Carrier communication main micro-networking transmission method and system

Abstract

This invention discloses a carrier communication master-distributor micro-network transmission method and system, comprising: collecting carrier communication state data and generating node timing state records; constructing a master-distributor carrier communication graph structure based on the node timing state records, mapping nodes as graph nodes and establishing graph edge relationships; inputting the master-distributor carrier communication graph structure and node timing state records into an improved DCRNN network to construct a carrier causal precursor; generating causal succession chains based on the carrier causal precursors, and constructing a master-distributor micro-network structure based on the intersection relationships of the causal succession chains; constructing a time compression window, generating causal folding pieces, and associating data in the same chain to generate cross-cycle causal closed records; generating node probability state vectors, updating the master-distributor micro-network structure, and executing carrier communication transmission. This invention achieves stable transmission in a master-distributor micro-network by constructing a carrier causal precursor, causal succession chains, and cross-cycle causal closed records.

Description

Technical Field

[0001] This invention relates to the field of smart grid communication technology, and in particular to a carrier communication master-distributor micro-network transmission method and system. Background Technology

[0002] In low-voltage distribution transformer areas, master nodes, distribution-side nodes, and end nodes typically interact via power line carrier communication to complete power data acquisition, distribution status monitoring, and end-device communication. Existing carrier communication micro-networking methods generally establish a communication topology based on communication success rate, signal strength, response delay, bit error rate, or fixed relay paths, and then complete data uploading through master node polling or relay forwarding.

[0003] However, low-voltage distribution lines are characterized by numerous branches, frequent load changes, unstable phase-line coupling, significant impedance abrupt changes, and strong sudden noise interference. The same node may exhibit alternating states of success, failure, delay, and retransmission in different communication cycles. Existing methods primarily focus on single-cycle or local link quality, making it difficult to reflect the stable forward transmission capability and transmission-taking capability of a node across multiple communication cycles. This can easily lead to inaccurate micro-group partitioning and unstable selection of receiving nodes.

[0004] Existing technologies typically handle short-term communication failures at nodes by immediately retransmitting, switching routes, or reconfiguring the network. This can easily increase carrier channel occupancy, causing local anomalies to spread into network-wide scheduling changes. For data that can be recovered across cycles, existing methods lack compression and reconstruction mechanisms based on causal relationships and probabilistic evolutionary update mechanisms, making it difficult to maintain the continuity of the master-slave micro-network structure and the stability of data uploads.

[0005] Therefore, how to provide a carrier communication master-slave micro-network transmission method and system is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0006] One objective of this invention is to propose a carrier communication master-distributor micro-network transmission method and system. This invention constructs a master-distributor carrier communication graph structure and combines it with an improved DCRNN network to extract the temporal diffusion trajectory of nodes, establishing a carrier causal precursor, a causal succession chain, and a cross-cycle causal closed record. It utilizes a time compression window and probabilistic state vectors to achieve dynamic updating and adaptive transmission control of the master-distributor micro-network structure, completing stable networking and reliable transmission in the low-voltage distribution radio area carrier communication process. It has the advantages of accurate micro-group division, fewer retransmissions, strong anti-interference ability, high communication continuity, and good networking stability.

[0007] A carrier communication master-slave micro-network transmission method according to an embodiment of the present invention includes:

[0008] Collect carrier communication status data of master nodes, distribution side nodes and end nodes in low-voltage distribution radio area in multiple consecutive communication cycles, and generate node timing status records;

[0009] Construct a primary and secondary carrier communication graph structure based on the node timing status records. The primary node, secondary node, and terminal node are used as graph nodes, and the node pairs with successful communication records are used as graph edges. Write the communication success rate, average response delay, bit error rate, and delay variation for each graph edge.

[0010] The main and secondary carrier communication graph structure and the corresponding node timing state records are input into the improved DCRNN network. The timing diffusion trajectory corresponding to each node is extracted. Based on the sequential and continuous relationships between the timing diffusion trajectories of different nodes, node causal precedence pairs are generated. Based on all node causal precedence pairs, a carrier causal precedence body is constructed.

[0011] Based on the carrier causal precedence, a causal succession chain is generated according to the transmission order of the node causal precedence pairs. The starting succession node, the continuing succession node and the ending succession node in each causal succession chain are identified. The main and distribution micro-network structure is constructed according to the intersection relationship between each causal succession chain.

[0012] Construct a time compression window covering multiple communication cycles, mark incomplete transmission data as data to be closed, generate causal folding pieces based on the causal succession chain corresponding to the data to be closed, and generate cross-cycle causal closure records by associating data in the same chain.

[0013] Based on the node timing state record, carrier causal precursor, causal succession chain, and cross-cycle causal closure record, a probability state vector is generated for each node, the master-supplier micro-network structure is updated, and carrier communication transmission between the master node, the supplier node, and the end node is executed.

[0014] Optionally, the carrier communication status data includes node identifier, phase line identifier, branch number, communication success marker, communication failure marker, upload timestamp, response delay, bit error rate, retransmission count, and noise intensity.

[0015] Optionally, the generation of node timing state records includes:

[0016] Read the node identifier, phase line identifier, branch number, communication success mark, communication failure mark, upload timestamp, response delay, bit error rate, retransmission count and noise intensity of each node within a continuous communication cycle;

[0017] The communication status data corresponding to the same node are collected according to the node identifier;

[0018] Based on the upload timestamp, the collected communication status data is sorted chronologically to generate a communication status sequence for the corresponding node;

[0019] The communication success marker, communication failure marker, response delay, bit error rate, retransmission count, and noise intensity in the communication state sequence are associated in the order of the communication cycle to generate the time sequence state record of the corresponding node.

[0020] Optionally, the step of constructing the primary and secondary carrier communication graph structure based on the node timing state records includes:

[0021] Read the node identifier, node type, phase line identifier, branch number, communication cycle number, communication success mark, communication failure mark, upload timestamp, response delay and bit error rate from the node timing status record, and register the master node, the auxiliary node and the end node as nodes to be built in the map respectively;

[0022] The communication records of each node to be mapped are aligned according to the communication cycle number. Node pairs that have carrier communication interaction within the same communication cycle are extracted. The communication success mark, communication failure mark, response delay and bit error rate corresponding to each node pair are collected into node pair communication records.

[0023] Based on the communication records of each node pair, the total number of communications and the number of successful communications are counted. The proportion of successful communications in the total number of communications is determined as the communication success rate, and the node pairs with successful communication records are determined as candidate graph edges.

[0024] Read the response delay and bit error rate corresponding to each candidate graph edge, calculate the average response delay, average bit error rate and response delay change amplitude of adjacent communication cycles according to the communication cycle order, and determine the statistical result of the response delay change amplitude of adjacent communication cycles as the delay change amount;

[0025] The registered master node, partner node, and end node are used as graph nodes, and the candidate graph edges are used as the connection relationships between the graph nodes. The corresponding communication success rate, average response delay, bit error rate, and delay change are written into the candidate graph edges to generate the master-partition carrier communication graph structure.

[0026] Optionally, constructing a carrier causal precedence body based on all node causal precedence pairs includes:

[0027] Using the communication cycle as the time step, the communication success marker, communication failure marker, upload timestamp, response delay, bit error rate, retransmission count, and noise intensity of each node are constructed as node feature vectors, which are then stacked in the order of graph nodes to generate the spatiotemporal input tensor of the improved DCRNN network.

[0028] An adaptive multi-channel adjacency mapping layer is added to the input of the original DCRNN. ​​Based on the communication success rate and bit error rate of the graph edges, two channels, the backbone diffusion adjacency matrix and the noise suppression adjacency matrix, are dynamically generated. During each forward calculation, the two channels are weighted and merged using learnable fusion coefficients to obtain a diffusion adjacency matrix adapted to the current carrier environment.

[0029] The spatiotemporal input tensor is subjected to diffusion convolution using a diffusion adjacency matrix. The output of the diffusion convolution is fed into a gated recurrent unit. A causal attention gate is embedded inside the gated recurrent unit. The causal attention gate uses a multi-head attention mechanism to assign different causal weights to the historical hidden states and performs weighted updates on the hidden states at the current time step, outputting a node temporal diffusion state sequence.

[0030] The node temporal diffusion state sequence is input into the temporal drift normalization layer. The temporal drift normalization layer uses the average upload time of the communication cycle as a benchmark to perform amplitude scaling and centering on the upload time residual of each node, generating the node hidden state after drift normalization.

[0031] After the drift is normalized, the hidden state of the node is input into the dual-channel timing readout layer. One channel outputs the node's leading stability vector, and the other channel outputs the receiving stability vector. The node's causal leading pairs are generated by combining the leading stability, receiving stability and the order of node upload time. All node causal leading pairs are then connected and combined to form a carrier causal leading body.

[0032] A cross-entropy loss function is constructed using the true preceding order label of the communication cycle and the outputs of node preceding stability and succession stability. An adjacent time step smoothing regularization term is added to jointly optimize the improved DCRNN network. Batch size, learning rate and learning rate decay strategy are set, and early stopping is triggered when the loss on the validation set does not decrease significantly.

[0033] Optionally, the construction of the master-supplier micro-network structure based on the intersection relationships between each causal link includes:

[0034] Read all node causal precedence pairs in the carrier causal precedence body, and extract the preceding node, succeeding node, causal propagation direction and communication cycle marker from each node causal precedence pair;

[0035] The causal precedence pairs of nodes are sorted according to the causal transmission direction and communication cycle marker. The causal precedence pairs of nodes whose successor node is the same as the predecessor node of the next node's causal precedence pair and whose communication cycles are adjacent are sequentially connected to generate a causal succession chain.

[0036] Read the order of nodes within each causal succession chain, mark the node at the head of the chain as the starting succession node, mark the node between the head and tail of the chain that connects two causal pairs as the continuing succession node, and mark the node at the tail of the chain as the ending succession node.

[0037] Compare the starting node, continuing node, and ending node in different causal succession chains. Establish inter-chain sharing relationships for causal succession chains with the same node. Establish inter-chain continuity relationships for causal succession chains where the ending node of one causal succession chain is the same as the starting node of another causal succession chain. Generate a set of intersecting succession chains based on the inter-chain sharing relationship and the inter-chain continuity relationship.

[0038] After deduplicating all nodes in the same causal link set, a microgroup node set is generated. Nodes belonging to two or more causal links are identified as microgroup linking nodes. The order of nodes within each causal link is used as the transmission order within the microgroup. The shared relationship between links and the continuity relationship between links are used as the linking relationship between microgroups. A master-slave microgroup network structure is generated, which includes microgroup identifier, microgroup node set, microgroup linking nodes, transmission order within the microgroup, and linking relationship between microgroups.

[0039] Optionally, the step of generating a causal fold based on the causal chain corresponding to the data to be closed, and generating a cross-period causal closure record by associating data in the same chain, includes:

[0040] Read the microgroup identifier, causal succession chain, node order within the chain, and communication cycle marker in the main and auxiliary microgroup network structure. Select a continuous communication cycle according to the preset window length and write the node upload data, node receive data, transmission status marker, and frame check marker within the continuous communication cycle into the time compression window.

[0041] Within the time compression window, read the transmission status flag and frame verification flag of each node's data, mark data with upload failure, reception timeout, frame verification failure or chain interruption as data to be closed, and write node identifier, data frame identifier, communication cycle identifier and the causal chain identifier to be closed for the data to be closed.

[0042] Based on the causal chain to which the data to be closed belongs, extract the node position, previous successor node, current successor node and end successor node corresponding to the data to be closed in the chain, and combine the data frame identifier, communication cycle identifier, node position in the chain and successor node information of the data to be closed to generate a causal folding piece;

[0043] Read the causal relay chain identifier corresponding to the causal folding piece, extract different communication cycle data belonging to the same causal relay chain within the time compression window, and associate the different communication cycle data according to the node order, communication cycle order and data frame identifier to generate the same chain data association relationship;

[0044] The causal folding piece, the data association relationship of the same chain, the corresponding causal successor chain identifier, the data status to be closed and the closure result mark are combined to generate a cross-cycle causal closure record, and the cross-cycle causal closure record is written into the corresponding master-slave micro-network structure.

[0045] Optionally, updating the master-distributor micro-network structure and performing carrier communication transmission between the master node, the distribution-side node, and the end node includes:

[0046] Read the number of successful communication, the number of failed communication, the number of cross-cycle causal closures, the number of connection failures, and the number of causal connection chain breaks for each node within a preset statistical window, and use the counting results as node status statistical parameters.

[0047] Based on the node state statistical parameters, calculate the stable transmission probability, short-term instability probability, long-term attenuation probability, recovery probability, and connection failure probability corresponding to the node, combine them to generate a node probability state vector, and write it into the node probability state vector index table.

[0048] The main and sub-microgroup network structure is updated according to the node probability state vector index table. Nodes with stable transmission probability higher than the first threshold are retained in the original microgroup. Nodes with short-term instability probability between the first and second thresholds are marked as temporary fluctuation nodes. Nodes with long-term attenuation probability higher than the second threshold or failure probability higher than the third threshold are removed from the main transmission path of the original microgroup and reassigned as receiving nodes.

[0049] A communication scheduling table is generated based on the updated master-distributor micro-group network structure. The communication scheduling table includes micro-group identifier, core uploading node, receiving node, ordinary node, intra-group transmission order, uploading time slot allocation, and retransmission threshold. The master node organizes the master node, distribution side node, and end node to perform carrier communication transmission according to the communication scheduling table.

[0050] A carrier communication master-distributor micro-network transmission system according to an embodiment of the present invention includes:

[0051] The timing status generation module is used to collect carrier communication status data of the master node, the matching node, and the end node, and generate node timing status records.

[0052] The communication graph construction module is used to construct the primary and secondary carrier communication graph structure based on the node timing status records.

[0053] The causal precedence construction module is used to input the primary and secondary carrier communication graph structure and node timing state records into the improved DCRNN network to generate node causal precedence pairs and construct carrier causal precedence entities;

[0054] The micro-network construction module is used to generate causal succession chains based on carrier causal precursors and to construct a master-supplier micro-network structure.

[0055] The causal closure generation module is used to construct time compression windows and generate causal folded pieces and cross-period causal closure records;

[0056] The probabilistic state update module is used to generate a probabilistic state vector based on the node time-series state record, carrier causal precursor, causal succession chain, and cross-cycle causal closure record, update the master-supplier micro-network structure, and execute carrier communication transmission.

[0057] The beneficial effects of this invention are:

[0058] This invention improves the DCRNN network to extract the temporal diffusion trajectory of nodes and constructs a carrier causal precedence based on the sequential and continuous relationships between nodes. This enables the main distribution micro-network to no longer rely solely on single-cycle link quality, signal strength, or fixed routes for division. It can more accurately identify nodes in the low-voltage distribution area that have stable precedence transmission capabilities and the ability to undertake transmission, thereby improving the rationality and stability of the micro-network structure.

[0059] This invention constructs a master-supplier micro-network structure through a causal relay chain, and uses a time compression window to convert incomplete transmission data into causal folded pieces. It also associates data in the same chain to generate cross-cycle causal closed records, enabling transmission failures caused by short-term noise, load start-up and shutdown, or phase line coupling fluctuations to be closed within multiple communication cycles, reducing immediate retransmissions and repeated polling, and reducing carrier channel occupancy.

[0060] This invention generates a probability state vector based on node time-series state records, carrier causal precursors, causal succession chains, and cross-cycle causal closed records, and updates the master-supplier micro-network structure accordingly. This allows the short-term fluctuations, long-term attenuation, and recovery states of nodes to be continuously characterized, avoiding local anomalies that could trigger network-wide reorganization, and improving the continuity, anti-interference, and adaptive capabilities of carrier communication transmission between master nodes, supplier nodes, and end nodes. Attached Figure Description

[0061] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0062] Figure 1 This is a flowchart of a carrier communication master-distributor micro-network transmission method proposed in this invention;

[0063] Figure 2 This is a schematic diagram of the structure of a carrier communication master-distributor micro-network transmission system proposed in this invention. Detailed Implementation

[0064] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0065] refer to Figure 1 A carrier communication master-distributor micro-network transmission method, comprising:

[0066] Collect carrier communication status data of master nodes, distribution side nodes and end nodes in low-voltage distribution radio area in multiple consecutive communication cycles, and generate node timing status records;

[0067] Construct a primary and secondary carrier communication graph structure based on the node timing status records. The primary node, secondary node, and terminal node are used as graph nodes, and the node pairs with successful communication records are used as graph edges. Write the communication success rate, average response delay, bit error rate, and delay variation for each graph edge.

[0068] The main and secondary carrier communication graph structure and the corresponding node timing state records are input into the improved DCRNN network. The timing diffusion trajectory corresponding to each node is extracted. Based on the sequential and continuous relationships between the timing diffusion trajectories of different nodes, node causal precedence pairs are generated. Based on all node causal precedence pairs, a carrier causal precedence body is constructed.

[0069] Based on the carrier causal precedence, a causal succession chain is generated according to the transmission order of the node causal precedence pairs. The starting succession node, the continuing succession node and the ending succession node in each causal succession chain are identified. The main and distribution micro-network structure is constructed according to the intersection relationship between each causal succession chain.

[0070] Construct a time compression window covering multiple communication cycles, mark incomplete transmission data as data to be closed, generate causal folding pieces based on the causal succession chain corresponding to the data to be closed, and generate cross-cycle causal closure records by associating data in the same chain.

[0071] Based on the node timing state record, carrier causal precursor, causal succession chain, and cross-cycle causal closure record, a probability state vector is generated for each node, the master-supplier micro-network structure is updated, and carrier communication transmission between the master node, the supplier node, and the end node is executed.

[0072] In this embodiment, the carrier communication status data includes node identifier, phase line identifier, branch number, communication success marker, communication failure marker, upload timestamp, response delay, bit error rate, retransmission count, and noise intensity.

[0073] In this embodiment, the generation of node timing state records includes:

[0074] Read the node identifier, phase line identifier, branch number, communication success mark, communication failure mark, upload timestamp, response delay, bit error rate, retransmission count and noise intensity of each node within a continuous communication cycle;

[0075] The communication status data corresponding to the same node are collected according to the node identifier;

[0076] Based on the upload timestamp, the collected communication status data is sorted chronologically to generate a communication status sequence for the corresponding node;

[0077] The communication success marker, communication failure marker, response delay, bit error rate, retransmission count, and noise intensity in the communication state sequence are associated in the order of the communication cycle to generate the time sequence state record of the corresponding node.

[0078] In this embodiment, the step of constructing the primary and secondary carrier communication graph structure based on the node timing state records includes:

[0079] Read the node identifier, node type, phase line identifier, branch number, communication cycle number, communication success mark, communication failure mark, upload timestamp, response delay and bit error rate from the node timing status record, and register the master node, the auxiliary node and the end node as nodes to be built in the map respectively;

[0080] The communication records of each node to be mapped are aligned according to the communication cycle number. Node pairs that have carrier communication interaction within the same communication cycle are extracted. The communication success mark, communication failure mark, response delay and bit error rate corresponding to each node pair are collected into node pair communication records.

[0081] Based on the communication records of each node pair, the total number of communications and the number of successful communications are counted. The proportion of successful communications in the total number of communications is determined as the communication success rate, and the node pairs with successful communication records are determined as candidate graph edges.

[0082] Read the response delay and bit error rate corresponding to each candidate graph edge. Calculate the average response delay, average bit error rate, and the change in response delay between adjacent communication cycles according to the communication cycle order. The statistical result of the change in response delay between adjacent communication cycles is determined as the delay change. Specifically, the calculation of the average response delay, average bit error rate, and change in response delay between adjacent communication cycles according to the communication cycle order is as follows:

[0083] The sorted response delay sequence is traversed sequentially, and the response delay of each communication cycle is accumulated to obtain the cumulative delay value. At the same time, the communication cycle count is recorded. After the traversal is completed, the cumulative delay value is divided by the communication cycle count to obtain the average response delay of the candidate graph edge.

[0084] The sorted bit error rate sequence is traversed sequentially, and the bit error rate of each communication cycle is added together to obtain the cumulative bit error rate value. The cumulative bit error rate value is divided by the number of communication cycles to obtain the average bit error rate of the candidate graph edge.

[0085] Calculate the absolute value of the difference between adjacent response delays in the order of communication cycles, sum all the absolute differences to obtain the cumulative difference value, record the adjacent difference count, divide the cumulative difference value by the adjacent difference count to obtain the change amplitude of response delay between adjacent communication cycles, and use the change amplitude as the delay change amount of the candidate graph edge;

[0086] The registered master node, matching node, and end node are used as graph nodes, and candidate graph edges are used as the connection relationships between graph nodes. The corresponding communication success rate, average response time, bit error rate, and delay variation are written into the candidate graph edges to generate the primary and matching carrier communication graph structure. The specific process for generating the primary and matching carrier communication graph structure is as follows:

[0087] Each registered node is assigned a unique integer index in sequence, generating a node index lookup table. Five two-dimensional arrays with equal number of rows and columns and initial values ​​of zero are created. The first two-dimensional array serves as an adjacency matrix to mark node connectivity, while the other four two-dimensional arrays serve as the communication success rate matrix, average response delay matrix, average bit error rate matrix, and delay variation matrix, respectively. Next, the candidate graph edges are traversed. After reading a candidate graph edge, a 1 is written to the corresponding row and column position in the adjacency matrix, indicating that the node pair has a connection. At the same time, the communication success rate, average response delay, average bit error rate, and delay variation of the graph edge are written to the same row and column positions in the communication success rate matrix, average response delay matrix, average bit error rate matrix, and delay variation matrix, respectively. After traversing all candidate graph edges, the node index lookup table is combined with the five two-dimensional arrays to generate the primary and secondary carrier communication graph structure.

[0088] In this embodiment, constructing a carrier causal precedence body based on all node causal precedence pairs includes:

[0089] Using the communication cycle as the time step, the communication success marker, communication failure marker, upload timestamp, response delay, bit error rate, retransmission count, and noise intensity of each node are constructed as node feature vectors, which are then stacked in the order of graph nodes to generate the spatiotemporal input tensor of the improved DCRNN network.

[0090] The improvement to the construction of the DCRNN network is specifically as follows:

[0091] An input layer is set up, with input data consisting of a three-dimensional spatiotemporal tensor stacked in the order of communication cycles. The first dimension of the tensor is the communication cycle index, the second dimension is the node index, and the third dimension is a seven-dimensional node feature vector, including communication success marker, communication failure marker, upload timestamp, response latency, bit error rate, retransmission count, and noise intensity. Then, an adaptive multi-channel adjacency mapping layer is added after the input layer. Based on the graph edge communication success rate and bit error rate, a backbone diffusion adjacency channel and a noise suppression adjacency channel are generated, respectively. During the forward computation, the two channels are weighted and merged using learnable fusion weights to obtain a diffusion adjacency matrix adapted to the current carrier environment. The spatiotemporal tensor and the merged diffusion adjacency matrix are fed into a diffusion convolution module, which uses a bidirectional random walk diffusion method to spatially propagate the node features. Residual connections are then superimposed after the convolution output. Subsequently, the diffusing convolution output is fed into a gated recurrent unit (ROU), where a causal attention gate is embedded. This causal attention gate employs a multi-head attention mechanism, assigning sequentially ordered causal weights to historical hidden states and performing a weighted update of the hidden state at the current time step. A temporal drift normalization layer is added at the output of the gated RNU. This layer uses the average upload time of the communication cycle as a benchmark to scale and center the residual of the node upload time, reducing the impact of sudden delays on the model. Finally, a dual-channel temporal readout layer is set up. The first channel outputs the node-ahead stability vector, and the second channel outputs the node-bearing stability vector. The two channels share weights and use a linear activation function. The output results are used to generate node causal-ahead pairs and further construct a carrier causal-ahead body.

[0092] An adaptive multi-channel adjacency mapping layer is added to the input of the original DCRNN. ​​Two channels, a backbone diffusion adjacency matrix and a noise suppression adjacency matrix, are dynamically generated based on the communication success rate and bit error rate of the graph edges. During each forward computation, the two channels are weighted and merged using learnable fusion coefficients to obtain a diffusion adjacency matrix adapted to the current carrier environment. Specifically, the two channels—the backbone diffusion adjacency matrix and the noise suppression adjacency matrix—are dynamically generated based on the communication success rate and bit error rate of the graph edges.

[0093] When generating the backbone diffusion adjacency matrix and the noise suppression adjacency matrix, all graph edges in the current batch are used as objects. The communication success rate and bit error rate of each graph edge are calculated, and the maximum and minimum values ​​of the two indicators in the batch are recorded as the upper and lower limits of normalization.

[0094] For each graph edge, the following operations are performed: the communication success rate is linearly normalized to the maximum and minimum values ​​to obtain a reliability score between zero and one; the bit error rate is normalized in the same way to obtain a noise score between zero and one; the normalized reliability score is directly written into the corresponding row and column positions of the backbone diffusion adjacency matrix to represent the positive weight of the graph edge during spatial diffusion; the normalized noise score is then written into the corresponding row and column positions of the noise suppression adjacency matrix to represent the suppression strength of the graph edge in the noise suppression channel; if the graph structure is undirected, the same values ​​are synchronously written into the symmetrical row and column positions of the matrix to maintain symmetry; after all graph edges are processed, the dynamic generation of two channels is completed. In the forward calculation stage, the two channels are weighted and merged using learnable fusion coefficients to obtain a diffusion adjacency matrix adapted to the current carrier environment;

[0095] A diffusion convolution is performed on the spatiotemporal input tensor using a diffusion adjacency matrix. The output of the diffusion convolution is fed into a gated recurrent unit (ROU). A causal attention gate is embedded within the ROU. This causal attention gate employs a multi-head attention mechanism to assign different causal weights to historical hidden states, and then performs a weighted update of the hidden state at the current time step, outputting a node temporal diffusion state sequence. Specifically, the causal attention gate uses a multi-head attention mechanism to assign different causal weights to historical hidden states.

[0096] When setting a causal attention gate inside the gated recurrent unit, the hidden state of the current time step is copied as a query vector, and the set of hidden states of the same node in past communication cycles is copied as key vectors and value vectors. Independent linear mappings are performed on the query vector, key vector, and value vector, and they are divided into several equal-length sub-vectors according to the mapping method to form a multi-head structure. For each sub-vector head, the similarity between the query sub-vector and the corresponding key sub-vector is calculated first, and then a strict lower triangular mask is applied to the similarity in chronological order to mask future information. Then, normalization is performed in the time dimension to obtain the causal weights of each historical time step. The causal weights of the sub-vector head are multiplied with the value sub-vectors of the same head and the weighted sum is calculated in the time dimension to obtain the causal weighted output of the sub-vector head. The causal weighted outputs of all sub-vector heads are concatenated and merged into a single vector through linear mapping. The single vector is multiplied and superimposed with the candidate hidden states of the gated recurrent unit element-wise to realize the causal weighted update of the current hidden state by different historical hidden states. The updated hidden state is used as the output of the node temporal diffusion state sequence at that time step.

[0097] The node temporal diffusion state sequence is input into the temporal drift normalization layer. The temporal drift normalization layer uses the average upload time of the communication cycle as a benchmark, and performs amplitude scaling and centering on the upload time residuals of each node to generate the drift-normalized hidden states of the nodes. Specifically, the generation of the drift-normalized hidden states of the nodes is as follows:

[0098] Extract the upload timestamp sequence within the current batch by grouping nodes, calculate the arithmetic mean of the node upload timestamp sequence and use it as the reference upload time for the node in this batch; sequentially traverse the upload timestamps of each communication cycle of the node, subtract the reference upload time from each upload timestamp to obtain the upload time residual, then calculate the maximum absolute value of all upload time residuals of the node, and use the maximum value to perform a division operation on all upload time residuals to achieve residual amplitude scaling, so that the range of the scaled residual values ​​is limited to between -1 and +1; then calculate the average value of the scaled residual sequence again, and use each scaled residual to subtract the average value to complete the centering process, to obtain the centered residual sequence; concatenate the centered residual sequence with the node temporal diffusion state sequence in the feature dimension, replace the original upload time feature component, and obtain the node hidden state after drift normalization;

[0099] After drift normalization, the hidden state of the nodes is input into a dual-channel timing readout layer. One channel outputs the node's ahead stability vector, and the other channel outputs the follow-up stability vector. The node causal-ahead pairs are generated by combining the ahead stability, follow-up stability, and node upload time sequence. All node causal-ahead pairs are then concatenated to form a carrier causal-ahead vector, where:

[0100] The one-channel output node-ahead stability vector is as follows:

[0101] In each communication cycle, the upload time residual in the hidden state of the node after drift normalization is read. The upload time residuals of all nodes in the same cycle are sorted, and the node with the smallest residual value is determined as the preceding node in the current cycle. The number of cycles in which each node is determined as a preceding node is accumulated, and the result is divided by the total number of statistical cycles to obtain the occurrence ratio of preceding nodes. The ratio is used as the preceding stability component of the node. The preceding stability components of each node are collected in the order of node index to form the preceding stability vector of the node.

[0102] The other channel output carries the stability vector, specifically:

[0103] In each communication cycle, the forwarding success flag and receiving success flag of the node are read according to the carrier communication graph structure. The sum of the forwarding success flag and receiving success flag of all nodes in the same cycle is calculated. The node with a sum greater than zero and successfully uploaded is determined as the accepting node in the current cycle. The number of cycles in which each node is determined as the accepting node is accumulated and divided by the total number of statistical cycles to obtain the proportion of accepting nodes. The proportion is used as the node accepting stability component. The accepting stability components of each node are collected in the order of node index to form the node accepting stability vector.

[0104] The carrier causal antecedent is formed as follows:

[0105] Within each communication cycle, the upload time sequence of nodes is read. The node with the earliest upload time and corresponding preceding stability higher than a set threshold is recorded as the cycle preceding node, and the node with a forwarding relationship with the preceding node and corresponding receiving stability higher than a set threshold is recorded as the cycle receiving node. Node pairs are recorded sequentially along the forwarding path from the preceding node to the receiving node. The preceding node is used as the causal starting point and the receiving node as the causal ending point. A preceding-receiving tag is written for each node pair and the current communication cycle number is appended. All node pairs are traversed in ascending order of communication cycle number. Node pairs with the same starting and ending points in adjacent communication cycles are connected into the same causal chain segment. Causal chain segments with a common starting or ending point and continuous communication cycles are merged until they can no longer be merged, generating multiple node causal chains. A unified chain identifier is assigned to each node causal chain. The chain structure is described by the chain identifier, the node order within the chain, and the cycle sequence within the chain. The entire set of causal chains is stored as a carrier causal preceding entity.

[0106] A cross-entropy loss function is constructed using the true preceding order label of the communication cycle and the outputs of node preceding stability and succession stability. An adjacent time step smoothing regularization term is added to jointly optimize the improved DCRNN network. Batch size, learning rate, and learning rate decay strategies are set, and early stopping is triggered when the validation set loss does not show a significant decrease.

[0107] The cross-entropy loss function is constructed as follows:

[0108] Within each training batch, a one-hot precedence label vector is generated for each node based on the true precedence order label of the communication cycle, marking whether the node is actually preceded as one and zero, respectively. Normalization is performed on the node precedence stability vector output by the improved DCRNN network, normalizing the precedence stability of each node to a probability distribution. Then, the negative log-likelihood between the one-hot precedence label and the normalized precedence probability is calculated for each node, and the precedence branch loss is obtained by summing the negative log-likelihoods of all nodes. The acceptance branch loss is calculated based on the true acceptance label vector and the normalized acceptance probability distribution. Finally, the precedence branch loss and the acceptance branch loss are linearly superimposed according to set weights to obtain the joint cross-entropy loss.

[0109] The training batch size is set to 64, the initial learning rate is set to 0.001, and the learning rate is decayed by a factor of 0.9 every ten iterations. Early termination is triggered when the validation set loss does not decrease significantly for five consecutive iterations.

[0110] In this embodiment, the construction of the master-supplier micro-network structure based on the intersection relationship between each causal link includes:

[0111] Read all node causal precedence pairs in the carrier causal precedence body, and extract the preceding node, succeeding node, causal propagation direction and communication cycle marker from each node causal precedence pair;

[0112] The causal precedence pairs of nodes are sorted according to the causal propagation direction and communication cycle marker. Causal precedence pairs where the preceding node of one causal precedence pair is the same as the preceding node of the following node's causal precedence pair and their communication cycles are adjacent are sequentially connected to generate a causal succession chain. Specifically, generating the causal succession chain involves:

[0113] Read all node causal precedence pairs, and extract the preceding node identifier, receiving node identifier, causal propagation direction, and communication cycle marker for each node causal precedence pair. Arrange all node causal precedence pairs in ascending order according to the communication cycle marker, and continue arranging them according to the link order corresponding to the causal propagation direction within the same communication cycle. Starting from the node causal precedence pair of the earliest communication cycle, traverse each pair sequentially, comparing the receiving node identifier of the current node causal precedence pair with the preceding node identifiers of each node causal precedence pair in the next communication cycle. When the node identifiers are the same and the next communication cycle is adjacent to the current communication cycle, connect the two node causal precedence pairs sequentially. If the same receiving node corresponds to multiple connectable node causal precedence pairs, select the node causal precedence pair with the smallest response delay and the lowest bit error rate as the connection object. Continue to perform the above connection until there are no node causal precedence pairs that meet the connection conditions. Finally, combine the sequentially connected node causal precedence pairs into a causal receiving chain, and write the chain identifier, the order of nodes in the chain, the order of causal precedence pairs in the chain, and the corresponding communication cycle sequence.

[0114] Read the order of nodes within each causal succession chain, mark the node at the head of the chain as the starting succession node, mark the node between the head and tail of the chain that connects two causal pairs as the continuing succession node, and mark the node at the tail of the chain as the ending succession node.

[0115] Compare the starting, continuing, and ending nodes in different causal succession chains. Establish inter-chain sharing relationships for causal succession chains with the same nodes. Establish inter-chain continuity relationships for causal succession chains where the ending node of one causal succession chain matches the starting node of another. Generate a set of intersecting succession chains based on the inter-chain sharing and inter-chain continuity relationships. Specifically, the generation of the set of intersecting succession chains based on the inter-chain sharing and inter-chain continuity relationships is as follows:

[0116] Read the chain identifier, starting node, continuing node, and ending node of all causal connection chains to create a chain node list for each causal connection chain; compare the chain node lists of different causal connection chains one by one; when two causal connection chains have the same node, write an inter-chain shared marker between the two corresponding causal connection chains and record the shared node identifier; when the ending node of the first causal connection chain is the same as the starting node of the second causal connection chain, write an inter-chain continuation marker between the two corresponding causal connection chains and record the continuation node identifier; add causal connection chains with inter-chain shared markers or inter-chain continuation markers to the same temporary set, and continue to read other causal connection chains associated with each causal connection chain in the temporary set until no new associated causal connection chains exist; remove duplicate causal connection chains in the temporary set, write the set identifier, set-internal chain identifier, shared node identifier, continuation node identifier, and inter-chain relationship type to generate a converged connection chain set;

[0117] After deduplicating all nodes in the same causal link set, a microgroup node set is generated. Nodes belonging to two or more causal links are identified as microgroup receiving nodes. The order of nodes within each causal link is used as the transmission order within the microgroup. The shared relationship between links and the continuity relationship between links are used as the inter-microgroup receiving relationship. A master-slave microgroup network structure is generated, which includes microgroup identifier, microgroup node set, microgroup receiving nodes, transmission order within the microgroup, and inter-microgroup receiving relationship. The generation of the master-slave microgroup network structure is as follows:

[0118] Read all causal connections in each convergence connection chain set, and extract the node identifiers, intra-chain node order, inter-chain shared relationships, and inter-chain succession relationships contained in each causal connection chain. Delete node identifiers that appear repeatedly in the same convergence connection chain set, keeping only one node identifier, and generate the corresponding microgroup node set. Count the number of times each node in the microgroup node set appears in different causal connection chains, mark nodes that appear at least twice as microgroup connection nodes, and mark nodes that appear only once as microgroup ordinary nodes. Then, generate the intra-microgroup transmission order according to the intra-chain node order of each causal connection chain, and write the shared nodes corresponding to the inter-chain shared relationships and the succession nodes corresponding to the inter-chain succession relationships into the inter-microgroup connection relationships. Assign a microgroup identifier to each microgroup node set, and combine the microgroup identifier, microgroup node set, microgroup connection node, microgroup ordinary node, intra-microgroup transmission order, and inter-microgroup connection relationships to generate the master-supplier microgroup network structure.

[0119] In this embodiment, the step of generating a causal folding piece based on the causal succession chain corresponding to the data to be closed, and generating a cross-period causal closure record by associating data in the same chain, includes:

[0120] Read the microgroup identifier, causal succession chain, node order within the chain, and communication cycle marker in the main and auxiliary microgroup network structure. Select a continuous communication cycle according to the preset window length. Write the node upload data, node receive data, transmission status marker, and frame check marker within the continuous communication cycle into the time compression window, where the preset window length is 5 continuous communication cycles.

[0121] The construction of the time compression window is as follows:

[0122] The current communication cycle number is read, and the current communication cycle is used as the window termination cycle. Four consecutive communication cycles are read forward to form a window cycle set containing five consecutive communication cycles. The causal link chains in the master-slave micro-network structure are read one by one according to the micro-group identifier, and the node order and corresponding communication cycle marker of each causal link chain are extracted. Then, within the window cycle set, the node upload data, node receive data, transmission status marker, and frame check marker are read sequentially according to the micro-group identifier, causal link identifier, node order, and communication cycle number. For each read data record, the window identifier, micro-group identifier, causal link identifier, node position, communication cycle number, data frame identifier, upload data field, receive data field, transmission status field, and frame check field are written. Finally, all data records are arranged from earliest to latest according to the communication cycle number, and arranged according to the node order within the same communication cycle to generate a time compression window.

[0123] Within the time compression window, read the transmission status flag and frame verification flag of each node's data, mark data with upload failure, reception timeout, frame verification failure or chain interruption as data to be closed, and write node identifier, data frame identifier, communication cycle identifier and the causal chain identifier to be closed for the data to be closed.

[0124] Based on the causal chain to which the data to be closed belongs, extract the node position, preceding node, current node, and ending node corresponding to the data within the chain. Combine the data frame identifier, communication cycle identifier, node position within the chain, and node information of the data to be closed to generate a causal fold piece. Specifically, generating the causal fold piece involves:

[0125] Read the data frame identifier, node identifier, communication cycle identifier, and causal connection chain identifier of the data to be closed; then find the node position within the corresponding causal connection chain where the node identifier is located, and write the node position into the chain position field; next, read the node adjacent to the node position before the current node position as the preceding node, read the node corresponding to the current node position as the current node, and read the tail node of the causal connection chain as the terminating node; if the node corresponding to the data to be closed is at the head of the chain, write a null value flag into the preceding node field; then extract the data length, frame check flag, transmission status flag, and failure reason flag of the data to be closed; combine the data frame identifier, communication cycle identifier, causal connection chain identifier, chain position field, preceding node, current node, terminating node, data length, frame check flag, transmission status flag, and failure reason flag in a fixed field order to generate a causal folding piece;

[0126] Read the causal relay chain identifier corresponding to the causal folding piece, extract different communication cycle data belonging to the same causal relay chain within the time compression window, and associate the different communication cycle data according to the node order within the chain, the communication cycle order, and the data frame identifier to generate a same-chain data association relationship. Specifically, generating the same-chain data association relationship involves:

[0127] Read the causal succession chain identifier, chain position field, and data frame identifier from the causal folding piece; filter all data records that match the causal succession chain identifier within the time compression window, and remove data records with inconsistent causal succession chain identifiers; arrange the filtered data records from earliest to latest according to the communication cycle number, and arrange them from front to back according to the chain node position within the same communication cycle; then, using the data frame identifier of the causal folding piece as a reference, find data records with the same data frame identifier, and write the found data records into the same frame association set; simultaneously, find the preceding and following node data records that are adjacent to the chain position of the causal folding piece, and write the found data records into the adjacent association set; combine the causal folding piece, the same frame association set, the adjacent association set, the communication cycle arrangement order, and the chain node arrangement order to generate the same chain data association relationship;

[0128] The causal folding fragment, the data association relationship of the same chain, the corresponding causal successor chain identifier, the data status to be closed, and the closure result marker are combined to generate a cross-period causal closure record. This cross-period causal closure record is then written into the corresponding master-slave micro-network structure. Specifically, the generation of the cross-period causal closure record involves:

[0129] Read the data frame identifier, communication cycle identifier, causal succession chain identifier, chain position field, preceding succession node, current succession node, and ending succession node from the causal folding piece; then read the same-frame association set and adjacent association set in the same-chain data association relationship, check whether there is a data record in the same-frame association set that has passed frame verification and has a successful transmission status, and check whether there is a successful preceding node record or a successful following node record in the adjacent association set; when there is a data record in the same-frame association set that has passed frame verification and a successful adjacent node record in the adjacent association set, mark the closure result as closure successful; when the same-frame association... If the set does not contain a data record that has passed frame verification, or if the adjacent association set does not contain a record of a successfully adjacent node, the closure result mark is written as closure failure. Then, the data state to be closed, the closure result mark, the data frame identifier used for closure, the communication cycle identifier participating in the closure, the node identifier participating in the closure, and the closure time mark are combined in a fixed field order to generate a cross-cycle causal closure record. Finally, according to the microgroup identifier and causal succession chain identifier in the cross-cycle causal closure record, the microgroup record and causal succession chain record in the corresponding master-distributor microgroup network structure are located, and the cross-cycle causal closure record is written into the closure record field of the corresponding record.

[0130] In this embodiment, updating the master-distributor micro-network structure and performing carrier communication transmission between the master node, the distribution node, and the end node includes:

[0131] Read the number of successful communication, the number of failed communication, the number of cross-cycle causal closures, the number of connection failures, and the number of causal connection chain breaks for each node within a preset statistical window, and use the counting results as node status statistical parameters.

[0132] Based on the node state statistical parameters, the stable transmission probability, short-term instability probability, long-term attenuation probability, recovery probability, and connection failure probability corresponding to the node are calculated, and a node probability state vector is generated by combining these probabilities and written into the node probability state vector index table. Specifically, the calculation of the stable transmission probability, short-term instability probability, long-term attenuation probability, recovery probability, and connection failure probability corresponding to the node is as follows:

[0133] The system reads the target node's successful upload count, failed upload count, closure recovery count, successful connection count, failed connection count, consecutive failure cycles, and consecutive offline cycles within the target statistics window. It adds the successful upload count to the failed upload count to obtain the total upload count; divides the successful upload count by the total upload count to obtain the stable transmission probability; divides the failed upload count in the last two communication cycles by the total upload count in the last two communication cycles to obtain the short-term instability probability; adds the consecutive failure cycles and consecutive offline cycles and divides by the total number of communication cycles in the target statistics window to obtain the long-term decay probability; divides the closure recovery count by the sum of the failed upload count and the closure recovery count to obtain the recovery probability; divides the failed connection count by the sum of the successful connection count and the failed connection count to obtain the connection failure probability; when any denominator is zero, the corresponding probability is written to zero; and generates a node probability state vector by combining the stable transmission probability, short-term instability probability, long-term decay probability, recovery probability, and connection failure probability in that order.

[0134] The main and secondary micro-group network structure is updated according to the node probability state vector index table. Nodes with stable transmission probability higher than the first threshold are retained in the original micro-group. Nodes with short-term instability probability between the first and second thresholds are marked as temporary fluctuation nodes. Nodes with long-term attenuation probability higher than the second threshold or failure to accept the connection probability higher than the third threshold are removed from the main transmission path of the original micro-group and reassigned as receiving nodes. The first threshold is set to 0.75, the second threshold is set to 0.60, and the third threshold is set to 0.50.

[0135] A communication scheduling table is generated based on the updated master-distributor micro-group network structure. The communication scheduling table includes micro-group identifier, core uploading node, receiving node, ordinary node, intra-group transmission order, uploading time slot allocation, and retransmission threshold. The master node organizes the master node, distribution side node, and end node to perform carrier communication transmission according to the communication scheduling table.

[0136] refer to Figure 2 A carrier communication master-distribution micro-network transmission system, comprising:

[0137] The timing status generation module is used to collect carrier communication status data of the master node, the matching node, and the end node, and generate node timing status records.

[0138] The communication graph construction module is used to construct the primary and secondary carrier communication graph structure based on the node timing status records.

[0139] The causal precedence construction module is used to input the primary and secondary carrier communication graph structure and node timing state records into the improved DCRNN network to generate node causal precedence pairs and construct carrier causal precedence entities;

[0140] The micro-network construction module is used to generate causal succession chains based on carrier causal precursors and to construct a master-supplier micro-network structure.

[0141] The causal closure generation module is used to construct time compression windows and generate causal folded pieces and cross-period causal closure records;

[0142] The probabilistic state update module is used to generate a probabilistic state vector based on the node time-series state record, carrier causal precursor, causal succession chain, and cross-cycle causal closure record, update the master-supplier micro-network structure, and execute carrier communication transmission.

[0143] Example 1: In a low-voltage distribution transformer area carrier communication acquisition scenario, the system connects 1 master node, 5 distribution-side nodes, and 84 end nodes, distributed across 3 phase lines and 7 branches. The system continuously acquires data for 80 communication cycles. Within each communication cycle, the master node receives communication success markers, communication failure markers, upload timestamps, response delays, bit error rates, retransmission counts, and noise levels from each node. Raw statistics show that the 84 end nodes generated a total of 6720 communication records, including 5846 successful records and 874 failed records. The average response delay was 286 milliseconds, the average bit error rate was 0.043, and the cumulative retransmission count was 912. Significant short-term fluctuations were observed in branches 4 and 6. Specifically, the average noise level of the end nodes in branch 4 increased from 0.31 to 0.68, and the response delay of some nodes suddenly increased from 180 milliseconds to 520 milliseconds within adjacent communication cycles.

[0144] The system first generates node timing status records. Taking the end node N028 as an example, its successful communication within five consecutive communication cycles is marked as 1, 1, 0, 0, 1; upload timestamp offsets are 132 ms, 139 ms, no upload, no upload, 151 ms; response latency is 174 ms, 181 ms, timeout, timeout, 196 ms; bit error rate is 0.018, 0.021, 0.096, 0.088, 0.026; and retransmission count is 0, 0, 2, 2, 0. Traditional methods would trigger rerouting after two consecutive failures at this node, while this method records it as a short-term instability sample and retains its cross-cycle continuity.

[0145] When constructing the primary and secondary carrier communication graph structure, the system registers 90 communication nodes as graph nodes and establishes candidate graph edges for node pairs with successful communication records. Taking the secondary node P02 and the terminal node N028 as an example, they communicated successfully 72 times in 80 communication cycles, with a communication success rate of 90.00%, an average response delay of 205 milliseconds, an average bit error rate of 0.027, and a response delay variation of 38 milliseconds between adjacent communication cycles. The secondary node P03 and N028 communicated successfully 43 times, with a communication success rate of 53.75%, an average response delay of 391 milliseconds, an average bit error rate of 0.071, and a delay variation of 104 milliseconds. The system writes the connection relationship from P02 to N028 into the adjacency matrix and writes the corresponding communication success rate, average response delay, bit error rate, and delay variation into the graph edge attribute array.

[0146] When improving the training of the DCRNN network, the system uses the first 60 communication cycles as training samples and the last 20 communication cycles as validation samples. Each training sample contains node features for 5 consecutive communication cycles. The node features consist of communication success markers, communication failure markers, upload timestamps, response latency, bit error rate, retransmission count, and noise intensity. The input tensor size is 5×90×7. In the training samples, N012 had 5 successful transmissions in 5 cycles, an average response latency of 142 milliseconds, an average bit error rate of 0.014, and an average noise intensity of 0.24, and was marked as a pre-stable node; N028 had 3 successful transmissions in 5 cycles, an average response latency of 276 milliseconds, an average bit error rate of 0.050, and an average noise intensity of 0.61, and was marked as a short-term fluctuating node; P02 had 41 successful receptions in 5 cycles and 2 failed receptions, and was marked as a reception-stable node; N061 had 1 successful transmission in 5 cycles, 6 retransmissions, and an average noise intensity of 0.74, and was marked as a long-term decaying node.

[0147] The improved DCRNN network first reads the edge attributes of the communication graph, generates a backbone diffusion adjacency channel based on the communication success rate, and generates a noise suppression adjacency channel based on the bit error rate. The two channels are weighted and merged to form a diffusion adjacency matrix. After the spatiotemporal input tensor undergoes diffusion convolution, the hidden states of adjacent nodes P02, N012, and N028 are propagated to the same local diffusion region. The average diffusion response of N012 is 0.82, and the diffusion response of N028 recovers from 0.39 to 0.63 during the failure cycle. After entering the gated recurrent unit, the causal attention gate weights the historical hidden states. The attention weights of N012 in the past 5 cycles are 0.13, 0.17, 0.20, 0.23, and 0.27, respectively, indicating that its ahead state is continuously enhanced. The weight of N028 decreases to 0.08 during the failure cycle, but rises back to 0.19 during the recovery cycle, and is not judged as a permanent anomaly. After time-shift normalization, the upload time residual of N028 was compressed from a maximum offset of 344 milliseconds to 0.72, preventing sudden delays from directly disrupting the causal relationship.

[0148] The dual-channel timing readout layer outputs node-ahead stability and continuity stability. The preceding stability of N012 is 0.91, and the continuity stability is 0.44; the preceding stability of P02 is 0.62, and the continuity stability is 0.93; the preceding stability of N028 is 0.58, and the continuity stability is 0.37; the preceding stability of N061 is 0.21, and the continuity stability is 0.18. The system generates node causal preceding pairs based on upload time order and continuity relationships. For example, N012 points to P02, N028 points to P02, and P02 points to the master node. Successively continuing node causal preceding pairs are merged into a carrier causal preceding pair. This ultimately results in 29 stable causal chains and 13 fluctuating causal chains.

[0149] During the construction of the master-distributor micro-group network structure, the system connects causal pairs of nodes according to the causal propagation direction, generating causal succession chains. Taking chain C07 as an example, the node order within the chain is N012, N028, P02, and the master node. N012 is marked as the starting succession node, N028 and P02 are marked as continuing succession nodes, and the master node is marked as the ending succession node. After comparing all causal succession chains, the system found that P02 appears in 6 causal succession chains and P04 appears in 4 causal succession chains. Both are identified as microgroup succession nodes. Finally, the system generates 5 master-distributor microgroups, with an average of 16.8 nodes per microgroup. The average link success rate within the microgroup is improved from 86.4% in traditional fixed groups to 94.1%.

[0150] In the time compression window processing, the window length is set to 5 consecutive communication cycles. During the 37th communication cycle, data frame F028-37 of N028 is not fully transmitted. The system reads its causal successor chain C07, marks the data as pending closure, and generates a causal fold. The causal fold field includes the data frame identifier F028-37, communication cycle identifier 37, chain position 2, preceding successor node N012, current successor node N028, end successor node master node, and the failure reason marked as frame verification failure. The system searches for data in the same chain within the same time compression window and finds that frames F028-36 and F028-39 have passed verification. Furthermore, P02 retains the received digest of N028 in the 37th cycle. The system associates F028-37 with data in the adjacent cycle's same chain, generating a cross-cycle causal closure record and marking the closure result as successful. No immediate retransmission is triggered during this process, saving two carrier transmissions.

[0151] During the probability state vector calculation, N028 successfully uploaded 16 times, failed to upload 4 times, recovered from closure 3 times, failed to accept 1 time, and had 0 consecutive offline cycles within the statistics window. The system generated a stable transmission probability of 0.80, a short-term instability probability of 0.40, a long-term attenuation probability of 0.00, a recovery probability of 0.43, and a failure to accept 0.11. It was marked as a temporary fluctuation node and retained in the original microgroup. N061 successfully uploaded 7 times, failed to upload 13 times, recovered from closure 1 time, failed to accept 7 times, had 4 consecutive offline cycles, a long-term attenuation probability of 0.20, and a failure to accept 0.54. It was removed from the main transmission path of the original microgroup and reassigned to P05 for acceptance.

[0152] In the comparative experiment, the traditional fixed microgroup method uploaded 6720 frames, with 5846 frames successfully transmitted (86.99% success rate), an average response latency of 286 milliseconds, 912 retransmissions, 11 re-networking attempts, and 14 falsely identified offline nodes. The traditional link quality dynamic routing method successfully transmitted 6127 frames (91.18% success rate), with an average response latency of 251 milliseconds, 684 retransmissions, 7 re-networking attempts, and 8 falsely identified offline nodes. The method of this invention successfully transmitted 6548 frames (97.44% success rate), with an average response latency of 198 milliseconds, 263 retransmissions, 2 re-networking attempts, 2 falsely identified offline nodes, and recovered 218 data frames across cycle closures.

[0153] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A carrier communication master-distribution micro-network transmission method, characterized in that, include: Collect carrier communication status data of master nodes, distribution side nodes and end nodes in low-voltage distribution radio area in multiple consecutive communication cycles, and generate node timing status records; Construct a primary and secondary carrier communication graph structure based on the node timing status records. The primary node, secondary node, and terminal node are used as graph nodes, and the node pairs with successful communication records are used as graph edges. Write the communication success rate, average response delay, bit error rate, and delay variation for each graph edge. The main and secondary carrier communication graph structure and the corresponding node timing state records are input into the improved DCRNN network. The timing diffusion trajectory corresponding to each node is extracted. Based on the sequential and continuous relationships between the timing diffusion trajectories of different nodes, node causal precedence pairs are generated. Based on all node causal precedence pairs, a carrier causal precedence body is constructed. Based on the carrier causal precedence, a causal succession chain is generated according to the transmission order of the node causal precedence pairs. The starting succession node, the continuing succession node and the ending succession node in each causal succession chain are identified. The main and distribution micro-network structure is constructed according to the intersection relationship between each causal succession chain. Construct a time compression window covering multiple communication cycles, mark incomplete transmission data as data to be closed, generate causal folding pieces based on the causal succession chain corresponding to the data to be closed, and generate cross-cycle causal closure records by associating data in the same chain. Based on the node timing state record, carrier causal precursor, causal succession chain, and cross-cycle causal closure record, a probability state vector is generated for each node, the master-supplier micro-network structure is updated, and carrier communication transmission between the master node, the supplier node, and the end node is executed.

2. The carrier communication master-distribution micro-network transmission method according to claim 1, characterized in that, The carrier communication status data includes node identifier, phase line identifier, branch number, communication success marker, communication failure marker, upload timestamp, response delay, bit error rate, retransmission count, and noise intensity.

3. The carrier communication master-distribution micro-network transmission method according to claim 1, characterized in that, The generated node timing state record includes: Read the node identifier, phase line identifier, branch number, communication success mark, communication failure mark, upload timestamp, response delay, bit error rate, retransmission count and noise intensity of each node within a continuous communication cycle; The communication status data corresponding to the same node are collected according to the node identifier; Based on the upload timestamp, the collected communication status data is sorted chronologically to generate a communication status sequence for the corresponding node; The communication success marker, communication failure marker, response delay, bit error rate, retransmission count, and noise intensity in the communication state sequence are associated in the order of the communication cycle to generate the time sequence state record of the corresponding node.

4. The carrier communication master-distribution micro-network transmission method according to claim 1, characterized in that, The construction of the primary and secondary carrier communication graph structure based on the node timing status records includes: Read the node identifier, node type, phase line identifier, branch number, communication cycle number, communication success mark, communication failure mark, upload timestamp, response delay and bit error rate from the node timing status record, and register the master node, the auxiliary node and the end node as nodes to be built in the map respectively; The communication records of each node to be mapped are aligned according to the communication cycle number. Node pairs that have carrier communication interaction within the same communication cycle are extracted. The communication success mark, communication failure mark, response delay and bit error rate corresponding to each node pair are collected into node pair communication records. Based on the communication records of each node pair, the total number of communications and the number of successful communications are counted. The proportion of successful communications in the total number of communications is determined as the communication success rate, and the node pairs with successful communication records are determined as candidate graph edges. Read the response delay and bit error rate corresponding to each candidate graph edge, calculate the average response delay, average bit error rate and response delay change amplitude of adjacent communication cycles according to the communication cycle order, and determine the statistical result of the response delay change amplitude of adjacent communication cycles as the delay change amount; The registered master node, partner node, and end node are used as graph nodes, and the candidate graph edges are used as the connection relationships between the graph nodes. The corresponding communication success rate, average response delay, bit error rate, and delay change are written into the candidate graph edges to generate the master-partition carrier communication graph structure.

5. The carrier communication master-distribution micro-network transmission method according to claim 1, characterized in that, The construction of the carrier causal precedence body based on all node causal precedence pairs includes: Using the communication cycle as the time step, the communication success marker, communication failure marker, upload timestamp, response delay, bit error rate, retransmission count, and noise intensity of each node are constructed as node feature vectors, which are then stacked in the order of graph nodes to generate the spatiotemporal input tensor of the improved DCRNN network. An adaptive multi-channel adjacency mapping layer is added to the input of the original DCRNN. ​​Based on the communication success rate and bit error rate of the graph edges, two channels, the backbone diffusion adjacency matrix and the noise suppression adjacency matrix, are dynamically generated. During each forward calculation, the two channels are weighted and merged using learnable fusion coefficients to obtain a diffusion adjacency matrix adapted to the current carrier environment. The spatiotemporal input tensor is subjected to diffusion convolution using a diffusion adjacency matrix. The output of the diffusion convolution is fed into a gated recurrent unit. A causal attention gate is embedded inside the gated recurrent unit. The causal attention gate uses a multi-head attention mechanism to assign different causal weights to the historical hidden states and performs weighted updates on the hidden states at the current time step, outputting a node temporal diffusion state sequence. The node temporal diffusion state sequence is input into the temporal drift normalization layer. The temporal drift normalization layer uses the average upload time of the communication cycle as a benchmark to perform amplitude scaling and centering on the upload time residual of each node, generating the node hidden state after drift normalization. After the drift is normalized, the hidden state of the node is input into the dual-channel timing readout layer. One channel outputs the node's leading stability vector, and the other channel outputs the receiving stability vector. The node's causal leading pairs are generated by combining the leading stability, receiving stability and the order of node upload time. All node causal leading pairs are then connected and combined to form a carrier causal leading body. A cross-entropy loss function is constructed using the true preceding order label of the communication cycle and the outputs of node preceding stability and succession stability. An adjacent time step smoothing regularization term is added to jointly optimize the improved DCRNN network. Batch size, learning rate and learning rate decay strategy are set, and early stopping is triggered when the loss on the validation set does not decrease significantly.

6. The carrier communication master-distribution micro-network transmission method according to claim 1, characterized in that, The construction of the master-supplier micro-network structure based on the intersection relationships between each causal link includes: Read all node causal precedence pairs in the carrier causal precedence body, and extract the preceding node, succeeding node, causal propagation direction and communication cycle marker from each node causal precedence pair; The causal precedence pairs of nodes are sorted according to the causal transmission direction and communication cycle marker. The causal precedence pairs of nodes whose successor node is the same as the predecessor node of the next node's causal precedence pair and whose communication cycles are adjacent are sequentially connected to generate a causal succession chain. Read the order of nodes within each causal succession chain, mark the node at the head of the chain as the starting succession node, mark the node between the head and tail of the chain that connects two causal pairs as the continuing succession node, and mark the node at the tail of the chain as the ending succession node. Compare the starting node, continuing node, and ending node in different causal succession chains. Establish inter-chain sharing relationships for causal succession chains with the same node. Establish inter-chain continuity relationships for causal succession chains where the ending node of one causal succession chain is the same as the starting node of another causal succession chain. Generate a set of intersecting succession chains based on the inter-chain sharing relationship and the inter-chain continuity relationship. After deduplicating all nodes in the same causal link set, a microgroup node set is generated. Nodes belonging to two or more causal links are identified as microgroup linking nodes. The order of nodes within each causal link is used as the transmission order within the microgroup. The shared relationship between links and the continuity relationship between links are used as the linking relationship between microgroups. A master-slave microgroup network structure is generated, which includes microgroup identifier, microgroup node set, microgroup linking nodes, transmission order within the microgroup, and linking relationship between microgroups.

7. The carrier communication master-distribution micro-network transmission method according to claim 1, characterized in that, The process of generating causal folds based on the causal chain corresponding to the data to be closed, and generating cross-period causal closure records by associating data from the same chain, includes: Read the microgroup identifier, causal succession chain, node order within the chain, and communication cycle marker in the main and auxiliary microgroup network structure. Select a continuous communication cycle according to the preset window length and write the node upload data, node receive data, transmission status marker, and frame check marker within the continuous communication cycle into the time compression window. Within the time compression window, read the transmission status flag and frame verification flag of each node's data, mark data with upload failure, reception timeout, frame verification failure or chain interruption as data to be closed, and write node identifier, data frame identifier, communication cycle identifier and the causal chain identifier to be closed for the data to be closed. Based on the causal chain to which the data to be closed belongs, extract the node position, previous successor node, current successor node and end successor node corresponding to the data to be closed in the chain, and combine the data frame identifier, communication cycle identifier, node position in the chain and successor node information of the data to be closed to generate a causal folding piece; Read the causal relay chain identifier corresponding to the causal folding piece, extract different communication cycle data belonging to the same causal relay chain within the time compression window, and associate the different communication cycle data according to the node order, communication cycle order and data frame identifier to generate the same chain data association relationship; The causal folding piece, the data association relationship of the same chain, the corresponding causal successor chain identifier, the data status to be closed and the closure result mark are combined to generate a cross-cycle causal closure record, and the cross-cycle causal closure record is written into the corresponding master-slave micro-network structure.

8. The carrier communication master-distribution micro-network transmission method according to claim 1, characterized in that, The update of the master-distributor micro-network structure, and the execution of carrier communication transmission between the master node, the distribution node, and the end node, includes: Read the number of successful communication, the number of failed communication, the number of cross-cycle causal closures, the number of connection failures, and the number of causal connection chain breaks for each node within a preset statistical window, and use the counting results as node status statistical parameters. Based on the node state statistical parameters, calculate the stable transmission probability, short-term instability probability, long-term attenuation probability, recovery probability, and connection failure probability corresponding to the node, combine them to generate a node probability state vector, and write it into the node probability state vector index table. The main and sub-microgroup network structure is updated according to the node probability state vector index table. Nodes with stable transmission probability higher than the first threshold are retained in the original microgroup. Nodes with short-term instability probability between the first and second thresholds are marked as temporary fluctuation nodes. Nodes with long-term attenuation probability higher than the second threshold or failure probability higher than the third threshold are removed from the main transmission path of the original microgroup and reassigned as receiving nodes. A communication scheduling table is generated based on the updated master-distributor micro-group network structure. The communication scheduling table includes micro-group identifier, core uploading node, receiving node, ordinary node, intra-group transmission order, uploading time slot allocation, and retransmission threshold. The master node organizes the master node, distribution side node, and end node to perform carrier communication transmission according to the communication scheduling table.

9. A carrier communication master-distributor micro-network transmission system, comprising the carrier communication master-distributor micro-network transmission method according to any one of claims 1 to 8, characterized in that, include: The timing status generation module is used to collect carrier communication status data of the master node, the matching node, and the end node, and generate node timing status records. The communication graph construction module is used to construct the primary and secondary carrier communication graph structure based on the node timing status records. The causal precedence construction module is used to input the primary and secondary carrier communication graph structure and node timing state records into the improved DCRNN network to generate node causal precedence pairs and construct carrier causal precedence entities; The micro-network construction module is used to generate causal succession chains based on carrier causal precursors and to construct a master-supplier micro-network structure. The causal closure generation module is used to construct time compression windows and generate causal folded pieces and cross-period causal closure records; The probabilistic state update module is used to generate a probabilistic state vector based on the node time-series state record, carrier causal precursor, causal succession chain, and cross-cycle causal closure record, update the master-supplier micro-network structure, and execute carrier communication transmission.

Images