Self-adaptive sensing method, device and system for power quality disturbance of intelligent power distribution network

Abstract

The invention discloses an intelligent power distribution network electric energy quality disturbance self-adaptive sensing method, device and system. The intelligent power distribution network electric energy quality disturbance self-adaptive sensing method comprises the steps of performing denoising processing on an obtained intelligent power distribution network electric energy quality disturbance signal; analyzing the de-noised electric energy quality disturbance signal of the intelligent power distribution network, calculating a hierarchical weighted permutation entropy considering a multi-phase grouping-Teager operator as a disturbance feature, and generating a disturbance feature vector; and inputting the disturbance feature vector into a pre-trained disturbance recognition model to obtain a disturbance recognition result. According to the method, the hierarchical weighted permutation entropy considering the multi-phase grouping-Teager operator is calculated as the disturbance characteristic according to the difference of characteristic information included between different types of disturbance signals, the disturbance characteristic vector is formed by selecting the disturbance characteristic with the large difference, disturbance identification is completed, and the identification efficiency and accuracy can be greatly improved.

Description

Technical Field

[0001] This invention belongs to the field of power quality detection technology, specifically relating to an adaptive sensing method, device, and system for power quality disturbances in smart distribution networks. Background Technology

[0002] Distributed renewable energy refers to relatively small-scale energy generation systems scattered across geographical locations, typically including solar photovoltaic, wind power, small hydropower, and biomass energy. Compared to traditional large-scale centralized power plants, distributed renewable energy offers advantages such as flexibility, environmental friendliness, and high utilization efficiency. Smart distribution networks, on the other hand, utilize advanced sensing, communication, and information technologies to transform and upgrade traditional power grids, achieving intelligent, efficient, and sustainable development of the power system. Connecting distributed renewable energy to a smart distribution network involves linking geographically dispersed small-scale renewable energy generation systems (such as solar photovoltaic, wind power, and small hydropower) to the smart distribution network, which can improve the utilization rate of renewable energy. However, when distributed renewable energy is connected to a smart distribution network, the power quality (PQDs) of the smart distribution network is susceptible to disturbances. These disturbances can disrupt the integrity of the smart distribution network's voltage data, making it impossible to determine monitoring accuracy, resulting in long monitoring times and low accuracy.

[0003] Therefore, rapidly and accurately detecting and identifying the disturbance type of PQDs signals is of great significance for improving power quality, equipment condition monitoring, and disturbance fault management. PQDs signals contain a wealth of power system operation information, and extracting disturbance characteristics from PQDs signals is crucial for identifying disturbance types. Traditional methods for identifying PQDs include threshold-based methods, time-domain analysis methods, and frequency-domain analysis methods. For example, threshold-based methods determine whether power quality has been disturbed by setting predefined thresholds. For instance, thresholds can be set based on voltage or current waveform characteristics to detect waveform changes outside the normal range. The disadvantage of this method is that it requires pre-determining the thresholds and may not be applicable to different power grid environments and load conditions. Another example is time-domain analysis methods, which directly analyze waveforms in the time domain, including observing waveform shape, amplitude, rise time, fall time, and other characteristics. For example, disturbances can be identified by detecting abrupt changes, drastic changes, or periodic waveforms. However, time-domain analysis methods may not be sensitive enough to some complex disturbances and are significantly affected by noise. Later, frequency domain analysis methods were proposed. These methods transform signals into the frequency domain for analysis, typically involving Fourier transform or wavelet transform. By analyzing frequency domain characteristics, such as harmonics and harmonic distortion, power quality disturbances can be identified. However, frequency domain analysis methods may fail to capture some short-term, non-periodic disturbances and may be ineffective with low spectral resolution. In recent years, artificial intelligence has been increasingly used to solve power quality disturbance problems caused by distributed renewable energy access to smart distribution networks. For example, algorithms such as support vector machines, decision trees, and neural networks can train models based on historical disturbance data to achieve disturbance identification. These methods improve identification accuracy and can adapt to complex disturbance scenarios. However, these methods also have limitations: support vector machines require time to adjust parameters; decision trees tend to ignore the correlation between data; and neural networks require a large amount of data to train the model.

[0004] It is evident that ensuring the accuracy of disturbance identification while improving processing speed is an urgent problem to be solved in the identification of power quality disturbances generated after distributed renewable energy sources are connected to the smart distribution network. Summary of the Invention

[0005] To address the aforementioned issues, this invention proposes an adaptive sensing method, device, and system for power quality disturbances in smart distribution networks. Based on the differences in feature information contained in different types of disturbance signals, the invention calculates the hierarchical weighted permutation entropy considering the multiphase grouping-Teager operator as the disturbance feature, selects disturbance features with significant differences to form a disturbance feature vector, and completes disturbance identification, which can greatly improve identification efficiency and accuracy.

[0006] To achieve the above-mentioned technical objectives and effects, the present invention is implemented through the following technical solution:

[0007] In a first aspect, the present invention provides an adaptive sensing method for power quality disturbances in a smart distribution network, comprising:

[0008] The acquired power quality disturbance signals of the smart distribution network are denoised.

[0009] The power quality disturbance signal of the denoised smart distribution network is analyzed, and the hierarchical weighted permutation entropy considering the multiphase grouping-Teager operator is calculated as the disturbance feature to generate the disturbance feature vector.

[0010] The perturbation feature vector is input into a pre-trained perturbation recognition model to obtain the perturbation recognition result.

[0011] In conjunction with the first aspect, optionally, the denoising process for the acquired power quality disturbance signal includes:

[0012] The parameters of ICEEMDAN are optimized using the GWO algorithm to obtain the optimized ICEEMDAN.

[0013] The optimized ICEEMDAN was used to decompose the acquired power quality disturbance signal of the smart distribution network to obtain the intrinsic mode function components.

[0014] Calculate the correlation coefficient values ​​between each intrinsic mode function component and the acquired power quality disturbance signal of the smart distribution network;

[0015] By comparing each correlation coefficient value with a preset standard coefficient threshold, the intrinsic mode function components that do not meet the requirements are removed, and the remaining intrinsic mode function components are reconstructed to obtain the denoised power quality disturbance signal of the smart distribution network.

[0016] In conjunction with the first aspect, the parameters optimized using the GWO algorithm may optionally include the white noise amplitude weight and the number of times white noise is added.

[0017] In conjunction with the first aspect, optionally, the method for generating training samples for the perturbation recognition model is as follows:

[0018] Hierarchical decomposition of power quality disturbance signals in smart distribution networks;

[0019] Construct a predetermined number of feature scales based on the results of hierarchical decomposition;

[0020] For different feature scales, the corresponding hierarchical weighted permutation entropy considering the multiphase grouping-Teager operator is calculated respectively;

[0021] Retain the first with the largest variance Each feature scale corresponds to a hierarchical weighted permutation entropy that considers the multiphase grouping-Teager operator, which serves as the feature vector of the perturbation sample.

[0022] Training samples are constructed based on the feature vectors of the perturbation samples and their corresponding perturbation categories.

[0023] In conjunction with the first aspect, optionally, the hierarchical decomposition of the power quality disturbance signal in the smart distribution network includes:

[0024] Based on power quality disturbance signals of smart distribution networks Generate a sequence of signals with the same frequency , , , This represents the total number of sampling points;

[0025] For the same frequency signal sequence Perform hierarchical decomposition to obtain the first-level approximate signal. and first layer detail signal ;

[0026] For the first layer approximate signal Decomposition yields the second-level approximate signal. and second layer detail signals ;

[0027] For the second layer approximation signal Decomposition yields the third-level approximate signal. and third layer detail signals ;

[0028] For the third layer approximation signal The fourth-level approximate signal is obtained by decomposition. .

[0029] In conjunction with the first aspect, optionally, constructing a preset number of feature scales based on the hierarchical decomposition results includes:

[0030] Based on the first layer of approximate signal Construct the first feature scale , ;

[0031] Based on the first layer detail signal Constructing the second feature scale , ;

[0032] Based on the second-layer approximation signal Constructing the third feature scale , ;

[0033] Based on the second layer detail signal Constructing the fourth feature scale , ;

[0034] Based on the third-layer approximation signal Constructing the fifth feature scale , ;

[0035] Based on the third layer detail signal Constructing the sixth feature scale , ;

[0036] Based on the fourth layer approximation signal Construct the seventh feature scale , .

[0037] In the formula, For scaling function, It is a wavelet function.

[0038] In conjunction with the first aspect, optionally, for different feature scales, the corresponding hierarchical weighted permutation entropy considering the multiphase grouping-Teager operator is calculated, including:

[0039] For power quality disturbance signals in smart distribution networks The l-th sub-signal obtained after hierarchical decomposition Calculate its multiphase grouping-Teager weights;

[0040] Based on the multiphase grouping-Teager weights of all sub-signals and different feature scales, the corresponding hierarchical weighted permutation entropy is calculated respectively.

[0041] In conjunction with the first aspect, optionally, the multiphase grouping-Teager weights are calculated, including:

[0042] For each sub-signal Phase space recombination is performed to obtain the recombined phase space matrix. , ,in, For time delay, The embedding dimension for phase space recombination; For the first One recombined vector; , represents the index of the recombined vector, and K is the total number of valid recombined vectors;

[0043] Based on the phase space matrix Calculate the multiphase grouping-Teager weight, the formula for which the multiphase grouping-Teager weight is calculated is:

[0044]

[0045] In the formula, For the l-th sub-signal Multiphase grouping - Teager weights , The total number of sub-signals, For the first The phase space of the recombined vectors - Teager instantaneous metric parameters.

[0046] In conjunction with the first aspect, optionally, based on the multiphase grouping-Teager weights of all sub-signals and different feature scales, the corresponding hierarchical weighted permutation entropy is calculated, including:

[0047] Each phase space matrix Convert to symbolic sequence , In the formula, This refers to the process of converting symbol sequences;

[0048] Based on the symbol sequence Calculate the pattern Weighted probability of occurrence , , ,in, For indicator functions;

[0049] Based on different feature scales, and the aforementioned pattern Weighted probability of occurrence Calculate the corresponding hierarchical weighted permutation entropy. ,in, The index representing the feature scale. Indicates the first Each feature scale Indicates the first Hierarchical weighted permutation entropy at each feature scale Represents the set of all possible permutation patterns. , This represents the normalization factor.

[0050] Combining the first aspect, optionally, the top [cases] with the largest variance are retained. The entropy of the hierarchical weighted permutation, which considers the multiphase grouping-Teager operator, corresponding to each feature scale, is used as the feature vector of the perturbed sample. The calculation formula is as follows:

[0051] ,

[0052] In the formula, They represent the first The class of perturbations with the largest variance The hierarchical weighted permutation entropy corresponding to each feature scale , This represents the total number of disturbance types. Indicates the first The feature vector of the perturbation sample.

[0053] In a second aspect, the present invention provides an adaptive sensing device for power quality disturbances in a smart distribution network, comprising:

[0054] The noise reduction module is used to perform noise reduction processing on the acquired power quality disturbance signals of the smart distribution network.

[0055] The disturbance feature vector generation module is used to analyze the power quality disturbance signal of the denoised smart distribution network, calculate the hierarchical weighted permutation entropy considering the multi-phase grouping-Teager operator as the disturbance feature, and generate the disturbance feature vector.

[0056] The disturbance identification module is used to input the disturbance feature vector into a pre-trained disturbance identification model to obtain the disturbance identification result.

[0057] Thirdly, the present invention provides an adaptive sensing system for power quality disturbances in a smart distribution network, including a storage medium and a processor;

[0058] The storage medium is used to store instructions;

[0059] The processor is configured to operate according to the instructions to perform the method according to any one of the first aspects.

[0060] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0061] Based on the differences in feature information contained in different types of disturbance signals, this invention calculates the hierarchical weighted permutation entropy that takes into account the multiphase grouping-Teager operator as the disturbance feature, selects the disturbance features with large differences to form the disturbance feature vector, and completes the disturbance identification, which can greatly improve the identification efficiency and accuracy. Attached Figure Description

[0062] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly described 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, wherein:

[0063] Figure 1A schematic diagram of the original voltage signal under harmonic disturbances after distributed renewable energy is connected to a smart distribution network;

[0064] Figure 2 A schematic diagram illustrating the fitness curve of ICEEMDAN optimized using the GWO algorithm;

[0065] Figure 3 This is a flowchart illustrating the decomposition of power quality disturbance signals in one embodiment of the present invention;

[0066] Figure 4 The time-domain plot of the IMF component obtained by decomposing the power quality disturbance signal;

[0067] Figure 5 A graph showing the correlation coefficient matrix between the various IMF components;

[0068] Figure 6 A joint histogram of the correlation coefficients between the IMF components of the eight perturbations and the original signal;

[0069] Figure 7 This is a comparison chart of the original power quality disturbance signal and the reconstructed signal;

[0070] Figure 8 Scatter plot of WPE values ​​for 8 types of disturbances;

[0071] Figure 9 FGRT-HWPE curves for 8 types of perturbations at 7 characteristic scales;

[0072] Figure 10 This is a structural diagram of an LSTM network in one embodiment of the present invention;

[0073] Figure 11 This is a structural diagram of a BiLSTM network in one embodiment of the present invention;

[0074] Figure 12 This is a structural diagram of the Attention network in one embodiment of the present invention;

[0075] Figure 13 This is a structural diagram of a BiLSTM-Attention perturbation recognition model in one embodiment of the present invention;

[0076] Figure 14 This is a flowchart illustrating the adaptive sensing method for power quality disturbances in a smart distribution network according to one embodiment of the present invention.

[0077] Figure 15 A graph showing the training accuracy of a BiLSTM network;

[0078] Figure 16 The training loss curve for the BiLSTM network;

[0079] Figure 17 This is a power quality disturbance classification label diagram in one embodiment of the present invention;

[0080] Figure 18 To Figure 17 The corresponding confusion matrix diagram;

[0081] Figure 19 The three-dimensional cloud map and the equal-precision surface projection map show the recognition accuracy of eight disturbances under six different disturbance recognition models. Detailed Implementation

[0082] 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 a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0083] Furthermore, if the embodiments of this invention involve descriptions such as "first" or "second," these descriptions are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined with "first" or "second" may explicitly or implicitly include at least one of those features. Additionally, the technical solutions of the various embodiments can be combined with each other, but this must be based on the ability of those skilled in the art to implement them. If the combination of technical solutions is contradictory or impossible to implement, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection claimed by this invention.

[0084] Example 1

[0085] This invention provides an adaptive sensing method for power quality disturbances in a smart distribution network, comprising the following steps:

[0086] (1) Denoise the acquired power quality disturbance signals of the smart distribution network;

[0087] (2) Analyze the power quality disturbance signal of the denoised smart distribution network, calculate the hierarchical weighted permutation entropy considering the multi-phase grouping-Teager operator as the disturbance feature, and generate the disturbance feature vector;

[0088] (3) Input the perturbation feature vector into the pre-trained perturbation recognition model to obtain the perturbation recognition result.

[0089] In one specific embodiment of the present invention, the denoising process for the acquired power quality disturbance signal includes:

[0090] The GWO algorithm is used to optimize the parameters of ICEEMDAN to obtain the optimized ICEEMDAN.

[0091] The optimized ICEEMDAN was used to decompose the acquired power quality disturbance signal of the smart distribution network to obtain the intrinsic mode function components.

[0092] Calculate the correlation coefficient values ​​between each intrinsic mode function component and the acquired power quality disturbance signal of the smart distribution network;

[0093] By comparing each correlation coefficient value with a preset standard coefficient threshold, the intrinsic mode function components that do not meet the requirements are removed, and the remaining intrinsic mode function components are reconstructed to obtain the denoised power quality disturbance signal of the smart distribution network.

[0094] In the above scheme, the GWO algorithm is combined with ICEEMDAN to decompose the perturbation signal and complete the signal denoising and reconstruction by using the correlation coefficient relationship. This can avoid severe mode mixing and reduce interference from complex signals while preserving the perturbation features.

[0095] In one specific embodiment of the present invention, the parameters optimized using the GWO algorithm include the white noise amplitude weight and the number of times white noise is added.

[0096] In the above scheme, GWO is used to optimize the white noise amplitude weight and the number of times white noise is added to ICEEMDAN, so as to realize automatic parameter selection and overcome the drawbacks of manual parameter selection in traditional methods.

[0097] In one specific embodiment of the present invention, the method for generating training samples for the perturbation recognition model is as follows:

[0098] Hierarchical decomposition of power quality disturbance signals in smart distribution networks;

[0099] Construct a predetermined number of feature scales based on the results of hierarchical decomposition;

[0100] For different feature scales, the corresponding hierarchical weighted permutation entropy considering the multiphase grouping-Teager operator is calculated respectively;

[0101] Retain the first with the largest variance Each feature scale corresponds to a hierarchical weighted permutation entropy that considers the multiphase grouping-Teager operator, which serves as the feature vector of the perturbation sample.

[0102] Training samples are constructed based on the feature vectors of the perturbation samples and their corresponding perturbation categories.

[0103] In the above scheme, based on the differences in feature information contained between different types of disturbance signals, the hierarchical weighted permutation entropy considering the multiphase grouping-Teager operator is calculated as the disturbance feature. The disturbance features with large differences are selected to form the disturbance feature vector, thus completing the disturbance identification and greatly improving the identification efficiency and accuracy.

[0104] In one specific embodiment of the present invention, the hierarchical decomposition of the power quality disturbance signal in the smart distribution network includes:

[0105] Based on power quality disturbance signals of smart distribution networks Generate a sequence of signals with the same frequency , , , This represents the total number of sampling points;

[0106] For the same frequency signal sequence Decomposition yields the first-level approximate signal. and detailed signals ;

[0107] Approximate signal of the first layer Decomposition yields the second-level approximate signal. and second layer detail signals ;

[0108] For the second layer approximation signal Decomposition yields the third-level approximate signal. and third layer detail signals ;

[0109] For the third layer approximation signal The fourth-level approximate signal is obtained by decomposition. .

[0110] In one specific embodiment of the present invention, constructing a first preset number of feature scales based on the hierarchical decomposition results includes:

[0111] Based on the first layer of approximate signal Construct the first feature scale , ;

[0112] Based on the first layer detail signal Constructing the second feature scale , ;

[0113] Based on the second-layer approximation signal Constructing the third feature scale , ;

[0114] Based on the second layer detail signal Constructing the fourth feature scale , ;

[0115] Based on the third-layer approximation signal Constructing the fifth feature scale , ;

[0116] Based on the third layer detail signal Constructing the sixth feature scale , ;

[0117] Based on the fourth layer approximation signal Construct the seventh feature scale , .

[0118] Based on the above scheme, it is possible to construct seven characteristic scales based on different frequency bands. , , , , , , .

[0119] In one specific embodiment of the present invention, the corresponding hierarchical weighted permutation entropy considering the phase space-Teager operator is calculated, including:

[0120] For power quality disturbance signals in smart distribution networks The l-th sub-signal Calculate its phase space-Teager weights;

[0121] Based on the l-th sub-signal The phase space-Teager weights and different feature scales are used to calculate the corresponding hierarchical weighted permutation entropy.

[0122] In one specific embodiment of the present invention, calculating the multiphase grouping-Teager weight includes:

[0123] For each sub-signal Phase space recombination is performed to obtain the recombined phase space matrix. , ,in, For time delay, The embedding dimension for phase space recombination; For the first One recombined vector; , represents the index of the recombined vector, and K is the total number of valid recombined vectors;

[0124] Based on the phase space matrix Calculate the multiphase grouping-Teager weight, the formula for which the multiphase grouping-Teager weight is calculated is:

[0125]

[0126] In the formula, For the l-th sub-signal Multiphase grouping - Teager weights , The total number of sub-signals, For the first The phase space of the recombined vectors - Teager instantaneous metric parameters.

[0127] In the above scheme, by calculating the phase space-Teager instantaneous metric parameters and the multiphase grouping-Teager weights, the sensitivity of the Teager operator to instantaneous metrics can be preserved, while incorporating phase space dynamic information.

[0128] In one specific embodiment of the present invention, based on the multiphase grouping-Teager weights of all sub-signals and different feature scales, the corresponding hierarchical weighted permutation entropy is calculated, including:

[0129] Each phase space matrix Convert to symbolic sequence , In the formula, This refers to the process of converting symbol sequences;

[0130] Based on the symbol sequence Calculate the pattern Weighted probability of occurrence , , ,in, For indicator functions;

[0131] Based on different feature scales, and the aforementioned pattern Weighted probability of occurrence Calculate the corresponding hierarchical weighted permutation entropy. ,in, The index representing the feature scale. Indicates the first Each feature scale Indicates the first Hierarchical weighted permutation entropy at each feature scale Represents the set of all possible permutation patterns. , This represents the normalization factor.

[0132] In the specific implementation process, based on the above scheme, it can be calculated that the first... perturbation in the first Hierarchical weighted permutation entropy at each feature scale , , , Representation pattern Next The weighted probability of occurrence of perturbations can be based on the first... The type g disturbance signal is obtained using the above scheme. Furthermore, the type g disturbance can be obtained in... , , , , , , The hierarchical weighted permutation entropy matrix under the 7 feature scales is shown in the following formula:

[0133] .

[0134] In one specific embodiment of the present invention, the top sample with the largest variance is retained. The hierarchical weighted permutation entropy of the combined phase space-Teager operator corresponding to each feature scale is used as the perturbation sample feature vector, and the calculation formula is as follows:

[0135] ,

[0136] In the formula, They represent the first The class of perturbations with the largest variance The hierarchical weighted permutation entropy corresponding to each feature scale , This represents the total number of disturbance types. Indicates the first The feature vector of the perturbation sample.

[0137] In one specific embodiment of the present invention, the perturbation recognition model is a BiLSTM-Attention perturbation recognition model. The BiLSTM-Attention perturbation recognition model includes a connected BiLSTM network and an Attention network. The BiLSTM network consists of two parts: forward computation and backward computation. In the forward computation at time t, the forget gate controls... Output at time and The information to be filtered out from the input at any given time is shown in the following formula:

[0138]

[0139] in, Indicates in The output vector of the forget gate at time step; What it means is The input vector at time step; What it means is The input vector at time t; and This represents the Sigmoid activation function; , These correspond to the weight matrices updated by the forget gate; This represents the bias vector. Simultaneously, the input gate returns the new perturbation feature information to the current state, and updates the state information through the Sigmoid activation function and the tanh layer, as shown in the following equation:

[0140]

[0141] in, for The output vector of the input gate at any given time; , These are the weight matrices updated by the input gate, respectively; This is the bias vector. The tanh function is also used to determine the bias vector. The candidate cell state at time t is shown in the following equation:

[0142]

[0143] in, for The output vector of the candidate unit state at time step; , These are the weight matrices updated for the candidate units. This is the bias vector. Then, the cell state at time t is updated based on the forget gate, input gate, and candidate cell states:

[0144]

[0145] in, for Output vector of the unit state at time step; This is the element-wise product of vectors. The output gate extracts the vectors using the sigmoid activation function. and Information, and with Multiplying the mapping elements yields The output vector at time t is shown in the following equation:

[0146]

[0147]

[0148] in, for The output vector of the output gate at each time step; , These are the weight matrices for updating the output gate; It is the bias vector; for The output vector at time step.

[0149] Similarly, backward computation yields the result of concatenating the hidden states from the two directions to output a bidirectional hidden state sequence.

[0150] The purpose of the Attention network is to filter key information that is beneficial to target recognition from the large amount of information processed by BiLSTM, and to assign different weights according to the degree of influence of the information on the target in order to achieve time series prediction.

[0151] In the above scheme, by combining the BiLSTM network and the Attention network, the recognition process is faster and the results are more accurate.

[0152] The following describes in detail the adaptive sensing method for power quality disturbances in smart distribution networks according to a specific embodiment of the present invention.

[0153] Step 1: Simulate eight types of power quality disturbances generated after distributed power sources are intelligently integrated into the smart distribution network using simulation software. Specifically, based on the relevant IEEE Std 1159-2019 standard and the given mathematical model of power quality disturbances, generate eight types of disturbance signals (including normal signals) C1 to C8 in MATLAB. The fundamental frequency of the disturbance signals is set to 50Hz, the continuous sampling time is 0.2 s, and a total of 1000 samples are collected. The initial phase and other parameters of each disturbance are completely randomized. 100 samples are generated for each type of disturbance, for a total of 800 samples. The initial signal of each disturbance is as follows: Figure 1 As shown.

[0154] Step 2: Optimize the ICEEMDAN parameters using the GWO algorithm, denoted as GWO-ICEEMDAN. The parameters include the white noise amplitude weight and the number of white noise additions. Specifically, the optimization range of the white noise amplitude weight Nstd is set to [0.15, 0.6], the optimization range of the number of white noise additions NE is [50, 600], the population size is 10, the number of iterations is 30, and the fitness function is the minimum envelope entropy function. Figure 2 This is the fitness value variation curve of GWO-ICEEMDAN. (From...) Figure 2It can be seen that after 21 iterations, the fitness values ​​of each perturbation tend to be minimized and then stabilize. Taking harmonic perturbation as an example, the optimal parameter combination of ICEEMDAN at this time is [Nstd, NE]=[0.59044, 295]. Figure 3 The flowchart for GWO-ICEEMDAN is as follows: First, parameters are initialized, followed by the positions of α, β, and δ wolves, with ω wolf set as the surrogate wolf. Next, the positions of the search surrogate pack are initialized. Then, the iterative process begins, performing the following operations on each search surrogate wolf: First, determine if the surrogate wolf's position is better than α wolf's position. If so, further determine if the surrogate wolf's objective function value lies between α and β wolves. If so, change β wolf's position to the current surrogate wolf's position. If not, check if the objective function value lies between β and δ wolves. If so, change δ wolf's position to the current surrogate wolf's position. If the surrogate wolf's position is not better than α wolf's, proceed according to the corresponding logic. Next, the random walk algorithm is executed, the wolf pack positions are updated, and it is checked whether the maximum number of iterations has been reached. If not, the loop continues; if reached, the GWO search parameter results are output. Finally, the parameters are substituted into ICEEMDAN. The entire process iteratively optimizes the wolf pack positions, utilizes GWO search parameters, and combines these with subsequent processing to form a complete algorithm logic.

[0155] Step 3: Use the optimized ICEEMDAN to decompose each disturbance signal to obtain IMFs. Specifically, input the waveform data of different disturbance types into the optimized ICEEMDAN. Figure 4 This is the time-domain plot corresponding to the IMF components.

[0156] Step 4: First, calculate the correlation coefficient between each IMF to ensure the data is reasonable. Then, calculate the standard threshold and the correlation coefficient between each component and the original signal. Taking harmonic disturbance as an example, the results are shown in Table 1.

[0157] Table 1. Correlation coefficients of each IMF of harmonic disturbance with the original signal

[0158]

[0159] Calculate threshold threshold The calculation formula is:

[0160] ,

[0161] Figure 5 The correlation coefficient matrix diagram between the various IMF components is shown. Figure 6 This represents the correlation coefficient between each component and the original signal. The threshold value is calculated. The value is 0.1545. Components IMF1, IMF2, IMF3, IMF4, IMF5, and IMF6 are valid components, and other components are discarded.

[0162] Step 5: Denoising and reconstructing the disturbed signal by comparing the threshold and correlation coefficient, calculating the similarity between the reconstructed and reconstructed signals, and retaining IMF1, IMF2, IMF3, IMF4, IMF5, and IMF6 to complete the denoising and reconstruction. The result is as follows: Figure 7 As shown, the correlation coefficient between the reconstructed signal and the original signal is 0.9768, proving that the reconstructed signal retains the useful information of the original signal while removing useless information.

[0163] Step 6: Analyze the reconstructed signal and extract perturbation features based on the Hierarchical Weighted Permutation Entropy (FGRT-HWPE) combined with the multiphase grouping-Teager operator. Set the embedding dimension, time delay, and number of decomposition levels for the Hierarchical Weighted Permutation Entropy (HWPE), and then use the HWPE to obtain the entropy values ​​of the eight perturbation signals at seven feature scales. Select feature scales with significant entropy differences to construct perturbation feature samples. The specific implementation is as follows: First, set the parameters of HWPE, including the embedding dimension. Time delay Number of decomposition layers First, to fully assess the differences between different types of power quality disturbance signals, the weighted permutation entropy (WPE) was used to analyze eight disturbance signals, and the results are as follows: Figure 8 As shown. By Figure 8 It can be concluded that the WPE values ​​of some disturbance types are quite chaotic and lack differentiation; at the same time, it can be seen that the WPE values ​​of some disturbances, such as voltage fluctuations, are relatively small. This may be because when the disturbance parameters are randomly generated, the disturbance amplitude is not large, approximating a normal signal, thus resulting in a small entropy value. Furthermore, the above analysis also indicates that using a multi-level, multi-feature scale method is necessary. Next, HWPE analysis considering the multiphase grouping-Teager operator was performed on the signal after ICEEMDAN decomposition and reconstruction, obtaining the hierarchically weighted entropy of the eight types of power disturbances at seven feature scales, as shown below. Figure 9 As shown. By Figure 9 It can be seen that the hierarchical weighted permutation entropy exhibits certain differences across various feature scales. To increase the discriminative power of the features and improve the accuracy of perturbation identification, four feature scales with significant differences in various perturbations (e.g., feature scales 2, 3, 4, and 5) are selected to construct perturbation feature samples. For perturbation voltage oscillation, one of the feature samples is constructed as follows:

[0164] ,

[0165] Step 7: Establish a BiLSTM-Attention perturbation recognition model. This model combines a BiLSTM network with an Attention network, which can effectively improve the accuracy of power quality perturbation recognition. The LSTM structure is as follows: Figure 10 As shown in the figure. BiLSTM, while retaining the advantages of LSTM, enhances the learning efficiency and utilization of data by extracting forward and backward information. BiLSTM consists of a forget gate, an input gate, and an output gate to regulate the state of the control units. The specific structure of BiLSTM is shown in the figure. Figure 11 As shown in the diagram, the Attention network filters key information from a large amount of data that is beneficial to target recognition, and assigns different weights based on the degree of influence of the information on the target to achieve perturbation state prediction. The specific structure of the Attention network is shown in the diagram. Figure 12 As shown.

[0166] Steps 8 and 6 have already constructed the disturbance feature samples. These samples are then divided into training and testing samples at an 8:2 ratio to train the BiLSTM-Attention disturbance recognition model, and the final power quality disturbance recognition result is output. The structure of the BiLSTM-Attention disturbance recognition model proposed in this invention is as follows: Figure 13 As shown, the model consists of an input layer, a BiLSTM network, an Attention network, and an output layer. The BiLSTM network extracts data features and uses the extracted results as input to the Attention network; the Attention network calculates feature weights to highlight the influence of features on the result; the output layer outputs the final result through a Softmax function. The model improves data utilization and thus increases the accuracy of perturbation recognition by incorporating an attention mechanism. Figure 14 The diagram shows the flowchart of the power quality disturbance identification process using the method proposed in this embodiment (denoted as GWO-ICEEMDAN-HWPE-BiLSTM-AM). The specific implementation of disturbance identification is as follows: the extracted power quality disturbance features are randomly divided into training samples and test samples at an 8:2 ratio. From 100 samples of each disturbance type, 80 samples are randomly selected as training samples, and 20 as test samples. The BiLSTM-Attention disturbance identification model (referred to as BiLSTM-AM) is used for disturbance identification. The training iterations are set to 30, with 100 iterations per training session. Figure 15 , Figure 16 The figures show the accuracy and loss curves of the BiLSTM network during training, respectively. The BiLSTM-AM classification prediction label map for the test samples is shown below. Figure 17 As shown, the confusion matrix is ​​as follows: Figure 18 As shown. From Figure 18It can be observed that all 6 types of power quality disturbance samples were accurately identified, while the remaining samples were misidentified. Specifically, one sample in type 4 (C4: oscillation) was incorrectly identified as a type 5 sample (voltage interruption), with an accuracy rate of 95.00%; two samples in type 7 (C7: pulse) were incorrectly identified as type 8 samples (C8: harmonics), with an accuracy rate of 90.00%. In the entire test sample, 157 samples were accurately identified, and 3 samples were incorrectly identified, with an overall identification accuracy rate of 98.12%.

[0167] To effectively verify the effectiveness of Attention Network (AM) and Bidirectional Long Short-Term Memory Network (BiLSTM) in the recognition process, BiLSTM-AM, BiLSTM, LSTM-AM, BiLSTM-PSO, LSTM, and SVM were selected for comparison.

[0168] The comparison metrics include accuracy (Acc), precision (P), recall (R), and... The score is defined as follows: Precision (P) represents the proportion of results correctly predicted as belonging to the class by the model. Recall (R) represents the proportion of results correctly identified as belonging to the class by the model. score , The closer the score is to 1, the better the recognition performance. Specific metrics are shown in the table below.

[0169] Table 2. Disturbance identification results for each model

[0170]

[0171] Table 2 shows that, firstly, BiLSTM-AM achieves the best results across all metrics, indicating that this model performs better than the other methods in disturbance identification. Secondly, the identification accuracy is BiLSTM-AM > BiLSTM-PSO > BiLSTM, and LSTM-AM > LSTM, demonstrating the effectiveness of the attention mechanism in optimizing identification performance. Finally, the identification accuracy is BiLSTM > LSTM > SVM, and BiLSTM-AM > LSTM-AM. This demonstrates the effectiveness of BiLSTM in identifying power quality disturbances generated after distributed renewable energy sources are integrated into the smart distribution network.

[0172] Figure 19 This demonstrates the specific identification capabilities of the six models for eight types of disturbances. It further demonstrates the effectiveness of BiLSTM-AM in identifying power quality disturbances generated after distributed renewable energy sources are integrated into the smart distribution network.

[0173] Example 2

[0174] Based on the same inventive concept as Embodiment 1, this embodiment of the invention provides an adaptive sensing device for power quality disturbances in a smart distribution network, comprising:

[0175] The noise reduction module is used to perform noise reduction processing on the acquired power quality disturbance signals of the smart distribution network.

[0176] The disturbance feature vector generation module is used to analyze the power quality disturbance signal of the denoised smart distribution network, calculate the hierarchical weighted permutation entropy considering the multi-phase grouping-Teager operator as the disturbance feature, and generate the disturbance feature vector.

[0177] The disturbance identification module is used to input the disturbance feature vector into a pre-trained disturbance identification model to obtain the disturbance identification result.

[0178] The rest are the same as in Example 1.

[0179] Example 3

[0180] Based on the same inventive concept as in Embodiment 1, this embodiment of the invention provides an adaptive sensing system for power quality disturbances in a smart distribution network, including a storage medium and a processor;

[0181] The storage medium is used to store instructions;

[0182] The processor is configured to operate according to the instructions to perform the method according to any one of the first aspects.

[0183] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0184] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1A device that provides the functions specified in one or more boxes.

[0185] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0186] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0187] The embodiments of the present invention have been described above with reference to the accompanying drawings. However, the present invention is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of the present invention without departing from the spirit and scope of the claims. All of these forms are within the protection scope of the present invention.

[0188] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of this invention is defined by the appended claims and their equivalents.

Claims

1. An adaptive sensing method for power quality disturbances in a smart distribution network, characterized in that, include: The acquired power quality disturbance signals of the smart distribution network are denoised. The power quality disturbance signal of the denoised smart distribution network is analyzed, and the hierarchical weighted permutation entropy considering the multiphase grouping-Teager operator is calculated as the disturbance feature to generate the disturbance feature vector. The perturbation feature vector is input into a pre-trained perturbation recognition model to obtain the perturbation recognition result.

2. The adaptive sensing method for power quality disturbances in a smart distribution network according to claim 1, characterized in that: The noise reduction process for the acquired power quality disturbance signal includes: The parameters of ICEEMDAN are optimized using the GWO algorithm to obtain the optimized ICEEMDAN. The optimized ICEEMDAN was used to decompose the acquired power quality disturbance signal of the smart distribution network to obtain the intrinsic mode function components. Calculate the correlation coefficient values ​​between each intrinsic mode function component and the acquired power quality disturbance signal of the smart distribution network; By comparing each correlation coefficient value with a preset standard coefficient threshold, the intrinsic mode function components that do not meet the requirements are removed, and the remaining intrinsic mode function components are reconstructed to obtain the denoised power quality disturbance signal of the smart distribution network.

3. The adaptive sensing method for power quality disturbances in a smart distribution network according to claim 2, characterized in that: The parameters optimized using the GWO algorithm include the white noise amplitude weight and the number of times white noise is added.

4. The adaptive sensing method for power quality disturbances in a smart distribution network according to claim 1, characterized in that, The method for generating training samples for the disturbance recognition model: Hierarchical decomposition of power quality disturbance signals in smart distribution networks; Construct a predetermined number of feature scales based on the results of hierarchical decomposition; For different feature scales, the corresponding hierarchical weighted permutation entropy considering the multiphase grouping-Teager operator is calculated respectively; Retain the first with the largest variance Each feature scale corresponds to a hierarchical weighted permutation entropy that considers the multiphase grouping-Teager operator, which serves as the feature vector of the perturbation sample. Training samples are constructed based on the feature vectors of the perturbation samples and their corresponding perturbation categories.

5. The adaptive sensing method for power quality disturbances in a smart distribution network according to claim 4, characterized in that, The hierarchical decomposition of power quality disturbance signals in smart distribution networks includes: Based on power quality disturbance signals of smart distribution networks Generate a sequence of signals with the same frequency , , , This represents the total number of sampling points; For the same frequency signal sequence Perform hierarchical decomposition to obtain the first-level approximate signal. and first layer detail signal ; For the first layer approximate signal Decomposition yields the second-level approximate signal. and second layer detail signals ; For the second layer approximation signal Decomposition yields the third-level approximate signal. and third layer detail signals ; For the third layer approximation signal The fourth-level approximate signal is obtained by decomposition. .

6. The adaptive sensing method for power quality disturbances in a smart distribution network according to claim 5, characterized in that: The construction of a preset number of feature scales based on the hierarchical decomposition results includes: Based on the first layer of approximate signal Construct the first feature scale , ; Based on the first layer detail signal Constructing the second feature scale , ; Based on the second-layer approximation signal Constructing the third feature scale , ; Based on the second layer detail signal Constructing the fourth feature scale , ; Based on the third-layer approximation signal Constructing the fifth feature scale , ; Based on the third layer detail signal Constructing the sixth feature scale , ; Based on the fourth layer approximation signal Construct the seventh feature scale , ; In the formula, For scaling function, It is a wavelet function.

7. The adaptive sensing method for power quality disturbances in a smart distribution network according to claim 4, characterized in that: For different feature scales, the corresponding hierarchical weighted permutation entropy considering the multiphase grouping-Teager operator was calculated, including: For power quality disturbance signals in smart distribution networks The l-th sub-signal obtained after hierarchical decomposition Calculate its multiphase grouping-Teager weights; Based on the multiphase grouping-Teager weights of all sub-signals and different feature scales, the corresponding hierarchical weighted permutation entropy is calculated respectively.

8. The adaptive sensing method for power quality disturbances in a smart distribution network according to claim 7, characterized in that, Calculating the multiphase grouping-Teager weights includes: For each sub-signal Phase space recombination is performed to obtain the recombined phase space matrix. , ,in, For time delay, The embedding dimension for phase space recombination; For the first One recombined vector; , represents the index of the recombined vector, and K is the total number of valid recombined vectors; Based on the phase space matrix Calculate the multiphase grouping-Teager weight, the formula for which the multiphase grouping-Teager weight is calculated is: ， In the formula, For the l-th sub-signal Multiphase grouping - Teager weights , The total number of sub-signals, For the first The phase space of the recombined vectors - Teager instantaneous metric parameters.

9. The adaptive sensing method for power quality disturbances in a smart distribution network according to claim 8, characterized in that, Based on the multiphase grouping-Teager weights of all sub-signals and different feature scales, the corresponding hierarchical weighted permutation entropy is calculated, including: Each phase space matrix Convert to symbolic sequence , In the formula, This refers to the process of converting symbol sequences; Based on the symbol sequence Calculate the pattern Weighted probability of occurrence , , ,in, For indicator functions; Based on different feature scales, and the aforementioned pattern Weighted probability of occurrence Calculate the corresponding hierarchical weighted permutation entropy. ,in, The index representing the feature scale. Indicates the first Each feature scale Indicates the first Hierarchical weighted permutation entropy at each feature scale Represents the set of all possible permutation patterns. , This represents the normalization factor.

10. The adaptive sensing method for power quality disturbances in a smart distribution network according to claim 4, characterized in that, Retain the first with the largest variance The entropy of the hierarchical weighted permutation, which considers the multiphase grouping-Teager operator, corresponding to each feature scale, is used as the feature vector of the perturbed sample. The calculation formula is as follows: ， In the formula, They represent the first The class of perturbations with the largest variance The hierarchical weighted permutation entropy corresponding to each feature scale , This represents the total number of disturbance types. Indicates the first The feature vector of the perturbation sample.

11. An adaptive sensing device for power quality disturbances in a smart distribution network, characterized in that, include: The noise reduction module is used to perform noise reduction processing on the acquired power quality disturbance signals of the smart distribution network. The disturbance feature vector generation module is used to analyze the power quality disturbance signal of the denoised smart distribution network, calculate the hierarchical weighted permutation entropy considering the multi-phase grouping-Teager operator as the disturbance feature, and generate the disturbance feature vector. The disturbance identification module is used to input the disturbance feature vector into a pre-trained disturbance identification model to obtain the disturbance identification result.

12. An adaptive sensing system for power quality disturbances in an intelligent distribution network, characterized in that, Including storage media and processor; The storage medium is used to store instructions; The processor is configured to operate according to the instructions to perform the method according to any one of claims 1-10.

Images