A self-healing gateway and method for power communication edge computing power based on hardware and software coupling

Abstract

This invention provides a self-healing gateway and method for power communication edge computing based on hardware and software coupling, including a multimodal communication unit, a hardware adaptive protection unit, and an edge computing decision unit. The multimodal communication unit is used to collect physical layer parameters of each communication channel in real time and realize seamless switching of heterogeneous links according to instructions. The hardware adaptive protection unit includes a multi-level surge protection circuit, an isolation transformer, an active noise monitoring circuit, and a power amplifier cascaded between the power line inlet and the power line carrier baseband chip. The edge computing decision unit is connected to the multimodal communication unit and the hardware adaptive protection unit through a bus, and calculates the comprehensive link cost of each heterogeneous link based on the collected physical layer parameters, through dynamic evolution evaluation weight based on information entropy and priority coefficient associated with power service type. It controls the multimodal communication unit to transmit core power service data along selected links or optimal relay paths.

Description

Technical Field

[0001] This invention belongs to the field of power communication technology, specifically relating to a power communication edge computing power self-healing gateway and method based on hardware and software coupling. Background Technology

[0002] Power communication gateways serve as the communication bridge between the end of the power distribution network and the cloud master station, undertaking the tasks of uploading and downloading data such as relay protection commands, fault alarms, and equipment status acquisition. The end nodes of the power distribution network are numerous, widely distributed, and deployed in harsh environments, facing various extreme conditions such as lightning surges, power frequency interference, and impedance distortion. Furthermore, core services such as relay protection require end-to-end transmission latency of no more than 20ms and zero transmission errors, placing extremely high demands on communication reliability. Currently, the mainstream communication methods in the power communication field include fiber optics, 5G, power line carrier, and low-power wireless, while edge computing technology is also gradually becoming widespread, reducing the cloud load through local data processing. Surge protection typically uses transient suppression diodes to achieve single-stage discharge, preventing damage to communication chips from lightning strikes and operational overvoltages.

[0003] In existing technologies, ZigBee network fault gateway self-healing solutions based on edge computing are geared towards smart home scenarios. They achieve self-healing by having a backup node take over the core gateway functions, but only support dual-mode ZigBee and Wi-Fi communication, lacking hardware protection and channel compensation mechanisms, and thus unable to cope with strong electromagnetic interference and high-voltage surges in power environments. Machine learning-based ZigBee network intelligent diagnosis and self-healing systems employ a cloud-edge collaborative architecture, with core decisions relying on the cloud. Edge nodes lack autonomy and dedicated hardware design, making them unsuitable for extreme power conditions. Dynamic gateway switching technology for SMS services only implements software-level gateway switching, completely neglecting hardware protection and physical layer optimization, and cannot solve channel distortion and interference problems in power communication.

[0004] Existing technologies generally suffer from the following shortcomings: narrow applicability to application scenarios and lack of resistance to extreme interference; insufficient hardware-software collaboration and no proactive hardware compensation capability; single communication mode and limited link evaluation dimensions; single self-healing mechanism and lack of cross-node relay optimization capability; insufficient depth of edge computing application and weak autonomous decision-making capability; reliance on cloud-based decision-making, making it difficult to guarantee security and reliability. Currently, a deeply coupled self-healing solution combining hardware adaptive protection and edge intelligent decision-making is still lacking for the strong electromagnetic interference environment at the end of the power distribution network. Existing gateways cannot meet the communication needs of the end of the power distribution network, easily forming offline islands, which may lead to relay protection failure or maloperation, causing line tripping, equipment damage, or even large-scale power outages, seriously threatening the safe and stable operation of the power system. Summary of the Invention

[0005] To address the shortcomings and deficiencies of existing technologies, this invention provides a power communication edge computing self-healing gateway and method based on hardware-software coupling. This invention employs a three-domain isolated modular architecture consisting of a multimodal communication unit, an edge computing decision unit, and a hardware adaptive protection unit. Through closed-loop linkage of hardware parameter acquisition, software algorithm decision-making, hardware instruction execution, and parameter feedback optimization, it achieves autonomous perception, dynamic scheduling, and millisecond-level self-healing of the power communication link. This invention constructs a two-layer self-healing system combining link-level and network-wide approaches. In the event of a single link failure, seamless switching between heterogeneous links and hierarchical compression of non-core services ensure the transmission of core services. In the event of a local full-link failure, a cross-node Mesh networking self-healing process is triggered. An improved path optimization algorithm is employed, using the Lagrange multiplier method to transform the dual hard constraints of end-to-end transmission delay and remaining node power into penalty terms, constructing a comprehensive cost function. Dual hard constraint verification is introduced during node relaxation updates to generate the globally optimal relay path. Real-time channel compensation is achieved through noise sampling synchronized with the symbol transmission time slot of the power line carrier baseband chip, combined with dynamic gain and phase adjustment, and a variational spatiotemporal evolution prediction model is introduced to achieve channel pre-compensation. The weights of the link evaluation dimensions are dynamically adjusted using the information entropy method, and the comprehensive link cost is calculated in conjunction with power service priorities, making the link evaluation more aligned with the needs of power scenarios.

[0006] The specific technical solution adopted by this invention to solve its technical problem is as follows:

[0007] A power communication edge computing self-healing gateway based on hardware and software coupling includes a multimodal communication unit, a hardware adaptive protection unit, and an edge computing decision unit;

[0008] The multimodal communication unit integrates multiple heterogeneous communication modules to collect physical layer parameters of each communication channel in real time and achieve seamless switching of heterogeneous links according to instructions.

[0009] The hardware adaptive protection unit includes a multi-stage surge protection circuit, an isolation transformer, an active noise monitoring circuit, and a power amplifier cascaded between the power line inlet and the power line carrier baseband chip; the active noise monitoring circuit is configured to acquire instantaneous power line interference voltage and impedance parameters in a manner synchronized with the symbol transmission time slot of the power line carrier baseband chip; the power amplifier is configured to receive instructions from the edge computing decision unit and perform dynamic adjustment of transmission gain and phase;

[0010] The edge computing decision unit is connected to the multimodal communication unit and the hardware adaptive protection unit via a bus, and is configured as follows:

[0011] Based on the collected physical layer parameters, the comprehensive link cost of each heterogeneous link is calculated by using dynamic evolution evaluation weights based on information entropy and priority coefficients associated with power business types.

[0012] When the overall link cost of the main link exceeds the preset threshold, the multi-mode communication unit is controlled to switch to the available heterogeneous link with the lowest overall link cost based on the relationship between the available bandwidth of the link and the bandwidth required by the service. The non-core services are then subjected to hierarchical compression to ensure the transmission of core services.

[0013] When the combined link cost of all local uplink heterogeneous links exceeds a preset threshold, the cross-node Mesh networking self-healing process is triggered. The multimodal communication unit scans the surrounding adjacent gateway nodes to obtain the status parameters of the adjacent nodes, and an improved path optimization algorithm is executed based on the network topology to generate the optimal relay path that simultaneously satisfies the end-to-end transmission delay constraint and the node's remaining power constraint.

[0014] The multimodal communication unit is controlled to transmit core power service data along a selected link or optimal relay path;

[0015] The improved path optimization algorithm is configured to: convert the time delay constraint and power constraint into penalty terms using the Lagrange multiplier method, construct a Lagrange comprehensive cost function that includes the comprehensive path cost, time delay penalty and power penalty; and introduce double hard constraint verification during the node relaxation update process, only updating the path cost and node state when the candidate path simultaneously satisfies the time delay constraint and power constraint.

[0016] Furthermore, the edge computing decision unit is also used to run a variational spatiotemporal evolution prediction model. The input of the model is a historical noise voltage sequence and a power line impedance parameter sequence, and the output is the predicted mean and variance of the noise voltage at future times. The edge computing decision unit calculates the optimal gain based on the prediction results to achieve pre-compensation of the power line carrier channel. The dynamic evolution evaluation weight based on information entropy is calculated by the information entropy method. In each sampling period, the weight value is dynamically adjusted according to the dispersion of the data in each evaluation dimension. The weight value is positively correlated with the dispersion of the dimension data.

[0017] Furthermore, the improved path optimization algorithm also includes a pre-screening step for invalid nodes, which performs three-layer filtering on all nodes based on battery life, signal strength, and communication distance to generate a mask of valid nodes. The algorithm performs dynamic re-optimization every 10ms, comparing the combined cost of the new path with the current running path. If the cost difference exceeds 10% or the path changes, the algorithm switches to the new path.

[0018] Furthermore, the tiered compression of non-core services specifically involves dividing power services into four levels: Level 1 services are prohibited from compression, Level 2 services are only allowed to be compressed without loss, Level 3 services are allowed to be compressed without loss up to 30%, and Level 4 services are allowed to be compressed with loss or without loss up to 90%. The compression ratio is dynamically adjusted based on the ratio of available link bandwidth to service demand bandwidth, and is updated every 10ms.

[0019] Furthermore, the acquisition of status parameters of adjacent nodes includes remaining power and real-time bandwidth utilization. The edge computing decision unit calculates the remaining battery life of adjacent nodes through a battery life prediction model and selects nodes with a remaining battery life of not less than 48 hours as valid relay nodes.

[0020] Furthermore, the multi-stage surge protection circuit is a three-stage cascaded structure consisting of a GDT ceramic gas discharge tube, a PTC resettable fuse, and a TVS transient suppression diode array. The cascading order is as follows: power line inlet, GDT ceramic gas discharge tube, PTC resettable fuse, TVS transient suppression diode array, isolation transformer, and power line carrier baseband chip. The inter-electrode capacitance of the GDT ceramic gas discharge tube does not exceed 1.5pF, the parasitic capacitance of the PTC resettable fuse does not exceed 100pF, and the parasitic capacitance of the TVS transient suppression diode array does not exceed 80pF.

[0021] Furthermore, the sampling period of the active noise monitoring circuit is no greater than 12.5ns, and the synchronization error with the symbol transmission time slot of the power line carrier baseband chip is no more than 1μs; the sampled data is transmitted to the edge computing decision unit through the LVDS interface in the direct memory access mode, and the data format is 16-bit signed integer plus 32-bit microsecond-level timestamp.

[0022] Furthermore, the edge computing decision unit is also configured to: after controlling data transmission, collect the channel state and transmission performance parameters after the decision is executed, and dynamically correct the comprehensive link cost, optimal relay path and channel compensation parameters based on the feedback parameters to form a continuously optimized closed-loop control.

[0023] Furthermore, when the improved path optimization algorithm fails to find an effective relay path that satisfies the dual hard constraints, the edge computing decision unit triggers a local data temporary storage mechanism to store the core power business data in local non-volatile memory, which will be automatically uploaded to the cloud master station after the link is restored.

[0024] Furthermore, a self-healing method for edge computing power in power communication based on hardware and software coupling, characterized in that it operates on the above-mentioned gateway and includes the following steps:

[0025] The physical layer parameters of each heterogeneous communication channel are collected in the first cycle, and the instantaneous interference voltage and impedance parameters of the power line are collected in a way that is synchronized with the symbol transmission time slot of the power line carrier baseband chip.

[0026] Based on the collected physical layer parameters, the comprehensive link cost of each heterogeneous link is calculated by using dynamic evolution evaluation weights based on information entropy and priority coefficients associated with power business types.

[0027] When the overall link cost of the main link exceeds a preset threshold, the system switches to the available heterogeneous link with the lowest overall link cost based on the relationship between the available bandwidth of the link and the bandwidth required by the service. The system also performs hierarchical compression on non-core services to ensure the transmission of core services.

[0028] When the combined link cost of all local uplink heterogeneous links exceeds a preset threshold, scan the surrounding adjacent gateway nodes and obtain their status parameters to construct the network topology.

[0029] An improved path optimization algorithm is used to solve the global optimal relay path. The algorithm transforms the end-to-end transmission delay constraint and the node remaining power constraint into penalty terms through the Lagrange multiplier method, constructs the Lagrange comprehensive cost function, and introduces double hard constraint verification during the node relaxation update process. The path cost and node state are updated only when the candidate path simultaneously satisfies the delay constraint and the power constraint.

[0030] Transmit core power service data along the selected link or the optimal relay path, and monitor the link status in a second cycle, dynamically updating the comprehensive link cost and self-healing decision.

[0031] Compared to existing technologies, this invention and its preferred solution achieve active compensation for power line carrier channels through a deeply coupled hardware and software collaborative optimization mechanism, effectively improving communication reliability in environments with strong electromagnetic interference. It constructs a dual-layer self-healing system combining link-level and network-wide approaches, capable of handling link failures of varying degrees and completely solving the "offline island" problem at the end of the power distribution network. The improved path optimization algorithm simultaneously considers communication quality, transmission latency, and node lifespan, generating optimal relay paths that meet the stringent requirements of power services and preventing secondary network failures. All core decisions are completed locally at the edge, without relying on cloud computing power, significantly improving the security and response speed of the power communication system. Through a service-level scheduling mechanism, the transmission of core power services is prioritized under bandwidth constraints, ensuring the safe and stable operation of the power system. A closed-loop feedback mechanism continuously optimizes system parameters, enabling adaptation to the complex and ever-changing operating environment of the power distribution network. Attached Figure Description

[0032] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments:

[0033] Figure 1 This is a basic flowchart illustrating the implementation of the self-healing scheme for edge computing power in power communication according to an embodiment of the present invention. Detailed Implementation

[0034] To make the features and advantages of the present invention more apparent and understandable, specific embodiments are described below in detail:

[0035] It should be noted that the following detailed descriptions are exemplary and intended to provide further explanation of this application. Unless otherwise specified, all technical and scientific terms used in this specification have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains.

[0036] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the exemplary embodiments according to this application. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0037] This invention proposes a deeply coupled hardware and software edge computing self-healing gateway and method for power communication. It employs a modular hardware architecture with three-domain isolation and a lightweight software algorithm system. Through a closed-loop architecture of hardware parameter acquisition, software algorithm decision-making, hardware instruction execution, and parameter feedback optimization, it achieves autonomous perception, dynamic scheduling, and millisecond-level self-healing of the power communication link. The system consists of a multimodal communication unit, an edge computing decision-making unit, and a hardware adaptive protection unit. The system adopts a heterogeneous bus architecture: the physical layer's raw sampled signals are aggregated in real time through a high-speed LVDS interface, while the edge decision-making unit collaborates deeply with each communication module and AI acceleration unit via a PCIe / SPI bus, thereby decoupling data throughput and control commands.

[0038] In this embodiment, the multimodal communication unit integrates four communication modules: fiber optic, 5G, power line carrier, and low-power wireless, supporting seamless switching between heterogeneous channels. This unit collects physical layer parameters such as signal-to-noise ratio, latency, packet loss rate, and power consumption per bit for each channel in real time, as well as power line noise characteristics and impedance parameters, and uploads them to the edge computing decision unit. Simultaneously, it receives software instructions to adjust the transmission gain, phase, transmission power, and operating mode, completing data encapsulation, transmission, and decapsulation.

[0039] The edge computing decision unit is equipped with a low-power SoC chip with AI acceleration capabilities, running lightweight algorithms such as comprehensive link cost assessment, hierarchical service scheduling, power line carrier channel compensation, and improved path optimization. All core decisions are completed locally at the edge, and only the self-healing results are synchronized to the cloud. It does not rely on cloud computing power and meets the requirements of local autonomy, security and reliability of the power system.

[0040] The hardware adaptive protection unit is designed specifically for complex power environments, comprising a GDT-PTC-TVS three-level cascaded surge protection array, a high-permeability isolation transformer, an active noise monitoring circuit, and a power amplifier. This unit achieves magneto-electric isolation of the power line, high-voltage surge discharge, and overcurrent and overtemperature protection. Simultaneously, it acquires instantaneous interference voltages on the power line in real time, receives software compensation commands, and dynamically adjusts the transmission gain and phase, forming a collaborative optimization with the software algorithm.

[0041] As a preferred embodiment, the GDT-PTC-TVS three-level cascaded protection circuit is cascaded in the following order: power line inlet → GDT → PTC → TVS → isolation transformer → PLC baseband chip. Its main parameters are shown in Table 1. This circuit, through the coordinated operation of the three levels of protection, achieves a complete protection process from high-voltage surge discharge to residual voltage clamping, while ensuring extremely low parasitic capacitance to avoid attenuation of the broadband power line carrier signal.

[0042] Table 1. Main parameters of the GDT-PTC-TVS three-level cascaded protection circuit.

[0043]

[0044] A high-permeability isolation transformer is located between the three-stage protection circuit and the PLC baseband chip, achieving magnetoelectric isolation between the high-voltage side of the power line and the signal processing side, while ensuring distortion-free transmission of the power line carrier signal. As a preferred embodiment, its main matching parameters are shown in Table 2.

[0045] Table 2 Main Matching Parameters of High Permeability Isolation Transformers

[0046]

[0047] The active noise monitoring circuit is a key hardware module for realizing software-hardware coupled channel compensation. It is responsible for real-time acquisition of power line noise characteristics, providing accurate physical layer input for the software channel compensation algorithm. As a preferred embodiment, its main parameters and binding relationship with the software and hardware are shown in Table 3.

[0048] Table 3 Main parameters and coupling bonding instructions for the active noise monitoring circuit

[0049]

[0050] The comprehensive link cost assessment model considers four dimensions: signal-to-noise ratio, latency, energy consumption, and environmental robustness. It uses the information entropy method to dynamically evolve the assessment weights and adjusts the priority normalization coefficients based on the type of power service. Within a 10ms sampling period, parameters are collected from m groups of heterogeneous links across the four assessment dimensions to form an observation matrix. , where x ijThis represents the parameter value of the i-th link in the j-th evaluation dimension. After forward-normalizing the inverse metrics (latency, energy consumption), standardization is performed using the extreme value method:

[0051]

[0052] in, and Let be the minimum and maximum values ​​of all link parameters in the j-th evaluation dimension, respectively. The value is the standardized parameter value, and its range is [0,1].

[0053] Calculate the information entropy for each dimension to measure the dispersion of the data in that dimension:

[0054]

[0055] in, Let be the information entropy of the j-th evaluation dimension. When all links have the same parameter value in this dimension, =1 indicates that this dimension provides no distinguishing information; when the parameter value difference is the greatest... =0 indicates that this dimension provides the most discriminative information.

[0056] Coefficient of difference This reflects the importance of that dimension; the larger the difference coefficient, the more important the dimension. The dynamic weights are:

[0057]

[0058] in, Let be the dynamic weight of the j-th evaluation dimension, and the sum of all weights is 1.

[0059] The overall link cost function is:

[0060]

[0061] Wherein, α, β, γ, and δ are service priority normalization coefficients, which are dynamically adjusted according to different service types, and their values ​​range from [0,1], and α+β+γ+δ=1. As a preferred embodiment, in relay protection service, α=0.2, β=0.5, γ=0.2, and δ=0.1, because relay protection has the highest time delay requirement, so β ​​takes a larger value. In environmental monitoring service, α=0.2, β=0.2, γ=0.5, and δ=0.1, because environmental monitoring has high energy consumption requirements. , , , The dynamic weights are for four dimensions: signal-to-noise ratio, latency, energy consumption, and environmental robustness. Let be the signal-to-noise ratio of the j-th link, in dBm; The maximum signal-to-noise ratio across all links; Let be the transmission delay of the j-th link, in milliseconds (ms). For the maximum end-to-end delay allowed in power services, in relay protection services =100ms; The unit bit power consumption of the j-th link is expressed in J / bit. This represents the total available energy consumption corresponding to the remaining power of the node, in J; is the environmental robustness coefficient of the j-th link, with a value range of [0,1]. The larger the value, the less the link is affected by environmental interference and the more stable it is.

[0062] The dynamic gain and phase compensation model for power line carrier channels compensates for signal phase shifts through a phase correction factor and achieves real-time gain compensation by combining active noise monitoring. The complex impedance of power lines is derived. Where R is the electric field line resistance, X L For power line reactance, X C This is the power line capacitive reactance. The phase correction factor is:

[0063]

[0064] This factor is used to compensate for the signal phase shift caused by power line impedance, ensuring that the signal phase at the receiving end is correct.

[0065] The real-time gain compensation formula is:

[0066]

[0067] Among them G base As a reference gain, set according to the power line carrier communication standard, in a preferred embodiment, G base =20dB; k is the feedback adjustment slope, used to control the sensitivity of gain adjustment, with a value range of [0.5, 2]. In a preferred embodiment, k = 1; V noise (t) represents the real-time interference voltage, acquired by the active noise monitoring circuit; V ref As a reference noise voltage, the value is taken as the average value of the power line background noise. In a preferred embodiment, V ref =0.1V.

[0068] The edge computing decision unit synchronously calculates the gain adjustment and phase correction values ​​according to the above formula, and sends them to the power amplifier and power line carrier baseband chip for execution to achieve real-time channel compensation.

[0069] The mapping relationship between the ADC value acquired by the active noise monitoring circuit and the actual noise voltage is as follows: , where ADCcode is a signed integer value obtained by 16-bit ADC sampling, with a value range of [-32768, 32767].

[0070] To achieve adaptive gain adjustment under noise fluctuations, this scheme introduces a variational spatiotemporal evolution prediction model to predict noise at future times. As a preferred embodiment, the specific implementation of this model is as follows:

[0071] The model input is a noise voltage sequence {V} over the past N time steps. noise (t-N+1),V noise (t-N+2),...,V noise The model outputs the predicted mean μ(t+1:t+M) and variance σ of the noise voltage over the next M time steps. The model also includes the corresponding electric field line impedance parameter sequence {Z(t-N+1), Z(t-N+2),...,Z(t)}. In a preferred embodiment, N=20. 2 (t+1: t+M), in a preferred embodiment, M=5.

[0072] Variational distribution q(x) t The algorithm employs a diagonal Gaussian distribution, with parameters generated by a feedforward neural network. The input consists of historical noise and an impedance sequence, and the output is the mean and log-variance of the Gaussian distribution. The prior distribution p(x) t |z {1：t-1} A first-order Markov process is used, assuming that the hidden state at the current time step depends only on the hidden state at the previous time step, and the transition distribution follows a Gaussian distribution. Model training employs a stochastic gradient variational inference algorithm to maximize the lower bound of evidence. E q (·) represents the variational distribution Expectations; Let be the log-likelihood function, representing the state of a given hidden state. Noise was observed under the following conditions The probability of; (·||·) represents the KL divergence, used to measure the variational distribution. With prior distribution The differences between them.

[0073] During the iterative solution process, the Adam optimizer is used to update the neural network parameters, with a learning rate of 1e-4, a batch size of 32, and 1000 iterations. During model deployment, a prediction is performed every 10ms, and the optimal gain is calculated based on the mean and variance of the predicted noise. ,in, Var(z) represents the predicted noise mean. t) represents the predicted noise variance; λ and η are adjustment coefficients, calibrated by actual measurements in the power distribution area. In a preferred embodiment, λ=0.8 and η=0.2.

[0074] This model can predict noise fluctuations in advance, achieve pre-adjustment of gain, and further reduce the bit error rate of signal transmission.

[0075] The battery life prediction model is used to select effective relay nodes in a mesh network, avoiding secondary network failures caused by insufficient battery power. E rem The remaining power of the node is expressed in Ah; n represents the number of communication modes, which in this scheme is n=4, corresponding to four communication modes: fiber optic, 5G, power line carrier, and low-power wireless; k i Let k be the power consumption coefficient for the i-th communication mode, expressed in A / Mbps. It represents the current consumption per 1 Mbps of data transmitted. In a preferred embodiment, k1 = 0.01 A / Mbps for fiber optic communication, k2 = 0.1 A / Mbps for 5G communication, k3 = 0.05 A / Mbps for power line carrier communication, and k4 = 0.02 A / Mbps for low-power wireless communication. i denoted as the real-time bandwidth utilization rate for the i-th communication mode, in Mbps.

[0076] Nodes with a battery life of less than 48 hours are set to refuse to act as relays, prioritizing their own data uploads.

[0077] The relay path optimization in a power mesh self-healing gateway is essentially a multi-objective combinatorial optimization problem with dual hard constraints. The standard Dijkstra's algorithm can only solve for the single-objective unconstrained shortest path and cannot be directly adapted. This scheme improves Dijkstra's algorithm by introducing the Lagrange multiplier method to transform the dual hard constraints into relaxable penalty terms, constructing a single cost function. The mathematical description of the problem is:

[0078]

[0079] Where P represents any path from source node v0 (local gateway) to target node vn-1 (cloud access gateway); J(P) is the total comprehensive link cost of the path, calculated by the comprehensive link cost evaluation model; D(P) is the end-to-end transmission delay of the path, and d(i,j) is the single-link transmission delay from node i to node j, in milliseconds; E min (P) represents the minimum remaining battery percentage for all nodes in the path, e i T represents the remaining battery percentage of node i, with a value ranging from [0,1]; limit =100ms is the mandatory delay threshold for power relay protection services; R limit=0.3 is the minimum remaining power threshold for a node; if it is below 30%, it is prone to power depletion failure.

[0080] The Lagrange synthesis cost function is: Where λ=0.6 and μ=0.4 are Lagrange penalty coefficients, calibrated through actual measurements in the power distribution area. When the path delay is less than or equal to the threshold or the remaining power of the node is greater than or equal to the threshold, the corresponding penalty term is 0, which does not affect the path ranking; when the constraint is violated, the penalty term increases linearly with the degree of violation, forcing the algorithm to prioritize the path that satisfies the constraint.

[0081] During the algorithm execution, the state from the source node to any node u is a triplet. Let represent the minimum Lagrange total cost, cumulative transmission delay, and minimum remaining power ratio on the path, respectively. For any edge (u,v), the new path state is derived as follows:

[0082] Cumulative delay of the new path:

[0083]

[0084] Minimum power consumption for new route:

[0085]

[0086] Total link cost of the new path:

[0087]

[0088] The comprehensive cost of the new path Lagrange:

[0089]

[0090] The improved relaxation update rule adds double hard constraints to the standard Dijkstra algorithm to enforce verification. The state of node v is only updated when the following three conditions are met simultaneously:

[0091]

[0092] If the conditions are met, then perform the update:

[0093]

[0094] Where prev[v] represents the predecessor node of node v on the optimal path, which is used for the final path backtracking.

[0095] The complete process of the improved Dijkstra algorithm is as follows: First, parameters such as the network topology adjacency matrix, comprehensive link cost matrix, link delay matrix, and node remaining power array are collected in real time from the gateway hardware module, and the global parameter T is initialized. limit=100ms, R limit =0.3, λ=0.6, μ=0.4. Then, a pre-screening of invalid nodes is performed, with three layers of filtering on all nodes: nodes with battery life less than 48 hours, signal strength less than -75dBm, or communication distance greater than 500m are marked as invalid, and the source node and target node are forcibly set as valid. Next, the algorithm state array is initialized, the Lagrange cost array is initialized to infinity, the source node cost is 0; the cumulative delay array is initialized to 0; the minimum path power array is initialized to 1.0, and the source node power is its own remaining power; the predecessor node array is initialized to -1; and the node access flag array is initialized to false.

[0096] The main loop iterates n times, selecting the node u with the minimum Lagrange cost from all unvisited valid nodes in each iteration. If no valid node is accessible, the algorithm terminates early and returns an empty path. Node u is marked as visited; if u is the target node, the main loop exits early. All adjacent nodes v of node u are traversed; if v is visited or invalid, it is skipped. The new path state is calculated using the above formula and verified according to the improved relaxation update rule. If the condition is met, the state of node v is updated.

[0097] After the main loop ends, if the Lagrange cost of the target node is still infinite, it indicates that there is no effective path satisfying the double hard constraints. This triggers a local data storage mechanism. As a preferred approach, Level 1 and Level 2 core business data can be stored in local Flash memory for at least 72 hours. Once the link is restored, it will be automatically uploaded to the cloud master station in chronological order. Otherwise, the algorithm backtracks from the target node along the predecessor node array to the source node, reversing the backtracking result to obtain the optimal relay path. The algorithm performs dynamic re-optimization every 10ms, comparing the combined cost of the new path with the currently running path. If the cost difference exceeds 10% or the path changes, the algorithm switches to the new path, achieving continuous self-healing of the Mesh network.

[0098] This algorithm will converge to the optimal path satisfying the double hard constraints within a finite number of steps. The number of nodes in a power mesh network is finite, and the main loop will terminate after at most n iterations; all link costs and delays are non-negative, therefore the Lagrange total cost has a lower bound of 0; each relaxation update strictly reduces the Lagrange cost of a node, preventing loops; the double hard constraints only eliminate infeasible paths and do not affect the optimality of feasible paths. The substantial differences between the improved Dijkstra algorithm and the standard Dijkstra algorithm are shown in Table 4.

[0099] Table 4. Comparison between the improved Dijkstra's algorithm and the standard Dijkstra's algorithm

[0100]

[0101] In this scheme, the solution space refers to the set of all possible nodes that the algorithm needs to traverse during path optimization. By pre-screening invalid nodes, this scheme reduces the solution space from all nodes to the set of valid nodes, significantly reducing the computational cost of the algorithm.

[0102] The edge data compression mechanism prioritizes power services based on their importance, ensuring the transmission of primary services under bandwidth constraints. Service priority and compression permissions are shown in Table 5.

[0103] Table 5 Priority Classification and Reduction Authority for Electricity Business

[0104]

[0105] For compressible Level 3 and Level 4 services, the unique device address identifier must be 100% preserved with an accuracy of ≤1. The data includes core electrical parameters such as timestamps, switch status, fault codes, event type codes, and RMS voltage / current values, as well as data checksums and encrypted signature fields. Compressible redundant content includes continuously sampled redundant data points, non-critical frames from video surveillance, high-frequency detail information from images, repeated fields from historical statistical data, and non-real-time logs and debugging information.

[0106] The triggering of compression actions and the selection of compression ratios are strongly tied to the available link bandwidth and the bandwidth required by the service. When the available link bandwidth is greater than or equal to 1.2 times the required service bandwidth, all services are not compressed; when the available bandwidth is between 0.8 and 1.2 times the required service bandwidth, 30% lossless compression for Level 3 services and 50% compression for Level 4 services are triggered; when the available bandwidth is between 0.5 and 0.8 times the required service bandwidth, 30% lossless compression for Level 3 services and 70% compression for Level 4 services are triggered, and non-real-time Level 4 service transmission is suspended; when the available bandwidth is less than 0.5 times the required service bandwidth, 30% lossless compression for Level 3 services and 90% extreme compression for Level 4 services are triggered, all non-core Level 4 service transmission is suspended, and only the transmission of Level 1-2 core services is guaranteed. The compression ratio is updated every 10ms based on the link cost assessment results and dynamically adjusted to ensure bandwidth resources for core services.

[0107] To adapt to the edge computing power constraints of gateways, a lightweight combination algorithm of LZ77 lossless compression and Huffman entropy coding is used for Level 2 and Level 3 service data. The code size is less than 50KB, the memory usage is less than 100KB, and the single packet data compression / decompression latency is less than 1ms, which can run in real time on the edge chip. For image and video data of Level 4 services, a lightweight MJPEG compression algorithm is used, which supports dynamic adjustment of the compression ratio to ensure priority transmission of the original bitstream of Level 1 services under bandwidth-constrained conditions.

[0108] The key to this solution is building a deeply coupled hardware and software collaborative optimization mechanism. The active noise monitoring circuit of the hardware adaptive protection unit and the channel parameter acquisition module of the multimodal communication unit collect physical parameters such as power line noise characteristics, channel signal-to-noise ratio, latency, energy consumption, and impedance distortion at 1ms and 10ms periods. These parameters are transmitted in real-time to the edge computing decision unit via a high-speed bus, driving the real-time execution of algorithms such as comprehensive link cost calculation, gain compensation factor derivation, and topology optimization. Based on the parameters collected by the hardware, the software algorithm of the edge computing unit generates decision commands in real-time, including link switching, service classification, gain adjustment, phase correction, and topology reconstruction. These commands are then sent to the corresponding hardware modules via the high-speed bus, driving hardware operations such as channel switching in the multimodal communication unit, power amplifier gain adjustment in the hardware adaptive protection unit, and neighborhood scanning in the low-power wireless module. After executing the software commands, the hardware modules collect the channel state and transmission performance parameters in real-time and feed them back to the edge computing decision unit. The software algorithm then dynamically corrects and optimizes link costs, compensation factors, and relay paths based on the feedback parameters, forming a closed-loop linkage and continuous optimization between hardware and software.

[0109] The self-healing method of this solution is implemented through five steps: multimodal perception, business dimensionality reduction processing, neighborhood path discovery, topology self-healing reconstruction, and relay transmission, with an overall execution cycle of no more than 50ms.

[0110] like Figure 1 As shown, the self-healing process of this solution forms a complete closed-loop operation system. Multimodal perception, as the starting point of the process, is responsible for collecting operational data and environmental information from power communication edge devices; business dimensionality reduction processing extracts key business features and simplifies data dimensions, prioritizing the transmission of core services; when all local uplinks fail, the process enters the neighborhood path discovery phase, searching for available communication paths and computing resources in the vicinity; topology self-healing reconstruction adjusts the network topology according to the fault situation, generating the globally optimal relay path; finally, data transmission is completed through multi-node relay. The link status information collected during the relay transmission process is fed back to the multimodal perception and business dimensionality reduction processing stages in real time, achieving continuous dynamic optimization of the self-healing process.

[0111] The multimodal sensing step, with a 10ms cycle, synchronously collects physical layer parameters of each channel and the remaining power of the gateway, constructs an observation matrix, calculates a dynamic weight vector, adjusts priority coefficients according to the current transmission service type, and calculates the comprehensive link cost of the main link. When the comprehensive link cost of the main link exceeds a threshold of 0.6, a service dimensionality reduction step is triggered to prioritize the current service and calculate the available bandwidth and the bandwidth required by the service. If bandwidth is sufficient, it directly switches to the heterogeneous link with the lowest comprehensive link cost; if bandwidth is insufficient, it performs edge data compression on non-core services, with the compression ratio dynamically adjusted according to the bandwidth gap, releasing bandwidth resources for core services.

[0112] When the combined link cost of all heterogeneous links on the local network exceeds a threshold, all uplinks on the local network are deemed to have failed, triggering the neighborhood path discovery step. The low-power wireless module broadcasts a beacon in the 433MHz band, scans neighboring gateway nodes within a 500m physical distance, collects the remaining battery power and real-time bandwidth utilization of neighboring nodes, calculates battery life using a battery life prediction model, filters out effective nodes with a battery life greater than or equal to 48 hours, obtains the uplink availability of effective nodes, and constructs a network topology adjacency matrix based on the local node and effective nodes.

[0113] The topology self-healing reconfiguration step uses minimizing the total comprehensive link cost as the core criterion, employs an improved Dijkstra's algorithm, introduces a Lagrange optimization model, and sets constraints such as path latency not exceeding 100ms and node remaining power not less than 30%. An objective function is constructed and iteratively solved until convergence, yielding the globally optimal relay path sequence. The edge computing decision unit synchronizes the path sequence to all participating relay nodes, completing the mesh topology self-healing reconfiguration.

[0114] In the relay transmission process, the edge computing decision unit encapsulates the core power business data, adds path identifiers and heartbeat detection fields, and sends the data to the first relay node via a low-power wireless module according to the path sequence. Each relay node then relays the data sequentially until it is uploaded to the cloud master station. During transmission, all nodes perform heartbeat detection every 10ms, collect link parameters in real time, and update the overall link cost. If the overall link cost of a relay link exceeds a threshold, path recalculation is immediately triggered, dynamically adjusting the relay path to ensure data transmission continuity.

[0115] In summary, this solution addresses the pain points of strong electromagnetic interference, easy link interruption, and limited edge computing power at the end of power distribution networks. It proposes a deeply coupled hardware and software self-healing gateway and method for power communication edge computing. Through a three-level cascaded protection system (GDT-PTC-TVS) and a high-permeability isolation transformer, hardware interference immunity and electromagnetic isolation are achieved in the power environment. Furthermore, through an active noise monitoring circuit coupled with hardware and software and a channel compensation model, active gain and phase compensation for the power line carrier channel are achieved, significantly reducing the signal transmission bit error rate under strong interference.

[0116] This solution constructs a comprehensive link cost assessment model with multi-dimensional dynamic weights, making link assessment more scientific and aligned with the needs of power services. It proposes a two-layer self-healing system at the link level and the entire network level. In the event of a single link failure, core transmission is guaranteed through seamless switching of heterogeneous links and service compression. In the event of a local full link failure, relay transmission is achieved through cross-node Mesh networking, completely solving the problem of "offline islands" in the power industry. It introduces battery life prediction and power constraint relay screening to avoid secondary network failures. All core decisions are completed locally at the edge, without relying on cloud computing power, thus improving the security and reliability of power communication.

[0117] The overall execution cycle of this solution does not exceed 50ms, meeting the latency requirements of core power business. It can be directly deployed in harsh industrial environments such as power distribution network areas and distribution boxes, providing reliable communication guarantees for the safe and stable operation of the power system.

[0118] This invention is not limited to the above-described preferred embodiments. Anyone inspired by this invention can derive other forms of self-healing gateways and methods for power communication edge computing based on hardware and software coupling. All equivalent variations and modifications made within the scope of the claims of this invention shall fall within the scope of this invention.

Claims

1. A self-healing gateway for power communication edge computing power based on hardware and software coupling, characterized in that: The edge computing decision unit is connected to the multimodal communication unit and the hardware adaptive protection unit via a bus, and is configured as follows: Based on the collected physical layer parameters, the comprehensive link cost of each heterogeneous link is calculated by using dynamic evolution evaluation weights based on information entropy and priority coefficients associated with power business types. When the overall link cost of the main link exceeds the preset threshold, the multi-mode communication unit is controlled to switch to the available heterogeneous link with the lowest overall link cost based on the relationship between the available bandwidth of the link and the bandwidth required by the service. The non-core services are then subjected to hierarchical compression to ensure the transmission of core services. When the combined link cost of all local uplink heterogeneous links exceeds a preset threshold, the cross-node Mesh networking self-healing process is triggered. The multimodal communication unit scans the surrounding adjacent gateway nodes to obtain the status parameters of the adjacent nodes, and an improved path optimization algorithm is executed based on the network topology to generate the optimal relay path that simultaneously satisfies the end-to-end transmission delay constraint and the node's remaining power constraint. The multimodal communication unit is controlled to transmit core power service data along a selected link or optimal relay path; The improved path optimization algorithm is configured to: convert the time delay constraint and power constraint into penalty terms using the Lagrange multiplier method, construct a Lagrange comprehensive cost function that includes the comprehensive path cost, time delay penalty and power penalty; and introduce double hard constraint verification during the node relaxation update process, only updating the path cost and node state when the candidate path simultaneously satisfies the time delay constraint and power constraint.

2. The power communication edge computing power self-healing gateway based on hardware and software coupling according to claim 1, characterized in that: The edge computing decision unit is also used to run a variational spatiotemporal evolution prediction model. The input of the model is a historical noise voltage sequence and a power line impedance parameter sequence, and the output is the predicted mean and variance of the noise voltage at future times. The edge computing decision unit calculates the optimal gain based on the prediction results to achieve pre-compensation of the power line carrier channel. The dynamic evolution evaluation weight based on information entropy is calculated by the information entropy method. In each sampling period, the weight value is dynamically adjusted according to the dispersion of the data in each evaluation dimension. The weight value is positively correlated with the dispersion of the dimension data.

3. The power communication edge computing power self-healing gateway based on hardware and software coupling according to claim 1, characterized in that: The improved path optimization algorithm also includes a pre-screening step for invalid nodes, which performs three-layer filtering on all nodes based on battery life, signal strength, and communication distance to generate a mask of valid nodes. The algorithm performs dynamic re-optimization every 10ms, comparing the combined cost of the new path with the current running path. If the cost difference exceeds 10% or the path changes, the algorithm switches to the new path.

4. The power communication edge computing power self-healing gateway based on hardware and software coupling according to claim 3, characterized in that: The specific implementation of tiered compression for non-core services is as follows: power services are divided into four levels. Level 1 services are prohibited from compression, Level 2 services are only allowed to be compressed without loss, Level 3 services are allowed to be compressed without loss up to 30%, and Level 4 services are allowed to be compressed with loss or without loss up to 90%. The compression ratio is dynamically adjusted according to the ratio of available bandwidth of the link to the bandwidth required by the service, and is updated every 10ms.

5. A power communication edge computing power self-healing gateway based on hardware and software coupling according to claim 1, characterized in that: The status parameters of the adjacent nodes are obtained, including the remaining power and real-time bandwidth utilization. The edge computing decision unit calculates the remaining battery life of the adjacent nodes through the battery life prediction model and selects nodes with a remaining battery life of not less than 48 hours as valid relay nodes.

6. The power communication edge computing power self-healing gateway based on hardware and software coupling according to claim 1, characterized in that: The multimodal communication unit integrates multiple heterogeneous communication modules to collect physical layer parameters of each communication channel in real time and achieve seamless switching of heterogeneous links according to instructions. The hardware adaptive protection unit includes a multi-stage surge protection circuit, an isolation transformer, an active noise monitoring circuit, and a power amplifier cascaded between the power line inlet and the power line carrier baseband chip. The active noise monitoring circuit is configured to acquire instantaneous power line interference voltage and impedance parameters in a manner synchronized with the symbol transmission time slot of the power line carrier baseband chip. The power amplifier is configured to receive instructions from the edge computing decision unit and perform dynamic adjustments to the transmission gain and phase.

7. A power communication edge computing power self-healing gateway based on hardware and software coupling according to claim 6, characterized in that: The sampling period of the active noise monitoring circuit is no greater than 12.5ns, and the synchronization error with the symbol transmission time slot of the power line carrier baseband chip is no more than 1μs. The sampled data is transmitted to the edge computing decision unit through the LVDS interface in the direct memory access mode, and the data format is 16-bit signed integer plus 32-bit microsecond-level timestamp. The multi-stage surge protection circuit is a three-stage cascaded structure consisting of a GDT ceramic gas discharge tube, a PTC resettable fuse, and a TVS transient suppression diode array. The cascaded sequence is as follows: power line inlet, GDT ceramic gas discharge tube, PTC resettable fuse, TVS transient suppression diode array, isolation transformer, and power line carrier baseband chip. The inter-electrode capacitance of the GDT ceramic gas discharge tube does not exceed 1.5pF, the parasitic capacitance of the PTC resettable fuse does not exceed 100pF, and the parasitic capacitance of the TVS transient suppression diode array does not exceed 80pF.

8. A power communication edge computing self-healing gateway based on hardware and software coupling according to claim 1, characterized in that: The edge computing decision unit is also configured to: after controlling data transmission, collect the channel state and transmission performance parameters after the decision is executed, and dynamically correct the comprehensive link cost, optimal relay path and channel compensation parameters based on the feedback parameters to form a continuously optimized closed-loop control.

9. A power communication edge computing power self-healing gateway based on hardware and software coupling according to claim 1, characterized in that: When the improved path optimization algorithm fails to find an effective relay path that satisfies the dual hard constraints, the edge computing decision unit triggers a local data temporary storage mechanism to store core power business data in local non-volatile memory, which will be automatically uploaded to the cloud master station after the link is restored.

10. A self-healing method for edge computing power in power communication based on hardware and software coupling, characterized in that, Operating on the power communication edge computing self-healing gateway according to any one of claims 1 to 9 includes the following steps: The physical layer parameters of each heterogeneous communication channel are collected in the first cycle, and the instantaneous interference voltage and impedance parameters of the power line are collected in a way that is synchronized with the symbol transmission time slot of the power line carrier baseband chip. Based on the collected physical layer parameters, the comprehensive link cost of each heterogeneous link is calculated by using dynamic evolution evaluation weights based on information entropy and priority coefficients associated with power business types. When the overall cost of the main link exceeds a preset threshold, the system switches to the available heterogeneous link with the lowest overall cost based on the relationship between the available bandwidth and the bandwidth required by the service. The system also performs tiered compression on non-core services to ensure the transmission of core services. When the combined link cost of all local uplink heterogeneous links exceeds a preset threshold, scan the surrounding adjacent gateway nodes and obtain their status parameters to construct the network topology. An improved path optimization algorithm is used to solve for the globally optimal relay path. The algorithm transforms the end-to-end transmission delay constraint and the node remaining power constraint into penalty terms using the Lagrange multiplier method, constructs the Lagrange comprehensive cost function, and introduces double hard constraint verification during the node relaxation update process. The path cost and node status are updated only when the candidate path simultaneously satisfies the delay constraint and the power constraint. Core power service data is transmitted along the selected link or the optimal relay path, and the link status is monitored in a second cycle to dynamically update the comprehensive link cost and self-healing decision.

Images