A distributed new energy power distribution network fault diagnosis system

Abstract

The application relates to the technical field of distributed new energy distribution network fault diagnosis, and discloses a distributed new energy distribution network fault diagnosis system which comprises a data integration module, a sorting and classifying module, an information fusion diagnosis module and an intelligent terminal. The system uses a waveform data feature extraction module to analyze the data distribution characteristics by using a time domain, a frequency domain and a sample entropy feature extraction method, realizes preliminary screening of a fault diagnosis model, outputs a low-dimensional fault feature set, and inputs the low-dimensional fault feature set into a sample training module and a sorting module, realizes training and sorting of fault data, outputs a diagnosis result, further combines protection action conditions and circuit breaker switch position information through the information fusion diagnosis module, sorts suspicious fault elements, carries out information fusion diagnosis, and outputs the diagnosis result to the intelligent terminal. The system realizes automatic fault diagnosis, analysis means and rapid active decision of the distributed new energy distribution network, and improves the adaptability of a fault diagnosis model algorithm to complex characteristics of the distribution network.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of distributed new energy power distribution network fault diagnosis, and particularly relates to a distributed new energy power distribution network fault diagnosis system. BACKGROUND

[0002] In recent years, with high proportion of new energy and high proportion of power electronic equipment connected to the power grid, the randomness, volatility, intermittency and heterogeneous characteristics of power grid faults are increasingly obvious, which comprehensively and systematically tests the links of dispatching decision support, the algorithm iteration and support means of the traditional fault identification model are difficult to continue, and the efficiency and quality of dispatching decision are easily affected. First, the information relied on by dispatching decision has the characteristics of mass, many points, large area and rush, and the judgment of real power grid fault relies on the experience and response ability of dispatchers to a high degree, and the intelligentization and automation development of fault identification field is seriously lagging behind. Secondly, the existing power distribution network fault identification model algorithm is extremely dependent on the primary system state information, and the anti-interference degree is poor. In recent years, the development trend of secondary system evidence information is increasingly similar to that of the primary system, and the timeliness and independence are poor, and the quality level of fault identification directly determines the level of dispatching decision. Finally, the algorithm updating and iteration algorithm of the existing fault identification model has obvious predictability and periodicity characteristics, and cannot well adapt to the dispatching decision demand in the background of new energy occupying a dominant position in power supply structure, intermittent, volatile output and complex, variable and other unstable situations of new power system, and optimization is urgently needed.

[0003] The main contradiction of fault identification and fault arrangement of the distributed new energy power distribution network dispatching level is concentrated in the dispatching decision support system which has been developed in the past in the aspects of data acquisition and visual display, ignoring data mining and fault arrangement, and too much information also causes unnecessary interference, which puts forward strict requirements on the fault identification and disposal experience of dispatching management personnel. The technical principle of the existing dispatching decision system is built on the basis of the fault identification system of impedance method analysis, and the technical means is the unified modeling of fixed value parameters and equipment account, and such expected strong model framework cannot cope with the polymorphism, heterogeneity and unmeasurable fault characteristics of new energy as the main body. SUMMARY

[0004] (I) Technical problems to be solved

[0005] In view of the defects of the prior art, the application provides a distributed new energy power distribution network fault diagnosis system, which has the advantages of improving the adaptability of the fault diagnosis model algorithm to the complex characteristics of the power distribution network, and solves the above technical problems.

[0006] (II) Technical scheme

[0007] To achieve the above object, the application provides the following technical scheme: a distributed new energy power distribution network fault diagnosis system, comprising a data integration module, a data cleaning module, a data standardization processing module, a waveform data feature extraction module, a sample training module, a sorting and classification module, an information fusion diagnosis module and an intelligent terminal;

[0008] The data integration module acquires recording wave master station data and protection master station data, is used for acquiring the reliability of distributed new energy power distribution network fault information, and performs weighted fusion calculation;

[0009] The data cleaning module cleans invalid and redundant data generated in data generation, data collection and data transmission, the data standardization processing module performs data processing and converts into standard data;

[0010] The waveform data feature extraction module is used for processing information before, during and after the occurrence of new energy power distribution network faults, screening out effective data, extracting fault data features and transmitting to the sample training module to generate a fault sample library;

[0011] The sample training module acquires the processed fault data of the recording wave master station and the protection master station, is used for establishing a fault diagnosis sample set, and compares and analyzes the fault data features in the fault sample library, after processing, performs fault feature sample training, and outputs the pre-judgment result to the sorting and classification module;

[0012] The sorting and classification module is used for sorting fault data and transmitting to the information fusion diagnosis module;

[0013] The signal fusion diagnosis module realizes multi-element heterogeneous data standardization and fault recognition after filtering, and outputs the fault diagnosis discrimination result after recognition to the intelligent terminal;

[0014] The intelligent terminal is used for giving a decision scheme of fault discrimination.

[0015] As a preferred technical scheme of the application, the recording wave master station data comprises local fault recording waveform features and opposite side fault recording waveform features, and the protection master station data comprises protection setting value, protection action signal and switch position signal;

[0016] The data integration module has the following specific working process:

[0017] A1, the local fault recording waveform features and the opposite side fault recording waveform features in the recording wave master station data are subjected to fault feature preliminary judgment and fault reliability evaluation, and whether the features of the two sides of the fault information are unified is confirmed;

[0018] A2, obtaining the protection setting value, protection action signal and switch position signal from the protection master station, sorting the suspicious fault element set, and calculating the fault credibility index of each element in the suspicious fault element set to obtain the fault characterization probability of each element;

[0019] A3, calculating the fault credibility index E of the distributed new energy distribution network, and the specific expression is as follows:

[0020] E=P(E L，t <E0)

[0021]

[0022] Wherein, E is the fault credibility index, P(E L，t <E0) represents the occurrence probability of E L，t <E0, E L，t is the fault credibility index of the system fault loss load ratio L at time t, E0 represents the credibility index of the limit state 0 point, Q r，t , Q n，t respectively represent the fault load and normal operation load of the system at time t.

[0023] As a preferred technical scheme of the application, the data cleaning module performs quality detection on the data output by the data integration module, discriminates outlier data, fills in missing data and smooths noise data, adopts a group intelligent optimization association rule algorithm for data cleaning, and calculates the accuracy rate of the distributed new energy distribution network fault data after cleaning The specific expression is as follows:

[0024]

[0025] Wherein, TP represents that the true value is positive class, TN represents that the true value is negative class, and M is the total number of sample data, when the accuracy rate of the cleaned data is The distributed new energy distribution network fault data is output to the data standardization processing module, when the accuracy rate of the cleaned data is lower than , and is filtered out after reconfirmation.

[0026] As a preferred technical scheme of the application, the data standardization processing module includes a data fault tolerance processing unit and a data standardization conversion unit, the data fault tolerance processing unit is used for processing the case that there is missing or redundancy in the data, and specifically includes the following steps:

[0027] B1, scanning the fault file to make an initial judgment;

[0028] B2, if there is missing information attribute, the attribute list containing attribute type and compensation information is inquired, and the missing attribute is supplemented;

[0029] B3, if there is data missing, using cubic spline interpolation or linearly calculated mean value to fill in;

[0030] B4, the redundant data is deleted;

[0031] The data standardization conversion unit is used for processing the data processed by the data fault tolerance processing unit to standardize, and the specific calculation expression is as follows:

[0032]

[0033] Wherein, represents the input zth real vector function in dimension, C represents the number of fault classes, e represents the natural constant, represents the summation of C data, represents the zth real vector in dimension.

[0034] As a preferred technical scheme of the present application, the waveform data feature extraction module judges the fault type and fault property of the fault recording main station data and the signal-keeping main station data processed by the standardization processing module: respectively extracts the time domain feature of the waveform data signal of normal operation, intra-zone fault and extra-zone fault frequency domain feature and sample entropy feature SE to construct a distributed new energy distribution network multi-domain fault feature set, and the distributed new energy distribution network multi-domain fault feature set is mapped to a low-dimensional fault feature set;

[0035] The time domain feature Through the instantaneous frequency and amplitude changing with time by empirical mode decomposition, ten time domain features are extracted, including peak value, peak-peak value, root mean square, absolute mean value, square root amplitude, variance, skewness, kurtosis, waveform index and margin index, to form a time domain feature vector

[0036] The frequency domain feature Three time domain features are extracted, including mean square frequency, frequency variance and peak frequency, to form a three-dimensional frequency domain fault feature vector

[0037] The sample entropy feature SE extracts the first four IMF components as effective components by VMD decomposition of the distributed new energy distribution network fault signal, and extracts the sample entropy to construct a four-dimensional sample entropy feature vector:

[0038] SE = [SampEn1, SampEn2, SampEn3, SampEn4]

[0039] Where SampEn1, SampEn2, SampEn3, and SampEn4 represent the four-dimensional sample entropy.

[0040] As a preferred embodiment of the present invention, the sorting and classification module uses the K-means clustering algorithm on the pre-diagnosis results output by the sample training module to complete the sorting and classification of fault data in the distributed new energy distribution network. The specific steps are as follows:

[0041] C1. Randomly select k initial cluster centers from the fault dataset used for sample training.

[0042] C2. Calculate the cluster centers of the remaining data objects. Using Euclidean distance, find the cluster centers that are closest to the target data object. Simultaneously, waveform data objects are assigned to cluster centers. The specific calculation expression for the corresponding cluster is as follows:

[0043]

[0044] in, Representing μ and The distance between the similarities, For the first There are 1 cluster center, where μ represents a data object. Represents the dimensions of a data object. For μ and The Each attribute value;

[0045] C3. Calculate the average value of the data objects in each cluster again as the new cluster center, and perform the next iteration until the cluster centers no longer change.

[0046] Compared with the prior art, the present invention provides a distributed new energy distribution network fault diagnosis system, which has the following beneficial effects:

[0047] This invention analyzes data distribution characteristics using time-domain, frequency-domain, and sample entropy feature extraction methods through a waveform data feature extraction module. This enables preliminary screening of the fault diagnosis model, outputting a low-dimensional fault feature set, which is then input into a sample training and sorting module to train and sort fault data, outputting diagnostic results. The information fusion diagnosis module further combines protection action information and circuit breaker switch change information to identify suspicious faulty components for information fusion diagnosis, outputting the diagnostic results to a smart terminal. This provides a means for fault diagnosis and analysis, as well as rapid and proactive decision-making in the automation of distributed new energy distribution networks, improving the adaptability of the fault diagnosis model algorithm to the complex characteristics of distribution networks. Attached Figure Description

[0048] Figure 1 Structure diagram of the present application;

[0049] Figure 2 Standard data file conversion diagram of the present application;

[0050] Figure 3 Distributed power distribution network fault multi-domain feature extraction diagram of the present application;

[0051] Figure 4 Fault data enhancement process diagram of the present application. DETAILED DESCRIPTION

[0052] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work fall within the scope of protection of the present application.

[0053] Please refer to Figures 1-4 A distributed new energy power distribution network fault diagnosis system, comprising a data integration module, a data cleaning module, a data standardization processing module, a waveform data feature extraction module, a sample training module, a sorting and classification module, an information fusion diagnosis module and an intelligent terminal.

[0054] The data integration module obtains the recording wave master station data and the protection signal master station data, and is used to obtain the reliability of the distributed new energy power distribution network fault information. The formula for weighted fusion calculation is F=W·S, wherein F is the fusion result, W is the weight, and S is the data source. Different weights are allocated to different data sources to integrate the data of multiple data sources, thereby improving the accuracy and reliability of data fusion.

[0055] The recording wave master station data includes the self-side fault recording waveform feature and the opposite-side fault recording waveform feature. The protection signal master station data includes the protection setting value, the protection action signal and the switch position signal.

[0056] The specific working process of the data integration module is as follows:

[0057] A1, the self-side fault recording waveform feature and the opposite-side fault recording waveform feature in the recording wave master station data are subjected to fault feature preliminary judgment and fault reliability evaluation, and whether the features of the two-side fault information are unified is confirmed.

[0058] A2, the protection setting value, the protection action signal and the switch position signal are obtained from the protection signal master station, the suspicious fault element set is sorted, the fault reliability index of each element in the suspicious fault element set is calculated one by one, and the fault representation probability of each element is obtained.

[0059] A3, calculate the fault credibility index E of the distributed new energy distribution network, and the specific expression is as follows:

[0060] E = P (E L，t <E0)

[0061]

[0062] Wherein, E is the fault credibility index, P (E L，t <E0) indicates the occurrence probability of E L，t <E0, E L，t is the fault credibility index of the system fault loss load ratio L at time t, E0 indicates the credibility index of the limit state 0 point, Q r，t , Q n，t respectively indicate the fault load and normal operation load of the system at time t, wherein the higher the fault credibility index E, the more real the fault waveform data feature extraction, and vice versa, the lower the fault credibility index E, the more serious the fault waveform data feature extraction distortion, so that each suspicious fault element with low credibility index should be weighted fault diagnosis and transmitted to the data cleaning module;

[0063] The data cleaning module cleans the invalid and redundant data generated in the data generation, collection and data transmission. The data standardization processing module processes the data and converts it into standard data.

[0064] The data cleaning module detects the quality of the data output by the data integration module, identifies the outlier data, fills in the missing data and smooths the noise data, eliminates invalid data, fills in valuable data, uses group intelligence optimization association rule algorithm for data cleaning, and mines the association rules between data fault information feature index and distributed new energy distribution network fault. Remove invalid and redundant data generated in the data generation, collection and data transmission process. Calculate the accuracy of the distributed new energy distribution network fault data after cleaning The specific expression is as follows:

[0065]

[0066] Wherein, TP indicates that the true value is positive class, TN indicates that the true value is negative class, and M is the total number of sample data. When the accuracy of the cleaned data is The distributed new energy distribution network fault data is output to the data standardization processing module. When the accuracy of the cleaned data is lower than , it is filtered out after reconfirmation.

[0067] The data standardization processing module comprises a data fault-tolerant processing unit and a data standardization conversion unit.

[0068] B1, scanning the fault file to make an initial judgment;

[0069] B2, if there is missing information attribute, querying an attribute list comprising attribute types and compensation information, and supplementing the missing attribute;

[0070] B3, if there is data missing, using a cubic spline interpolation or a linear average value seeking method to fill in;

[0071] Let S(x) be a cubic spline interpolation function, and x represent an interpolation node, then the data function S(X) needs to satisfy: on each interval [xj, xj+1]j=0,1,...,n, x is given (n+1) interpolation nodes on a specific data interval; S(x) has continuous 2-order derivative on a certain data interval; S(xj)=yj, j=0,1,2,...,n, y is the value of the corresponding data function on a specific data interval, by cubic spline interpolation of the missing data, the interpolation mean error can reach ±1.78mm. For the case of less data missing, the linear average value seeking method is used to fill in; + 1] on each interval [xj, xj

[0072] B4, deleting the redundant data;

[0073] The data standardization conversion unit receives the data processed by the data fault-tolerant processing unit, carries out multi-source data conversion into a unified standard for data inconsistency from multiple manufacturers or systems, and carries out numerical value processing on non-numerical information to convert it into dimensionless pure numerical value, which is convenient for unified analysis and measurement. For non-standard data, granularity conversion is needed, which can be divided into coarse granularity and fine granularity, so that it falls into a small specific interval, although some details are discarded, but the granularity of the data is more meaningful and easier to obtain effective features. The classification method of fixed distance or fixed ratio is used to complete the granularity conversion of current amplitude, voltage amplitude, time interval and other attributes. The standard data file conversion process is shown in the figure;

[0074] The standard space data exchange file interacts with the recording wave master station data and the signal preservation master station data processed by the data cleaning module in real time, and the non-standard or non-standard data identified needs to be transmitted to the data cleaning module again for processing. The data standardization method adopts a softmax function, which can "compress" a k-dimensional vector z containing any real number into another k-dimensional real vector sigma(z). Make the range of each element between (0, 1), and the sum of all elements is 1;

[0075] The data standardization conversion unit is used for processing data processed by the data fault-tolerant processing unit to standardize the data, and the specific calculation expression is as follows:

[0076]

[0077] wherein, represents input zth real vector function in n-dimensional space, C represents the number of fault classes, e represents a natural constant, represents summing C data, represents the zth real vector in n-dimensional space;

[0078] The waveform data feature extraction module is used for processing information of changes before, during and after the fault occurrence of the new energy power distribution network, screening out effective data, extracting fault data features, and transmitting to the sample training module to generate a fault sample library.

[0079] The waveform data feature extraction module uses the fault recording main station data and the signal-keeping main station data processed by the standardization processing module to determine the fault type and the fault property, and respectively takes 1000 groups of waveform data signals for normal operation, intra-zone fault and extra-zone fault, with 1200 sampling points in each group of data. Time domain features of waveform signals in three states are extracted frequency domain features and sample entropy features SE to construct a multi-domain fault feature set of the distributed new energy power distribution network. The fault feature information mainly includes fault interval, fault time, fault phase, trip time, reclosing time, second fault time, second trip time and second fault phase.

[0080] The time domain features extract the instantaneous frequency and amplitude that change with time through empirical mode decomposition (EMD), and the extraction method selects ten time domain features including peak value, peak-peak value, root mean square, absolute mean value, square root amplitude, variance, skewness, kurtosis, waveform index and margin index to form a time domain feature vector. The distributed new energy power distribution network system takes 1000 groups of data for each operating state, and a total of 3000 groups of samples in three states form the dimensional time domain fault feature set and the dimensionless time domain fault feature set. In the early stage of the fault, the signal mutation caused by the fault is weak and difficult to identify in the noise environment. The time-frequency representation of the signal is obtained by using HHT, and the energy distribution of the signal can be analyzed to extract the time domain features of the fault data.

[0081] The frequency domain feature extraction method includes three time domain features, i.e., mean square frequency, frequency variance and peak frequency, to form a three-dimensional frequency domain fault feature vector. ​The distributed new energy distribution network system has 1000 groups of data for each operating state, and 3000 groups of samples in total for three states to form a frequency domain feature set. When the distribution network fails, there are a large number of high-frequency transient components in the recording fault signal, and the high-frequency energy ratio of the fault component is much larger than that of the non-fault component. The frequency domain is integrated to obtain a high-resolution feature that can reflect the statistical energy distribution characteristics of the data in the frequency domain, frequency variation degree (FVD), energy distortion degree (EDD), and amplitude fault degree (AFD), so as to extract the frequency domain features of the fault data;

[0082] The sample entropy feature extraction method selects the first four IMF components as effective components to extract sample entropy and construct a four-dimensional sample entropy feature vector: SE = [SampEn1, SampEn2, SampEn3, SampEn4], wherein SampEn represents sample entropy, SampEn1, SampEn2, SampEn3, and SampEn4 represent four-dimensional sample entropy. The distributed new energy distribution network system has 1000 groups of data for each operating state, and 3000 groups of samples in total for three states to form an entropy feature set. An index representing the degree of component failure is obtained, and the entropy features of the fault data are extracted.

[0083] The multi-domain fault feature extraction method comprises the following steps:

[0084] Step 1: Extracting time domain and frequency domain features of distributed new energy distribution network operating state signals;

[0085] Step 2: VMD decomposition of distributed new energy distribution network fault signals to extract IMF component sample entropy features, and selection of effective features according to the correlation coefficient method;

[0086] Step 3: Combining time domain features, frequency domain features, and sample entropy features to construct a multi-domain feature set;

[0087] Step 4: Global information extraction and fusion of the multi-domain fault feature set, mapping the multi-dimensional features to a low-dimensional space;

[0088] Step 5: End of algorithm, output of a new feature set, i.e. a low-dimensional fault feature set;

[0089] The sample training module obtains the processed fault data of the recording master station and the fault data of the fault data feature of the fault sample library for comparison and analysis processing. After processing, the fault feature sample training is performed, and the pre-judgment result is output to the sorting and classification module;

[0090] The sample training module includes a fault diagnosis sample set, a fault sample library, KELM model training, output of a pre-diagnosis result, and receives data from the waveform data feature extraction module. An overlapping sampling data enhancement technique is used for fault features of the received low-dimensional data. Each sample has an overlapping part with the previous sample. A sample length l is first given, and then a moving step r is set. The overlapping sample length is l-r. The smaller r is, the more overlapping parts there are, and the more samples that can be collected.

[0091] The calculation formula of the sample number is as follows:

[0092]

[0093] Where m represents the sample number, s represents the fault feature signal length of the collected low-dimensional data, l represents the sample length, and r represents the moving step.

[0094] The fault diagnosis sample set is obtained by sample entropy feature extraction in the waveform data feature extraction module. It is found that the sample entropy value of the fault phase is much smaller than that of the non-fault phase, and the fault phase is preliminarily judged. After VMD decomposition, the sample entropy values of the four intrinsic modal components (IMF) of each phase voltage are calculated to form the fault diagnosis sample set.

[0095] The fault sample library is generated by extracting and processing the fault features from the waveform information after collection and processing. The effective data reflecting the fault process are selected, and the fault features are extracted and processed by the waveform data feature extraction module. Each group of fault component three-phase 12 sample entropy values extracted by the sample entropy feature extraction in the waveform data feature extraction module are used as fault features, and are input to the KELM model to select the data processed by the waveform data feature extraction module for training. At time t, the fault feature quantities of each data of the wave recording master station and the signal protection master station form a data set, which is named X t 1， t ...x i，t ...x n，t ], where n represents the number of sample categories, i represents the i-th data fault feature quantity, and x i，t represents the value of the i-th data fault feature quantity at time t. After training, the fault type matching is input to the pre-diagnosis result.

[0096] The sorting and classification module is used for sorting fault data and transmitting to the information fusion diagnosis module.

[0097] The sorting and classification module uses the K-means clustering algorithm for the pre-diagnosis result output by the sample training module, completes the sorting and classification of the distributed new energy distribution network fault data, and the specific steps are as follows:

[0098] ​C1, randomly select k initial cluster centers from the sample training fault data set

[0099] C2, calculate the Euclidean distance between the remaining data objects and the cluster centers Find the cluster center closest to the target data object At the same time, the waveform data object is assigned to the cluster center corresponding cluster, the specific calculation expression is as follows:

[0100]

[0101] wherein, denotes the distance between the similarity between μ and , the m-th cluster center, μ represents a data object, and m represents the dimension of the data object is the m-th attribute value of μ and

[0102] C3, calculate the average value of the data objects in each cluster as the new cluster center again, and perform the next iteration until the cluster center no longer changes;

[0103] The data after classification diagnosis can be roughly divided into three categories: short circuit fault, complex fault and disturbance data. The short circuit fault and complex fault data are input to the information fusion diagnosis module for processing; the fault element reliability is obtained, the diagnosis result is output, and the disturbance data is filed and accepts the information fusion request of other associated data;

[0104] The signal fusion diagnosis module realizes fault recognition after multi-element heterogeneous data standardization and filtering, and outputs the fault diagnosis discrimination result to the intelligent terminal;

[0105] The signal fusion diagnosis module adopts an intelligent fault-tolerant and lossless compatible method based on the COMTRADE standard to solve the problems of multi-source heterogeneous, inconsistent sampling rate, different time synchronization, unit and other interference information fusion precision of different system data, standardize the multi-source heterogeneous data, and output the distributed distribution network fault pre-diagnosis result to the intelligent terminal. The accuracy rate of the information fusion diagnosis module fault identification is greater than 99.5%;

[0106] The intelligent terminal is used to give the decision scheme of fault discrimination;

[0107] ​​​​The intelligent terminal obtains data processed by the signal fusion diagnosis module, automatically analyzes new fault characteristics caused by factors such as limited fault current amplitude, phase change, and equivalent impedance property change, which are affected by intermittent and fluctuating output of distributed new energy and complex and variable factors, adapts to fault characteristic analysis in different stages of the whole process of high-proportion distributed new energy and high-proportion power electronic equipment characteristics, and automatically generates a fault discrimination decision scheme within 10 minutes.

[0108] While embodiments of the application have been shown and described, it is to be understood that the embodiments described are merely exemplary of the principles and application of the present application. Numerous modifications, changes, variations, substitutions, and equivalents will occur to those of ordinary skill in the art without departing from the spirit and scope of the application as defined by the following claims and their equivalents.

Claims

1. A distributed new energy distribution network fault diagnosis system, characterized in that: It includes a data integration module, a data cleaning module, a data standardization processing module, a waveform data feature extraction module, a sample training module, a sorting and classification module, an information fusion and diagnosis module, and an intelligent terminal; The data integration module acquires waveform recording master station data and information protection master station data, which are used to obtain the credibility of fault information in distributed new energy distribution network and perform weighted fusion calculation. The data cleaning module cleans invalid and redundant data generated during data generation, collection, and transmission; the data standardization module processes the data and converts it into standard data. The waveform data feature extraction module is used to process the information on the changes before, during and after the occurrence of faults in the new energy distribution network, filter out effective data, extract fault data features, and transmit them to the sample training module to generate a fault sample library. The waveform data feature extraction module determines the fault type and nature of the fault recording master station data and the information protection master station data processed by the standardization processing module: extracting the time-domain features of the waveform data signals for normal operation, intra-area fault, and extra-area fault respectively. Frequency domain characteristics and sample entropy features A multi-domain fault feature set of distributed new energy distribution network is constructed, and the multi-domain fault feature set of distributed new energy distribution network is mapped in low dimension to obtain a low-dimensional fault feature set. The time-domain features By using empirical mode decomposition to measure the instantaneous frequency and amplitude that change over time, a time-domain feature vector is extracted, consisting of ten time-domain features: peak value, peak-to-peak value, root mean square (RMS), absolute mean, root square amplitude, variance, skewness, kurtosis, waveform index, and margin index. ; The frequency domain features A three-dimensional frequency domain fault feature vector is constructed by extracting three time-domain features: mean square frequency, frequency variance, and kurtosis frequency. ; The sample entropy features By performing VMD decomposition on the fault signals of distributed new energy distribution networks, the first four IMF components are extracted as effective components, and sample entropy is extracted to construct a four-dimensional sample entropy feature vector: ； in, This represents the entropy of the four-dimensional sample. The sample training module acquires the processed fault data from the waveform recording master station and the information protection master station to establish a fault diagnosis sample set, and compares and analyzes it with the fault data features in the fault sample library. After processing, fault feature sample training is performed, and the prediction results are output to the sorting and classification module. The sorting and classification module is used to sort fault data and transmit it to the information fusion and diagnosis module; The information fusion diagnostic module realizes the standardization and filtering of multi-dimensional heterogeneous data and the identification of faults, and outputs the identification and diagnosis results to the smart terminal. The intelligent terminal is used to provide a decision-making scheme for fault diagnosis.

2. The distributed new energy distribution network fault diagnosis system according to claim 1, characterized in that: The waveform recording master station data includes the waveform characteristics of faults on this side and the waveform characteristics of faults on the opposite side; the protection information master station data includes protection settings, protection action signals and switch position signals. The specific workflow of the data integration module is as follows: A1. Perform preliminary judgment of fault morphology characteristics and fault reliability assessment on the fault recording morphology characteristics of the local side and the opposite side in the main recording station data to confirm whether the fault information characteristics of the two sides are consistent. A2. Obtain protection settings, protection action signals and switch position signals from the main station of the protection information system, organize the set of suspected faulty components, calculate the fault confidence index for each component in the set of suspected faulty components, and obtain the fault characterization probability of each component. A3. Calculate the fault reliability index of distributed new energy distribution networks. The specific expression is as follows: ； ； in, As a reliability indicator of the fault, express The probability of occurrence, for System failure load ratio Fault reliability index A reliability index representing the limit state 0. These respectively represent the system in The load after the failure and the normal operating load at any given time.

3. The distributed new energy distribution network fault diagnosis system according to claim 2, characterized in that: The data cleaning module performs quality checks on the data output by the data integration module, identifies outliers, fills in missing data, and smooths noisy data. It employs a swarm intelligence optimization association rule algorithm for data cleaning and calculates the accuracy of the cleaned fault data from the distributed new energy distribution network. The specific expression is as follows: ； in, This indicates that the actual value is a positive class. This indicates that the actual value is negative. Given the total number of sample data, the accuracy of the cleaned data is... 95% of the fault data from distributed renewable energy distribution networks is output to the data standardization processing module. When the accuracy of the cleaned data is lower than... At 95%, it was filtered out after reconfirmation.

4. The distributed new energy distribution network fault diagnosis system according to claim 1, characterized in that: The data standardization processing module includes a data fault tolerance processing unit and a data standardization transformation unit. The data fault tolerance processing unit is used to handle situations where there are missing or redundant data, and specifically includes the following steps: B1. Scan for faulty files to make an initial assessment; B2. If any information attribute is missing, query the attribute list, which contains the attribute type and compensation information, and supplement the missing attribute. B3. If there is missing data, use cubic spline interpolation or linear averaging to fill in the missing data. B4. Redundant data will be deleted; The data standardization and transformation unit is used to standardize the data processed by the data fault tolerance processing unit. The specific calculation expression is as follows: ； in, Indicates input Victor A real vector function, Indicates the number of fault classes. Represents the natural constant. This indicates summing up C data points. Indicates in The first in the dimension A real vector.

5. A distributed new energy distribution network fault diagnosis system according to claim 1, characterized in that: The sorting and classification module uses the K-means clustering algorithm on the pre-diagnostic results output by the sample training module to complete the sorting and classification of fault data in the distributed new energy distribution network. The specific steps are as follows: C1. Randomly select from the fault dataset used for sample training Initial cluster centers ; C2. Calculate the cluster centers of the remaining data objects. Using Euclidean distance, find the cluster centers that are closest to the target data object. Simultaneously, waveform data objects are assigned to cluster centers. The specific calculation expression for the corresponding cluster is as follows: ； Where, d express and The distance between the similarities, For the first Cluster centers, Represents a data object. Represents the dimensions of a data object. For μ and The Each attribute value; C3. Calculate the average value of the data objects in each cluster again as the new cluster center, and perform the next iteration until the cluster centers no longer change.

Images