Intelligent power grid fault analysis method and system based on big data

Abstract

The invention relates to the technical field of intelligent power grid fault analysis, in particular to an intelligent power grid fault analysis method and system based on big data. Comprising the following steps: analyzing a power grid physical wiring diagram to construct a node admittance matrix, and extracting modal feature vectors through feature decomposition to generate a topological fingerprint database; synchronously collecting wide-area phasor data streams, dynamically intercepting a data time window, extracting resonance frequency band signals, and generating a fault feature data set through feature enhancement; a modal confidence factor is dynamically evaluated, when the dominant modal confidence coefficient is lower than a threshold value, redundant modals in the node-modal association map are activated, and fault feature vectors are fused and reconstructed; and matching the reconstruction vector with a fingerprint database, performing triple verification, and outputting a fault positioning coordinate or starting a safety protection instruction in combination with a confidence scoring decision. By introducing a dynamic confidence mechanism and a redundancy modal compensation strategy, the accuracy and fault-tolerant capability of fault identification in a new energy disturbance environment are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of smart grid fault analysis, in particular to a smart grid fault analysis method and system based on big data. BACKGROUND

[0002] In the new power system, large-scale access of new energy significantly improves the dynamic nature of the power grid, and frequent power disturbance and topology change complicate the fault location problem. Although the wide-area measurement system provides high spatiotemporal resolution data support, it still faces challenges in how to extract key modal response features with high fault correlation from it. The commonly used fault recognition methods currently rely on static modal features, and lack dynamic evaluation of modal stability and reliability. When the dominant modal degenerates or energy disperses, there is a lack of effective alternative mechanism, resulting in poor fault judgment robustness and high misjudgment rate.

[0003] Therefore, there is an urgent need for a smart grid fault analysis method and system based on big data, which can dynamically identify key modal features in a multi-disturbance, multi-source input environment, and improve the accuracy of fault location and the overall robustness of the system. SUMMARY

[0004] (1) Technical problem to be solved

[0005] The purpose of the present application is to provide a smart grid fault analysis method and system based on big data to solve the problem of unstable fault feature extraction caused by modal frequency drift and dominant modal failure under the condition of high proportion of new energy access and frequent change of power grid topology.

[0006] (2) Technical solution

[0007] To achieve the above purpose, on the one hand, the present application provides a smart grid fault analysis method based on big data, which comprises:

[0008] Step S1: Obtain a power grid physical connection diagram; analyze the power grid physical connection diagram to construct a node admittance matrix; extract modal feature vectors from the node admittance matrix through eigenvalue decomposition; generate dynamic fingerprint parameters according to the modal feature vectors to construct a topology fingerprint library.

[0009] Step S2: Synchronously collect the phasor data stream of the wide-area measurement system, dynamically intercept the data time window corresponding to the dispatching instruction period; extract the current and voltage signals of the power grid inherent resonant frequency band within the data time window; perform fault feature enhancement processing on the current and voltage signals to generate a fault feature data set.

[0010] Step S3: obtaining a corresponding modal confidence factor through dynamic evaluation according to the modal feature vector; when the modal confidence factor of the dominant modal is lower than a preset activation threshold, activating a pre-constructed node-modal correlation graph, selecting a redundant modal in the node-modal correlation graph, and fusing the feature components corresponding to the dominant modal in the fault feature data set and the redundant modal to generate a fault feature vector.

[0011] Step S4: matching the fault feature vector with a topology fingerprint library, and sequentially performing Kirchhoff's law verification, topology connection consistency verification, and traveling wave reflection characteristic verification; calculating a confidence score according to the distribution result of the modal confidence factor, and when the confidence score is greater than a preset score threshold and the verification is passed, outputting a fault positioning coordinate; otherwise, starting a power grid safety protection instruction, the power grid safety protection instruction including locking automatic reclosing, starting fault recording, or sending a dispatching alarm signal.

[0012] Further, the method for constructing a topology fingerprint library according to the modal feature vector includes:

[0013] The modal feature vector is a low-energy modal vector with a feature value less than a preset modal energy threshold; a multi-dimensional dynamic fingerprint parameter is constructed according to the modal feature vector, including a resonance frequency offset Δf=(f1-f0) / f0 and an impedance derivative norm ‖dZ / dt‖, where f1 is a real-time measurement value, f0 is a reference frequency, Z is a node impedance matrix element, and t is a time variable synchronized with a wide-area measurement system at all times.

[0014] A topology hash value is generated through a digital signature algorithm according to the dynamic fingerprint parameter; and the topology hash value is stored in the topology fingerprint library.

[0015] Further, the method for constructing the node-modal correlation graph includes:

[0016] The electrical connection degree D between nodes is calculated according to the modulus of the node impedance matrix element mm ;

[0017] The modal response similarity S is calculated according to the included angle cosine between the modal feature vectors ij ;

[0018] The correlation strength W=D mm ×S ij is calculated according to the electrical connection degree and the modal response similarity; a triple [node, correlation strength, modal] is generated and stored in a graph database structure.

[0019] Further, the calculation process of the modal confidence factor includes:

[0020] The drift amplitude of the monitoring modal frequency is monitored, a history memory item reflecting the recent disturbance state of the modal is constructed, the time decay weighting is performed according to the history memory item and the initial confidence value, the confidence degree is calculated in combination with the correlation between the modal and the power disturbance, and a modal confidence factor is obtained.

[0021] Further, the method of obtaining the corresponding confidence factor through dynamic evaluation according to the modal feature vector comprises:

[0022] The initial modal weight factor is set according to the category of the power grid line.

[0023] The power output fluctuation rate of the new energy access point is monitored in real time, and when the power output fluctuation rate exceeds the preset disturbance threshold, the initial modal weight factor is dynamically adjusted to obtain a modal weight factor.

[0024] When the access area is a photovoltaic field station, the weight of the frequency-related modal component is increased, and the weight of the impedance derivative-related modal component is reduced.

[0025] When the access area is a wind power field station, the weight of the impedance derivative-related modal component is increased, and the weight of the frequency-related modal component is reduced.

[0026] The modal weight factor and the modal feature vector are fused through weighting to generate a fault feature vector.

[0027] Further, the method further comprises:

[0028] The Shannon information entropy of the modal response amplitude and the relative drift amplitude of the frequency are calculated, when the Shannon information entropy exceeds the preset information entropy threshold or the relative drift amplitude exceeds the drift amplitude threshold, the confidence factor of the corresponding modal is down-regulated according to the preset penalty coefficient, and the corrected confidence factor is used for weighting operation on the fault feature vector.

[0029] Further, the method further comprises:

[0030] In the topology fingerprint matching process, the overall modal information entropy of the fault feature data set is calculated.

[0031] When the overall modal information entropy is less than the information entropy partition threshold, a full-amount modal matching strategy is adopted.

[0032] When the overall modal information entropy is greater than the information entropy partition threshold, a redundant modal weighted matching strategy is adopted, and the modal confidence factor is used as a matching constraint condition.

[0033] Further, the method further comprises:

[0034] The output confidence score is obtained by weighted fusion calculation according to the modal confidence factor distribution and the feature matching similarity; when the confidence score is lower than a preset score threshold, the automatic output of the fault positioning coordinates is rejected, and the fault recording data is returned; the confidence score is used as a fusion judgment quantity in a dispatching system to prioritize the positioning results.

[0035] Based on the same inventive concept, in another aspect, the application also provides a smart grid fault analysis system based on big data, which comprises:

[0036] A topology fingerprint library construction module is configured to obtain a physical wiring diagram of a power grid, analyze the physical wiring diagram to construct a node admittance matrix, extract a modal feature vector from the node admittance matrix through characteristic decomposition, and generate a dynamic fingerprint parameter to construct a topology fingerprint library according to the modal feature vector.

[0037] A real-time data preprocessing module is configured to synchronously collect phasor data streams of a wide-area measurement system, dynamically intercept data time windows corresponding to dispatching instruction cycles, extract current and voltage signals of inherent resonance frequency bands of the power grid within the data time windows, and generate a fault feature data set by performing fault feature enhancement processing on the current and voltage signals.

[0038] A fault feature reconstruction module is configured to obtain a corresponding modal confidence factor through dynamic evaluation according to the modal feature vector, activate a pre-constructed node-modal correlation graph when the modal confidence factor of a dominant modal is lower than a preset activation threshold, select redundant modals in the node-modal correlation graph, and generate a fault feature vector by fusing feature components corresponding to the dominant modal and the redundant modals in the fault feature data set.

[0039] A multi-dimensional verification decision module is configured to match the fault feature vector with the topology fingerprint library, and sequentially perform Kirchhoff's law verification, topology connection consistency verification, and traveling wave reflection characteristic verification, calculate a confidence score according to the distribution result of the modal confidence factor, output fault positioning coordinates when the confidence score is greater than a preset score threshold and the verification is passed, and otherwise, start a power grid safety protection instruction, wherein the power grid safety protection instruction comprises locking automatic reclosing, starting fault recording, or sending a dispatching alarm signal.

[0040] (3) Beneficial effects

[0041] Compared with the prior art, the application has the following beneficial effects:

[0042] 1. The application introduces a modal confidence factor and an uncertainty measurement mechanism to realize quantitative modeling and online discrimination of modal response credibility, and can still stably extract effective features in the case of degradation of a dominant modal.

[0043] 2. A collaborative control strategy of redundant modal compensation and matching score is proposed to support rapid modal switching and feature reconstruction when the main mode is unavailable, which significantly improves the fault identification accuracy and fault tolerance robustness of the system under multi-source disturbance environment. Attached Figure Description

[0044] Figure 1 This is a block diagram of the smart grid fault analysis method based on big data according to Embodiment 1 of the present invention;

[0045] Figure 2 This is a block diagram of the smart grid fault analysis system based on big data according to Embodiment 2 of the present invention. Detailed Implementation

[0046] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0047] Before providing examples, it is necessary to describe the application scenarios of this invention. This invention is applicable to complex power grid environments with a wide-area power system, particularly those with a high proportion of renewable energy integration. In such grids, the volatility and intermittency of renewable energy sources (such as wind power and photovoltaics) significantly increase the instability of the grid's modal response, causing methods based on static topology, fixed modal features, or rule bases to exhibit low confidence and increased false positive rates in fault location. Furthermore, the frequent adjustments to current power grid operation and scheduling result in a dynamically changing line topology, making traditional solutions relying on static models or fixed path propagation features difficult to adapt to. In scenarios with frequent renewable energy disturbances, the modal energy distribution reflected in time-series signals such as current and voltage is prone to drift or even failure of the dominant mode, leading to the failure of conventional matching strategies and the inability to effectively locate fault areas. Therefore, this invention focuses on addressing the following issues: high renewable energy integration ratio and modal response instability; frequent topology adjustments and failure of traditional fingerprint features; fault signals being masked or weakened by disturbances, resulting in decreased confidence; and the need for the system to possess online adaptive identification and fault-tolerant matching capabilities. Based on the above-mentioned practical needs, this invention constructs a dynamic modal fingerprint matching mechanism, introduces modal confidence factors and uncertainty measurement system, and realizes dynamic and reliable extraction and fault-tolerant location of fault features. It is particularly suitable for typical complex power grid scenarios with significant new energy disturbances, flexible power grid structure and strong state uncertainty.

[0048] Example 1: As Figure 1 As shown, this embodiment provides a smart grid fault analysis method based on big data, the method including:

[0049] Step S1: Obtain the power grid physical connection diagram; analyze the power grid physical connection diagram to construct a node admittance matrix; extract modal characteristic vectors by eigenvalue decomposition of the node admittance matrix; generate dynamic fingerprint parameters according to the modal characteristic vectors to construct a topology fingerprint library.

[0050] Step S2: Synchronously collect the phasor data stream of the wide-area measurement system, dynamically intercept the data time window corresponding to the dispatching instruction period; extract the current and voltage signals of the power grid inherent resonance frequency band within the data time window; perform fault feature enhancement processing on the current and voltage signals to generate a fault feature data set.

[0051] Step S3: Obtain the corresponding modal confidence factor by dynamic evaluation according to the modal characteristic vectors; when the modal confidence factor of the dominant modal is lower than a preset activation threshold, activate the pre-constructed node-modal correlation graph, select the redundant modal in the node-modal correlation graph; and fuse the feature components corresponding to the dominant modal and the redundant modal in the fault feature data set to generate a fault feature vector.

[0052] Step S4: Match the fault feature vector with the topology fingerprint library, and sequentially perform Kirchhoff's law verification, topology connection consistency verification, and traveling wave reflection characteristic verification; calculate a confidence score according to the distribution result of the modal confidence factor; when the confidence score is greater than a preset score threshold and the verification is passed, output the fault positioning coordinates; otherwise, start the power grid safety protection instruction, which includes locking the automatic reclosing, starting the fault recording, or sending the dispatching alarm signal.

[0053] For example, the scenario of this embodiment is a single-phase ground fault of a wind farm power collection line accompanied by power fluctuations caused by turbulence.

[0054] In particular, the dominant modal refers to a modal component with an energy proportion greater than 30% or a weight ranking first; dynamic evaluation includes monitoring frequency drift amplitude, constructing disturbance history memory items, and calculating power disturbance correlation; the feature component is a sub-element of the modal characteristic vector, representing the electrical parameters of a single modal.

[0055] Further, the method of generating dynamic fingerprint parameters according to the modal characteristic vectors to construct a topology fingerprint library comprises:

[0056] The modal characteristic vector is a low-energy modal vector with a characteristic value less than a preset modal energy threshold; the resonance frequency offset Δf = (f1-f0) / f0 and the impedance derivative norm ‖dZ / dt‖ are calculated according to the modal characteristic vector, and a multi-dimensional dynamic fingerprint parameter is constructed; wherein f1 is the real-time measurement value, f0 is the reference frequency, Z is the node impedance matrix element, and t is the time variable which is always synchronized with the wide-area measurement system.

[0057] According to the dynamic fingerprint parameter, a topology hash value is generated by a digital signature algorithm; and the topology hash value is stored in a topology fingerprint library.

[0058] An exemplary power grid physical connection diagram is parsed, including nodes #201-#210 of the wind farm, and a 582-dimensional node admittance matrix is generated; the node admittance matrix is constructed according to an IEEE 39-node standard model, and elements Y ij = G ij +jB ij are calculated from branch impedance parameters. At the same time, a wind farm station category label is written into a topology library for subsequent scene association of mode matching.

[0059] Eigenvalue decomposition is performed, and a low-energy mode vector of a characteristic value λ < 0.15 is extracted, λ max = 3.07. The preset mode energy threshold is dynamically set according to the size of the power grid, and in this embodiment, the threshold is set to λ = 0.05 when the number of nodes of the wind farm is 10, λ max ≈ 0.15. Due to the increase in the density of resonance modes caused by the expansion of the node scale, the threshold needs to be reduced to avoid mode aliasing, so if the number of regional nodes is greater than 500, the threshold is adjusted to 0.03. max

[0060] According to the selected mode components, a resonance frequency offset Δf = (52.3-50) / 50 = 0.046 and an impedance derivative norm ‖dZ / dt‖ = 22.1 are calculated, which are used to construct dynamic fingerprint parameters.

[0061] In this embodiment, Z in the impedance derivative norm ‖dZ / dt‖ is defined as an element of a node impedance matrix, which is obtained by inverting a node admittance matrix, and its real-time value is calculated and updated based on voltage and current data of a wide-area measurement system. The impedance derivative norm is normalized according to a reference capacity of 100 MVA of the IEEE 39-node system, and the unit is pu / s.

[0062] The time variable t is based on a dispatch master station clock, the time window length dt for derivative calculation is consistent with the dispatch period T, and in this embodiment, T = dt = 200 ms, which ensures synchronization with the time scale of the phasor data stream.

[0063] After the dynamic fingerprint parameters are serialized, a hash value d4e5f6...a9b1 is generated by inputting the dynamic fingerprint parameters into an SHA-256 algorithm and stored in a topology fingerprint library to support subsequent fault feature comparison.

[0064] Further, the method for constructing the node-mode association graph comprises:

[0065] The electrical connection degree D mn between nodes is calculated according to the modulus of the elements of the node impedance matrix. ​

[0066] The modal response similarity S is calculated according to the included angle cosine between modal eigenvectors ij；

[0067] The correlation strength W=D is calculated according to the electrical connection degree and the modal response similarity mn ×S ij A triple [node, correlation strength, mode] is generated and stored as a graph database structure.

[0068] For example, when the system detects that the dominant modal confidence factor 0.65 is less than the activation threshold 0.7, the redundant modal compensation mechanism is triggered.

[0069] The system calls a pre-defined node-mode correlation atlas; selects the k redundant modal components most relevant to the current fault target area from the atlas as compensation candidates. k is dynamically set according to the number of nodes in the power grid area.

[0070] For example, when the target node #201 fails, the three redundant modes with the highest correlation strength in the atlas are selected, such as #205, #209, and #210.

[0071] Subsequently, these redundant modes are introduced into the current fault feature reconstruction process and fused with the dominant modal features to generate a reconstructed eigenvector with more stable structure and higher disturbance robustness. This process ensures that the fault feature recognition ability and system judgment continuity can be maintained even in the case of degradation or failure of the dominant mode.

[0072] Further, the calculation process of the modal confidence factor includes:

[0073] The drift amplitude of the modal frequency is monitored to construct a history memory item reflecting the recent disturbance state of the mode; the history memory item and the initial confidence value are weighted according to the time decay; the confidence is calculated in combination with the correlation between the mode and the power disturbance to obtain the modal confidence factor.

[0074] For example, the past three period confidence [0.85, 0.76, 0.68] and the time decay factor η=0.8.

[0075] The dynamic modal confidence factor is used to quantify the stability and reliability of each modal component under the current power grid state, and is used to guide the weighting of the fault feature vector and the triggering mechanism of the redundant mode. To calculate the modal confidence factor, first monitor the relative drift amplitude of the current modal frequency compared to the reference frequency. If the frequency fluctuates dramatically, it means that the mode is more affected by the disturbance and the stability decreases, so the confidence decreases. The system extracts the activation records of the mode from the last n dispatching periods to construct a history memory item Applying a time decay factor η can achieve dynamic evaluation of the recent activation degree; where h i(k) is the activation state of the mode in the kth period, activated is 1, and not activated is 0. The number of history periods n = min(5, fault feature time window length / scheduling period), which ensures that the memory items cover the typical disturbance decay period.

[0076] The correlation factor between the mode amplitude and the new energy power disturbance rate is combined to comprehensively evaluate the confidence level. Finally, the system outputs the mode confidence factor of the fusion drift feature Historical stability and disturbance correlation p i of the mode for guiding the weighted processing of the fault feature vector and the redundant mode triggering mechanism. Wherein, g is an experience set sensitivity factor, the range is usually [3, 7], which is adjusted according to the scene disturbance intensity, and here g = 5; Df i is the real-time frequency offset of mode i, is the confidence degree of the last period, wherein, is the normalized standard deviation of the new energy power fluctuation rate, and the new energy type coefficient K = 0.5 in the wind power field and K = 0.3 in the photovoltaic field.

[0077] Further, the method of obtaining the corresponding confidence factor according to the mode feature vector through dynamic evaluation comprises:

[0078] According to the category of the power grid line, an initial mode weight factor is set.

[0079] The power output fluctuation rate of the new energy access point is monitored in real time, and when the power output fluctuation rate exceeds the preset disturbance threshold, the initial mode weight factor is dynamically adjusted to obtain a mode weight factor.

[0080] When the access area is a photovoltaic station, the weight of the frequency-related mode component is increased, and the weight of the impedance derivative-related mode component is reduced.

[0081] When the access area is a wind power station, the weight of the impedance derivative-related mode component is increased, and the weight of the frequency-related mode component is reduced.

[0082] The mode weight factor and the mode feature vector are fused by weighting to generate a fault feature vector.

[0083] For example, the wide area measurement system vector data stream is synchronously collected, and the sampling rate of 4kHz meets the IEEE C37.118.2 standard.

[0084] The data window is intercepted according to the scheduling period T = 200ms; the time length is dynamically adjusted according to the real-time load of the scheduling system.

[0085] The current and voltage signals in the 55-60 Hz frequency band are extracted, and the inherent resonance frequency band is determined through historical fault statistics, covering 90% of the grid resonance points.

[0086] The wavelet packet decomposition is used to extract the resonance frequency band features, and the data dimension is compressed through the sparse autoencoder, and the signal-to-noise ratio is improved from 15 dB to 23 dB, which is greater than the index of 20 dB.

[0087] To realize the dynamic evaluation of the modal confidence factor, in a certain wind farm access area, the system first sets the initial modal weight factor according to the type of the grid line. By default, the weight of the frequency-related modal component and the impedance derivative-related modal component is 0.5.

[0088] Because the wind farm power fluctuation rate is 32% greater than the preset disturbance threshold of 25%, the dynamic correction mechanism is triggered. Because the wind farm access disturbance is more likely to cause impedance response changes, the impedance derivative modal weight is increased from 0.5 to 0.7, and the frequency modal weight is reduced from 0.5 to 0.3. The frequency component Z and the impedance derivative quantity f in the modal feature vector are extracted, and the modified modal weight vector W' = [0.3, 0.7] is generated, and the modified fault feature vector V f = W'·[Z, f] is generated by weighting and fusing. The fused fault feature vector V f can highlight the impedance derivative-related modal, and the stability and positioning accuracy of feature extraction are improved in the environment dominated by wind power disturbance.

[0089] In particular, the disturbance threshold is set according to the anti-disturbance ability grading of the new energy station, and the dynamic correction mechanism is only applicable to the defined wind and photovoltaic scenarios.

[0090] Further, the method further comprises:

[0091] The Shannon information entropy of the modal response amplitude and the relative drift amplitude of the frequency are calculated; when the Shannon information entropy exceeds the preset information entropy threshold, or the relative drift amplitude exceeds the drift amplitude threshold, the confidence factor of the corresponding modal is lowered according to the preset penalty coefficient; and the corrected confidence factor is used for weighting operation on the fault feature vector.

[0092] In an exemplary single-phase ground fault scenario of a certain wind power collection line, the system first extracts the response amplitude and frequency fluctuation information of each modal from the current fault feature data set. For each modal component, the system calculates its amplitude Shannon information entropy H i and the relative frequency drift amplitude Δf i / f0. In this embodiment, the amplitude Shannon information entropy of the modal #207 is 0.41, which is greater than the preset information entropy threshold of 0.3; at the same time, the relative frequency drift amplitude of the modal is 0.12, which is greater than the drift amplitude threshold of 0.05.

[0093] Any condition is considered to be modal instability, triggering the confidence factor down; select the weight ratio of more than 30% of the modal as the dominant modal, the dominant modal confidence The corrected confidence factor is 0.8xC t ≈0.61, wherein 0.8 is a penalty coefficient determined by training historical failure data. Then normalize all modal confidence factors to ensure consistency when participating in the weighting process of the fault feature vector; wherein k is the number of historical periods.

[0094] The correction mechanism is established based on historical failure statistics, ensuring that the system can identify modal instability in real time and improve the reliability of the final matching score by dynamically adjusting the confidence.

[0095] In particular, the preset information entropy threshold and drift amplitude threshold are obtained by training historical failure data, and the training set contains 327 failure records.

[0096] Further, the method further comprises:

[0097] In the topology fingerprint matching process, the overall modal information entropy of the fault feature data set is calculated.

[0098] When the overall modal information entropy is less than the information entropy partition threshold, a full-quantity modal matching strategy is adopted.

[0099] When the overall modal information entropy is greater than the information entropy partition threshold, a redundant modal weighted matching strategy is adopted, and the modal confidence factor is used as a matching constraint condition.

[0100] Illustratively, in the topology fingerprint matching process, the system first calculates the overall modal information entropy of the current fault feature data set to evaluate the concentration or dispersion of modal energy distribution. When the overall modal information entropy is less than the information entropy partition threshold 0.5, it indicates that the modal energy distribution is relatively concentrated, and at this time, the full-quantity modal matching strategy is preferentially executed to match all valid modal information to improve accuracy.

[0101] If the overall modal information entropy is higher than the information entropy partition threshold, it indicates that the modal features show a dispersion trend, and there may be abnormal situations such as degradation of the dominant modal and enhancement of signal disturbance. At this time, the system automatically switches to the redundant modal weighted matching strategy, which preferentially considers the modal components with high confidence factors and uses them as matching constraint conditions to suppress the interference of unstable modal on the matching result.

[0102] The strategy switching mechanism ensures that the system can still perform stable and robust fault matching and positioning judgment under different disturbance environments.

[0103] In particular, the information entropy partition threshold is set to 0.5 according to the scale of the power grid.

[0104] Further, the method further comprises:

[0105] According to the modal confidence factor distribution and the feature matching similarity, the output confidence score is calculated by weighted fusion calculation; when the confidence score is lower than the preset score threshold, the automatic output of the fault positioning coordinates is rejected, and the fault recording data is returned and the expert diagnosis system is intervened; the confidence score is used as a fusion judgment quantity in the dispatching system, and the positioning result priority is sorted.

[0106] For example, the fault feature matching similarity is 0.88 obtained by the cosine similarity algorithm.

[0107] Three verifications are performed:

[0108] Kirchhoff's law verification: the deviation 4.2% is less than the threshold value 5%, and the verification is passed; wherein the threshold value standard 5% is based on IEEE 1159 §4.3.

[0109] Topological connection consistency: the SCADA matching rate 98% is greater than the threshold value 95%, and the verification is passed; wherein the threshold value standard 95% is based on DL / T 1230-2021.

[0110] Traveling wave reflection characteristics: Δt=45.6μs, within the standard of 28.8-53.6μs, the verification is passed; wherein the threshold value standard is a new solidified threshold value suitable for the new energy short line high fluctuation scene.

[0111] The confidence score S is obtained by weighted fusion of the modal confidence factor distribution and the feature matching similarity, wherein w i is the above-mentioned modified modal weight, ξ is the modal confidence weight, and ζ is the feature matching similarity weight. The weights ξ and ζ are dynamically adjusted according to the overall modal information entropy H. According to the analysis of 327 historical fault data, when H>0.3, the modal instability probability is >80%; therefore, when H<0.3, the values ξ=0.4 and ζ=0.6 are taken to make the feature matching more reliable; when H≥0.3, the values ξ=0.7 and ζ=0.3 are taken to make the modal confidence more priority.

[0112] Since S=0.67 is less than the score threshold value 0.7 set according to the dispatching regulations, the automatic output of the positioning coordinates is rejected. The power grid safety protection instruction is started, the grid-connected breaker of the wind farm is locked out, the fault recording data is returned, the dispatching alarm signal is sent to the regional monitoring center. S is used as a fusion judgment quantity, and the present fault is marked as Class-B priority, which needs to be manually reviewed, and the wind farm #201-#210 power collection line is cut off.

[0113] Embodiment 2: based on the same inventive concept as inFigure 2 The embodiment also provides a smart grid fault analysis system based on big data, which comprises:

[0114] A topology fingerprint library construction module is configured to acquire a physical wiring diagram of a power grid, analyze the physical wiring diagram of the power grid to construct a node admittance matrix, extract a modal feature vector from the node admittance matrix through eigenvalue decomposition, and generate a dynamic fingerprint parameter to construct a topology fingerprint library according to the modal feature vector.

[0115] A real-time data preprocessing module is configured to synchronously collect a phasor data stream of a wide-area measurement system, dynamically intercept a data time window corresponding to a dispatch instruction period, extract current and voltage signals of a natural resonant frequency band of the power grid within the data time window, and generate a fault feature data set by performing fault feature enhancement processing on the current and voltage signals.

[0116] A fault feature reconstruction module is configured to obtain a corresponding modal confidence factor through dynamic evaluation according to the modal feature vector, activate a pre-constructed node-modal correlation graph when a modal confidence factor of a dominant modal is lower than a preset activation threshold, select a redundant modal in the node-modal correlation graph, and fuse a feature component corresponding to the dominant modal in the fault feature data set with the redundant modal to generate a fault feature vector.

[0117] A multi-dimensional verification decision module is configured to match the fault feature vector with the topology fingerprint library, and sequentially perform Kirchhoff's law verification, topology connection consistency verification, and traveling wave reflection characteristic verification, calculate a confidence score according to a distribution result of the modal confidence factor, output a fault positioning coordinate when the confidence score is greater than a preset score threshold and the verification is passed, and otherwise, start a power grid safety protection instruction, wherein the power grid safety protection instruction comprises locking automatic reclosing, starting fault recording, or sending a dispatch alarm signal.

[0118] It should be noted that, as for the system in the above embodiment, the specific manner in which each module performs an operation has been described in detail in the embodiment related to the method, and will not be described in detail here.

[0119] Finally, it should be noted that, although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions recorded in the foregoing embodiments, or equivalently replace some of the technical features, and any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. A smart grid fault analysis method based on big data, characterized in that, The method includes: Obtain the physical wiring diagram of the power grid; parse the physical wiring diagram of the power grid to construct the node admittance matrix; extract the modal feature vectors from the node admittance matrix through eigenvalue decomposition; generate dynamic fingerprint parameters based on the modal feature vectors to construct a topological fingerprint database; The system synchronously acquires phasor data streams from a wide-area measurement system and dynamically extracts data time windows corresponding to scheduling command cycles; it extracts current and voltage signals from the inherent resonant frequency band of the power grid within the data time window; and it performs fault feature enhancement processing on the current and voltage signals to generate a fault feature dataset. Based on the modal feature vector, the corresponding modal confidence factor is obtained through dynamic evaluation; when the modal confidence factor of the dominant mode is lower than the preset activation threshold, the pre-constructed node-modal association graph is activated, and the redundant modes in the node-modal association graph are selected; the feature components corresponding to the dominant mode in the fault feature dataset are fused with the redundant modes to generate a fault feature vector; The fault feature vector is matched with the topology fingerprint database, and Kirchhoff's laws, topology connectivity consistency, and traveling wave reflection characteristics are verified sequentially. A confidence score is calculated based on the distribution of modal confidence factors. When the confidence score is greater than a preset score threshold and the verification is passed, the fault location coordinates are output. Otherwise, a power grid safety protection command is initiated, which includes blocking automatic reclosing, initiating fault recording, or sending a dispatch alarm signal.

2. The smart grid fault analysis method based on big data according to claim 1, characterized in that, The method for generating dynamic fingerprint parameters and constructing a topological fingerprint database based on the modality feature vector includes: The modal feature vector is a low-energy modal vector with an eigenvalue less than a preset modal energy threshold. Based on the modal feature vector, the resonant frequency offset Δf = (f1-f0) / f0 and the impedance derivative norm ‖dZ / dt‖ are calculated to construct multidimensional dynamic fingerprint parameters. Among them, f1 is the real-time measurement value, f0 is the reference frequency, Z is the node impedance matrix element, and t is the time variable in which the reference is always synchronized with the wide-area measurement system. A topological hash value is generated using a digital signature algorithm based on the dynamic fingerprint parameters; the topological hash value is then stored in a topological fingerprint database.

3. The smart grid fault analysis method based on big data according to claim 2, characterized in that, The method for constructing the node-modal association graph includes: The electrical connectivity D between nodes is calculated based on the magnitudes of the elements of the node impedance matrix. mn ; The modal response similarity S is calculated based on the cosine of the angle between the modal feature vectors. ij ; The correlation strength W = D is calculated based on the electrical connectivity and modal response similarity. mn ×S ij Generate triples [node, association strength, modality] and store them as a graph database structure.

4. The smart grid fault analysis method based on big data according to claim 3, characterized in that, The calculation process of the modal confidence factor includes: Monitor the drift amplitude of the modal frequency and construct a historical memory term reflecting the recent perturbation state of the mode; perform time decay weighting based on the historical memory term and the initial confidence value; calculate the confidence level by combining the correlation between the mode and the power perturbation to obtain the modal confidence factor.

5. The smart grid fault analysis method based on big data according to claim 4, characterized in that, The method for obtaining the corresponding confidence factor based on the modality feature vector through dynamic evaluation includes: The initial modal weighting factor is set according to the type of power grid line; Real-time monitoring of the power output volatility of the new energy access point; when the power output volatility exceeds a preset disturbance threshold, the initial mode weighting factor is dynamically adjusted to obtain the mode weighting factor. When the access area is a photovoltaic power station, the weight of frequency-related mode components is increased and the weight of impedance derivative-related mode components is decreased. When the access area is a wind farm, increase the weight of impedance derivative-related mode components and decrease the weight of frequency-related mode components; The modal weighting factor and the modal feature vector are weighted and fused to generate a fault feature vector.

6. The smart grid fault analysis method based on big data according to claim 4, characterized in that, The method further includes: Calculate the Shannon information entropy of the modal response amplitude and the relative drift amplitude of the frequency; when the Shannon information entropy exceeds a preset information entropy threshold, or the relative drift amplitude exceeds a drift amplitude threshold, adjust the confidence factor of the corresponding mode according to a preset penalty coefficient; use the corrected confidence factor for weighted calculation of the fault feature vector.

7. The smart grid fault analysis method based on big data according to claim 6, characterized in that, The method further includes: During the topological fingerprint matching process, the overall modal information entropy of the fault feature dataset is calculated; When the overall modal information entropy is less than the information entropy partition threshold, a full modal matching strategy is adopted. When the overall modal information entropy is greater than the information entropy partitioning threshold, a redundant modal weighted matching strategy is adopted, and the modal confidence factor is used as the matching constraint.

8. The smart grid fault analysis method based on big data according to claim 7, characterized in that, The method further includes: The confidence score is obtained by weighted fusion calculation based on the distribution of modal confidence factors and feature matching similarity. When the confidence score is lower than the preset score threshold, the automatic output of fault location coordinates is rejected, and fault recording data is triggered to be transmitted back. The confidence score is used as the priority ranking of the fusion judgment result location in the scheduling system.

9. A smart grid fault analysis system based on big data, characterized in that, The system includes: A topology fingerprint database construction module is used to obtain the physical wiring diagram of the power grid; parse the physical wiring diagram of the power grid to construct a node admittance matrix; extract modal feature vectors from the node admittance matrix through feature decomposition; and generate dynamic fingerprint parameters based on the modal feature vectors to construct a topology fingerprint database. The real-time data preprocessing module is used to synchronously acquire phasor data streams from the wide-area measurement system, dynamically extract data time windows corresponding to scheduling command cycles, extract current and voltage signals within the inherent resonant frequency band of the power grid within the data time window, and perform fault feature enhancement processing on the current and voltage signals to generate a fault feature dataset. The fault feature reconstruction module is used to obtain the corresponding modal confidence factor through dynamic evaluation based on the modal feature vector; when the modal confidence factor of the dominant mode is lower than the preset activation threshold, the pre-constructed node-modal association graph is activated, and redundant modes in the node-modal association graph are selected; the feature components corresponding to the dominant mode in the fault feature dataset are fused with the redundant modes to generate a fault feature vector; The multidimensional verification decision module is used to match the fault feature vector with the topology fingerprint database, and sequentially perform Kirchhoff's law verification, topology connection consistency verification, and traveling wave reflection characteristic verification; calculate the confidence score based on the distribution results of the modal confidence factor; when the confidence score is greater than the preset score threshold and the verification is passed, output the fault location coordinates; otherwise, initiate the power grid safety protection command, which includes blocking automatic reclosing, initiating fault waveform recording, or sending a dispatch alarm signal.

Images