The present invention will be further described below with reference to the accompanying drawings and embodiments.
 It should be noted that the following detailed description is exemplary and intended to provide further explanation of the application. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs.
 It should be noted that the terms used herein are for the purpose of describing particular embodiments only and are not intended to limit exemplary embodiments in accordance with the present application; as used herein, unless the context clearly dictates otherwise, the singular forms are also intended to include Plural forms, furthermore, should also be understood that when the terms "comprising" and/or "comprising" are used in this specification, they indicate the presence of features, steps, operations, devices, components and/or combinations thereof.
 A method for detecting fiber network link defect data based on a deep belief network, comprising the following steps:
 Step 1, optical fiber network link data collection, according to the optical fiber network link of the power company, the fiber grating sensor is arranged at the position between the central computer and each front-end computer and the link, and then the optical fiber network link data collection is performed.
 In order to ensure that the collected optical fiber network link defect data conforms to the processing specifications of the subsequent deep learning algorithm, the data needs to be preprocessed, including denoising, interpolation, equalization, scaling, and duration processing.
 Step 2: Since the optical fiber network link data will be affected by noise during the acquisition process, the useful data will be covered and the accuracy of the subsequent data detection will be disturbed. Therefore, the wavelet threshold denoising method is used to remove the optical fiber network link defect data. noise processing. Define the fiber network link data as:
 where s(t) is the real data on the link; n(t) is the variance α 2 Gaussian white noise, obeying N(0, α 2 ) distribution, so wavelet transform is performed on the noisy link data to obtain a set of wavelet coefficients, and the denoising process is completed by thresholding the coefficients and then performing wavelet reconstruction.
 Step 3: During the data collection process, missing values will inevitably occur due to various reasons, and the most important requirement for time series data analysis is to ensure the integrity of the data. Therefore, by calculating the mean and covariance of the data, the mean is used to replace the data. After the regression equation converges, the estimated value of the last regression is used as the missing value for data interpolation.
 Step 4: Data imbalance is the main reason for the performance degradation of the classifier during training caused by subsequent deep learning. Therefore, the SMOTE algorithm is used for data balance processing. The sampling nearest neighbor algorithm calculates the K nearest neighbors of each minority class sample. Randomly pick N samples from the K nearest neighbors for random linear interpolation. Construct new minority class samples. Synthesize the new samples with the original data to generate a new training set to enrich the data.
 In step 5, the dimension of the collected optical fiber network link defect data may be inconsistent due to its own data characteristics. This will affect the classifier convergence speed and training time in the later stage. Therefore, Min-Max and Z-score normalization methods are used to scale the data to keep the data scale consistent. This is because the collected optical fiber network link defect data may have inconsistent dimensions due to its own data characteristics. The specific formula is as follows;
 Min-Max normalization:
 Z-score normalization:
 Among them, x^' normalized data, x is the original data, max and min are the maximum and minimum values of the sample data, a is the mean value of the corresponding feature, and b is the standard deviation.
 Step 6: Since the collected fiber network link defect data is temporal, it is a long sequence with timestamps, which is not conducive to subsequent feature extraction. Therefore, the TSSA time series segmentation method is used to discretize the long sequence and divide it into Fixed-size segments, resulting in equally spaced time series segments.
 Step 7: The optical fiber network link data has time characteristics, so a time series feature extraction method based on rule iteration is used, so the sliding window averaging method is used to generate a feature sequence with an iterative method, and statistical features, signal features and physical features are extracted from it. features, and then encapsulate these features into a feature set for subsequent classifiers.
Step 8: The fiber network link defect data samples to be detected are used as input values for forward propagation through the deep confidence network to obtain feature values, which are then input into the support vector description anomaly detection and judgment model, and the model is used for defect data identification. , complete the data detection. Specifically, the model is a deep belief network (DBN) analysis model. The fiber network link defect data samples to be detected are used as input values to propagate forward through the deep confidence network to obtain eigenvalues, which are then input into the Support Vector Description Anomaly Detection and Judgment Model (SVDD), which is used to identify defect data. , complete data detection, such as figure 1 shown.
 The invention proposes a method for detecting optical fiber network link defect data based on a deep confidence network, which realizes optical fiber network link defect data detection in combination with the unbalanced company link data. It can fully consider the unbalanced link data. During the process of data collection and transmission in the network link, it can effectively detect the situation of data defects due to various reasons.