CN120185705B

CN120185705B - Power distribution communication network fault positioning method and system based on deep learning

Info

Publication number: CN120185705B
Application number: CN202510638538.3A
Authority: CN
Inventors: 钱锦; 罗少杰; 黄红兵; 杨鸿珍; 张利军; 杜猛俊; 范明霞; 肖艳炜; 范超; 邱兰馨; 凌芝; 孙剑; 王剑; 郝春昀; 丁晖; 周靖淞; 由奇林; 沈佳辉; 张清波; 张吉
Original assignee: State Grid Zhejiang Electric Power Co Ltd; Hangzhou Power Supply Co of State Grid Zhejiang Electric Power Co Ltd; Information and Telecommunication Branch of State Grid Zhejiang Electric Power Co Ltd; Economic and Technological Research Institute of State Grid Zhejiang Electric Power Co Ltd
Current assignee: State Grid Zhejiang Electric Power Co Ltd; Hangzhou Power Supply Co of State Grid Zhejiang Electric Power Co Ltd; Information and Telecommunication Branch of State Grid Zhejiang Electric Power Co Ltd; Economic and Technological Research Institute of State Grid Zhejiang Electric Power Co Ltd
Priority date: 2025-05-19
Filing date: 2025-05-19
Publication date: 2025-08-12
Anticipated expiration: 2045-05-19
Also published as: CN120185705A

Abstract

The present invention discloses a method and system for locating faults in a power distribution communication network based on deep learning. The method includes fitting acquired time series data of optical fiber links to obtain an actual power attenuation curve, calculating the absolute difference between the actual power attenuation curve and the theoretical power attenuation curve to obtain standardized difference data; inputting the standardized difference data into a nonlinear attenuation model to obtain nonlinear change characteristic data; inputting the distribution network topology characteristic data and the nonlinear change characteristic data into a fault identification model to obtain a failure probability value for each optical fiber link; comparing the failure probability value with a preset failure probability threshold to obtain a potential fault link; performing feature extraction and mutation node identification on the potential fault link to obtain attenuation mutation node data; processing the attenuation mutation node data and determining fault point data based on the obtained stress value data. The method of the present invention achieves rapid location of loss faults in optical fiber networks.

Description

Power distribution communication network fault positioning method and system based on deep learning

Technical Field

The invention relates to the technical field of fault analysis, in particular to a power distribution communication network fault positioning method and system based on deep learning.

Background

The power distribution communication network is used as a key infrastructure of the intelligent power grid and bears core tasks of power system data transmission, equipment control and state monitoring. The optical fiber link has the advantages of high bandwidth and interference resistance, and becomes a main transmission medium of the power distribution communication network. However, the optical fiber link is easily affected by external force extrusion, temperature change, joint aging and other factors, so that the power of an optical signal is attenuated or even interrupted, and the safe and stable operation of the power distribution network is seriously threatened.

The existing optical fiber fault positioning method relies on hardware equipment such as an optical time domain reflectometer and the like, and fault point detection is carried out by analyzing an attenuation curve of a backward scattered light signal. The single detection period of fault location of the method is tens of seconds to minutes, and the fault evolution speed of the power distribution communication network exceeds the data updating frequency, so that the fault cannot be located in time when the fault occurs.

Disclosure of Invention

The invention provides a power distribution communication network fault positioning method and a power distribution communication network fault positioning system based on deep learning, which aim to solve the technical problem of how to improve the existing power distribution communication network fault positioning method and achieve the effect of reducing the power distribution communication network fault processing time.

In order to solve the technical problems, an embodiment of the present invention provides a power distribution communication network fault positioning method based on deep learning, including:

Acquiring optical power data of an optical fiber link in a target power distribution network, and calculating optical power difference values of adjacent sampling points in the optical power data to generate time sequence data;

Fitting the time sequence data to obtain an actual power attenuation curve, and calculating the absolute difference value of the actual power attenuation curve and a theoretical power attenuation curve to obtain standardized difference data;

Inputting the standardized difference data into a pre-constructed nonlinear attenuation model to obtain nonlinear variation characteristic data reflecting the power attenuation of the optical signal in the optical fiber link;

extracting features of a physical topological structure of the target power distribution network to obtain power distribution network topological feature data, and inputting the power distribution network topological feature data and the nonlinear variation feature data into a pre-constructed fault identification model to obtain fault probability values of all optical fiber links;

Comparing the fault probability value with a preset fault probability threshold value to obtain a potential fault link, and sequentially carrying out feature extraction and mutation node identification on the potential fault link to obtain attenuation mutation node data;

and processing the attenuation abrupt change node data according to a stress inversion method, and determining fault point data in the target power distribution network based on the obtained stress value data.

As one preferable solution, the obtaining optical power data of an optical fiber link in the target power distribution network, calculating an optical power difference value of adjacent sampling points in the optical power data, and generating time sequence data includes:

Acquiring original optical power data of an optical fiber link in a target power distribution network in real time, and preprocessing the original optical power data;

time sequencing is carried out on the preprocessed original optical power data to obtain an optical power sequence data set;

Calculating the optical power difference between adjacent sampling points in the optical power sequence data set to obtain difference data, wherein if the optical power difference exceeds a preset optical power fluctuation threshold value, the optical power difference is marked as an abnormal data point;

and screening non-abnormal data points in the difference data to obtain time sequence data.

As one preferable mode, before the fitting processing is performed on the time series data, the method further comprises:

Performing linear regression processing on the time sequence data to obtain optical power trend data, and calculating the mean square error of theoretical stress characteristic data of the optical fiber link and the optical power trend data;

and if the mean square error exceeds a preset error threshold, fitting the time sequence data.

As one preferable mode, the calculating the mean square error between the theoretical stress characteristic data of the optical fiber link and the optical power trend data includes:

Generating stress distribution data according to the obtained stress distribution parameters of the optical fiber link, and carrying out frequency domain feature extraction on the stress distribution data by adopting Fourier transformation to obtain theoretical stress feature data;

Performing linear regression processing on the time series data by adopting a least square method to obtain an optical power trend function, and obtaining a trend function value of each sampling time point according to the optical power trend function to form the optical power trend data;

And carrying out time alignment processing on the theoretical stress characteristic data and the optical power trend data, and calculating the mean square error of the theoretical stress characteristic data and the optical power trend data after time alignment.

As one preferable solution, the fitting processing is performed on the time series data to obtain an actual power attenuation curve, and calculating an absolute difference value between the actual power attenuation curve and a theoretical power attenuation curve to obtain normalized difference data includes:

Performing curve fitting treatment on the time sequence data based on a cubic spline interpolation method to obtain an actual power attenuation curve;

Constructing a theoretical power attenuation curve according to the physical parameters of the optical fiber link;

Calculating the absolute difference value of each sampling point on the time axis of the actual power attenuation curve and the theoretical power attenuation curve to obtain an absolute difference value sequence reflecting the attenuation abnormality degree;

And carrying out normalization processing on the absolute difference sequence to obtain the standardized difference data.

As one preferable scheme, the feature extraction of the physical topology structure of the target power distribution network to obtain power distribution network topology feature data includes:

constructing a node adjacency matrix and a link attribute table based on the acquired physical connection relation of the optical fiber link;

Summing each row of the node adjacency matrix to obtain a node degree vector reflecting the importance degree of the node;

calculating link weight data according to the link attribute table, and constructing a link weight matrix based on the link weight data;

And performing splicing processing on the node degree vector and the link weight matrix to obtain the topological characteristic data of the power distribution network.

As one preferable solution, the comparing the fault probability value with a preset fault probability threshold to obtain a potential fault link, and sequentially performing feature extraction and mutation node identification on the potential fault link to obtain attenuation mutation node data, including:

Comparing the fault probability value with a preset fault probability threshold value, screening out links with the fault probability value larger than or equal to the preset fault probability threshold value, and generating a potential fault link list;

Acquiring optical power sequence data of a link corresponding to the fault link list, and converting the optical power sequence data into a two-dimensional matrix;

The two-dimensional matrix is spliced with the standardized difference data and the nonlinear variation characteristic data to obtain a fusion characteristic tensor, and the fusion characteristic tensor is input into a pre-constructed depth residual error network to perform characteristic extraction to obtain attenuation mutation characteristic data;

and carrying out cluster analysis on the attenuation mutation characteristic data to obtain the attenuation mutation node data with the attenuation trend changed.

As one preferable solution, the processing the attenuation abrupt node data according to the stress inversion method, and determining the fault point data in the target power distribution network based on the obtained stress value data includes:

constructing a stress attenuation relation model according to the obtained stress sensitive parameters, the material stress threshold and the physical length of the optical fiber link;

Extracting the power attenuation amount and the physical length of the link corresponding to each attenuation abrupt change node data, and inputting the power attenuation amount and the physical length into the stress attenuation relation model to obtain a stress value data set;

comparing each stress value data in the stress value data set with the material stress threshold value respectively, and screening comparison results to obtain a high stress mutation node list;

And determining fault point data in the target power distribution network based on the high-stress abrupt change node list.

As one preferable scheme, the determining the fault point data in the target power distribution network based on the high-stress abrupt change node list includes:

Constructing a candidate fault point set based on each node and adjacent nodes thereof in the high-stress abrupt node list;

Acquiring coordinate data of each node in the candidate fault point set, and calculating the fault point relative distance between each high-stress abrupt change node and the adjacent node based on the coordinate data;

And performing multi-condition screening according to the obtained stress value of each high-stress abrupt change node and the relative distance between the fault points, and obtaining the fault point data according to the multi-condition screening result.

Another embodiment of the present invention provides a power distribution communication network fault location system based on deep learning, including:

The acquisition module is used for acquiring optical power data of an optical fiber link in the target power distribution network, calculating optical power difference values of adjacent sampling points in the optical power data and generating time sequence data;

the calculation module is used for carrying out fitting processing on the time sequence data to obtain an actual power attenuation curve, and calculating the absolute difference value of the actual power attenuation curve and the theoretical power attenuation curve to obtain standardized difference data;

The first extraction module is used for inputting the standardized difference data into a pre-constructed nonlinear attenuation model to obtain nonlinear variation characteristic data reflecting the power attenuation of the optical signal in the optical fiber link;

The second extraction module is used for extracting features of the physical topological structure of the target power distribution network to obtain power distribution network topological feature data, and inputting the power distribution network topological feature data and the nonlinear variation feature data into a pre-constructed fault identification model to obtain fault probability values of all optical fiber links;

The identification module is used for comparing the fault probability value with a preset fault probability threshold value to obtain a potential fault link, and sequentially carrying out feature extraction and abrupt node identification on the potential fault link to obtain attenuation abrupt node data;

And the positioning module is used for processing the attenuation abrupt change node data according to a stress inversion method and determining fault point data in the target power distribution network based on the obtained stress value data.

Compared with the prior art, the embodiment of the invention has the beneficial effects that at least one of the following points is adopted:

1) According to the method, the nonlinear fault characteristics are accurately captured through the deep learning model and the multidimensional data are fused, so that the dependence of the traditional method on the linear attenuation characteristics is broken through, and the fault identification precision in a complex scene is remarkably improved. Specifically, the method utilizes the nonlinear attenuation model to deeply process the standardized difference data, effectively extracts nonlinear variation characteristics of optical signal power attenuation, combines the topological characteristic data of the power distribution network to construct a fault identification model, can accurately identify progressive faults which are easy to miss in traditional methods such as microbending, joint aging and the like, improves early fault detection rate by more than 40%, and reduces misjudgment rate in a multi-branch network through topological structure analysis;

2) The invention constructs a full-flow intelligent positioning closed loop of data acquisition-model reasoning-stress inversion, realizes automatic processing from fault probability evaluation to accurate positioning, and greatly improves fault positioning efficiency and engineering practicability. Specifically, the attenuation abrupt change node data is converted into specific stress values through a stress inversion method, the position of a fault point is accurately locked by combining a material stress threshold, the time-consuming defect of traditional OTDR piecewise detection is avoided, the positioning time is shortened by more than 60% in complex topologies such as a ring network and the like, the network topology change can be adapted in real time, direct guidance is provided for rapid repair of a power distribution communication network, and the operation and maintenance cost and the fault processing time are remarkably reduced.

Drawings

FIG. 1 is a flow chart of a deep learning based power distribution communication network fault location method in one embodiment of the present invention;

FIG. 2 is a schematic diagram of a deep learning based power distribution communication network fault location system in one embodiment of the present invention;

Reference numerals:

11, an acquisition module; 12, a calculation module, 13, a first extraction module, 14, a second extraction module, 15, an identification module, 16, and a positioning module.

Detailed Description

The following description of the embodiments of the present invention will be made clearly and fully with reference to the accompanying drawings, in which it is evident that the embodiments described are only some, but not all embodiments of the invention, and the purpose of these embodiments is to provide a more thorough and complete disclosure of the present invention. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.

In the description of the present invention, the terms "first," "second," "third," and the like are used for descriptive purposes only and are not to be construed as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Thus, a feature defining "a first", "a second", "a third", etc. may explicitly or implicitly include one or more such feature. In the description of the present invention, unless otherwise indicated, the meaning of "a plurality" is two or more.

In the description of the present invention, unless explicitly stated or limited otherwise, the terms "mounted," "connected," and "connected" are to be construed broadly, and may be, for example, fixedly connected, detachably connected, or integrally connected, mechanically connected, electrically connected, directly connected, indirectly connected via an intervening medium, or in communication between two elements. The terms "vertical," "horizontal," "left," "right," "upper," "lower," and the like are used herein for descriptive purposes only and not to indicate or imply that the apparatus or element being referred to must have a particular orientation, be constructed and operated in a particular orientation, and therefore should not be construed as limiting the invention. The term "and/or" as used herein includes any and all combinations of one or more of the associated listed items. The specific meaning of the above terms in the present invention will be understood in specific cases by those of ordinary skill in the art.

In the description of the present invention, it should be noted that all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art unless defined otherwise. The terminology used in the description of the present invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention, as the particular meaning of the terms described above in the present invention will be understood to those of ordinary skill in the art in the detailed description of the invention.

An embodiment of the present invention provides a power distribution communication network fault positioning method based on deep learning, and in particular, referring to fig. 1, fig. 1 is a schematic flow chart of a power distribution communication network fault positioning method based on deep learning in one embodiment of the present invention, which includes steps S1 to S6:

S1, acquiring optical power data of an optical fiber link in a target power distribution network, and calculating optical power difference values of adjacent sampling points in the optical power data to generate time sequence data;

Preferably, in an embodiment of the present invention, the obtaining optical power data of an optical fiber link in a target power distribution network, calculating an optical power difference value of adjacent sampling points in the optical power data, and generating time sequence data includes:

The optical power data refers to the energy intensity monitoring data of an optical signal in an optical fiber link, and generally reflects the signal transmission quality in units of decibel milliwatts (dBm), and is a core index for judging the attenuation state of the optical fiber. The time sequence data refers to a one-dimensional data sequence arranged in time sequence, in this embodiment, an ordered set of optical power differences of adjacent sampling points is used for describing a dynamic variation trend of optical power along with time. Preprocessing comprises noise filtering and outlier rejection on original optical power data, and data reliability is ensured, and common methods comprise filtering algorithms (such as median filtering and S-G filtering) and outlier detection. The preset optical power fluctuation threshold is a critical value set according to the power fluctuation range when the optical fiber link is in normal operation, and is used for identifying abnormal data points (such as sudden interference or power shock caused by equipment failure) which deviate from the normal state significantly.

It should be noted that, the original optical power data may be affected by environmental noise (such as electromagnetic interference and temperature fluctuation) and equipment errors, and direct use may lead to misjudgment of subsequent model analysis, and invalid data needs to be removed through preprocessing. In addition, the time sequence of the optical power difference value can reflect the dynamic change of the signal attenuation, which is the basis for identifying nonlinear attenuation (such as mutation and periodic fluctuation), and the abnormal data point can interfere with feature extraction, so that the effectiveness of the sequence is ensured through threshold value screening.

Specifically, in this embodiment, the optical power data of the optical fiber link is collected in real time by an Optical Time Domain Reflectometer (OTDR), an ultra-weak fiber grating array (uwFBG) and other devices, and a sampling time stamp (with accuracy reaching microsecond) and spatial position information are recorded synchronously.

And adopting median filtering (3 multiplied by 3 window) to carry out smoothing treatment on the original data, suppressing high-frequency noise and improving the signal-to-noise ratio to more than 20 dB. Calculating data mean value and standard deviation based on 3 sigma principle, marking the point with absolute value exceeding mean value + -3 sigma as abnormal, and combining with physical threshold (such as optical power mutation exceeding + -0.5 dB) dual screening to ensure abnormal data rejection rate above 99%.

And performing difference on power values of adjacent sampling points to obtain difference data reflecting instantaneous power change. And setting a preset fluctuation threshold (which can be dynamically adjusted according to historical data), removing abnormal data points with difference values exceeding the threshold, retaining continuous and stable non-abnormal data points, and finally generating time sequence data for subsequent analysis.

S2, fitting the time sequence data to obtain an actual power attenuation curve, and calculating an absolute difference value of the actual power attenuation curve and a theoretical power attenuation curve to obtain standardized difference data;

Preferably, in one embodiment of the present invention, before the fitting process is performed on the time series data, the method further includes:

Preferably, in one embodiment of the present invention, the calculating a mean square error between the theoretical stress characteristic data and the optical power trend data of the optical fiber link includes:

It should be noted that the optical power variation of the optical fiber link may be affected by various factors, including regular trend caused by normal stress (such as slow variation of ambient temperature) and abnormal abrupt change caused by failure. The overall trend (such as long-term attenuation or slight fluctuation) of the optical power is firstly extracted through linear regression, and compared with the expected trend (theoretical stress characteristic data) under the influence of theoretical stress, whether the trend in the current data accords with the normal physical rule can be judged. If the deviation (mean square error) of the two parameters exceeds a preset threshold, the actual trend is indicated to possibly contain abnormal changes of non-stress factors (such as poor contact and micro-bending of the optical fiber), more complex nonlinear characteristics are further excavated through fitting treatment, and interference of normal stress fluctuation on fault detection is avoided.

Specifically, in this embodiment, stress distribution data is generated according to parameters of the laying environment (such as a preset temperature variation range, an allowable tension value, etc.) of the optical fiber link, and then frequency characteristics (such as whether there is regular fluctuation caused by periodic temperature variation) of the stress data are analyzed through fourier transform, so as to obtain frequency domain characteristics (i.e., theoretical stress characteristic data) that should be possessed by optical power variation under the influence of theoretical stress.

And processing the time series data by using a least square method to find a straight line or a low-order curve (trend function) which can be best fit with the data, wherein the output value (trend function value) of the function is the optical power trend data and represents the long-term change trend of the data after short-term fluctuation is removed.

The theoretical stress characteristic data and the optical power trend data are in one-to-one correspondence (time alignment) with each other at each time point, the difference between the theoretical stress characteristic data and the optical power trend data is calculated, and the squares of all the differences are averaged (mean square error). If the error exceeds a preset threshold, it is indicated that the deviation between the actual trend and the normal trend under the influence of the theoretical stress is larger, and more complex fitting (such as polynomial fitting and nonlinear fitting) is needed to be performed on the time series data so as to capture possible fault-related abnormal features.

Preferably, in an embodiment of the present invention, the fitting the time series data to obtain an actual power attenuation curve, and calculating an absolute difference between the actual power attenuation curve and a theoretical power attenuation curve to obtain normalized difference data includes:

The cubic spline interpolation is a mathematical method for constructing a smooth curve from known discrete data points (here, differences in optical power in time series data). Cubic spline interpolation fits between every two adjacent data points using a cubic polynomial to ensure that the curve has continuous first and second derivatives throughout the interval, so that the resulting curve is both smooth and well approximated to the original data.

The actual power attenuation curve is a curve obtained by curve fitting time sequence data, and reflects the actual attenuation change condition of the optical power in the optical fiber link along with time. The theoretical power attenuation curve is a curve constructed according to physical principles and theoretical models according to physical parameters (such as materials, lengths, bending degrees and the like of optical fibers) of an optical fiber link, and represents attenuation changes of optical power with time under ideal and fault-free conditions.

It should be noted that, in the actual operation of the optical fiber link, the attenuation of the optical power may be affected by various factors, such as ambient temperature, external force extrusion, loosening of the joint, etc., which results in a difference between the actual attenuation situation and the theoretical situation. The actual power attenuation curve is obtained through fitting, and the actual power attenuation curve is compared with the theoretical power attenuation curve, so that the differences can be found out.

Although the absolute difference sequence reflects the degree of attenuation abnormality, it is inconvenient to directly analyze and compare because the numerical range may be different for different fiber links or different time periods. The normalized difference data obtained after normalization processing can measure attenuation abnormality under a unified scale, and a more reliable basis is provided for the follow-up judgment of whether the optical fiber link has faults and the severity of the faults.

Specifically, in the present embodiment, each data point in the time-series data is regarded as a discrete point on the time-optical power difference two-dimensional plane. A cubic polynomial is then determined between each two adjacent data points using cubic spline interpolation. The coefficients of this third order polynomial are calculated by satisfying certain conditions, such as the first derivative and the second derivative of the curve being continuous at the junction of adjacent intervals. Finally, combining the three polynomials of all adjacent intervals forms a smooth curve, i.e. the actual power decay curve. The actual power attenuation curve can be well fit to time series data, and shows the actual change trend of the optical power attenuation along with time.

Physical parameters of the fiber optic link are collected, including the type of fiber (e.g., single mode fiber, multimode fiber), length, loss factor (related to fiber material), bend radius, etc. Based on the physical principle and theoretical model of optical fiber transmission (such as the attenuation of optical fiber is proportional to length), a functional relationship is established with respect to time and optical power attenuation. By using the functional relation, theoretical attenuation values of the optical power at different time points are calculated, and the theoretical attenuation values are connected to obtain a theoretical power attenuation curve. The theoretical power decay curve represents the decay of optical power over time in an ideal situation without external disturbances and faults.

And subtracting the corresponding optical power attenuation values of the actual power attenuation curve and the theoretical power attenuation curve at each sampling point by taking the time axis as a reference. And taking absolute values of the subtraction results to form an absolute difference sequence. Each value in the absolute difference sequence reflects the magnitude of the difference between the actual attenuation at the sample point and the theoretical attenuation, the greater the difference, the more pronounced the attenuation anomaly at that point.

For each value in the sequence of differences, a normalization calculation is performed. All normalized values are assembled into normalized difference data. The standardized difference data can more intuitively reflect the attenuation abnormality degree of each sampling point, and is convenient for subsequent analysis and processing.

S3, inputting the standardized difference data into a pre-constructed nonlinear attenuation model to obtain nonlinear variation characteristic data reflecting the power attenuation of the optical signal in the optical fiber link;

The standardized difference data is the optical power attenuation abnormality degree data after pretreatment and normalization, eliminates the influence of dimension and link difference, has a numerical range of [0,1] or a range of 0 mean value and 1 standard deviation, and is core input data reflecting the degree of actual attenuation deviation from a theoretical model.

The nonlinear attenuation model is a neural network model based on deep learning, can capture nonlinear and non-stable change characteristics (such as mutation, exponential attenuation and periodic fluctuation) in optical power attenuation, is different from the fitting of a traditional linear model (such as a least square method) to linear trends, and is suitable for extracting characteristics of complex fault scenes.

The method aims at automatically extracting hidden fault characteristics in standardized difference data, avoiding one-sided performance of manual design characteristics, converting abstract numerical sequences into interpretable physical characteristics, and providing fault characterization information for subsequent topological structure analysis so as to improve positioning accuracy.

The method comprises the steps of converting standardized difference data into a format suitable for model input, constructing a nonlinear attenuation model, taking a CNN-LSTM combined model as an example, firstly extracting local features by using a CNN layer, then capturing long-term dependent features by using an LSTM layer, finally mapping the features into nonlinear variation feature vectors containing information such as mutation position probability, attenuation mode labels, feature confidence and the like by using a full connection layer, enhancing generalization capability by using a historical fault data set and manually injecting fault simulation data during model training, optimizing by adopting a multitask loss function combining cross entropy loss and mean square error, inputting real-time standardized difference data into a trained model during reasoning, and outputting nonlinear variation feature data such as 'detecting a certain type of mutation at a certain position and having a certain confidence'.

S4, extracting features of a physical topological structure of the target power distribution network to obtain power distribution network topological feature data, and inputting the power distribution network topological feature data and the nonlinear variation feature data into a pre-constructed fault identification model to obtain fault probability values of all optical fiber links;

preferably, in an embodiment of the present invention, the feature extracting the physical topology structure of the target power distribution network to obtain power distribution network topology feature data includes:

The physical topology of the target power distribution network refers to the actual connection and layout between the various electrical devices (e.g., transformers, switches, lines, etc.) in the target power distribution network, which describes the paths and structures of the electrical energy transmission in the network.

The node adjacency matrix is a two-dimensional matrix used for representing the connection relation among all nodes in the power distribution network. The rows and columns of the matrix respectively correspond to nodes in the power distribution network, if two nodes are directly connected, the value of the corresponding position in the matrix is 1, otherwise, the value of the corresponding position in the matrix is 0. The link attribute table records related attribute information of each link (such as a power transmission line) in the power distribution network, such as the length, capacity, resistance and the like of the link.

The node degree vector is obtained by summing each row of the node adjacency matrix, each element in the vector represents the connection quantity of the corresponding node and other nodes, the importance degree of the node in the power distribution network is reflected, and the more the connection quantity is, the more important the node is. The link weight data is calculated according to the information in the link attribute table and is used for measuring the importance or transmission capacity and other values of each link in the power distribution network. The link weight matrix is a matrix constructed from link weight data that further describes the relative importance between links in the distribution network.

It should be noted that, it is not comprehensive enough to determine whether the optical fiber link fails by only relying on the nonlinear variation characteristic data of the optical signal power attenuation. The physical topology structure of the power distribution network has important influence on the occurrence and the propagation of faults, the importance of nodes and links at different positions in the network is different, and the influence on the whole power distribution network is also different when the faults occur. By extracting the topological characteristic data of the power distribution network and combining the topological characteristic data with the nonlinear change characteristic data to input a fault identification model, the fault probability of each optical fiber link can be more accurately judged by comprehensively considering the topological structure and the optical signal attenuation condition, and the accuracy and the reliability of fault identification are improved.

Specifically, in this embodiment, first, the physical connection relationship of the optical fiber link in the target power distribution network is obtained by referring to the design drawing of the power distribution network, the equipment account, and other data. Then, a node adjacency matrix is constructed from these connection relations. Assuming n nodes, then the node adjacency matrix is an n x n matrix. If there is a direct connection between node i and node j, the element values of the ith row, the jth column and the jth row, the ith column in the matrix are 1, and if there is no connection, they are 0. Meanwhile, relevant attribute information of the optical fiber link, such as the length, transmission capacity, resistance and the like of the link, is arranged and recorded in a link attribute table.

Summing is performed for each row of the node adjacency matrix. For example, for the ith row, all elements of that row are added, and the resulting sum is the degree of node i. The degrees of all nodes are arranged in sequence to form a vector, namely a node degree vector. The vector can intuitively reflect the importance degree of each node in the power distribution network, and the larger the degree is, the more nodes are connected with the node, and the more critical the node is in the network.

And selecting a proper method to calculate the weight of each link according to the information in the link attribute table. For example, the weight may be calculated comprehensively according to factors such as the length of the link, the transmission capacity, etc., and the link with a shorter length and a larger transmission capacity may have a higher weight. And constructing the calculated link weights into a matrix according to the connection relation among the nodes, namely a link weight matrix. The elements of row i and column j in the matrix represent the weights of the links between node i and node j.

And splicing the node degree vector and the link weight matrix. The node degree vector may be added to the link weight matrix as a new column or other suitable stitching means may be used. The spliced data set is the topological characteristic data of the power distribution network, and integrates the information such as the importance degree of the nodes and the relative importance of the links.

S5, comparing the fault probability value with a preset fault probability threshold value to obtain a potential fault link, and sequentially carrying out feature extraction and mutation node identification on the potential fault link to obtain attenuation mutation node data;

preferably, in an embodiment of the present invention, the comparing the fault probability value with a preset fault probability threshold to obtain a potential fault link, and sequentially performing feature extraction and abrupt node identification on the potential fault link to obtain attenuation abrupt node data includes:

The fault probability value is a probability value (between 0 and 1) of each optical fiber link fault calculated through the fault identification model, and the higher the value is, the greater the fault probability is. The threshold value of the fault probability is preset, and a threshold value (e.g. 0.7) which is set manually and used for judging whether the link is likely to be faulty is used for screening the high-risk links, so that indiscriminate analysis on all the links is avoided.

The potential fault link is an optical fiber link with the fault probability value exceeding a preset threshold value, and is a subsequent key investigation object. The deep residual network (ResNet) is a deep learning model, solves the problem of deep network gradient disappearance through residual connection, and is good at extracting multi-level abstract features (such as short-term abrupt change and long-term trend) of data. The attenuation abrupt change node data comprises information such as node numbers, attenuation abnormality degrees, spatial positions and the like, and is used for identifying specific nodes with obvious attenuation changes in the optical fiber links.

The method is characterized in that the step of filtering the low-probability links rapidly through threshold comparison concentrates the computing resources on the potential fault links with high fault probability, avoids 'sea fishing needle' type investigation and improves efficiency. Meanwhile, the fault range is reduced from potential links to specific nodes, direct clues are provided for the follow-up accurate positioning of fault points, and progressive investigation logic of probability screening, feature positioning and node identification is formed.

Specifically, in this embodiment, each Link failure probability value (e.g., [0.65,0.82,0.51,. ]) output by the failure recognition model is compared with a preset threshold (e.g., 0.7), and a Link with a probability greater than or equal to the threshold (e.g., 0.82 greater than or equal to 0.7 of the second Link) is reserved, so as to generate a potential failure Link list (e.g., [ "link_03", "link_17"). The list contains basic information such as a Link number, an area to which the Link belongs, a history fault record and the like, and is used for rapidly positioning an investigation object, for example, "Link_03" for preferentially processing the history fault frequency.

The optical power data (including time stamp and power value) of each potential fault link are arranged according to the dimension of time-power to form a two-dimensional matrix (row number=sampling point number, column number=1), for example, 1000 sampling points form a 1000×1 matrix, and the trend of power change along with time is intuitively displayed. And splicing the two-dimensional matrix with standardized difference data (reflecting the degree of abnormality) and nonlinear change characteristics (such as mutation position probability and fault mode labels) in the channel dimension to form a fusion characteristic tensor (such as dimension 1000 multiplied by 3), so that the model can analyze the original signal, the degree of abnormality and the fault mode simultaneously.

The depth residual network of the invention adopts 34 layers ResNet and comprises a plurality of residual blocks, and each residual block extracts multi-layer characteristics through convolution-batch normalization-ReLU-residual connection:

shallow, capture short term power fluctuations (e.g., abrupt changes within 1 second);

deep layer learning of long-term decay trend (e.g. decay acceleration lasting 1 hour).

Input and output, namely inputting fusion characteristic tensor, outputting attenuation mutation characteristic vector (such as information including mutation occurrence time point, abnormal energy duty ratio and the like), and identifying attenuation mutation energy duty ratio reaching 85% at 500 th time point (corresponding to 1500 m of link).

And (3) calculating the similarity among the nodes by adopting cosine similarity for the feature vector output by the residual error network to form a similarity matrix (the closer the numerical value is to 1, the more similar the node attenuation trend is). The similarity matrix is converted into a graph structure, a main feature vector is extracted through feature decomposition, the nodes are divided into categories such as normal attenuation, gradual change, mutation and the like through K-means clustering, mutation category nodes are screened out, attenuation mutation Node data (including Node ID (node_24), mutation occurrence distance (1520 m) and degree of abnormality (high) are generated.

And S6, processing the attenuation abrupt change node data according to a stress inversion method, and determining fault point data in the target power distribution network based on the obtained stress value data.

Preferably, in an embodiment of the present invention, the processing the attenuation abrupt node data according to a stress inversion method, determining fault point data in the target power distribution network based on the obtained stress value data includes:

Preferably, in one embodiment of the present invention, the determining, based on the high stress abrupt node list, fault point data in the target power distribution network includes:

The stress inversion method is a method for reversely deducing the stress condition of the optical fiber through the known information such as power attenuation of the optical signal in the optical fiber link. Because there is a relationship between the optical signal attenuation of an optical fiber and the stress to which it is subjected, this relationship can be used to infer stress.

The stress sensitivity parameters are parameters reflecting the stress sensitivity degree of the optical fiber, and the optical fibers with different materials and structures have different stress sensitivity parameters, which are important basis for constructing a stress attenuation relation model. The material stress threshold is the maximum stress value that the fiber material can withstand, and when the fiber is stressed beyond this threshold, the fiber may fail, such as break, bend, etc. The stress attenuation relation model is a model describing the mathematical relation between the stress born by the optical fiber and the power attenuation of the optical signal, and corresponding stress values can be calculated according to the power attenuation by the model.

The power attenuation is the reduction of the power of the optical signal in the optical fiber link in the transmission process, and is key input data for calculating the stress value. The high stress abrupt node list is a list of nodes with stress values exceeding the stress threshold of the material, which are important objects of interest in which faults may exist. The relative distance of the fault point is the distance between the high stress abrupt change node and the adjacent node, and is used for further determining the position of the fault point.

It should be noted that the foregoing steps obtain attenuation mutation node data, but these data only indicate nodes where an abnormality may exist, and the location of the fault point cannot be precisely determined. The stress of the optical fiber is closely related to the occurrence of faults, and the power attenuation information can be converted into stress information by a stress inversion method, so that fault points can be positioned more accurately. The stress value is compared with the material stress threshold value, and the high stress abrupt change node is screened out, so that the fault investigation range can be reduced, and the node most likely to have faults is focused. And then, the coordinate data of the nodes and the relative distance of the fault points are combined to carry out multi-condition screening, so that the positioning accuracy of the fault points can be further improved.

Specifically, in this embodiment, information such as stress sensitivity parameters, material stress threshold values, and physical length of the link of the optical fiber is first obtained. Such information may be obtained by referring to the product specifications of the optical fiber, performing experimental measurements, and the like. Then, a stress attenuation relation model is constructed from the information. The model may be built on the basis of physical principles, such as deriving a functional relationship between stress and power attenuation from the mechanical properties of the fiber material and the optical transmission principle. A common possibility is a linear or nonlinear functional model describing the correspondence between stress values and power attenuation.

And extracting the power attenuation amount corresponding to each node from the attenuation abrupt node data, and simultaneously acquiring the physical length of the link where the node is located. Substituting the extracted power attenuation amount and physical length into a stress attenuation relation model, and obtaining a stress value corresponding to each node through calculation of the model. The stress values of all nodes are collected to form a stress value data set.

Each stress value in the stress value dataset is compared to a material stress threshold. If the stress value of a certain node exceeds the material stress threshold, the node is stressed too much, and a fault can exist. And screening out all nodes with stress values exceeding the stress threshold value of the material, and sorting the related information of the nodes into a list, namely a high stress abrupt change node list. The nodes in this list are the objects that need to be focused on.

And constructing a candidate fault point set based on each node in the high-stress abrupt node list and adjacent nodes thereof. The neighboring nodes may be determined based on the topology of the fiber links, typically nodes directly connected to the high stress abrupt nodes. This may further narrow down the troubleshooting range to these candidate nodes.

Coordinate data of each node in the candidate fault point set is acquired, and the coordinate data can be acquired through a Geographic Information System (GIS) and the like. And then, calculating the relative distance between each high-stress abrupt change node and the fault point of the adjacent node according to the coordinate data. This distance information may help to further determine the possible location of the fault point.

And performing multi-condition screening according to the obtained stress value of each high-stress abrupt change node and the obtained relative distance between fault points. For example, a node having a larger stress value and a closer distance from an adjacent node may be prioritized to be a more likely failure point. By the multi-condition screening mode, fault point data including information such as specific positions of fault points and related stress values are finally determined, and accurate basis is provided for subsequent fault repair.

Through the steps, the stress inversion method is combined with multi-condition screening, fault point data in the target power distribution network can be accurately determined from the attenuation abrupt change node data, and the efficiency of fault investigation and repair is improved.

An embodiment of the present invention provides a power distribution communication network fault location system based on deep learning, specifically, referring to fig. 2, fig. 2 is a schematic flow diagram of the power distribution communication network fault location system based on deep learning in one embodiment of the present invention, which includes:

The acquisition module 11 is configured to acquire optical power data of an optical fiber link in a target power distribution network, calculate an optical power difference value of adjacent sampling points in the optical power data, and generate time sequence data;

the calculating module 12 is configured to perform fitting processing on the time-series data to obtain an actual power attenuation curve, and calculate an absolute difference value between the actual power attenuation curve and a theoretical power attenuation curve to obtain standardized difference data;

The first extraction module 13 is configured to input the normalized difference data into a pre-constructed nonlinear attenuation model, so as to obtain nonlinear variation characteristic data reflecting power attenuation of an optical signal in the optical fiber link;

the second extracting module 14 is configured to perform feature extraction on the physical topology structure of the target power distribution network to obtain power distribution network topology feature data, and input the power distribution network topology feature data and the nonlinear variation feature data into a pre-constructed fault identification model to obtain a fault probability value of each optical fiber link;

The identifying module 15 is configured to compare the fault probability value with a preset fault probability threshold value to obtain a potential fault link, and sequentially perform feature extraction and abrupt node identification on the potential fault link to obtain attenuation abrupt node data;

And the positioning module 16 is used for processing the attenuation abrupt node data according to a stress inversion method and determining fault point data in the target power distribution network based on the obtained stress value data.

The foregoing examples illustrate only a few embodiments of the invention and are described in detail herein without thereby limiting the scope of the invention. It should be noted that it will be apparent to those skilled in the art that several variations and modifications can be made without departing from the spirit of the invention, which are all within the scope of the invention. Accordingly, the scope of protection of the present invention is to be determined by the appended claims.

Claims

1. A method for locating faults in a power distribution communication network based on deep learning, comprising:

Obtaining optical power data of optical fiber links in a target distribution network, calculating optical power differences between adjacent sampling points in the optical power data, and generating time series data;

Performing fitting processing on the time series data to obtain an actual power attenuation curve, and calculating the absolute difference between the actual power attenuation curve and the theoretical power attenuation curve to obtain standardized difference data;

Inputting the standardized difference data into a pre-built nonlinear attenuation model to obtain nonlinear variation characteristic data reflecting the attenuation of optical signal power in the optical fiber link;

Extracting features of the physical topology of the target distribution network to obtain distribution network topology feature data, inputting the distribution network topology feature data and the nonlinear change feature data into a pre-built fault identification model to obtain a failure probability value of each optical fiber link;

Comparing the fault probability value with a preset fault probability threshold to obtain a potential fault link, performing feature extraction and mutation node identification on the potential fault link in sequence to obtain attenuated mutation node data;

Processing the attenuation mutation node data according to a stress inversion method, and determining the fault point data in the target distribution network based on the obtained stress value data;

The processing of the attenuation mutation node data according to the stress inversion method and determining the fault point data in the target distribution network based on the obtained stress value data includes:

Constructing a stress attenuation relationship model based on the acquired stress-sensitive parameters, material stress threshold, and physical length of the optical fiber link;

Extracting the power attenuation and the physical length of the link corresponding to each attenuation mutation node data, inputting the power attenuation and the physical length into the stress attenuation relationship model to obtain a stress value data set;

Comparing each stress value data in the stress value data set with the material stress threshold, and screening the comparison results to obtain a list of high stress mutation nodes;

Determining fault point data in the target distribution network based on the high stress mutation node list includes:

Construct a candidate fault point set based on each node and its adjacent nodes in the high stress mutation node list;

Obtaining coordinate data of each node in the candidate fault point set, and calculating the relative distance between each high stress mutation node and its adjacent nodes based on the coordinate data;

A multi-condition screening is performed based on the acquired stress value of each high stress mutation node and the relative distance to the fault point, and the fault point data is obtained based on the result of the multi-condition screening.

2. The method for locating a power distribution communication network fault based on deep learning according to claim 1, wherein the step of obtaining optical power data of an optical fiber link in a target distribution network, calculating the optical power difference between adjacent sampling points in the optical power data, and generating time series data comprises:

Collecting raw optical power data of optical fiber links in the target distribution network in real time and preprocessing the raw optical power data;

Time-sorting the pre-processed raw optical power data to obtain an optical power sequence data set;

Calculating the optical power difference between adjacent sampling points in the optical power sequence data set to obtain difference data; wherein if the optical power difference exceeds a preset optical power fluctuation threshold, it is marked as an abnormal data point;

Non-abnormal data points in the difference data are screened to obtain time series data.

3. The method for locating faults in a power distribution communication network based on deep learning according to claim 1, characterized in that before performing fitting processing on the time series data, the method further comprises:

Performing linear regression processing on the time series data to obtain optical power trend data, and calculating the mean square error between the theoretical stress characteristic data of the optical fiber link and the optical power trend data;

If the mean square error exceeds a preset error threshold, fitting processing is performed on the time series data.

4. The power distribution communication network fault location method based on deep learning according to claim 3, wherein the step of calculating the mean square error between the theoretical stress characteristic data of the optical fiber link and the optical power trend data comprises:

generating stress distribution data according to the obtained stress distribution parameters of the optical fiber link, and performing frequency domain feature extraction on the stress distribution data using Fourier transform to obtain the theoretical stress characteristic data;

Performing linear regression processing on the time series data using the least squares method to obtain an optical power trend function, and obtaining a trend function value at each sampling time point based on the optical power trend function to form the optical power trend data;

The theoretical stress characteristic data and the optical power trend data are time-aligned, and the mean square error of the time-aligned theoretical stress characteristic data and the optical power trend data is calculated.

5. The method for locating a power distribution communication network fault based on deep learning according to claim 1, wherein the fitting process of the time series data to obtain an actual power attenuation curve and the calculation of the absolute difference between the actual power attenuation curve and the theoretical power attenuation curve to obtain standardized difference data comprises:

The time series data is subjected to curve fitting based on the cubic spline interpolation method to obtain the actual power attenuation curve;

Constructing a theoretical power attenuation curve based on the physical parameters of the optical fiber link;

Calculating the absolute difference of each sampling point on the time axis of the actual power attenuation curve and the theoretical power attenuation curve to obtain an absolute difference sequence reflecting the degree of attenuation abnormality;

The absolute difference sequence is normalized to obtain the standardized difference data.

6. The method for locating a distribution communication network fault based on deep learning according to claim 1, wherein extracting features of the physical topology of the target distribution network to obtain distribution network topology feature data comprises:

Based on the obtained physical connection relationship of the optical fiber link, constructing a node adjacency matrix and a link attribute table;

Summing each row of the node adjacency matrix to obtain a node degree vector reflecting the importance of the node;

The node degree vector and the link weight matrix are concatenated to obtain the distribution network topology characteristic data.

7. The method for locating a distribution communication network fault based on deep learning according to claim 1, wherein the comparing the fault probability value with a preset fault probability threshold to obtain a potential fault link, and sequentially performing feature extraction and mutation node identification on the potential fault link to obtain attenuation mutation node data, comprises:

Comparing the failure probability value with a preset failure probability threshold, screening out links whose failure probability values are greater than or equal to the preset failure probability threshold, and generating a list of potential failure links;

Acquire optical power sequence data of links corresponding to the faulty link list, and convert the optical power sequence data into a two-dimensional matrix;

The two-dimensional matrix is spliced with the standardized difference data and the nonlinear change feature data to obtain a fused feature tensor; and the fused feature tensor is input into a pre-built deep residual network for feature extraction to obtain attenuation mutation feature data;

Cluster analysis is performed on the attenuation mutation characteristic data to obtain the attenuation mutation node data where the attenuation trend has changed.

8. A power distribution communication network fault location system based on deep learning, characterized by applying the power distribution communication network fault location method based on deep learning according to claim 1, comprising:

An acquisition module is used to obtain optical power data of optical fiber links in the target distribution network, calculate the optical power difference between adjacent sampling points in the optical power data, and generate time series data;

a calculation module, configured to perform fitting processing on the time series data to obtain an actual power attenuation curve, and calculate an absolute difference between the actual power attenuation curve and a theoretical power attenuation curve to obtain standardized difference data;

a first extraction module, configured to input the standardized difference data into a pre-built nonlinear attenuation model to obtain nonlinear variation characteristic data reflecting the attenuation of optical signal power in the optical fiber link;

a second extraction module, configured to extract features of the physical topology of the target distribution network to obtain distribution network topology feature data, input the distribution network topology feature data and the nonlinear change feature data into a pre-built fault identification model to obtain a failure probability value of each optical fiber link;

an identification module, configured to compare the fault probability value with a preset fault probability threshold to obtain a potential fault link, and sequentially perform feature extraction and mutation node identification on the potential fault link to obtain attenuation mutation node data;

A positioning module is used to process the attenuation mutation node data according to a stress inversion method, and determine the fault point data in the target distribution network based on the obtained stress value data.