Fault intelligent switching and self-healing method for electric power communication network integrating optical fiber and wireless

Abstract

The invention discloses an optical fiber and wireless fused power communication network fault intelligent switching and self-healing method, which comprises the following steps: constructing a multi-mode link health portrait, arranging four-in-one sensors at all nodes, and generating a health vector through edge calculation; mining a double-domain fault propagation chain, constructing a dynamic heterogeneous graph, and predicting fault propagation through memory attenuation GNN; pre-generating a hybrid redundant path, and selecting a main and standby path pair based on a game algorithm; establishing a programmable self-healing decision engine, and outputting an adaptive instruction; and zero-interruption service migration and closed-loop verification are realized. The method solves the problems of lack of physical constraints, insufficient dynamic adaptation and low efficiency of edge deployment in the prior art, and improves the reliability of the power communication network.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of power communication network operation and fault processing, and particularly relates to a power communication network fault intelligent switching and self-healing method combining optical fibers and wireless. BACKGROUND

[0002] The existing power communication network adopts a hybrid networking mode mainly using optical fibers and supplemented by wireless to meet the requirements of wide-area coverage and high reliability. In actual operation, the optical fiber link relies on a fixed topology and is easily affected by physical damage (such as optical cable fracture and vibration interference); the wireless link faces stability problems caused by environmental factors such as radio frequency interference and temperature and humidity fluctuations, and the faults of the two types of links may cause risks such as interruption of power dispatching instruction transmission and loss of monitoring data.

[0003] The existing technology has obvious defects in dealing with the above problems: first, it lacks multi-dimensional real-time perception of the link health state and only relies on a single index (such as optical power or signal strength) to judge the fault, which is easy to cause misjudgment due to the lack of physical constraints; second, it has insufficient fault propagation path prediction capability and cannot lock the potential failure area in advance, has poor dynamic adaptability, and is often only passively responded after the fault occurs; third, the path generation and service migration during fault switching are inefficient, rely on manual configuration or fixed strategies, and are difficult to meet the "zero interruption" requirement of power services, and the resource occupation and time delay problems are prominent in edge deployment.

[0004] Therefore, the application provides a power communication network fault intelligent switching and self-healing method combining optical fibers and wireless. SUMMARY

[0005] In view of the defects in the prior art, the application aims to solve the problems of the lack of physical constraints, insufficient dynamic adaptation and low efficiency in edge deployment in the prior art, and to realize accurate fault perception, early warning, rapid switching and self-healing, improve network reliability and provide support for dispatching.

[0006] To achieve the above purpose, the application adopts the following technical solutions.

[0007] The power communication network fault intelligent switching and self-healing method combining optical fibers and wireless comprises the following steps:

[0008] S1. Multi-modal link health portrait construction, a four-in-one sensor of micro optical power, radio frequency interference, temperature and humidity, and vibration is arranged at each node, and after sampling, the four types of original signals are processed by an event-driven edge computing unit (FPGA); the abnormal weight after processing is fused with the basic features to generate a "link health portrait" vector in real time;

[0009] S2. Dual-domain fault propagation chain mining, based on link health portrait, build dynamic model to mine potential fault propagation path, and lock the failure risk area in advance;

[0010] S3. Hybrid redundancy path real-time pre-generation, based on fault propagation chain, generate master and backup path pairs to ensure quick switching in case of failure;

[0011] S4. Programmable self-healing decision engine, build dynamic decision model, output self-healing instructions adapted to the scene;

[0012] S5. Zero-interruption service migration and closed-loop verification, according to the self-healing instructions output by S4, through double buffering mechanism and hierarchical migration strategy, realize business non-aware switching, and based on the verification result, continuously optimize the decision model.

[0013] Further, the four types of original signals include optical power signals, radio frequency interference signals, temperature and humidity signals, and vibration signals.

[0014] Further, the event-driven edge computing unit (FPGA) processes the four types of original signals, including:

[0015] S111. Joint feature extraction, extract time domain and frequency domain features from four types of signals respectively, and then obtain fusion feature vectors through feature splicing;

[0016] S112. Abnormal weight scoring, give each dimension in the joint feature an abnormal weight (reflecting the degree of deviation from the normal range of that dimension), the calculation formula is:

[0017]

[0018] Where, x i is the measured value of the i-th feature, is the mean value of the feature under normal operating conditions; σ i is the standard deviation of the feature under normal conditions, used to measure the degree of dispersion of the data; α i is the importance coefficient of the feature, which is determined according to the degree of influence of each type of signal on the link health;

[0019] S113. Spatio-temporal alignment, time stamp calibration and spatial coordinate matching of features collected at different nodes and different times to eliminate spatio-temporal deviation.

[0020] Further, the link health portrait is a 6-dimensional link health vector H, which is refreshed every second, and H is defined as:

[0021] H=[ω o ,ω r ,ω t ,ω h ,ω v,t up ];

[0022] Where, ω o ω r ω t ω h ω v These are the abnormal weights for optical power, radio frequency interference, temperature, humidity, and vibration, respectively; t up To update the timestamp, accurate to the second, the moment the image vector was generated is recorded.

[0023] Furthermore, the step of constructing a dynamic model based on link health profiles to mine potential fault propagation paths includes:

[0024] S211. Dynamic heterogeneous graph construction: The link health profile H generated in stage S1 is used as the core attribute of the node. Fiber optic links are regarded as "hard connection edges" and wireless links are defined as "soft connection edges" to construct the topology graph.

[0025] To accurately reflect changes in network state over time, a timestamp is added to each node and edge in the topology graph, resulting in the following dynamic heterogeneous graph:

[0026] G(t) = (V, E, T);

[0027] V is the node set, containing all communication nodes and their corresponding link health profiles; E is the edge set, describing the fiber optic or wireless connection relationships between nodes; T is the time set, recording the time information of each node and edge state.

[0028] S212. Memory decay graph neural network modeling and forward propagation: A graph neural network with memory decay (GNN) is used to perform rolling forward calculation on the "node-edge-time" triple to simulate the fault propagation trend;

[0029] In graph neural networks with memory decay, the memory decay mechanism is as follows: a memory factor λ(t) is introduced, and the influence of historical states decays over time, as shown in the formula:

[0030]

[0031] β is the attenuation coefficient, t c t represents the current time; p For historical time;

[0032] Rolling forward computation is performed on the "node-edge-time" triple: the node state is updated through the GNN layer, and the probability matrix P∈R of cascading failures occurring in each node within the future time period Δt is output. N×N In the probability matrix P ij This represents the probability that a failure in node i causes a failure in node j (N is the total number of nodes);

[0033] S213. Fault propagation chain screening: Set a probability threshold and retain P. ij Edges with a probability greater than or equal to a set probability threshold are filtered out as low-probability associations; starting from the initial fault node, the propagation path is traced backwards, and chains with a length exceeding 5 hops are removed (ensuring that the chain length is ≤ 5 hops); finally, a set of "dual-domain fault propagation chains" is obtained, which is represented as:

[0034] C = {c1, c2, ..., c} k ,…,c m};

[0035] Among them, each chain c k A sequence of nodes is represented as:

[0036] c k ={v k1 ,v k2 ,…,v kL (L≤5).

[0037] Furthermore, the step of generating a primary / backup path pair based on the fault propagation chain includes:

[0038] S311. Redundant path generation: Based on the fault propagation chain in S2, generate K candidate redundant paths in the fiber optic domain and the wireless domain respectively.

[0039] The fiber domain uses "wavelength-time slot joint coding" to generate K candidate paths, with the coding format being (λ,τ), where λ is the wavelength number and τ is the time slot number. Conflicts are avoided by combining different wavelengths and time slots.

[0040] The wireless domain uses "frequency hopping-power joint coding" to generate K candidate paths, with the coding format being (f, p), where f is the frequency hopping frequency and p is the transmit power, and anti-interference is achieved through dynamic frequency hopping and power adjustment;

[0041] S312. Lightweight game theory algorithm for resource bidding: Under the three-dimensional constraints of "latency-bandwidth-energy consumption", a lightweight game theory algorithm is designed to select the primary and backup path pair that satisfies the three-dimensional constraints of "latency-bandwidth-energy consumption" and has the lowest cost from the candidate paths.

[0042] In the lightweight game theory algorithm, the cost function C is defined as a weighted sum of three-dimensional indicators:

[0043] C=ω D ·D+ω B ·(B max -B)+ω E ·E;

[0044] D is the transmission delay of the path; B is the bandwidth currently available for the path; E is the energy consumption of the path when transmitting data; B max The maximum available bandwidth that the network can provide; ω D ω B ω E ω is the weighting coefficient. D +ω B +ω E =1;

[0045] S313. Switching trigger condition setting: A dual threshold triggering mechanism is adopted. When the health profile score of any node in the main path is less than the health threshold, or the failure probability of the propagation chain in which the main path is located is greater than the probability threshold, the path switching is triggered.

[0046] Furthermore, the step of constructing a dynamic decision-making model and outputting self-healing instructions adapted to the scenario includes:

[0047] S411. Decision tree model construction: Map the primary and backup path pairs generated in S3 to the leaf nodes of the decision tree. Use the link health profile H as the input feature and the path availability as the output label to train the decision tree and achieve a fast mapping from input to path status.

[0048] S412. Reinforcement Learning Meta-Policy Network Update: A meta-policy network is designed based on reinforcement learning, with the reward function as follows:

[0049] R′=γ·success-(1-γ)·delay;

[0050] γ∈[0,1] represents the weight; success is the handover success identifier; delay is the handover delay penalty; the network updates the weights online every 5 seconds.

[0051] S413. Self-healing command output: When the switching conditions set by S3 are met, the engine immediately generates a self-healing command to dynamically optimize the decision-making logic.

[0052] Self-healing instructions include:

[0053] Fiber priority, prioritize the use of the fiber backup path, parameter is (λ) new τ new , routing table o) where λ new Number the new wavelength, τ new The new time slot number is used, and the routing table o contains the node hopping sequence of the fiber optic path;

[0054] Wireless priority, prioritizes the use of the wireless backup path, parameter is (f new p new , routing table r), where f new For the new frequency hopping frequency, p newFor the new transmit power, the routing table r contains the node hop sequence of the wireless path;

[0055] Hybrid collaboration enables both fiber optic and wireless paths to share the load, with parameters being a combination of the two types of parameters mentioned above.

[0056] Furthermore, the process of seamless service switching is achieved through a double-buffering mechanism and a tiered migration strategy.

[0057] S511. Dual-buffered zero-copy data migration channel is enabled. A dual-buffered channel is established between the source node and the destination node. The primary buffer receives current business data in real time, while the backup buffer preloads the data to be migrated. The "zero-copy" technology is adopted to skip the CPU data copying step through DMA transfer, thereby reducing migration latency.

[0058] S512. Business-level migration: Data is migrated in a tiered manner according to business priority through a dual-buffered channel. Specific rules are as follows:

[0059] Priority 1 (Protection-related services): including relay protection signals, security control commands, etc., requiring migration latency ≤10ms; priority is given to occupying the dual-buffer channel bandwidth through preemptive scheduling, allocating ≥60% of the bandwidth to ensure migration is completed in milliseconds;

[0060] Priority 2 (Monitoring services): including telemetry data, video surveillance streams, etc., with a migration latency of ≤100ms allowed; after the migration of priority 1 services is completed, 30% of the bandwidth will be used for transmission to achieve second-level migration;

[0061] Priority 3 (Management Services): Includes file transfer, remote configuration commands, etc., with a migration latency of ≤60s allowed; asynchronous migration is carried out using remaining bandwidth to ensure that high-priority services are not affected.

[0062] Furthermore, the method for continuously optimizing the decision model based on the verification results:

[0063] After the migration is complete, data integrity is verified by reverse sequence code injection, and decision parameters are optimized based on the results:

[0064] Reverse sequence code injection, the destination node generates a random sequence code: S = [s1, s2, ... s i …,s n (n = 1024, each s) i The received sequence (a 32-bit integer) is sent to the source node via the reverse channel (new path). The source node compares the received sequence with its local backup sequence and calculates the matching rate. If M=1, the verification is considered successful; if M<1, the verification is considered unsuccessful.

[0065] Result processing:

[0066] If M < 1 (verification failed), the rollback mechanism is immediately triggered to switch the business back to the original path and raise the health threshold and probability threshold in S3 to reduce the probability of false switching.

[0067] If M=1 (verification successful), the current link health profile H, fault chain, and migration parameters are written into the experience base for use by the reinforcement learning meta-policy network of S4 during updates, thus realizing the self-evolution of decision logic.

[0068] In summary, due to the adoption of the above technical solution, the beneficial technical effects of the invention are as follows:

[0069] First, by constructing a multimodal link health profile, the time and frequency domain characteristics of four types of signals, including optical power and radio frequency interference, are integrated, and anomalies are assigned weights and spatiotemporal alignment is achieved. This solves the problem of the lack of physical constraints in the existing technology for judging faults with a single indicator, making the description of the link health status more comprehensive, providing accurate input for subsequent decision-making, and reducing the misjudgment rate.

[0070] Secondly, based on dynamic heterogeneous graph and memory decay graph neural network mining of dual-domain fault propagation chain, combined with hybrid redundant path pre-generation and lightweight game algorithm, it realizes the early prediction of potential faults and optimal path planning, overcomes the shortcomings of insufficient dynamic adaptation of existing technology, can actively adapt to the link state changes in complex environments such as wind field, and provide stable path support for scheduling command transmission.

[0071] Finally, a programmable self-healing decision engine and a zero-disruption service migration mechanism, combined with decision trees and reinforcement learning, enable dynamic decision-making. Double-buffered zero-copy technology and a hierarchical migration strategy ensure seamless service switching, resolving the inefficiency of edge deployment. Simultaneously, closed-loop verification and experience-based iterative mechanisms continuously optimize the decision logic, significantly improving the success rate and efficiency of fault self-healing, meeting the stringent requirements of power communication networks for high reliability and low latency. Attached Figure Description

[0072] Figure 1 This is a flowchart of a method for intelligent fault switching and self-healing in power communication networks that integrates fiber optic and wireless technologies. Detailed Implementation

[0073] To make the objectives, technical solutions, and advantages of the invention clearer, the invention will be further described in detail below with reference to embodiments. It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the scope of the invention.

[0074] like Figure 1 As shown, a method for intelligent fault switching and self-healing in power communication networks that integrates fiber optic and wireless technologies includes the following steps:

[0075] S1. Multimodal link health profile construction: Miniature optical power, radio frequency interference, temperature and humidity, and vibration four-in-one sensors are deployed at all nodes. The event-driven edge computing unit (FPGA) performs joint feature extraction, anomaly weighting and spatiotemporal alignment processing on these four types of raw signals to generate a "link health profile" vector in real time for subsequent decision input.

[0076] For sensor selection, an optical power sensor is used to accurately measure the optical power signal in the fiber optic link, with a measurement accuracy of ±0.1dBm, which is crucial for evaluating the transmission quality of the fiber optic link. An RF interference sensor focuses on capturing RF interference signals around the wireless link, with a sensitivity of up to 120dBm, enabling timely detection of weak interference sources. High-precision temperature and humidity sensors are selected, with temperature measurement errors controlled within ±0.5℃ and humidity measurement errors not exceeding ±3%RH, to accurately reflect the impact of ambient temperature and humidity on the communication link. An acceleration-type vibration sensor can detect vibrations as low as 0.01m / s². 2 The vibration changes can be effectively monitored to detect physical disturbances.

[0077] The installation layout is designed to rationally plan the sensor placement based on the equipment structure and surrounding environment of different nodes. At fiber optic nodes, optical power sensors are tightly connected next to the fiber optic connectors to ensure direct and accurate acquisition of optical signal strength; radio frequency interference sensors are installed near the wireless antennas to maximize the detection of interference signals within the wireless frequency band. Temperature and humidity sensors are installed near the equipment's ventilation openings for quick and accurate measurement of environmental parameters around the equipment; vibration sensors are fixed to the equipment's support structure to effectively capture vibrations caused by external forces or internal faults.

[0078] To achieve real-time monitoring of the link status, the sensor sampling frequency is set to 5Hz, meaning 5 data acquisitions per second. This high sampling frequency allows for timely detection of subtle changes in the link status, ensuring the high timeliness of the subsequently generated link health profile. The collected data is transmitted in real-time to the edge computing unit via a high-speed data transmission line, awaiting further processing.

[0079] The event-driven edge computing unit (FPGA) performs joint feature extraction, anomaly weighting, and spatiotemporal alignment on these four types of raw signals, specifically including:

[0080] S121. Joint feature extraction: In-depth time-domain (e.g., mean, variance, peak value) and frequency-domain (e.g., peak value, bandwidth) feature extraction is performed on the four types of sensor signals respectively.

[0081] In the time domain, for the optical power signal, its mean μ is calculated. o The formula is:

[0082]

[0083] Among them, P o,i Let n be the optical power value collected in the i-th sampling, and n be the number of samplings.

[0084] The variance formula is:

[0085]

[0086] Variance can reflect the degree of fluctuation in optical power signals;

[0087] The same mean and variance calculation methods are used for radio frequency interference signals, temperature and humidity signals, and vibration signals.

[0088] In the frequency domain, the time-domain signal is converted into a frequency-domain signal using Fast Fourier Transform (FFT) to obtain features such as spectral peaks and bandwidth. For frequency-domain feature extraction of optical power signals, the acquired optical power time-domain sequence P is first processed. o (t) is subjected to FFT transformation to obtain the frequency domain representation P. o (f), and then the spectral peak p is found through a search algorithm. max,o :

[0089] p max,o =max(|P0(f)|);

[0090] Finally, the time-domain and frequency-domain features of the four types of signals are concatenated to form a joint feature vector containing rich information.

[0091] S122. Anomaly Weight Assignment: To accurately assess the degree of anomaly in each feature dimension, the concept of anomaly weight is introduced. For each dimension in the joint features, an anomaly weight ω is assigned using the following formula. i :

[0092]

[0093] Where, x i Let be the measured value of the i-th feature. This is the mean of the characteristic under normal operating conditions, which is obtained through statistical analysis of a large amount of historical normal data; σ i α represents the standard deviation of this feature under normal conditions, used to measure the dispersion of the data; i The importance coefficient for this feature is determined based on the degree of impact of various signals on link health. For example, optical power plays a decisive role in the normal operation of fiber optic links, so its importance coefficient can be set to 0.4; radio frequency interference has a significant impact on wireless links, so it is set to 0.3; temperature, humidity, and vibration have relatively small impacts on the link, so they are set to 0.1; ω iThe larger the value, the greater the deviation of the feature from the normal range, that is, the more severe the abnormality.

[0094] S123. Spatiotemporal alignment: Since the timing of data collection by sensors at different nodes may vary slightly, and the nodes are also located in different spatial positions, spatiotemporal alignment is required to ensure data consistency and comparability. In terms of time, high-precision clock synchronization technologies, such as the Global Positioning System (GPS) or Network Time Protocol (NTP), are used to calibrate the timestamps of sensors at each node, unifying the time base of all data. In terms of space, a unique geographic coordinate is assigned to each node, and a mapping relationship between node location and communication link is established. In this way, feature data collected from different nodes at different times are integrated according to a unified spatiotemporal standard, eliminating spatiotemporal bias and providing an accurate data foundation for subsequent analysis and modeling.

[0095] The link health profile vector is a vector H with 6 or more dimensions, where H is defined as:

[0096] H = [ω] o ,ω r ,ω t ,ω h ,ω v ,t up ];

[0097] Where, ω o ω r ω t ω h ω v These are the abnormal weights for optical power, radio frequency interference, temperature, humidity, and vibration, respectively. These weights intuitively reflect the degree of impact of each factor on link health; up To update the timestamp, accurate to the second, the generation time of this profile vector is recorded. To ensure real-time tracking of link health status, the vector is refreshed every second, meaning a new link health profile vector is generated every second to promptly reflect dynamic changes in link status.

[0098] S2. Dual-domain fault propagation chain mining: Using the link health profile of S1 as node attributes and the fiber-wireless topology as edges, a dynamic heterogeneous graph is constructed. A graph neural network with memory decay is used to simulate forward propagation of the node-edge-time triplet, and outputs the probability matrix of cascading failures occurring at each node within a future time period Δt (e.g., Δt = 0.5s). Threshold truncation and backtracking pruning are performed on this matrix to obtain a set of "dual-domain fault propagation chains" with a chain length ≤ 5 hops, which are used to identify potential failure areas in advance.

[0099] S211. Dynamic heterogeneous graph construction: The link health profile H generated in stage S1 is used as the core attribute of the node. Fiber optic links are regarded as "hard connection edges" and wireless links are defined as "soft connection edges" to construct the topology graph.

[0100] Each node represents a communication node in the power communication network, and its health status is quantitatively described through a link health profile vector. In terms of edge definition, fiber optic links are considered "hard-connected edges" because fiber optic communication is relatively stable and reliable, and its connection relationship is relatively fixed within a certain period. Wireless links, on the other hand, are defined as "soft-connected edges" because wireless communication is easily affected by environmental factors, and the link status exhibits significant dynamic changes.

[0101] To accurately reflect changes in network state over time, a timestamp is added to each node and edge in the topology graph, resulting in the following dynamic heterogeneous graph:

[0102] G(t) = (V, E, T);

[0103] V is the node set, containing all communication nodes and their corresponding link health profiles; E is the edge set, describing the fiber optic or wireless connection relationships between nodes; T is the time set, recording the time information of each node and edge state.

[0104] In this way, it is possible to capture the dynamic evolution of network topology and link health status at different times.

[0105] S212. Memory decay graph neural network modeling and forward propagation: A graph neural network with memory decay (GNN) is used to perform complex rolling forward calculations on the "node-edge-time" triples in order to accurately simulate the propagation trend of faults in the network.

[0106] The memory decay mechanism introduces a memory factor λ(t) to simulate the characteristic that a network's memory of historical states gradually weakens over time; the formula for calculating the memory factor is:

[0107]

[0108] Where β is the attenuation coefficient, which is adjusted according to the dynamic characteristics of the actual network, and generally takes a value between 0.1 and 0.5; t c t represents the current time; p For historical time;

[0109] Forward propagation computation, through the multi-layered structure of the Generative Neural Network (GNN), progressively updates the node states. In each layer of the GNN, nodes pass information and update their states based on the states of their neighbors and the weights of edges. After multiple layers of computation, the output is a probability matrix P∈R representing the probability of cascading failures occurring at each node within a future time interval Δt (e.g., Δt = 0.5s).N×N , where P ij This represents the probability that a failure in node i will cause a failure in node j (N is the total number of nodes in the network). During the calculation, the parameters of the GNN model are trained and optimized using a large amount of historical data to improve the accuracy of predicting the probability of failure propagation.

[0110] S213. Fault propagation chain screening: To filter out low-probability fault propagation associations, a probability threshold is set, and P is retained in the probability matrix. ij Edges with a probability greater than or equal to a set probability threshold represent fault propagation paths with a high probability, while edges below the threshold are considered unlikely to cause fault propagation under the current circumstances and are removed.

[0111] Backtracking pruning involves tracing back along the preserved edges from the initial fault node to construct a fault propagation chain. To avoid excessively long propagation chains that would lead to overly complex analysis and limited practical significance, the maximum length of the propagation chain is set to 5 hops. This means that propagation paths that pass through a maximum of 5 nodes from the initial fault node are preserved, while those exceeding 5 hops are pruned. After these two steps, the final set of "two-domain fault propagation chains" C = {c1, c2, ..., c...} is obtained. k ,…,c m}, where each chain c k A sequence of nodes is represented as: c k ={v k1 ,v k2 ,…,v kL (L≤5) clearly shows the possible fault propagation paths and the nodes involved.

[0112] S3. Real-time pre-generation of hybrid redundancy paths: Based on the fault propagation chain in S2, K candidate redundant paths are generated in the fiber domain and the wireless domain respectively; a lightweight game theory algorithm is used to auction resources between the two domains to select the primary and backup path pair that meets the three-dimensional constraints of "latency-bandwidth-energy consumption" and has the lowest cost.

[0113] S311. Redundant path generation:

[0114] In the fiber optic domain, a wavelength-time slot joint coding technique is used to generate K candidate paths. In fiber optic communication, different wavelengths and time slots can be used to distinguish different communication channels and avoid signal collisions. The coding format is (λ, τ), where λ is the wavelength number. Assuming the fiber optic system supports 100 different wavelengths, i.e.

[0115] λ∈(1,100); τ is the time slot number, and each wavelength can be divided into 60 time slots, i.e., τ∈[1,60]. By combining different wavelengths and time slots, a large number of candidate paths can be generated, for example, (λ=5, τ=10).

[0116] This represents the communication path established on the 10th time slot of the 5th wavelength. When generating candidate paths, factors such as the bandwidth and loss of the fiber optic link are considered, and an optimization algorithm is used to select a suitable wavelength-time slot combination to ensure the quality and availability of the path.

[0117] The wireless domain uses a "frequency hopping-power joint coding" technique to generate K candidate paths. The coding format is (f, p), where f is the frequency hopping frequency. Within the commonly used wireless frequency band of 2.4 GHz to 5 GHz, multiple frequency hopping channels are divided according to actual needs, such as one channel every 10 MHz.

[0118] f∈[2.4GHz,2.41GHz,2.42GHz,…,5GHz];

[0119] p represents the transmit power. Based on the performance of the wireless device and the communication distance requirements, the transmit power range is set to p∈[10dBm,30dBm]. By dynamically adjusting the frequency hopping frequency and transmit power, interference in wireless communication can be effectively resisted, and the reliability of communication can be improved. For example, when there is strong interference in the 2.45GHz band in a certain area, the candidate path can choose to hop to other frequencies with less interference and appropriately increase the transmit power to ensure signal strength.

[0120] S312. Lightweight Game Theory Algorithm Resource Auction:

[0121] Under the three-dimensional constraints of "latency (D)-bandwidth (B)-energy consumption (E)", a lightweight game theory algorithm is designed to select the optimal primary and backup path pairs from the candidate paths.

[0122] The cost function, defined as the path cost C, is a weighted sum of three-dimensional indicators, and the formula is:

[0123] C=ω D ·D+ω B ·(B max -B)+ω E ·E;

[0124] in,

[0125] ω D ω B ω E ω is the weighting coefficient. D +ω B +ω E =1, and these weighting coefficients are adjusted based on the service's sensitivity to latency, bandwidth, and energy consumption. For example, for services with extremely high real-time requirements, ω = 1. D It can be set to 0.6; for services with high bandwidth requirements, ω B Set it to 0.5, etc.

[0126] B max D is the maximum available bandwidth that the network can provide; E is the transmission delay of the path, which can be obtained by measurement or estimation, such as by sending test data packets and recording the round-trip time; B is the bandwidth that the path can currently provide, which can be obtained by monitoring network traffic and link status; and E is the energy consumption when the path transmits data, which can be calculated based on the power parameters of the device and the transmission time.

[0127] The bidding rules treat each candidate path as a "bidder," participating in the auction with a cost C as their bid. During the bidding process, each candidate path calculates its cost C based on its latency, bandwidth, and energy consumption. The path with the lowest cost wins the bid and is selected as the optimal path. When selecting primary and backup paths, priority is given to ensuring that the primary and backup paths belong to different domains (i.e., one is an optical path, and the other is a wireless path). If the requirement of different domains cannot be met, a path within the same domain but using different resources (such as different wavelengths, different frequency hopping frequencies, etc.) is selected to minimize the possibility of simultaneous failure of the primary and backup paths due to a common cause of failure.

[0128] S313. Switching trigger condition setting: A dual-threshold triggering mechanism is adopted to ensure timely path switching when network failure risks occur; when the health profile score H of any node on the main path... score When the score falls below the health threshold, a switchover is triggered; health profile score H score The following is obtained by comprehensively calculating the anomaly weights in the link health profile vector H:

[0129]

[0130] Health thresholds are typically set based on historical data and business needs, such as a health threshold of 0.6. When H... score A value less than 0.6 indicates that the node's health condition is poor and may affect normal communication on the main path. Alternatively, when the failure probability P of the fault propagation chain to which the main path belongs... chain A switchover is also triggered when the probability threshold is exceeded. The failure probability P of the fault propagation chain. chain It is obtained by accumulating the probability of fault propagation among nodes on the chain.

[0131] S4. Programmable self-healing decision engine: This engine maps the primary / backup path pair of S3 to decision tree nodes, using link health profiles as input and path availability as output to construct a programmable self-healing decision engine. The engine is updated online based on a reinforcement learning-based meta-policy network. When the triggering condition is met, the engine outputs three types of self-healing instructions: "fiber priority," "wireless priority," or "hybrid collaboration." Specifically, these include:

[0132] S411. Decision tree model construction: The primary and backup path pairs generated in S3 (such as "fiber primary path + wireless backup path", "wireless primary path + fiber backup path", etc.) are mapped one by one to the leaf nodes of the decision tree. Each leaf node uniquely corresponds to a set of path pairs and is associated with the basic attributes of the path (such as the wavelength range of the fiber path and the frequency band range of the wireless path).

[0133] The input and output are the link health profile vector H generated by S1, and the output label is the path availability (0-unavailable, 1-available). The decision tree is trained to realize the fast mapping from input to path status. The output is the path availability label (0 indicates unavailable, 1 indicates available), which is used to determine whether the path can stably carry services.

[0134] S412. Reinforcement Learning Meta-Policy Network Update: Based on reinforcement learning, a meta-policy network is designed. Its core is to optimize decision-making logic through continuous interaction with the network environment, aiming to "maximize the success rate of path switching." The reward function is:

[0135] R′=γ·success-(1-γ)·delay;

[0136] γ∈[0,1] represents the weight; success is the handover success flag (1 for success, 0 for failure); delay is the handover delay penalty (the greater the delay, the higher the penalty); the network updates the weights online every 5 seconds.

[0137] S413. Self-healing command output: When the switching conditions set in S3 (when the health profile score of any node in the main path is less than the health threshold, or the failure probability of the propagation chain in which the main path is located is greater than the probability threshold) are met, the engine immediately generates a self-healing command to dynamically optimize the decision-making logic.

[0138] Self-healing instructions include:

[0139] Fiber priority, prioritize the use of the fiber backup path, parameter is (λ) new τ new , routing table o) where λ new Number the new wavelength, τ new The new time slot number is used, and the routing table o contains the node hopping sequence of the fiber optic path;

[0140] Wireless priority, prioritizes the use of the wireless backup path, parameter is (f new p new , routing table r), where f new For the new frequency hopping frequency, p new For the new transmit power, the routing table r contains the node hop sequence of the wireless path;

[0141] Hybrid collaboration enables both fiber optic and wireless paths to share the load, with parameters being a combination of the two types of parameters mentioned above.

[0142] S5. Zero-interruption service migration and closed-loop verification: Based on the self-healing instructions of S4, a double-buffered zero-copy data migration channel is enabled between the source and destination nodes, and the migration granularity is graded according to service priority. After the migration is completed, the integrity of the service is verified by reverse injection of known sequence codes, and the decision parameters are optimized based on the results.

[0143] Double-buffered zero-copy data migration channel enabled:

[0144] Deploy a double-buffered storage structure between the source node (the starting point of the original path) and the destination node (the ending point of the new path):

[0145] The main buffer receives and caches current service data (such as power monitoring messages and protection signals) in real time to ensure continuous transmission of services during switching without data loss.

[0146] The backup buffer preloads the business data to be migrated (pre-generated based on the path parameters in the S4 command) and forms a data synchronization mechanism with the main buffer (checking data consistency every 10ms).

[0147] Data migration is achieved using "zero-copy" technology. Through a DMA (Direct Memory Access) controller, data in the backup buffer is directly transferred to the receiver buffer of the destination node, skipping the CPU data copying step, reducing migration latency (to ≤5ms), and avoiding service interruption.

[0148] Business-level migration: Data is migrated in a tiered manner according to business priority, using a dual-buffered channel. Specific rules are as follows:

[0149] Priority 1 (Protection-related services): including relay protection signals, security control commands, etc., requiring migration latency ≤10ms; preemptive scheduling is used to prioritize the use of dual-buffer channel bandwidth (allocating ≥60% of the bandwidth) to ensure migration is completed in milliseconds;

[0150] Priority 2 (Monitoring services): including telemetry data, video surveillance streams, etc., with a migration latency of ≤100ms allowed; after the migration of priority 1 services is completed, 30% of the bandwidth will be used for transmission to achieve second-level migration;

[0151] Priority 3 (Management Services): Includes file transfer, remote configuration commands, etc., with a migration latency of ≤60s allowed; asynchronous migration is carried out using remaining bandwidth to ensure that high-priority services are not affected.

[0152] Feedback and optimization:

[0153] After the migration is complete, data integrity is verified by reverse sequence code injection, and decision parameters are optimized based on the results.

[0154] Reverse sequence code injection: The destination node generates random sequence codes:

[0155] S = [s1, s2, ... s i …,s n (n = 1024, each s) i The received sequence (a 32-bit integer) is sent to the source node via the reverse channel (new path). The source node compares the received sequence with its local backup sequence and calculates the matching rate. If M=1, the verification is considered successful; if M<1, the verification is considered unsuccessful.

[0156] If M < 1 (verification failed): immediately trigger the rollback mechanism (switch back to the original path, rollback time ≤ 20ms), switch the business back to the original path (rollback time ≤ 20ms), and at the same time raise the health threshold and probability threshold in S3 to reduce the probability of false switching;

[0157] If M=1 (verification successful): Write the current link health profile H, fault chain, and migration parameters (such as switching delay and bandwidth allocation) into the experience base for use by the reinforcement learning meta-policy network of S4 during updates, so as to realize the self-evolution of decision logic.

[0158] The above description is a preferred embodiment of the invention and is not intended to limit the scope of the invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the invention should be included within the scope of protection of the invention.

Claims

1. A method for intelligent fault switching and self-healing in power communication networks integrating fiber optic and wireless technologies, characterized in that: Includes the following steps: S1. Multimodal link health profile construction: Miniature optical power, radio frequency interference, temperature and humidity, and vibration four-in-one sensors are deployed at all nodes. After sampling, the four types of raw signals are processed by the event-driven edge computing unit FPGA. The processed anomaly weights are fused with basic features to generate a "link health profile" vector in real time; S2. Dual-domain fault propagation chain mining: Based on the link health profile, a dynamic model is built to mine potential fault propagation paths and identify failure risk areas in advance. S3. Real-time pre-generation of hybrid redundant paths: Based on the fault propagation chain, primary and backup path pairs are generated to ensure rapid switching in case of failure. S4. Programmable self-healing decision engine: Constructs a dynamic decision model and outputs self-healing instructions that adapt to the scenario. S5. Zero-disruption service migration and closed-loop verification: Based on the self-healing instructions output by S4, the system achieves seamless service switching through a double buffering mechanism and a hierarchical migration strategy, and continuously optimizes the decision model based on the verification results.

2. The intelligent fault switching and self-healing method for power communication networks integrating fiber optic and wireless technologies according to claim 1, characterized in that, The four types of raw signals include optical power signals, radio frequency interference signals, temperature and humidity signals, and vibration signals.

3. The intelligent fault switching and self-healing method for power communication networks integrating fiber optic and wireless technologies according to claim 1, characterized in that, The event-driven edge computing unit FPGA processes four types of raw signals, including: S111. Joint feature extraction: extract time-domain and frequency-domain features for the four types of signals respectively, and then obtain a fused feature vector by feature concatenation; S112. Anomaly Weight Assignment: Assign anomaly weights to each dimension of the joint features. The calculation formula is as follows: Where, x i Let be the measured value of the i-th feature. σ is the mean of this feature under normal operating conditions. i α represents the standard deviation of this feature under normal conditions, used to measure the dispersion of the data; i The importance coefficient for this feature is determined based on the degree of impact of various signals on link health; S113. Spatiotemporal alignment: Timestamp calibration and spatial coordinate matching are performed on features collected at different nodes and at different times to eliminate spatiotemporal deviations.

4. The intelligent fault switching and self-healing method for power communication networks integrating fiber optic and wireless technologies according to claim 1, characterized in that, The link health profile is a 6-dimensional link health vector H, refreshed once per second. H is defined as follows: H=[ω o ,oh r ,oh t ,oh h ,oh v ,t up ]; Where, ω o ω r ω t ω h ω v These are the abnormal weights for optical power, radio frequency interference, temperature, humidity, and vibration, respectively; t up To update the timestamp, accurate to the second, the moment the image vector was generated is recorded.

5. The intelligent fault switching and self-healing method for power communication networks integrating fiber optic and wireless technologies according to claim 1, characterized in that, The steps for constructing a dynamic model based on link health profiles to mine potential fault propagation paths include: S211. Dynamic heterogeneous graph construction: The link health profile H generated in stage S1 is used as the core attribute of the node. Fiber optic links are regarded as "hard connection edges" and wireless links are defined as "soft connection edges" to construct the topology graph. To accurately reflect changes in network state over time, a timestamp is added to each node and edge in the topology graph, resulting in the following dynamic heterogeneous graph: G(t) = (V, E, T); V is the node set, containing all communication nodes and their corresponding link health profiles; E is the edge set, describing the fiber optic or wireless connection relationships between nodes; T is the time set, recording the time information of each node and edge state. S212. Memory decay graph neural network modeling and forward propagation: A graph neural network with memory decay is used to perform rolling forward calculation on the "node-edge-time" triple to simulate the fault propagation trend. In graph neural networks with memory decay, the memory decay mechanism is as follows: a memory factor λ(t) is introduced, and the influence of historical states decays over time, as shown in the formula: β is the attenuation coefficient, t c t represents the current time; p For historical time; Rolling forward computation is performed on the "node-edge-time" triple: the node state is updated through the GNN layer, and the probability matrix P∈R of cascading failures occurring in each node within the future time period Δt is output. N×N In the probability matrix P ij This represents the probability that a failure in node i causes a failure in node j, where N is the total number of nodes; S213. Fault propagation chain screening: Set a probability threshold and retain P. ij Edges with a probability greater than or equal to a set probability threshold are filtered out as low-probability associations; starting from the initial fault node, the propagation path is traced backwards, and chains longer than 5 hops are removed; finally, a set of "dual-domain fault propagation chains" is obtained, which is represented as: C={c1,c2,…c k ,…,c m }; Among them, each chain c k A sequence of nodes is represented as: c k ={v k1 ,v k2 ,…,v kL }(L≤5)。 6. The intelligent fault switching and self-healing method for power communication networks integrating fiber optic and wireless technologies according to claim 1, characterized in that, The steps for generating primary / backup path pairs based on the fault propagation chain include: S311. Redundant path generation: Based on the fault propagation chain in S2, generate K candidate redundant paths in the fiber optic domain and the wireless domain respectively. The fiber domain uses "wavelength-time slot joint coding" to generate K candidate paths, with the coding format being (λ,τ), where λ is the wavelength number and τ is the time slot number. Conflicts are avoided by combining different wavelengths and time slots. The wireless domain uses "frequency hopping-power joint coding" to generate K candidate paths, with the coding format being (f, p), where f is the frequency hopping frequency and p is the transmit power, and anti-interference is achieved through dynamic frequency hopping and power adjustment; S312. Lightweight game theory algorithm for resource bidding: Under the three-dimensional constraints of "latency-bandwidth-energy consumption", a lightweight game theory algorithm is designed to select the primary and backup path pair that satisfies the three-dimensional constraints of "latency-bandwidth-energy consumption" and has the lowest cost from the candidate paths. In the lightweight game theory algorithm, the cost function C is defined as a weighted sum of three-dimensional indicators: C=ω D ·D+ω B ·(B max -B)+ω E ·E; D is the transmission delay of the path; B is the bandwidth currently available for the path; E is the energy consumption of the path when transmitting data; B max The maximum available bandwidth that the network can provide; ω D ω B ω E ω is the weighting coefficient. D +ω B +ω E =1; S313. Switching trigger condition setting: A dual threshold triggering mechanism is adopted. When the health profile score of any node in the main path is less than the health threshold, or the failure probability of the propagation chain in which the main path is located is greater than the probability threshold, the path switching is triggered.

7. The intelligent fault switching and self-healing method for power communication networks integrating fiber optic and wireless technologies according to claim 1, characterized in that, The steps for constructing a dynamic decision-making model and outputting self-healing instructions adapted to the scenario include: S411. Decision tree model construction: Map the primary and backup path pairs generated in S3 to the leaf nodes of the decision tree. Use the link health profile H as the input feature and the path availability as the output label to train the decision tree and achieve a fast mapping from input to path status. S412. Reinforcement Learning Meta-Policy Network Update: A meta-policy network is designed based on reinforcement learning, with the reward function as follows: R′=γ·success-(1-γ)·delay; γ∈[0,1] represents the weight; success is the handover success flag; delay is the handover delay penalty; the network updates the weights online every 5 seconds. S413. Self-healing command output: When the switching conditions set by S3 are met, the engine immediately generates a self-healing command to dynamically optimize the decision-making logic. Self-healing instructions include: Fiber priority, prioritizing the use of backup fiber paths, parameter: λ new τ new and routing table o; where λ new Number the new wavelength, τ new The new time slot number is used, and the routing table o contains the node hopping sequence of the fiber optic path; Wireless priority, prioritizes the use of the wireless backup path, parameter: f new p new and routing table r; where f new For the new frequency hopping frequency, p new For the new transmit power, the routing table r contains the node hop sequence of the wireless path; Hybrid collaboration enables both fiber optic and wireless paths to share the load, with parameters being a combination of the two types of parameters mentioned above.

8. The intelligent fault switching and self-healing method for power communication networks integrating fiber optic and wireless technologies according to claim 1, characterized in that, The steps for achieving seamless service switching through a double buffering mechanism and a tiered migration strategy are as follows: S511. Dual-buffered zero-copy data migration channel is enabled. A dual-buffered channel is established between the source node and the destination node. The primary buffer receives current business data in real time, while the backup buffer preloads the data to be migrated. The "zero-copy" technology is adopted to skip the CPU data copying step through DMA transfer, thereby reducing migration latency. S512. Business-level migration: Data is migrated in a tiered manner according to business priority through a dual-buffered channel. Specific rules are as follows: Priority 1: Includes relay protection signals, security control commands, etc., requiring migration delay ≤10ms; prioritize the use of dual buffer channel bandwidth through preemptive scheduling, allocate ≥60% of the bandwidth, and ensure migration is completed in milliseconds; Priority 2: Includes telemetry data, video surveillance streams, etc., with a migration latency of ≤100ms allowed; after the migration of priority 1 services is completed, 30% of the bandwidth will be used for transmission to achieve second-level migration; Priority 3: Includes file transfer, remote configuration commands, etc., with a migration latency of ≤60s; utilizes remaining bandwidth for asynchronous migration to ensure that high-priority services are not affected.

9. The intelligent fault switching and self-healing method for power communication networks integrating fiber optic and wireless technologies according to claim 1, characterized in that, The method for continuously optimizing the decision model based on verification results: After the migration is complete, data integrity is verified by reverse sequence code injection, and decision parameters are optimized based on the results: Reverse sequence code injection, the destination node generates random sequence codes: S = [s1, s2, ... s i …,s n (n = 1024, each s) i The received sequence (a 32-bit integer) is sent to the source node via the reverse channel (new path). The source node compares the received sequence with its local backup sequence and calculates the matching rate. If M=1, the verification is considered successful; if M<1, the verification is considered unsuccessful. Result processing: If M < 1, the rollback mechanism is immediately triggered to switch the service back to the original path and raise the health threshold and probability threshold in S3 to reduce the probability of false switching. If M=1, the current link health profile H, fault chain, and migration parameters are written into the experience base for use by the reinforcement learning meta-policy network of S4 during updates, thus realizing the self-evolution of decision logic.

Images