Optical fiber communication line fault automatic diagnosis method and system based on neural network

Abstract

The invention discloses an optical fiber fault diagnosis method and device based on a convolutional neural network (CNN), and is particularly applied to fault detection and positioning in an optical fiber communication system. According to the method, optical power data are monitored in real time, a deep learning technology is combined, and various fault types in the optical fiber, such as optical fiber fracture and bending, are accurately diagnosed. According to the method, the convolutional neural network is utilized to perform feature extraction on the optical power data acquired in real time, and the fault types are classified through the multi-layer neural network. Based on the trained model, the fault occurrence time and position can be accurately predicted, and an efficient troubleshooting scheme is provided.

Description

Technical Field

[0001] The present invention relates to the field of optical communication technologies, and particularly to an automatic fault diagnosis method and system for optical fiber communication lines based on neural networks. Specifically, the present invention relates to using real-time monitored optical power data combined with a neural network model to automatically diagnose the types and locations of faults in an optical fiber communication link, thereby achieving efficient fault management and network maintenance of an optical fiber communication system. Background Art

[0002] With the wide application of optical fiber communication technologies, the importance of optical fiber lines in long-distance and high-speed data transmission has been increasing. To ensure the stability and reliability of an optical fiber communication system, real-time monitoring of the health status of an optical fiber link has become an important requirement. Optical power monitoring is one of the key means to evaluate whether an optical fiber line is operating normally. However, since faults in an optical fiber line usually manifest as abnormal fluctuations in optical power, and the location and type of these faults when they occur are complex, an efficient and intelligent method is needed to automatically diagnose and locate faults.

[0003] Currently, traditional optical fiber fault diagnosis methods usually rely on manual intervention or rule-based detection algorithms. The limitations of these methods are that they cannot adapt to various complex fault patterns in real time, and it is often difficult to accurately locate the fault location. Therefore, how to achieve fast and accurate diagnosis of faults based on optical power data through intelligent means has become an urgent problem to be solved. Summary of the Invention

[0004] In view of the problems existing in the prior art, the present invention provides an automatic fault diagnosis method and system for optical fiber lines based on neural networks. By collecting real-time optical power data and combining it with a neural network model, it can achieve fast diagnosis and location of the types and locations of optical fiber line faults, overcoming the deficiencies in the traditional technology.

[0005] Specifically, one aspect of the present invention is to provide an automatic fault diagnosis method for optical fiber lines based on neural networks, which is characterized by including the following steps:

[0006] Step S1: Real-time monitor the optical power data of the optical fiber line, and perform data preprocessing, including smoothing processing and normalization;

[0007] Step S2: Use a convolutional neural network (CNN) to extract features from the processed optical power data;

[0008] Step S3: Use a long short-term memory network (LSTM) to model the temporal relationship of the optical power data and analyze the time and location of the fault occurrence;

[0009] Step S4: Based on the features extracted by CNN and LSTM, diagnose the fault type in the optical fiber line and locate the position where the fault occurs;

[0010] Among them, the features extracted by CNN and LSTM in step S4 include: combining the fully connected layer of the neural network, the model can output the probability values of multiple fault types; the output of the model is a vector y = [y 1 , y 2 ,..., y k , where k is the number of fault types. Calculate the probability P(fault type i ) of each fault type through the Softmax function:

[0011]

[0012] Furthermore, the preprocessing steps of the optical power data include:

[0013] Step S11: Smooth the optical power data using the moving average method;

[0014] Step S12: Normalize the optical power data to the range of [0, 1].

[0015] Furthermore, the convolutional neural network (CNN) extracts features through the following steps:

[0016] Step S21: Perform a convolution operation on the optical power data to extract local features;

[0017] Step S22: Further reduce the feature dimension through the pooling layer.

[0018] Furthermore, the long short-term memory network (LSTM) models the temporal relationship through the following steps:

[0019] Step S31: Use the forget gate, input gate, and output gate to update the internal state of the network respectively, calculate the cell state and output the current state; among them, the forget gate (f(t)): controls the degree of forgetting of the previous memory and determines which information should be discarded from the cell state;

[0020] Among them, the input gate (i(t)): controls whether the current input data should be stored and affects the update of the cell state;

[0021] Among them, the output gate (o(t)): determines whether the current cell state should be output to the outside;

[0022] Cell state update (C(t)): Update the cell state by integrating historical information and the current input, so as to capture long-term dependencies;

[0023] Step S32: Identify the fault mode of the optical fiber line by learning the time dependence in the optical power data.

[0024] Furthermore, the fault types include fiber breakage, fiber bending, poor connection, etc.

[0025] Another aspect of the present invention is to provide an automatic fault diagnosis system for optical fiber lines based on a neural network,

[0026] which is characterized by including:

[0027] A data acquisition module for real-time acquisition of optical power data of the optical fiber line;

[0028] A data preprocessing module for smoothing and normalizing the acquired optical power data;

[0029] A feature extraction module, including a convolutional neural network (CNN), for extracting features from the processed optical power data;

[0030] A time series analysis module, including a long short-term memory network (LSTM), for modeling the time series relationship of the optical power data;

[0031] A fault diagnosis module for diagnosing the fault type in the optical fiber line and locating the fault occurrence position based on the outputs of the feature extraction module and the time series analysis module;

[0032] Among them, the features extracted by the fault diagnosis module based on CNN and LSTM include: combining the fully connected layer of the neural network, and the model can output probability values of multiple fault types; the output of the model is a vector y = [y 1 , y 2 ,..., y k , where k is the number of fault types. Calculate the probability P(fault type i ) of each fault type through the Softmax function:

[0033]

[0034] Furthermore, the data preprocessing module includes:

[0035] A smoothing processing unit for smoothing the optical power data using the moving average method;

[0036] A normalization unit for normalizing the optical power data to the range of [0, 1].

[0037] Furthermore, the feature extraction module performs local feature extraction through a convolutional neural network (CNN), and reduces the feature dimension through a pooling layer.

[0038] Furthermore, the timing analysis module models the timing dependence of optical power data through a Long Short-Term Memory network (LSTM) to identify fault patterns in the optical fiber line.

[0039] Furthermore, the fault diagnosis module is used to output the diagnosis results of the fault type and location, and determine the location of the fault point according to the change trend of the optical power data.

[0040] Compared with the existing technical solutions, the present invention has at least the following beneficial effects:

[0041] 1) The present invention combines a neural network model (CNN and LSTM) with optical power data, enabling efficient and accurate optical fiber fault diagnosis and location. By adopting a combined model of Convolutional Neural Network (CNN) and Long Short-Term Memory network (LSTM), the advantages of both are fully exploited. CNN has excellent capabilities in feature extraction and can automatically learn complex feature patterns from the original optical power data, while LSTM can capture long-term dependencies in time series data. Through the combination of CNN and LSTM, the present invention can achieve deep learning and time-dependence modeling of optical power data, thereby accurately and quickly diagnosing optical fiber faults and locating the fault positions without relying on manual intervention. This method can monitor the status of the optical fiber communication system in real time and detect potential faults in a timely manner, greatly improving the efficiency and accuracy of optical fiber fault detection.

[0042] 2) The fault type classification method adopted by the present invention can accurately identify various optical fiber fault types, such as fiber breakage, bending, etc. Fiber breakage is one of the most common optical fiber fault types, usually resulting in a sudden drop or complete loss of optical power. By monitoring the optical power data, especially the sudden power drop, the CNN and LSTM networks can accurately identify the fault pattern of fiber breakage. Bending will cause an increase in the loss of optical fiber signal transmission, usually manifested as a gradual decrease in optical power. The impact of bending faults on optical power is relatively slow, so the LSTM network can capture the power change trend over a long time and identify the occurrence of bending faults. Other faults: including other fault types that affect optical power, such as loose or damaged fiber connection points. Through the deep learning method, the model can learn the characteristics of these faults from a large amount of training data and accurately classify them according to the pattern of optical power data. Moreover, the present invention can infer the time when the fault occurs based on the power data. For example, in the case of fiber breakage, the power will drop sharply, and the model can infer the time point when the fault occurs by monitoring the change of the power value in real time. In the case of fiber bending, the power change is relatively slow, and the LSTM can infer the approximate time range when the fault occurs by capturing the historical power fluctuation trend.

[0043] 3) The method does not rely on manual intervention, can realize automatic fault detection, and improve the operation and maintenance efficiency of the optical fiber communication system. The automatic fault detection method of the present invention is based on a deep neural network model, and can perform real-time monitoring and diagnosis of optical fiber faults without manual intervention. Traditional optical fiber fault detection methods usually rely on manual analysis or simple threshold judgment, which often requires a lot of manual intervention and experience judgment, and it is difficult to handle complex fault modes and large-scale optical fiber communication networks. The present invention, by using a neural network model, can automatically complete the detection, diagnosis and positioning of faults without manual intervention, wherein automatic detection: the optical power data processing method of the present invention can automatically receive real-time optical power signals and analyze them through trained CNN and LSTM models. This means that the operation and maintenance personnel of the optical fiber communication system do not need manual intervention, and the system can run autonomously and monitor the health status of the optical fiber in real time. Real-time fault diagnosis: Through the application of deep learning models, the speed and accuracy of fault diagnosis are greatly improved. The system can immediately identify when a fault occurs in the optical fiber, and infer the fault type, occurrence time and location by analyzing the optical power data. Whether it is a break, bend or other fault type, the system can diagnose through an automated method, avoiding the delay and error of manual diagnosis. Automated fault detection and diagnosis can effectively reduce the workload of manual monitoring and improve the system's operation and maintenance efficiency. Operation and maintenance personnel only need to handle according to the system's automatic alarms and reports, without frequent manual inspections and judgments of fault points. By reducing human intervention, the system can achieve more efficient and reliable operation and maintenance management, especially in the case of large-scale fiber-optic communication networks, where automated methods are particularly important.

[0044] Through the application of neural network models, the present invention successfully realizes the automated detection, diagnosis and location of optical fiber faults, significantly improves the operating efficiency of the optical fiber communication system, and reduces the need for manual intervention, thereby improving the reliability and operation and maintenance efficiency of the optical fiber communication system. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 It is the overall architecture of the optical fiber communication fault automatic diagnosis system of the present invention;

[0046] Figure 2 It is a model architecture combining a convolutional neural network (CNN) and a long short-term memory network (LSTM) of the present invention;

[0047] Figure 3 It is a schematic diagram of optical power data processing and fault location of the present invention;

[0048] Figure 4 It is a flow chart of fault management and operation and maintenance of the optical fiber communication network of the present invention.

[0050] 100. Automatic Optical Fiber Communication Fault Diagnosis System, 10. Data Acquisition Module, 20. Data Preprocessing Module, 30. Feature Extraction Module, 31. Convolutional Layer, 32. Pooling Layer, 33. Fully Connected Layer, 34. LSTM Layer, 35. Output Layer, 40. Time Series Analysis Module, 50. Fault Diagnosis Module

[0051] The present invention will be further described in detail below. However, the following examples are merely simple examples of the present invention and do not represent or limit the scope of the claimed rights of the present invention. The scope of protection of the present invention shall be subject to the claims. Specific Embodiment

[0052] The technical solution of the present invention will be further described below in conjunction with the accompanying drawings and specific embodiments.

[0053] As Figure 1 shown, it is the overall architecture of the automatic optical fiber communication fault diagnosis system of the present invention. The automatic optical fiber communication fault diagnosis system 100 includes a data acquisition module 10, a data preprocessing module 20, a feature extraction module 30, a time series analysis module 40, and a fault diagnosis module 50.

[0054] 1. Optical Power Data Acquisition and Preprocessing

[0055] First, the optical power data on the optical fiber line is collected in real time through the optical module to be measured. Assume that at time point t, the optical power is P(t), where t ∈ [t1, t2] is the time range during detection.

[0056] The optical power data P(t) usually presents as a continuous signal varying with time. In the data preprocessing stage, we use the following two methods to process the optical power data:

[0057] Smoothing to obtain F smoothed (t): Use the moving average method to smooth the power data and remove noise.

[0058]

[0059] where n is the window size, usually taken as 5 - 10 to reduce the influence of external environmental interference.

[0060] Normalization to obtain P normalized (t): Normalize the optical power data to the range of [0, 1] to reduce the influence of power differences between different optical fiber lines:

[0061]

[0062] where P min and P max are respectively the minimum and maximum values in the collected power data.

[0063] 2. Convolutional Neural Network (CNN) Feature Extraction

[0064] As Figure 2 shown, it is the model architecture combining the Convolutional Neural Network (CNN) and the Long Short-Term Memory Network (LSTM) provided by the present invention. This model is used for feature extraction and time series analysis of optical power data.

[0065] After preprocessing the optical power data, we send the data into the Convolutional Neural Network (CNN) for feature extraction. CNN can effectively identify local features in the optical power data, such as power mutations, fluctuations, etc.

[0066] The operations of the convolutional layer are as follows:

[0067]

[0068] Among them, F conv (t) represents the output of the convolutional layer, W i is the convolutional kernel (filter), b is the bias term, k is the size of the convolutional kernel, and this k value is usually set to 3 or 5.

[0069] A smaller convolutional kernel (k = 3) has a smaller computational amount. Especially when used in a multi-layer network, the cumulative effect of the computational amount will be more obvious. It can gradually extract more complex features by increasing the number of layers and usually has a better effect in practical applications.

[0070] A larger convolutional kernel (k = 5) has a larger computational amount because it involves more multiplication and addition operations. If the computational efficiency is not considered during network design, using a larger convolutional kernel may bring a large computational overhead.

[0071] Through the convolutional layer, local patterns in the optical power signal can be extracted, such as the trend of power change, mutation points, or stable intervals in the signal. Next, CNN further reduces the dimension of the data (i.e., the width and height of the space) through the pooling layer (such as max pooling), extracts more representative features, and retains the most important feature information at the same time. This dimension compression not only reduces the complexity of subsequent calculations but also helps the network achieve translational invariance to a certain extent and improve the robustness of the model.

[0072] 3. Long Short-Term Memory Network (LSTM) Modeling

[0073] To process the time series characteristics of optical power data, we adopt the Long Short-Term Memory Network (LSTM). LSTM can effectively process time series data and capture long-term dependencies in the data, which is very important for detecting faults in optical fiber lines.

[0074] The state of LSTM is updated through the following formula:

[0075] Forget gate: Controls the degree to which previous memories are forgotten.

[0076] f(t) = σ(W f ·[h(t - 1), P smoothed (t)] + b f ) (4)

[0077] In optical fiber fault diagnosis, not all historical information is helpful for current fault detection. Through the forget gate, LSTM can automatically select which historical information is important based on real-time optical power data, thus effectively eliminating irrelevant past information and maintaining the flexibility and accuracy of the model.

[0078] Input gate: Controls whether the current input should be saved.

[0079] i(t) = σ(W i ·[h(t - 1), P smoothed (t)] + b i ) (5)

[0080] During the optical fiber fault diagnosis process, different fault types affect the optical power in different ways. The input gate enables LSTM to automatically adjust the storage strategy for new information according to the change patterns in the power data, ensuring that the model responds sensitively to power changes at critical moments.

[0081] Output gate: Controls the output of the current state.

[0082] o(t) = σ(W o ·[h(t - 1), P smoothed (t)] + b o ) (6)

[0083] During the optical fiber fault diagnosis process, accurately judging the time and type of fault occurrence is of great significance for subsequent fault location and repair. The design of the output gate enables LSTM to accurately identify fault patterns in complex optical power fluctuations, avoiding misjudgment problems that may exist in simple threshold judgment methods.

[0084] Cell state update: Updates the cell state by integrating historical information and the current input.

[0085] C(t) = f(t)·C(t - 1) + i(t)·tanh(W C ·[h(t - 1), P smoothed (t)] + b C ) (7)

[0086] In fiber optic fault diagnosis, faults do not occur instantaneously. They may develop gradually or occur suddenly. Through continuous updates of the cell state, LSTM can gradually capture the development process of fiber optic faults, ensuring that the time point of fault occurrence can be identified in a timely manner and avoiding the problem of information loss in static feature extraction methods for time series data.

[0087] Final output: Output the current LSTM state based on the cell state.

[0088] h(t) = o(t) · tanh(C(t)) (8)

[0089] Among them, f(t), i(t), and o(t) are the activation functions of the forget gate, input gate, and output gate respectively, W f , W i , W o , W C are weight matrices, b f , b i , b o , b C are bias terms, and σ is the sigmoid and tanh activation functions.

[0090] The output of LSTM is not just a single judgment result, but a comprehensive evaluation based on long-term historical data and current power data, greatly improving the accuracy of fault detection. Different from traditional threshold-based fault diagnosis methods, LSTM can dynamically judge whether a fiber optic fault occurs and make a judgment based on the complex pattern of power fluctuations.

[0091] In the LSTM network, the above-mentioned cell state is the core part of the information transmitted in the network and is used to store information about long-term dependencies. It flows between time steps, transmitting the long-term memory of the sequence. LSTM solves the problem of vanishing gradients in standard RNNs when dealing with long-term dependencies through the cell state. The update of the cell state depends on the gating mechanism, and these gates (input gate, forget gate, and output gate) determine which information should be remembered or forgotten.

[0092] Through the LSTM network, the model can identify the temporal dependencies in the optical power data, such as the power decline trend, mutations, or fluctuations when a fiber optic line fails. Using the LSTM network is in line with fiber optic line monitoring. The power changes caused by fiber optic faults usually exhibit a certain temporal dependence. For example, the power attenuation caused by fiber optic bending is usually a gradual change, while a fiber optic break may cause a power mutation. Regardless of the type of fault, the change in optical power over time is regular, and these regularities may only become apparent over a long period of time. Therefore, it is necessary to effectively model the temporal dependencies. The long short-term memory network is a special type of recurrent neural network, and its core advantage lies in its ability to effectively process and learn long-term dependencies in time series data. When processing time series data, the LSTM can retain important information from previous time steps and dynamically adjust the memory based on the current input data.

[0093] By inputting the optical power data into the LSTM network, the model can identify and learn the power change trends before and after a fiber optic fault. Specifically, the LSTM can learn the temporal dependencies in the following ways:

[0094] a. Power decline trend:

[0095] For progressive faults such as fiber optic bending, the power decline trend of the optical power is usually slow and continuous. The LSTM network can capture the trend of the optical power gradually decreasing over time and infer the fault occurrence time of the fiber optic based on historical data. For example, when the power gradually decreases at several consecutive monitoring points and does not recover, the LSTM can identify this gradual change trend and, under the temporal dependence learning of the model, make a fault prediction.

[0096] b. Mutation:

[0097] For faults such as fiber optic breaks, the power may mutate at a certain moment, that is, the optical power suddenly drops to a lower value (such as below -20 dBm). The LSTM can identify this "mutation" feature of the power and capture this change in a short time. Especially when there are obvious differences in the power changes before and after the time point of the break, the LSTM will capture this sudden change through the data in the time window and predict the occurrence of the fault.

[0098] c. Fluctuation:

[0099] In some cases (such as slight bending or unstable connections), the power change may show a fluctuating state. By analyzing the fluctuation characteristics of the optical power, the LSTM can determine whether these fluctuation patterns are related to faults. For example, if periodic fluctuations appear in the monitoring data and these fluctuations are different from the normal power change patterns, the LSTM can identify these abnormal fluctuations and associate them with specific types of faults.

[0100] 4. Fault Diagnosis and Location

[0101] As Figure 3 shown, it is a schematic diagram of optical power data processing and fault location of the present invention, including data acquisition, preprocessing, feature extraction, time series analysis, and final fault diagnosis processes. It shows the relationship between optical power data at different time points and fault location.

[0102] Through the combination of CNN and LSTM networks, the system can perform fault diagnosis based on optical power data. Specifically, the fault diagnosis process includes the following steps:

[0103] Fault type classification: Using the features extracted by CNN and LSTM, combined with the fully connected layer of the neural network, the model can output the probability values of multiple fault types, such as fiber breakage, fiber bending, poor connection, etc. Suppose the output of the model is a vector y = [y 1 , y 2 ,..., y k , where k is the number of fault types. Calculate the probability of each fault type through the Softmax function:

[0104]

[0105] There are 3 possible fault types here: fiber breakage (Fault Type 1), fiber bending (Fault Type 2), and normal state (Fault Type 3). The output vector of the model may be like this, for example:

[0106] y = [0.85, 0.10, 0.05]

[0107] Among them, y 1 = 0.85: indicates that the probability of fiber breakage is 85%. y 2 = 0.10: indicates that the probability of fiber bending is 10%. y 3 = 0.05: indicates that the probability of the fiber being normal is 5%. In this way, the type of fiber fault can be judged according to the output vector, and the corresponding fault type can be obtained (for example, here the model judges that the fault is most likely fiber breakage because the value of y1 is the largest).

[0108] k is the number of fault types. That is, the number of different fault types that the model can classify. For example, if the fault types we consider are: fiber breakage, fiber bending, fiber contamination, connector failure, etc., then k = 4, and the output vector of the model will have 4 elements, and each element corresponds to the probability of a fault type.

[0109] If there are only 3 fault types, the output vector will be of length 3. If there are more fault types, the dimension of the output vector will be higher.

[0110] Fault location: Combining the trend of power loss and the diagnostic results output by the neural network, the model can further estimate the fault location based on the characteristics of the fault type.

[0111] Through the above steps, the system can diagnose the fault type of the optical fiber line in a short time and locate the fault point Fp (Fault Position) through the trend of optical power change.

[0112] F p = f(t 0 , ΔP) (10)

[0113] where ΔP is the change in optical power, and t 0 is the time point when the change occurs. The function f is a positioning model designed according to the transmission characteristics of the optical fiber and the power attenuation law.

[0114] Through training with a large amount of optical power data in normal and faulty conditions, LSTM can predict the corresponding state for each optical power data point. The output of the LSTM model will be combined with a time window for prediction. As time progresses, LSTM not only depends on the power data at the current moment but also takes into account the power changes in the previous few moments. This way of considering historical data enables LSTM to identify the early signals of faults and infer the time points.

[0115] In addition, LSTM calculates the power fluctuation trend at each moment and judges whether the current power change conforms to the fault pattern based on the learning results of historical data. If the current power fluctuation matches the previously recorded fault pattern, LSTM will infer that the fault occurred within a certain time range around the current time point.

[0116] Specifically, the LSTM network can map the input time-series optical power (optical power varying with time) data to a set of output results, and the output results include the time when the fault occurs. Based on the changes in time dependence in the optical power data, LSTM can accurately predict the time point when the fault occurs. For example, by analyzing the rate and trend of power decline, LSTM can make a prediction before the power decline reaches a certain threshold. A key piece of information:

[0117] Assume that the optical power data sequence is P(t) (where t is the timestamp). LSTM calculates the power value at the current moment t through the historical power data P(t - k), P(t - k + 1), P(t - 1):

[0118] h(t) = LSTM(P(t - k), P(t - k + 1),..., P(t - 1)) (11)

[0119] Among them, h(t) is the output of the LSTM, representing the predicted state at the current moment. By comparing the predicted state h(t) with the actual optical power P(t), the LSTM can determine whether there is an abnormality and infer the time window of the fault. When the difference between the state h(t) output by the LSTM and the actual power value P(t) exceeds a certain predetermined threshold, the system will determine that this moment may be the moment when the fault occurs. If the predictions at multiple consecutive time points differ significantly from the actual power values, the system will infer that the fault occurred within this period of time, and then determine the time range of the fault occurrence.

[0120] 5. Model Training and Optimization

[0121] To ensure that the neural network can perform accurate fault diagnosis under various environments and fault conditions, the training data includes various different types of fault samples, and factors such as different lengths of optical fiber lines and different light source intensities are considered. The cross-validation method is used to train the neural network model, and the loss function is used to optimize the accuracy of the model. Commonly used loss functions include the cross-entropy loss function and the mean squared error loss function, and their specific forms are as follows:

[0122] Cross-entropy loss function (for classification):

[0123]

[0124] Mean squared error loss function (for regression or localization):

[0125]

[0126] Among them, y true,i is the true label, y pred,i is the predicted result, and N is the number of samples.

[0127] Through training, the neural network can gradually optimize the parameters and improve the accuracy and real-time performance of fault diagnosis.

[0128] As Figure 4 shown, it is the flowchart of fault management and operation and maintenance of the optical fiber communication network of the present invention, showing the automated process of fault management and operation and maintenance in the optical fiber communication network, including links such as fault diagnosis, fault alarm, and location repair.

[0129] An automatic fault diagnosis method for optical fiber communication lines based on a neural network, the method includes the following steps:

[0130] Step S1: Monitoring step, showing that the network monitoring device collects optical power data in real time;

[0131] Step S2: Fault diagnosis step, showing that the collected data is subjected to fault diagnosis through the neural network model of the present invention;

[0132] Step S3: Fault alarm step. If a fault is diagnosed, the system will trigger a fault alarm;

[0133] Step S4: Fault location and repair step. Finally, the fault is located according to the diagnosis result and the repair process is triggered.

[0134] By setting optical power monitoring nodes in a certain optical fiber communication network, and then through the above fault diagnosis method, optical power data is obtained in real time, and a convolutional neural network (CNN) is applied to identify the fault type and infer the location and time of the fault occurrence. The optical power monitoring data used in the experiment comes from each monitoring point of the optical fiber network.

[0135] Optical power data acquisition:

[0136] Suppose the network consists of 5 monitoring points, and the optical power value of the optical fiber is regularly collected at each node. The unit of the optical power value is dBm, and the sampling time interval of each monitoring node is 10 seconds. Each monitoring point records one optical power data point per second. We selected optical power data under 5 different fault situations for simulation.

[0137] Situation 1: Optical power at each point and time point in the normal optical fiber communication state

[0138] Table 1-1 Optical power in the normal optical fiber communication state

[0139]

[0140] Under normal circumstances, the optical power changes of all monitoring points are relatively stable and the values are close.

[0141] Situation 2: Optical power at each point and time point in the optical fiber break fault state

[0142] Table 1-2 Optical power in the optical fiber break fault state

[0143]

[0144] When the optical fiber break fault occurs, the optical power of the monitoring point drops significantly, and the optical power suddenly drops below -20 dBm after the fault occurs.

[0145] Situation 3: Optical power at each point and time point in the optical fiber bend fault state

[0146] Table 1-3 Optical power in the optical fiber bend fault

[0147]

[0148] When the optical fiber is bent, the optical power gradually decreases, and there may be fluctuations between different monitoring points. The power reduction caused by the bend fault is relatively slow, and the changes of monitoring points 1, 3, and 5 are relatively large.

[0149] Fault identification using a convolutional neural network:

[0150] After obtaining the above experimental data, we used a convolutional neural network (CNN) to train and identify faults in the optical power time series data.

[0151] Data preprocessing:

[0152] Normalize the optical power data of each monitoring point.

[0153] Each data point forms a sequence as the input to the CNN.

[0154] Network architecture:

[0155] Input layer: The size of the input data is (N×T), where N is the number of monitoring points and T is the number of time steps.

[0156] Convolutional layer: Use multiple convolutional kernels for feature extraction. The size of the convolutional kernel is 3×3 and the stride is 1.

[0157] Pooling layer: Use a max-pooling layer to reduce the data dimension. The size of the pooling window is 2×2.

[0158] Fully connected layer: Output the classification result to determine the fault type.

[0159] Fault identification and speculation:

[0160] After training, the CNN can accurately classify different fault types according to the change pattern of the optical power. For the above data, the network can identify:

[0161] Normal state: Output "no fault".

[0162] Fiber break: Output "fiber break", and speculate that the fault occurred around 20 seconds through the time information output by the model.

[0163] Fiber bend: Output "fiber bend", and speculate that the fault occurred around 30 seconds through the time information output by the model.

[0164] Fault speculation:

[0165] Based on the time series of the optical power change, the system can speculate the location and time of the fault. For example:

[0166] In the case of a fiber break, the model will output that the fault occurred at 20 seconds and speculate that the monitoring points where the fault occurred are monitoring points 1-5.

[0167] In the case of a fiber bend, the model will output that the fault occurred at 30 seconds and speculate that the monitoring points where the fault occurred are monitoring points 1, 3, and 5.

[0168] Result analysis:

[0169] Accuracy: The fault recognition accuracy of the CNN model in this experiment is 98%.

[0170] Real-time performance: This fault detection system can give the fault type and location in real time within 5 seconds after the fault occurs.

[0171] By monitoring the optical power data in real time and applying deep learning models, the present invention can accurately identify the fault type in the optical fiber and quickly infer the time and location of the fault after the fault occurs.

[0172] The preferred embodiments of the present invention have been described in detail above. However, the present invention is not limited to the specific details in the above embodiments. Within the scope of the technical concept of the present invention, various simple modifications can be made to the technical solutions of the present invention, and these simple modifications all fall within the protection scope of the present invention.

[0173] In addition, it should be noted that, in the case of no conflict, the various specific technical features described in the above specific embodiments can be combined in any suitable way. To avoid unnecessary repetition, the present invention will not separately describe various possible combination methods.

[0174] Furthermore, any combination can be made between various different embodiments of the present invention, as long as it does not violate the idea of the present invention, and it should also be regarded as the content disclosed by the present invention.

Claims

1. A method for automatic diagnosis of optical fiber line faults based on neural network, characterized in that: The following steps are involved: Step S1: real-time monitoring of optical power data of the optical fiber line, and data preprocessing, including smoothing and normalization; Step S2: using a convolutional neural network (CNN) to perform feature extraction on the processed optical power data; Step S3: Use a long short-term memory network (LSTM) to model the temporal relationship of the optical power data and analyze the time and location of the fault. Step S4: Based on the features extracted by CNN and LSTM, diagnose the fault type in the optical fiber line and locate the location where the fault occurs; The features extracted based on CNN and LSTM in step S4 include: combining the fully connected layer of the neural network, the model can output probability values ​​of multiple fault types; the output of the model is a vector y = [y1, y2, ..., y k ], where k is the number of fault types. The probability P (fault type i ):

2. The method according to claim 1, characterized in that: The optical power data preprocessing step comprises: Step S11: smoothing the optical power data using a sliding average method; Step S12: Normalize the optical power data to the range of [0, 1].

3. The method according to claim 1, characterized in that The convolutional neural network (CNN) extracts features through the following steps: Step S21: performing a convolution operation on the optical power data to extract local features; Step S22: further reduce the feature dimension through the pooling layer.

4. The method according to claim 1, characterized in that: The LSTM network models temporal relationships through the following steps: Step S31: Use the forget gate, input gate and output gate to update the internal state of the network respectively, calculate the cell state and output the current state; The forget gate (f(t)) controls the degree of forgetting of previous memories and determines which information should be discarded from the cell state; The input gate (i(t)) controls whether the current input data should be stored and affects the update of the cell state; The output gate (o(t)) determines whether the current cell state responds to external output; Cell state update (C(t)): Update the cell state by integrating historical information and current input to capture long-term dependencies; Step S32: Identify the failure mode of the optical fiber line by learning the time dependency in the optical power data.

5. The method according to claim 1, characterized in that: The fault types include optical fiber breakage, optical fiber bending, poor connection, etc.

6. An automatic optical fiber line fault diagnosis system based on neural network, characterized in that: include: Data acquisition module, used to collect optical power data of optical fiber lines in real time; A data preprocessing module is used to smooth and normalize the collected optical power data; A feature extraction module, including a convolutional neural network (CNN), for extracting features from the processed optical power data; A timing analysis module, including a long short-term memory network (LSTM), is used to model the timing relationship of optical power data; A fault diagnosis module, used to diagnose the fault type in the optical fiber line and locate the location where the fault occurs based on the outputs of the feature extraction module and the timing analysis module; The features extracted based on CNN and LSTM in the fault diagnosis module include: combined with the fully connected layer of the neural network, the model can output probability values ​​of multiple fault types; the output of the model is a vector y = [y1, y2, ..., y k ], where k is the number of fault types. The probability P (fault type i ):

7. The system according to claim 6, characterized in that The data preprocessing module comprises: A smoothing processing unit, used for smoothing the optical power data using a sliding average method; The normalization unit is used to normalize the optical power data to the range of [0,1].

8. The system according to claim 6, characterized in that The feature extraction module performs local feature extraction through a convolutional neural network (CNN) and reduces feature dimensions through a pooling layer.

9. The system according to claim 6, characterized in that The timing analysis module models the timing dependency of optical power data through a long short-term memory network (LSTM) to identify fault modes in optical fiber lines.

10. The system according to claim 6, characterized in that The fault diagnosis module is used to output the diagnosis results of the fault type and the fault location, and determine the location of the fault point according to the change trend of the optical power data.

Images