Method for semantic communication of ipv6 terminal behavior in a substation

Abstract

This application relates to the field of power system communication network technology, specifically disclosing a semantic communication method for IPv6 terminals in substations. It introduces graph neural network embedding dimensionality reduction technology to construct a semantic ontology state graph from the terminal's multi-dimensional operating parameters. This graph is then compressed into a fixed-length semantic state vector with a fixed bit width through graph convolution aggregation and hash discretization. Furthermore, this vector is injected bit-by-bit into the reserved alignment bits of the IPv6 hop-by-hop option extension header to achieve implicit in-path communication. This allows the white-box switch's programmable pipeline to perform line-speed extraction of the semantic vector based on a fixed offset. Nanosecond-level parallel matching and verification are then performed using a three-state content-addressable memory and a pre-compiled rule flow table, directly determining the legality of terminal behavior at the hardware layer. The matching result drives deterministic network slice scheduling, achieving continuous protection of terminal behavior semantics and deterministic transmission guarantee for control services without introducing additional upload latency.

Description

Technical Field

[0001] This application relates to the field of power system communication network technology, and more specifically, to a semantic communication method for IPv6 terminal behavior in substations. Background Technology

[0002] With the accelerated transformation of smart substations, applications such as intelligent inspection, digital operations, and high-precision status perception are emerging continuously. Multiple services, including control and monitoring, coexist, and the number of terminal devices is increasing dramatically. Traditional IPv4 networks, due to insufficient address capacity and limited protocol extension capabilities, struggle to meet the flexible access needs of large-scale intelligent terminals. Furthermore, the high cost and heavy maintenance burden of external security gateways limit the further evolution of smart substation communication architectures. Against this backdrop, leveraging the ample address space and native extended header mechanism of the IPv6 protocol to abstract real-time terminal behavior into semantic features and embed them into the protocol layer for in-line communication—thereby enhancing the security of interaction protocols and continuously protecting terminal behavior—has become an important direction for the development of power information and communication network technologies.

[0003] However, in the engineering implementation of semantic behavioral communication for IPv6 terminals in substations, there are underlying technical contradictions that have not yet been fully resolved by the industry. Specifically, the behavioral semantics of terminal devices have high-dimensional and multimodal inherent characteristics. If such complex semantic information is directly filled into the IPv6 extended header in variable-length form for communication, it will exceed the hardware parsing capability limit of the data plane pipeline of the white-box switch's underlying programmable switching chip. The fundamental reason is that the data plane of existing switching chips can only achieve constant-time line-rate matching processing for fixed offsets and fixed-length fields. When the semantic features carried in the message exhibit variable-length attributes, the chip's parsing window depth is insufficient to cover the complete semantic range, forcing the message to be sent to the control plane for software-level parsing. This uploading operation introduces millisecond-level or even higher levels of operating system interruption and queuing delay, creating an irreconcilable conflict with the microsecond-level deterministic end-to-end transmission delay required by control services such as relay protection in smart substations, seriously threatening power production safety and communication quality assurance. Therefore, there is an urgent need for a communication method that can directly perform line-rate extraction, verification, and deterministic forwarding of terminal behavior semantics in the data plane hardware pipeline without triggering control plane transmission. Summary of the Invention

[0004] To address the aforementioned technical problems, this application is proposed. Embodiments of this application provide a semantic communication method for IPv6 terminal behavior in substations.

[0005] According to one aspect of this application, a semantic communication method for IPv6 terminal behavior in a substation is provided, comprising: Step 1: Based on the multi-dimensional operating parameters of the terminal, perform graph-structured fusion modeling of the node attribute features and communication edge relationships of the terminal to obtain the terminal behavior semantic ontology state graph. Step 2: Perform terminal behavior semantic dimensionality reduction and feature extraction based on graph neural networks on the terminal behavior semantic ontology state graph to obtain a fixed-length semantic state vector. Step 3: Inject the fixed-length semantic state vector bit by bit into the reserved alignment spare bits of the hop-by-hop option extension header of the Internet Protocol version 6 (IP6), and concatenate and encapsulate it with the substation control business data payload to obtain a business message carrying the semantic vector. Step 4: In the white-box switch data plane hardware pipeline, the service packets carrying semantic vectors are truncated at line speed based on a fixed offset to obtain fixed-length semantic state vectors. The fixed-length semantic state vectors are then matched and verified in parallel with the pre-compiled semantic state transition rule flow table using a three-state content addressing memory. The semantic legality matching result instruction is output and a forwarding packet stripped of semantic vectors is generated. Step 5: Based on the semantic legitimacy matching result, perform deterministic network slice scheduling and forwarding on the forwarded packets stripped of semantic vectors to output a deterministic slice forwarding data stream.

[0006] Compared with existing technologies, this application provides a semantic communication method for IPv6 terminals in substations. By introducing graph neural network embedding dimensionality reduction technology, it constructs a semantic ontology state graph from the terminal's multi-dimensional operating parameters. This graph is then compressed into a fixed-length semantic state vector with a fixed bit width through graph convolution aggregation and hash discretization, eliminating the variable-length attribute of semantic features. Based on this, the fixed-length vector is injected bit-by-bit into the reserved alignment spare bits of the IPv6 hop-by-hop option extension header to achieve implicit in-path communication. This allows the white-box switch's programmable pipeline on the data plane to complete line-speed extraction of the semantic vector based on a fixed offset without variable-length parsing. Nanosecond-level parallel matching and verification are performed using a tri-state content-addressable memory and a pre-compiled rule flow table, directly determining the legality of terminal behavior at the hardware layer. The matching result drives deterministic network slice scheduling, bypassing the control plane throughout, thereby simultaneously achieving continuous protection of terminal behavior semantics and microsecond-level deterministic transmission guarantee for control services without introducing additional upload latency. Attached Figure Description

[0007] The above and other objects, features, and advantages of this application will become more apparent from the more detailed description of the embodiments of this application in conjunction with the accompanying drawings. The drawings are provided to further illustrate the embodiments of this application and form part of the specification. They are used together with the embodiments of this application to explain this application and do not constitute a limitation thereof. In the drawings, the same reference numerals generally represent the same components or steps.

[0008] Figure 1This is a flowchart of a semantic communication method for IPv6 terminal behavior in a substation according to an embodiment of this application; Figure 2 This is a schematic diagram of the data flow of the semantic communication method for IPv6 terminal behavior in a substation according to an embodiment of this application; Figure 3 This is a flowchart of step three of the semantic communication method for substation IPv6 terminal behavior according to an embodiment of this application; Figure 4 The flowchart illustrates the process of converting the hop-by-hop option extension header skeleton into a hop-by-hop option extension header carrying semantics, which involves offset injection overlay of a fixed-length semantic state vector according to the semantic communication method for substation IPv6 terminal behavior in this application. Figure 5 This is a flowchart of step four of the semantic communication method for IPv6 terminal behavior in a substation according to an embodiment of this application. Detailed Implementation

[0009] Hereinafter, exemplary embodiments according to this application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of this application, and not all embodiments of this application. It should be understood that this application is not limited to the exemplary embodiments described herein.

[0010] In the context of IPv6 terminal behavior semantic communication in smart substations, a fundamental contradiction exists between the high-dimensional complexity of terminal behavior semantics and the hardware limitations of white-box switch data plane chips, which only support fixed-length field line-rate processing. The uncontrollable latency introduced by variable-length semantic messages triggering control plane transmission will breach the deterministic transmission baseline of control-related services. To address this issue, this solution first starts with multi-dimensional operating parameters of the terminal, such as relay protection device operating status codes, sensor sampling period intervals, network communication port numbers, and historical access control permission identifiers. It then performs graph-structured fusion modeling of node attribute features and communication edge relationships to construct a behavior semantic ontology state graph that can fully represent the terminal's current operational intent and communication interaction mode. Subsequently, local adjacency sampling and multi-layer graph convolution feature aggregation are applied to this state graph to compress the high-dimensional topological semantics layer by layer and discretize and hash it into a fixed-length semantic state vector with the same width as the spare bits in the IPv6 extended header, completely eliminating the hardware bottleneck of variable-length parsing. The vector is injected bitwise into the reserved bits of the hop-by-hop option extension header and concatenated with the service payload to encapsulate a standard message. After entering the white-box switch, the programmable parser performs line-rate truncation based on a fixed offset. The tri-state content addressing memory performs parallel matching and verification with the pre-compiled semantic state transition rule flow table within nanoseconds and outputs a legality instruction. Finally, based on this instruction, deterministic network slicing scheduling and forwarding are performed on the message. No control plane intervention is required throughout the process, realizing the synchronous guarantee of continuous protection of terminal behavior semantics and microsecond-level deterministic communication.

[0011] Figure 1 This is a flowchart of a semantic communication method for IPv6 terminal behavior in a substation according to an embodiment of this application. Figure 2 A schematic diagram of the data flow of the semantic communication method for substation IPv6 terminal behavior according to an embodiment of this application is shown below. Figure 1 and Figure 2 As shown, the substation IPv6 terminal behavior semantic communication method according to an embodiment of this application includes: Step 1, based on the terminal's multi-dimensional operating parameters, performing graph-structured fusion modeling of the terminal's node attribute features and communication edge relationships to obtain a terminal behavior semantic ontology state graph; Step 2, performing terminal behavior semantic dimensionality reduction and feature extraction based on graph neural networks on the terminal behavior semantic ontology state graph to obtain a fixed-length semantic state vector; Step 3, injecting the fixed-length semantic state vector bit by bit into the reserved alignment spare bits of the hop-by-hop option extension header of the Internet Protocol version 6 (IP6), and integrating it with the substation control-type service data payload. Step 4: In the white-box switch data plane hardware pipeline, the service packets carrying semantic vectors are spliced ​​and encapsulated to obtain a fixed-length semantic state vector based on a fixed offset. The fixed-length semantic state vector is then matched and verified in parallel with the pre-compiled semantic state transition rule flow table using a three-state content addressing memory. The semantic legality matching result instruction is output and a forwarding packet stripped of semantic vector is generated. Step 5: Based on the semantic legality matching result instruction, the forwarding packet stripped of semantic vector is subjected to deterministic network slice scheduling and forwarding to output a deterministic slice forwarding data stream.

[0012] Specifically, in step one, based on the terminal's multi-dimensional operating parameters, a graph-structured fusion model is performed on the terminal's node attribute features and communication edge relationships to obtain a terminal behavior semantic ontology state graph. It is worth noting that the terminal's multi-dimensional operating parameters include the relay protection device's operating status code, sensor sampling period interval, terminal network communication port number, and historical access control permission identifier. It should be understood that, due to the large number of heterogeneous terminal devices deployed in smart substations, the network-level operating behavior of each terminal is not determined by a single-dimensional state parameter, but rather by multiple heterogeneous parameters such as its own device operating status, historical security permissions, communication timing characteristics, and network interaction modes. Furthermore, these parameters have complex topological relationships and spatiotemporal coupling relationships, and a single flat vector or sequence model cannot fully characterize the structured interactive dependencies between terminals. Therefore, in the technical solution of this application, a graph-structured fusion model is performed on the terminal's node attribute features and communication edge relationships based on the terminal's multi-dimensional operating parameters to obtain a terminal behavior semantic ontology state graph. This unifies the static attributes of the terminal devices and the dynamic communication relationships between devices into a graph data representation framework with a topological structure. In this way, while preserving the complete semantic information of terminal behavior and the relationship structure between devices, a structured standard input can be provided for subsequent graph neural network dimensionality reduction calculations, avoiding a decrease in the representation accuracy of semantic state vectors due to missing information or lost relationships.

[0013] More specifically, in the embodiments of this application, step one includes: vectorizing the operating status code of the relay protection equipment and concatenating and aggregating it with the permission vector of the historical access control permission identifier in the channel dimension to obtain the terminal node attribute feature matrix; estimating the time decay factor scalar of the sensor sampling period interval and performing spatiotemporal feature joint mapping with the high-dimensional embedding vector of the terminal network communication port number to obtain the terminal communication edge relationship matrix; using the terminal node attribute feature matrix as the vertex feature set and the terminal communication edge relationship matrix as the directed edge feature set, performing entity relationship alignment and graph representation fusion based on the substation physical network connection adjacency matrix to obtain the terminal behavior semantic ontology state graph.

[0014] Specifically, in a particular example of this application, firstly, the operating status code of the relay protection equipment is processed into a status vector, that is, the current discrete status code is mapped into a status vector in a continuous numerical space according to a preset equipment operating status coding table. Simultaneously, the historical access control permission identifier is processed into a permission vector. Then, the status vector and the permission vector are concatenated and aggregated along the channel dimension. After linear transformation by a preset node mapping weight matrix and normalization by a nonlinear activation function, the terminal node attribute feature matrix is ​​obtained. This calculation process is expressed as follows:

[0015] in, This is the output terminal node attribute feature matrix; The state vector is obtained by state vectorization of the operating status code of relay protection equipment. The permission vector is obtained by vectorizing the historical access control permission identifier; Concatenate aggregation operators for channel dimensions; The preset node mapping weight matrix; This is the bias term vector; The activation function is nonlinear. In the substation scenario, the state vector carries a numerical representation of the relay protection device's current operating mode, such as closed standby, open protection, or maintenance lockout. The permission vector reflects the security trust level of the terminal in the historical access control policy. The concatenation and aggregation of the two at the channel dimension enables the terminal node attribute feature matrix to simultaneously encode the physical operating state and logical security identity of the device, providing a dual semantic basis for the node features in the subsequent graph structure, which combines functional and security attributes.

[0016] Next, a scalar estimation of the time decay factor is performed on the sensor sampling period interval. This is achieved by applying a reciprocal normalization operation to the sampling period interval to obtain a scalar time decay factor reflecting the activity level of the terminal's real-time sampling frequency. Simultaneously, the terminal's network communication port number is mapped to a high-dimensional embedding vector using an embedding layer. Then, a spatiotemporal feature joint mapping is performed between the time decay factor scalar and the high-dimensional embedding vector. After a linear transformation of the edge feature mapping weight matrix and processing with a hyperbolic tangent activation function, the terminal communication edge relationship matrix is ​​obtained. This calculation process is expressed as follows:

[0017] in, This is the output terminal communication edge relationship matrix; The time decay factor is a scalar obtained by normalizing the inverse of the sensor sampling period interval. The shorter the sampling period, the larger the value of this factor, indicating that the terminal communication activity is higher. The high-dimensional embedding vector is obtained by mapping the terminal network communication port number through the embedding layer. The edge feature mapping weight matrix; The hyperbolic tangent activation function is used. In the substation scenario, the time decay factor scalar characterizes the data reporting frequency of the sensor under the current operating conditions. This scalar increases when the terminal enters the high-frequency inspection mode. Meanwhile, the high-dimensional embedding vector transforms the port number from a discrete identifier into a continuous feature expression that reflects the communication service type and protocol level. The joint mapping of the spatiotemporal features of the two enables the terminal communication edge relationship matrix to simultaneously encode the temporal rhythm and network level attributes of the communication interaction between terminals, providing a relational measure that integrates spatiotemporal semantics for the weight assignment of directed edges in the graph structure.

[0018] Then, the terminal node attribute feature matrix is ​​used as the vertex feature set, and the terminal communication edge relationship matrix is ​​used as the directed edge feature set. Based on the substation physical network connection adjacency matrix, entity relationship alignment and graph representation fusion are performed to obtain the terminal behavior semantic ontology state graph. Specifically, a directed graph topology data structure is instantiated in memory. Based on the actual fiber optic or Ethernet connection relationships between terminal devices recorded in the substation physical network connection adjacency matrix, it is determined which nodes are allowed to have valid directed edge connections. The terminal node attribute feature matrix is ​​aligned and assigned to the corresponding vertex positions according to the device entity index, and the terminal communication edge relationship matrix is ​​assigned to the corresponding directed edge positions according to the connectivity relationships indicated by the adjacency matrix. After completing the joint mounting of node attributes and edge attributes on the unified topology structure, the terminal behavior semantic ontology state graph is output. In the substation scenario, the physical network adjacency matrix of the substation reflects the actual physical connection topology between various terminal devices in the substation. Entity relationship alignment based on this matrix ensures that the logical connections in the graph structure are strictly consistent with the actual layout of the physical network. This allows the terminal behavior semantic ontology state graph to not only carry the operational state semantics of each terminal itself, but also accurately preserve the communication interaction structure between terminals based on the physical network. This provides a topologically reliable structured input for subsequent graph neural networks to perform neighborhood information transmission and feature aggregation on this graph.

[0019] Specifically, in step two, the terminal behavior semantic ontology state graph is subjected to terminal behavior semantic dimensionality reduction and feature extraction based on graph neural networks to obtain a fixed-length semantic state vector. It should be understood that since the terminal behavior semantic ontology state graph is essentially a high-dimensional topology structure containing multiple nodes and edges, the amount of semantic information it carries far exceeds the fixed bit width that can be accommodated by the reserved bits in the hop-by-hop option extension header in the white-box switch's data plane hardware pipeline. If all the features in the graph structure are directly encoded into the packet without effective dimensionality reduction and compression, variable-length semantic fields will inevitably be generated, triggering the switching chip's parsing window overflow and causing the packet to be sent to the control plane, thereby compromising the deterministic latency guarantee of control services. Therefore, in the technical solution of this application, the terminal behavior semantic ontology state graph is further subjected to terminal behavior semantic dimensionality reduction and feature extraction based on graph neural networks to obtain a fixed-length semantic state vector. This preserves the core semantic discrimination information of terminal behavior while compressing the high-dimensional graph structure features into a fixed-length bit sequence with a width strictly equal to the spare bits in the extension header. This fundamentally eliminates the variable-length attribute of semantic features, enabling subsequent white-box switch data plane programmable pipelines to complete line-speed truncation and hardware-level matching verification of semantic vectors based on a fixed offset without variable-length parsing, ensuring that the control plane transmission is not triggered throughout the process.

[0020] More specifically, in the embodiments of this application, step two includes: performing local high-frequency adjacency sampling and topological subgraph extraction on the terminal behavior semantic ontology state graph to obtain a local semantic topological subgraph; performing multi-layer feature aggregation on the local semantic topological subgraph through a multi-layer graph convolutional network to obtain a high-dimensional continuous semantic feature tensor; and performing feature dimensionality reduction compression and discretized hash calculation on the high-dimensional continuous semantic feature tensor to obtain a fixed-length semantic state vector.

[0021] Specifically, in a concrete example of this application, a local high-frequency adjacency sampling and topology subgraph extraction are first performed on the terminal behavior semantic ontology state graph to obtain a local semantic topology subgraph. Specifically, taking the terminal node currently undergoing a behavior state change as the sampling starting point, a multi-round random walk process is performed on the terminal behavior semantic ontology state graph. The transition probability is calculated based on the weight attributes of the directed edges between each neighboring node. High-frequency accessed core nodes and their associated paths are retained, while long-tailed noise nodes with access frequencies below a preset threshold are pruned and removed. The retained nodes and their corresponding directed edges are then reconstructed into a lightweight local semantic topology subgraph. In a substation scenario, when a relay protection terminal switches from standby mode to closing execution mode, this terminal serves as the sampling starting point. High-frequency communication neighbor nodes such as measurement and control devices and merging units that directly interact with it in the physical network are prioritized for sampling and retention, while remote monitoring terminals with minimal communication are pruned and excluded. This allows the local semantic topology subgraph to centrally carry the core intent evolution trajectory closely related to the current behavior change event, effectively reducing the scale and complexity of subsequent graph convolution calculations.

[0022] Next, a multi-layer graph convolutional network is used to aggregate features from the local semantic topological subgraph to obtain a high-dimensional continuous semantic feature tensor. Specifically, the adjacency structure and initial node features of the local semantic topological subgraph are input into a pre-trained multi-layer graph convolutional network. In each convolutional operation, each node collects feature information from its first-order neighboring nodes based on topological connectivity, performs weighted aggregation with its own features, and then generates the hidden representation of the current layer through a non-linear activation function. This iterative propagation layer by layer enables each node to ultimately fuse the global context semantics within its multi-hop neighborhood. The feature propagation process of a single-layer graph convolution is represented as follows:

[0023] in, For the first The node hidden feature representation matrix output by layer graph convolution; To modify the activation function of the linear unit; Adding self-loop connections to the adjacency matrix of the local semantic topological subgraph, i.e. ,in This is the original adjacency matrix of the local semantic topological subgraph. It is the identity matrix; for The corresponding degree matrix; It is a symmetrically normalized Laplacian matrix used to achieve normalized aggregation and propagation of neighborhood features; For the first The node feature input matrix of the layer, when l=0, is the initial attribute feature of each node in the local semantic topological subgraph; For the first The learnable graph convolutional weight matrix of the layer. In the substation scenario, the symmetric normalized Laplacian matrix structure ensures that each terminal node is weighted and balanced according to its connectivity when aggregating neighborhood information, avoiding the dominance of high-connectivity core exchange nodes in the aggregation process. Simultaneously, the introduction of self-loop connections ensures that each terminal node retains its original semantic information during feature updates without being completely covered by neighborhood features. After multi-layer iterative aggregation, a global average pooling operation is applied to the hidden representations of all nodes in the final layer, fusing node-level features into graph-level features and outputting a high-dimensional continuous semantic feature tensor.

[0024] Then, feature dimensionality reduction compression and discretization hashing are performed on the high-dimensional continuous semantic feature tensor to obtain a fixed-length semantic state vector. Specifically, the high-dimensional continuous semantic feature tensor is input into a fully connected compression layer for linear low-dimensional space mapping. After matrix multiplication of the compression weight matrix, its dimension is compressed from the high-dimensional hidden space to the target dimension space with the same width as the reserved bits in the IPv6 hop-by-hop option extension header. Subsequently, a step hash function is applied to the compression result for discretization, forcing continuous floating-point values ​​to be mapped into a binary bit sequence containing only zeros and ones. Finally, after bit-width alignment and truncation, a fixed-length semantic state vector is generated. This calculation process is expressed as:

[0025] in, This is the output fixed-length semantic state vector; This is a step hash function that performs hard binarization on each component of the input vector, outputting one when the component value is greater than or equal to zero, and outputting zero when it is less than zero. The input is a high-dimensional continuous semantic feature tensor; This is the learnable compressible weight matrix of the fully connected compressed layer, responsible for mapping the high-dimensional feature space to the target low-dimensional space; This is the bias term vector of the fully connected compression layer. In the substation scenario, the fully connected compression layer compresses the hundreds-dimensional continuous feature space output by the graph convolutional network into a 128-dimensional target space with the same width as the reserved bits in the hop-by-hop option extension header by using a compression weight matrix. The discretization processing of the step hash function converts the compressed continuous floating-point representation into a pure binary bit sequence, making the final fixed-length semantic state vector fully compatible with the IPv6 header field in terms of bit format. It can be directly injected bit by bit without any additional encoding format conversion, thus providing a strictly fixed-length and hardware-natively recognizable input format for the data plane hardware pipeline's line-rate truncation based on fixed offset and parallel bit matching of the tri-state content-addressable memory in subsequent steps.

[0026] Specifically, in step three, the fixed-length semantic state vector is injected bit-by-bit into the reserved alignment spare bits of the hop-by-hop option extension header of Internet Protocol version 6 (IP6), and then concatenated and encapsulated with the substation control service data payload to obtain a service message carrying the semantic vector. It should be understood that although the fixed-length semantic state vector obtained in the aforementioned steps already possesses a fixed-width binary format compatible with hardware header fields, it is, as an isolated bit sequence, not yet embedded in any standardized network communication protocol structure. Therefore, it cannot be transmitted along with the service data stream in the substation IPv6 network, nor can it be identified and located by the data plane hardware pipeline of the white-box switches along the route based on the protocol specification. Therefore, in the technical solution of this application, the fixed-length semantic state vector is further injected bit-by-bit into the reserved alignment spare bits of the hop-by-hop option extension header of IP6, and then concatenated and encapsulated with the substation control service data payload to obtain a service message carrying the semantic vector. This utilizes the native extension header mechanism of IP6 to implicitly encode semantic information in a fixed offset position allowed by the protocol specification to achieve in-path communication. In this way, without adding extra protocol fields or changing the variable-length attribute of the message, the semantic vector and the service payload can be encapsulated together in a standard IPv6 message for transmission. This ensures that the programmable parser of the data plane of the white-box switch along the way can directly locate and extract the semantic vector based on the fixed offset known in the protocol specification, without the need for deep variable-length parsing of the message.

[0027] Figure 3 This is a flowchart of step three of the semantic communication method for IPv6 terminal behavior in a substation according to an embodiment of this application. Figure 3As shown, step three includes: S31, initializing and generating the Internet Protocol version 6 (IP6) basic header, and constructing a hop-by-hop option extension header skeleton based on the type-length-value format, and allocating a range of all-zero reserved bits with the same width as the fixed-length semantic state vector in the skeleton; S32, performing offset injection overlay on the fixed-length semantic state vector, and converting the hop-by-hop option extension header skeleton into a hop-by-hop option extension header carrying semantics; S33, concatenating and encapsulating the I6 basic header, the hop-by-hop option extension header carrying semantics, and the substation control business data payload according to the network byte order to obtain a business message carrying a semantic vector.

[0028] Accordingly, in step S31, the Internet Protocol version 6 (IP6) basic header is initialized and generated, and a hop-by-hop option extension header skeleton is constructed based on the type-length-value format. A reserved bit range of all zeros with the same width as the fixed-length semantic state vector is allocated within the skeleton. It should be understood that since the fixed-length semantic state vector generated in the aforementioned steps needs to be embedded into a standardized network communication protocol structure for transmission along the route in the substation IPv6 network, and the IPC version 6 hop-by-hop option extension header is the only extension header type in the protocol specification that requires inspection and processing by each hop routing device, its type-length-value-based option structure naturally supports carrying fixed-length additional data in the reserved padding alignment area. However, before actually injecting the semantic vector, a header skeleton conforming to the protocol specification must be pre-constructed, and a precisely equal-width bit storage space must be reserved as the injection target range. Therefore, in the technical solution of this application, the basic header of Internet Protocol version 6 (IP6) is initialized and generated, and a hop-by-hop option extension header skeleton is constructed based on the type-length-value format. A range of all-zero reserved bits with a width equal to that of the fixed-length semantic state vector is allocated within the skeleton. This provides a standardized carrier container with a compliant format and fixed offset for subsequent bit-by-bit semantic vector injection operations. This ensures that the final encapsulated message fully conforms to the I6 standard specification in structure, enabling the programmable parsers of white-box switches along the route to accurately locate the semantic vector's storage position using a deterministic fixed offset based on the known protocol format. This guarantees the feasibility and determinism of subsequent line-rate data plane truncation operations.

[0029] More specifically, in a specific example of this application, the network protocol stack of the substation terminal first initializes and generates the Internet Protocol version 6 (IPv6) basic header according to the current service communication requirements. In this basic header, the version number field is filled with 6, the traffic category field is set according to the service level of the current service, the flow label field is assigned a value according to the terminal's session identifier, the hop count limit field is set to the maximum hop value of the substation's network topology, the source address field is filled with the IPv6 global unicast address of the current terminal, the destination address field is filled with the IPv6 global unicast address of the destination terminal, and the value of the next header field is set to zero to declare to the network device that there is a hop-by-hop option extension header immediately following the basic header. Next, a fixed-length contiguous byte space is allocated in memory. A hop-by-hop option extension header skeleton is constructed according to the type-length-value format. The extension header length field is calculated based on the bit width of the fixed-length semantic state vector plus the fixed overhead of the option control field, and is encoded and padded in units of eight bytes. The option type field is set to a preset experimental option code, with its highest two bits set to zero to indicate that nodes along the path should skip processing if the option is not recognized. The option data length field is set to the number of bytes in the fixed-length semantic state vector. Then, a contiguous interval of all-zero bits is allocated after the option data length field. The bit width of this interval is strictly equal to the bit width of the fixed-length semantic state vector, i.e., 128 bits. All bits are initialized to zero to serve as the target write area for subsequent semantic vector offset injection. In the substation scenario, setting the next header field in the basic header to zero ensures that all white-box switches within the substation can identify the existence of the hop-by-hop option extension header according to the protocol specification and trigger the corresponding hardware processing logic when parsing the message. The configuration of setting the highest two bits of the option type in the hop-by-hop option extension header skeleton to zero ensures that the option will not cause the message to be dropped even on some un-upgraded intermediate network nodes. The precise equal-width allocation of the all-zero reserved bit interval provides a target storage space with clear boundaries and fixed offsets for subsequent bit-by-bit clearing and bit-by-bit OR overwrite operations, making the entire header skeleton fully ready at the format level to accept the injection of fixed-length semantic state vectors.

[0030] Accordingly, in step S32, the fixed-length semantic state vector is offset-injected and overwritten, converting the hop-by-hop option extension header skeleton into a hop-by-hop option extension header carrying semantics. It should be understood that since the aforementioned steps have already constructed the hop-by-hop option extension header skeleton and allocated a reserved interval of all-zero bits with a width equal to that of the fixed-length semantic state vector, the actual bit content of the fixed-length semantic state vector needs to be written into this reserved interval to complete the embedding of semantic information at the protocol layer. This writing operation must precisely overwrite the semantic bit sequence to the specified offset position without destroying the original values ​​of the remaining control fields in the hop-by-hop option extension header skeleton. Therefore, in the technical solution of this application, the fixed-length semantic state vector is further offset-injected and overwritten, converting the hop-by-hop option extension header skeleton into a hop-by-hop option extension header carrying semantics, thereby embedding the terminal behavior semantic information with bit-level precision into the fixed offset position allowed by the protocol specification. In this way, while maintaining the integrity of the hop-by-hop option extension header format and protocol compliance, the semantic vector is transmitted with the message in an implicitly encoded form, providing a deterministic bit positioning basis for subsequent line-rate interception of the data plane of the white-box switch based on a fixed offset.

[0031] More specifically, in a concrete example of this application, a fixed-length semantic state vector is directly injected into the reserved bit range of the hop-by-hop option extension header skeleton using pure bit-level operations. First, a bitwise AND clearing operation is performed on the target reserved bit range in the hop-by-hop option extension header skeleton, generating a reverse mask where all bits in the target injection range are zero and all other bits are one. This reverse mask is then bitwise ANDed with the hop-by-hop option extension header skeleton, forcibly clearing all bits in the target range while leaving control information such as the option type field and option length field unchanged. Next, a logical left shift alignment operation is performed on the fixed-length semantic state vector, shifting it left to the bit position corresponding to the starting offset of the reserved bit range in the hop-by-hop option extension header skeleton, ensuring that the least significant bit of the semantic vector is strictly aligned with the least significant bit of the target injection range. Then, a bitwise OR operation is performed between the left-shifted and aligned fixed-length semantic state vector and the zeroed-out hop-by-hop option extension header skeleton. This fills the zeroed-out target interval with the bit content of the semantic vector, ultimately outputting the hop-by-hop option extension header carrying the semantics. This injection process is represented as follows:

[0032] in, Expand the header with semantically-carrying hop-by-hop options in the output; Expand the header skeleton for the hop-by-hop options input; This is the bitwise AND operator; To preserve the bit-inverted result of the bitmask, the corresponding bit in the target injection interval is zero and the corresponding bit in the non-injection region is one, which is used to clear the target injection interval while retaining the original data in the non-injection region; For the bitwise OR operator, the aligned semantic vector is overwritten into the target range that has been cleared. The fixed-length semantic state vector generated by the aforementioned steps; This is a logical left shift operator used to align a fixed-length semantic state vector to the beginning of the target bit range; This specifies the starting offset for the bit range preserved in the hop-by-hop option extension header. In the substation scenario, the bitwise AND clearing operation ensures that protocol control fields such as option type code and option data length in the hop-by-hop option extension header skeleton are not contaminated by semantic data during injection. The logical left shift alignment operation ensures that each bit of the fixed-length semantic state vector falls precisely within the physical boundary of the option value region in the type-length-value structure. The bitwise OR overwrite operation writes the semantic vector as a static bit block into the cleared target space all at once.

[0033] In particular, it is worth noting that in the message encapsulation stage of IPv6 terminal behavior semantic communication in substations, the first embodiment of the original sub-step S32 adopts a pure bit-level bit operation method to directly overwrite and inject the fixed-length terminal behavior semantic state vector into the reserved bits of the hop-by-hop option extension header skeleton. This operation is completed in the underlying implementation through only three steps: bitwise AND clearing, logical left shift alignment, and bitwise OR overwriting. Essentially, it treats the semantic vector as a static bit block that is completely independent of the rest of the message and has no contextual association, and mechanically moves it.

[0034] However, in the actual industrial control communication scenarios of smart substations, there is a strict content-level semantic-load consistency binding relationship between the fixed-length terminal behavior semantic state vector and the original data payload of substation control services. Specifically, the fixed-length terminal behavior semantic state vector represents the terminal's current operational intent, such as a relay protection device performing a closing operation or a sensor entering a high-frequency inspection mode, while the original data payload of substation control services carries the specific instruction parameters and control data corresponding to this operational intent. In normal communication, a specific behavior semantic vector must and can only be transmitted together with its corresponding specific service payload in the same message, forming an inseparable logical dual. The fundamental weakness of the first embodiment is that the semantic vector injection calculation process does not depend on the payload content at all, and the payload data does not participate in the semantic vector encoding operation in any form. This complete separation at the computational level means that there is no verifiable mathematical binding relationship between the injected semantic vector and the payload of its message. In real-world attack scenarios at substations, malicious attackers can intercept a legitimate packet using network sniffing techniques. They can then extract the legitimate semantic vector from the packet using a known fixed offset and graft this vector onto a forged packet carrying malicious control commands. Since the grafted semantic vector remains legitimate, the subsequent parallel matching and verification of the three-state content-addressed memory on the white-box switch's data plane will be unable to distinguish between the forged and legitimate packets. The malicious control commands will then be deterministically sliced ​​and forwarded with high priority to the substation's core protection equipment, constituting a semantic vector grafting replay attack. This could potentially lead to serious power safety incidents such as relay protection malfunctions or failures to operate.

[0035] To address the security vulnerability of missing content binding between semantic vectors and business payloads, a second embodiment is proposed, which includes the following steps. Figure 4 This document describes a flowchart illustrating the process of converting a hop-by-hop option extension header skeleton into a semantically-driven hop-by-hop option extension header, based on an embodiment of the semantically-driven semantic communication method for substation IPv6 terminals according to this application. Figure 4 As shown, step S32 includes: S321, performing equal-width fixed-length hash compression on all byte sequences of substation control business data payload to obtain a payload content binding fingerprint with the same bit width as the fixed-length semantic state vector; S322, applying a cyclic right shift operation to the fixed-length semantic state vector based on the dynamic rotation offset calculated by taking the modulus of the total bit width using the payload content binding fingerprint, and then performing bit-by-bit XOR fusion with the payload content binding fingerprint to obtain a payload binding semantic vector; S323, injecting the payload binding semantic vector into the reserved bit range of the hop-by-hop option extension header skeleton to obtain a hop-by-hop option extension header carrying semantics.

[0036] In step S321, the entire byte sequence of the substation control business data payload is subjected to equal-width fixed-length hash compression to obtain a payload content binding fingerprint with the same bit width as the fixed-length semantic state vector. It should be understood that, in the aforementioned first embodiment, the injection calculation process of the fixed-length semantic state vector is completely independent of the content of the substation control business data payload; the two are completely separated at the computational level. This results in no verifiable mathematical binding relationship between the injected semantic vector and the payload of its corresponding message. Malicious attackers can use network sniffing to intercept legitimate messages and extract the legitimate semantic vector using a known fixed offset, then graft it onto a forged message carrying malicious control instructions to constitute a semantic vector grafting replay attack. Furthermore, the subsequent binding and fusion operation requires the two operands involved in the calculation to have strictly identical bit widths. The length of the substation control business data payload dynamically changes with different business types and cannot directly participate in fixed-length bit operations. Therefore, in the technical solution of this application, a fixed-length hash compression with equal bit width is performed on the entire byte sequence of the substation control business data payload to obtain a payload content binding fingerprint with the same bit width as the fixed-length semantic state vector. This compresses the variable-length payload content into a fixed-length hash value with a strictly equal bit width to the fixed-length semantic state vector, thus satisfying the requirement for operand bit width consistency in subsequent binding fusion operations. In this way, the avalanche effect characteristic of cyclic redundancy hashing ensures that tampering with any bit in the payload will cause approximately 50% of the bits in the payload content binding fingerprint to be flipped, providing a highly sensitive content summary basis for subsequent binding calculations and ensuring that an inseparable mathematical association at the content level is established between the semantic vector and the payload.

[0037] In a specific example of this application, the entire byte sequence of substation control service data payload in the current message is first obtained as the input object for hash operation. The length of this byte sequence dynamically changes depending on the service type. When the terminal performs a closing operation, the payload carries closing command parameters and control data; when the terminal enters high-frequency inspection mode, the payload carries telemetry sampling data and inspection parameters. The byte length of the payload varies under different service scenarios. Next, a 128-bit cyclic redundancy hash operation is applied to the entire byte sequence of substation control service data payload, compressing the variable-length payload content into a fixed-length hash value with the same bit width as the fixed-length semantic state vector. The output is the fingerprint bound to the payload content. This process is represented as:

[0038] in, A fingerprint is bound to the output payload content, with a fixed bit width of 128 bits; It is a 128-bit cyclic redundancy hash function; This is the complete byte sequence of the original data payload for substation control services. In the substation scenario, the reason for first performing payload content binding fingerprint extraction is that subsequent binding fusion operations require the two operands involved in the calculation to have strictly identical bit widths. However, the length of the substation control service data payload dynamically changes depending on the service type, making it impossible to directly participate in fixed-length bit operations. Specifically, when a relay protection device performs a closing operation, its payload may only contain tens of bytes of instruction parameters, while when a sensor performs high-frequency inspection, its payload may contain hundreds of bytes of telemetry data. This service-driven difference in payload length makes it impossible to directly align the substation control service data payload with the fixed-length semantic state vector (128 bits wide) at the bit level. Through compression mapping using a 128-bit cyclic redundancy hash function, regardless of the original payload length, it is uniformly compressed into a 128-bit payload content binding fingerprint, thus ensuring it has the same bit width as the fixed-length semantic state vector. This satisfies the strict requirement for operand format consistency in subsequent cyclic rotation XOR binding fusion calculations. Furthermore, the avalanche effect of the cyclic redundancy hash function ensures that any tampering with any bit in the payload will cause approximately 50% of the bits in the payload content binding fingerprint to be flipped. This means that in the actual attack scenarios faced by substations, if an attacker makes even the slightest tampering with any instruction parameter or control data in the substation control business data payload, the payload content binding fingerprint output by the hash will undergo a global and drastic change. This provides a highly sensitive content digest basis for subsequent binding calculations, enabling any payload tampering behavior to be detected by hardware at the hardware level in the subsequent reverse binding restoration stage.

[0039] In step S322, the dynamic rotation offset calculated based on the payload content binding fingerprint modulo the total bit width is applied to the fixed-length semantic state vector by performing a cyclic right shift operation, and then fused bit-by-bit with the payload content binding fingerprint to obtain the payload binding semantic vector. It should be understood that since the aforementioned steps have already obtained a payload content binding fingerprint with a strictly equal bit width to the fixed-length semantic state vector, if only a simple fixed offset XOR operation is used to fuse the fixed-length semantic state vector with the payload content binding fingerprint, although the two vectors can be merged into one in form, an attacker, once aware of the fixed XOR pattern, can easily reconstruct the original semantic vector and complete the grafting through inverse XOR, failing to fundamentally block the semantic vector grafting replay attack path. Therefore, to completely block this attack path, a data-dependent dynamic rotation mechanism is introduced. In the technical solution of this application, the dynamic rotation offset calculated by modulo the total bit width based on the payload content binding fingerprint is further applied to the fixed-length semantic state vector by performing a cyclic right shift operation and then XORing it bit by bit with the payload content binding fingerprint to obtain the payload binding semantic vector. This introduces a data-dependent dynamic rotation mechanism so that the physical arrangement of semantic bits is disturbed by the payload content, thereby completely blocking the attacker's attack path of reverse reconstruction through a fixed pattern. In this way, an unforgeable mathematical binding relationship can be established between the fixed-length semantic state vector and the substation control business data payload. In the substation adversarial scenario, if the attacker grafts the payload binding semantic vector onto messages of different payloads, the fingerprint calculated by the receiving end using the new payload will generate completely different rotation amounts and XOR bases. The reverse binding reconstruction process will output a globally disordered pseudo-semantic vector. This pseudo-vector cannot hit any legitimate rule entries in the three-state content addressing memory of the white-box switch, thereby achieving line-speed interception of grafted replay attacks at the hardware pipeline level.

[0040] In a specific example of this application, firstly, low-order bits are extracted from the payload content binding fingerprint and modulo the total bit width to calculate a dynamic rotation offset completely determined by the payload content. The value of this offset depends entirely on the specific content of the substation control business data payload in the current message; different payloads will produce different rotation offsets, while the same payload will always produce the same rotation offset. Next, a cyclic right shift operation of this dynamic rotation offset is applied to the fixed-length semantic state vector, shifting the entire bit sequence to the right by a specified number of bits. Overflowing low-order bits are refilled into high-order bits, ensuring that the vector bit width remains constant, thus perturbing the physical arrangement of the semantic bits with the payload content. Then, the rotated result is XORed bit-by-bit with the payload content binding fingerprint itself to achieve an inseparable mathematical fusion of the rotated semantic vector and the payload fingerprint, ultimately outputting the payload binding semantic vector. This step is represented as:

[0041] in, Bind semantic vectors to the output payload; The input is a fixed-length terminal behavior semantic state vector; As a circular right shift operator, it shifts the entire bit sequence to the right by a specified number of bits, and the overflowing low-order bits are refilled into the high-order bits to ensure that the vector width remains constant. The dynamic rotation offset is determined by the fingerprint bound to the payload content, which is part of the total bit width. The rotation amount is obtained through modulo operation, ensuring that the rotation amount is always within the legal range; This is a constant representing the total bit width of the semantic state vector, which is 128 in this embodiment. A bit-by-bit XOR operator is used to achieve the inseparable mathematical fusion of the rotated semantic vector and the payload fingerprint; A fingerprint is bound to the load content generated in the previous step. In the substation scenario, the dynamic rotation offset is obtained by modulo operation of the total bit width using the fingerprint bound to the load content. This design has critical security significance. Since the dynamic rotation offset is entirely determined by the specific content of the substation control business data load, the dynamic rotation offset generated by the load content when the relay protection device performs a closing operation will inevitably be different from the dynamic rotation offset generated by the load content when the sensor performs an inspection operation. The modulo operation ensures that no matter how large the value of the fingerprint bound to the load content is, the rotation amount is always constrained within the legal range of 0 to 127 bits and will not cause out-of-bounds overflow. The cyclic right shift operation rearranges the physical position of each bit in the fixed-length semantic state vector by the rotation amount driven by the load content. Different loads will cause the same bit in the semantic vector to fall in completely different physical positions. This positional perturbation makes it impossible for an attacker to determine the true arrangement order of each bit in the original semantic vector without knowing the current load content, even if the attacker intercepts the final load-bound semantic vector. The bitwise XOR operation further integrates the numerical information of the fingerprint bound to the payload content into each bit of the semantic vector based on the rotation perturbation, so that each bit of the final output payload-bound semantic vector carries both semantic information and payload content information, which are inseparable.

[0042] In the context of countermeasures in substations, if an attacker grafts a semantic vector bound to a payload onto a message with a different payload, the fingerprint calculated by the receiver using the new payload will generate completely different rotations and XOR bases. The reverse binding restoration process will output a globally disordered pseudo-semantic vector. This pseudo-vector cannot hit any legitimate rule entries in the three-state content addressing memory of the white-box switch, thus achieving line-speed interception of grafted replay attacks at the hardware pipeline level. Specifically, after the data plane hardware pipeline of the receiving white-box switch extracts the payload binding semantic vector, it can perform inverse XOR and inverse rotation operations using the hash fingerprint of the substation control business data payload in the same message to restore the original fixed-length semantic state vector for three-state content addressing memory matching. If the payload is tampered with during transmission or the semantic vector is grafted to other messages, the payload content binding fingerprint generated by the new payload is completely different from the original payload. The rotation amount and XOR basis used in the inverse operation are fundamentally changed, and the restoration process will inevitably produce erroneous results. The matching will fail directly and trigger hardware-level message discarding or slice isolation. Thus, the semantic description and command payload in substation control communication messages are unforgeable with extremely low computational cost.

[0043] In step S323, the payload binding semantic vector is injected into the reserved bit range of the hop-by-hop option extension header skeleton to obtain a hop-by-hop option extension header carrying semantics. It should be understood that since the aforementioned steps have already generated a payload binding semantic vector strongly coupled with the substation control business data payload content through cyclic rotation XOR binding fusion calculation, this vector, as a bit sequence carrying both semantic and payload content information, is still stored in the terminal memory. It needs to be written into the reserved bit range of the hop-by-hop option extension header skeleton before it can be transmitted through the network with the message. The injection operation itself still uses bit-level bitwise operations, but the object being written has changed from the unbound original fixed-length semantic state vector to a payload binding semantic vector strongly coupled with the payload content. Therefore, in the technical solution of this application, the payload binding semantic vector is further injected into the reserved bit range of the hop-by-hop option extension header skeleton to obtain a hop-by-hop option extension header carrying semantics, thereby embedding the binding semantic bit sequence, which is inseparable from the payload content, into the fixed offset position allowed by the protocol specification to complete the message encapsulation of the second embodiment. In this way, the bit sequence carried by the hop-by-hop option extension header in the final message is no longer a static semantic block unrelated to the payload, but a bound semantic vector strongly coupled with the current message payload content. Even if an attacker intercepts the injected bit sequence in the final message, the sequence cannot be used independently of the current payload context.

[0044] In a specific example of this application, the physical starting offset of the all-zero reserved bit interval in the hop-by-hop option extension header skeleton is first located. A reserved bit mask is generated where all bits in the corresponding position of the target injection interval are one and all other bits are zero. This mask is then bitwise inverted to obtain a reverse mask where all bits in the target interval are zero instead of all bits in the target interval. Next, a bitwise AND operation is performed between the reverse mask and the hop-by-hop option extension header skeleton, forcibly clearing all bits in the target injection interval while leaving protocol control information such as the option type field and option length field unchanged. Then, a logical left shift operation is performed on the payload binding semantic vector to align it to the start position of the target bit interval. A bitwise OR operation is then performed with the cleared hop-by-hop option extension header skeleton to overwrite the bit content of the payload binding semantic vector into the cleared target interval all at once. Finally, the hop-by-hop option extension header carrying the semantics is output. This process is represented as follows:

[0045] in, Expand the header for hop-by-hop options carrying semantics in the output; Expand the head skeleton for the input jump-by-jump options; For bitwise AND operator, To preserve the bit-inverted result of the bitmask, it is used to clear the target injection interval while retaining the original data in the non-injection region; For bitwise OR operation, the aligned binding vector is overwritten into the target range that has been cleared; Bind semantic vectors to the payload generated in the previous step; This is a logical left shift operator used to align the bound vector to the beginning of the target bit range; The starting offset of the reserved bit range in the header is extended for hop-by-hop options. In the substation scenario, the execution mechanism of this injection operation at the bitwise operation level is exactly the same as the injection of the original fixed-length semantic state vector in the first embodiment. Both are completed through three steps: bitwise AND clearing, logical left shift alignment, and bitwise OR overwriting. However, its inherent security attributes have undergone fundamental changes. In the first embodiment, an original fixed-length semantic state vector completely independent of the payload is written. Attackers can extract it from legitimate messages and graft it into any forged message for continued use. In this second embodiment, a payload-bound semantic vector is written after cyclic rotation XOR fusion driven by payload content. Each bit of this vector simultaneously encodes terminal behavior semantic information and the content fingerprint information of the current payload, which are mathematically inseparable.

[0046] In summary, through the three-step binding calculation described above, the second embodiment establishes a hardware-verifiable mathematical binding relationship between the semantic vector and the service payload without increasing the overhead of extra bytes in the message or changing the hop-by-hop option extended header format specification. This is achieved by utilizing a cyclic rotation XOR operation driven by the payload content fingerprint. On the receiving side of the substation white-box switch, after the data plane hardware pipeline extracts the bound semantic vector, it can restore the original semantic state vector by performing inverse rotation and inverse XOR operations using the hash fingerprint of the payload in the same message. This vector is then used for three-state content-addressed memory matching. If the payload is tampered with during transmission or the semantic vector is grafted onto other messages, the restoration process will inevitably produce incorrect results, the matching will fail directly, and hardware-level message discarding or segmentation isolation will be triggered. This mechanism achieves the unforgeable binding of semantic descriptions and command payloads in substation control communication messages with extremely low computational cost (only one hash, one circular shift, and one XOR operation, all of which can be completed within a single clock cycle). It effectively resists the threat to power control security posed by semantic vector grafting and replay attacks, and ensures the credibility and robustness of the deterministic network slicing scheduling mechanism in adversarial environments.

[0047] Accordingly, in step S33, the Internet Protocol version 6 (IP6) basic header, the hop-by-hop options extension header carrying semantics, and the substation control service data payload are concatenated and encapsulated according to network byte order to obtain a service message carrying semantic vectors. It should be understood that since the aforementioned steps have already generated the hop-by-hop options extension header carrying semantics, at this point, the IPC basic header, the hop-by-hop options extension header carrying semantics, and the substation control service data payload are stored as independent data blocks in different buffers in the terminal memory. They have not yet formed a complete protocol data unit that can be recognized by the network interface card and sent to the physical link. Therefore, they must be strictly and orderly encapsulated and concatenated at the bitstream level according to the standard message structure of IPC version 6. Therefore, in the technical solution of this application, the basic header of Internet Protocol version 6 (IPv6), the hop-by-hop options extension header carrying semantics, and the substation control service data payload are further concatenated and encapsulated according to network byte order to obtain a service packet carrying semantic vectors. This integrates the protocol control information, semantic bearer information, and service content information scattered in different memory areas into a standard IPv6 packet with a compliant format and complete structure. This ensures that the final output service packet carrying semantic vectors fully complies with the IPv6 specification requirements in terms of format. The programmable parsers of the white-box switches along the route can sequentially identify the basic header, locate the extension header, and extract the semantic vector from it according to the fixed field offset relationship defined in the protocol standard, while ensuring that the substation control service data payload can reach the target terminal intact.

[0048] More specifically, in a concrete example of this application, the total byte length of the hop-by-hop options extension header carrying semantics and the substation control service data payload is first calculated. This total length value is then filled back into the payload length field of the Internet Protocol version 6 (IPL) base header, enabling the receiving end and along-path devices to accurately determine the total amount of data following the base header based on this field. Next, in the transmission buffer of the terminal network protocol stack, according to the network byte order (big-endian), the 40-byte fixed-length data of the IPL base header is placed at the beginning of the buffer. Fields such as version number, traffic type, flow label, payload length, next header, and hop count limit have already been filled in the aforementioned steps. Then, the hop-by-hop options extension header carrying semantics is continuously written into the buffer immediately following the IPL base header, with its physical start position strictly aligned to the 41st byte at the end of the base header. The payload-bound semantic vector carried within it is located within the option value region of the type-length-value structure in the extension header, with the offset being completely fixed and predictable relative to the first byte of the message. Furthermore, the entire byte sequence of the substation control business data payload is written to the end of the buffer immediately after the hop-by-hop option extended header carrying semantics. The three data segments form a continuous bit stream concatenated in memory. The network interface card's sending engine pushes the entire continuous buffer data as a complete Ethernet frame payload into the physical link, and finally outputs a business message carrying semantic vectors. In substation scenarios, the design of placing the basic header of the Internet Protocol version 6 (IP6) at the very beginning of the message ensures that all white-box switches within the substation can immediately read the routing information after receiving the message and trigger the hardware parsing process of the hop-by-hop option extension header based on the declaration that the next header field has a value of zero. The physical arrangement of the hop-by-hop option extension header carrying semantics immediately following it allows the data plane programmable parser to directly jump to the starting position of the extension header with the basic header fixed at a length of 40 bytes as the offset reference to extract semantic vectors without having to traverse and scan the message byte by byte. The substation control business data payload located at the end of the message ensures that the actual business content, such as relay protection closing instructions or telemetry sampling data, can be completely transparently transmitted to the target terminal for business logic processing after semantic verification and slice scheduling by the white-box switch.

[0049] Specifically, in step four, the service packets carrying semantic vectors are linearly truncated based on a fixed offset in the data plane hardware pipeline of the white-box switch to obtain fixed-length semantic state vectors. A tri-state content-addressable memory is then used to perform parallel matching and verification between the fixed-length semantic state vectors and the pre-compiled semantic state transition rule flow table. A semantic validity matching result instruction is output, and a forwarding packet stripped of the semantic vector is generated. It should be understood that since the aforementioned steps have already encapsulated the payload-bound semantic vector in the hop-by-hop option extension header of the service packet carrying the semantic vector and sent it to the substation network via the physical link, when the packet arrives at the white-box switch ingress port, the semantic vector extraction, terminal behavior validity verification, and forwarding packet reassembly need to be completed at linear speed in the data plane hardware pipeline. The entire process must be completed within nanoseconds without triggering control plane transmission; otherwise, uncontrollable queuing and processing delays will be introduced, thereby compromising the microsecond-level deterministic transmission guarantee of control services such as relay protection. Therefore, in the technical solution of this application, the service packets carrying semantic vectors are further truncated at line speed based on a fixed offset in the data plane hardware pipeline of the white-box switch to obtain a fixed-length semantic state vector. A tri-state content-addressable memory is then used to perform parallel matching and verification between the fixed-length semantic state vector and the pre-compiled semantic state transition rule flow table. A semantic validity matching result instruction is output, and a forwarding packet stripped of the semantic vector is generated. This allows for the direct completion of terminal behavior semantic extraction and verification, as well as packet forwarding preparation, at the data plane hardware level. In this way, the fixed-length field line-speed processing capability of the white-box switch's programmable switching chip can be utilized to simultaneously complete the three operations of semantic vector truncation, validity determination, and packet reassembly within a single hardware pipeline cycle, bypassing the control plane throughout the process. This ensures that the end-to-end transmission latency of control-type service packets is not affected by any additional semantic verification process.

[0050] Figure 5 This is a flowchart of step four of the semantic communication method for IPv6 terminals in a substation according to an embodiment of this application. Step four includes: S41, performing bit-level truncation of the service packet carrying the semantic vector in the programmable parser based on a fixed offset to obtain a fixed-length semantic state vector, and removing the extended header during the reassembly in the inverse parser to generate a forwarding packet stripped of the semantic vector; S42, inputting the fixed-length semantic state vector as a lookup key into the three-state content addressing memory, performing full-table concurrent XOR and mask matching with the pre-compiled semantic state transition rule flow table to obtain a hardware rule action index; S43, accessing the action data table in the static random access memory based on the hardware rule action index as an address pointer, and pulling and decoding the network control parameters of the matched entries to obtain a semantic legality matching result instruction.

[0051] More specifically, in a concrete example of this application, when a service packet carrying semantic vectors first enters from the physical port of a white-box switch, the packet bitstream is fed into the ingress pipeline buffer of the data plane programmable parser. The programmable parser internally maintains a finite state machine. The initial state of this finite state machine is Ethernet frame header parsing. After identifying the Ethernet type field in the Ethernet frame header and confirming its value as the hexadecimal value 0x8600 corresponding to IPv6, the finite state machine jumps to the Internet Protocol version 6 (IPv6) base header parsing state. In this state, the programmable parser reads the next header field fixed at the sixth byte position in the base header. Recognizing its value as zero indicates the presence of a hop-by-hop options extension header immediately following the base header, and the finite state machine then jumps to the hop-by-hop options extension header parsing state. Since the length of the Internet Protocol version 6 (IP6) base header is fixed at 40 bytes, and the option type field and option data length field in the hop-by-hop option extension header each occupy one byte, the offset from the first byte of the message (excluding the Ethernet frame header) to the starting position of the semantic vector storage area is a fixed 42 bytes, or 336 bits. Based on this fixed offset, the programmable parser directly jumps its internal read pointer to the physical start position of the reserved bit range, and uses hardware bit-level truncation instructions to continuously copy 128 bits of data from the message bit stream as a fixed-length semantic state vector, and writes it into the message header vector register inside the chip for use in the subsequent match-action pipeline stage. This truncation operation can be completed within a single clock cycle of the programmable parser. Essentially, it is a fixed-length read operation on a fixed address range of the chip's internal message buffer, without involving any loop traversal or conditional branch judgment, which fully meets the hard requirement of constant-time operation for the data face of the programmable switching chip.

[0052] Next, after the programmable parser completes the truncation of the fixed-length semantic state vector, it passes the complete message bitstream along with the parsed message header vector to the matching-action pipeline stage for processing. After the matching-action pipeline completes all entry lookups and action executions, the message enters the inverse parser for reassembly. The inverse parser determines the structure of the final output message based on a pre-programmed message reassembly template. This template explicitly specifies that the output message contains only the Internet Protocol version 6 (IP6) basic header and the substation control business data payload, excluding the hop-by-hop option extension header. During the reassembly process, the reverse parser reads forty bytes of the IPv6 base header from the message buffer, modifies the value of the next header field from zero to the protocol number corresponding to the upper-layer transport protocol, pointing to the transport layer protocol in the payload, and subtracts the number of bytes occupied by the hop-by-hop options extension header from the payload length field to reflect the actual payload length after removing the extension header. Then, it skips the byte range occupied by the hop-by-hop options extension header and directly concatenates the entire byte sequence of the substation control service data payload to the modified base header, forming a forwarded message stripped of semantic vectors. This forwarded message no longer carries any hop-by-hop options extension headers; its format is a standard IPv6 base header plus transport layer payload structure. This eliminates the need for subsequent network devices along the forwarding path to parse the already verified semantic extension headers again, reducing the processing burden on devices along the route.

[0053] Then, in the matching-action pipeline stage, the matching engine inside the chip reads the fixed-length semantic state vector from the message header vector register and sends it as the unique lookup key to the comparison array of the tri-state content-addressable memory chip. The storage array of the tri-state content-addressable memory is pre-loaded with a pre-compiled semantic state transition rule flow table compiled and issued by the control plane. Each row in this flow table corresponds to one rule, and each rule consists of a 128-bit matching feature key and a 128-bit tri-state mask. The hardware comparison circuit of the tri-state content-addressable memory performs concurrent comparison operations between the fixed-length semantic state vector and all rule entries in the storage array within a single clock cycle. Specifically, for each rule, the following operation is performed: a bit-by-bit XOR operation is performed between the fixed-length semantic state vector and the matching feature key of the rule to obtain a difference vector; then a bit-by-bit AND operation is performed between the difference vector and the tri-state mask of the rule. If the final result is an all-zero bit vector, the rule is considered to have matched. When multiple rules are matched simultaneously, the priority encoding circuit of the tri-state content-addressable memory automatically selects the matched entry with the smallest row number, i.e., the highest priority, and outputs its physical row number as the hardware rule action index. This matching process is represented as:

[0054] in, This is the output hardware rule action index; To find the minimum index value that meets the conditions, when a multiple hit occurs in a tri-state content-addressed memory, the function outputs the regular index with the highest priority, i.e., the smallest row number. The input is a fixed-length semantic state vector; This is a bitwise XOR operator. The XOR result is zero when the corresponding bits of the input key value and the rule key value are the same, and one when they are different. The matching feature key value of the i-th rule in the pre-compiled semantic state transition rule flow table is used to encode a valid terminal behavior state pattern or state transition pattern. This is the bitwise AND operator, used to apply a mask to mask differences in irrelevant bits; The three-state mask for the i-th rule in the pre-compiled semantic state transition rule flow table is as follows: a bit with a value of 1 indicates that the bit must be exactly consistent in the matching decision, and a bit with a value of 0 indicates that the bit is irrelevant, that is, whether its value is 0 or 1 does not affect the matching result. A zero-bit vector indicates that a match is successful if the fixed-length semantic state vector within the effective bit range of the mask is completely consistent with the rule key value. This is the set of all rule entries for the precompiled semantic state transition rule flow table. In the substation scenario, the precompiled semantic state transition rule flow table is pre-generated by the semantic policy compiler of the control plane based on the substation terminal behavior semantic ontology knowledge base. Each rule corresponds to a known legal terminal behavior state or a legal state transition sequence. The introduction of the three-state masking mechanism allows a single rule to cover multiple subtle variations of the same type of behavior in a wildcard manner. For example, the semantic vector of a relay protection device during the closing operation window may differ in the low-order bits due to fine-tuning of the sampling period. By setting the mask bit corresponding to the low-order bits to zero, a single rule can cover all legal variations of this type of behavior without having to establish an exact matching entry for each specific sampling parameter. This effectively controls the flow table size and reduces the storage resource occupation of the three-state content addressing memory. Bitwise XOR operation ensures that every valid bit in the fixed-length semantic state vector participates in the matching judgment, so that any bit deviation caused by payload tampering or semantic vector grafting will be detected. The highest priority hit strategy ensures that when a terminal behavior matches multiple rules of different strictness at the same time, the system always selects the rule with the strictest security constraints to respond, avoiding the risk of lenient rules being maliciously exploited.

[0055] Furthermore, the matching engine sends the output hardware rule action index as an address pointer to the static random access memory (SRAM) module via the chip's internal bus. The SRAM pre-stores an action data table that corresponds one-to-one with the rule entries in the tri-state content-addressable memory (TRM). Each storage unit contains a set of network control parameters that need to be executed when the rule is matched. After receiving the address pointer, the SRAM completes the read operation of the corresponding storage unit within a single clock cycle, pulling and outputting the preset network control parameters to the action execution engine. This parameter set includes fields such as action type identifier, quality of service queue number, physical output port number, and time-sensitive network slice priority. The action execution engine decodes the retrieved parameters. When the action type identifier indicates "allow," it encapsulates scheduling parameters such as the service quality queue number and outgoing port number into a semantic validity matching result instruction and passes it to the subsequent traffic manager for deterministic slice scheduling. When the action type identifier indicates "block and drop," the action execution engine directly sends a drop instruction to the packet buffer to remove the forwarded packet stripped of its semantic vector from the pipeline. When the action type identifier indicates "slice isolation," it remaps the packet's queue number to a low-priority best-effort slice queue for restricted forwarding. In substation scenarios, when the fixed-length semantic state vector carried in a message sent by a relay protection terminal within the closing operation window successfully matches the corresponding legal rule entry for the closing action after being matched by a three-state content-addressable memory, the action data of the corresponding storage unit in the static random access memory will indicate permission to proceed and allocate it to a high-priority deterministic time-sensitive network slice queue, ensuring that the closing control command reaches the target relay protection device with a microsecond-level deterministic delay. However, if the fixed-length semantic state vector is tampered with, causing a global error when the receiving side performs reverse binding restoration, or if an attacker grafts the semantic vector of other messages onto the current message, resulting in a discrepancy between the vector content and the actual content, the error will be mitigated. When all valid patterns in the rule base do not match, the tri-state content addressing memory will output a default miss flag. At this time, the hardware rule action index points to the preset default action entry in the static random access memory action data table. The type of the default action is identified as blocking / dropping or slice isolation. Based on this, the action execution engine immediately removes the abnormal message from the forwarding pipeline or downgrades it to the isolation slice, thereby achieving line-speed interception of abnormal messages at the hardware pipeline level. The entire matching verification and action execution process is completed inside the data plane chip without generating any control plane interruption requests, ensuring that the deterministic transmission of substation control services is not affected by the delay of the security verification process.

[0056] Specifically, in step five, deterministic network slicing scheduling is performed on the forwarded packets stripped of semantic vectors based on the semantic legitimacy matching result instruction to output a deterministic slice forwarding data stream. It should be understood that since the preceding steps have already completed hardware-level legitimacy verification of terminal behavior semantics and output semantic legitimacy matching result instructions and forwarded packets stripped of semantic vectors, the action type identifier and quality of service parameters carried in the semantic legitimacy matching result instruction have not yet been transformed into hardware scheduling signals that can be directly recognized by the backplane bus and outgoing port traffic manager of the white-box switch. The forwarded packets stripped of semantic vectors remain in the output buffer of the matching-action pipeline, waiting for the scheduling instruction to guide them into the correct physical outgoing port queue. If the semantic verification result is not strongly bound to the packet forwarding scheduling mechanism, high-priority control packets that have passed legitimacy verification may be assigned to a non-deterministic best-effort path for transmission, thus losing deterministic latency guarantees. Therefore, in the technical solution of this application, deterministic network slice scheduling and forwarding are further performed on the forwarded packets stripped of semantic vectors based on the semantic legitimacy matching result instruction to output a deterministic slice forwarding data stream. This allows the result of terminal behavior semantic verification to directly drive the hardware scheduling decision of network slices, ensuring that control packets that have passed the legitimacy verification are accurately diverted to the time-sensitive network slice queue with microsecond-level latency guarantee for deterministic forwarding. In this way, end-to-end closed-loop linkage between terminal behavior semantic verification and deterministic network slice scheduling can be achieved, ensuring that only packets that pass the semantic legitimacy verification can obtain the allocation of high-priority deterministic transmission resources. Abnormal packets have been intercepted or isolated in the aforementioned steps and will not occupy the limited bandwidth resources of deterministic slices, thereby ensuring that control services such as substation relay protection maintain microsecond-level end-to-end deterministic latency indicators while the semantic security protection mechanism is operating.

[0057] More specifically, in this embodiment, step five includes: parsing the legality identifier and service level parameters in the semantic legality matching result instruction, mapping them to the outgoing port number, time-sensitive queue number, and backplane transmission priority, and structurally packaging them to generate hardware scheduling metadata; concatenating the hardware scheduling metadata in the form of a pre-addition code before the header of the forwarding packet stripped of the semantic vector, synthesizing an internal switching in-band metadata packet that can be identified and transmitted on the backplane bus of the switch; the traffic manager reads the scheduling instruction at the top of the internal switching in-band metadata packet and directs it to the time-sensitive network slice queue, stripping all internal metadata when sending at the outgoing port and popping it at line speed to output a deterministic slice forwarding data stream.

[0058] Specifically, in a concrete example of this application, the action execution engine in the white-box switch's data plane pipeline first parses the legality identifier and service level parameters encapsulated in the semantic legality matching result instruction. The legality identifier indicates whether the current packet has passed the terminal behavior semantic legality verification, and the service level parameter indicates the substation service priority category to which the current packet belongs. The action execution engine maps the legality identifier and service level parameters to three hardware-level scheduling parameters based on a preset resource mapping configuration table: the outgoing port number determines which physical interface of the white-box switch the packet will be sent from to reach the network segment where the target terminal is located; the time-sensitive queue number determines which time-sensitive network gating queue with deterministic latency guarantee the packet should enter at that outgoing port; and the backplane transmission priority determines the preemption right level of the packet when passing through the switch's internal switching network board. The action execution engine structures and packages the above three scheduling parameters according to the internal data structure specifications of the switching chip, generating fixed-byte-length hardware scheduling metadata. The format of this metadata strictly matches the frame header identification format of the switch's internal backplane bus to ensure that the backplane controller can directly interpret the scheduling instructions within. In a substation scenario, when a relay protection device performs a closing operation, its service level parameter is identified as the highest priority control service after passing semantic verification. The time-sensitive queue number generated by the mapping points to the gated scheduling queue reserved for relay protection services in the outgoing port. The backplane transmission priority is set to the highest preemption level to ensure that the message can preempt the transmission time slots of all low-priority traffic during the transmission process on the switching network board.

[0059] Next, the packet reassembly engine inside the white-box switch appends the hardware scheduling metadata as a pre-addition code to the header of the semantic vector-stripped forwarding packet, synthesizing an internal switching in-band metadata packet. Specifically, the packet reassembly engine allocates a new contiguous storage space in the chip's internal buffer, writes the fixed-byte length hardware scheduling metadata to the beginning of this storage space, and then continuously writes the entire byte sequence of the semantic vector-stripped forwarding packet. The two data segments form a continuous bit stream in physical memory, constituting an internal switching in-band metadata packet that can be recognized and transmitted on the switch's internal backplane bus. This internal switching in-band metadata packet is the standard carrier format for transmitting packets and scheduling information between various functional modules of the data plane inside the white-box switch. The hardware scheduling metadata at the top is used by the backplane controller and flow manager to read scheduling instructions, while the semantic vector-stripped forwarding packet at the end is the actual network packet that ultimately needs to be sent from the physical port. In substation scenarios, the design of the pre-addition code allows the flow manager to obtain all scheduling information by reading only the metadata of a fixed length at the beginning of the message when it receives the metadata message in the internal switching band. No parsing operation is required on the subsequent actual business message content, thereby minimizing the time overhead of scheduling decisions.

[0060] Then, after the white-box switch's traffic manager receives the internal switching in-band metadata packet from the internal switching network board, it reads the outgoing port number from the top hardware scheduling metadata and directs the packet to the queue scheduling module of the corresponding physical outgoing port. The queue scheduling module further reads the time-sensitive queue number and pushes the packet into the designated time-sensitive network gating queue to wait for transmission. The time-sensitive network gating queue opens and closes the transmission gating at fixed time intervals according to the pre-configured gating scheduling list. When the transmission gating is opened within the preset time window, the packet at the head of the queue is popped and sent to the transmission engine of the outgoing port. In the substation scenario, the time window configuration of the gating scheduling list is strictly aligned with the communication cycle of the substation relay protection business, ensuring that the closing control command is sent within the deterministic time window of the gating opening, and the upper limit of its end-to-end transmission delay is strictly constrained to the microsecond range and the jitter is controlled at the nanosecond level.

[0061] Furthermore, when a packet is popped from the time-sensitive network gating queue and arrives at the sending engine of the physical output port, the sending engine automatically identifies and strips the hardware scheduling metadata at the very beginning of the packet, pushing only the subsequent forwarded packet, stripped of its semantic vector, into the physical medium interface layer for Ethernet frame encapsulation and line-rate transmission. The hardware scheduling metadata, after being completely removed from the output port, will not appear in the transmitted packets on the physical link; its scope is strictly limited to the transmission of scheduling information between modules within the data plane of the white-box switch. After being popped at line rate by the output port sending engine, a deterministic slice forwarding data stream is finally output. Each packet in this data stream is a standard IPv6 packet that has undergone terminal behavior semantic validity verification and has had all internal scheduling overhead stripped, reaching the target terminal device via the substation's fiber optic link with deterministic latency and jitter metrics. In substation scenarios, deterministic slice forwarding data streams carry critical control-related business content such as relay protection closing commands and telemetry sampling data that have been verified by semantic security protection. After receiving the messages in the data stream, the target relay protection equipment can directly extract the substation control-related business data payload for business logic processing. The entire semantic communication process forms a complete end-to-end closed loop from semantic modeling, dimensionality reduction, and encapsulation on the terminal side to interception, verification, scheduling, and forwarding on the network side. Without increasing any control plane processing latency, it simultaneously achieves continuous semantic protection of terminal behavior and microsecond-level deterministic transmission guarantee for control-related business.

[0062] In summary, the semantic communication method for IPv6 terminal behavior in substations according to the embodiments of this application is explained. It introduces graph neural network embedding dimensionality reduction technology to construct a semantic ontology state graph from the multidimensional operating parameters of the terminal. This graph is then compressed into a fixed-length semantic state vector with a fixed bit width through graph convolution aggregation and hash discretization, eliminating the variable-length attribute of semantic features. Based on this, the fixed-length vector is injected bit-by-bit into the reserved alignment spare bits of the IPv6 hop-by-hop option extension header to achieve implicit in-path communication. This allows the white-box switch's programmable pipeline on the data plane to complete the line-speed extraction of the semantic vector based on a fixed offset without variable-length parsing. Nanosecond-level parallel matching and verification are performed using a tri-state content-addressable memory and a pre-compiled rule flow table, directly determining the legality of terminal behavior at the hardware layer. The matching result drives deterministic network slice scheduling, bypassing the control plane throughout, thereby simultaneously achieving continuous protection of terminal behavior semantics and microsecond-level deterministic transmission guarantee for control-type services without introducing additional upload latency.

[0063] The various embodiments of this disclosure have been described above. These descriptions are exemplary and not exhaustive, nor are they limited to the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments. The terminology used herein is chosen to best explain the principles, practical application, or improvement of the technology in the market, or to enable others skilled in the art to understand the embodiments disclosed herein.

Claims

1. A semantic communication method for IPv6 terminals in substations, characterized in that, include: Step 1: Based on the multi-dimensional operating parameters of the terminal, perform graph-structured fusion modeling of the node attribute features and communication edge relationships of the terminal to obtain the terminal behavior semantic ontology state graph. Step 2: Perform terminal behavior semantic dimensionality reduction and feature extraction based on graph neural networks on the terminal behavior semantic ontology state graph to obtain a fixed-length semantic state vector. Step 3: Inject the fixed-length semantic state vector bit by bit into the reserved alignment spare bits of the hop-by-hop option extension header of the Internet Protocol version 6 (IP6), and concatenate and encapsulate it with the substation control business data payload to obtain a business message carrying the semantic vector. Step 4: In the white-box switch data plane hardware pipeline, the service packets carrying semantic vectors are truncated at line speed based on a fixed offset to obtain fixed-length semantic state vectors. The fixed-length semantic state vectors are then matched and verified in parallel with the pre-compiled semantic state transition rule flow table using a three-state content addressing memory. The semantic legality matching result instruction is output and a forwarding packet stripped of semantic vectors is generated. Step 5: Based on the semantic legitimacy matching result, perform deterministic network slice scheduling and forwarding on the forwarded packets stripped of semantic vectors to output a deterministic slice forwarding data stream.

2. The substation IPv6 terminal behavior semantic communication method according to claim 1, characterized in that, The terminal's multi-dimensional operating parameters include the relay protection equipment's operating status code, sensor sampling period interval, terminal network communication port number, and historical access control permission identifier.

3. The semantic communication method for IPv6 terminal behavior in substations according to claim 2, characterized in that, Step one includes: The status codes of relay protection equipment are vectorized and then concatenated with the permission vectors of historical access control permission identifiers along the channel dimension to obtain the terminal node attribute feature matrix. A scalar estimation of the time decay factor is performed on the sensor sampling period interval, and spatiotemporal feature joint mapping is performed with the high-dimensional embedding vector of the terminal network communication port number to obtain the terminal communication edge relationship matrix. The terminal node attribute feature matrix is ​​used as the vertex feature set, and the terminal communication edge relationship matrix is ​​used as the directed edge feature set. Entity relationship alignment and graph representation fusion are performed based on the substation physical network connection adjacency matrix to obtain the terminal behavior semantic ontology state graph.

4. The semantic communication method for IPv6 terminal behavior in substations according to claim 1, characterized in that, Step two includes: Local high-frequency adjacency sampling and topological subgraph extraction are performed on the terminal behavior semantic ontology state graph to obtain a local semantic topological subgraph. A high-dimensional continuous semantic feature tensor is obtained by performing multi-layer feature aggregation on local semantic topological subgraphs through multi-layer graph convolutional networks. We perform feature dimensionality reduction compression and discretization hashing on high-dimensional continuous semantic feature tensors to obtain fixed-length semantic state vectors.

5. The semantic communication method for IPv6 terminal behavior in substations according to claim 1, characterized in that, Step three includes: Initialize and generate the sixth version of the Internet Protocol Basic header, and construct a hop-by-hop option extended header skeleton based on the type-length-value format, and allocate a range of all-zero reserved bits in the skeleton with a width equal to that of the fixed-length semantic state vector; Offset injection overlay is applied to the fixed-length semantic state vector, and the hop-by-hop option extended header skeleton is converted into a hop-by-hop option extended header carrying semantics; The Internet Protocol version 6 (IP6) basic header, the hop-by-hop options extension header with semantics, and the substation control business data payload are concatenated and encapsulated in network byte order to obtain a business message carrying semantic vectors.

6. The semantic communication method for IPv6 terminals in substations according to claim 1, characterized in that, Step four includes: Based on a fixed offset, the programmable parser performs bit-level truncation on the service packets carrying semantic vectors to obtain fixed-length semantic state vectors, and removes the extended header during the reassembly in the inverse parser to generate forwarding packets stripped of semantic vectors. The fixed-length semantic state vector is used as the lookup key value and input into the three-state content addressing memory. It is then subjected to full table concurrent XOR and mask matching with the pre-compiled semantic state transition rule flow table to obtain the hardware rule action index. The action data table in the static random access memory is accessed based on the hardware rule action index as the address pointer. The network control parameters of the hit table entries are pulled and decoded to obtain the semantic legality matching result instruction.

7. The semantic communication method for IPv6 terminal behavior in substations according to claim 1, characterized in that, Step five includes: The legality identifier and service level parameters in the semantic legality matching result instruction are parsed and mapped to the outgoing port number, time-sensitive queue number and backplane transmission priority, and then structured and packaged to generate hardware scheduling metadata. The hardware scheduling metadata is appended to the header of the forwarded message after the semantic vector has been stripped, in the form of a pre-addition code, to synthesize an internal switching in-band metadata message that can be recognized and transmitted on the switch's internal backplane bus. The traffic manager reads the scheduling instructions at the top of the internal switching in-band metadata packets and directs them to the time-sensitive network slice queue. When sending at the outgoing port, all internal metadata is stripped and popped at line speed to output a deterministic slice forwarding data stream.

8. The semantic communication method for IPv6 terminals in substations according to claim 5, characterized in that, Offset injection overlay is applied to the fixed-length semantic state vector, and the hop-by-hop option extended header skeleton is converted into a hop-by-hop option extended header carrying semantics, including: Perform equal-width fixed-length hash compression on the entire byte sequence of substation control business data payload to obtain a payload content binding fingerprint with the same bit width as the fixed-length semantic state vector. Based on the dynamic rotation offset calculated by taking the modulo of the total bit width using the payload content binding fingerprint, a cyclic right shift operation is applied to the fixed-length semantic state vector, and then bit-by-bit XOR fusion is performed with the payload content binding fingerprint to obtain the payload binding semantic vector. The payload binding semantic vector is injected into the reserved bit range of the hop-by-hop option extension header skeleton to obtain the hop-by-hop option extension header carrying semantics.

Images