IPv6 white box network construction system for smart substation

Abstract

This application provides an IPv6 white-box network construction system for smart substations, relating to the field of power communication network technology. It utilizes IPv6 flow tag fields to establish hardware-level identification credentials for services of different priorities at the chip entry point, while embedding network element status information in a folded and compressed form into the IPv6 hop-by-hop extended header for synchronous transmission with the packet. Furthermore, lightweight congestion prediction inference is pushed down to the white-box switch control plane to quantify link risks in real time. If the risk exceeds a safety threshold, a high-priority control flow is bypassed through an event-driven mechanism to free up the chip parsing cycle, and a pre-calculated backup path is issued based on segmented routing, achieving dynamic decoupling of telemetry tasks and critical services at the data plane micro-instruction level. Finally, based on network-wide execution feedback, cascaded sensing topology aggregation and risk assessment are performed, and telemetry acquisition is resumed after the risk dissipates, forming a closed-loop self-healing mechanism.

Description

Technical Field

[0001] This application relates to the field of power communication network technology, and more specifically, to an IPv6 white-box network construction system for smart substations. Background Technology

[0002] With the accelerated advancement of smart substation transformation, the types and quantities of communication services within the substation have increased dramatically due to intelligent inspection, digital operations, and the access of massive numbers of sensing terminals. Multi-granularity services such as protection and control, sampling, and video surveillance place extremely stringent demands on the deterministic latency and reliability of network transmission. Traditional IPv4-based substation communication networks are struggling to adapt to these evolving trends in terms of address space, protocol extension capabilities, and the ability to support differentiated multi-service capabilities. Software-defined network architectures based on white-box switches, with their advantages of hardware-software decoupling, programmability, and low cost, are gradually becoming the core direction for the next stage of smart substation communication network evolution. Against this backdrop, building an IPv6 white-box network for smart substations, enabling the network to possess inherent intelligent capabilities for sensing, analysis, and decision-making, has become an urgent need to ensure the real-time performance and deterministic communication quality of substation relay protection services.

[0003] However, given the limited hardware resources of white-box switches, there is a severe conflict between the in-band network telemetry tasks upon which intrinsic intelligence relies and the critical control services of substations, resulting in a significant competition for underlying chip forwarding resources. Specifically, to achieve holistic network status awareness, white-box devices need to perform IPv6 extended header parsing and encapsulation operations at high frequency. These operations consume a significant amount of parsing pipeline cycles and static random access memory resources within the switching chip. Meanwhile, the tolerance for latency jitter in substation protection and control messages such as GOOSE and sampled values ​​is close to zero, yet they share the same hardware processing pipeline with telemetry encapsulation traffic at the chip entry point. Existing outgoing port queue priority scheduling mechanisms can only perform scheduling at the outgoing end and cannot implement intelligent identification and forced resource yielding the instant a message enters the chip parsing pipeline. This leads to critical control commands being blocked at the chip entry point during telemetry data bursts or AI sampling, causing irreversible deterministic latency degradation. The root cause of this resource competition problem lies in the lack of an effective coordination mechanism between the flexibility requirements of white-box network operating systems and the rigid physical constraints of commercial switching chip pipelines. This results in an irreconcilable system-level conflict between improving network intelligent sensing capabilities and ensuring the absolute deterministic transmission of critical services.

[0004] Therefore, an optimized IPv6 white-box network construction system for smart substations is desired. Summary of the Invention

[0005] To address the aforementioned technical problems, this application is proposed. Embodiments of this application provide an IPv6 white-box network construction system for smart substations, comprising: The service flow perception and flow label mapping module is used to extract protocol features and determine priority of the acquired original message flow of substation secondary equipment, and map the service priority features to the IPv6 flow label field to obtain the substation service flow and service classification metadata with labels. The telemetry acquisition and folded header encapsulation module is used to acquire the status of the chip queue depth and port rate of each network element through which the tagged substation business flow passes, and to compress and encapsulate the multi-dimensional status indicators into a folded hop-by-hop extended header to obtain a status-coded extended header message. The intrinsic intelligent reasoning and risk quantification module is used to predict the link congestion probability of state-coded extended header messages based on a lightweight inference engine deployed on the control plane of the white-box switch in order to obtain the real-time congestion risk coefficient. The resource preemption and deterministic path reconstruction module is used to force the high-priority control flow in the tagged substation business flow to bypass the telemetry encapsulation pipeline to release the chip parsing cycle when the real-time congestion risk coefficient exceeds the preset safety threshold, and to issue a pre-calculated backup path for the high-priority control flow and generate a preemption forwarding instruction. The global topology aggregation and self-healing recovery module is used to aggregate the path switching status of all white-box network elements and refresh the global topology based on the execution feedback of the preemptive forwarding command, and to restore the telemetry acquisition task in the risk-eliminated network segment to obtain the self-healing network state diagram.

[0006] Compared to existing technologies, the IPv6 white-box network construction system for smart substations provided in this application utilizes IPv6 flow label fields to establish hardware-level identification credentials for services of different priorities at the chip entry point. Simultaneously, it embeds network element status information in a folded and compressed form into the IPv6 hop-by-hop extended header and transmits it synchronously with the packet, avoiding packet fragmentation and bandwidth storms caused by telemetry data. Based on this, lightweight congestion prediction inference is pushed down to the white-box switch control plane to quantify link risks in real time. Once the risk exceeds the safety threshold, an event-driven mechanism forces high-priority control flows to skip the telemetry encapsulation pipeline to release the chip parsing cycle, and pre-calculated backup paths are issued based on segmented routing, achieving dynamic decoupling of telemetry tasks and critical services at the data plane micro-instruction level. Finally, based on the network-wide execution feedback, cascaded topology aggregation and risk assessment are performed, and telemetry acquisition is resumed after the risk dissipates, forming a closed-loop self-healing system. This solution requires no additional hardware computing power, and by collaboratively approaching from the protocol layer and data plane scheduling layer, it achieves intrinsic intelligent perception of network status while ensuring zero packet loss and low latency control. 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 1 This is a system block diagram of an IPv6 white-box network construction system for smart substations according to an embodiment of this application; Figure 2 This is a schematic diagram of the data flow in an IPv6 white-box network construction system for smart substations according to an embodiment of this application; Figure 3 This is a block diagram of the resource preemption and deterministic path reconstruction module in the IPv6 white-box network construction system for smart substations according to an embodiment of this application. Figure 4 This is a block diagram of the global topology aggregation and self-healing recovery module in the IPv6 white-box network construction system for smart substations according to an embodiment of this application. Figure 5 This is a block diagram of a cascaded sensing risk retention estimation unit in an IPv6 white-box network construction system for smart substations 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 IPv6 white-box network of a smart substation, the high-frequency in-band telemetry operations required for intrinsic intelligence and protection and control services such as GOOSE / SV face irreconcilable resource contention at the chip parsing pipeline level. Existing outgoing port scheduling mechanisms cannot yield resources the instant packets enter the chip, causing irreversible degradation of deterministic latency for critical services during telemetry bursts. To address this issue, a system for constructing an IPv6 white-box network for smart substations is proposed. First, a programmable data plane parser extracts protocol features from the original packet streams of the substation's secondary equipment, mapping service priorities into IPv6 flow tag fields. This allows the switching chip to quickly identify the differences between control and sensing flows at the entry point. Then, when each network element triggers in-band telemetry sampling, a folding compression algorithm encodes multi-dimensional status indicators such as queue depth and port utilization into a fixed-length hop-by-hop extended header, transmitting network status information synchronously with service packets with minimal byte overhead. Finally, a shallow perceptron model deployed on the control plane fuses physical layer status and service layer priorities. Congestion probability prediction is performed, and a real-time risk coefficient is output. When the risk coefficient exceeds the safety threshold, the data plane immediately initiates event-driven resource preemption logic, forcibly bypasses the telemetry encapsulation pipeline for high-priority control flow, and injects pre-calculated segmented route backup paths into it. The telemetry bypass command and rerouting command are atomically combined into a single composite forwarding command and handed over to the chip's underlying layer for execution. Finally, a cascaded propagation adjacency matrix is ​​constructed based on the network-wide execution feedback to quantify the traffic injection intensity of upstream network segment preemption events on downstream shared network segments. The enhanced comprehensive risk retention rate guides the determination of telemetry recovery timing. After the risk dissipates, telemetry acquisition is restarted in an orderly manner and the global topology is refreshed, forming a complete closed loop from perception, decision-making, execution to self-healing.

[0011] Figure 1 This is a system block diagram of an IPv6 white-box network construction system for smart substations according to an embodiment of this application. Figure 2 This is a schematic diagram of data flow in an IPv6 white-box network construction system for smart substations according to an embodiment of this application. Figure 1 and Figure 2As shown, the IPv6 white-box network construction system 100 for smart substations according to an embodiment of this application includes: a service flow perception and flow label mapping module 110, used to extract protocol features and determine priorities from the acquired original message flow of substation secondary equipment, and map the service priority features to the IPv6 flow label field to obtain tagged substation service flows and service classification metadata; a telemetry acquisition and folded header encapsulation module 120, used to collect the status of the chip queue depth and port rate of each network element through which the tagged substation service flow passes, and compress and encapsulate the multi-dimensional status indicators into a folded hop-by-hop extended header to obtain a status-coded extended header message; and an intrinsic intelligent reasoning and risk quantification module 130, used for deployment-based... A lightweight inference engine in the white-box switch control plane performs link congestion probability prediction on state-coded extended header messages to obtain a real-time congestion risk coefficient. A resource preemption and deterministic path reconstruction module 140 is used to forcibly bypass the telemetry encapsulation pipeline for high-priority control flows in tagged substation service flows to release chip parsing cycles when the real-time congestion risk coefficient exceeds a preset safety threshold. It also issues pre-calculated backup paths for high-priority control flows and generates preemption forwarding instructions. A global topology aggregation and self-healing recovery module 150 is used to aggregate the path switching status of all white-box network elements and refresh the global topology based on the execution feedback of the preemption forwarding instructions. It also restores telemetry acquisition tasks in risk-eliminating network segments to obtain a self-healing network state diagram.

[0012] In the aforementioned IPv6 white-box network construction system 100 for smart substations, the service flow perception and flow label mapping module 110 is used to extract protocol features and determine priorities from the acquired raw message flows of substation secondary equipment, and map the service priority features to IPv6 flow label fields to obtain tagged substation service flows and service classification metadata. It is worth noting that here, the raw message flows of substation secondary equipment include protection control messages, sampled value messages, video surveillance data, and sensor monitoring data. It is understandable that in the IPv6 white-box network of a smart substation, the raw message flow of the substation's secondary equipment simultaneously contains multiple service types, including protection and control messages, sampled value messages, video surveillance data, and sensor monitoring data. The commercial ASIC chips of the white-box switch perform uniform processing on all arriving messages without differentiation at the entry parsing pipeline. The chips themselves lack the native ability to distinguish between critical control flows and non-critical sensing flows at the pipeline entry point. This directly leads to the situation where, when in-band telemetry encapsulation tasks suddenly occupy the parsing cycle, protection and control messages with zero latency tolerance, such as GOOSE and sampled values, are also blocked in the entry queue and cannot receive priority processing. Therefore, in the technical solution of this application, protocol features are extracted and priority is determined from the acquired raw message flow of the substation's secondary equipment. The service priority features are then mapped to the IPv6 flow tag field to obtain tagged substation service flows and service classification metadata. This allows for the establishment of hardware-recognizable service identity credentials for each message at the chip entry point. This enables the subsequent telemetry acquisition module and resource preemption module to immediately distinguish between high-priority control flow and best-effort sensing flow based on the flow tag field under data plane line-rate conditions. This provides a preliminary classification criterion for event-driven resource concession and deterministic path reconstruction, fundamentally eliminating the risk of critical business latency degradation caused by indiscriminate chip processing.

[0013] Specifically, in this embodiment, the service flow perception and flow label mapping module includes: a message feature extraction and caching unit, used to perform parallel feature extraction on the original message flow of the substation secondary equipment through a programmable data plane parser to obtain the Ethernet type field, application identifier field, virtual LAN priority label and message payload length, and to concatenate the message multidimensional feature matrix according to a fixed-dimensional tensor, while storing the original message payload in the chip entry static random access memory to form a cached original message flow; a service priority determination unit, used to perform intelligent service priority determination on the message multidimensional feature matrix to obtain service classification metadata; and a flow label injection and encapsulation unit, used to retrieve the cached original message flow from the chip memory, and write the category identifier in the service classification metadata into the flow label field of the corresponding message IPv6 header to generate a tagged substation service flow.

[0014] More specifically, in a concrete example of this application, firstly, when the raw message stream from the substation secondary equipment arrives at the white-box switch inlet, the programmable data plane parser performs bit-level parallel feature extraction on each data packet in the message stream. This programmable data plane parser, based on parsing logic written in P4, runs at the first stage of the white-box switch inlet pipeline. Its parsing state machine is configured to start from a fixed offset position in the Ethernet frame header and sequentially read bytes 12 to 13 to obtain the Ethernet type field. A value of 0x88B8 indicates that the message is an IEC 61850 GOOSE message; a value of 0x88BA indicates that the message is a sampled value message; and values ​​of 0x0800 or 0x86DD indicate IPv4 or IPv6 encapsulated video surveillance data and sensor monitoring data, respectively. After extracting the Ethernet type field, the parser continues to shift towards the application protocol data unit area in the message payload direction, reading the fixed two-byte application identifier field. This field uniquely identifies the specific logical node and control block instance in the IEC 61850 protocol system. For protection control messages, its value ranges from 0x0000 to 0x3FFF, and for sampled value messages, its value ranges from 0x4000 to 0x7FFF. The parser simultaneously extracts the three-bit priority code point from the outer VLAN tag as the virtual LAN priority tag. This value is pre-configured in the substation network according to the IEC 62439-3 standard as follows: Level 6 or 7 corresponds to relay protection services; Level 4 or 5 corresponds to process layer sampled value services; and Levels 0 to 3 correspond to monitoring and management services. Finally, the parser obtains the packet payload length from the payload length field of the IPv6 basic header. This value is in bytes; for a typical GOOSE trip message, this value is 60 to 150 bytes; for a message with 1920 sampling points, this value is approximately 1200 bytes; and for a high-definition video surveillance frame, this value is close to the maximum transmission unit limit of 1400 bytes. These four features are extracted in parallel within a single pipeline traversal cycle of the parser, consuming no more than one-tenth of the total depth of the entry parsing pipeline. After extraction, the parser arranges the Ethernet type field, application identifier field, VLAN priority tag, and packet payload length in a fixed-dimensional order and performs tensor concatenation to generate a four-dimensional packet multi-dimensional feature matrix. Its data structure is a fixed-length metadata vector, stored in the metadata bus register group output by the parser.Simultaneously with feature extraction, the parser writes the complete, unmodified raw packet data—including the complete Ethernet frame header, IPv6 basic header, and all payload content—to the static random access memory (SRAM) on the switching chip's ingress side via the bypass write channel of the ingress pipeline. The storage space occupied by each packet is dynamically allocated based on its actual frame length. The memory controller assigns a unique memory pointer to each cached entry and appends this pointer to the metadata bus, thus forming a cached stream of raw packets. This caching mechanism ensures that raw packets are not discarded due to pipeline congestion during subsequent priority determination and flow tag injection operations. Simultaneously, the memory pointer mechanism guarantees that subsequent modules can accurately locate and retrieve specific packets for editing operations.

[0015] Next, a service priority intelligent determination operation is performed on the message multidimensional feature matrix to generate service classification metadata. This determination operation is performed in the matching action table stage of the white-box switch data plane, specifically implemented as a pre-configured weighted scoring table, which pre-stores weight parameters corresponding to various services in the substation. The specific process of the determination operation is as follows: the value of the Ethernet type field is read from the message multidimensional feature matrix and the weight mapping dictionary is queried. When the Ethernet type field value is 0x88B8, the mapped Ethernet type weight value is 1.0; when the value is 0x88BA, it is mapped to 0.9; when the value is 0x86DD and the destination port points to the video streaming protocol, it is mapped to 0.2; and when the value is 0x86DD and the destination port points to the sensor data aggregation port, it is mapped to 0.15. Similarly, the application identifier weight value is queried from the application identifier field. The application identifier for protection tripping is mapped to 1.0, the application identifier for process layer sampling is mapped to 0.7, and the application identifier for management information is mapped to 0.1. The Ethernet type weight value is multiplied by a preset Ethernet type weight coefficient of 0.35, and the application identifier weight value is multiplied by a preset application identifier weight coefficient of 0.40. These two are then summed, and the result of dividing the VLAN priority label by its maximum value (7) and multiplying by the VLAN priority weight coefficient of 0.15 is added. Finally, the product of the load attenuation compensation coefficient of 0.10 and a negative exponential function with a natural constant as the base and a length attenuation constant of 0.003 multiplied by the packet payload length is added. The calculation of the priority score is expressed as follows:

[0016] in, This represents the calculated business priority score, which is a continuous value between zero and one. The weighting factor for the Ethernet type is shown; in this example, it is set to 0.35. This represents the Ethernet type weight value obtained by matching the Ethernet type field against a preset dictionary, with a value range of zero to one. This represents the weighting coefficient of the application identifier, which is 0.40 in this example; This represents the application identifier weight value obtained based on the application identifier field, with a value ranging from zero to one. The weighting coefficient represents the priority of the virtual LAN, and in this example, it is set to 0.15; This represents the extracted virtual LAN priority tag, with values ​​ranging from zero to seven. This represents the load attenuation compensation coefficient, which is 0.10 in this example. This represents the length attenuation constant, which is 0.003 in this example; This indicates the message payload length, in bytes. In this formula, the first two terms... This constitutes a static priority base based on protocol type and application identifier. A protection trip message with Ethernet type IEC 61850 GOOSE, after these two calculations, can obtain a basic score of 0.35×1.0+0.40×1.0=0.75, while a video surveillance message only obtains a basic score of 0.35×0.2+0.40×0.1=0.11. The significant score difference between the two ensures that protection control services maintain absolute priority under any competitive conditions. (Third item) The VLAN priority code points pre-configured according to the IEC 62439-3 standard in the substation network are introduced into the IPv6 layer evaluation framework. When the VLAN priority label of the protection control message is seven, the contribution value of this item is 0.15 × 7 / 7 = 0.15, realizing cross-layer connectivity from Layer 2 priority semantics to Layer 3 flow labels. (Fourth item) By applying a score reduction mechanism to messages with long payloads using an exponential decay method, and substituting the payload length of a GOOSE trip message of 80 bytes into the calculation, a score of 0.10×e is obtained. -0.003×80 =0.10 × 0.787 = 0.079, while substituting the effective payload length of the video surveillance frame as 1400 bytes, we get 0.10 × e -0.003×1400=0.10×0.015=0.0015. This attenuation term makes the contribution of the large packet sensing service in this dimension approach zero, thus effectively locking its comprehensive score in the low priority range. Combining the above four calculations, the comprehensive priority score of a typical GOOSE protection trip message is 0.75+0.15+0.079=0.979, while the comprehensive priority score of a video surveillance frame is 0.11+0.064+0.0015=0.176, with a score difference of more than five times. After the judgment calculation is completed, the comprehensive priority score is compared with the preset classification threshold. Messages with a score greater than or equal to 0.80 are marked as high priority control class, messages with a score between 0.40 and 0.80 are marked as medium priority sampling class, and messages with a score less than 0.40 are marked as best-effort sensing class. The above category identifiers and score values ​​are jointly structured and encapsulated to generate service classification metadata. This metadata is attached to the metadata bus in the form of a fixed eight-byte length structure and maintains a one-to-one binding relationship with the memory pointer of the corresponding message.

[0017] Then, the message editing pipeline of the switching chip is used to perform flow label injection and encapsulation operations to generate tagged substation service flows. The message editing pipeline is located after the matching action table stage of the white-box switch's inlet pipeline and before the outlet queue scheduling stage. This pipeline receives the memory pointer and service classification metadata transmitted in the metadata bus, and sequentially retrieves the corresponding message entities from the cached original message flow based on the memory pointer. The message editing pipeline reads the service category identifier from the service classification metadata, encodes this identifier as a 20-bit wide value, and writes it to the corresponding position in the flow label field of the corresponding message's IPv6 basic header. The IPv6 flow label field is located in the lower 20 bits of the first to third bytes of the IPv6 header. Its original value is all zeros in the original message stream of the substation secondary equipment. The message editing pipeline fills in the category identifier into this field through a bit overwrite operation. The high-priority control class is encoded as 0xFFFFF (all 20 bits are one), indicating the highest priority identity. The medium-priority sampling class is encoded as 0x80000 (the highest bit is one, and the rest are zeros). The best-effort awareness class is encoded as 0x00001 (the least significant bit is one). After the encoding is written, the message editing pipeline recalculates the hop-by-hop options check fields in the IPv6 header to ensure message validity. Then, the message with the flow label injected is pushed into the downstream transmission pipeline to form a tagged substation service flow.

[0018] Furthermore, the tagged substation service flow carries a 20-bit stream tag credential that can be identified by single-cycle line-rate mask matching of the three-state content-addressable memory of the switching chip. This credential will be used in subsequent modules to identify high-priority control messages that require forced bypass of the telemetry encapsulation pipeline within nanosecond-level time accuracy. Simultaneously, the priority score value carried in the service classification metadata will be concatenated across layers with the underlying physical state vector in the endogenous intelligent inference module. After extreme value scaling, it will be used as input features for the shallow perceptron model to participate in congestion probability prediction calculations. The priority sensitivity factor will dynamically amplify the risk perception sensitivity of high-priority services based on this score value, ensuring that when a micro-burst congestion trend occurs in the link traversed by the critical control flow, the system can trigger a protective resource preemption response with a lower risk threshold than the perceived service. The total execution time of the above flow tag injection operation does not exceed 15% of the total delay of the white-box switch inlet pipeline, corresponding to a measured value of approximately 120 to 180 nanoseconds. This time overhead is far less than the maximum transmission delay budget of four milliseconds allowed by the substation GOOSE message, and will not have a perceptible impact on the end-to-end deterministic delay of protection and control services.

[0019] In the aforementioned IPv6 white-box network construction system 100 for smart substations, the telemetry acquisition and folded header encapsulation module 120 is used to collect the chip queue depth and port rate of each network element through which the tagged substation service flow passes, and compress and encapsulate the multi-dimensional status indicators into a folded hop-by-hop extended header to obtain a status-coded extended header message. It should be understood that the inherent intelligence capability of the smart substation IPv6 white-box network relies on real-time perception of the entire network link status. The lightweight inference engine of the white-box switch control plane needs to obtain physical layer operating indicators such as queue depth, port rate, and delay offset at the chip level of each network element to accurately predict the link congestion probability. However, the traditional out-of-band centralized polling acquisition method has the drawbacks of large sampling delay and occupying independent management bandwidth. Directly inserting complete raw telemetry data into each hop-by-hop node would cause the IPv6 extended header to expand beyond the network's maximum transmission unit limit, thus triggering packet fragmentation. The fragmentation operation itself consumes additional chip parsing cycles and exacerbates the resource contention between the control flow and the telemetry flow. Therefore, in the technical solution of this application, the chip queue depth and port rate of each network element through which the tagged substation service flow passes are further collected for status. The multi-dimensional status indicators are compressed and encapsulated into a folded hop-by-hop extended header to obtain a status-coded extended header message. This allows the hardware operating status of each network element to be synchronously transmitted to the downstream inference module with extremely low byte overhead along with the service message without triggering message fragmentation. This enables the subsequent intrinsic intelligent inference and risk quantification modules to directly deconstruct the link physical status characteristics from the data plane message without relying on out-of-band acquisition channels, providing a real-time input data source with microsecond-level accuracy for congestion probability prediction. Simultaneously, the transmission overhead of telemetry information is compressed to a fixed four bytes or less, preventing the expansion of the extended header from encroaching on the processing bandwidth of the chip parsing pipeline.

[0020] Specifically, in this embodiment, the telemetry acquisition and folded header encapsulation module includes: a chip status acquisition and flow caching unit, used to trigger in-band network telemetry sampling logic when a tagged substation service flow enters the white-box switch inlet pipeline, read the queue depth, port utilization, and delay offset time of the current outgoing port from the chip's real-time status data based on the system clock and hardware counter, and aggregate them to generate multi-dimensional status telemetry indicators, while storing the original service flow into the hardware pipeline cache to form a cached tagged service flow; a status indicator folding and compression unit, used to perform multi-dimensional status indicator folding and compression calculations on the values ​​of each item in the multi-dimensional status telemetry indicators to obtain a fixed-length folded and compressed status vector; and an extended header insertion and encoding unit, used to read the cached tagged service flow from the cache using a data plane message editor, dynamically insert a hop-by-hop option extended header between its IPv6 basic header and upper-layer payload, write the folded and compressed status vector into the data field of the extended header, and update the payload length field to generate a status-encoded extended header message.

[0021] More specifically, in a specific example of this application, firstly, when a tagged substation service flow enters the ingress pipeline of the white-box switch, the first-level parser of the ingress pipeline identifies a non-zero value in the flow tag field of the IPv6 header of the packet and triggers the pre-programmed in-band network telemetry sampling logic. This sampling logic performs a status reading operation based on the system clock and hardware counter inside the switching chip, and reads three key indicators from the hardware register group of the chip's real-time status data: the queue depth of the current outgress port, which is obtained from the queue occupancy counter of the chip's outgress buffer management unit, and its physical meaning is the total number of bytes of data packets residing in the buffer of the outgress port at the current moment; the port utilization rate, which is obtained by reading the cumulative number of bytes sent in the current second from the port traffic statistics register and dividing it by the theoretical maximum throughput of the port's physical link bandwidth; and the delay offset time, which is calculated by the difference between the system clock timestamp when the packet enters the ingress pipeline and the sending timestamp written when the packet leaves the previous hop outgress. The above three indicators are assembled into a fixed-structure triplet through hardware aggregation logic, forming a multi-dimensional status telemetry indicator. The data format of this triplet is a 32-bit unsigned integer for queue depth, a 32-bit unsigned integer for port utilization, and a 32-bit unsigned integer for delay offset, totaling 12 bytes stored in the metadata bus register. Simultaneously with the status read operation, the ingress pipeline stores the complete message data of the tagged substation service flow into the hardware pipeline buffer through a bypass write channel. The buffer controller allocates a unique pipeline buffer pointer to this message and attaches it to the metadata bus, thus forming a buffered tagged service flow. This ensures that the original message is not lost due to pipeline occupancy during subsequent folding / compression and header insertion operations.

[0022] Next, a multidimensional state index folding and compression calculation is performed on three values ​​in the multidimensional state telemetry index to generate a fixed-length folded and compressed state vector. This calculation is performed in pure hardware within the arithmetic logic unit of the white-box switch's data plane. The specific process is as follows: the original value of the queue depth is normalized by dividing it by the maximum physical capacity threshold of the outgoing port queue. This maximum capacity threshold is obtained by dividing the total outgoing buffer of the switching chip by the number of ports. In a typical substation white-box switch, this value is 512 kilobytes per port. The normalized ratio is multiplied by the queue depth quantization resolution scalar and rounded down, converting the continuous normalization ratio into an 8-bit unsigned integer. The same normalization and quantization operation is performed on the port utilization rate. Its original throughput rate is divided by the theoretical maximum transmission bandwidth of the physical link and multiplied by the port utilization quantization resolution scalar, then rounded down to an 8-bit unsigned integer. The delay offset time is divided by the maximum tolerable delay boundary value specified by the substation control network and multiplied by the delay offset quantization resolution scalar, then rounded down to an 8-bit unsigned integer. Finally, the three 8-bit integers are multiplied by different powers of two, logically shifted left, and then assembled. The queue depth quantization value is shifted left by sixteen bits to occupy the high 8-bit interval, the port utilization quantization value is shifted left by eight bits to occupy the middle 8-bit interval, and the delay offset quantization value occupies the low 8-bit interval. These three are then added together to form a fixed 32-bit (4-byte) integer scalar, which is the folded compression state vector. The mathematical expression of the above folded compression calculation is as follows:

[0023] in, The calculated folded compressed state vector is represented as a 32-bit unsigned integer scalar. This indicates the actual dwell depth of the chip exit queue at the current moment, extracted from multidimensional state telemetry indicators, in bytes. This indicates the maximum physical capacity threshold for the hardware of the exit queue, which is 512 kilobytes in this example; The quantization resolution scalar represents the queue depth, with a value of 255, which is the maximum value of an eight-bit unsigned integer. This indicates the real-time throughput rate of the current port, in bits per second. This indicates the theoretical maximum transmission bandwidth limit of the current physical link. For a gigabit Ethernet link within a substation, this value is 1,000,000,000 bits per second. The quantization resolution scalar representing port utilization, with a value of 255; This indicates the actual processing delay offset of the message within the switching chip, in nanoseconds. This represents the maximum tolerable time delay boundary value specified by the substation control network for critical operations. In this example, it is taken as 4,000,000 nanoseconds, or four milliseconds, according to the IEC 61850 standard. The quantization resolution scalar representing the time delay offset, with a value of 255; This represents the floor operator. In this formula, the first term... The normalized quantization value of the queue depth is pushed to the 24th to 16th bit range of the 32-bit vector. The value in this range directly reflects the physical occupancy of the current egress port buffer. When this eight-bit value is close to 255, it indicates that the egress queue is about to overflow. The subsequent inference module can use this to determine whether the link is in the early stage of micro-burst congestion. (Second item) The port utilization quantification value is placed in the range of bits 15 to 8. The value in this range represents the real-time proportion of link bandwidth consumption. In substation scenarios, when multiple video surveillance streams and sensor data streams are active simultaneously, this value will rise rapidly, providing the inference engine with a quantitative basis for the intensity of bandwidth contention. (Third item) Occupying the lowest 7th to 0th bit range, this value directly quantifies the ratio of the processing latency experienced by the message within the chip to the maximum tolerance boundary. When this ratio exceeds 50%, it indicates that the chip's parsing pipeline is under high load. The three quantized values ​​achieve interference-free assembly within a single 32-bit integer through different left shifts, eliminating the possibility of carry overflow between intervals. Furthermore, the entire folding operation involves only four basic arithmetic operations: division, multiplication, rounding, and shifting. The ASIC chip of the white-box switch can complete all calculations within a single clock cycle through combinational logic circuits.

[0024] Then, the data plane message editor is used to perform hop-by-hop extension header insertion and encoding operations to generate a status-coded extension header message. The message editor reads the corresponding message entity from the tagged service flow in the hardware pipeline buffer according to the pipeline buffer pointer, and dynamically inserts a standard IPv6 hop-by-hop options extension header between the IPv6 basic header and its upper-layer payload. The format of this hop-by-hop options extension header conforms to the RFC 8200 specification. Its next header field is set to the value of the next header field in the original IPv6 basic header to maintain the integrity of the protocol stack's chain structure. The header extension length field is set to zero to indicate that the total length of the extension header is eight bytes, i.e., the minimum legal length. The option type field is set to a custom experimental option type value of 0x1E, indicating that the option carries in-band telemetry data. The option data length field is set to four to indicate that the option data area occupies four bytes. The message editor writes the 32-bit value of the folded compressed state vector into the four-byte space of the options data field of the extended header in network byte order. It then updates the next header field of the IPv6 basic header to zero, i.e., the protocol number of the hop-by-hop options extension header, and increments the payload length field of the IPv6 basic header by eight bytes to reflect the change in total length due to the newly inserted extension header. After the above insertion operation is completed, the message editor recalculates the frame check sequence of the message and pushes the encapsulated message into the egress transmission queue, generating a state-coded extended header message.

[0025] It is worth noting that the total byte increase of the status-coded extended header message on the physical link is strictly controlled to eight bytes. This increment is far less than the remaining space of more than 1300 bytes between the 1500-byte maximum transmission unit of the gigabit Ethernet link within the substation and the 140-byte frame length of a typical GOOSE message, ensuring that the insertion of the hop-by-hop extended header will not trigger the IPv6 fragmentation mechanism. When the status-coded extended header message passes through multiple white-box network elements along the forwarding path, each intermediate node repeats the above collection, folding, and insertion process at the entry pipeline, appending the folded and compressed status vector of this node to the subsequent option data area of ​​the existing extended header or concatenating a new hop-by-hop option extended header after the original extended header. When the status-coded extended header message finally arrives at the control plane of the white-box switch equipped with the endogenous intelligent inference and risk quantification module, its hop-by-hop option extended header has accumulated the hardware status coding information of all the network elements along the way. This provides a complete set of physical status data along the path for subsequent extended header decoding and feature restoration operations, as well as cross-layer context data fusion. Meanwhile, the service classification metadata is synchronously transmitted to the endogenous intelligent inference and risk quantification module for calculation.

[0026] In the aforementioned IPv6 white-box network construction system 100 for smart substations, the intrinsic intelligent inference and risk quantification module 130 is used to predict link congestion probability in state-coded extended header messages based on a lightweight inference engine deployed on the control plane of the white-box switch, in order to obtain a real-time congestion risk coefficient. It should be understood that since the folded compressed state vector carried in the state-coded extended header message is a chip hardware-level indicator encoded in a high-density bit-based assembly, the subsequent resource preemption and deterministic path reconstruction modules require a clear, continuously numerical risk criterion that can be directly compared with a preset security threshold to determine whether to trigger event-driven telemetry bypass and path switching operations. The folded compressed state vector itself is only a snapshot of the static state of each network element, and cannot directly reflect whether the link will evolve into a micro-burst congestion state within a short future time window. Furthermore, the purely physical layer indicator is not associated with service priority information, and cannot reflect the differentiated threat level posed by the same physical congestion level to services of different priorities. Therefore, in the technical solution of this application, a lightweight inference engine deployed on the control plane of the white-box switch is further used to predict the link congestion probability of the state-coded extended header message to obtain a real-time congestion risk coefficient. This coefficient is then used to fuse the hardware state coding information of each network element along the path with the service priority semantics and input it into a nonlinear prediction model, outputting a continuous risk value normalized to a probability range of zero to one. This allows subsequent resource preemption and deterministic path reconstruction modules to obtain a unified decision-making basis that combines physical layer state awareness and service layer priority sensitivity. It enables early warnings to be issued with a lower trigger threshold at an early stage when congestion trends appear on the links traversed by critical control flows, achieving a more proactive risk assessment than a passive alarm mechanism based solely on queue thresholds.

[0027] Specifically, in this embodiment, the endogenous intelligent reasoning and risk quantification module includes: a header decoding and state restoration unit, used to perform extended header decoding and feature restoration on the state-encoded extended header message to obtain the underlying physical state vector; a cross-layer feature fusion unit, used to perform cross-layer context data fusion on the business classification metadata and the underlying physical state vector to obtain a multi-dimensional reasoning feature tensor; and a perceptron reasoning and risk quantification unit, used to input the multi-dimensional reasoning feature tensor into the control plane shallow perceptron model for endogenous intelligent reasoning and risk quantification to obtain the real-time congestion risk coefficient.

[0028] More specifically, in a particular example of this application, firstly, the data plane parsing probe of the white-box switch control plane performs extended header decoding and feature restoration operations on the arriving state-coded extended header messages. The parsing probe reads the four-byte option data field in the IPv6 hop-by-hop option extension header of the packet in a non-destructive manner, and obtains the 32-bit integer value of the folded compressed state vector carried therein. This value is then subjected to bitwise AND masking and logical right shift operations to restore the quantized values ​​of each physical indicator: The 32-bit value is ANDed with the mask 0xFF0000, and then shifted right by 16 bits to obtain an 8-bit quantized value of the queue depth. This quantized value is divided by the quantization resolution scalar 255 and multiplied by the maximum physical capacity threshold of the queue, 512 kilobytes, to restore the normalized queue depth; the 32-bit value is ANDed with the mask 0x00FF00, and then shifted right by 8 bits to obtain an 8-bit quantized value of the port utilization rate. Dividing this by 255 gives the normalized port utilization rate; the 32-bit value is ANDed with the mask 0x0000FF to obtain an 8-bit quantized value of the delay offset. Dividing this by 255 and multiplying by the maximum tolerable delay boundary value of 4,000,000 nanoseconds restores the current processing delay. After obtaining the current latency value, the control plane reads the latency value recorded in the previous sampling period from the local buffer. It subtracts the previous period's latency from the current latency and divides by the time interval between two samples. Using discrete-time first-order difference operations, it calculates the current link's latency growth rate. A positive value indicates latency deterioration, while a negative value indicates latency improvement. The parsing probe assembles the normalized queue depth and latency growth rate into a two-dimensional vector to generate the underlying physical state vector. Simultaneously, it updates the previous period's latency buffer with the current latency value for use in the next sampling period. The state-coded extended header message is not modified during decoding and remains in the data plane forwarding pipeline awaiting further processing.

[0029] Next, a cross-layer context data fusion operation is performed on the service category metadata and the underlying physical state vector to generate a multi-dimensional inference feature tensor. The control plane reads the service category metadata bound to the current message from the metadata bus and extracts the priority score value. This score value has been calculated in the aforementioned service flow perception and flow label mapping module. For GOOSE protection trip messages, this value is 0.979, and for video surveillance frames, it is 0.176. This priority score is used as a scalar of the third dimension and concatenated with the normalized queue depth and latency growth rate in the underlying physical state vector to form a three-dimensional data volume. The minimum-maximum scaling operation is performed on each element in the three-dimensional data volume. The normalized queue depth value range is mapped from zero to one to zero to one while remaining unchanged. The latency growth rate value range is linearly mapped from the minimum and maximum values ​​of its historical observations to the zero-to-one interval. The priority score value range is mapped from zero to one to zero to one while remaining unchanged. This eliminates the dimensional differences between the three dimensions and generates a multidimensional inference feature tensor that meets the requirements of the neural network input format. This tensor is a one-dimensional floating-point vector with a fixed length of 3.

[0030] Then, the multidimensional inference feature tensor is input into the control plane shallow perceptron model to perform endogenous intelligent inference and risk quantification calculation to obtain the real-time congestion risk coefficient. This shallow perceptron model is deployed on the control plane CPU of the white-box switch. The model structure is a single-layer perceptron plus a nonlinear activation function. The specific inference calculation process is as follows: Two physical layer features, normalized queue depth and latency growth rate, are extracted from the multidimensional inference feature tensor. These are multiplied by their respective learning weight coefficients and summed to obtain a physical layer linear combination value. This linear combination value is multiplied by a multiplicative amplification adjustment term consisting of a priority score and a preset sensitivity factor, so that when a high-priority control flow passes through this link, the same level of physical congestion produces a larger risk response amplitude. The amplified result is subtracted from a preset activation threshold bias to set the basic danger level for triggering nonlinear mutations. Finally, the above linear combination result is input into the logistic activation function to map it to a continuous probability interval from zero to one. The output value is the real-time congestion risk coefficient. The mathematical expression of the above inference calculation is as follows:

[0031] in, This represents the calculated real-time congestion risk coefficient, with a value range of zero to one. When it infinitely approaches one, it indicates that the link is about to experience severe congestion. The normalized queue depth, extracted from the multidimensional inference feature tensor, reflects the current physical crowding level of the switching chip's egress cache; Δ This represents the latency growth rate extracted from the multidimensional inference feature tensor, used to capture the dynamic trend of latency deterioration during micro-bursts of traffic; This represents the learning weight coefficient assigned to the normalized queue depth in the inference engine. In this example, it is 2.8, obtained through offline training, which determines the proportion of the absolute queue length's influence on risk assessment. represents the learning weight coefficient assigned to the latency growth rate in the inference engine. In this example, the value is 3.5, which determines the proportion of the impact of the latency deterioration rate on risk assessment. The weight coefficient with a value greater than the queue depth reflects that the latency deterioration trend has a stronger early warning significance for protection and control services than the static queue length in the substation scenario; μμ represents the priority sensitivity factor, which is 1.2 in this example. It is used to amplify the distress response sensitivity of high-priority services multiple times when the same physical congestion risk occurs. This indicates the priority score extracted from the business category metadata; This represents the activation threshold bias, which is set to 2.0 in this example. It determines the baseline risk level that triggers the logical stith function to transition from the low-probability region to the high-probability region. In this formula, the inner linear combination... This constitutes the congestion intensity metric for the pure physical layer. When the normalized queue depth is 0.7 and the delay growth rate is 0.5, the combined value is 2.8 × 0.7 + 3.5 × 0.5 = 3.71; multiplicative adjustment term. Introducing service priority semantics into the physical layer metric amplification process, the adjustment term value for a GOOSE packet with a priority score of 0.979 is 1 + 1.2 × 0.979 = 2.175, and for a video surveillance frame with a score of 0.176, the adjustment term value is 1 + 1.2 × 0.176 = 1.211. This means that under the same physical congestion conditions, the effective risk perceived by a GOOSE packet is amplified to approximately 1.8 times that of a video stream. This differentiated amplification mechanism ensures that the system's risk trigger threshold for protection and control services is significantly lower than that for perceived services. The overall result of the exponential term is 3.71 × 2.175 - 2.0 = 6.07, which, when substituted into the logistic function, yields... 1 / (1+e -6.07 The value is 0.9977, which far exceeds the preset safety threshold of 0.85, immediately triggering event-driven resource preemption logic. For video surveillance frames under the same physical condition, the exponent term result is 3.71 × 1.211 - 2.0 = 2.49, which, when substituted into the logical stith function, yields... 1 / (1+e -2.49 The value is 0.923. Although it also exceeds the threshold, its trigger margin is much smaller than that of the GOOSE message, which reflects the system's differentiated protection strategy for services with different priorities.

[0032] Furthermore, the real-time congestion risk coefficient, as the core output data object of this module, will enter the resource preemption and deterministic path reconstruction module along with the tagged substation business flows. In this module, the real-time congestion risk coefficient will be compared with the preset safety threshold and physical limit threshold to calculate the resource reallocation weight index. When the index is greater than zero, the system will send a resource preemption trigger signal to the data plane through a hardware interrupt pin, initiating the telemetry bypass and backup path switching operation for high-priority control flows. The entire inference calculation process takes no more than 50 microseconds to execute on the CPU of the white-box switch control plane. This delay is far less than 1.3% of the maximum transmission delay budget of four milliseconds for substation GOOSE messages, and will not have a perceptible impact on end-to-end deterministic delay. At the same time, the total number of parameters of the shallow perceptron model does not exceed ten floating-point numbers, and the memory occupation is less than one hundred bytes, which meets the constraints of limited computing resources of the white-box switch control plane.

[0033] In the aforementioned IPv6 white-box network construction system 100 for smart substations, the resource preemption and deterministic path reconstruction module 140 is used to forcibly bypass the telemetry encapsulation pipeline for high-priority control flows in tagged substation service flows to release the chip parsing cycle when the real-time congestion risk coefficient exceeds a preset safety threshold. It also issues a pre-calculated backup path for the high-priority control flow and generates a preemption forwarding instruction. It should be understood that since the real-time congestion risk coefficient output by the intrinsic intelligent inference and risk quantification module has clearly quantified the probability of micro-burst congestion occurring in the current link within a short future time window, but this risk coefficient itself is only a predictive value, if it is not converted into actual hardware operations on the chip data plane, the protection control message will continue to share the same entry parsing pipeline as the telemetry encapsulation task. When the chip parsing cycle is filled by telemetry operations, the end-to-end deterministic latency of GOOSE and sampled value messages will inevitably exceed the four-millisecond tolerance boundary. Furthermore, the traditional outgoing port queue priority scheduling mechanism cannot trace back to the chip entry parsing stage to intervene in resource allocation. Therefore, in the technical solution of this application, when the real-time congestion risk coefficient exceeds a preset safety threshold, the high-priority control flow in the tagged substation service flow is forcibly bypassed from the telemetry encapsulation pipeline to release the chip parsing cycle. A pre-calculated backup path is then issued for the high-priority control flow, generating a preemptive forwarding instruction. This achieves physical-level dynamic decoupling of telemetry tasks and critical control services at the micro-instruction level of the chip entry pipeline, instantly returning the parsing cycle occupied by the telemetry encapsulation operation to the protection control message. In this way, preventative resource switching can be completed before the congestion risk actually evolves into packet loss or timeout, allowing GOOSE and sampled value messages to bypass the telemetry parsing steps and be forwarded rapidly via a congestion-free backup path, fundamentally preventing the erosion of deterministic latency of critical services by telemetry tasks.

[0034] Figure 3 This is a block diagram of the resource preemption and deterministic path reconstruction module in an IPv6 white-box network construction system for smart substations according to an embodiment of this application. Figure 3 As shown, the resource preemption and deterministic path reconstruction module 140 includes: a risk assessment and preemption triggering unit 141, used to assess the real-time congestion risk coefficient and preset safety threshold and physical limit threshold to obtain a resource reallocation weight index, wherein when the resource reallocation weight index is greater than zero, a resource preemption trigger signal is sent to the data plane through a hardware interrupt pin, and the tagged substation service flow is suspended in the chip entry cache; a flow label matching and bypass instruction unit 142, used to perform flow label matching on the IPv6 flow labels in the tagged substation service flow after the data plane receives the resource preemption trigger signal to generate a message to generate a telemetry bypass micro-instruction; and a backup path allocation and instruction synthesis unit 143, used to extract a list of segment identifiers that avoid congestion nodes from the pre-calculated path table to generate a rerouting micro-instruction, and to atomically synthesize the telemetry bypass micro-instruction, the rerouting micro-instruction, and the memory pointer context of the tagged substation service flow to obtain a preemption forwarding instruction.

[0035] More specifically, in a particular example of this application, firstly, after the white-box switch control plane receives the real-time congestion risk coefficient output by the intrinsic intelligent inference and risk quantification module, it performs a risk assessment operation on the coefficient and a pre-stored preset security threshold and physical limit threshold in the system to calculate the resource reallocation weight index. In this example, the preset security threshold is configured as 0.85, indicating that when the real-time congestion risk coefficient reaches this value, the system determines that the link has entered a micro-burst danger period; the physical limit threshold is configured as 0.95 in this example, indicating that when the real-time congestion risk coefficient exceeds this value, the link is in a physical limit state about to overflow. The calculation logic for the resource reallocation weight index is as follows: When the real-time congestion risk coefficient is lower than the preset safety threshold, the output of the maximum value function is zero, and the system maintains its normal operating mode without performing any preemption operations; when the real-time congestion risk coefficient is between the preset safety threshold and the physical limit threshold, the difference between the real-time congestion risk coefficient and the preset safety threshold is calculated and divided by the difference between the theoretical maximum risk value and the preset safety threshold to obtain the normalized basic weight; when the real-time congestion risk coefficient further exceeds the physical limit threshold, the indicator function activates the emergency state penalty amplification coefficient, applying additional multiplicative amplification to the basic weight to accelerate resource stripping. The calculation expression of the above resource reallocation weight index is as follows:

[0036] in, This represents the calculated resource reallocation weight index. When this value is strictly greater than zero, the preemption logic is triggered. This represents the real-time congestion risk coefficient of the input, which is 0.9977 in the example above, calculated for the link through which the GOOSE message passes. This represents the preset safety threshold, which is 0.85 in this example. α represents the theoretical maximum congestion risk value, which is constant at 1.0 in the normalized model; α represents the emergency penalty amplification factor, which is 0.5 in this example, used to accelerate the tilting transfer of chip resources to the control flow when the risk is extremely high. This indicates an indicator function that outputs one when the value of the expression within its parentheses is greater than zero, and zero otherwise. This represents the physical limit threshold, which is 0.95 in this example, a value strictly greater than the preset safety threshold. In this formula, The function ensures that the resource reallocation weight exponent remains zero as long as the real-time congestion risk coefficient has not exceeded the preset safety threshold. This prevents the system from mistakenly triggering preemption operations under normal micro-fluctuations, thus avoiding unnecessary path switching that could disrupt network stability. (Fractional part) The risk increment exceeding the safety threshold is normalized to the range of zero to one. Its physical meaning is the relative position of the current risk level between the safety boundary and the theoretical limit. A larger value indicates that it is closer to the physical limit of chip cache overflow. Multiplicative adjustment term. When the risk coefficient exceeds the physical limit threshold of 0.95, the weight index is amplified by an additional 1.5 times. This reflects the need for the system to perform resource stripping at a more aggressive speed to prevent irreversible packet loss when the link is about to overflow. Substituting the real-time congestion risk coefficient of 0.9977 for the aforementioned GOOSE message into the calculation, the basic normalized weight is (0.9977-0.85) / (1.0-0.85)=0.985. Since 0.9977 is greater than the physical limit threshold of 0.95, the indicator function output is one. The final resource redistribution weight index is 0.985×(1+0.5×1)=0.985×1.5=1.477, which is strictly greater than zero. After the control plane confirms that the resource reallocation weight index is greater than zero, it sends a high-level resource preemption trigger signal to the data plane ASIC chip through the hardware interrupt pin on the white-box switch motherboard. The propagation delay of this signal is no more than ten nanoseconds. At the same time, the control plane sends a suspend command to the data plane pipeline to suspend the forwarding of the tagged substation service flow currently located in the chip entry buffer, so as to prevent the message from continuing to enter the parsing pipeline occupied by the telemetry task before the preemption logic configuration is completed.

[0037] Next, when the data plane ASIC chip detects a high-level transition of the resource preemption trigger signal, it immediately activates the emergency matching pipeline of the hardware-level tri-state content-addressable memory to perform a flow tag matching operation to generate a telemetry bypass micro-instruction. The tri-state content-addressable memory pre-stores flow tag matching rules corresponding to each service category. The emergency matching pipeline reads the IPv6 header flow tag field of each packet in the suspended tagged substation service flow from the chip's entry buffer one by one, and performs line-rate mask matching on its 20-bit value against the pre-stored rules. When the flow tag field value is 0xFFFFF, i.e., high-priority control class encoding, a match is found. The action field of the tri-state content-addressable memory immediately outputs the corresponding telemetry bypass micro-instruction. This micro-instruction is a fixed-format internal chip control word whose function is defined as: skipping all stages of lookup, folding compression calculation, and writing operations of the hop-by-hop option extension header in the data plane pipeline, and directly pushing the matched packet from the entry parsing stage to the exit routing lookup stage, thereby freeing up the parsing pipeline cycle originally used for telemetry encapsulation. For medium-priority sampling and best-effort aware packets with flow label field values ​​of 0x80000 or 0x00001, if a match is not found, these packets remain suspended and will resume normal forwarding after congestion subsides. Once the telemetry bypass micro-instruction is generated, it is output to the instruction synthesis stage along with the memory pointer of the matched packet.

[0038] Then, the control plane extracts backup path information that avoids the currently congested node from the pre-computed path table to generate rerouting micro-instructions, and performs an atomic synthesis operation with the telemetry bypass micro-instructions to obtain preemptive forwarding instructions. The pre-computed path table is pre-computed by the SDN central controller based on the substation global topology during the system initialization phase and distributed to the local storage of each white-box switch. The table maintains at least one segmented route backup path that avoids critical aggregation nodes for each source-destination node pair. The control plane extracts the IPv6 destination address from the high-priority control packets hit in the tagged substation service flow, queries the corresponding backup path entry in the pre-computed path table, and obtains a list of segment identifiers consisting of multiple IPv6 segment identifiers. This list defines the sequence of intermediate forwarding nodes that the packet should pass through from the current node to the destination node. All nodes in the list have been verified during the pre-computed phase to not contain currently congested network elements. The control plane encapsulates the segment identifier list into a segmented routing header format to generate a rerouting microinstruction. This microinstruction is defined as follows: inserting a segmented routing extension header after the basic IPv6 header of the packet and writing the segment identifier list into it; replacing the packet's IPv6 destination address with the first-hop node address in the segment identifier list. During the instruction synthesis phase, the telemetry bypass microinstruction, the rerouting microinstruction, and the memory pointer context of the hit packet in the tagged substation service flow are atomically synthesized into a single, uninterruptible composite operation instruction, namely, a preemptive forwarding instruction. This atomic synthesis ensures that the bypass operation and path switching operation are completed as a single transaction at the chip execution level, eliminating the risk of inconsistencies such as a packet only bypassing without switching paths or only switching paths without bypassing telemetry due to intermediate states.

[0039] Furthermore, the preemption forwarding command, as the output data object of this module, will be passed to the command execution and status feedback unit in the global topology aggregation and self-healing recovery module. This unit receives and executes the command through the underlying hardware abstraction interface of the white-box switch to complete the actual update operation of the data plane flow table. The preemption forwarding command simultaneously encodes telemetry bypass actions and segmented routing path switching actions. The underlying ASIC chip can complete all configurations within a single flow table write cycle, enabling high-priority control flows to be rapidly emitted from the designated physical port via a congestion-free backup path without any telemetry encapsulation steps. The entire link time from the issuance of the resource preemption trigger signal to the completion of the preemption forwarding command execution does not exceed 200 microseconds. This value accounts for only five percent of the four-millisecond maximum transmission latency budget for GOOSE messages, ensuring that the end-to-end deterministic latency of protection control messages during event-driven resource switching will not exceed the tolerance boundary due to the switching operation itself.

[0040] In the aforementioned IPv6 white-box network construction system 100 for smart substations, the global topology aggregation and self-healing recovery module 150 is used to aggregate the path switching status of all white-box network elements and refresh the global topology based on the execution feedback of the preemptive forwarding command, and to restore the telemetry acquisition task in the risk-eliminating network segment to obtain a self-healing network state diagram. It is understandable that after the resource preemption and deterministic path reconstruction modules independently execute the preemption and forwarding instructions on each white-box network element, the traffic distribution of the entire network has undergone local path shifts. Some network segments bear additional traffic load from upstream because high-priority control flows are rerouted to backup paths. Furthermore, the network elements in the bypass telemetry encapsulation pipeline lose their real-time perception of the link status of their own nodes during the preemption state. If the SDN central controller does not collect the execution results of each network element in a timely manner and synchronously refresh the network topology status, it will be unable to determine which network segments have dissipated micro-burst traffic and which network segments are still under cascading traffic impact. Rashly resuming telemetry collection will occupy the chip parsing cycle that is already under high load again, thus causing secondary congestion. On the other hand, permanently prohibiting telemetry collection will cause the network to lose its inherent intelligent perception capability and thus be unable to predict subsequent new congestion. Therefore, in the technical solution of this application, based on the execution feedback of the preemptive forwarding command, the path switching status of all white-box network elements in the network is aggregated and the global topology is refreshed. Telemetry acquisition tasks are then restored in the risk-reduced network segment to obtain a self-healing network state diagram. This establishes a closed-loop control circuit from preemptive execution to state aggregation and then to telemetry recovery, enabling the system to gradually restore the inherent intelligent sensing coverage of the entire network while ensuring that the deterministic delay of critical control flows is not compromised. This allows for the safe and orderly reactivation of in-band telemetry acquisition functions in each network segment after micro-burst congestion subsides, avoiding vicious oscillation cycles caused by premature telemetry recovery. Simultaneously, it provides the SDN central controller with a unified view of the latest path distribution and telemetry coverage across the entire network to support subsequent global routing optimization decisions.

[0041] Figure 4 This is a block diagram of the global topology aggregation and self-healing recovery module in an IPv6 white-box network construction system for smart substations according to an embodiment of this application. Figure 4As shown, the global topology aggregation and self-healing recovery module 150 includes: an instruction execution and status feedback unit 151, used to receive and execute preemption forwarding instructions through the underlying hardware abstraction interface of the white-box switch, and extract local status feedback to obtain a single-node execution feedback matrix; a cascaded sensing risk retention estimation unit 152, used to perform topology aggregation and comprehensive risk retention estimation on the single-node execution feedback matrix to obtain a global affected network segment risk map; and a telemetry recovery and status reconstruction unit 153, used to traverse the comprehensive risk retention of each network segment in the global affected network segment risk map, and when it is lower than the preset security recovery baseline, send a bypass lock release signaling to the corresponding network segment white-box device to restore the in-band telemetry acquisition task, merge the status information of the recovered network segment and the still preempting network segment into the latest segmented routing table and telemetry coverage of the entire network for system-level reconstruction, and generate a self-healing network status map.

[0042] Accordingly, the instruction execution and status feedback unit 151 is used to receive and execute preemption forwarding instructions through the underlying hardware abstraction interface of the white-box switch, and to extract local status feedback to obtain a single-node execution feedback matrix. It should be understood that since the preemption forwarding instructions generated by the resource preemption and deterministic path reconstruction module are composite operation instructions that have undergone logical compilation in the control plane, encoding two hardware configuration requirements—telemetry bypass action and segmented routing path switching action—these instructions have not yet taken effect in the actual forwarding table of the underlying ASIC switching chip. If they are not translated into chip-recognizable underlying register write operations through the hardware abstraction interface, the high-priority control flow will still forward along the original congested path. Simultaneously, if the SDN central controller cannot obtain the actual switching status of each network element, subsequent cascaded risk assessment and telemetry recovery decisions will lose their reliable data foundation. Therefore, in the technical solution of this application, the white-box switch's underlying hardware abstraction interface further receives and executes the preemption forwarding command, and performs local state feedback extraction to obtain a single-node execution feedback matrix. This is used to translate the logical commands of the control plane into physical operations of the data plane, and after the operation is completed, the key parameters of the execution result are extracted and reported to the SDN central controller. In this way, the SDN central controller can obtain accurate time and port change information of each network element path switching, providing a complete and time-accurate input data source for subsequent cascading propagation adjacency matrix construction and enhanced comprehensive risk retention calculation.

[0043] More specifically, in a concrete example of this application, firstly, the underlying hardware abstraction interface (HAPI) of the white-box switch receives a preemption forwarding command from the control plane. This interface conforms to the switch abstraction interface standard specification, translating the unified operation request issued by the upper-layer application into a sequence of underlying register writes that can be recognized by the specific ASIC chip. The HAPI parses the two composite operations encoded in the preemption forwarding command: for the telemetry bypass action, the interface inserts a high-priority rule entry with the flow tag field value 0xFFFFF as the matching key into the matching action table of the ASIC chip's inlet pipeline, with the action field being skipping all stages of parsing and encapsulation of the hop-by-hop option extension header; for the segmented routing path switching action, the interface writes the segment identifier list into the corresponding entry in the egress forwarding table, updating the next-hop outgoing port of the high-priority control flow from the original outgoing port number to the physical port number corresponding to the backup path. The above two table entry write operations are completed in an atomic transaction manner, with a hardware time of no more than 50 microseconds.

[0044] Next, after completing the flow table update, the underlying ASIC switching chip returns a hardware confirmation signal indicating successful operation to the underlying driver. Upon receiving this signal, the underlying driver immediately calls the precise time protocol hardware clock module inside the switching chip to read the current absolute timestamp. This timestamp, with nanosecond precision, records the precise physical moment when the flow table update actually took effect. Then, the underlying driver reads the device identifier of this network element from the chip configuration register, extracts the original outgoing port number before the handover from the original content of the preemption forwarding instruction, and reads the new outgoing port number after the handover from the updated outgoing forwarding table entry. These four pieces of information are then structured and encapsulated according to a fixed field format: the device identifier occupies four bytes, the original outgoing port number occupies two bytes, the new outgoing port number occupies two bytes, and the absolute timestamp occupies eight bytes, totaling sixteen bytes, generating a single-node execution feedback matrix. In this matrix, the device identifier enables the controller to locate the specific physical network element location; the combination of the original and new outgoing port numbers allows the controller to trace the link changes before and after the path handover and construct a cascading propagation adjacency matrix accordingly; and the absolute timestamp allows the controller to calculate the time elapsed since the preemption occurred for the parameter input of the exponential time decay function.

[0045] Furthermore, the white-box switch asynchronously reports the single-node execution feedback matrix to the SDN central controller via the gRPC bidirectional flow channel of the northbound management interface. The reporting operation is completed within one millisecond after the instruction is executed. The single-node execution feedback matrix, as the output data object of this unit, will be aggregated with similar matrices reported by other white-box network elements across the entire network on the controller side, and enter the cascaded risk retention estimation unit. This provides the new outgoing port number information of each node for the construction of the cascaded propagation adjacency matrix, and provides the preemption status indication value and execution timestamp information for the calculation of the ontological risk retention.

[0046] Correspondingly, the cascaded sensing risk retention estimation unit 152 is used to perform topology aggregation and comprehensive risk retention estimation on the single-node execution feedback matrix to obtain a global risk map of affected network segments. It should be understood that since the single-node execution feedback matrix reported by multiple white-box network elements across the entire network only records the independent preemption execution time, device identification, and port change information of each node, these discrete single-node data have not yet established a network segment-level correlation. If the SDN central controller does not aggregate these data according to the network segment affiliation in the physical topology and calculate the current risk level of each network segment, it cannot determine which network segments' micro-burst traffic has dissipated to the point where telemetry acquisition can be safely restored, or which network segments are still in a state of congestion sequelae and need to continue maintaining telemetry bypass protection. Misjudging the timing of telemetry recovery will directly lead to the chip parsing cycle being filled again after telemetry recovery, thus triggering secondary congestion. Therefore, in the technical solution of this application, topological aggregation and comprehensive risk retention estimation are further performed on the single-node execution feedback matrix to obtain a global risk map of affected network segments. This integrates the local execution status information scattered across various network elements into a quantitative risk distribution view at the network segment level. This provides a clear, segment-level risk assessment basis for subsequent telemetry recovery and state reconstruction units, enabling the controller to have sufficient decision-making information when comparing risk retention with the preset safety recovery baseline segment by segment, thus avoiding premature restoration of telemetry data collection for network segments still under traffic impact.

[0047] More specifically, in a concrete example of this application, firstly, the SDN central controller receives the single-node execution feedback matrix asynchronously reported by each white-box network element across the entire network via the gRPC channel. Based on the device identifier field in each feedback record, it queries the global link layer discovery protocol base table to which the network element belongs for the physical network segment number. All feedback records are grouped and aggregated according to the network segment affiliation relationship to form a feedback set indexed by the network segment. For each network segment, the controller extracts the absolute timestamp of each node from its feedback set to determine the preemption status indication value: if the difference between the absolute timestamp of a node and the current time is less than the preset preemption status effective window and the new outgoing port number of the node is inconsistent with the original outgoing port number, it is determined that the node is still in the preemption handover state, and its preemption status indication value is set to one; otherwise, it is set to zero.

[0048] Next, the controller performs an independent weighted time decay summation operation on each network segment to calculate the overall risk retention rate. The logic of this operation is as follows: for each white-box network element node within the target network segment, its centrality weight in the global substation topology is multiplied by the preemption status indication value, then multiplied by a negative exponential decay factor with a base of the natural constant and an exponent of the product of the time-series decay constant and the elapsed time since the preemption occurred. Finally, the above multiplication results for all nodes within the network segment are summed to obtain the overall risk retention rate of that network segment. The mathematical expression of this calculation is as follows:

[0049] in, Indicates network segment Ontology risk retention rate calculated based on its own node status; Indicates association with network segment The total number of white-box network element nodes; Represents network element nodes In the global substation topology, the centrality weight is set to 0.8 for the core aggregation switch and 0.3 for the edge access switch. This represents the preemption status indicator value of node ii, which is extracted and determined from the single-node execution feedback matrix. The time-series decay constant representing risk recovery has a value of 0.5; This indicates the current absolute timestamp when the controller performs this calculation; This represents the nodes extracted from the single-node execution feedback matrix. The absolute timestamp of the preemptive forwarding instruction. In this formula, the exponential decay factor... This reflects the physical law that the negative impact of preemption events on link security gradually diminishes over time, and the centrality weight... This ensures that changes in the state of the core aggregation switch have a greater impact on the overall risk score of the network segment than changes in the edge access switch. In the first embodiment, the controller directly appends the intrinsic risk retention values ​​of each network segment to the topology subgraph constructed by the link layer discovery protocol base table to generate a global risk map of affected network segments.

[0050] Specifically, in the IPv6 white-box network of a smart substation, when a micro-burst congestion occurs in a network segment, triggering event-driven resource preemption logic, high-priority control flows (such as GOOSE / SV protection messages) will be rerouted to pre-calculated backup paths by a segmented routing mechanism. The original comprehensive risk retention calculation mechanism treats each network element as an independent entity, multiplying the preemption status indication value of each node by a centrality weight, applying exponential time decay, and then directly weighting and summing them to obtain the risk score for each network segment. This independent summation logic implicitly assumes an unverified premise: the risk states of different network segments do not affect each other.

[0051] However, the actual physical topology of the substation communication network determines that this assumption is not valid. Specifically, when the high-priority control flow of segment A is preempted and rerouted, these control flows do not disappear into thin air, but are actually injected into adjacent or shared segments B and C along the SRv6 backup path. These downstream segments actually bear the additional traffic load from the upstream segment A without any local preemption event. In the typical star and ring hybrid topology of a substation, the core aggregation layer often serves as the backup path aggregation point for multiple edge segments. Once multiple edge segments trigger preemption switching simultaneously, the cascading transfer traffic borne by the core segment will rapidly accumulate. The first embodiment completely lacks the ability to perceive this causal propagation chain between segments. Its calculated comprehensive risk retention only reflects the historical preemption status of the target segment's own nodes, completely ignoring the traffic pressure that neighboring segments are continuously injecting into it through shared backup paths.

[0052] This defect can lead to serious misjudgments at the control level in actual operation: although all nodes of a core network segment are in a normal state (the inherent risk retention rate is close to zero), it is actually being subjected to cascading traffic impacts from multiple upstream network segments. The SDN controller, based on an excessively low risk score, incorrectly determines that the network segment has recovered and immediately issues a telemetry recovery signal to reactivate in-band telemetry acquisition (INT). The reactivation of the telemetry encapsulation pipeline immediately consumes the already strained chip parsing cycle. Coupled with the dual pressure of upstream traffic transfer, the deterministic latency of high-priority control flow deteriorates further, creating a vicious cycle of misjudgment of safety → telemetry recovery → secondary congestion → re-preemption, seriously threatening the real-time performance and reliability of substation relay protection.

[0053] To address the aforementioned shortcomings, a second embodiment is proposed. Specifically, in another specific example of this application, Figure 5 This is a block diagram of a cascaded sensing risk retention estimation unit in an IPv6 white-box network construction system for smart substations, according to an embodiment of this application. Figure 5 As shown, the cascaded risk retention estimation unit 152 includes: a cascaded propagation adjacency matrix construction unit 1521, used to perform backup path association matching on the new port numbers of each node after switching in the single-node execution feedback matrix based on the link layer discovery protocol basic table, identify the set of downstream network segments actually traversed by the backup paths rerouted by each preemptive switching node, construct a directed adjacency relationship using the reciprocal of the hop distance from each source network segment to the downstream network segment on the backup path as the propagation strength, and generate a cascaded propagation adjacency matrix; and an enhanced risk retention calculation unit 1522, used to perform cascaded perception comprehensive risk retention estimation on the cascaded propagation adjacency matrix and the single-node execution feedback matrix to obtain a global risk map of affected network segments.

[0054] Specifically, the cascaded propagation adjacency matrix construction unit 1521 is used to perform backup path association matching on the new port number of each node after switching in the single node execution feedback matrix based on the link layer discovery protocol basic table, identify the set of downstream network segments actually traversed by the backup path rerouted by each preemptive switching node, construct a directed adjacency relationship with the reciprocal of the hop distance from each source network segment to the downstream network segment on the backup path as the propagation strength, and generate a cascaded propagation adjacency matrix. It is understandable that, since the calculation of the comprehensive risk retention in the first embodiment only performs an independent weighted summation of the states of nodes within each network segment, it does not model the cascading traffic propagation dependency between network segments due to backup path sharing. In the actual physical topology of the substation, when a network segment triggers a preemptive switch, its high-priority control flow is rerouted to the backup path by the segmented routing mechanism and is actually injected into the adjacent or shared downstream network segments in the topology. These downstream network segments actually bear the additional traffic load from the upstream without the occurrence of a local preemptive event. If this causal propagation dependency between network segments is not explicitly quantified, the subsequent calculation of the enhanced comprehensive risk retention will not be able to detect which network segments are being impacted by upstream transferred traffic. Therefore, in the technical solution of this application, based on the link layer discovery protocol base table, backup path association matching is performed on the new port numbers of each node after switching in the single node execution feedback matrix. This identifies the set of downstream network segments actually traversed by the backup paths rerouted by each preemptive switching node. A directed adjacency relationship is constructed using the reciprocal of the hop count distance from each source network segment to the downstream network segment on the backup path as the propagation strength, generating a cascading propagation adjacency matrix. This transforms the originally isolated node switching events into a topologically related causal propagation graph, explicitly depicting the cascading traffic impact dependencies between network segments across the entire network. This enables the subsequent enhanced risk retention calculation unit to simultaneously perceive the real traffic pressure being transferred to it by neighboring network segments through backup paths when assessing the risk level of the target network segment, providing topological-level causal correlation basis for the controller's telemetry recovery decisions.

[0055] More specifically, in a concrete example of this application, firstly, after the SDN central controller receives the single-node execution feedback matrix reported by the white-box network elements across the entire network, it needs to establish causal propagation dependencies between network segments. Since the single-node execution feedback matrix records the new port numbers of each node after switching, and the global link layer discovery protocol base table maintains complete physical topology connection information, the downstream network segments actually traversed by the backup path rerouted by each node that has undergone preemptive switching can be accurately traced by associating and matching the new port numbers with the base table. The specific operation is as follows: The controller extracts the new outgoing port number of each node from the single-node execution feedback matrix one by one, and queries the peer device identifier and peer port number connected to the port number in the link layer discovery protocol base table, thereby determining the first hop adjacent node of the backup path and its network segment; then, starting from the adjacent node, it queries the network segment number of each subsequent hop node in the segment identifier list stored in the pre-calculated path table in turn, until the end node of the backup path is reached, and collects all the network segment numbers passed along the way to form the downstream network segment set of the source network segment, and at the same time records the hop distance from the source network segment to each downstream network segment.

[0056] Next, based on the aforementioned path coverage relationships, a directed adjacency matrix is ​​constructed, and the adjacency matrix is ​​cascaded for propagation. Each element in the matrix represents the propagation strength from the source network segment to the target network segment, which is quantified by calculating the reciprocal of the hop count distance between the two network segments on the backup path. The calculation is expressed as follows:

[0057] in, This represents the propagation strength of segment m to segment n in the cascaded propagation adjacency matrix; This represents the hop count distance from network segment m to network segment n on the backup path of network segment m; This represents the set of all downstream network segments that the backup path of network segment m actually traverses after a preemption occurs. It should be noted that a smaller hop distance indicates a stronger propagation intensity. This accurately reflects the physical cascading effect in hybrid star and ring topologies of substations, where network segments closer to the preemption source experience greater impact from transferred traffic. This results in the core aggregation network segment accumulating higher cascading risk due to its proximity to the backup paths of multiple edge network segments. Specifically, when the backup path of network segment m directly passes through its adjacent core aggregation network segment n, the hop distance... As one, the intensity of transmission A value of one, or the maximum value, indicates that the core network segment has withstood the full impact of the rerouting traffic from the source network segment; while when the backup path takes three hops to reach the remote network segment k, the hop count distance... Third, the intensity of transmission Only one-third, indicating that the cascading traffic impact on the remote network segment has significantly decreased as the propagation path extends. In a typical substation topology, the core aggregation layer switch often serves as the first-hop aggregation point for backup paths of multiple edge network segments. When multiple edge network segments trigger preemptive switching simultaneously, the core network segment will receive a cascading contribution of propagation strength of one from each source network segment. These contributions are weighted by the coupling coefficient and accumulated in the subsequent calculation of enhanced comprehensive risk retention, resulting in a comprehensive risk score for the core network segment that is much higher than that of any single edge network segment. This ensures that the controller maintains the most conservative telemetry recovery strategy for the core aggregation network segment.

[0058] Then, after the controller completes the calculation of the propagation strength between each network segment that has experienced preemptive handover and its downstream network segments, it fills all non-zero elements into the corresponding positions of the cascaded propagation adjacency matrix. The matrix elements between network segment pairs not covered by backup paths remain zero. The dimension of this matrix is ​​a square matrix of the total number of network segments multiplied by the total number of network segments. Its sparsity depends on the actual coverage of backup paths in the substation topology. In a typical substation star and ring hybrid topology, the number of non-zero elements is proportional to the product of the number of edge network segments that have experienced preemptive handover and the length of each backup path. The generated cascaded propagation adjacency matrix, as the output data object of this unit, will enter the enhanced risk retention calculation unit together with the single-node execution feedback matrix. In this unit, the propagation intensity coefficient in the cascaded propagation adjacency matrix will be used as a weighting factor when the risk retention of each source network segment propagates to the target network segment and participates in the linear superposition operation. This makes the final risk score of the target network segment include not only the historical preemption state contribution of its own nodes, but also the external cascaded risk contribution injected to it by all upstream source network segments through backup paths, thereby realizing a qualitative change from single-point irrelevant assessment to global causal correlation assessment.

[0059] Specifically, the degree calculation unit 1522 is used to perform a comprehensive risk retention estimation of the cascaded propagation adjacency matrix and the single-node execution feedback matrix to obtain a global risk map of affected network segments. It should be understood that since the cascaded propagation adjacency matrix has fully depicted the directed propagation dependency relationship between network segments due to backup path sharing, but this matrix itself is only a static topology description and has not yet been integrated with the current actual preemption status and time decay information of each network segment, if only the ontological risk retention degree in the first embodiment is used as the risk score for each network segment, the core network segment located at the convergence point of multiple upstream network segment backup paths will obtain a risk score close to zero when all its nodes have recovered. The controller will then incorrectly issue telemetry recovery signaling to this network segment, causing secondary congestion. Therefore, in the technical solution of this application, a comprehensive risk retention estimate is further performed on the cascaded propagation adjacency matrix and the single-node execution feedback matrix to obtain a global risk map of affected network segments. This allows for the linear superposition and fusion of the ontological risk retention and the external cascaded risk contribution from the upstream source network segments, so that the final risk score of each network segment simultaneously reflects information from both its own state and the pressure transferred from its neighbors. This enables the controller to obtain a unified quantitative basis that combines ontological state awareness and topology cascade awareness when determining when telemetry data can be safely restored, effectively preventing misjudgments of safety and vicious oscillation cycles caused by underestimation of single-point assessments.

[0060] More specifically, in a concrete example of this application, after constructing the cascading propagation adjacency matrix, the calculation of the enhanced comprehensive risk retention degree begins. The reason for introducing a cascading correction term on top of the original weighted summation is that the backup path in the substation topology is not a virtual logical channel, but a real physical link. Rerouting traffic generated by upstream network segment preemption will inevitably flow into the downstream network segment along this physical link, resulting in quantifiable bandwidth encroachment. First, the first stage of the calculation process calculates the ontological risk retention degree of each network segment based on the state indication value and execution timestamp of each node in the single-node execution feedback matrix, multiplying each node's centrality weight by its state indication value and superimposing an exponential time decay function. The specific operation is as follows: For each target network segment jj, the SDN central controller traverses all white-box network element nodes belonging to that segment. It extracts the preemption status indicator value and absolute timestamp of each node from the single-node execution feedback matrix. The centrality weight of each node is multiplied by the preemption status indicator value, then multiplied by an exponential time decay factor, and the results are accumulated one by one to obtain the ontological risk retention degree of that network segment based solely on the historical preemption status of its own nodes. This calculation is expressed as follows:

[0061] in, Indicates network segment Ontology risk retention calculated solely based on its own node state; This represents the total number of white-box network element nodes associated with network segment j; This represents the centrality weight of network element node i in the global substation topology; This indicates the preemption status indicator value of node i. The value is 1 when the node is still in the preemption switching state, and 0 otherwise. The time-series decay constant representing risk recovery; Represents the current absolute timestamp; This represents the absolute timestamp of node i executing the preemption forwarding instruction.

[0062] Next, in the second stage of the calculation process, based on the cascading propagation adjacency matrix, all upstream source network segments are traversed. The intrinsic risk retention of each source network segment is multiplied by the corresponding propagation strength coefficient and accumulated to obtain the external cascading risk contribution borne by the target network segment. Then, the intrinsic risk retention and the external cascading risk contribution are linearly superimposed and fused to obtain the enhanced comprehensive risk retention. Specifically, for each target network segment j, the controller traverses all non-zero elements in the j-th column of the cascading propagation adjacency matrix. Each non-zero element... For an upstream source network segment m that is injecting transferred traffic into network segment j through a backup path, determine the inherent risk retention degree of source network segment m. Multiply by propagation intensity The single-source cascade contribution of network segment m to network segment j is obtained. The single-source cascade contributions of all upstream source network segments are then summed and multiplied by the cascade propagation coupling coefficient. The total contribution of exogenous cascade risks is obtained, and finally, the enhanced comprehensive risk retention is obtained by adding the ontological risk retention to the contribution of exogenous cascade risks. This calculation is expressed as follows:

[0063] in, Indicates network segment Enhanced composite risk retention rate after cascading propagation correction; This represents the cascading propagation coupling coefficient, used to adjust the weight of external risk contributions in the overall assessment. It is set by the substation network administrator based on topology density and service traffic characteristics. This represents the total number of network segments where preemption / handover events occurred. This represents the propagation strength of segment m to target segment j in the cascaded propagation adjacency matrix; This indicates the inherent risk retention rate of the source network segment m. It should be noted that even if a network segment itself does not directly experience any preemption events (i.e., ...), ... As long as there is a backup path for the upstream source network segment that passes through this network segment, the cascading item... A reasonable positive risk score will still be assigned to each segment to prevent the controller from making false security judgments about that network segment. Finally, the enhanced comprehensive risk retention rate of each network segment will be appended to the topology subgraph constructed by the link layer discovery protocol base table to generate a global risk map of affected network segments containing quantified risk distribution.

[0064] In summary, by introducing a cascading adjacency matrix and integrating it into the calculation logic of comprehensive risk retention, the second embodiment enables the SDN controller to determine when telemetry acquisition can be safely restored, no longer relying solely on the historical state of the target network segment's own nodes, but simultaneously sensing the real traffic pressure being transferred to it by neighboring network segments through backup paths. In the typical scenario where the core aggregation network segment of a substation serves as the common aggregation point for multiple backup paths, the second embodiment can accurately identify implicitly high-risk network segments that, despite not having preemption events themselves, have actually been impacted by multiple cascading traffic flows. This allows the controller to maintain a more conservative telemetry recovery strategy for such network segments, delaying the reactivation of the INT encapsulation pipeline, thereby reserving sufficient chip parsing cycles for high-priority control flows. Ultimately, this improvement effectively blocks the vicious cycle of misjudgment of safety → restoration of telemetry → secondary congestion → re-preemption, ensuring that the end-to-end deterministic latency of substation relay protection messages in white-box network global path switching scenarios is not disrupted by cascading effects.

[0065] Correspondingly, the telemetry recovery and state reconstruction unit 153 is used to traverse the comprehensive risk retention of each network segment in the global affected network segment risk map. When it is lower than the preset security recovery baseline, it sends a bypass lock release signaling to the corresponding network segment white-box device to restore the in-band telemetry acquisition task. It merges the state information of the recovered network segment and the network segment still in contention with the latest segmented routing table and telemetry coverage of the entire network for system-level reconstruction to generate a self-healing network state map. It is understandable that since the global affected network segment risk map has provided an enhanced comprehensive risk retention value for each network segment after cascading propagation correction, the risk assessment result itself is only a static quantitative snapshot. If actual telemetry recovery operations are not performed based on the assessment result, the healthy network segments whose risks have dissipated will continue to be in a telemetry bypass state and lose the ability to perceive their own link status in real time. The coverage of the network's inherent intelligent perception will gradually shrink to a blind spot over time. Permanently prohibiting telemetry collection will prevent the system from predicting subsequent new congestion. At the same time, if the affected network segments that are still under the impact of cascading traffic are incorrectly restored to telemetry, it will trigger chip parsing cycle contention again. Therefore, in the technical solution of this application, the comprehensive risk retention of each network segment in the global affected network segment risk map is further traversed. When it is lower than the preset security recovery baseline, a bypass lockout signaling is issued to the corresponding network segment white-box device to resume the in-band telemetry acquisition task. The status information of the recovered network segment and the network segment still in contention is merged and the latest segmented routing table and telemetry coverage of the entire network are reconstructed at the system level to generate a self-healing network state diagram. In this way, the inherent intelligent sensing coverage of the entire network is restored in an orderly manner segment by segment while ensuring that the deterministic delay of high-priority control flow is not destroyed. In this way, a complete closed-loop control loop can be formed from risk assessment to telemetry recovery to unified refresh of the entire network status, enabling the system to autonomously recover to the endogenous intelligent sensing state with full coverage after the micro-burst congestion subsides. At the same time, it provides the SDN central controller with a unified decision view of the entire network path distribution and telemetry coverage gaps.

[0066] More specifically, in a specific example of this application, firstly, the SDN central controller reads the enhanced comprehensive risk retention value of each network segment in the global affected network segment risk map and compares this value with a preset security recovery baseline. In this example, the preset security recovery baseline is set to 0.10. The physical meaning of this baseline value is: when the enhanced comprehensive risk retention of a network segment drops below this value, it indicates that the preemption state of the network segment's own nodes has sufficiently decayed, and the cascaded transfer traffic injected into it by the upstream source network segment via the backup path has dissipated to a level insufficient to cause a perceptible load on the chip parsing pipeline. At this time, restoring the in-band telemetry acquisition task of this network segment will not trigger resource contention between the chip parsing cycle and the high-priority control flow. For network segments with an enhanced comprehensive risk retention value below 0.10, the controller marks them as recoverable; for network segments with an enhanced comprehensive risk retention value still higher than or equal to 0.10, the controller maintains their current bypass preemption state without change, ensuring that the high-priority control flow continues to be forwarded via the backup path without triggering telemetry encapsulation operations.

[0067] Next, the controller sends a bypass lockout signaling message to all white-box devices within all network segments marked as recoverable via the gRPC management channel. This signaling message is a standard hardware abstract interface control command. Upon receiving the signaling message, the white-box switch performs two operations: clearing the telemetry bypass rule entries previously inserted for high-priority control flows in the ingress pipeline matching action table, allowing subsequent packets to re-enter the folding, compression, and encapsulation logic of the hop-by-hop option extended header; and restoring the forwarding entries of high-priority control flows in the egress forwarding table from the physical ports of the backup path to the physical ports of the original primary path, allowing the control flows to return to the normal optimal path forwarding. After the above two operations are completed, the in-band network telemetry acquisition task of this network segment resumes normal execution. Subsequent tagged substation service flows passing through this network segment will re-trigger the chip status acquisition and folding header encapsulation process, restoring the real-time perception capability of the link queue depth, port utilization, and time delay offset.

[0068] Then, the controller performs a system-level reconstruction operation on the status information of the healthy network segments that have recovered telemetry and the affected network segments that are still performing bypass preemption. Specifically, the controller extracts the current status label of each network segment from the global affected network segment risk map, indicating whether it has recovered or is still preempting. It then summarizes the currently effective segmented routing table entries for each network segment into the latest segmented routing table for the entire network. This table reflects whether the actual forwarding path of each high-priority control flow at the current moment is the primary path or the backup path. Simultaneously, it statistically analyzes the telemetry coverage status of each network segment, marking network segments that have recovered telemetry as telemetry active and those still in bypass mode as telemetry paused, forming a complete mapping of the entire network's telemetry coverage. The controller merges the latest segmented routing table for the entire network, the telemetry coverage mapping, and the enhanced comprehensive risk retention values ​​of each network segment into a unified network-wide status data structure, and generates a self-healing network status diagram in the form of a directed graph.

[0069] It is worth mentioning that each directed edge in the self-healing network state diagram is labeled with its telemetry acquisition status as active or bypassed, and each node is labeled with its current preemption switching status and enhanced comprehensive risk retention value. The global attributes of the graph record the network-wide telemetry coverage rate, i.e., the percentage of telemetry-active network segments out of the total number of network segments. As the final output data object, the self-healing network state diagram provides the SDN central controller with a complete and unified view of the network-wide path distribution, telemetry coverage gaps, and residual risk distribution. Based on this view, the controller can determine in subsequent scheduling cycles whether it is necessary to recalculate the globally optimal path to reclaim the link bandwidth occupied by the backup path, whether it is necessary to dynamically adjust the telemetry sampling frequency to adapt to the current network load level, and whether it is necessary to update the alternative entries in the pre-calculated path table to cope with topology changes. The entire closed-loop control cycle from the execution of the preemptive forwarding command to the generation of the self-healing network state diagram does not exceed 100 milliseconds, meeting the substation network management's operation and maintenance requirements for second-level topology convergence. At the same time, the cascaded sensing risk assessment mechanism ensures the conservatism and accuracy of telemetry recovery decisions, effectively blocking the vicious oscillation cycle caused by misjudgment of safety, and ensuring that the end-to-end deterministic delay of substation relay protection messages in global path switching scenarios is not destroyed by telemetry recovery operations.

[0070] In summary, the IPv6 white-box network construction system for smart substations according to the embodiments of this application is explained. It utilizes IPv6 flow label fields to establish hardware-level identification credentials for services of different priorities at the chip entry point. Simultaneously, it embeds network element status information in a folded and compressed form into the IPv6 hop-by-hop extended header and transmits it synchronously with the packet, avoiding packet fragmentation and bandwidth storms caused by telemetry data. Based on this, lightweight congestion prediction inference is pushed down to the white-box switch control plane to quantify link risks in real time. Once the risk exceeds the safety threshold, an event-driven mechanism forces high-priority control flows to skip the telemetry encapsulation pipeline to release the chip parsing cycle. Pre-calculated backup paths are then issued based on segmented routing, achieving dynamic decoupling of telemetry tasks and critical services at the data plane micro-instruction level. Finally, topology aggregation and risk assessment are performed based on network-wide execution feedback. Telemetry acquisition resumes after the risk dissipates, forming a closed-loop self-healing system. This solution requires no additional hardware computing power, collaboratively approaching from the protocol layer and data plane scheduling layer, achieving intrinsic intelligent perception of network status while ensuring zero packet loss and low latency control.

Claims

1. A white-box IPv6 network construction system for smart substations, characterized in that, include: The service flow perception and flow label mapping module is used to extract protocol features and determine priority of the acquired original message flow of substation secondary equipment, and map the service priority features to the IPv6 flow label field to obtain the substation service flow and service classification metadata with labels. The telemetry acquisition and folded header encapsulation module is used to acquire the status of the chip queue depth and port rate of each network element through which the tagged substation business flow passes, and to compress and encapsulate the multi-dimensional status indicators into a folded hop-by-hop extended header to obtain a status-coded extended header message. The intrinsic intelligent reasoning and risk quantification module is used to predict the link congestion probability of state-coded extended header messages based on a lightweight inference engine deployed on the control plane of the white-box switch in order to obtain the real-time congestion risk coefficient. The resource preemption and deterministic path reconstruction module is used to force the high-priority control flow in the tagged substation business flow to bypass the telemetry encapsulation pipeline to release the chip parsing cycle when the real-time congestion risk coefficient exceeds the preset safety threshold, and to issue a pre-calculated backup path for the high-priority control flow and generate a preemption forwarding instruction. The global topology aggregation and self-healing recovery module is used to aggregate the path switching status of all white-box network elements and refresh the global topology based on the execution feedback of the preemptive forwarding command, and to restore the telemetry acquisition task in the risk-eliminated network segment to obtain the self-healing network state diagram.

2. The IPv6 white-box network construction system for smart substations according to claim 1, characterized in that, The raw message stream of substation secondary equipment includes protection and control messages, sampled value messages, video surveillance data, and sensor monitoring data.

3. The IPv6 white-box network construction system for smart substations according to claim 1, characterized in that, The service flow awareness and flow tag mapping module includes: The message feature extraction and caching unit is used to perform parallel feature extraction on the original message stream of the substation secondary equipment through a programmable data plane parser to obtain the Ethernet type field, application identifier field, virtual LAN priority label and message payload length, and to generate a message multidimensional feature matrix by splicing tensors according to a fixed dimension order. At the same time, the original message payload is stored in the chip entry static random access memory to form a cached original message stream. The service priority determination unit is used to intelligently determine the service priority of the message multidimensional feature matrix in order to obtain service classification metadata. The flow label injection and encapsulation unit is used to retrieve the cached raw message flow from the chip memory and write the category identifier in the service classification metadata into the flow label field of the corresponding message IPv6 header to generate a tagged substation service flow.

4. The IPv6 white-box network construction system for smart substations according to claim 1, characterized in that, The telemetry acquisition and header folding encapsulation module includes: The chip status acquisition and flow buffer unit is used to trigger the in-band network telemetry sampling logic when the tagged substation service flow enters the white box switch inlet pipeline. Based on the system clock and hardware counter, it reads the queue depth, port utilization and delay offset time of the current output port from the chip real-time status data and aggregates them to generate multi-dimensional status telemetry indicators. At the same time, it stores the original service flow into the hardware pipeline buffer to form a cached tagged service flow. The state index folding and compression unit is used to perform multi-dimensional state index folding and compression calculations on the values ​​of each item in the multi-dimensional state telemetry index to obtain a fixed-length folded and compressed state vector. The Extended Header Insertion and Encoding Unit is used to read cached tagged service flows from the cache using the data plane message editor, dynamically insert hop-by-hop option extended headers between the IPv6 basic header and the upper-layer payload, write folded compressed state vectors into the data field of the extended header, and update the payload length field to generate a state-encoded extended header message.

5. The IPv6 white-box network construction system for smart substations according to claim 1, characterized in that, The intrinsic intelligent reasoning and risk quantification module includes: The header decoding and state restoration unit is used to perform extended header decoding and feature restoration on the state-coded extended header message to obtain the underlying physical state vector. The cross-layer feature fusion unit is used to perform cross-layer context data fusion on business classification metadata and underlying physical state vectors to obtain a multi-dimensional inference feature tensor. The perceptron reasoning and risk quantification unit is used to input multi-dimensional reasoning feature tensors into the control plane shallow perceptron model to perform endogenous intelligent reasoning and risk quantification to obtain real-time congestion risk coefficients.

6. The IPv6 white-box network construction system for smart substations according to claim 1, characterized in that, The resource preemption and deterministic path reconstruction module includes: The risk assessment and preemption triggering unit is used to assess the real-time congestion risk coefficient and the preset safety threshold and physical limit threshold to obtain the resource reallocation weight index. When the resource reallocation weight index is greater than zero, a resource preemption triggering signal is sent to the data plane through the hardware interrupt pin, and the tagged substation service flow is suspended in the chip entry buffer. The flow label matching and bypass instruction unit is used to perform flow label matching on the IPv6 flow labels in the tagged substation service flow after the data plane receives a resource preemption trigger signal in order to generate a message and generate a telemetry bypass micro-instruction. The backup path allocation and instruction synthesis unit is used to extract a list of segment identifiers that avoid congested nodes from the pre-calculated path table to generate rerouting micro-instructions, and to atomically synthesize the telemetry bypass micro-instructions, rerouting micro-instructions, and the memory pointer context of the tagged substation service flow to obtain preemptive forwarding instructions.

7. The IPv6 white-box network construction system for smart substations according to claim 1, characterized in that, The global topology aggregation and self-healing recovery module includes: The instruction execution and status feedback unit is used to receive and execute preemption forwarding instructions through the underlying hardware abstraction interface of the white-box switch, and to extract local status feedback to obtain the single-node execution feedback matrix. The cascaded sensing risk retention estimation unit is used to perform topological aggregation and comprehensive risk retention estimation on the feedback matrix of a single node to obtain a global risk map of affected network segments. The telemetry recovery and state reconstruction unit is used to traverse the comprehensive risk retention of each network segment in the global affected network segment risk map. When it is lower than the preset security recovery baseline, it sends a bypass lock release signaling to the corresponding network segment white-box device to restore the in-band telemetry acquisition task. It merges the state information of the recovered network segment and the network segment still in contention with the latest segmented routing table and telemetry coverage of the entire network for system-level reconstruction, generating a self-healing network state map.

8. The IPv6 white-box network construction system for smart substations according to claim 1, characterized in that, The cascaded sensing risk retention estimation unit includes: The cascaded propagation adjacency matrix construction unit is used to perform backup path association matching on the new port number of each node after switching in the single node execution feedback matrix based on the basic table of the link layer discovery protocol. It identifies the set of downstream network segments actually traversed by the backup path rerouted by each preemptive switching node, and constructs directed adjacency relationship with the reciprocal of the hop distance from each source network segment to the downstream network segment on the backup path as the propagation strength, thereby generating the cascaded propagation adjacency matrix. The enhanced risk retention calculation unit is used to perform cascaded perception of the adjacency matrix of cascaded propagation and the single-node execution feedback matrix to estimate the comprehensive risk retention of the network segments to obtain the global risk map of affected network segments.

Images