Fault identification method and system for distributed intelligent power distribution system

Abstract

The invention relates to the technical field of fault identification, and discloses a fault identification method and system for a distributed intelligent power distribution system, and the method comprises the steps: carrying out the wavelet decomposition and threshold de-noising of a historical power waveform of a power distribution system according to the number of historical wavelet decomposition layers, and obtaining a de-noised wavelet coefficient set, calculating the residual entropy of each layer of historical wavelet decomposition layer number to obtain an entropy sequence, screening out an initial wavelet decomposition layer number corresponding to a target entropy value in the entropy sequence, decomposing the historical power waveform to each layer in a candidate wavelet decomposition layer number set obtained by discrete normal disturbance, extracting a root-mean-square value and frequency spectrum energy distribution, and generating a fault identification result. And calculating the analysis accuracy of each layer in combination with a fault type label, screening out a target wavelet decomposition layer number, and performing fault identification on the real-time waveform by using the target wavelet decomposition layer number to obtain a target fault identification result. According to the method, the wavelet decomposition layer number is effectively selected, so that the quality of the signal feature set and the precision and efficiency of obstacle recognition are improved.

Description

Technical Field

[0001] This invention relates to the field of fault identification technology, and in particular to a fault identification method and system for a distributed intelligent power distribution system. Background Technology

[0002] In distributed intelligent power distribution systems, the large-scale integration of distributed power sources leads to increasingly complex system structures and results in highly volatile operating conditions and diverse fault types. Traditional fault identification methods based on current and voltage thresholds are prone to misjudgment or missed detection in complex scenarios. Therefore, researchers widely adopt fault identification methods based on signal processing and intelligent algorithms.

[0003] Among them, wavelet transform, due to its excellent time-frequency localization characteristics, is widely used in the analysis and feature extraction of fault signals in power distribution systems. Specifically, in the fault identification process, the voltage and current transient waveforms of the power distribution network first need to be decomposed into signal components of different frequency bands through wavelet decomposition. Then, fault feature vectors are constructed based on the coefficient energy, entropy value, or statistical characteristics of each decomposition layer. Finally, the vectors are input into a classifier (such as support vector machine, neural network, or fuzzy logic system) for fault type identification.

[0004] In this process, the number of wavelet decomposition layers is a key controllable parameter: if the number of decomposition layers is too low, the fault features are difficult to separate sufficiently, resulting in incomplete feature extraction, which reduces the distinguishability of different types of faults and increases the risk of misjudgment and missed judgment; if the number of decomposition layers is too high, the signal will be decomposed at too many scales, which not only increases redundant features, but also weakens the salience of fault features, further increasing the training and computational overhead of the classifier and reducing the real-time performance of the system.

[0005] Therefore, the wavelet decomposition level directly affects the signal preprocessing and feature extraction stages in fault identification methods. It relates to both the quality of the feature set and the accuracy and efficiency of fault identification. Since both excessively high and excessively low wavelet decomposition levels have drawbacks, how to rationally select the wavelet decomposition level has become a key issue in fault identification of distributed intelligent power distribution systems. Summary of the Invention

[0006] This invention provides a fault identification method for a distributed intelligent power distribution system. Its main purpose is to solve the problems that result in low quality of signal feature sets and low accuracy and efficiency of obstacle identification due to defects caused by excessively high or low wavelet decomposition layer values.

[0007] In a first aspect, to achieve the above objectives, the present invention provides a fault identification method for a distributed intelligent power distribution system, comprising: The historical power waveforms and several historical wavelet decomposition levels of the power distribution system are obtained. Based on the historical wavelet decomposition levels, the historical power waveforms are decomposed into wavelets and denoised with thresholds to obtain a set of denoised wavelet coefficients. The residual entropy corresponding to the historical wavelet decomposition level is calculated based on the set of denoised wavelet coefficients to obtain an entropy sequence, and the historical wavelet decomposition level corresponding to the target entropy value in the entropy sequence is selected as the initial wavelet decomposition level. Discrete normal perturbation is applied to the initial wavelet decomposition level to obtain several candidate wavelet decomposition levels; The historical power waveform is decomposed into wavelet decomposition to each of the candidate wavelet decomposition levels, and the root mean square value and spectral energy distribution of each candidate wavelet decomposition level are extracted. Obtain the fault type label corresponding to the historical power waveform, generate fault identification results based on the root mean square value and the spectral energy distribution, and calculate the analysis accuracy of each layer based on the fault identification results and the fault type label; The target wavelet decomposition level is selected based on the analysis accuracy, and the target wavelet decomposition level is used to identify faults in the distributed power distribution system to obtain the target fault identification result.

[0008] Secondly, the present invention also provides a fault identification system for a distributed intelligent power distribution system, the system comprising: The historical power waveform decomposition module is used to obtain the historical power waveform of the power distribution system and a number of historical wavelet decomposition levels. Based on the historical wavelet decomposition levels, the historical power waveform is decomposed into wavelet and thresholded denoising to obtain a set of denoised wavelet coefficients. The initial wavelet decomposition layer selection module is used to calculate the residual entropy corresponding to the historical wavelet decomposition layer based on the denoised wavelet coefficient set, obtain the entropy sequence, and select the historical wavelet decomposition layer corresponding to the target entropy value in the entropy sequence as the initial wavelet decomposition layer. The candidate wavelet decomposition level generation module is used to perform discrete normal perturbation on the initial wavelet decomposition level to obtain several candidate wavelet decomposition levels. The wavelet decomposition layer feature extraction module is used to decompose the historical power waveform into each of the candidate wavelet decomposition layers, and extract the root mean square value and spectral energy distribution of each candidate wavelet decomposition layer. The fault feature analysis module is used to obtain the fault type label corresponding to the historical power waveform, generate fault identification results based on the root mean square value and the spectrum energy distribution, and calculate the analysis accuracy of each layer based on the fault identification results and the fault type label. The target wavelet decomposition level application module is used to select the target wavelet decomposition level based on the analysis accuracy, and use the target wavelet decomposition level to identify faults in the distributed power distribution system to obtain the target fault identification result.

[0009] This invention introduces a residual entropy optimization mechanism and a discrete normal perturbation strategy to adaptively select the optimal wavelet decomposition level, effectively overcoming the limitation of traditional methods where level selection relies on experience. This significantly improves the discriminative power and accuracy of fault features. Employing an entropy sequence-based level selection and perturbation generation mechanism avoids the extensive computations required for traditional manual trial and error, achieving rapid and automatic optimization of the wavelet decomposition level, significantly reducing computational costs and improving processing efficiency. By fusing multi-dimensional features (root mean square value, spectral energy distribution) and machine learning models, it can adapt to different operating conditions and fault types in power distribution systems, exhibiting strong generalization ability and robustness. The optimized wavelet decomposition level can more accurately capture fault features in power waveforms, reduce noise interference, and generate a more discriminative feature set, providing a reliable data foundation for subsequent fault diagnosis. Attached Figure Description

[0010] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments of the present invention will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0011] Figure 1 A flowchart illustrating a fault identification method for a distributed intelligent power distribution system according to an embodiment of the present invention; Figure 2 This is a schematic diagram of a fault identification system for a distributed intelligent power distribution system according to an embodiment of the present invention; The objectives, features, and advantages of this invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0012] To enable those skilled in the art to better understand the technical solutions of this disclosure, and to fully understand and implement the process of how this disclosure applies technical means to solve technical problems and achieve corresponding technical effects, the technical solutions in the embodiments of this disclosure will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this disclosure, not all embodiments. The embodiments of this disclosure and the various features within them can be combined with each other without conflict, and the resulting technical solutions are all within the protection scope of this disclosure. All other embodiments obtained by those skilled in the art based on the embodiments of this disclosure without creative effort should fall within the protection scope of this disclosure.

[0013] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this disclosure are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this disclosure described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0014] This application provides a fault identification method for a distributed intelligent power distribution system. This method can be executed by software or hardware installed on terminal devices or server-side devices. The server-side includes, but is not limited to, a single server, a server cluster, a cloud server, or a cloud server cluster. The server can be an independent server or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDNs), and big data and artificial intelligence platforms.

[0015] Reference Figure 1 The diagram shown is a flowchart illustrating a fault identification method for a distributed intelligent power distribution system according to an embodiment of the present invention. In this embodiment, the fault identification method for a distributed intelligent power distribution system includes: S1. Based on historical power waveforms, perform wavelet decomposition and threshold denoising to obtain a set of denoised wavelet coefficients. Specifically, obtain the historical power waveforms of the power distribution system and several historical wavelet decomposition levels, and perform wavelet decomposition and threshold denoising on the historical power waveforms according to the historical wavelet decomposition levels to obtain a set of denoised wavelet coefficients.

[0016] In this embodiment of the invention, historical power waveform refers to waveform data of voltage or current changes over time collected and stored during the past operation of the power distribution system. Historical wavelet decomposition levels refer to various wavelet decomposition levels preset in previous experiments or analyses, used to attempt signal decomposition at different scales.

[0017] For each historical wavelet decomposition level, the historical power waveform is decomposed into multi-scale approximation coefficients and detail coefficients. A threshold denoising method is applied to the detail coefficients to remove noise interference, resulting in a set of denoised wavelet coefficients corresponding to different decomposition levels.

[0018] Specifically, the step of performing wavelet decomposition and threshold denoising on the historical power waveform based on the historical wavelet decomposition level to obtain a set of denoised wavelet coefficients includes: Obtain waveform windows, and perform mean-removal processing on each waveform window of the historical power waveform to obtain a standard power waveform; One of the historical wavelet decomposition layers is selected as the initial candidate layer; The standard power waveform is decomposed into discrete wavelet layers to the initial candidate layer using a preset mother wavelet, resulting in a set of approximation coefficients and detail coefficients. Calculate the noise standard deviation of the set of detail coefficients for each of the initial candidate layers; Obtain the resampling window length, and calculate the denoising threshold based on the noise standard deviation and the resampling window length; The approximation coefficients and the set of detail coefficients are subjected to soft thresholding based on the denoising threshold to obtain a set of denoised wavelet coefficients.

[0019] In detail, when analyzing historical power waveforms, the long-term original signal is divided into multiple waveform windows (usually fixed-length time segments, such as 256 points or 512 points). The mean value within each waveform window is subtracted to eliminate DC bias and baseline drift, resulting in a standard power waveform with zero mean, which is more conducive to wavelet decomposition and noise analysis.

[0020] Select a suitable mother wavelet function (such as Daubechies wavelet db4, Symlet wavelet, etc.) to perform discrete wavelet transform on the standard power waveform, decompose it to the current historical wavelet decomposition level, and the decomposition results include: approximation coefficients: representing the low-frequency part of the signal and preserving the main trend; detail coefficients: representing the high-frequency details of the signal in different frequency bands, including noise and transient features.

[0021] For each set of detail coefficients, the standard deviation of the noise is estimated using the median absolute deviation:

[0022] in, This represents the noise standard deviation for the i-th historical wavelet decomposition level. This represents the set of detail coefficients for the i-th historical wavelet decomposition level.

[0023] Let the resampling window length be N, the denoising threshold is usually set using Donoho's universal threshold formula:

[0024] in, This represents the noise standard deviation for the i-th historical wavelet decomposition level. Indicates the length of the resampling window.

[0025] A soft thresholding function is applied to the set of detail coefficients obtained from the decomposition to denoise them. Coefficients smaller than the threshold are compressed to 0, and coefficients larger than the threshold are shrunk.

[0026] in, This represents the k-th element of the set of detail coefficients for the i-th historical wavelet decomposition level. Represents a symbolic function. Indicates the threshold. This represents the k-th element of the set of denoised wavelet coefficients for the i-th historical wavelet decomposition level.

[0027] By decomposing power waveforms based on historical wavelet decomposition levels and applying threshold denoising, the rationality of decomposition can be improved by fully utilizing historical experience in determining the number of levels, avoiding distortion caused by blindly selecting the number of levels. Furthermore, by adaptively setting the threshold using noise standard deviation and resampling window length, noise and valid signals can be distinguished more accurately. This allows for the removal of random interference while preserving the true characteristics of the power waveform as much as possible, resulting in a more stable and reliable set of denoised wavelet coefficients. This provides a more accurate data foundation for subsequent power quality analysis and anomaly detection.

[0028] S2. Calculate the residual entropy of the wavelet coefficients to generate an entropy sequence and determine the initial wavelet decomposition level. Specifically, calculate the residual entropy corresponding to the historical wavelet decomposition level based on the denoised wavelet coefficient set to obtain an entropy sequence, and select the historical wavelet decomposition level corresponding to the target entropy value in the entropy sequence as the initial wavelet decomposition level.

[0029] In this embodiment of the invention, residual entropy is an indicator describing the complexity and randomness of the residual signal (the difference between the original signal and the reconstructed signal) after denoising. A smaller entropy value indicates better noise removal and a more concise and ordered signal. Entropy sequence is a set of entropy values ​​obtained by calculating the residual entropy for each different decomposition level, reflecting the differences in denoising effect at each level.

[0030] Using the set of denoised wavelet coefficients, the corresponding residual entropy is calculated at each historical wavelet decomposition level to form an entropy sequence. The target entropy value that best distinguishes between signal and noise is selected from the entropy sequence, and the decomposition level corresponding to the target entropy value is selected as the initial wavelet decomposition level.

[0031] Specifically, the step of calculating the residual entropy corresponding to the historical wavelet decomposition level based on the denoised wavelet coefficient set to obtain the entropy sequence includes: Perform inverse wavelet transform on the denoised wavelet coefficient set to obtain the denoised reconstructed signal; The residual signal for each of the historical wavelet decomposition levels is calculated based on the standard power waveform and the denoised reconstructed signal. The number of bins is calculated based on the resampling window length, and the residual signal is used to calculate the signal frequency according to the number of bins. A probability distribution is generated based on the number of boxes and the signal frequency. The residual entropy of each standard power waveform in each of the historical wavelet decomposition levels is calculated based on the probability distribution. The average of all residual entropies for each of the historical wavelet decomposition levels is taken as the representative entropy, and all the representative entropies are summarized into an entropy sequence.

[0032] In detail, after wavelet decomposition and threshold denoising, the resulting set of approximate coefficients and detail coefficients are still in the wavelet domain. In order to recover the time domain signal, it is necessary to perform inverse wavelet transform on the denoised wavelet coefficient set, reconstruct and synthesize the coefficients of each layer layer by layer, and finally obtain a denoised reconstructed signal that has removed most of the noise but still retains the main features and energy.

[0033] The residual signal represents the difference between the original standard signal and the denoised signal, reflecting how much unremoved noise or distortion remains after wavelet denoising at a certain decomposition level. The calculation formula is as follows:

[0034] in, Represents standard power waveforms. This represents the denoised and reconstructed signal. This represents the residual signal.

[0035] Number of bins: Determined based on the window length, used to divide the range of residual signal values ​​into intervals. The frequency of the residual signal falling within each bin interval is counted to obtain the signal frequency of each bin. After frequency counting, the probability distribution can be obtained, calculated as follows:

[0036] in, This represents the signal frequency of the j-th sub-box. This represents the total number of sampling points for the residual signal. Let represent the probability of the j-th bin.

[0037] Residual entropy describes the uncertainty and randomness of the residual signal. The lower the entropy, the better the noise removal effect and the more ordered the signal. The calculation formula is shown below:

[0038] in, Let represent the probability of the j-th bin. Indicates the number of boxes. This represents the residual entropy.

[0039] For each historical wavelet decomposition level, multiple residual entropy values ​​for waveform windows are obtained. To obtain an overall evaluation of the decomposition effect of the historical wavelet decomposition levels, the residual entropy of all windows in the historical wavelet decomposition levels is averaged to obtain a representative entropy, reflecting the overall signal complexity or uncertainty after denoising. The representative entropies of all historical wavelet decomposition levels are arranged sequentially to form an entropy sequence, which is used to analyze the denoising effect and signal feature preservation of each decomposition level, providing a basis for selecting the optimal initial wavelet decomposition level.

[0040] Inverse wavelet transform restores the denoised wavelet coefficients to the time domain signal, enabling the residual signal to accurately reflect the difference between the noise and the original signal. By calculating the residual entropy through binning and probability distribution, the uncertainty of the signal under each decomposition layer is quantitatively described. By taking the average to form a representative entropy and summarizing it into an entropy sequence, the performance of each decomposition layer can be intuitively compared, helping to select the optimal number of decomposition layers, thereby improving the effectiveness of feature extraction and the accuracy and stability of subsequent fault identification.

[0041] Specifically, the step of selecting the historical wavelet decomposition level corresponding to the target entropy value in the entropy sequence as the initial wavelet decomposition level includes: The total number of historical wavelet decomposition layers is counted, and it is determined whether the total number of decomposition layers is less than a preset number of layers. If the total number of decomposition layers is less than the preset number of layers, then the minimum initial candidate layer corresponding to the minimum representative entropy in the entropy sequence is taken as the initial wavelet decomposition layer. If the total number of decomposition layers is greater than or equal to the preset number of layers, then calculate the first difference of the representative entropy corresponding to each of the initial candidate layers; Calculate the second-order difference based on the first-order difference, and select the representative entropy corresponding to the smallest second-order difference as the target entropy value; The initial candidate layer number corresponding to the target entropy value is used as the initial wavelet decomposition layer number.

[0042] In detail, the total number of historical wavelet decomposition layers is counted and compared with a layer threshold. If the total number of layers is less than the threshold, the smallest initial candidate layer corresponding to the smallest representative entropy in the entropy sequence is directly selected as the initial wavelet decomposition layer. If the total number of layers is greater than or equal to the threshold, the optimal layer is determined using the entropy change trend. The first difference of the representative entropy sequence is calculated to reflect the rate of change of entropy between adjacent layers, and the calculation formula is shown below:

[0043] in, The representative entropy represents the (i+1)th historical wavelet decomposition level. The representative entropy represents the number of the i-th historical wavelet decomposition layer. Let represent the first-order difference of the i-th historical wavelet decomposition level.

[0044] Further calculation of the second-order difference is used to reflect the acceleration or inflection point of the entropy change curve. The representative entropy corresponding to the minimum value of the second-order difference is found, which represents the point where the entropy curve changes most drastically, i.e., the optimal trade-off between signal denoising effect and information preservation. The calculation formula is as follows:

[0045] in, This represents the first-order difference of the i-th historical wavelet decomposition level. This represents the first-order difference of the (i+1)th historical wavelet decomposition level. Let represent the second difference of the i-th historical wavelet decomposition level.

[0046] This method intelligently selects the initial wavelet decomposition level for different historical wavelet decomposition levels. When the total number of decomposition levels is small, the level with the lowest entropy is directly selected to ensure optimal denoising and signal order. When the number of decomposition levels is large, the entropy curve trend is analyzed using first-order and second-order difference analysis to find the level corresponding to the inflection point of the entropy curve, balancing denoising effect and signal feature preservation. This method considers both the absolute number of levels and the dynamic information of entropy changes, making the selection of the initial decomposition level more reasonable and improving the accuracy and stability of subsequent feature extraction and fault identification.

[0047] S3. The initial wavelet decomposition layer is subjected to discrete normal perturbation to obtain several candidate wavelet decomposition layers.

[0048] In this embodiment of the invention, a standard deviation parameter is set with the initial wavelet decomposition level as the center. Several integer values ​​are randomly generated under a normal distribution and these integer values ​​are limited to a reasonable range of levels (e.g., from 1 to the maximum decomposition level). Each perturbation will generate a new candidate level near the initial wavelet decomposition level. After multiple samplings, a set of candidate wavelet decomposition levels can be obtained.

[0049] Specifically, the step of performing a discrete normal perturbation on the initial wavelet decomposition level to obtain several candidate wavelet decomposition levels includes: Obtain the disturbance intensity factor and the normal distribution standard deviation, and generate random disturbance values ​​based on the disturbance intensity factor and the normal distribution standard deviation; Based on the random perturbation value, the initial wavelet decomposition layer number is subjected to discrete normal perturbation to obtain a set of candidate layers to be pruned; Boundary pruning and rounding are performed on the set of candidate wavelet decomposition layers to obtain several candidate wavelet decomposition layers.

[0050] In detail, the perturbation strength factor determines the standard deviation of the normal distribution, i.e., the magnitude of the perturbation. A random variable is sampled from the standard normal distribution N(0,1) according to the perturbation strength factor and scaled accordingly:

[0051] in, Indicates random perturbation value, This represents the standard deviation of a normal distribution. This represents the perturbation intensity factor. It ensures that the magnitude of random perturbation values ​​is controlled and exhibits normal distribution characteristics.

[0052] Add the initial wavelet decomposition level to the random perturbation value to obtain the set of candidate levels to be pruned:

[0053] in, Indicates the initial wavelet decomposition level. Indicates random perturbation value, This indicates the number of candidate layers to be pruned.

[0054] Determine whether each candidate layer to be pruned exceeds the allowed range. If it does, prune the candidate layers to the boundary value, discard the decimal part of all candidate layers to be pruned or round it to the nearest integer, and finally obtain a set of candidate wavelet decomposition layers that meet the range constraints.

[0055] By introducing a perturbation intensity factor and the standard deviation of the normal distribution, a set of candidate layers with randomness and diversity is generated based on the initial wavelet decomposition layers. This avoids the local optimum problem caused by choosing a single layer. Through boundary pruning and rounding, it is ensured that all candidate layers are within a reasonable integer range. This preserves the exploratory nature while ensuring the effectiveness and usability of the layers, thereby increasing the probability of finding the optimal wavelet decomposition layer in the subsequent optimization search.

[0056] S4. Decompose the historical power waveform into each candidate layer and extract the fault features of each layer. The fault features include the root mean square value and the spectral energy distribution features.

[0057] In this embodiment of the invention, the root mean square (RMS) value refers to a statistical measure of the effective value of a signal, reflecting the magnitude of the signal energy, and is often used to measure the intensity or power characteristics of power waveforms. Spectral energy distribution refers to the energy distribution of a signal across different frequency ranges, used to analyze the main frequency components and energy concentration areas of the signal.

[0058] Historical power waveforms are sequentially decomposed according to candidate wavelet decomposition levels to obtain sub-signals at different frequencies and scales. At each candidate decomposition level, the root mean square (RMS) value of the sub-signal is calculated to measure signal energy intensity, and the corresponding spectral energy distribution is extracted to reflect the signal's energy characteristics across various frequency bands. By comparing the RMS values ​​and spectral energy distributions at different levels, it is possible to analyze which decomposition level is more suitable for characterizing power waveform features.

[0059] Specifically, the step of decomposing the historical power waveform into wavelet decomposition levels for each candidate wavelet decomposition level, and extracting the root mean square value and spectral energy distribution for each candidate wavelet decomposition level, includes: Each of the candidate wavelet decomposition layers is selected as the target candidate layer; The historical power waveform is decomposed into the target candidate layer using wavelet decomposition to obtain the candidate wavelet coefficient set of the target candidate layer; The total number of coefficients in the candidate wavelet coefficient set is counted, and the root mean square value of the candidate wavelet coefficient set is calculated based on the total number of coefficients. Wavelet reconstruction is performed on the candidate wavelet coefficient set to obtain the time-domain candidate signal; The power spectrum is obtained by performing a fast Fourier transform on the time-domain candidate signal. Obtain the frequency range and divide the frequency range into low-frequency band, mid-frequency band and high-frequency band; The power spectrum is integrated according to the low-frequency band, the mid-frequency band, and the high-frequency band respectively to obtain the low-frequency band energy distribution, the mid-frequency band energy distribution, and the high-frequency band energy distribution; The low-frequency energy distribution, the mid-frequency energy distribution, and the high-frequency energy distribution are combined into a spectrum energy distribution.

[0060] In detail, a preset number of candidate wavelet decomposition levels are selected one by one as target candidate levels, and wavelet decomposition is performed on the waveform until the target candidate level is reached, resulting in a set of candidate wavelet coefficients (including approximation coefficients and detail coefficients) corresponding to the target candidate level. The candidate wavelet coefficient set is statistically analyzed to determine the total number of all coefficients, and the root mean square (RMS) value is calculated using these coefficients. This RMS value reflects the overall strength of the signal energy after wavelet decomposition at the target candidate level. The calculation formula is shown below:

[0061] in, Indicates the total number of coefficients. This represents the square of the a-th coefficient in the candidate wavelet coefficient set.

[0062] The approximation coefficients and detail coefficients in the candidate wavelet coefficient set are gradually reversed according to the corresponding wavelet basis functions and decomposition levels to restore the time-domain signal corresponding to the original waveform. The components of the signal at various scales and frequency bands are then superimposed to restore a complete time-domain candidate signal for subsequent frequency domain or feature analysis.

[0063] The time-domain candidate signal is converted from a time series to a frequency domain representation after being processed by the Fast Fourier Transform. The waveform that originally changed with time is decomposed into the superposition relationship of different frequency components, and the energy of each frequency component is intuitively obtained and presented in the form of power spectrum, thus reflecting the energy distribution characteristics of the signal in the frequency domain.

[0064] The frequency range of the signal to be analyzed is determined and divided into low-frequency, mid-frequency, and high-frequency bands. Energy is statistically analyzed in each frequency band. The power spectrum is integrated within each band (i.e., the power spectrum values ​​for that band are summed) to obtain the energy distributions for the low-frequency, mid-frequency, and high-frequency bands. These distributions reflect the degree of energy concentration of the signal in different frequency ranges. The energy distributions of the three bands are then summarized into a spectral energy distribution vector, which describes the overall frequency energy characteristics of the signal. The calculation formula is shown below:

[0065]

[0066]

[0067]

[0068] in, This represents the power value of the power spectrum at frequency f. Indicates the energy distribution in the low-frequency band. Indicates the energy distribution in the mid-frequency band. Indicates the energy distribution in the high-frequency band. This represents the energy distribution of the spectrum.

[0069] By performing layer-by-layer wavelet decomposition on historical power waveforms and extracting the root mean square (RMS) value and spectral energy distribution of each candidate decomposition layer, the energy characteristics of the signal at different frequency scales can be comprehensively characterized. The RMS value reflects the overall strength of the signal at each level, while the spectral energy distribution reveals the energy concentration of the signal in the low, medium, and high frequency bands. This approach takes into account both time and frequency domain information, making the characteristic analysis of power waveforms more accurate and comprehensive, which is beneficial for further fault identification.

[0070] S5. Obtain the fault type label corresponding to the historical power waveform, generate fault identification results based on the root mean square value and the spectrum energy distribution characteristics, and calculate the analysis accuracy of each layer based on the fault identification results and the fault type label.

[0071] In this embodiment of the invention, the fault type label is a mark or tag used to identify the fault category corresponding to the historical power waveform. In a power system, different fault types (such as short circuit, ground fault, voltage drop, current anomaly, etc.) will produce different characteristic manifestations on the voltage and current waveforms.

[0072] The actual fault type labels corresponding to historical power waveforms are obtained, and the root mean square value and spectral energy distribution extracted from each candidate wavelet decomposition layer are used as features to input the random forest model to generate fault identification results. The identification results are compared with the actual fault type labels, and the proportion of correct identification is statistically analyzed to calculate the analysis accuracy of each decomposition layer.

[0073] Specifically, generating the fault identification result based on the root mean square value and the spectral energy distribution includes: Candidate feature vectors are generated based on the root mean square value and the spectral energy distribution; The initial fault characteristics of the candidate feature vectors are identified using a pre-defined random forest model; The initial fault features are fused by voting to obtain historical fault features and their confidence levels. Historical fault features corresponding to confidence levels greater than or equal to the confidence threshold are selected, and fault identification results for each candidate wavelet decomposition level are generated based on the selected historical fault features.

[0074] In detail, the root mean square value and spectral energy distribution extracted from each candidate wavelet decomposition layer are combined to form candidate feature vectors, containing the energy characteristics of the signal at different frequency scales. These feature vectors are input into a pre-trained random forest model. The model sequentially passes the feature vectors through multiple decision trees within it. Each tree judges the feature value based on the splitting conditions at its nodes, progressing from the root node to the leaf nodes, obtaining the classification result for each tree. The random forest statistically analyzes the voting results of all trees, generating initial fault features and confidence values ​​corresponding to the feature vectors through majority voting or probability weighting. Based on a set confidence threshold, historical fault features with confidence values ​​greater than or equal to the threshold are selected, retaining high-reliability features for subsequent analysis or model optimization.

[0075] In detail, the step of calculating the analysis accuracy of each layer based on the fault identification result and the fault type label includes: The fault identification result is compared with the fault type label to determine whether the fault identification result is consistent with the fault type label; If the fault identification result is inconsistent with the fault type label, then the fault identification of the historical power waveform is marked as incorrect; If the fault identification result matches the fault type label, then the fault identification of the historical power waveform is marked as correct, and the number of correct identifications is counted. Obtain the total number of fault identification results, and calculate the analysis accuracy of each candidate wavelet decomposition layer based on the number of correct identifications and the total number of identifications.

[0076] In detail, the fault identification result of each historical power waveform is compared with the corresponding actual fault type label to determine whether they match. If the identification result does not match the actual label, the waveform is marked as an incorrect identification; if they match, it is marked as a correct identification, and the number of correct identifications is accumulated. The total number of identifications for all historical waveforms is counted, and combined with the number of correct identifications, the analysis accuracy of each candidate wavelet decomposition level is calculated, i.e., the proportion of correct identifications, thereby evaluating the effectiveness of the features of that level in fault identification. The calculation formula is as follows:

[0077] in, Indicates the candidate wavelet decomposition level. The analysis accuracy is represented by the number of candidate wavelet decomposition levels. This represents the number of correctly identified values ​​under the candidate wavelet decomposition level. This represents the total number of recognitions under the candidate wavelet decomposition level.

[0078] By generating candidate feature vectors from root mean square values ​​and spectral energy distributions and inputting them into a pre-defined random forest model for initial fault identification, and then obtaining highly reliable historical fault features through voting fusion and confidence level filtering, the reliability of fault identification can be effectively improved. The identification results are compared with actual fault type labels, and the number of correctly identified features at each wavelet decomposition layer is counted, and the analysis accuracy is calculated, allowing for a quantitative evaluation of the feature effectiveness of each candidate decomposition layer. The overall method considers a closed loop of feature extraction, model prediction, and result verification, making the fault identification process both accurate and controllable. This provides a reliable basis for fault detection and diagnosis in power systems and helps select the most suitable decomposition layer for signal analysis, thereby improving identification performance and system security.

[0079] S6. Based on the analysis accuracy, select the target wavelet decomposition level and use the target wavelet decomposition level to perform fault identification on the distributed power distribution system to obtain the target fault identification result.

[0080] In this embodiment of the invention, the target wavelet decomposition layer with the best performance is selected based on the accuracy, which can most effectively characterize the power waveform features. The target wavelet decomposition layer is used to perform wavelet decomposition on the power signal of the distributed power distribution system, and the corresponding feature vector is extracted as input for fault identification. Finally, the target fault identification result is generated, ensuring that the selected decomposition layer has high reliability in both signal feature expression and fault discrimination, thereby improving the overall identification accuracy.

[0081] In detail, the step of selecting the target wavelet decomposition level based on the analysis accuracy, and using the target wavelet decomposition level to perform fault identification on the distributed power distribution system to obtain the target fault identification result includes: Based on the analysis accuracy, the candidate wavelet decomposition layer corresponding to the highest analysis accuracy is selected; The selected candidate wavelet decomposition levels are used as the target wavelet decomposition levels. The waveform to be detected during the operation of the distributed power distribution system is obtained, and the waveform to be detected is decomposed into wavelet to the target wavelet decomposition level to obtain the target wavelet coefficient set. Calculate the root mean square value and the target spectral energy distribution based on the target wavelet coefficient set; The random forest model is used to perform anomaly feature analysis on the root mean square value of the target and the energy distribution of the target spectrum to obtain the target fault features of the waveform to be detected. Target fault identification results are generated based on the target fault characteristics.

[0082] In detail, the analysis accuracy of each candidate wavelet decomposition level is compared, and the level with the highest analysis accuracy is selected as the target wavelet decomposition level. During the operation of the distributed power distribution system, real-time power waveforms to be detected are acquired. These waveforms contain potential fault information. The waveforms to be detected are input into wavelet decomposition processing and decomposed according to the selected target wavelet decomposition level. This decomposes the signal into a set of approximation coefficients and detail coefficients corresponding to the target wavelet decomposition level, i.e., the target wavelet coefficient set. Through this process, it is ensured that the extracted signal features come from the most effective target wavelet decomposition level in historical verification, thereby more accurately reflecting the fault characteristics in the real-time waveform and providing reliable input for subsequent fault identification.

[0083] The target root mean square value and target spectral energy distribution are calculated for the target wavelet coefficient set, reflecting the characteristics of the waveform to be detected in terms of energy intensity and frequency distribution, respectively. The random forest model will pass the feature vector through multiple decision trees in sequence. Each tree judges whether the feature value matches the fault mode during training based on the splitting condition at the node. The judgment results of each tree are voted or weighted by probability to generate an abnormal feature prediction of the waveform to be detected. The random forest outputs the target fault features and confidence of the waveform to be detected, thereby identifying the possible fault types and abnormal behaviors in the waveform. Based on the target fault features and confidence, the target fault identification result of the waveform to be detected in the distributed power distribution system is generated.

[0084] By selecting the target wavelet decomposition level corresponding to the highest accuracy based on the analysis accuracy, and using this target wavelet decomposition level to perform wavelet decomposition on the waveform to be detected in the distributed power distribution system, it is ensured that feature extraction is based on the most effective decomposition level in historical validation, thereby obtaining a high-quality set of target wavelet coefficients. Further calculation of the target root mean square value and target spectral energy distribution, and inputting them into a random forest model for anomaly feature analysis, allows for accurate identification of fault features in the waveform. This fully utilizes both time-domain and frequency-domain information. Furthermore, using the selected target wavelet decomposition level for wavelet decomposition of the waveform to be detected not only preserves the most discriminative frequency components and energy features but also reduces the interference of redundant information on the model's judgment. This enables the random forest model to more accurately capture fault features in the waveform during anomaly feature analysis, thereby generating more stable and reliable target fault identification results.

[0085] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0086] like Figure 2 The diagram shown is a functional block diagram of a fault identification system for a distributed intelligent power distribution system provided in an embodiment of the present invention.

[0087] This disclosure provides a fault identification system for a distributed intelligent power distribution system, which corresponds one-to-one with the fault identification method for a distributed intelligent power distribution system described in the above embodiments. For example... Figure 2 As shown, the fault identification system 100 of the distributed intelligent power distribution system includes a historical power waveform decomposition module 101, an initial wavelet decomposition level selection module 102, a candidate wavelet decomposition level generation module 103, a wavelet decomposition level feature extraction module 104, a fault feature analysis module 105, and a target wavelet decomposition level application module 106. Detailed descriptions of each functional module are as follows: The historical power waveform decomposition module 101 is used to acquire the historical power waveform of the power distribution system and a number of historical wavelet decomposition levels, and to perform wavelet decomposition and threshold denoising on the historical power waveform according to the historical wavelet decomposition levels to obtain a set of denoised wavelet coefficients. The initial wavelet decomposition layer selection module 102 is used to calculate the residual entropy corresponding to the historical wavelet decomposition layer based on the denoised wavelet coefficient set, obtain an entropy sequence, and select the historical wavelet decomposition layer corresponding to the target entropy value in the entropy sequence as the initial wavelet decomposition layer. The candidate wavelet decomposition layer generation module 103 is used to perform discrete normal perturbation on the initial wavelet decomposition layer to obtain a number of candidate wavelet decomposition layers. The wavelet decomposition layer feature extraction module 104 is used to decompose the historical power waveform into each of the candidate wavelet decomposition layers, and extract the root mean square value and spectral energy distribution of each candidate wavelet decomposition layer. The fault feature analysis module 105 is used to obtain the fault type label corresponding to the historical power waveform, generate fault identification results based on the root mean square value and the spectrum energy distribution, and calculate the analysis accuracy of each layer based on the fault identification results and the fault type label. The target wavelet decomposition level application module 106 is used to select the target wavelet decomposition level based on the analysis accuracy, and use the target wavelet decomposition level to perform fault identification on the distributed power distribution system to obtain the target fault identification result.

[0088] In one embodiment, the historical power waveform decomposition module 101 performs wavelet decomposition and threshold denoising on the historical power waveform according to the historical wavelet decomposition level to obtain a set of denoised wavelet coefficients, which are used for: Obtain waveform windows, and perform mean-removal processing on each waveform window of the historical power waveform to obtain a standard power waveform; One of the historical wavelet decomposition layers is selected as the initial candidate layer; The standard power waveform is decomposed into discrete wavelet layers to the initial candidate layer using a preset mother wavelet, resulting in a set of approximation coefficients and detail coefficients. Calculate the noise standard deviation of the set of detail coefficients for each of the initial candidate layers; Obtain the resampling window length, and calculate the denoising threshold based on the noise standard deviation and the resampling window length; The approximation coefficients and the set of detail coefficients are subjected to soft thresholding based on the denoising threshold to obtain a set of denoised wavelet coefficients.

[0089] In one embodiment, the initial wavelet decomposition level selection module 102 calculates the residual entropy corresponding to the historical wavelet decomposition level based on the denoised wavelet coefficient set, and obtains an entropy sequence for: Perform inverse wavelet transform on the denoised wavelet coefficient set to obtain the denoised reconstructed signal; The residual signal for each of the historical wavelet decomposition levels is calculated based on the standard power waveform and the denoised reconstructed signal. The number of bins is calculated based on the resampling window length, and the residual signal is used to calculate the signal frequency according to the number of bins. A probability distribution is generated based on the number of boxes and the signal frequency. The residual entropy of each standard power waveform in each of the historical wavelet decomposition levels is calculated based on the probability distribution. The average of all residual entropies for each of the historical wavelet decomposition levels is taken as the representative entropy, and all the representative entropies are summarized into an entropy sequence.

[0090] In one embodiment, the initial wavelet decomposition level filtering module 102, when performing the filtering of historical wavelet decomposition levels corresponding to the target entropy value in the entropy sequence as the initial wavelet decomposition level, is used for: The total number of historical wavelet decomposition layers is counted, and it is determined whether the total number of decomposition layers is less than a preset number of layers. If the total number of decomposition layers is less than the preset number of layers, then the minimum initial candidate layer corresponding to the minimum representative entropy in the entropy sequence is taken as the initial wavelet decomposition layer. If the total number of decomposition layers is greater than or equal to the preset number of layers, then calculate the first difference of the representative entropy corresponding to each of the initial candidate layers; Calculate the second-order difference based on the first-order difference, and select the representative entropy corresponding to the smallest second-order difference as the target entropy value; The initial candidate layer number corresponding to the target entropy value is used as the initial wavelet decomposition layer number.

[0091] In one embodiment, the candidate wavelet decomposition level generation module 103 performs a discrete normal perturbation on the initial wavelet decomposition level to obtain a plurality of candidate wavelet decomposition levels, for the purpose of: Obtain the disturbance intensity factor and the normal distribution standard deviation, and generate random disturbance values ​​based on the disturbance intensity factor and the normal distribution standard deviation; Based on the random perturbation value, the initial wavelet decomposition layer number is subjected to discrete normal perturbation to obtain a set of candidate layers to be pruned; Boundary pruning and rounding are performed on the set of candidate wavelet decomposition layers to obtain several candidate wavelet decomposition layers.

[0092] In one embodiment, the wavelet decomposition level feature extraction module 104 performs wavelet decomposition of the historical power waveform to each of the candidate wavelet decomposition levels, and extracts the root mean square value and spectral energy distribution of each candidate wavelet decomposition level, for the following purposes: Each of the candidate wavelet decomposition layers is selected as the target candidate layer; The historical power waveform is decomposed into the target candidate layer using wavelet decomposition to obtain the candidate wavelet coefficient set of the target candidate layer; The total number of coefficients in the candidate wavelet coefficient set is counted, and the root mean square value of the candidate wavelet coefficient set is calculated based on the total number of coefficients. Wavelet reconstruction is performed on the candidate wavelet coefficient set to obtain the time-domain candidate signal; The power spectrum is obtained by performing a fast Fourier transform on the time-domain candidate signal. Obtain the frequency range and divide the frequency range into low-frequency band, mid-frequency band and high-frequency band; The power spectrum is integrated according to the low-frequency band, the mid-frequency band, and the high-frequency band respectively to obtain the low-frequency band energy distribution, the mid-frequency band energy distribution, and the high-frequency band energy distribution; The low-frequency energy distribution, the mid-frequency energy distribution, and the high-frequency energy distribution are combined into a spectrum energy distribution.

[0093] In one embodiment, the fault feature analysis module 105 generates a fault identification result based on the root mean square value and the spectral energy distribution, for the following purposes: Candidate feature vectors are generated based on the root mean square value and the spectral energy distribution; The initial fault characteristics of the candidate feature vectors are identified using a pre-defined random forest model; The initial fault features are fused by voting to obtain historical fault features and their confidence levels. Historical fault features corresponding to confidence levels greater than or equal to the confidence threshold are selected, and fault identification results for each candidate wavelet decomposition level are generated based on the selected historical fault features.

[0094] In one embodiment, the fault feature analysis module 105 performs the calculation of the analysis accuracy of each layer based on the fault identification result and the fault type label, for the following purposes: The fault identification result is compared with the fault type label to determine whether the fault identification result is consistent with the fault type label; If the fault identification result is inconsistent with the fault type label, then the fault identification of the historical power waveform is marked as incorrect; If the fault identification result matches the fault type label, then the fault identification of the historical power waveform is marked as correct, and the number of correct identifications is counted. Obtain the total number of fault identification results, and calculate the analysis accuracy of each candidate wavelet decomposition layer based on the number of correct identifications and the total number of identifications.

[0095] In one embodiment, the target wavelet decomposition level application module 106 performs the following steps: selecting the target wavelet decomposition level based on the analysis accuracy, and using the target wavelet decomposition level to perform fault identification on the distributed power distribution system, obtaining the target fault identification result, for the following purposes: Based on the analysis accuracy, the candidate wavelet decomposition layer corresponding to the highest analysis accuracy is selected; The selected candidate wavelet decomposition levels are used as the target wavelet decomposition levels. The waveform to be detected during the operation of the distributed power distribution system is obtained, and the waveform to be detected is decomposed into wavelet to the target wavelet decomposition level to obtain the target wavelet coefficient set. Calculate the root mean square value and the target spectral energy distribution based on the target wavelet coefficient set; The random forest model is used to perform anomaly feature analysis on the root mean square value of the target and the energy distribution of the target spectrum to obtain the target fault features of the waveform to be detected. Target fault identification results are generated based on the target fault characteristics.

[0096] In this invention, the specific limitations of the fault identification system for a distributed intelligent power distribution system can be found in the above-described limitations of the fault identification method for a distributed intelligent power distribution system, and will not be repeated here. Each module in the aforementioned fault identification system for a distributed intelligent power distribution system can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device in hardware form, or stored in the memory of a computer device in software form, so that the processor can call and execute the operations corresponding to each module.

[0097] In the embodiments provided by this invention, it should be understood that the disclosed system can be implemented in other ways. For example, the system embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and other division methods may be used in actual implementation.

[0098] Furthermore, the functional modules in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or in the form of hardware plus software functional modules.

[0099] Therefore, the embodiments should be considered exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be embraced within the invention. No appended diagram markings in the claims should be construed as limiting the scope of the claims.

[0100] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0101] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0102] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is used as an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the system can be divided into different functional units or modules to complete all or part of the functions described above.

[0103] In the embodiments provided in this disclosure, it should be understood that the disclosed systems and methods can also be implemented in other ways. The system embodiments described above are merely illustrative; for example, the flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than those marked in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram and / or flowchart, and combinations of blocks in block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.

[0104] It should be noted that, in this disclosure, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element limited by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0105] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A fault identification method for a distributed intelligent power distribution system, characterized in that, The method includes: Wavelet decomposition and threshold denoising are performed on historical power waveforms to obtain a set of denoised wavelet coefficients. Calculate the residual entropy of the wavelet coefficients to generate an entropy sequence, and determine the initial wavelet decomposition level; The initial wavelet decomposition level is subjected to a discrete normal perturbation to generate multiple candidate wavelet decomposition levels. Historical power waveforms are decomposed into candidate layers, and fault features of each layer are extracted. The fault features include root mean square value and spectral energy distribution characteristics. Fault identification is performed based on fault characteristics, and the accuracy of each candidate layer is calculated by combining fault type labels. The target wavelet decomposition level is selected based on accuracy and used for fault identification in power distribution systems.

2. The fault identification method for a distributed intelligent power distribution system as described in claim 1, characterized in that, The wavelet decomposition and threshold denoising based on historical power waveforms yield a set of denoised wavelet coefficients, including: Historical power waveforms are preprocessed to obtain standard power waveforms; The historical wavelet decomposition level is selected as the initial candidate level. The standard power waveform is decomposed into an initial candidate level using discrete wavelet decomposition to obtain a set of approximation coefficients and detail coefficients. Calculate the denoising threshold, and then use the denoising threshold to perform soft thresholding on the approximation coefficients and detail coefficients to obtain the denoised wavelet coefficient set.

3. The fault identification method for a distributed intelligent power distribution system as described in claim 2, characterized in that, The calculation of the residual entropy of the wavelet coefficients to generate an entropy sequence includes: Perform inverse wavelet transform on the set of denoised wavelet coefficients to obtain the denoised reconstructed signal; Calculate the residual signal corresponding to each historical wavelet decomposition level based on the standard power waveform and the denoised reconstructed signal; The number of bins is determined based on the resampling window length, and the frequency distribution of the residual signal is calculated based on the number of bins. The probability distribution is generated based on the frequency distribution, and the residual entropy of each historical wavelet decomposition level is calculated. The average value of the residual entropy of each layer is calculated as the representative entropy, and the values ​​are summarized to form an entropy sequence.

4. The fault identification method for a distributed intelligent power distribution system as described in claim 3, characterized in that, Determining the initial wavelet decomposition level includes: The total number of historical wavelet decomposition layers is counted. When the total number is less than the preset threshold, the minimum entropy level in the entropy sequence is selected as the initial wavelet decomposition level. When the total number is greater than or equal to the preset threshold, calculate the first-order difference and second-order difference of the entropy represented by each layer; The number of layers corresponding to the least second-order difference is selected as the initial wavelet decomposition layer number.

5. The fault identification method for a distributed intelligent power distribution system as described in claim 4, characterized in that, The process of performing a discrete normal perturbation on the initial wavelet decomposition level to generate multiple candidate wavelet decomposition levels includes: Obtain the preset disturbance parameters; Based on the perturbation parameters, the initial wavelet decomposition layer number is discretized and normally perturbed to obtain the set of candidate layers to be pruned. Boundary processing is performed on the candidate layer set to be clipped to obtain several candidate wavelet decomposition layers.

6. The fault identification method for a distributed intelligent power distribution system as described in claim 5, characterized in that, The historical power waveform is decomposed into candidate layers, and fault features are extracted for each layer. These fault features include root mean square values ​​and spectral energy distribution characteristics, wherein: The steps for extracting the root mean square value are as follows: Select the candidate wavelet decomposition level as the target candidate level; Historical power waveforms are decomposed into target candidate layers to obtain a set of candidate wavelet coefficients; Calculate the root mean square value of the wavelet coefficient set; The steps for extracting spectral energy distribution characteristics are as follows: Wavelet reconstruction of the coefficient set yields a time-domain candidate signal, and fast Fourier transform of the time-domain candidate signal yields the power spectrum. The frequency range is divided into multiple frequency bands, and the energy distribution of each frequency band is calculated. The energy distribution of each frequency band is then summarized to form the spectral energy distribution characteristics.

7. The fault identification method for a distributed intelligent power distribution system as described in claim 6, characterized in that, The fault identification based on fault characteristics includes: The root mean square value and the spectral energy distribution are combined to generate candidate feature vectors; The initial fault characteristics of the candidate feature vector are identified by a preset machine model; The initial fault features are fused by voting to obtain historical fault features and their confidence levels; Based on a pre-set confidence threshold, historical fault features are screened to generate fault identification results for each candidate wavelet decomposition level.

8. The fault identification method for a distributed intelligent power distribution system as described in claim 1 or 7, characterized in that, The process of fault identification based on fault features and calculation of the accuracy of each candidate layer in conjunction with fault type labels includes: The fault identification results are compared with the fault type labels; the correct or incorrect identification status is marked according to the comparison results; the number of correct identifications is counted, and the accuracy rate is calculated and analyzed in combination with the total number of identifications.

9. The fault identification method for a distributed intelligent power distribution system as described in claim 7, characterized in that, The selection of the target wavelet decomposition level based on accuracy is used for fault identification in the power distribution system, specifically as follows: Select the candidate wavelet decomposition level corresponding to the highest analysis accuracy as the target level; The waveform to be detected is acquired and wavelet decomposition is performed at the target level to obtain the target wavelet coefficient set; Calculate the root mean square value and spectral energy distribution of the target wavelet coefficient set; Anomaly analysis is performed on the features based on a machine learning model to obtain fault characteristics; Fault identification results are generated based on the fault characteristics and used to identify faults in the power distribution system.

10. A system using the fault identification method for a distributed intelligent power distribution system as described in any one of claims 1-9, characterized in that, The system includes: The historical power waveform decomposition module is used to perform wavelet decomposition and threshold denoising based on historical power waveforms to obtain a set of denoised wavelet coefficients. The initial wavelet decomposition level selection module is used to calculate the residual entropy of the wavelet coefficients to generate an entropy sequence and determine the initial wavelet decomposition level. The candidate wavelet decomposition level generation module is used to perform discrete normal perturbation on the initial wavelet decomposition level to generate multiple candidate wavelet decomposition levels. The wavelet decomposition layer feature extraction module is used to decompose historical power waveforms into various candidate layers and extract fault features of each layer. The fault features include root mean square value and spectral energy distribution features. The fault feature analysis module is used to identify faults based on fault features and calculate the accuracy of each candidate layer in combination with fault type labels. The target wavelet decomposition level application module is used to select the target wavelet decomposition level based on accuracy, and is used for fault identification in power distribution systems.

Images