-expansion divergence are performed on the link behavior evolution sequence to generate a link co-evolution structure. Multi-level situation compression mapping and disturbance sensitivity hierarchical folding are performed on the link co-evolution structure to construct a multi-level situation manifold structure domain. Scheduling behavior semantic normalization mapping and link resource competition situation aggregation are performed on the multi-level situation manifold structure domain to form a set of scheduling self-organizing decision units.

[0008] In a preferred technical solution, the step of generating a coupled situation superdimensional embedded representation by performing high-dimensional situation embedding reprojection and nonlinear interactive unwrapping based on the electromagnetic-optical-information coupled situation correlation sequence is as follows: Based on the electromagnetic-optical-information coupled situation correlation sequence, time-series sliding decomposition is performed to construct a cross-domain path coupling mapping primitive set; link state sieving and path propagation direction realignment are performed on the cross-domain path coupling mapping primitive set to form a path cooperative correlation cluster; based on the path cooperative correlation cluster, multi-scale topological rejoining is performed to generate an implicit coupled topological skeleton network; a multi-layer path correlation sequence is formed by performing topological neighborhood hierarchical expansion on the implicit coupled topological skeleton network; relation tensor reconstruction is performed on the multi-layer path correlation sequence to obtain a reconstructed path tensor expression; based on the reconstructed path tensor expression, latent space manifold expansion mapping is performed to form a continuously evolving manifold structure domain; cross-domain semantic distributed alignment is performed on the continuously evolving manifold structure domain, and high-dimensional embedding projection and topological sparsification rearrangement are performed on the manifold structure to output a coupled situation superdimensional embedded representation.

[0009] In a preferred technical solution, the process of performing topological manifold phase migration constraint folding and perturbation propagation path evolution control on the coupled state hyperdimensional embedded representation to construct a multi-scale manifold potential skeleton structure is as follows: Based on the coupled state hyperdimensional embedded representation, perturbation energy level trajectory deconstruction is performed to form a cross-domain perturbation path cluster set; dominant propagation chain screening and path resonance aggregation are performed on the cross-domain perturbation path cluster set to generate a perturbation dominant topological ridge set; phase gradient hierarchical embedding is performed on the perturbation dominant topological ridge set to obtain hierarchical phase coupled topological clusters; topological constraint manifold folding is performed on the hierarchical phase coupled topological clusters to form a cross-phase skeleton mapping framework; perturbation propagation trajectory control is performed on the cross-phase skeleton mapping framework to establish hierarchical propagation constraint links; multi-scale topological extrapolation expansion is performed on the hierarchical propagation constraint links to form a stable propagation ridge structure domain; cross-scale phase fusion reconstruction is performed on the stable propagation ridge structure domain to output a multi-scale manifold potential skeleton structure.

[0010] In a preferred technical solution, the step of generating a fused resource situation tensor scheduling graph domain by performing path conflict latent phase folding adjustment and topological resonance resolution on the scheduling self-organizing decision unit set is as follows: Based on the scheduling self-organizing decision unit set, a graph sequence joint learning model is introduced to encode temporal features and learn propagation dependency representations for the occupancy evolution state, forming an implicit resource conflict evolution flow; the implicit resource conflict evolution flow is subjected to competition path latent phase folding migration, forming a phase-coupled conflict chain cluster; based on the phase-coupled conflict chain cluster, topological resonance-induced reconstruction is performed, forming a set of resonance propagation topological ridge chains; multi-domain conflict ablation adjustment is performed on the set of resonance propagation topological ridge chains, forming a stable propagation coupling domain; based on the stable propagation coupling domain, hierarchical topological collapse mapping is performed, forming a set of hierarchical propagation constraint contexts; cross-layer resource situation tensor coupling is performed on the set of hierarchical propagation constraint contexts, forming a resource situation tensor correlation domain; and topological manifold resonance regularization is performed on the resource situation tensor correlation domain to output the fused resource situation tensor scheduling graph domain.

[0011] In a preferred technical solution, the process of performing implicit phase folding migration on the implicit resource conflict evolution flow to form a phase-coupled conflict chain cluster is as follows: The implicit resource conflict evolution flow is structured and temporally unfolded according to the propagation event chain, and multiple propagation trajectory units are established for each competing path to form a cross-path temporal phase unit set; multi-source temporal anchoring mapping is performed on the cross-path temporal phase unit set to construct a cross-temporal anchoring constraint network; based on the cross-temporal anchoring constraint network, implicit phase folding migration is performed to form a continuous evolutionary conflict chain segment; topological re-representation reconstruction is performed on the continuous evolutionary conflict chain segment to form a multi-directional conflict coupling node group; path mirror folding is performed on the multi-directional conflict coupling node group to obtain a high-density coupled structural unit; closed propagation substructures are generated by performing implicit phase energy level hierarchical mapping on the high-density coupled structural units; cross-energy level hierarchical fusion and path connection reorganization are performed on the closed propagation substructure, and overall topological regularization and redundancy convergence are performed on the connected structure to output a phase-coupled conflict chain cluster.

[0012] In a preferred embodiment, the process of forming a stable propagation coupling domain by performing multi-domain conflict ablation modulation on the resonant propagation topological ridge chain set is as follows: The resonant propagation topological ridge chain set is phase-expanded along the propagation topological axis, and the node propagation behavior is mapped to form a continuous coherent arrangement structure; a propagation potential field is constructed for the path intersection region in the continuous coherent arrangement structure, and the intersection region of the propagation potential field is hierarchically localized to form a conflict potential field distribution band; based on the conflict potential field distribution band, phase decoupling modulation is performed, and path-dependent constraint relationships are established; boundary coherent migration is performed on the path-dependent constraint relationships to obtain a continuous drift morphological boundary structure; a path tensor expression state renormalization sequence is generated by tensor mapping recoding of the continuous drift morphological boundary structure; a phase constraint projection is applied to the path tensor expression state renormalization sequence to form a continuous coherent path tensor chain state; based on the continuous coherent path tensor chain state, topological potential energy redistribution modulation is performed to output a stable propagation coupling domain.

[0013] In a preferred technical solution, the process of fusing resource occupancy trajectory representation learning and nonlinear phase recoding based on the fused resource situation tensor scheduling body to form a dynamic priority scheduling decision primitive set is as follows: Based on the fused resource situation tensor scheduling body, spatiotemporal expansion and hidden state encoding of occupancy trajectories based on a sequence learning model are performed to obtain continuous trajectory unit clusters; cross-chain phase stretching reconstruction is performed on the continuous trajectory unit clusters, and neighborhood propagation constraints are introduced for trajectory folding and embedding to form a continuous evolutionary chain core; resource state tensor up-dimensional mapping is performed on the continuous evolutionary chain core, and along the propagation axis... Tensor stack reorganization is performed to obtain a unified tensor representation domain; nonlinear phase recoding fusion is performed based on the unified tensor representation domain, and phase spectrum splitting and re-aggregation operations are applied to the encoding results to generate a clustered distribution structure; cross-domain dynamic orchestration and rearrangement are performed on the clustered distribution structure to form a unified evolutionary sequence body; dynamic priority tensor labeling is performed on the unified evolutionary sequence body, and hierarchical tensor decomposition and reorganization are applied to the encoding results to obtain a hierarchical tensor structure; based on the hierarchical tensor structure, intra-layer consistent aggregation and cross-layer conflict inference scheduling processing based on a decision learning model are performed to output a dynamic priority scheduling decision primitive set.

[0014] In a preferred technical solution, the step of performing resource state tensor up-dimensional mapping on the through-evolutionary chain core and reorganizing the tensor stack along the propagation axis to obtain a unified tensor expression domain is as follows: Phase trajectory deconstruction is performed on the through-evolutionary chain core to construct a phase reference projection path; based on the phase reference projection path, propagation potential energy coupling is performed on adjacent trajectory units to form potential energy identification units, and gradient hierarchical mapping is performed on the potential energy identification units to establish a multi-level propagation potential energy sequence; non-Euclidean phase stretching mapping is performed on the multi-level propagation potential energy sequence to form a nonlinear continuous unfolded phase structure; based on the nonlinear continuous unfolded phase structure, tensor domain embedding mapping is performed to form a continuous adjacency relationship chain; cross-domain mirror coupling reconstruction is performed on the phase units in the hierarchical tensor space to obtain a composite propagation unit structure; adaptive orchestration is performed on the composite propagation unit structure along the propagation axis to generate an orchestration tensor structure; global topological convergence modulation is performed on the orchestration tensor structure, and structural folding resolution is performed on the cross-layer repeated propagation relationship to form a unified tensor expression domain.

[0015] In a preferred technical solution, the step of triggering the scheduling engine to perform mirror reverse isomorphic evolution and tunnel multi-domain resource collaborative self-adaptive adjustment based on the dynamic priority scheduling decision primitive set is as follows: The dynamic priority scheduling decision primitive set is axially arranged, and each decision primitive is mirror-mapped to form a mirror evolution reference sequence; the mirror evolution reference sequence is reverse-propagation backtracking rearrangement is performed to form a reverse mirror evolution chain; the resource propagation segments in the reverse mirror evolution chain are heterogeneous link penetration reorganization is performed to form a multi-domain through propagation structure; the resource scheduling paths in the multi-domain through propagation structure are collaboratively occupied and migrated to form a collaborative scheduling propagation sequence; the collaborative scheduling propagation sequence is dynamically phase-accompanied evolution is performed to form a continuous accompanying propagation structure; the continuous accompanying propagation structure is multi-domain resource self-adaptive reconstruction is performed to form a multi-domain collaborative propagation chain; the multi-domain collaborative propagation chain is mirror reverse convergence adjustment is performed, and the folded propagation nodes are uniformly axially regularized.

[0016] The technical effects and advantages of the machine learning-based digital optical cable scheduling method for power and communication cable tunnels in this invention are as follows: 1. This invention collects the electromagnetic-optical-communication coupled situational flow of a shared energy-communication tunnel and performs multi-source alignment processing on electromagnetic disturbance characteristics, optical-communication transmission load changes, and service flow characteristics under a unified time reference. Simultaneously, it generates a hyperdimensional embedded representation of the coupled situational flow by combining link topology relationships. Based on this, it performs topological manifold folding reconstruction, reconstructing the originally dispersed and heterogeneous tunnel link states into a structurally continuous set of scheduling self-organizing decision units. This enables multi-source situational information to form a consistent expression in spatial topology and time series, effectively reducing the state expression deviation and structural fragmentation problems between different links, improving the fusion consistency and structured organization capability of multi-dimensional resource situational data, thereby enhancing the input stability and correlation expression integrity of the subsequent scheduling modeling process. Furthermore, a topological folding mechanism is introduced during the decision unit generation process to reorganize implicit link connection relationships, transforming cross-link dependencies from discrete states into continuously expressible structural units, thereby improving the resolvability and structural expressive capability of multi-source information in complex tunnel environments.

[0017] 2. This invention compresses and reconstructs multi-link resource competition relationships at the structural level by performing path conflict latent phase folding adjustment and topological resonance resolution on the scheduling self-organizing decision unit set. This transforms the latent conflict from a dispersed propagation state into a unified and fused resource situation tensor scheduling graph domain. Simultaneously, based on this graph domain, representation learning is performed on the resource occupancy trajectory, and a dynamic priority scheduling decision primitive set is generated by combining a nonlinear phase recoding fusion mechanism. This transforms resource scheduling behavior from a static rule mapping to a dynamic evolutionary ranking structure, effectively improving the accuracy of resource competition identification and the consistency of priority determination in complex conflict scenarios. Furthermore, by triggering the scheduling engine to perform mirror reverse isomorphic evolution and tunnel multi-domain resource collaborative self-adaptive adjustment processing, the scheduling path undergoes structural mirror correction and multi-domain resource synchronous reconfiguration during the reverse evolution process. This enables the scheduling strategy to have cross-link consistent evolution capability, thereby significantly improving the adaptive adjustment capability and overall collaborative optimization level of tunnel multi-domain resource scheduling. Attached Figure Description

[0018] Figure 1 This is a flowchart illustrating the digital optical cable scheduling method for energy and information co-transmission cable tunnels based on machine learning, as described in this invention.

[0019] Specific technical solutions The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0020] Example 1, Figure 1This invention presents a machine learning-based digital optical cable scheduling method for energy-information co-transmission cable tunnels, comprising the following steps: S1, collect the electromagnetic-optical-communication coupled situational flow of the energy-communication tunnel, generate a coupled situational hyperdimensional embedded representation, and perform topological manifold folding and reconstruction to form a set of scheduling self-organizing decision units; In this embodiment, the electromagnetic-optical-communication coupled situational flow of the energy-communication tunnel is collected, a coupled situational hyperdimensional embedded representation is generated, and topological manifold folding and reconstruction are performed to form a set of scheduling self-organizing decision units, as detailed below: Collect electromagnetic-optical-communication coupling situation flow of the co-transmission tunnel (including changes in optical transmission load under occupancy status, changes in electromagnetic disturbance response, electromagnetic change sequence, link occupancy status, changes in optical transmission load, tunnel link spatial number, electromagnetic disturbance characteristic data, optical attenuation characteristic data, and service flow characteristic data), perform unified time alignment on data from different sources, and establish association mapping according to link topology to form electromagnetic-optical-communication coupling situation flow; A multimodal temporal learning model is introduced to preprocess the electromagnetic-optical-communication coupling situation flow by performing cross-source feature encoding and link semantic alignment. Self-supervised training is then conducted to learn the joint temporal mapping relationship between electromagnetic disturbances, optical attenuation, and service flow, forming an electromagnetic-optical-communication coupling situation association sequence with topology node identifiers and time index alignment. Specifically, the electromagnetic-optical-communication coupling situation flow is aligned to a unified time reference and expanded with tunnel topology node numbers. Electromagnetic disturbance data, optical attenuation data, and service flow data are segmented into a multimodal temporal sample set according to link units. The samples are then reconstructed using a sliding time window to form the training sequence input. Subsequently, the multimodal temporal learning model initialization structure is constructed, i.e., historical tunnel operation log data is divided into temporal training samples according to the link topology structure, and pre-training is performed based on occlusion reconstruction and adjacency prediction tasks. During the pre-training stage, node-level electrical... The magnetic change sequence performs a random time-slice occlusion reconstruction task, the optical signal attenuation sequence performs a future segment prediction task, and the business flow sequence performs a cross-node migration prediction task, enabling the model to synchronously learn time dependencies and cross-modal mapping relationships. Furthermore, a link topology adjacency graph is constructed, and adjacency relationships are used as constraint inputs in joint optimization during training. The model is iteratively updated until the loss converges, and then the encoded model is solidified to complete the model construction. In the inference phase, the current electromagnetic-optical-signal coupling situation flow is divided according to the same segmentation rules, and time-step feature encoding is performed on each link node to map electromagnetic disturbances, optical signal attenuation, and business flow features into a unified temporal embedding representation. At the same time, attention weight calculation between nodes is performed in combination with link adjacency relationships to complete semantic alignment. Then, the encoding results of all link nodes are recursively spliced ​​and continuously expanded according to the topological order to form an electromagnetic-optical-signal coupling situation association sequence. Based on the electromagnetic-optical-information coupled situational correlation sequence, high-dimensional situational embedding reprojection and nonlinear interactive unwrapping processing are performed to generate a coupled situational hyperdimensional embedded representation. We perform topological manifold phase migration constraint folding and perturbation propagation path evolution control on the coupled state hyperdimensional embedded representation to construct a multi-scale manifold potential skeleton structure; Based on the multi-scale fluid morphology potential skeleton structure, temporal decoupling and recalibration of link behavior trajectories and semantic alignment of situational evolution paths are performed to form a link behavior evolution sequence. Specifically, the multi-scale fluid morphology potential skeleton structure is decomposed into two types of structural units according to topological hierarchical identification: main chain branches and lateral coupling branches. The link behavior trajectories within each structural unit are expanded into time series record clusters according to node sequence numbers. Then, event-anchored time axis recalibration is performed on the time series record clusters, and the time indexes of each node are remapped. Afterward, the recalibrated trajectory sequence is... The system performs trajectory slicing, dividing continuous time segments into connectable trajectory unit blocks. Then, based on the topological adjacency relationship in the manifold skeleton, the trajectory unit blocks are connected across nodes, that is, the unit blocks in the same path direction are chained together according to the spatial topological order. Then, the spliced ​​trajectory chain is aligned by point-by-point mapping of multi-source semantic features. Finally, the electromagnetic disturbance intensity feature, optical signal attenuation feature and business flow feature are bound together under the same time reference, and the aligned trajectory chain is subjected to temporal order regularization and local adjustment, outputting the link behavior evolution sequence. A dynamic adjacency relationship reconstruction driven by cross-domain cooperative coupling and implicit topology self-unfolding divergence are performed on the link behavior evolution sequence to generate a link cooperative evolution structure. Specifically, the link behavior evolution sequence is divided into continuous behavior segments according to a sliding time window, and multi-source feature extraction is performed on the electromagnetic disturbance characteristics, optical signal attenuation characteristics, and service flow characteristics within each segment to form node behavior feature clusters. Then, based on the cross-link distance relationship between node behavior feature clusters, a candidate adjacency pair set is generated. Furthermore, temporal screening is performed on the candidate adjacency pair set, and link relationships that co-occur stably in adjacent time windows are appended. The newly created dynamic adjacency link list is then entered. Subsequently, the dynamic adjacency link list is incrementally evolved and scanned. The connection relationships added or disappeared in the previous and next time windows are updated segment by segment differentially, and the connection edge relationship records are rewritten. Based on the updated adjacency relationships, the link structure is extended into a multi-hop path extension, that is, the single-hop adjacency is extended into a propagable path link group. The path link group is then broken and ineffective connections are removed. Finally, the removed topology path group is aligned node by node with cross-domain features. The electromagnetic and optical signaling features of different links are mapped and synchronized under a unified time axis, and the link cooperative evolution structure is output. By performing multi-layer situational compression mapping and perturbation sensitivity hierarchical folding on the link co-evolution structure, high-frequency perturbation links and low-frequency stable links are expressed hierarchically, constructing a multi-level situational manifold structure domain. Specifically, the link co-evolution structure is divided into multi-scale behavioral segment units, and node-by-node feature extraction is performed on electromagnetic perturbation, optical signal attenuation, and service flow within each segment unit to form a multi-channel state description sequence. Then, cross-scale mapping expansion is performed on the multi-channel state description sequence, that is, the link states at different time granularities are projected onto a unified hierarchical representation space, and then the link... The perturbation response of the road nodes is sorted to generate a hierarchical sensitive identifier set. Then, high change response nodes and low change response nodes are written into the hierarchical sensitive identifier set respectively. Subsequently, structural compression is performed on the sensitive identifier set of each level, and repeated connection relationships within the same level are merged and redundant paths are pruned. Based on the compressed hierarchical topology, cross-layer propagation path alignment is performed, and the node connection relationship between adjacent levels is mapped node by node and the path offset is fine-tuned. Finally, the multi-layer topology structure that has been aligned is stacked in hierarchical order to output a multi-level situational manifold structure domain. A semantic normalization mapping of scheduling behavior and aggregation of link resource competition situation are performed on the multi-level situation manifold structure domain to form a set of self-organizing scheduling decision units. Specifically, the multi-level situation manifold structure domain is divided into several link situation layer units in hierarchical order, and scheduling behavior records (including link occupancy status, changes in optical transmission load, and changes in electromagnetic disturbance response) are extracted from the link nodes in each layer unit to form a scheduling behavior description sequence. Then, a unified semantic encoding mapping is performed on the scheduling behavior description sequence to convert the scheduling behavior in different links into a standard expression in the same behavior identifier space, and then the behavior identifiers are associated. Grouping involves merging nodes with shared link resources into the same competition set and then merging these sets across layers to form resource competition clusters. These clusters are then rearranged according to a continuous time window, with competition relationships in different time slices continuously spliced ​​and updated. Based on the spliced ​​competition relationships, node-level merging is performed, merging and compressing duplicate competition paths while retaining dominant conflict paths. Finally, the processed competition clusters are arranged sequentially and structurally regularized according to the topology hierarchy, outputting a set of self-organizing decision units for scheduling that includes node competition relationship identifiers, link occupancy status, and propagation priority ranking.

[0021] In this embodiment, based on the electromagnetic-optical-information coupled situational correlation sequence, high-dimensional situational embedding reprojection and nonlinear interactive unwrapping processing are performed to generate a coupled situational hyperdimensional embedded representation, as detailed below: Based on the electromagnetic-optical-communication coupling situational correlation sequence, a time-slip decomposition is performed to break down the electromagnetic disturbance propagation chain, optical energy attenuation chain, and service flow evolution chain into multi-source link evolution units, constructing a cross-domain path coupling mapping primitive set. Specifically, the electromagnetic-optical-communication coupling situational correlation sequence is progressively sliced, and the data in each time slice is segmented into link domains according to the tunnel link spatial number. Then, the electromagnetic disturbance changes, optical attenuation changes, and service flow migrations within each link domain are independently expanded to form corresponding electromagnetic disturbance propagation chains, optical energy attenuation chains, and service flow evolution chains. The business flow evolution chain is then processed. Subsequently, the same-source time index binding is performed on the three types of propagation chains. The multi-source propagation trajectories of the same link node in the same time slice are synchronously attached. Then, the direction of the synchronously attached propagation trajectory is scanned. Link segments with consistent propagation direction and continuous time are sequentially merged. After that, cross-window propagation drift tracking is performed on the merged link segments. That is, path association identifiers are established for link nodes with continuous migration relationship in adjacent time windows, and the associated nodes are arranged hierarchically according to the topological order of the propagation path, finally forming a cross-domain path coupling mapping primitive set. The process involves performing link state screening and path propagation direction alignment on the cross-domain path coupling mapping primitive set, and establishing path association identifiers for link nodes with cross-propagation relationships to form path collaborative association clusters. Specifically, the cross-domain path coupling mapping primitive set is expanded line by line, and the electromagnetic disturbance change sequence, optical transmission fluctuation sequence, and service flow migration sequence within each primitive are deconstructed at the field level. Then, the deconstructed link state sequences are continuously rearranged according to time index, that is, the state segments of the same link node in different time windows are spliced ​​end-to-end to form a continuous evolution chain of link states. After that, the continuous evolution chain is scanned in direction, and the segments with the same state change direction in adjacent time slices are compressed. Subsequently, the compressed link segments are compared in path direction, and the link segments with the same propagation direction and topological adjacency are sequentially concatenated and recombined. Then, the link nodes in the concatenated link sequence that show path convergence or node reuse in different propagation paths are cross-marked, and the cross-marked nodes are bidirectionally bound to their respective propagation links. Finally, the bound nodes are grouped and merged according to the spatial distribution relationship of the cross-marks, and the path collaborative association clusters are finally output. Based on the path cooperative association clusters, multi-scale topology reassembly is performed to align the electromagnetic path structure, optical communication path structure, and service path structure with path anchoring in the shared node space, thereby generating an implicitly coupled topology skeleton network. Specifically, the path cooperative association clusters are split and parsed, and the electromagnetic path fragments, optical communication path fragments, and service path fragments within each cluster are expanded at the link level to form a set of categorized path units. Then, the categorized path unit sets are uniformly rearranged, and the occurrence positions of the same node in different path types are aligned and indexed. Based on the aligned node sequence, a shared node scan is performed, and nodes with intersection relationships are cross-anchored and marked. The cross-anchored nodes are bidirectionally bound to their corresponding path fragments. Subsequently, the path fragments are reassembled across types, that is, the electromagnetic path, optical communication path, and service path are attached point by point at the shared nodes. Then, the attached path structure is hierarchically organized according to the topology level, and the continuously reachable path fragments within the same level are sequentially spliced ​​together to finally generate an implicitly coupled topology skeleton network. By performing a hierarchical expansion of the implicitly coupled topological skeleton network and a progressively expanding process for the expanded path connections, a multi-layered path association sequence is formed. Specifically, the implicitly coupled topological skeleton network is decomposed point by point into node connection units, and the direct connection edges and cross-node reachable connection edges of each node connection unit are extracted to form a basic connection mapping set. Then, the basic connection mapping set is hierarchically encoded, and one-hop connections, two-hop connections, and multi-hop connections are written into a hierarchical connection index table. Based on the hierarchical connection index table, a neighborhood expansion scan is performed, starting from each node... The process involves progressively extending the path along the connecting edges, gradually incorporating reachable nodes into the extended neighborhood set in stages, and simultaneously recording the hierarchical origin of the connecting paths between nodes. Then, the path relationships in the extended neighborhood set are sorted and organized according to the topological adjacency direction. Spatially adjacent and hierarchically continuous path segments are chained together. Subsequently, cross-level connection backfilling is performed on the chained path, that is, high-level paths are reconnected at the breaks in low-level paths, and the reconnected path chains are hierarchically ordered. Paths of different levels are uniformly rearranged in the order of expansion from low to high, and finally, a multi-level path association sequence is obtained. A nonlinear interactive unwrapping-driven relation tensor reconstruction is performed on multi-level path association sequences to obtain a reconstructed path tensor representation. Specifically, the multi-level path association sequences are expanded in two dimensions according to hierarchical identifiers and time indices. The node connection relationships in each path are then deconstructed segment by segment to extract electromagnetic disturbance correlation segments, optical transmission dependency segments, and service flow dependency segments, forming a multi-source relation segment set. Then, an interactive coupling scan is performed on the multi-source relation segment set to match and mark node connection relationships with cross-dependencies in different path levels, and write them into the relation coupling index unit. Finally, the relation coupling... The index units are grouped and organized according to the conflict type, and repeated connections, cross connections and skip connections are classified and recorded separately. Based on the classified relation units, the order is rearranged. Then, cross-level recursive expansion is performed on the rearranged relation units to extend and supplement the connection relationship between adjacent levels step by step. Then, the supplemented relation set is non-linearly interactively split, that is, high-strength coupling connection is decomposed into independent association units, and low-strength coupling connection is continuously maintained and mapped. Finally, the split association units are uniformly regularized according to the node dimension and the level dimension to generate the reconstructed path tensor representation. Based on the reconstructed path tensor representation, a latent space manifold expansion mapping process is performed. Discrete link nodes are projected onto a high-dimensional latent space through continuous neighborhood expansion, forming a continuously evolving manifold structure domain. Specifically, the reconstructed path tensor representation is decomposed according to the node dimension and the path dimension. The connection vector, propagation dependency record, and cross-link association identifier of each node at different path levels are extracted. The nodes are then mapped to initial distribution coordinates in the latent space. Local neighborhood initialization is then performed on these initial distribution coordinates, i.e., an initial adjacency index is established for topologically adjacent and path-continuous nodes. This adjacency index is then recursively expanded, starting from any node and based on the connection dependency strength and... The path continuation relationship is extrapolated step by step, and reachable nodes are included in the neighborhood range in circles according to the expansion level. The hierarchical source of the cross-layer connection path between nodes is recorded simultaneously. Then, the node connection relationship in the neighborhood range is sorted by direction. The node sequence in the same path propagation direction is organized into a chain according to the spatial adjacency order. Then, cross-layer breakpoint repair is performed on the chained node sequence. The path break positions between different expansion levels are connected and the relationship is completed segment by segment. Based on the repaired node sequence, the latent space relocation mapping is performed to convert the discrete node coordinates into a continuous distribution trajectory expression. Finally, the overall morphological reconstruction is performed on the continuous distribution trajectory to finally obtain the continuous evolution manifold structure domain. Cross-domain semantic distributed alignment is performed on the continuously evolving manifold structure domain. Then, high-dimensional embedding projection and topological sparsification rearrangement are performed on the semantically aligned manifold structure to output a high-dimensional embedding representation of the coupling situation. Specifically, the continuously evolving manifold structure domain is segmented and analyzed according to both time slices and topological hierarchy. The spatial distribution of nodes, electromagnetic disturbance response markers, optical link dependency markers, and service flow association markers within each segment are extracted and uniformly encoded into a set of cross-domain semantic feature units. Finally, semantic anchoring matching is performed on the set of cross-domain semantic feature units. Semantic units with propagation associations or node co-occurrence relationships are bound point by point, and a cross-domain semantic correspondence index structure is constructed. Then, the index structure is hierarchically aggregated, duplicate semantic connections are merged, and broken semantic connections are completed by chaining according to the neighborhood propagation path. Furthermore, a high-dimensional space remapping is performed on the manifold structure that has completed semantic binding, that is, the node trajectory is reprojected to a unified high-dimensional expression space according to the semantic correspondence. Then, the high-dimensional expression result is subjected to topological structure compression and rearrangement. Finally, the coupled situation hyperdimensional embedded representation is output.

[0022] In this embodiment, topological manifold phase migration constraint folding and perturbation propagation path evolution control are performed on the coupled state hyperdimensional embedded representation to construct a multi-scale manifold potential skeleton structure, as follows: Based on the coupled situation hyperdimensional embedded representation, disturbance energy level trajectories are deconstructed to form a cross-domain disturbance path cluster set. Specifically, the coupled situation hyperdimensional embedded representation is jointly expanded according to time index and topological node identifier to extract the electromagnetic disturbance response trajectory, optical link fluctuation trajectory, and service flow dependency trajectory of the node in a continuous time window. The three types of trajectories are then synchronized with time scales and their sequences are normalized to form a multi-domain disturbance trajectory primitive set. Next, disturbance event segmentation is performed on the multi-domain disturbance trajectory primitive set, dividing the sequence into several local disturbance segments. A node-level trajectory unit description structure is established for each segment. Subsequently, cross-domain disturbance path clusters are performed on the trajectory unit description structure. The node propagation association scan interactively marks units with path continuity or disturbance propagation associations and writes them into a newly created propagation association index chain. Further, the propagation association index chain is expanded by directional extrapolation. Starting from each node, it is extended in a stepwise manner along the disturbance intensity gradient and topological adjacency direction. Trajectory units that meet the propagation continuity condition are included in the expanded trajectory set layer by layer. At the same time, the cross-layer propagation transition relationship and path branch structure are recorded. Finally, the extended trajectory set is converged and aggregated by the backbone. Consistent trajectory chains with stable propagation directions are merged and compressed, and discontinuous weakly coupled branches are pruned and merged to form a cross-domain disturbance path cluster set. The process involves: screening dominant propagation chains and aggregating path resonances for the cross-domain disturbance path cluster set; suppressing and archiving intermittent jump paths to generate a set of dominant topological ridges for disturbances. Specifically, the cross-domain disturbance path cluster set is split and parsed according to the path cluster identifier, extracting the corresponding electromagnetic disturbance propagation subsequences, optical link fluctuation subsequences, and service flow dependency subsequences from each path cluster; merging and mapping the subsequences of the same node in different path clusters to form a set of cross-domain disturbance multi-source sequences; then, remapping sequences with inconsistent sampling in different path clusters within the cross-domain disturbance multi-source sequence set to the same time index; and performing segment-level recalibration on the remapped sequences to form a standardized disturbance path record set. Subsequently, the standardized disturbance path records are... The recording set performs dominant path screening by comparing path segments one by one, chaining highly continuous path segments together and writing them into the dominant propagation candidate set. Further, path intersection identification is performed on the dominant propagation candidate set, marking node intersections and propagation overlap relationships within the same time window and constructing an intersection association pair set. Then, resonance aggregation is performed on the intersection association pair set to combine paths with synchronous propagation characteristics into resonance propagation chain clusters. Discontinuity identification is then performed on the resonance propagation chain clusters to classify path segments with propagation interruption or jump characteristics into unstable branches and write them into the suppression archive set. Finally, the dominant propagation chain clusters and the suppression archive set are separated, reconstructed, and structurally regularized to finally output a set of perturbation dominant topological ridges. Based on the perturbation-dominant topological ridge set, a phase gradient hierarchical embedding is performed to map path nodes to multi-level phase energy bands. Local neighborhood coupling structures are then constructed within these phase energy bands to obtain hierarchical phase-coupled topological clusters. Specifically, the perturbation-dominant topological ridge set is structurally analyzed according to path identifiers and propagation directions to extract node sequences and corresponding perturbation response trajectories. Time axis recalibration and propagation order regularization are then performed on the node sequences to form standardized ridge sequence units. These standardized ridge sequence units are then hierarchically labeled, and nodes are mapped to multi-level phase energy bands according to the hierarchical labels. In the quantitative interval, a hierarchical phase energy level band distribution framework is constructed. Then, neighborhood coupling is constructed within the hierarchical phase energy level band distribution framework. Taking the node as the core, associated nodes are selected along the topological adjacency relationship and the same layer propagation relationship, and a set of coupling connection mapping relationships between nodes is generated. Further, structural aggregation is performed on the set of coupling connection mapping relationships. The node relationships in the highly connected and tightly connected regions are clustered and the sparse regions with low connections are selectively pruned and retained. Then, the merged neighborhood coupling structure is hierarchically aligned and reorganized according to the phase energy level band, and finally, a hierarchical phase coupling topological cluster is obtained. Based on the hierarchical phase-coupled topological cluster, topological constraint manifold folding is performed, and anchored alignment and topological boundary reconstruction are carried out on cross-phase connections to form a cross-phase skeleton mapping framework. Specifically, the hierarchical phase-coupled topological cluster is deconstructed layer by layer according to phase energy level bands, and the cross-layer propagation connection records are uniformly coded and labeled with directions to form a basic set of cross-phase connections. Then, constraint filtering is performed on the basic set of cross-phase connections to remove non-continuous connections, and the remaining connections are sorted at the segment level and the topological order is regularized to form a sequence of cross-layer connection units. Further, folding mapping is performed on the sequence of cross-layer connection units to fold and merge nodes with dependencies in different phase energy level bands according to their connection relationships, and anchored alignment is performed on the folding results to bind the cross-layer repeated nodes with equivalent propagation nodes to coordinates. On the basis of the aligned structure, boundary reconstruction is performed, that is, the broken segments in the cross-phase connections are repaired by path splicing and structural backfilling, and the overall topological rearrangement of the repaired connection relationships is performed to finally output the cross-phase skeleton mapping framework. Based on the cross-phase skeleton mapping framework, perturbation propagation trajectory control is performed. High-perturbation paths are mapped with skeleton priority and shape preservation, while low-perturbation paths are elastically weakened and structurally sparsified, thereby establishing a hierarchical propagation constraint link. Specifically, the cross-phase skeleton mapping framework is decomposed and analyzed according to the propagation direction, and the perturbation propagation trajectory sequence in each path unit is extracted. The trajectory sequence is then normalized on the time base and the node order is reorganized to form a standardized propagation path primitive set. Then, the perturbation response is graded on the standardized propagation path primitive set, that is, the path units are labeled with high response and low response, and the labeling results are written into the path response index structure. Subsequently, the path response index structure is structurally screened. The complete node chain connection relationship and cross-layer propagation association are retained for high response paths, while the structure is compressed for low response paths, and the discontinuous node connections are deleted and path segments are merged. Based on the processed path set, cross-phase hierarchical splicing is performed to connect the path segments in different phase layers in the order of consistent propagation direction, and the connection results are recursively rearranged hierarchically, finally outputting the hierarchical propagation constraint link. By performing multi-scale topological extrapolation on the hierarchical propagation constraint links, the local skeleton is continuously extended along the dominant propagation direction to form a stable propagation ridge structure domain. Specifically, the hierarchical propagation constraint links are segmented and expanded according to the propagation direction. Next, the node sequences within each link segment are read, and the node sequences are synchronized on the time axis and adjusted in topological order to form a set of normalized constraint path units. Then, the dominant propagation direction is identified on the set of normalized constraint path units, and the dominant propagation path segments are selected based on the consistency of the path continuity direction. Finally, endpoint locking is performed on the selected results. After annotation, multi-scale extrapolation construction is performed on the dominant propagation path segments. Starting with the locked node as the reference, the extension is carried out progressively along the topological adjacency relationship and propagation direction. Nodes that meet the propagation constraints are included in the extended path chain level by level, and the hierarchical transition relationship and path extension trajectory are recorded simultaneously. Then, the extended path chain is reconstructed by cross-segment splicing, and the dispersed path units are connected in a chain according to the propagation direction. The structural breakpoints are connected and repaired. Finally, redundant branches are cleaned up on the reconstructed path structure, and low-strength connection segments and non-dominant branches are trimmed and regularized to finally obtain the stable propagation ridge structure domain. A cross-scale phase fusion reconstruction is performed on the stable propagation ridge structural domain. Different levels of skeleton units are topologically homeomorphically aligned and structurally interlocked to output a multi-scale flow morphological potential skeleton structure. Specifically, the stable propagation ridge structural domain is partitioned according to hierarchical identifiers and spatial topological coordinates. The node sequences, connections, and propagation trajectories in each level of skeleton unit are standardized and regularized. Spatial reference unification processing is performed on the node sequences to form a standardized skeleton unit set. Then, cross-layer structural matching is performed on the standardized skeleton unit set, i.e., skeleton units with the same structure in different levels are paired and bound together, and a cross-layer mapping index relationship is established. Next, structural interlocking is performed on the cross-layer mapping index relationship, interactively connecting and merging the node connections of corresponding skeleton units, and merging and regularizing overlapping connection areas. Subsequently, cross-scale repair is performed on the interlocked structure, i.e., connection compensation and path backfilling are performed on broken paths and disconnected nodes, and topological rearrangement is performed on the repaired overall structure. Finally, the skeleton units at each level are uniformly aligned and structurally reorganized according to the mapping relationship, ultimately outputting a multi-scale flow morphological potential skeleton structure.

[0023] S2 generates a fusion resource situation tensor scheduling graph domain by adjusting the hidden phase folding and resolving the topological resonance of the execution path conflict of the scheduling self-organizing decision-making unit set. In this embodiment, by adjusting the hidden phase folding and resolving the topological resonance of the execution path conflict of the scheduling self-organizing decision unit set, a fused resource situation tensor scheduling graph domain is generated, as follows: Based on a set of self-organizing decision units for scheduling, a graph-sequence joint learning model is introduced to encode temporal features and learn propagation dependency representations for the occupancy evolution states of different link nodes, forming an implicit resource conflict evolution flow. Specifically, the self-organizing decision unit set for scheduling is structurally parsed and the links are expanded. Node sequence input samples are generated by aligning the propagation path numbers with time slices, and these samples are divided into a training set and a current inference set (where the training set contains node occupancy change sequences, link connection relationships, and conflict annotation data). Subsequently, the graph-sequence joint learning model is pre-trained. The pre-training process employs a self-supervised approach, jointly optimizing the node occupancy sequence time slice occlusion reconstruction task and the link adjacency prediction task, enabling the model to learn node temporal dependency features and propagation topological dependency features respectively. After training, the encoding is solidified to form a reasonable model, which is then used for inference. In this stage, the current set of decision units is input into the model. First, a sliding time window slicing process is performed on each node to divide the continuous occupancy state into fixed-step sequence segments and encode them as temporal input vectors. At the same time, the link topology is expanded by adjacency to extract the set of the previous and next hop associated nodes of each node and use them as structural input. Further, the model performs node-by-node recursive encoding on the temporal input vector to generate the occupancy evolution latent state vector, and performs neighborhood aggregation calculation on the structural input to generate the propagation dependency representation vector. The two types of vectors are fused and mapped node-by-node to form a unified node representation sequence. Then, the unified node representation sequence is continuously spliced ​​according to the propagation path order, and the conflict activation difference between adjacent nodes is calculated segment by segment. The high difference interval is marked as the conflict propagation segment. Finally, all segments are chained and continuously aligned to output the implicit resource conflict evolution flow. The implicit resource conflict evolution flow is subjected to competition path implicit phase folding migration, and competition nodes are subjected to cross-temporal propagation anchoring and path mirror folding to form a phase-coupled conflict chain cluster. Based on the phase-coupled conflict chain clusters, topological resonance-induced reconstruction is performed, and hierarchical resonance connection relationships are established for cross-path propagation nodes to form a set of resonance propagation topological ridge chains. Specifically, the phase-coupled conflict chain clusters are rearranged in a structured manner according to the cluster number and propagation direction label, and the phase change trajectory and propagation path of the nodes within each cluster are segmented and merged. Then, cross-chain path interaction identification is performed on the rearranged cluster sequence, and node pairs with propagation path intersection or phase change synchronization characteristics are associated and labeled, and a set of candidate resonance connection pairs is constructed. Subsequently, propagation screening is performed on the candidate resonance connection pair set, and node pairs that show a consistent evolution trend in the continuous propagation interval are connected and strengthened, and a cross-path hierarchical connection structure is established between the clusters. Further, topological connection is performed on the cross-path hierarchical connection structure, and resonance nodes in adjacent clusters are chained together according to the path direction, and interrupted connection segments are extended and supplemented. Finally, the overall topological arrangement of the completed cross-path connection structure is performed, the repeated connection relationships are compressed and integrated, and the set of resonance propagation topological ridge chains is output in hierarchical order. By performing multi-domain conflict ablation modulation on the set of resonant propagation topological ridge chains, and performing propagation boundary drift reconstruction and discontinuous path merging on the conflict intersection region, a stable propagation coupling domain is formed. Based on the stable propagation coupling domain, a hierarchical topology collapse mapping is performed to conformally integrate high-frequency competing links into dominant propagation, and elastic sparse folding and path weakening rearrangement are performed on low-association propagation branches to form a hierarchical propagation constraint network set. Specifically, the stable propagation coupling domain is partitioned according to topological hierarchy and propagation direction, and the link connection relationships and propagation trajectories within each partition are sequentially reorganized. Then, propagation strength labeling is performed on the reorganized link segments, high-frequency competing links and low-association branches are separated and classified, and the main path is locked for high-frequency competing links. The process involves connecting continuous, conflict-dense, and stable propagation path segments, merging redundant branches, compressing low-association propagation branches, pruning discontinuous node connections, rearranging and simplifying weakly coupled paths, and reconnecting the remaining branches into a low-intensity propagation sequence according to the topological direction. Furthermore, the high-frequency main path and low-frequency branches are hierarchically nested, with branches arranged as subordinate structures. Finally, the overall structure is hierarchically regularized and compressed to form a hierarchical propagation constraint network set. A cross-layer resource situation tensor coupling is performed on the hierarchical propagation constraint network set to form a resource situation tensor association domain. Specifically, the hierarchical propagation constraint network set is partitioned and decomposed, and the link node relationships, propagation trajectory directions, and resource occupancy changes within each partition are serialized and arranged. Then, nodes in the sequence of each layer that are in the same topological position or have propagation dependencies are marked for alignment and a cross-layer mapping node pair set is constructed. Subsequently, resource attribute binding is performed on the cross-layer mapping node pair set, and the changes in link occupancy status, propagation intensity, and interaction dependencies are written into the association record chain structure node by node. Further, tensor interleaving and recombination are performed on the association record chain, that is, the cross-layer node association relationships are arranged in an alternating manner according to topological order and time order, and duplicate path connections are compressed and merged. Then, the discontinuous association points are path supplemented and continuous. Based on the recombined cross-layer association structure, hierarchical aggregation is performed to group and merge similar propagation paths according to hierarchical direction and structure regularization, finally obtaining the resource situation tensor association domain. By performing topological manifold resonance regularization on the resource situation tensor correlation domain, the competitive propagation structures at different levels are continuously integrated and their topological boundaries are reconstructed. Then, the regularized propagation relationships are subjected to tensor-level hierarchical rearrangement, outputting a fused resource situation tensor scheduling graph domain. Specifically, the resource situation tensor correlation domain is divided into blocks according to hierarchical numbers and topological partitions. The competitive propagation relationships, node connection relationships, and resource occupancy changes within each block are serialized and arranged. Then, cross-layer synchronization identification is performed on the chain structures at each layer, structural units with consistent propagation rhythms are aligned and marked, and cross-layer resonance correlation pairs are constructed. The set is then subjected to continuous merging operations on the cross-layer resonance correlation set. Nodes with overlapping paths or propagation between adjacent layers are linked together, and discontinuous regions are extended and connected. Based on the merging cross-layer propagation structure, topological boundary reshaping is performed, the segmented nodes at the layer boundaries are re-divided and the connection relationships are rearranged, and the intersection regions are converged and the structure is merged. Finally, tensor-level rearrangement is performed on the regularized propagation relationships, and the propagation paths of different layers are staggered and grouped according to a unified topological order. The final output is the fused resource situation tensor scheduling graph domain.

[0024] In this embodiment, the implicit resource conflict evolution flow is subjected to a contention path implicit phase folding migration, and cross-temporal propagation anchoring and path mirroring folding are performed on the contention nodes to form a phase-coupled conflict chain cluster, as detailed below: The implicit resource conflict evolution flow is structured and temporally unfolded according to the propagation event chain, and multiple propagation trajectory units are established for each competition path, forming a cross-path temporal phase unit set. Specifically, the implicit resource conflict evolution flow is chained according to the event triggering order, and the conflict records are sequenced and reorganized according to the node propagation direction and resource occupation change trajectory. Then, the event chain is segmented according to the nodes of propagation path changes, and the intervals where path turning, resource contention, or propagation discontinuity occur are divided into independent propagation trajectory units. Subsequently, each independent propagation trajectory unit is bound to a corresponding node identifier and time index relationship, forming a multi-path temporal phase unit set. The set of propagation trajectory units is further subjected to cross-path interaction scanning. The node interaction behaviors that occur in the same time window in different competing paths are marked and cross-associated with trajectory units that have path overlap or resource conflict characteristics. Based on the bound trajectory units, sequential merging is performed. The same-direction propagation units are chained together, the reverse propagation units are reconstructed in direction and rearranged in links, and the splicing results are subjected to structural convergence. Redundant propagation segments are compressed and integrated, and discontinuous propagation segments are path-supplemented, finally obtaining the cross-path temporal phase unit set. Multi-source temporal anchoring mapping is performed on the cross-path temporal phase unit set to align and bind conflicting nodes with propagation synchronization, delayed coupling, and alternating triggering characteristics, thereby constructing a cross-path temporal anchoring constraint network. Specifically, the cross-path temporal phase unit set is arranged in a two-dimensional manner according to node identifier and time index, so that different path units form an aligned sequence under a unified time reference framework. Then, the interaction behavior of nodes within each time window is scanned and compared window by window. Conflicting nodes exhibiting synchronous triggering, delayed response, and alternating triggering characteristics are pattern-labeled and classified into various temporal features according to the triggering mode. After grouping, cross-path matching is performed on each temporal feature group, that is, nodes in different paths that present the same trigger rhythm or propagation dependency are paired and bound, and their interaction correspondence links are recorded. Then, the bound node pairs are chained together according to the temporal continuity relationship and propagation direction, the node pairs in the same trigger sequence are organized into continuous association segments in sequence, and the overlapping parts across sequences are merged. Then, the structure regularization is performed on the chained node relationship set, the repeated association relationships are compressed and fused, and the discontinuous connection relationships are path supplemented and the order is reorganized, finally forming a cross-temporal anchoring constraint network. Based on a cross-temporal anchoring constraint network, latent phase folding migration is performed to form continuously evolving conflict chains. Specifically, the node relationships in the cross-temporal anchoring constraint network are sequentially woven according to the time progression direction. Discrete conflict events are then linked according to the propagation connections between nodes. Next, latent phase migration is performed on the linked conflict events, connecting event nodes located in adjacent time intervals with continuous competition. Then, event nodes in different time windows but maintaining propagation dependencies are migrated and attached across windows. Finally,... The event chain that has been connected is folded and merged. Repeated propagation segments are clustered, concurrent competition paths are converged and spliced, and path segments with mirror propagation characteristics are symmetrically spliced. Furthermore, the propagation sequence is reconstructed on the folded conflict chain. The node order is rearranged according to the continuity of propagation direction and the time progression relationship. The reverse cross chain segments are transposed and the link order is adjusted. On the basis of the reconstructed propagation chain, continuous reorganization is performed. Discontinuous connection areas are reconnected, and redundant jump links are trimmed and merged, finally forming a continuously evolving conflict chain segment. The continuous evolution conflict chain segments are reconstructed by topological re-representation, and bidirectional conflict coupling mapping relationship clusters are generated in the path intersection area to form a multidirectional conflict coupling node group. Specifically, the continuous evolution conflict chain segments are reorganized in a chain according to the propagation direction and node connection order, and the competitive propagation relationship in the chain segment is topologically classified. Then, the path convergence area in the continuous propagation structure is interactively inspected, and the node combination with bidirectional competitive transmission relationship is marked accordingly. A bidirectional coupling connection chain is established based on the round-trip propagation relationship between the nodes. Subsequently, the bidirectional coupling connection chain is associated and grouped, and the nodes with the same intersection path or repeated competitive behavior are clustered and merged to form a local conflict coupling relationship group. Further, the local conflict coupling relationship group is connected by cross-chain progressive connection, that is, the associated nodes in different propagation chains are continuously connected in series along the propagation continuation direction, and the cross-over overlapping area is connected, compressed, and the link is converged. Then, the topological arrangement of the connected node relationship is performed. Finally, redundant cross connections are merged and pruned, and the discontinuous area is connected and the order is changed, outputting a multidirectional conflict coupling node group. A path mirroring and folding process is performed on a multi-directional conflicting coupled node group, and tensor compression is applied to the mirrored redundant connections to obtain high-density coupled structural units. Specifically, the multi-directional conflicting coupled node group is symmetrically distributed according to propagation direction and interaction relationship, and mirror matching and calibration are performed on node pairs with reverse competing links to form a bidirectional mapping sequence at the structural level for the corresponding propagation paths. Then, the mirror-matched bidirectional mapping sequence is folded and converged, and node segments with opposite propagation directions but consistent conflict characteristics are superimposed. Repeated interaction intervals are chain-compressed and path merged. Subsequently, the folded... The node connection relationships are subjected to tensor quantization rearrangement, which involves interleaving the node interaction relationships in different propagation chains according to the time progression and topological adjacency order, converging high-frequency repetitive connections, and weakening the boundaries and reducing the paths of low-frequency discrete connections. Furthermore, the rearranged structure is coupled and integrated, clustering the nodes with high correlation, shrinking the outer edges and simplifying the structure of weakly correlated paths. Finally, the integrated node system is regularized as a whole, compressing and merging the cross-redundant connections, and continuing the paths and connecting the structures in the discontinuous connection regions, ultimately obtaining a high-density coupled structural unit. By performing a hidden phase energy level hierarchical mapping on high-density coupled structural units, nodes are embedded into hierarchical energy level bands according to their phase states, generating closed propagation substructures. Specifically, the high-density coupled structural units are sequentially arranged according to the node interaction strength and propagation response rhythm, and the phase states of the nodes are segmented and marked, thus forming a phase distribution chain with continuously changing characteristics in the structure. Then, the phase distribution chain is hierarchically arranged, that is, nodes with similar phase characteristics and consistent propagation rhythms are assigned to the same energy level interval, and a hierarchical arrangement structure is established according to the energy level size. Subsequently, within the hierarchical energy level bands... The node relationships are embedded and mapped, and the nodes are placed into the corresponding energy level positions according to their phase affiliation. Cross-energy level connections are bound and docked to form cross-layer connection links. Furthermore, the node connection relationships within each energy level band are closed-loop spliced, that is, the node links that have the conditions for connection at the beginning and end or form a circular dependency are closed in a loop. For broken links, the path is continued and the node order is rearranged. Finally, the overall structure after the embedding and closure processing is regulated and organized, redundant cross connections are compressed and merged, and discontinuous links are sequentially connected to form a closed propagation substructure. The closed propagation substructure undergoes cross-level hierarchical fusion and path reorganization, and the reorganized structure is subjected to overall topology regularization and redundancy convergence, outputting a cluster of phase-coupled conflict chains. Specifically, the hierarchical coupling substructure is partitioned and grouped according to energy level identifiers and propagation directions, and the connection relationships and propagation trajectories of nodes within each energy level are serialized and organized, thus presenting a segmented distribution pattern in different hierarchical structures. Then, cross-level coupling docking is performed on the node relationships between adjacent energy levels, binding nodes with propagation continuation or dependency associations one by one, and constructing cross-level hierarchical fusion links based on the binding relationships. Subsequently, cross-level hierarchical fusion is performed. The link execution path is reorganized and connected, and the propagation paths distributed in different energy levels are connected in series according to the same direction. The overlapping paths are rearranged and replaced. Furthermore, the overall structure after the connection is reorganized and the topology is reorganized. The node connection relationship is arranged in a unified manner according to the propagation direction, the repeated connection segments are merged and compressed, and the broken chain segments are continued and the structure is repaired. Finally, the reorganized structure is subjected to redundancy convergence, the low-frequency interactive connection is eliminated, the high-frequency propagation path is preserved in order and shape and the hierarchical position is rearranged, and finally a phase-coupled conflict chain cluster organized according to the cross-energy-level hierarchical fusion relationship is formed.

[0025] In this embodiment, a stable propagation coupling domain is formed by performing multi-domain conflict ablation adjustment on the set of resonant propagation topological ridge chains and performing propagation boundary drift reconstruction and discontinuous path merging on the conflict intersection region, as detailed below: The set of resonant propagation topological ridge chains is phase-expanded along the propagation topological axis, and the propagation behavior of the nodes is mapped to form a continuous coherent arrangement structure. For the path intersection region in the continuous coherent arrangement structure, a propagation potential field is constructed, and the intersection region of the propagation potential field is hierarchically classified to form a conflict potential field distribution band. Specifically, the path intersection region is segmented according to the topological adjacency relationship, the interaction connection and propagation direction between nodes in each segment are uniformly identified, and the path relationship between the intersection nodes is converted into a directional propagation vector group. Then, the directional propagation vector group is overlaid on the topological space axis, and the corresponding propagation potential field distribution state is generated according to the directional consistency and aggregation intensity after overlay. Subsequently, the high intersection region in the propagation potential field distribution state is divided into gradient bands, that is, divided into layers according to the potential energy change from strong to weak, and the node connection relationship within each layer is reorganized in the same layer. Furthermore, cross-layer association marking is performed on the boundary nodes between adjacent layers, and nodes with path overlap or propagation continuation characteristics are aligned and bound. Finally, the layered structure after binding is merged into bands, and finally a conflict potential field distribution band is formed according to the hierarchical distribution of intersection intensity. Based on the conflict potential field distribution zone, phase decoupling adjustment is performed to separate the high potential energy intersection structure by phase shift and establish path dependency constraint relationship. Specifically, the conflict potential field distribution zone is segmented and numbered according to the potential energy gradient. The intersection points of nodes in each segment are topologically located and marked, and the connection relationship between the intersection nodes is phase-state encoded. Then, phase difference mapping is performed on the high-density intersection structure in the same segment, that is, the relationship between nodes with the same spatial position but different propagation directions is written into a phase shift vector group, and the phase shift vector group is separated and projected on the topological axis. Then, the nodes with propagation continuation relationship or path nesting relationship among the nodes that have completed phase separation are bound and labeled one by one, and the binding relationship is written into the path dependency constraint chain structure. Further, the constraint chain structure is oriented and cross-layer association is performed on the dependent nodes across segments to express the discontinuous dependent path as a continuous connection, and finally a set of path dependency constraint relationship based on phase shift separation is formed. A coherent boundary migration is performed on path-dependent constraints, and the phase pointing relationship of path connections is rewritten during the migration process to obtain a continuously drifting boundary structure. Specifically, the path-dependent constraints are serialized and arranged according to the topological hierarchy and propagation direction. Boundary positioning marks are applied to boundary interaction nodes, and the connection relationship between nodes is transcribed into a phase pointing vector group. Then, the phase pointing vector group is continuously slid-projected in the topological neighborhood space, and the phase pointing of the overlapping connection relationship that occurs during the migration process is rewritten and uniformly adjusted into a unidirectional continuous pointing structure. Subsequently, the neighborhood cross-segment connection is performed on the node connection relationship that has completed phase rewriting, that is, the nodes with propagation continuation conditions in adjacent boundary segments are connected one by one, and the direction of the connection path is shaped. Finally, the cross-segment connection structure is sequentially rearranged, and the discrete boundary connection is reorganized into a continuous progressive link according to the drift direction, ultimately forming a boundary structure with topological continuous drift characteristics. By performing tensor quantization mapping and recoding on the continuous drift morphological boundary structure, the broken segments and jump segments are uniformly projected as path tensor units, generating a path tensor quantization expression state reorganization sequence. Specifically, the continuous drift morphological boundary structure is segmented and identified according to topological coherence. Intervals with connection interruptions and trajectory abrupt changes are marked as broken, and the node connection relationships and propagation directions within each interval are structurally recorded. Then, tensor quantization encoding transformation is applied to the structured recorded content, and the node spatial position, connection direction, and propagation state are synchronously mapped to the multidimensional tensor representation domain to generate tensor quantization results. Subsequently, the tensor quantization results are cross-segment projection aligned in the topological neighborhood space. The tensor units between adjacent segments are arranged coaxially according to spatial adjacency, and the state of tensor units with jump differences is rewritten. Furthermore, the aligned tensor units are ordered according to the propagation direction, and the dispersed tensor structure is rearranged into a progressively arranged link according to the topological path, finally forming a path tensor quantization expression state reorganization sequence. Based on the path tensor expression state renormalization sequence, phase-constrained projection is applied to map the path tensor units to a unified coherent reference field. Cross-domain coherent state matching is then performed on the path tensor units to form a continuous coherent path tensor chain state. Specifically, the path tensor expression state renormalization sequence is indexed and arranged according to the propagation topology. The node connection relationship, propagation direction, and state characteristics of each path tensor unit are phase-calibrated, and a phase-constrained projection rule set is generated. Then, the tensor units in the phase-constrained projection rule set are sequentially sent to the unified coherent reference field for coordinate normalization projection. Cross-domain coherence comparison is then performed on the projected tensor units, that is, units with consistent propagation trajectories and phase deviation convergence characteristics are paired and marked. Subsequently, the tensor units that meet the conditions in the pairing and marking results are connected segment by segment according to the topological direction, and phase compensation is performed to fill the discontinuous connections that occur during the connection process. Finally, the completed tensor chain structure is serialized and recombined according to the propagation direction to form a continuous coherent path tensor chain state. Based on the continuous coherent path tensor chain state, topological potential energy redistribution modulation is performed. Potential energy compression normalization and local convergence control are applied to redundant overlapping paths, outputting a stable propagation coupling domain. Specifically, the continuous coherent path tensor chain state is serialized and numbered according to topological order, and the connection strength, overlap frequency, and propagation direction of each path unit are marked with potential energy to form a topological potential energy mapping sequence. Then, within the topological potential energy mapping space, the densely overlapping regions in the topological potential energy mapping sequence are redistributed, that is, the connection relationships of multiple repeated paths are merged and mapped, and redundant paths from the same source are compressed to a unified potential energy coordinate expression form. Subsequently, local convergence is performed on the compressed path structure, and node-level merging and path order rearrangement are performed on segments with repeated intersection nodes and high-frequency connection jumps. The connection relationships are then continuously organized according to the main propagation direction. Furthermore, the overall structure of the processed path is topologically reorganized, the remaining path units are reordered according to the potential energy gradient, and the boundary connection relationships are continuously connected, finally obtaining the stable propagation coupling domain.

[0026] S3, based on the fusion of resource situation tensor scheduling body, performs resource occupation trajectory representation learning and nonlinear phase recoding fusion to form a dynamic priority scheduling decision primitive set; In this embodiment, based on the fused resource situation tensor scheduling body, resource occupancy trajectory representation learning and nonlinear phase recoding are fused to form a dynamic priority scheduling decision primitive set, as follows: Based on the fused resource situation tensor scheduling body, spatiotemporal unfolding and hidden state encoding of occupancy trajectories are performed using a sequence learning model to obtain continuous trajectory unit clusters. Specifically, the historical resource occupancy records in the fused resource situation tensor scheduling body are first archived in two dimensions according to node identifier and timestamp, constructing a trajectory training sample set with labels (including node occupancy state change type and propagation path number). Then, the trajectory training sample set is divided into training sequences according to continuous time segments, and multi-scale sequence alignment samples are generated using a sliding window method. After that, the sequence learning model is pre-trained based on this training sequence set. During the pre-training process, time-step encoding is performed on the input sequence to map the node occupancy state to an embedding vector, and the sequence prediction error is processed through a backpropagation mechanism. Iterative updates are performed to bring the model to a stable hidden state representation. After pre-training, the current fused resource situation tensor scheduling volume is input into the model inference channel. Stepwise state encoding is performed on the real-time resource occupancy sequence, and a hidden state vector sequence corresponding to the time step is generated inside the model. Then, the hidden state segments in the hidden state vector sequence with jump amplitude exceeding the preset continuity threshold are reordered, and the hidden state segments with high continuity are aligned on the time axis. On this basis, the aligned hidden state sequence is axially expanded and mapped along the node propagation path direction. The hidden state results of different paths are uniformly projected onto the same temporal reference coordinate system, and the projected sequence is spliced ​​and completed between segments. Finally, the continuous trajectory unit cluster after model encoding and spatiotemporal expansion is output. Cross-chain phase stretching reconstruction is performed on continuous trajectory unit clusters, and neighborhood propagation constraints are introduced for trajectory folding and embedding to form a through-type evolutionary chain core. Specifically, the continuous trajectory unit clusters are divided into domains according to the propagation link direction, the occupancy state change position corresponding to each trajectory unit is phase-written, and the phase change segments in different links are mapped to a unified propagation axis. Then, cross-chain stretching migration is performed on trajectory segments with propagation continuity relationship, that is, the originally discrete trajectory segments are continued and spliced ​​along the propagation direction, and phase continuity supplementation is performed on the misaligned segments between adjacent links to obtain a continuous stretched trajectory structure. Further, the trajectory segments with interactive relationship in the continuous stretched trajectory structure in the topological neighborhood are folded and compressed according to the spatial adjacency order, and the trajectory connections with intersecting propagation directions are reversed and reconstructed. On the basis of the compressed trajectory structure, nested through-organization is performed, the cross-chain trajectory nodes are progressively connected according to the propagation continuity order, and the paths of the local overlapping segments are merged to finally form a through-type evolutionary chain core. By performing resource state tensor up-dimensional mapping on the kernel of the continuous evolution chain and reorganizing the tensor stack along the propagation axis, a unified tensor expression domain is obtained. Nonlinear phase recoding fusion is performed based on a unified tensor representation domain, and phase spectrum splitting and re-aggregation operations are applied to the encoding results to generate a clustered distribution structure. Specifically, the state tensors in the unified tensor representation domain are sequentially arranged according to the propagation path and node level, and phase characterization is performed on the resource occupancy fluctuation positions in each state tensor. The switching behavior, propagation lag relationship, and state transition segment in the link are transcribed into corresponding phase coding slices. Subsequently, nonlinear interleaving arrangement is performed on the phase coding slices, that is, the phase slices in different propagation paths are interleaved and superimposed according to the offset correlation relationship, and the local phase density is adjusted. The phase spectrum structure is obtained by compressing and folding the segments. Then, the phase spectrum structure is split into multiple independent phase sub-layers. The phase sub-layers are rearranged according to the propagation direction. Then, the split phase sub-layers are aggregated and bundled. The sub-layer segments with phase convergence characteristics are chained together and the boundary positions of adjacent sub-layers are transitioned and spliced. Finally, the merged phase structure is clustered and organized. The encoded segments with similar phase correlation are arranged in a clustered manner according to the topological adjacency order to output a clustered distribution structure. A cross-domain dynamic orchestration and rearrangement is performed on the clustered distribution structure. Resource behavior units from different paths are arranged in an interleaved nested manner according to propagation continuity and dependency chain strength to form a unified evolutionary sequence. Specifically, resource behavior units in the clustered distribution structure are classified into domains according to their source path identifiers, and the propagation trajectory, state transition nodes, and dependency triggering relationships of each unit are marked. Subsequently, cross-domain alignment matching is performed on behavior units from different source paths. Units that are adjacent on the propagation timeline and have the same direction are bound one by one, and a dependency chain association index structure is constructed based on the binding results. Gradient sorting is performed on the dependency chain association index, that is, high dependency strength units are preferentially embedded into the main propagation channel, and medium and low dependency units are assigned to auxiliary nested channels. During the embedding process, behavioral units from different path sources are interleaved and arranged, so that multi-source units form a progressive alternating layout in the sequence. Subsequently, local topology reorganization is performed on the progressive alternating layout, links are rewritten and the order is reorganized for segments with propagation breaks or sequence misalignments, and dependency supplementation and path connection are performed for continuous interruption regions. Finally, the reorganized overall sequence is uniformly converged and normalized along the propagation direction, and a unified evolution sequence body is output. Dynamic priority tensor labeling is performed on the unified evolutionary sequence body, and hierarchical tensor decomposition and recombination are applied to the encoding results to obtain a hierarchical tensor structure. Specifically, resource behavior units in the unified evolutionary sequence body are numbered and positioned according to the propagation path, and the dependency triggering relationship, propagation coherence, and path nesting depth of each unit are marked and characterized to form a priority identifier sequence. Subsequently, tensor projection is performed on the priority identifier sequence, that is, behavior units from different source paths are converted into multidimensional relational expressions, and the sequences are segmented and grouped according to the relational relationships. Further, the... The segmented tensor representation undergoes hierarchical decomposition, which involves assigning units with high dependency coupling and continuous propagation characteristics to the high-level structure, medium-coupling units to the mid-level structure, and weakly dependent units to the base layer structure. Units within each layer are then rearranged and chained according to the propagation direction. Subsequently, cross-layer mapping is performed on the tensor units between different levels, binding dependent units pairwise and rewriting the inter-layer connection paths. The bound structure is then reorganized across layers. Finally, the overall structure, after layering and reorganization, is regularized and output according to the hierarchical order, forming a layered tensor structure. Based on a hierarchical tensor structure, this system performs intra-layer consistent aggregation and cross-layer conflict inference scheduling based on a decision learning model, outputting a dynamic priority scheduling decision primitive set. Specifically, the process involves: first, constructing a sequence learning model pre-training process, dividing the hierarchical tensor structure samples into trajectory sequence segments according to time windows, and annotating each segment at the node level (including changes in resource occupancy state, propagation direction, and hierarchical transition relationships). Then, inputting the annotated sequences into a temporal coding network for supervised training. During training, a bidirectional sequence modeling structure is used to encode the dependencies between preceding and following sequences, and updates are performed by predicting the consistency error between the occupancy state and the hidden state at the next time step, allowing the model to converge to the trajectory hidden state representation space and complete the pre-training initialization. Subsequently, in the online execution phase, the hierarchical tensor structure is expanded into a temporal sequence hierarchically, and a structured input vector is generated for each node. The input vector (containing node connection direction encoding, local adjacency topology encoding, and resource occupancy change sequence encoding) is then processed. The trained model is input, and it outputs a hidden state vector at each time step. Cross-time step alignment calculations are performed on the hidden state sequence. Sliding window aggregation and similarity comparison are applied to continuous hidden states, and highly similar hidden state segments are continuously merged. Interpolation-based completion and reconstruction are performed on discontinuous hidden states at boundaries. Subsequently, spatiotemporal unfolding mapping is performed based on the hidden state sequence, reordering the hidden states according to the propagation axis direction into an unfolded sequence structure. Intra-layer scanning is performed on the unfolded sequence, merging and clustering hidden states with similar distribution characteristics to form intra-layer consistent unit clusters. Further, conflict inference is performed on the intra-layer consistent unit clusters, matching and identifying node pairs in different layers whose hidden states are close in distance but have opposite topological connection directions. Conflicting nodes undergo sequence rearrangement and connection relationship rewriting. Path backtracking scanning is performed on the rearranged cross-layer hidden state structure, reconstructing broken paths segment by segment according to the continuity of hidden states. Finally, the processed intra-layer consistent unit clusters and cross-layer reconstructed structures are uniformly sorted to obtain a dynamic priority scheduling decision primitive set.

[0027] In this embodiment, a unified tensor representation domain is obtained by performing resource state tensor up-dimensional mapping on the kernel of the continuous evolution chain and reorganizing the tensor stack along the propagation axis, as follows: Phase trajectory deconstruction is performed on the through-evolutionary chain core, rewriting the state transition paths of each node into continuous phase flow units, and constructing a phase reference projection path based on the propagation axis direction. Specifically, the node propagation behavior in the through-evolutionary chain core is segmented and labeled, and the resource occupation changes and path switching actions within each time segment are rewritten as events to form a node-level transition event chain. Then, the key change points in the transition event chain are phase symbolized, that is, resource occupation increases / decreases, path migrations, and state pauses are mapped to corresponding phase marker units, and the phase marker units are sequentially concatenated according to the propagation direction to form an initial phase flow structure. Furthermore, the initial phase flow structure is subjected to... Cross-node splicing involves connecting phase segments from different nodes that are continuously connected in the propagation time sequence at their endpoints, and compressing and merging the repeated phases at the connection points to form cross-node continuous phase flow units. Subsequently, structural perturbation resolution is performed on the cross-node continuous phase flow units to smoothly rearrange phase segments with excessively dense local jumps or directional offsets. Then, axial projection mapping is performed on the cross-node continuous phase flow units, which involves spatially aligning all phase units according to the propagation axis and continuously filling the gaps generated during the projection process. Finally, path concatenation shaping is performed on the axially aligned phase projection sequence to chain the scattered phase points according to the propagation order to obtain the phase reference projection path. Based on the phase reference projection path, propagation potential energy coupling characterization is performed on adjacent trajectory units to form potential energy identification units. Gradient hierarchical mapping is then performed on these potential energy identification units to establish a multi-level propagation potential energy sequence. Specifically, continuous phase points in the phase reference projection path are segmented and paired according to sliding neighborhood intervals. The phase transition amplitude, path turning angle, and resource occupancy changes of each pair of adjacent trajectory units are synchronously calibrated. These three types of changes are coupled and encoded under the same characterization to form propagation potential energy identification units. These propagation potential energy identification units are then arranged in a continuous chain, and adjacent difference scanning is performed on the arranged units. Subsequently, the potential energy change relationship between preceding and following units is progressively re-labeled, and sudden changes are identified. The changing segments undergo boundary segmentation and rewriting. Then, structured diversion is performed on all potential energy identification units. Units exhibiting high-frequency fluctuations and strong turning characteristics on the propagation trajectory are assigned to the high potential energy sequence, units exhibiting moderate continuous perturbation characteristics are assigned to the medium potential energy sequence, and units with gentle changes and weak dependencies are assigned to the low potential energy sequence. Units within each sequence are reordered and chained together according to the consistent propagation direction. Furthermore, cross-layer position alignment is performed between different potential energy sequences. Units with corresponding relationships on the propagation axis are bound point by point, and the binding relationships are recorded and indexed in an interleaved manner. Finally, each potential energy sequence is arranged vertically according to the gradient magnitude to form a multi-level propagation potential energy sequence. A non-Euclidean phase stretching mapping is performed on the multi-level propagation potential energy sequence to form a nonlinear continuously unfolded phase structure. Specifically, each layer unit in the multi-level propagation potential energy sequence is numbered point-by-point according to the propagation axis. Then, for each numbered position, the phase value and potential energy gradient value of the corresponding layer unit are collected. The difference between the multi-level units at that numbered position is then calculated, and the phase difference and potential energy difference are superimposed into a single transformation control quantity. Next, processing is advanced segment by segment according to the propagation axis from the starting position. Within each processing segment, the control quantity is mapped into three types of adjustment commands. In high potential energy gradient segments, equidistant subdivision nodes are inserted between adjacent phase points, and the connection order is rearranged. In the mid-potential gradient segment, the original spacing is maintained and the phase order is locally swapped and rearranged. In the low-potential gradient segment, adjacent phase points are compressed and merged and redundant connection points are deleted. At the same time, the first and last nodes of each processing segment are marked with boundaries, and the connection is rewritten at the junction of adjacent processing segments according to the phase direction of the preceding and following segments. The disconnected connection points are re-established across segments. Then, the structure after segmentation is uniformly scanned back along the propagation axis, and all cross-layer nodes are reprojected into the same axis coordinate sequence. The overlapping positions are deduplicated and the order is preserved. Finally, the reconstructed phase point sequence is connected point by point in axial order to form a continuously unfolded nonlinear phase structure. Based on a nonlinear continuously expanded phase structure, tensor domain embedding mapping is performed, and topological re-anchoring is applied to the embedding positions to form a continuous adjacency chain. Specifically, the phase points in the nonlinear continuously expanded phase structure are numbered and calibrated point-by-point along the propagation axis, and the phase state value, local potential gradient, and adjacent connection orientation of each phase point are recorded simultaneously. Then, based on the numbering sequence, each phase point is mapped to its corresponding tensor embedding coordinate position, and the embedding coordinates are segmented and classified according to the continuity of phase change. Points with smooth phase changes and consistent directions are grouped into the same embedding segment unit. Finally, the phase increment and decrement relationships are systematically... First, the process is converted to unidirectional propagation sorting, and the connection relationships of nodes within the segment are rewritten. Then, cross-segment connection is performed between different embedded segment units, and the end node of the previous segment is connected to the starting node of the next segment point by point according to phase continuity. The broken connections generated during the connection process are repaired and reconstructed. Furthermore, topological re-anchoring is performed on the overall embedded structure, and all connection relationships are scanned node by node. Cross connections and redundant connections are stripped one by one, and unique primary adjacency relationships are re-established according to the propagation direction. Finally, the re-anchored structure is traversed and output sequentially according to the propagation axis, and all nodes are connected in series according to the primary adjacency relationship to form a continuous adjacency relationship chain. Cross-domain mirror coupling reconstruction is performed on phase units in the hierarchical tensor space, and tensor overlap fusion processing is performed on the bound units to obtain a composite propagation unit structure. Specifically, phase units in the hierarchical tensor space are doubly located and marked according to their hierarchical number and propagation axis position, and the phase trajectory sequence, potential energy change curve, and adjacent pointing relationship of each unit are recorded simultaneously. Then, cross-layer mirror retrieval is performed on all units, that is, units that show symmetrical phase change direction, similar node arrangement structure, and corresponding axial position mapping relationship in different layers are matched and bound one by one. Subsequently, phase sequence alignment is performed on the bound unit pairs, that is, the phase trajectories on both sides are aligned point by point in time progression order. The process involves comparing and contrasting the nodes, rearranging and interpolating the positions where phase shifts or jumps occur, and then performing overlapping fusion on the aligned mirror units. This involves rearranging and combining the two sets of node sequences point by point in an interlaced manner, extracting nodes alternately in axial order and rearranging them, merging and connecting the overlapping nodes, and reassigning a unified number to the fused connection relationships. Then, a topological scan is performed on the fused structure, which involves disassembling the paths with forked connections, loop connections, or redundant intersections one by one and reconstraining them into a single continuous main chain structure according to the propagation direction. Finally, the reconstructed overall structure is sequentially traversed and chained along the hierarchical levels to generate a composite propagation unit structure. Adaptive orchestration is performed on the composite propagation unit structure along the propagation axis, and continuity tensor compensation is performed at the stack boundaries to generate an orchestration tensor structure. Specifically, the nodes in the composite propagation unit structure are numbered and mapped point by point along the propagation axis. The phase state, connection direction, and hierarchy of each node are written into the axial sequence record. Based on this axial sequence, continuous scanning is performed. Intervals where the phase changes between adjacent nodes are consistent and the connection directions are continuous are divided into orchestration segment units. Then, the node order within each orchestration segment unit is rearranged, and nodes that are arranged in reverse or staggered order are rearranged according to the propagation axis direction. The sequence is unidirectional and progressive, and the connection relationships between nodes are updated synchronously. Then, all orchestration fragment units are stacked according to the hierarchical boundaries, that is, the structure is divided into multiple stack units, and the upper and lower boundary node sets are extracted at the intersection of adjacent stacks. Then, the boundary node sets are connected point by point, and relay connection nodes are inserted between corresponding nodes in adjacent stacks that have broken connections. Based on the connection relationship after the connection is repaired, the rewriting process is performed, that is, duplicate connections and redundant paths are cleaned up one by one and renumbered. Finally, all orchestration fragments are sequentially spliced ​​and chained together along the propagation axis to form an orchestration tensor structure. Global topological convergence modulation is performed on the orchestrated tensor structure, and structural folding resolution is performed on cross-layer repeated propagation relationships, ultimately forming a unified tensor representation domain that continuously expands along the propagation axis. Specifically, nodes in the orchestrated tensor structure are sequentially numbered along the propagation axis, and the phase state, level affiliation, and preceding and following connection points of each node are simultaneously written. Then, using the axial number as the primary index, all node connection relationships are scanned path by path. Propagation relationships that appear in different levels but have consistent connection directions and highly overlapping path segments are marked as cross-layer repeated propagation relationships, and these relationships are uniformly archived. Subsequently, the global connection... The structure performs a main chain filtering operation, selecting a unique main connection path from multiple connection paths of each node according to propagation continuity and phase consistency, and stripping the remaining connection paths one by one and recording them as backup mapping relationships. Then, the stripped structure is folded, merging and rewriting the node sequences in the cross-layer repeated propagation paths that are completely consistent in axial order and synchronized in phase change, compressing multiple parallel repeated paths into a single continuous path, and rearranging the axial number of the merged path. Finally, the processed overall structure is traversed point by point along the propagation axis and chained out, ultimately forming a continuously unfolded unified tensor representation domain.

[0028] S4, based on the dynamic priority scheduling decision primitive set, triggers the scheduling engine to execute mirror reverse isomorphic evolution and tunnel multi-domain resource collaborative self-adaptive adjustment; In this embodiment, based on the dynamic priority scheduling decision primitive set, the scheduling engine is triggered to execute mirror reverse isomorphic evolution and tunnel multi-domain resource collaborative self-adaptive adjustment, as follows: The dynamic priority scheduling decision primitives are axially arranged, and each decision primitive is mirror-mapped to form a mirror evolution reference sequence. Specifically, the dynamic priority scheduling decision primitives are axially sorted, and the propagation start node, propagation end node, and relay connection position in each decision primitive are sequentially numbered. Then, all decision primitives are aligned along a unified propagation axis, that is, decision primitives that originally had backpropagation relationships are flipped according to the unified propagation direction, and their internal connection order is rearranged. Finally, mirror position mapping is performed on the decision primitives after the direction unification is completed, and the nodes of each decision primitive on the propagation axis are mirrored. The point arrangement order is rewritten symmetrically, and the connection relationship of the mirrored nodes is marked point by point to correspond with the original connection relationship. Further, boundary merging is performed on mirrored units with overlapping paths or repeated propagation segments, that is, the repeated propagation segments are folded and compressed, and the unidirectional connection is rearranged on the converging nodes. Then, progressive chaining is performed on the merged mirrored structure, and the termination node of the preceding mirrored unit and the starting node of the following mirrored unit are chained together according to the propagation continuation order. At the same time, the discontinuous connection at the chaining position is filled and reconnected. Finally, the overall mirrored chain structure is uniformly traversed and ordered according to the propagation axis to form a mirrored evolution reference sequence. The mirror evolution reference sequence is subjected to reverse propagation backtracking rearrangement, that is, the propagation path is pushed back segment by segment along the resource propagation termination direction. At the same time, repeated propagation segments at the turnaround connection points are compressed and merged to form a reverse mirror evolution chain. Specifically, each propagation chain segment in the mirror evolution reference sequence is located at the end of the propagation termination node, and then pushed back segment by segment along the reverse propagation direction with the termination node as the starting position. That is, the preceding connection node, propagation transition position and path turnaround point of the current chain segment are read in sequence, and the internal node order of the chain segment is rearranged according to the reverse propagation order. Then, the propagation chain segments that have been reversed are backtracked and spliced, that is, the end node of the previous chain segment is reversed and connected to the starting node of the next chain segment, and any interrupted connection points are re-connected. Insert a relay connection segment and rewrite the corresponding propagation direction identifier. Then, perform repeated scanning on the backtracking connection segments that appear during the backtracking process. That is, compare the segments with the same propagation path, consistent node arrangement and opposite propagation direction one by one, and fold and compress redundant nodes in the repeated segments to retain only a single continuous propagation link. Further, perform reverse connection and reorganization on the compressed propagation structure, reconnect the nodes with jump connections in the backtracking order, and perform staggered reversal arrangement on the cross propagation positions. Subsequently, perform unified axial convergence on all reverse chain segments, and connect the scattered chain segments in the reverse propagation order. Finally, perform sequence regularization and link rearrangement on the overall reverse propagation structure to form a reverse mirror evolution chain. Heterogeneous link penetration and reorganization are performed on resource propagation segments in the reverse mirror evolution chain, and the propagation direction at the continuation position is synchronously reoriented. Simultaneously, progressive nested reordering is performed on cross-propagation segments to form a multi-domain through-propagation structure. Specifically, each resource propagation segment in the reverse mirror evolution chain is segmented and grouped according to link type, propagation direction, and node connection level, and a corresponding mapping relationship is established for the connection endpoints in different links. Then, penetration reconnection is performed on segments with cross-link propagation associations, that is, the terminating node in one link is directly inserted into the relay node position in another link, and the connection order and propagation direction between the nodes on both sides are rewritten synchronously. Finally, direction reordering is performed on the propagation segments after penetration reconnection. The process involves: 1) readjusting the original reverse propagation arrangement to a unified direction of propagation; 2) swapping the positions and rearranging the links of reverse connection nodes that appear during the reversal process; 3) performing progressive nested reordering on segments with intersecting propagation paths, i.e., inserting intersecting links into the main propagation chain layer by layer according to the propagation time sequence; 4) performing staggered arrangement on overlapping nodes; 5) redistributing the positions of multiple connection paths at the same position according to the propagation sequence; 6) performing cross-domain link connection and reorganization on the overall structure after nesting, connecting discontinuous connection segments according to the propagation continuation direction, and merging and compressing duplicate connection positions; and 7) sequentially connecting and reconstructing all propagation segments along a unified propagation axis to form a multi-domain connected propagation structure. Cooperative occupation migration is performed on resource scheduling paths in the multi-domain traversal propagation structure to form a cooperative scheduling propagation sequence. Specifically, the resource scheduling paths in the multi-domain traversal propagation structure are deconstructed and calibrated path by path, and the occupation starting position, occupation duration segment, and path switching node of each path are recorded sequentially. The resource occupation timing between different paths is synchronized and aligned. Then, a cooperative occupation migration operation is performed on the aligned scheduling paths. That is, taking the replaceable occupation segment between paths as a unit, the entire sequence of occupied nodes in one path is migrated to the empty occupation window of another path, and the connection pointers and propagation connection relationships of the nodes before and after the migration are updated synchronously. Finally, the migrated path set is processed. The system performs a structural re-orchestration, reconnecting consecutive occupied segments in the propagation order and merging and compressing duplicate occupied nodes, retaining only single occupied records and main connection relationships. Furthermore, it performs nested reorganization on path segments with overlapping occupied relationships, embedding time-preceding occupied segments into the intermediate occupied gaps of subsequent paths and interleaving the connection relationships of nodes at the embedded boundary positions. Subsequently, it performs continuous splicing on the overall path structure, continuing interrupted occupied segments segment by segment according to the propagation direction of adjacent paths and converging redundant branch connections. Finally, it performs sequential rearrangement and chain integration of all paths along a unified propagation axis to form a collaborative scheduling propagation sequence. Dynamic phase co-evolution is performed on the cooperative scheduling propagation sequence, and phase migration is synchronously advanced along the propagation direction. Simultaneously, offset segments appearing during co-propagation are mirrored and reconnected to form a continuous co-propagation structure. Specifically, phase markers are sequentially written to each propagation node in the cooperative scheduling propagation sequence, and a phase state field and a propagation direction field are added to each node. Then, the node sequence is traversed point by point according to the propagation direction. During the traversal, the phase difference between adjacent nodes is continuously recorded, and a phase migration trajectory chain is constructed for each node. Then, the phase migration trajectory chain is synchronously advanced, that is, the phase state of the previous node is sequentially passed to the next node, and during the transmission... During the process, the phase value is updated by successive inheritance, and the node positions where phase changes occur are marked and registered. Further, the segments marked as offset are mirrored and reconnected, that is, the first half and the second half of the segment are flipped with the midpoint node of the offset segment as the axis of symmetry, and the node sequence after flipping is reassembled according to the original propagation path. At the same time, the connection points of the spliced ​​nodes are rewritten into a unidirectional continuous structure. Then, the overall link is connected for all reconnected segments, the broken nodes between adjacent segments are connected point by point according to the propagation direction, and duplicate connection relationships are cleaned up. Finally, the entire sequence is uniformly traversed and continuously arranged according to the propagation direction to form a continuous accompanying propagation structure. A multi-domain resource self-adaptive reconstruction is performed on the continuous co-propagation structure, and cross-domain replacement migration is performed on propagation segments with occupancy conflicts to form a multi-domain collaborative propagation chain. Specifically, each propagation node in the continuous co-propagation structure is identified and encoded according to resource type, occupancy status, and propagation direction. Then, nodes of different resource types are decomposed into domains, and nodes within the same resource domain are reordered according to the propagation path order. At the same time, a one-to-one mapping relationship table is established for cross-domain nodes. Next, the occupancy position relocation operation is performed on the nodes in the mapping relationship table, that is, the occupancy interval of the original node in the propagation axis is read and converted into an equivalent occupancy interval in the target resource domain, and a space is reserved in the target interval for the node to occupy the corresponding resource. The insertion position is determined, and the corresponding segment in the original path is stripped of its occupancy mark and the migration status is recorded. Then, the set of nodes with conflicting occupancy is screened point by point, and the multiple occupancy records of the same propagation position are split. Based on the cross-domain mapping relationship, the migrateable nodes are written into the free segment of the target path, and the connection pointer relationship between its predecessor node and successor node is updated. Furthermore, the updated propagation path in each resource domain is reordered, the broken segments are reassembled in chronological order, and the duplicate occupancy nodes are merged, retaining only the main occupancy record and the main connection path. Finally, all resource domain paths are uniformly traversed and chainedly merged along the propagation axis to generate a multi-domain collaborative propagation chain. The multi-domain collaborative propagation chain undergoes mirrored reverse convergence adjustment. Resource segments with opposite propagation directions but corresponding link structures are mirrored and folded. Unified axial regularization is then performed on the folded propagation nodes to ultimately complete the tunnel's multi-domain resource collaborative self-adaptive adjustment. Specifically, each propagation node in the multi-domain collaborative propagation chain is encoded node-by-node according to its propagation axis coordinates, direction identifier, and resource occupancy status. Subsequently, forward and reverse sequences are generated synchronously across the entire link. This involves constructing bidirectional traversal sequences based on the propagation start and end points, and comparing and matching node segments with identical structures but opposite directions in both sequences. Finally, mirror mapping is performed on the successfully matched segments, converting the reverse segments to their axial index positions. The forward sequence corresponds to the coordinate interval, and the nodes are stacked and aligned at the same axial position. At the same time, the resource usage records of the stacked nodes are merged and written at the field level, and duplicate connection paths are cleared, retaining only the primary adjacency pointer relationship. Then, the stacked node set is folded at the segment level, and continuous repeated or highly overlapping propagation segments are compressed and merged according to the axial interval. The predecessor and successor connection relationship pointers are regenerated for the interval boundary nodes. Subsequently, the entire link is axially consistent and rearranged. All nodes are reordered according to the unified propagation direction, and replacement nodes are inserted point by point at the cross-interval broken connection positions and the link pointer relationship is rebuilt. Finally, the rearranged propagation chain is traversed in its entirety and output in a chain-like manner to form the final multi-domain collaborative propagation chain structure.

[0029] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, in the form of a computer program product.

[0030] Those skilled in the art will recognize that the modules and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0031] In addition, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0032] The above description is merely a specific technical solution of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0033] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A machine learning-based digital optical cable scheduling method for energy and communication co-transmission cable tunnels, characterized in that, include: Collect the electromagnetic-optical-communication coupled situational flow of the energy-communication tunnel, generate a superdimensional embedded representation of the coupled situation, and perform topological manifold folding and reconstruction to form a set of scheduling self-organizing decision units; By adjusting the hidden phase folding and resolving the topological resonance of the execution path conflict of the scheduling self-organizing decision-making unit set, a fused resource situation tensor scheduling graph domain is generated. Based on the fusion of resource situation tensor scheduling, the resource occupation trajectory representation learning and nonlinear phase recoding are fused to form a dynamic priority scheduling decision primitive set; Based on the dynamic priority scheduling decision primitive set, the scheduling engine is triggered to execute mirror reverse isomorphic evolution and tunnel multi-domain resource collaborative self-adaptive adjustment.

2. The machine learning-based digital optical cable scheduling method for power and communication co-transmission cable tunnels according to claim 1, characterized in that, The process involves collecting the electromagnetic-optical-communication coupled situational flow of the energy-communication tunnel, generating a coupled situational hyperdimensional embedded representation, and performing topological manifold folding and reconstruction to form a set of scheduling self-organizing decision units, as detailed below: Collect the electromagnetic-optical-information coupling situation flow of the energy-information co-transmission tunnel, perform unified timing alignment, and establish an association mapping according to the link topology to form the electromagnetic-optical-information coupling situation flow; A multimodal temporal learning model is introduced to perform cross-source feature encoding and link semantic alignment preprocessing on the electromagnetic-optical-information coupling situation flow, and self-supervised training is performed to learn the joint temporal mapping relationship to form an electromagnetic-optical-information coupling situation association sequence. Based on the electromagnetic-optical-information coupled situational correlation sequence, high-dimensional situational embedding reprojection and nonlinear interactive unwrapping are performed to generate a coupled situational hyperdimensional embedded representation. We perform topological manifold phase migration constraint folding and perturbation propagation path evolution control on the coupled state hyperdimensional embedded representation to construct a multi-scale manifold potential skeleton structure; Based on the multi-scale flow morphology potential skeleton structure, the link behavior trajectory temporal decoupling and recalibration and the situation evolution path semantic alignment are performed to form a link behavior evolution sequence; Dynamic adjacency reconstruction and implicit topology self-expansion divergence are performed on the link behavior evolution sequence to generate a link cooperative evolution structure; By performing multi-level situation compression mapping and perturbation sensitivity hierarchical folding on the link cooperative evolution structure, a multi-level situation manifold structure domain is constructed. The semantic normalization mapping of scheduling behavior and the aggregation of link resource competition situation are performed on the multi-level situational manifold structure domain to form a set of scheduling self-organizing decision units.

3. The machine learning-based digital optical cable scheduling method for power and communication co-transmission cable tunnels according to claim 2, characterized in that, The electromagnetic-optical coupling situational correlation sequence is used to perform high-dimensional situational embedding reprojection and nonlinear interactive unwrapping to generate a coupled situational hyperdimensional embedding representation, as detailed below: Based on the electromagnetic-optical-information coupling situational correlation sequence, a time-series sliding decomposition is performed to construct a cross-domain path coupling mapping primitive set; Perform link state screening and path propagation direction realignment on the cross-domain path coupling mapping primitive set to form a path collaborative association cluster; Based on path-cooperative association clusters, perform multi-scale topology rejoining to generate an implicitly coupled topology skeleton network; By performing topological neighborhood hierarchical expansion on the implicitly coupled topological skeleton network, a multi-layer path association sequence is formed; Perform relation tensor reconstruction on the multi-level path association sequence to obtain the reconstructed path tensor representation; Based on the reconstructed path tensor representation, the latent space manifold expansion mapping is performed to form a continuously evolving manifold structure domain; Perform cross-domain semantic distributed alignment on the continuously evolving manifold structure domain, and perform high-dimensional embedding projection and topological sparsification rearrangement on the manifold structure to output a coupled situational hyperdimensional embedding representation.

4. The machine learning-based digital optical cable scheduling method for power and communication co-transmission cable tunnels according to claim 2, characterized in that, The process involves performing topological manifold phase migration constraint folding and perturbation propagation path evolution control on the coupled state hyperdimensional embedded representation to construct a multi-scale manifold potential skeleton structure, as detailed below: Based on the coupled situation hyperdimensional embedding characterization, the perturbation energy level trajectory is deconstructed to form a cross-domain perturbation path cluster set; Perform dominant propagation chain screening and path resonance aggregation on the set of cross-domain perturbation path clusters to generate a set of perturbation dominant topological ridges; Phase gradient hierarchical embedding is performed based on the perturbation-dominant topological ridge set to obtain a hierarchical phase-coupled topological cluster; Based on the hierarchical phase-coupled topological clusters, topological constraint manifold folding is performed to form a cross-phase skeleton mapping framework; Based on the cross-phase skeleton mapping framework, perturbation propagation trajectory control is performed, and a hierarchical propagation constraint link is established; By performing multi-scale topological extrapolation expansion on the hierarchical propagation constraint links, a stable propagation ridge structure domain is formed; Perform cross-scale phase fusion reconstruction on the stable propagation ridge domain to output a multi-scale flow potential skeleton structure.

5. The machine learning-based digital optical cable scheduling method for power and communication co-transmission cable tunnels according to claim 1, characterized in that, The process involves adjusting the hidden phase folding and resolving the topological resonance of the execution path conflict in the self-organizing decision-making unit set of the scheduling system to generate a fused resource situation tensor scheduling graph domain, as detailed below: Based on the self-organizing decision-making unit set of scheduling, a graph sequence joint learning model is introduced to encode the temporal features and learn the propagation dependency representation of the occupancy evolution state, forming an implicit resource conflict evolution flow; Perform hidden phase folding migration of the implicit resource conflict evolution flow to form a cluster of phase-coupled conflict chains; Based on the phase-coupled conflict chain cluster, topological resonance-induced reconstruction is performed to form a set of resonance-propagating topological ridge chains; By performing multi-domain conflict ablation modulation on the resonant propagation topological ridge chain set, a stable propagation coupling domain is formed; Based on the stable propagation coupling domain, a hierarchical topological collapse mapping is performed to form a set of hierarchical propagation constraint contexts. Perform cross-layer resource situation tensor coupling on the hierarchical propagation constraint context set to form a resource situation tensor correlation domain; By performing topological manifold resonance regularization on the resource status tensor correlation domain, a fused resource status tensor scheduling graph domain is output.

6. The machine learning-based digital optical cable scheduling method for power and communication co-transmission cable tunnels according to claim 4, characterized in that, The implicit resource conflict evolution flow is subjected to a contention path implicit phase folding migration to form a phase-coupled conflict chain cluster, as detailed below: The implicit resource conflict evolution flow is structured and temporally unfolded according to the propagation event chain, and multiple propagation trajectory units are established for each competition path to form a cross-path temporal phase unit set. Perform multi-source temporal anchoring mapping on the cross-path temporal phase unit set to construct an out-of-path temporal anchoring constraint network; Based on the cross-temporal anchoring constraint network, hidden phase folding migration is performed to form a continuous evolutionary conflict chain segment; Perform topological re-representation reconstruction on the continuously evolving conflict chain segments to form a multi-directional conflict coupling node group; Path mirroring folding is performed on a group of multi-directional conflicting coupled nodes to obtain a high-density coupled structural unit. Closed propagation substructures are generated by performing hidden phase energy level hierarchical mapping on high-density coupled structural units; The closed propagation substructure is subjected to cross-level hierarchical fusion and path reorganization, and the connected structure is subjected to overall topology regularization and redundancy convergence, outputting a cluster of phase-coupled conflict chains.

7. The machine learning-based digital optical cable scheduling method for power and communication co-transmission cable tunnels according to claim 4, characterized in that, The stable propagation coupling domain is formed by performing multi-domain conflict ablation modulation on the resonant propagation topological ridge chain set, as detailed below: The set of resonant propagation topological ridge chains is phase-expanded along the propagation topological axis, and the propagation behavior of the nodes is mapped to form a continuous coherent arrangement structure. For the path intersection region in the continuous coherent arrangement structure, a propagation potential field is constructed, and the intersection region of the propagation potential field is hierarchically assigned to form a conflict potential field distribution zone. Based on the conflict potential field distribution band, phase decoupling adjustment is performed, and path-dependent constraint relationships are established; Perform boundary coherent migration on path-dependent constraints to obtain a continuously drifting boundary structure; By performing tensor quantization mapping recoding on the continuous drift morphological boundary structure, a path tensor quantization expression state renormalization sequence is generated. Based on the path tensor expression state renormalization sequence, phase-constrained projection is applied to form a continuous coherent path tensor chain state; Based on the continuous coherent path tensor chain state, topological potential redistribution modulation is performed to output a stable propagation coupling domain.

8. The machine learning-based digital optical cable scheduling method for power and communication co-transmission cable tunnels according to claim 1, characterized in that, The resource situation tensor scheduling body, based on fusion resource occupancy trajectory representation learning and nonlinear phase recoding, is fused to form a dynamic priority scheduling decision primitive set, as follows: Based on the fusion resource situation tensor scheduling body, spatiotemporal expansion and hidden state encoding of occupied trajectories based on sequence learning model are performed to obtain continuous trajectory unit clusters; Cross-chain phase stretching reconstruction is performed on continuous trajectory unit clusters, and neighborhood propagation constraints are introduced to perform trajectory folding and embedding to form a continuous evolution chain core; By performing resource state tensor up-dimensional mapping on the core of the continuous evolution chain and reorganizing the tensor stack along the propagation axis, a unified tensor expression domain is obtained. Nonlinear phase recoding fusion is performed based on a unified tensor representation domain, and phase spectrum splitting and re-aggregation operations are applied to the encoding results to generate a clustered distribution structure. Perform cross-domain dynamic rearrangement and reordering on clustered distribution structures to form a unified evolutionary sequence; Dynamic priority tensor calibration is performed on the unified evolutionary sequence body, and hierarchical tensor decomposition and recombination are applied to the encoding result to obtain a hierarchical tensor structure; Based on a hierarchical tensor structure, it performs intra-layer consistent aggregation and cross-layer conflict inference scheduling based on a decision learning model, and outputs a dynamic priority scheduling decision primitive set.

9. The machine learning-based digital optical cable scheduling method for power and communication co-transmission cable tunnels according to claim 7, characterized in that, The process involves performing resource state tensor up-dimensional mapping on the kernel of the continuous evolution chain and reorganizing the tensor stack along the propagation axis to obtain a unified tensor representation domain, as detailed below: Phase trajectory deconstruction is performed on the core of the continuous evolution chain to construct the phase reference projection path; Based on the phase reference projection path, the propagation potential energy coupling characterization is performed on adjacent trajectory units to form potential energy identification units, and gradient hierarchical mapping is performed on the potential energy identification units to establish a multi-level propagation potential energy sequence. A non-Euclidean phase stretching mapping is performed on the multi-level propagation potential energy sequence to form a nonlinear continuous unfolded phase structure. Based on the nonlinear continuous expansion phase structure, tensor domain embedding mapping is performed to form a continuous adjacency chain; Cross-domain mirror coupling reconstruction is performed on the phase units in the hierarchical tensor space to obtain the composite propagation unit structure; Adaptive orchestration is performed on the composite propagation unit structure along the propagation axis to generate an orchestration tensor structure; Global topological convergence modulation is performed on the orchestrated tensor structure, and structural folding resolution is performed on the cross-layer repetitive propagation relationship to form a unified tensor representation domain.

10. The machine learning-based digital optical cable scheduling method for power and communication co-transmission cable tunnels according to claim 1, characterized in that, The process of triggering the scheduling engine to perform mirror reverse isomorphic evolution and tunnel multi-domain resource collaborative self-adaptive adjustment based on the dynamic priority scheduling decision primitive set is as follows: The dynamic priority scheduling decision primitive set is axially arranged, and each decision primitive is mirror-mapped to form a mirror evolution reference sequence; Perform backpropagation and rearrangement on the mirror evolution reference sequence to form a reverse mirror evolution chain; Heterogeneous link penetration and reorganization are performed on the resource propagation segments in the reverse mirror evolution chain to form a multi-domain through propagation structure; Perform coordinated occupation migration on the resource scheduling paths in the multi-domain interconnected propagation structure to form a coordinated scheduling propagation sequence; Dynamic phase co-evolution is performed on the coordinated scheduling propagation sequence to form a continuous co-propagation structure; Perform multi-domain resource self-adaptive reconstruction on the continuous co-propagation structure to form a multi-domain collaborative propagation chain; Perform mirror reverse convergence adjustment on the multi-domain collaborative propagation chain, and perform uniform axial regularization on the folded propagation nodes.

Images