Online Monitoring Method for Faults in Multi-Core Optical Cables Based on Optical Fiber Sensing Technology

Abstract

The present invention relates to the technical field of communication fault monitoring, specifically an online monitoring method for multi-core optical cable faults based on optical fiber sensing technology. First, a primary network topology diagram is established according to the direct communication relationship of core nodes, and key service optical cables are selected according to the key service types of the primary network topology diagram to establish a secondary network topology diagram; by clustering the historical performance data of the secondary links in the secondary network topology diagram, the performance status of the secondary links is judged in real time; the backbone index is calculated according to bandwidth and throughput and the backbone optical cable is determined; the ROTDR device is used to monitor the faults of the backbone optical cable, and super-resolution reconstruction and signal analysis are completed through the SR-LSTM model to locate the fault position and identify the fault type; combining the performance status, fault position and type, a state space and an action space are constructed through the Actor-Critic reinforcement learning framework, and the intelligent agent learns the optimal operation and maintenance strategy to realize the online monitoring of the faults of the backbone optical cable.

Description

Technical Field

[0001] The present invention relates to the technical field of communication fault monitoring, and in particular to a multi-core optical cable fault online monitoring method based on optical fiber sensing technology. Background Art

[0002] Optical cables, as essential infrastructure for modern information transmission and industrial control, play a vital role in power systems. Optical cable networks in power systems typically consist of multi-core cables, each transmitting different service data through a single fiber core. This structure significantly improves communication bandwidth and network reliability. Among these cables, trunk cables connect power dispatch centers, substations, distribution networks, and control centers, carrying critical services such as dispatch control, protection communications, and fault monitoring. A failure in a trunk cable can directly impact the stable operation of the power grid and potentially even cause a serious power outage. Therefore, performance monitoring and fault detection of multi-core cables, especially trunk cables, are crucial for the efficient operation of power systems and rapid fault response.

[0003] Raman optical time-domain reflectometry (ROTDR), a distributed fiber-optic sensing technology, has been widely used in multi-core optical cable fault monitoring. Based on the Raman scattering effect, ROTDR accurately measures the temperature distribution along the optical cable by transmitting optical pulses and measuring the intensity ratio of the Stokes and anti-Stokes light in the echo. This technology is not only suitable for detecting temperature-related faults (such as overheating or abnormal ambient temperature), but also leverages the structural characteristics of multi-core optical cables to monitor the independent performance of each optical fiber core. However, ROTDR has limitations in directly detecting non-temperature-related faults such as mechanical damage and fiber breaks. Furthermore, due to the high signal complexity in multi-core optical cables, ROTDR's accuracy in distinguishing between single-core fiber faults and overall cable performance anomalies within a multi-core cable needs to be further improved. Therefore, optimizing ROTDR's signal analysis capabilities in multi-core optical cable environments and combining it with intelligent analysis methods to improve the accuracy of fault detection and operational efficiency of trunk optical cables remain key challenges.

[0004] Therefore, an online fault monitoring method for multi-core optical cables based on optical fiber sensing technology is proposed. Summary of the Invention

[0005] The object of the present invention is to provide an online monitoring method for multi-core optical cable faults based on optical fiber sensing technology. The present invention performs clustering analysis on the historical performance data of critical service optical cables, judges the performance state of the optical cables, calculates the backbone index based on the bandwidth and throughput of the optical cables, and thus determines the backbone optical cables; the present invention performs signal super-resolution reconstruction and analysis on the time-domain reflection signal through an SR-LSTM model, and accurately obtains the fault location and fault type; the present invention establishes a state space based on the performance state, fault location and type, constructs an action space in combination with expert experience and historical operation and maintenance records, and optimizes the operation and maintenance strategy through an Actor-Critic reinforcement learning framework to achieve efficient performance maintenance and intelligent fault repair of the backbone optical cables.

[0006] To achieve the above object, the present invention provides the following technical solutions:

[0007] An online monitoring method for multi-core optical cable faults based on optical fiber sensing technology, comprising:

[0008] Establish a first-level network topology diagram according to the direct communication relationship of the core nodes; the core nodes include a power dispatching center, a substation, a distribution network and a control center;

[0009] Select critical service optical cables according to the service types between the first-level nodes of the first-level network topology diagram, and establish a second-level network topology diagram according to the critical service optical cables; the critical service optical cables are optical cables for processing critical services, and the critical services include dispatching control, protection communication and fault monitoring;

[0010] Collect and store the historical performance data between the second-level links of the second-level network topology diagram, establish a performance database, perform clustering on the historical performance data through the K-Means algorithm, and judge the performance state of the current second-level link according to the clustering result;

[0011] Calculate the backbone index according to the bandwidth and throughput of the second-level link, and establish the backbone optical cables of the second-level network topology diagram according to the backbone index;

[0012] Use a ROTDR device to monitor faults in the backbone optical cables, obtain time-domain reflection signals, and use an SR-LSTM model to perform super-resolution reconstruction and signal analysis on the time-domain reflection signals to obtain the fault location and fault type;

[0013] Establish a state space according to the performance state, the fault location and the fault type, and establish an action space according to expert experience and historical operation and maintenance records; through an Actor-Critic reinforcement learning framework, combine the state space and the action space, and use an agent to perform reinforcement learning to obtain an optimal operation and maintenance strategy.

[0014] Further, establishing the first-level network topology diagram includes:

[0015] Determine all the core nodes within the region based on the existing power network database; analyze the communication traffic between each of the core nodes through a network detection tool to determine the communication optical cables directly connected between the core nodes; use a topology graph tool to model the core nodes and the communication optical cables as a first-level network topology graph, where the core nodes are the first-level nodes of the first-level network topology graph, and the communication optical cables are the first-level links of the first-level network topology graph.

[0016] Furthermore, establishing the second-level network topology graph includes:

[0017] Conduct traffic analysis on each communication optical cable in the first-level network topology graph to obtain the service type of each communication optical cable; select key service optical cables according to the service type; use a topology graph tool to model the core nodes and the key service optical cables as a second-level network topology graph, where the core nodes are the second-level nodes of the second-level network topology graph, and the key service optical cables are the second-level links of the second-level network topology graph.

[0018] Furthermore, the historical performance data includes bandwidth, latency, throughput, packet loss rate, bit error rate, and optical power, and the performance status includes normal status, fiber aging, temperature drift, bandwidth overload, signal delay increase, and abnormal fiber dispersion.

[0019] Furthermore, determining the performance status of the current second-level link includes:

[0020] Collect the historical performance data of the second-level link, and perform data cleaning and data normalization on the historical performance data to obtain preprocessed data; use the elbow method to determine the optimal K value, cluster the preprocessed data to obtain performance status categories; cluster the performance data of the current second-level link, and output the performance status according to the clustering result. [[ID=I9]]

[0021] Furthermore, the calculation formula for the backbone index is:

[0022]

[0023] where M represents the backbone index, B represents the bandwidth, T represents the throughput, α1 represents the bandwidth weight, α2 represents the throughput weight, and α3 represents the utilization rate weight.

[0024] Furthermore, the SR-LSTM model includes:

[0025] An input layer for receiving the time-domain reflection signal;

[0026] Signal super-resolution layer, which is used to perform preliminary super-resolution processing on the time-domain reflection signal using the Fourier interpolation method to generate a low-order super-resolution signal; the preliminary super-resolution processing includes: performing a fast Fourier transform on the time-domain reflection signal to obtain a frequency-domain signal; filling zero values at the centrally symmetric positions of the frequency-domain signal to obtain an extended spectrum; performing an inverse fast Fourier transform on the extended spectrum to generate a low-order super-resolution signal;

[0027] LSTM layer, which is used to capture the temporal dependence relationship of the preliminary super-resolution signal to generate a high-order super-resolution signal;

[0028] Feature decoding layer, which is used to extract features related to the fault location and fault type according to the high-order super-resolution signal;

[0029] Output layer, which is used to output the prediction results of the fault location and fault type.

[0030] Further, obtaining the fault location and the fault type includes:

[0031] Using the ROTDR device to send an optical pulse to the backbone optical cable; recording the time delay and intensity information of the reflection signal to obtain the time-domain reflection signal; inputting the time-domain reflection signal into the SR-LSTM model to obtain the fault location and the fault type.

[0032] Further, obtaining the optimal operation and maintenance strategy:

[0033] S601: Establish the state space according to the performance state, the fault location and the fault type, and establish the action space according to the expert experience and the historical operation and maintenance records;

[0034] S602: Initialize the Actor network and the Critic network, and set the learning rate, discount factor, exploration strategy and reward function;

[0035] S603: The agent generates the current action through the Actor network according to the current state, and after interacting with the system environment, obtains an immediate reward and returns the next state;

[0036] S604: Calculate the updated value of the Critic network using the TD error, and update the Actor network according to the TD error;

[0037] S605: Determine whether the iteration meets the convergence condition. If the convergence condition is not met, then re-execute S603; the convergence condition is that the TD error is less than the error threshold;

[0038] S606: Obtain the optimal operation and maintenance strategy.

[0039] Furthermore, the calculation formula of the reward function is as follows:

[0040]

[0041] Among them, R represents the reward function, P exp represents the expected performance, P act represents the actual performance after operation and maintenance, T rep represents the operation and maintenance time, C res represents the operation and maintenance cost, ω1 represents the performance weight, ω2 represents the time weight, and ω3 represents the cost weight.

[0042] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0043] 1. By calculating the backbone index based on the bandwidth and throughput of the critical service optical cable and establishing the backbone optical cable, the present invention can accurately evaluate the importance and load level of the optical cable, and preferentially identify and construct high-priority communication paths carrying critical services. This method improves the scientificity and accuracy of network topology optimization by quantifying the distribution of network resources, ensures the efficient allocation of resources and the priority solution of key problems, and indirectly improves the efficiency of online fault monitoring.

[0044] 2. The present invention proposes and uses the SR-LSTM model to perform super-resolution reconstruction and signal analysis on the time-domain reflection signal, significantly improving the accuracy and efficiency of fault detection. By initially performing signal super-resolution processing to improve signal resolution and combining the LSTM layer to capture the temporal dependence relationship of the signal, the SR-LSTM model accurately extracts fault features while reconstructing high-resolution signals, achieving higher spatial resolution, more accurate fault location, and realizing the real-time and effectiveness of online multi-core optical cable fault monitoring.

[0045] 3. Through the Actor-Critic reinforcement learning framework, the present invention intelligently optimizes the online multi-core optical cable fault monitoring and maintenance process. This framework combines the state space and the action space, and through continuous interaction between the intelligent agent and the environment and iterative optimization of the optimal operation and maintenance strategy, it realizes the efficient maintenance, accurate fault location and repair of the performance state of the backbone optical cable. Through the precise reinforcement learning mechanism, this framework can dynamically learn the optimal decision-making path, significantly improving the real-time and accuracy of fault monitoring and effectively ensuring the stable operation of the backbone optical cable. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 is a flowchart of the online multi-core optical cable fault monitoring method based on fiber optic sensing technology provided by the present invention;

[0047] Figure 2 is a schematic structural diagram of the SR-LSTM model provided by the present invention;

[0048] Figure 3 This is the flowchart for learning the optimal operation and maintenance strategy provided by the present invention. Specific embodiments

[0049] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0050] The first embodiment provided by the present invention is as follows:

[0051] A multi-core optical cable fault online monitoring method based on optical fiber sensing technology, as Figure 1 shown, includes:

[0052] S100: Establish a first-level network topology diagram according to the direct communication relationship of core nodes; the core nodes include a power dispatching center, a substation, a distribution network, and a control center;

[0053] Further, establishing the first-level network topology diagram includes:

[0054] Determine all the core nodes in the region according to the existing power network database; analyze the communication traffic between each core node through a network detection tool to determine the communication optical cables directly connected between the core nodes; use a topology diagram tool to model the core nodes and the communication optical cables as a first-level network topology diagram, where the core nodes are the first-level nodes of the first-level network topology diagram, and the communication optical cables are the first-level links of the first-level network topology diagram.

[0055] Specifically, some data of the power network database of a certain city is shown in Table 1. Select the core nodes of this city according to the node name and node location; detect the communication IP addresses between nodes through the NetFlow network detection tool to determine the communication optical cables directly connected between the core nodes; model the core nodes and the communication optical cables as a first-level network topology diagram through the topology diagram tool Gephi.

[0056] Table 1 Some data of the power network database of a certain city

[0057]

[0058] Establishing a first-level network topology diagram can intuitively display the connection relationship between core nodes and communication optical cables, providing a clear visualization basis for the online monitoring of multi-core optical cables in the region.

[0059] S200: Select critical service optical cables based on the service types between the first-level nodes of the first-level network topology diagram, and establish a second-level network topology diagram based on the critical service optical cables; the critical service optical cables are optical cables for handling critical services, and the critical services include dispatching control, protection communication, and fault monitoring;

[0060] Further, establishing the second-level network topology diagram includes:

[0061] Conduct traffic analysis on each communication optical cable in the first-level network topology diagram to obtain the service type of each communication optical cable; select critical service optical cables according to the service type; use topology diagram tools to model the core nodes and the critical service optical cables as a second-level network topology diagram, where the core nodes are the second-level nodes of the second-level network topology diagram, and the critical service optical cables are the second-level links of the second-level network topology diagram.

[0062] Specifically, use the protocol analysis tool Wireshark to extract service type identifiers from the traffic data of the communication optical cables in the first-level network topology diagram, set the communication optical cables with transmission service types of dispatching control, protection communication, and fault monitoring as critical service optical cables, and use the topology diagram tool Gephi to model the critical service optical cables as a second-level network topology diagram.

[0063] Establishing a second-level network topology diagram can focus on critical service optical cables and their related core nodes, accurately depict the communication paths and dependency relationships of critical services, and provide a reliable data basis for solving the real-time online monitoring of the performance indicators of key optical cable lines.

[0064] S300: Collect and store the historical performance data between the second-level links of the second-level network topology diagram, establish a performance database, perform clustering on the historical performance data through the K-Means algorithm, and judge the performance status of the current second-level link according to the clustering results;

[0065] Further, the historical performance data includes bandwidth, latency, throughput, packet loss rate, bit error rate, and optical power, and the performance status includes fiber aging, temperature drift, bandwidth overload, signal delay increase, and abnormal fiber dispersion.

[0066] By collecting and analyzing historical performance data including bandwidth, latency, throughput, packet loss rate, bit error rate, and optical power, the operating conditions of optical cables can be comprehensively monitored, and combined with various classifications of performance statuses, it helps to accurately identify the performance status of critical service optical cables.

[0067] Further, judging the performance status of the current second-level link includes:

[0068] Collect the historical performance data of the secondary link, perform data cleaning and data normalization on the historical performance data to obtain preprocessed data; use the elbow method to determine the optimal K value, cluster the preprocessed data to obtain performance status categories; cluster the performance data of the current secondary link and output the performance status according to the clustering result.

[0069] Specifically, collect the historical performance data of the secondary link, including bandwidth, latency, throughput, packet loss rate, bit error rate, and optical power, and integrate these metrics into a vector. The representation form of the vector is as follows:

[0070] D = {x1, x2, x3,..., x n};

[0071] x i = {bandwidth, latency, throughput, packet loss rate, bit error rate, optical power};

[0072] Among them, D represents all historical performance data, n represents the data volume of historical performance data, and x i represents the metrics included in the i-th historical performance data. Subsequently, remove outliers and noise, fill in missing data, and normalize the data to between [0, 1]; find the optimal K value by comparing the sum of mean squared errors under different K values. The calculation formula for the sum of mean squared errors is:

[0073]

[0074] Among them, SSE represents the sum of mean squared errors, K represents the number of clusters, j represents the cluster index, and μ j represents the j-th cluster center. In a feasible implementation, the optimal K value is 5, and the performance status categories are fiber aging, temperature drift, bandwidth overload, signal delay increase, and fiber dispersion anomaly; cluster the performance data of the current secondary link and output the performance status according to the clustering result.

[0075] Through the real-time clustering analysis and performance status output of the current optical cable performance data, accurate monitoring and anomaly monitoring of the optical cable network can be achieved, providing efficient support for maintenance decision-making, performance optimization, and fault warning, thereby greatly improving the stability and operation efficiency of the optical cable network.

[0076] S400: Calculate the backbone index according to the bandwidth and throughput of the secondary link, and establish the backbone optical cable of the secondary network topology diagram according to the backbone index;

[0077] Furthermore, the calculation formula for the backbone index is:

[0078]

[0079] Wherein, M represents the backbone index, B represents the bandwidth, T represents the throughput, α1 represents the bandwidth weight, α2 represents the throughput weight, and α3 represents the utilization weight.

[0080] Specifically, the backbone index is calculated based on the bandwidth and throughput of the critical service optical cables, and the N critical service optical cables with the highest backbone index are selected to establish the backbone optical cables.

[0081] By calculating the backbone index based on the bandwidth and throughput of the critical service optical cables, the optical cables with relatively high load criticality in the network can be accurately identified, and the backbone optical cables in the secondary network topology diagram can be effectively refined. The establishment of the backbone optical cables helps to optimize the network resource allocation and provides technical support for the priority guarantee of critical services.

[0082] S500: Use the ROTDR device to measure the backbone optical cable to obtain the time-domain reflection signal, and use the SR-LSTM model to perform super-resolution reconstruction and signal analysis on the time-domain reflection signal to obtain the fault location and fault type;

[0083] Furthermore, the SR-LSTM model is as Figure 2 shown, including:

[0084] An input layer for receiving the time-domain reflection signal;

[0085] A signal super-resolution layer for performing preliminary super-resolution processing on the time-domain reflection signal using the Fourier interpolation method to generate a low-order super-resolution signal; the preliminary super-resolution processing includes: performing a fast Fourier transform on the time-domain reflection signal to obtain a frequency-domain signal; filling zero values at the centrosymmetric position of the frequency-domain signal to obtain an extended spectrum; performing an inverse fast Fourier transform on the extended spectrum to generate a low-order super-resolution signal;

[0086] An LSTM layer for capturing the temporal dependence relationship of the preliminary super-resolution signal to generate a high-order super-resolution signal;

[0087] A feature decoding layer for extracting features related to the fault location and fault type according to the high-order super-resolution signal;

[0088] An output layer for outputting the prediction results of the fault location and fault type.

[0089] Specifically, the input data of the input layer is the time-domain reflection signal, expressed as:

[0090] x = {x1, x2,..., x N};

[0091] where N represents the sampling time points; the signal super-resolution layer performs a fast Fourier transform on the time-domain reflection signal to obtain a frequency-domain signal, and the formula for obtaining the frequency-domain signal is:

[0092]

[0093] Among them, F[k] represents the frequency component of the time-domain reflection signal in the frequency domain, k represents the index of the frequency domain, x n represents the discrete points of the time-domain reflection signal, n represents the index of the sampling points, e -j2πkn / N represents the basis function of the Fourier transform; zero values are filled at the centrosymmetric position of the frequency-domain signal to obtain the extended spectrum, and the calculation formula of the extended spectrum is:

[0094]

[0095] Among them, F padded [k] represents the extended spectrum, M represents the length of the extended spectrum; the inverse fast Fourier transform is performed on the extended spectrum to generate the low-order super-resolution signal, and the calculation formula of the low-order super-resolution signal is:

[0096]

[0097] Among them, f super [n] represents the low-order super-resolution signal; the LSTM layer uses LSTM units to capture the temporal dependence of the low-order super-resolution signal, and this process can be expressed as:

[0098] h t ,c t = LSTM(f super [n] t ,h t-1 ,c t-1 );

[0099] Among them, t represents the time variable, h t represents the hidden state at time t, c t represents the cell state at time t, LSTM() represents the LSTM module, h t-1 represents the hidden state at time t-1, c t-1 represents the cell state at time t-1, the LSTM layer stacks multiple LSTM modules to increase the expressive power of the model and obtains the high-order super-resolution signal, and its expression form is:

[0100] H = {h1, h2, h3,..., h T};

[0101] Among them, H represents the high-order super-resolution signal; the feature decoding layer extracts the features related to the fault location and fault type according to the high-order super-resolution signal, and the feature decoding layer is implemented by a fully connected or convolutional neural network. In this embodiment, a fully connected layer is selected to implement feature decoding, and this process can be expressed as:

[0102] y location= Decoder1(H);

[0103] y type = Decoder2(H);

[0104] where y location represents the prediction of the fault location, and y type represents the prediction of the fault type. Decoder() represents the feature decoder; the output layer analyzes the output result of the feature decoding layer through the Softmax function to obtain the prediction result, and this process can be expressed as:

[0105]

[0106] where represents the prediction result of the fault location, represents the prediction result of the fault type, and Softmax() represents the Softmax function.

[0107] The SR-LSTM model provides an efficient and accurate time-domain reflectometry signal analysis method by combining signal processing and deep learning. This model integrates the advantages of traditional signal processing and deep learning, realizes efficient end-to-end analysis from the original signal to the prediction result, significantly improves the performance and reliability of optical cable fault monitoring, and provides technical guarantee for the monitoring and maintenance of backbone optical cables.

[0108] Furthermore, obtaining the fault location and the fault type includes:

[0109] Using the ROTDR device to send optical pulses to the backbone optical cable; recording the time delay and intensity information of the reflected signal to obtain the time-domain reflectometry signal; inputting the time-domain reflectometry signal into the SR-LSTM model to obtain the fault location and the fault type.

[0110] By using the ROTDR device to send optical pulses and recording the time delay and intensity information of the time-domain reflectometry signal, the performance data of the backbone optical cable can be efficiently collected. Inputting the collected signal into the SR-LSTM model, combining signal super-resolution processing and time-series analysis, the fault location and the fault type can be accurately extracted. This process realizes the online analysis from signal acquisition to fault diagnosis, improves the accuracy and speed of fault monitoring, and provides reliable technical support for optical cable maintenance and fault repair.

[0111] S600: Establish a state space according to the performance state, the fault location and the fault type, and establish an action space according to expert experience and historical operation and maintenance records; combine the state space and the action space through the Actor-Critic reinforcement learning framework, and use an agent to perform reinforcement learning to obtain the optimal operation and maintenance strategy.

[0112] Furthermore, the obtained optimal operation and maintenance strategy is as follows Figure 3 shown, including:

[0113] S601: Establish the state space according to the performance status, the fault location, and the fault type, and establish the action space according to the expert experience and the historical operation and maintenance records;

[0114] S602: Initialize the Actor network and the Critic network, and set the learning rate, the discount factor, the exploration strategy, and the reward function;

[0115] S603: The agent generates the current action through the Actor network according to the current state. After the current action interacts with the system environment, it obtains the immediate reward and returns the next state;

[0116] S604: Calculate the update value of the Critic network using the TD error, and update the Actor network according to the TD error;

[0117] S605: Determine whether the iteration meets the convergence condition. If the convergence condition is not met, re-execute S603; the convergence condition is that the TD error is less than the error threshold;

[0118] S606: Obtain the optimal operation and maintenance strategy.

[0119] Table 2 Settings of the state space and the action space of the Actor-Critic reinforcement learning framework

[0120]

[0121] Specifically, the settings of the state space and the action space are shown in Table 2, where the state space includes the performance status, the fault location, and the fault type, and the action space is the measures taken for different performance statuses and fault types. These measures are based on the historical operation and maintenance records and expert experience, where the expert experience is the solution proposed by technicians in this industry for the performance status and the fault type to truly solve the above problems; initialize the Actor network and the Critic network, and set the learning rate, the discount factor, the exploration strategy, and the reward function, where both the Actor network and the Critic network are common multi-layer perceptrons, and the exploration strategy is the ε-greedy strategy; the Actor network takes the current state s t as the input, calculates the probability of generating each action for the current state through the Softmax function, and then selects the current action a according to the probability t , and the agent sends the current action a t to the system environment and executes the current action a t , returns the next state s t+1 and the immediate reward Rt ; Calculate the update value of the Critic network using the TD error, and update the Actor network according to the TD error, where the calculation formula of the TD error is:

[0122] δ t =R t +γ×Q(s t+1 ,a t+1 ) - Q(s t ,a t );

[0123] Among them, δ t represents the TD error, R t represents the immediate reward, γ represents the discount factor, Q() represents the value expectation function of the state-action pair, Q(s t+1 ,a t+1 ) represents the value expectation of the next state-action pair, Q(s t ,a t ) represents the value expectation of the current state-action pair. Update the parameters in the Critic network through the gradient descent method to make it more accurately estimate the value of the state-action pair. The update formula is:

[0124]

[0125] Among them, θ c represents the parameters of the Critic network, α c represents the learning rate of the Critic network, represents the gradient of the Critic network. The Actor network updates the parameters by maximizing the expected return of the action. The gradient update formula is:

[0126]

[0127] Among them, θ a represents the parameters of the Actor network, α c represents the learning rate of the Actor network, logπ(a t |s t ) represents the action selection probability distribution generated by the Actor network; Judge whether the TD error is less than the error threshold. If it is less than the threshold, stop the update of the Actor network and the Critic network, and give the optimal operation and maintenance strategy. Otherwise, re-execute step S603. In a feasible implementation, the error threshold is 0.5; When the iteration ends, the optimal operation and maintenance strategy is finally obtained.

[0128] By establishing a state space by combining performance status, fault location, and fault type, and constructing an action space using expert experience and historical operation and maintenance records, the Actor-Critic reinforcement learning framework can dynamically adapt to the complex backbone optical cable maintenance requirements. Through continuous learning and optimization of the operation and maintenance strategy by the intelligent agent, accurate decision-making for the performance maintenance and fault repair of the backbone optical cable is achieved.

[0129] Furthermore, the calculation formula of the reward function is as follows:

[0130]

[0131] Among them, P exp represents the expected performance, P act represents the actual performance after operation and maintenance, T rep represents the operation and maintenance time, C res represents the operation and maintenance cost, ω1 represents the performance weight, ω2 represents the time weight, and ω3 represents the cost weight.

[0132] Specifically, the expected performance is the theoretical performance when the optical cable state is intact, including theoretical bandwidth, theoretical delay, theoretical throughput, theoretical packet loss rate, theoretical bit error rate, and theoretical optical power. The actual performance after operation and maintenance includes actual bandwidth, actual delay, actual throughput, actual packet loss rate, actual bit error rate, and actual optical power. The physical meaning represented by the division operation of the expected performance and the actual performance after operation and maintenance can be expressed by the following formula:

[0133]

[0134] The operation and maintenance cost is the operation and maintenance expense generated during operation and maintenance.

[0135] By comprehensively considering performance recovery, operation and maintenance time, and operation and maintenance cost, the reward function flexibly adjusts the influence weights of various factors in a weighted manner to achieve multi-objective optimization. By quantifying the difference between the expected performance and the actual performance, the performance improvement is explicitly incorporated into the reward calculation. At the same time, by combining the operation and maintenance time and cost, it guides the intelligent agent to give an efficient and economical way to help technicians complete fault repair and performance maintenance.

[0136] The second embodiment provided by the present invention is as follows:

[0137] In response to the call for the digital and intelligent development of electric power, a power company upgrades the technology of the fault monitoring system A for the backbone optical cable of the power communication network to achieve online monitoring of multi-core optical cable faults in the backbone optical cable of the power communication network. For this purpose, a new fault monitoring system B is adopted, which is equipped with the multi-core optical cable fault online monitoring method based on optical fiber sensing technology proposed by the present invention, including:

[0138] The multi-core optical cable fault online monitoring method based on optical fiber sensing technology, including:

[0139] Build a first-level network topology map according to the direct communication relationships of the core nodes, where the core nodes include a power dispatching center, a substation, a distribution network, and a control center;

[0140] Select key service optical cables according to the service types between the first-level nodes of the first-level network topology map, and build a second-level network topology map according to the key service optical cables. The key service optical cables are optical cables for processing key services, and the key services include dispatching control, protection communication, and fault monitoring;

[0141] Collect and store the historical performance data between the second-level links of the second-level network topology map, establish a performance database, cluster the historical performance data through the K-Means algorithm, and judge the performance status of the current second-level link according to the clustering results;

[0142] Calculate the backbone index according to the bandwidth and throughput of the second-level link, and establish the backbone optical cable of the second-level network topology map according to the backbone index;

[0143] Use the ROTDR device to monitor the faults of the backbone optical cable, obtain the time-domain reflection signal, and use the SR-LSTM model to perform super-resolution reconstruction and signal analysis on the time-domain reflection signal to obtain the fault location and fault type;

[0144] Establish a state space according to the performance status, the fault location, and the fault type, and establish an action space according to expert experience and historical operation and maintenance records; Combine the state space and the action space through the Actor-Critic reinforcement learning framework, and use an agent to perform reinforcement learning to obtain the optimal operation and maintenance strategy.

[0145] After implementing the fault monitoring system B based on the present invention, the performance comparison with the original fault monitoring system A is shown in Table 3. It can be seen from the comparison in the table that the fault monitoring system B based on the present invention is significantly superior to the original fault monitoring system A in multiple key performance indicators. First of all, in terms of the fault recognition rate, system B reaches 93%, which is 8 percentage points higher than 85% of system A, showing higher accuracy. Secondly, the fault space resolution of system B reaches 0.4 meters, which is much better than 1.5 meters of system A, indicating that system B can locate the fault point more accurately and improve the refinement degree of fault detection. In addition, the fault response time of system B only needs 1 minute, which is 2 minutes shorter than that of system A, significantly improving the fault response efficiency. In addition, system B also has functions that system A does not have, including providing line performance detection and online operation and maintenance solutions. These extended functions not only improve the comprehensive performance of the system, but also provide a scientific basis for subsequent operation and maintenance, making the operation and maintenance more efficient and intelligent. Generally speaking, the fault monitoring system B has been comprehensively upgraded in terms of accuracy, precision, efficiency and functional comprehensiveness, and has important value for the monitoring and operation and maintenance of the power communication network.

[0146] Table 3 Performance Comparison of Different Fault Monitoring Systems

[0147] Performance Fault Monitoring System A Fault Monitoring System B Fault Identification Rate 85％ 93％ Fault Spatial Resolution (m) 1.5 0.4 Fault Response Time (min) 3 1 Whether to Provide Line Performance Detection No Yes Whether to Provide Online Operation and Maintenance Solutions No Yes

[0148] The present invention establishes a first-level network topology diagram according to the direct communication relationship between the core nodes of the power network, and selects key service optical cables according to the service type to construct a second-level network topology diagram; by collecting the historical performance data of the second-level links, a performance database is established, and the K-Means algorithm is used for clustering analysis to judge the performance status of the second-level links; at the same time, the backbone index is calculated according to the bandwidth and throughput of the second-level links to determine the backbone optical cable; the ROTDR device is used to monitor the faults of the backbone optical cable, and the signal super-resolution reconstruction and analysis are carried out through the SR-LSTM model to accurately obtain the fault location and fault type; a state space is established based on the performance status, fault location and type, and an action space is constructed by combining expert experience and historical operation and maintenance records. The operation and maintenance strategy is optimized through the Actor-Critic reinforcement learning framework to realize the efficient performance maintenance and intelligent fault repair of the backbone optical cable. This method significantly improves the optical cable operation and maintenance efficiency and network reliability, and provides comprehensive monitoring and guarantee for key service optical cables.

[0149] Although the embodiments of the present invention have been shown and described, it will be understood by those of ordinary skill in the art that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the present invention, and the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. An online monitoring method for multi-core optical cable faults based on optical fiber sensing technology, characterized in that Including: Establish a first-level network topology diagram according to the direct communication relationship of the core nodes; The core nodes include a power dispatching center, a substation, a distribution network, and a control center; Select key service optical cables according to the service types between the first-level nodes of the first-level network topology diagram, and establish a second-level network topology diagram according to the key service optical cables; The key service optical cable is an optical cable for processing key services, and the key services include dispatching control, protection communication, and fault monitoring; Collect and store the historical performance data between the second-level links of the second-level network topology diagram, establish a performance database, cluster the historical performance data through the K-Means algorithm, and judge the performance status of the current second-level link according to the clustering results; Calculate the backbone index according to the bandwidth and throughput of the second-level link, and establish the backbone optical cable of the second-level network topology diagram according to the backbone index; Use the ROTDR device to monitor the faults of the backbone optical cable, obtain the time-domain reflection signal, and use the SR-LSTM model to perform super-resolution reconstruction and signal analysis on the time-domain reflection signal to obtain the fault location and fault type; Establish a state space according to the performance status, the fault location, and the fault type, and establish an action space according to expert experience and historical operation and maintenance records; combine the state space and the action space through the Actor-Critic reinforcement learning framework, and use an agent to perform reinforcement learning to obtain the optimal operation and maintenance strategy.

2. The online monitoring method for multi-core optical cable faults based on optical fiber sensing technology according to claim 1, wherein, Establishing the first-level network topology diagram includes: Determine all the core nodes in the area according to the existing power network database; analyze the communication traffic between each core node through a network detection tool to determine the directly connected communication optical cables between the core nodes; use a topology diagram tool to model the core nodes and the communication optical cables as a first-level network topology diagram, where the core nodes are the first-level nodes of the first-level network topology diagram, and the communication optical cables are the first-level links of the first-level network topology diagram.

3. The online monitoring method for multi-core optical cable faults based on optical fiber sensing technology according to claim 1, characterized in that, Establishing the second-level network topology diagram includes: Perform traffic analysis on each communication optical cable in the first-level network topology diagram to obtain the service type of each communication optical cable; select key service optical cables according to the service type; use a topology diagram tool to model the core nodes and the key service optical cables as a second-level network topology diagram, where the core nodes are the second-level nodes of the second-level network topology diagram, and the key service optical cables are the second-level links of the second-level network topology diagram.

4. The online monitoring method for multi-core optical cable faults based on optical fiber sensing technology according to claim 1, characterized in that, The historical performance data includes bandwidth, delay, throughput, packet loss rate, bit error rate, and optical power, and the performance status includes normal status, fiber aging, temperature drift, bandwidth overload, signal delay increase, and abnormal fiber dispersion.

5. The on-line monitoring method for multi-core optical cable faults based on optical fiber sensing technology according to claim 1, characterized in that, Judging the performance status of the current second-level link includes: Collect the historical performance data of the second-level link, perform data cleaning and data normalization on the historical performance data to obtain preprocessed data; use the elbow method to determine the optimal K value, cluster the preprocessed data to obtain the performance status category; cluster the performance data of the current second-level link, and output the performance status according to the clustering results.

6. The online monitoring method for multi-core optical cable faults based on optical fiber sensing technology according to claim 1, characterized in that The calculation formula of the backbone index is as follows: Where, M represents the backbone index, B represents the bandwidth, T represents the throughput, α1 represents the bandwidth weight, α2 represents the throughput weight, and α3 represents the utilization rate weight.

7. The on-line monitoring method for multi-core optical cable faults based on optical fiber sensing technology according to claim 1, characterized in that The SR-LSTM model includes: An input layer for receiving the time-domain reflection signal; A signal super-resolution layer for preliminarily super-resolving the time-domain reflection signal using the Fourier interpolation method to generate a low-order super-resolved signal; the preliminary super-resolution processing includes: performing a fast Fourier transform on the time-domain reflection signal to obtain a frequency-domain signal; filling zero values at the centrally symmetric position of the frequency-domain signal to obtain an extended spectrum; performing an inverse fast Fourier transform on the extended spectrum to generate a low-order super-resolved signal; An LSTM layer for capturing the temporal dependence relationship of the low-order super-resolved signal to generate a high-order super-resolved signal; A feature decoding layer for extracting features related to the fault location and fault type according to the high-order super-resolved signal; An output layer for outputting the prediction results of the fault location and fault type.

8. The on-line monitoring method for multi-core optical cable faults based on optical fiber sensing technology according to claim 1, characterized in that, Obtaining the fault location and the fault type includes: Sending an optical pulse to the backbone optical cable using the ROTDR device; recording the time delay and intensity information of the reflection signal to obtain the time-domain reflection signal; inputting the time-domain reflection signal into the SR-LSTM model to obtain the fault location and the fault type.

9. The online monitoring method for multi-core optical cable faults based on optical fiber sensing technology according to claim 1, characterized in that, Obtaining the optimal operation and maintenance strategy: S601: Establish the state space according to the performance state, the fault location, and the fault type, and establish the action space according to the expert experience and the historical operation and maintenance records; S602: Initialize the Actor network and the Critic network, and set the learning rate, discount factor, exploration strategy, and reward function; S603: The agent generates the current action through the Actor network according to the current state, and after interacting with the system environment, obtains an immediate reward and returns the next state; S604: Calculate the update value of the Critic network using the TD error, and update the Actor network according to the TD error; S605: Determine whether the iteration meets the convergence condition. If the convergence condition is not met, then re-execute S603; the convergence condition is that the TD error is less than the error threshold; S606: Obtain the optimal operation and maintenance strategy.

10. The on-line monitoring method for multi-core optical cable faults based on optical fiber sensing technology according to claim 9, characterized in that, The calculation formula of the reward function is as follows: where R represents the reward function, P exp represents the expected performance, P act represents the actual performance after operation and maintenance, T rep represents the operation and maintenance time, C res represents the operation and maintenance cost, ω1 represents the performance weight, ω2 represents the time weight, and ω3 represents the cost weight.

Images