Method and device for identifying voltage flicker source in power distribution network based on SVMD-HistGBT

Abstract

The invention discloses a method and device for identifying a voltage flicker source in a power distribution network based on SVMD-HistGBT, and relates to the field of electric energy quality monitoring, voltage flicker is divided into sine wave modulation type flicker, rectangular wave modulation type flicker and inter-harmonic modulation type flicker according to causes of the voltage flicker, and for the three typical types, the voltage flicker source in the power distribution network is identified. The sampling voltage is split into narrow-band intrinsic mode functions through successive variational mode decomposition, and feature extraction is carried out; and then, constructing a histogram gradient boosting tree classifier to learn three types of flickers, and realizing identification of the voltage flicker source in the high-proportion photovoltaic access power distribution network according to the extracted characteristics of the sampling voltage. According to the invention, different types of flicker signals can be effectively decomposed, so that identification of different voltage flicker sources is completed.

Description

Technical Field

[0001] This invention relates to the field of power quality detection, and more specifically to a method and apparatus for identifying voltage flicker sources in a power distribution network based on SVMD-HistGBT. Background Technology

[0002] With the integration of a high proportion of distributed photovoltaic (PV) power and the emergence of numerous nonlinear and fluctuating loads, voltage fluctuations and flicker in distribution networks are becoming increasingly prominent. Voltage fluctuations and flicker can not only cause malfunctions and failures in electrical equipment but also potentially jeopardize the safe and stable operation of the power grid. Therefore, the detection, identification, and tracing of voltage flicker in distribution networks under the background of high-proportion distributed PV integration are of great significance.

[0003] Currently, there are two main methods for monitoring and identifying voltage flicker sources: one is based on mechanistic analysis, and the other is based on data-driven methods. Mechanism-based methods primarily utilize frequency and time domain transformations to extract data features, which are then input into a classifier for type identification. Common feature extraction algorithms include S-transform, wavelet transform, Fourier transform, and Hilbert-Huang transform. The classifier determines the voltage flicker source to which the disturbance signal belongs, and common methods include support vector machines, neural networks, fuzzy comprehensive evaluation, and principal component analysis reduction. Mechanism-based methods offer strong physical interpretability, but their accuracy heavily relies on prior knowledge of the distribution network topology, load components, and device parameters. In cases of multiple sources overlapping or changes in operating conditions, the model is prone to mismatch, and the computational load increases rapidly with the scale of grid connection. Data-driven methods, on the other hand, can automatically extract latent features and possess good adaptability, but they typically require a large number of labeled samples, lack interpretability, and are difficult to directly apply to complex real-world operating conditions. In summary, existing research focuses on a limited range of voltage flicker sources and lacks systematic research on voltage fluctuation and flicker issues in novel scenarios such as high-proportion distributed photovoltaic grid integration.

[0004] Therefore, how to accurately identify voltage flicker in scenarios with a high proportion of distributed photovoltaic access is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0005] In view of the above problems, the present invention is proposed to provide a method and apparatus for identifying voltage flicker sources in a distribution network based on SVMD-HistGBT, which overcomes or at least partially solves the above problems, and can effectively decompose different types of flicker signals, thereby completing the identification of different voltage flicker sources.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: In a first aspect, embodiments of the present invention provide a method for identifying voltage flicker sources in a distribution network based on SVMD-HistGBT, comprising the following steps: Step 1: Acquire the voltage flicker signal to be identified; Step 2: Decompose the voltage flicker signal to be identified based on the successive variational mode decomposition (SVMD) method to obtain several intrinsic modes (IMFs); Step 3: Extract the corresponding signal features from several intrinsic mode functions (IMFs); Step 4: Input the signal features into the HistGBT classifier for identification to obtain the voltage flicker type.

[0007] Preferably, the process of decomposing the voltage flicker signal to be identified based on the successive variational mode decomposition (SVMD) method in step 2 is as follows: Step 21: Set the number of modes K and penalty parameters ; Step 22: Initialize residuals u(t) represents the voltage flicker signal to be identified; Step 23: Extract the i-th eigenmode from the voltage flicker signal u(t) to be identified. , This represents the i-th intrinsic mode component extracted in the current round. This represents the residual corresponding to the eigenmode extracted in the previous round; Step 24: Calculate the new residuals In each round of decomposition and extraction of intrinsic modes, the energy of the signal is reduced by subtracting the extracted intrinsic mode components from the current residual signal, so that the energy of the remaining signal gradually decreases with each round of decomposition, thus ensuring the convergence and integrity of the decomposition process. Step 25: If If i=K, then output all extracted eigenmodes; otherwise, let the current eigenmode... , i = i + 1, and return to step 23 to extract the next round of intrinsic modes.

[0008] Preferably, the voltage flicker signal to be identified is decomposed and represented based on the successive variational mode decomposition (SVMD) method as follows:

[0009] in, This represents the actual number of intrinsic modes extracted. ; The residual of the voltage flicker signal to be identified at time t is the residual term; This represents the k-th intrinsic mode (IMF) corresponding to the voltage flicker signal to be identified at time t. Each IMF corresponds to a component of the signal in a certain frequency band, such as the fundamental component, low-frequency trend, or high-frequency transient. SVMD decomposition not only improves the algorithm's ability to analyze photovoltaic voltage flicker signals, but also has a certain degree of noise suppression. By adaptively extracting the main modes during the decomposition process, SVMD can still accurately extract useful signal components even in high-noise environments.

[0010] Preferably, the signal features extracted from several intrinsic mode functions (IMFs) in step 3 include energy distribution features, frequency features, and time-domain statistical features; the energy distribution features include energy concentration, energy entropy, and residual energy ratio. Energy concentration is expressed as:

[0011]

[0012]

[0013]

[0014] in, This represents the energy of the k-th eigenmode; Indicates residual energy; Represents total energy; This represents the energy percentage of the k-th intrinsic mode, and is a sequence consisting of the energy percentages corresponding to all intrinsic modes. As a set of energy characteristics, this reflects the importance of this mode relative to the original photovoltaic voltage flicker signal; This indicates the number of intrinsic modes in the decomposition of the voltage flicker signal to be identified. ; This represents the eigenmode component corresponding to the k-th eigenmode; Represents the residual; N represents the number of sampling points; energy concentration reflects the degree of concentration of energy distribution among different modes. The larger the value, the more concentrated the energy is in a few modes, and the smaller the value, the more dispersed the energy is. Energy entropy is expressed as:

[0015] in, Let represent energy entropy; 'e' is a minimum value to avoid logarithmic infinity. Energy entropy is used to measure the uncertainty of energy distribution. For signals dominated by a single mode, When the energy is close to 0 and uniformly distributed across multiple modes, Larger; The residual energy ratio is expressed as:

[0016] in, The residual energy ratio indicates whether SVMD decomposition has adequately extracted the major components. For flashes with distinct characteristics, It should be relatively small, if A larger value may mean that the signal contains components that are difficult to decompose or that the noise level is high. Based on the center frequency of each local vibration mode The mean, median, range, and standard deviation of the center frequency are calculated as frequency characteristics; Based on a preset screening strategy, some intrinsic modes are selected to calculate the corresponding time-domain statistical features. These time-domain statistical features include extreme value features, higher-order statistics, amplitude indices, and zero-crossing rates. Under different flicker sources, the parameter combinations that dominate the IMF can reflect the differences in waveform morphology.

[0017] Preferably, a screening strategy is set based on the energy proportion and center frequency distribution of each intrinsic mode, specifically as follows: 1) Select intrinsic modes whose energy percentage is higher than a preset threshold; 2) Select the eigenmodes that rank in the top M positions according to their energy percentage, where M is a positive integer greater than or equal to 1; 3) The intrinsic mode whose center frequency is located within the main operating frequency band of voltage flicker.

[0018] Preferably, the maximum, minimum, and peak-to-peak values ​​of the selected intrinsic modes are calculated as extreme value features; the skewness, kurtosis, and other statistics of the sample sequences of the selected intrinsic modes are calculated as higher-order statistics; the root mean square value, mean absolute value, and other amplitude correlation quantities of the selected intrinsic modes are calculated as amplitude indicators; and the number of sign changes between adjacent sampling points of the selected intrinsic modes is used as the zero-crossing rate.

[0019] Preferably, the HistGBT classifier is represented as follows:

[0020] in, This represents the prediction result of the m-th iteration; Indicates the learning rate; Let represent the base learner obtained in the m-th training round. The HistGBT classifier is based on an additive model, progressively stacking weak learners (decision trees) to approximate the true objective function. The optimization objective is to minimize the overall risk of the samples under a given loss function. HistGBT, based on the Boosting concept, is used as the main classification model. HistGBT introduces a histogram approximation method on the gradient boosting tree framework, significantly reducing computational complexity and storage requirements while maintaining model accuracy, making it particularly suitable for the large-scale samples and high-dimensional features involved in this invention.

[0021] Preferably, the HistGBT classifier is trained using log loss L, as follows:

[0022] in, Indicates sample Indication function under voltage flicker type c; This represents the predicted probability, i.e., the softmax probability; C represents the total number of voltage flicker types.

[0023] Preferably, the voltage flicker types include sinusoidal wave modulation flicker, rectangular wave modulation flicker, and interharmonic modulation flicker. Historical voltage flicker signals corresponding to the three voltage flicker types are collected, and the corresponding signal features are extracted through steps 2 and 3 to form a training set. The HistGBT classifier is trained using the training set. The training process is as follows: The signal features corresponding to all intrinsic modes in the training set are concatenated to obtain a unified feature vector. The uniform feature vector is automatically discretized into histogram buckets according to fixed intervals; Histogram accumulation is performed based on the gradient of the histogram bucket and the Hessian quantity to calculate the split gain; Based on the gradient boosting mechanism, the split point is selected iteratively according to the split gain, a gradient boosting tree is constructed, and the feature patterns corresponding to the three voltage flicker types are learned layer by layer to predict the probability distribution of the three voltage flicker types.

[0024] Preferably, the HistGBT classifier divides continuous features into a fixed number of histogram buckets and accumulates gradients and Hessian quantities at the bucket level, thereby reducing computational complexity from... Reduced to Let the cumulative amount of the left and right subsets corresponding to a certain candidate split point be... and The nodes before the split form the left subset of the candidate split points, G L H represents the gradient of the left subset. L G represents the Hessian quantity of the left subset, and the nodes after the split form the right subset of the candidate split point. R H represents the gradient of the right subset. R The Hessian quantity of the right subset is represented by the parent node. G P H represents the gradient of the parent node. P Let the Hessian quantity of the parent node be represented, then the split gain is expressed as:

[0025] in, Indicates classification gain; For regularization parameters, To penalize the complexity of leaf nodes, the threshold is calculated by traversing all histogram buckets. The HistGBT classifier can efficiently select the optimal split point. The gradient G is the accumulated value of the first derivative of the sample loss function with respect to the classifier's predicted value, used to characterize the first-order change direction of the classifier's prediction error; the Hessian quantity H is the accumulated value of the second derivative of the sample loss function with respect to the classifier's predicted value, used to reflect the curvature information of the loss function.

[0026] Preferably, the expression for sinusoidal modulation flicker is:

[0027] in, This represents the voltage of the sinusoidal modulation flicker at time t; It is the fundamental frequency of the voltage. is the modulation frequency; m is the modulation intensities.

[0028] Preferably, the expression for rectangular wave modulation flicker is:

[0029] in, This represents the voltage of the rectangular wave modulated flicker at time t; It is the fundamental frequency of the voltage. is the modulation frequency; m is the modulation intensities.

[0030] The preferred expression for interharmonic modulation flicker is:

[0031] in, This represents the voltage of the harmonic modulation flicker at time t; It is the interharmonic frequency; Aspect ratio; This is the fundamental frequency of the voltage.

[0032] Secondly, embodiments of the present invention provide a voltage flicker source identification device in a distribution network based on SVMD-HistGBT, comprising: The signal acquisition module acquires the voltage flicker signal to be identified; The mode decomposition module decomposes the voltage flicker signal to be identified based on the successive variational mode decomposition method to obtain several intrinsic modes (IMFs). The feature extraction module extracts corresponding signal features from several intrinsic mode functions (IMFs). The signal recognition module deploys a pre-trained HistGBT classifier to identify signal features and output the voltage flicker type.

[0033] As can be seen from the above technical solution, compared with the prior art, this invention discloses a method and device for identifying voltage flicker sources in distribution networks based on SVMD-HistGBT, solving the problems of multi-form, non-stationary fluctuations, and difficulty in accurate identification of voltage flicker caused by high-proportion distributed photovoltaic access. This invention combines Sequential Variational Mode Decomposition (SVMD) and Histogram-based Gradient Boosting Trees (HistGBT) algorithms to achieve effective identification of multiple voltage flicker sources. First, the collected voltage flicker signal is decomposed by SVMD to obtain a series of Intrinsic Mode Functions (IMFs); second, the features of these IMFs are extracted to construct feature vectors; finally, the feature vectors are input into the trained HistGBT classifier to achieve automatic identification of voltage flicker sources, providing theoretical support for accurately identifying the causes and mitigation of voltage flicker in distribution networks. Beneficial effects include: (1) By performing layer-by-layer mode decomposition on the original voltage signal through SVMD, different frequency components are effectively separated and key time-frequency features are extracted, providing reliable feature input for subsequent classification and recognition, and ensuring the robustness of features in non-stationary flicker environment; (2) The HistGBT classifier used has advantages in feature utilization and residual fitting, and the recognition accuracy reaches 96.9%, which is significantly better than existing classifiers; (3) It can effectively decompose and identify three types of voltage flicker signals: sinusoidal modulation, rectangular modulation and interharmonic modulation, providing a feasible technical path for tracing and controlling voltage flicker in distribution networks with high proportion of photovoltaic access. Attached Figure Description

[0034] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0035] Figure 1 This is a flowchart of a method for identifying voltage flicker sources in a distribution network based on SVMD-HistGBT, provided in an embodiment of the present invention. Figure 2 This is a schematic diagram of the photovoltaic power generation system structure provided in an embodiment of the present invention; Figure 3This is a schematic diagram of the equivalent circuit model of a photovoltaic module provided in an embodiment of the present invention; Figure 4 This is a schematic diagram of a single-phase equivalent circuit of a grid-connected photovoltaic power generation system provided in an embodiment of the present invention; Figure 5 This is a schematic diagram of the power grid topology provided in the embodiments of the present invention; Figure 6 This is a schematic diagram of node voltage distribution provided in an embodiment of the present invention; Figure 7 This is a schematic diagram of voltage flicker waveform under photovoltaic access provided in the embodiments of the present invention; Figure 8 This is a schematic diagram of the SVMD decomposition results of different flicker sources provided in the embodiments of the present invention; Figure 9 This is a schematic diagram of the power spectral density of each mode provided in the embodiments of the present invention; Figure 10 This is a schematic diagram of the confusion matrix analysis results provided in the embodiments of the present invention; Figure 11 This is a schematic diagram of the training and learning curves of HistGBT provided in the embodiments of the present invention; Figure 12 This diagram illustrates an application example of a voltage flicker source identification device based on SVMD-HistGBT in a power distribution network, as provided in an embodiment of the present invention. Detailed Implementation

[0036] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0037] This invention discloses a method for identifying voltage flicker sources in a distribution network based on SVMD-HistGBT, such as... Figure 1 As shown, it includes the following steps: S1: Acquire the voltage flicker signal to be identified; S2: Based on SVMD, the voltage flicker signal to be identified is decomposed to obtain several intrinsic mode functions (IMFs); S3: Extract the corresponding signal features from several intrinsic mode functions (IMFs); S4: Input the signal characteristics into the HistGBT classifier for identification to obtain the voltage flicker type.

[0038] Furthermore, the process of decomposing the voltage flicker signal to be identified based on SVMD in S2 is as follows: S21: Set the modality number K and penalty parameters ; S22: Initialize residuals u(t) represents the voltage flicker signal to be identified; S23: Extract the i-th eigenmode from the voltage flicker signal u(t) to be identified. , This represents the i-th intrinsic mode component extracted in the current round. This represents the residual corresponding to the eigenmode extracted in the previous round; S24: Calculate the new residual Reduce signal energy; S25: If If i=K, then output all extracted eigenmodes; otherwise, let the current eigenmode... , i=i+1, and return to S23 to extract the next round of intrinsic modes.

[0039] Furthermore, the voltage flicker signal to be identified is decomposed and represented based on SVMD as follows:

[0040] in, This represents the actual number of intrinsic modes extracted. ; The residual of the voltage flicker signal to be identified at time t is the residual term; This represents the k-th intrinsic mode corresponding to the voltage flicker signal to be identified at time t.

[0041] Furthermore, the signal features extracted from several intrinsic mode IMFs in S3 include energy distribution features, frequency features, and time-domain statistical features; the energy distribution features include energy concentration, energy entropy, and residual energy ratio. Energy concentration is expressed as:

[0042]

[0043]

[0044]

[0045] in, This represents the energy of the k-th eigenmode; Indicates residual energy; Represents total energy; This represents the energy percentage of the k-th intrinsic mode, and is a sequence consisting of the energy percentages corresponding to all intrinsic modes. As a set of energy characteristics, this reflects the importance of this mode relative to the original photovoltaic voltage flicker signal; This indicates the number of intrinsic modes in the decomposition of the voltage flicker signal to be identified. ; This represents the eigenmode component corresponding to the k-th eigenmode; Represents the residual; N represents the number of sampling points; energy concentration reflects the degree of concentration of energy distribution among different modes. The larger the value, the more concentrated the energy is in a few modes, and the smaller the value, the more dispersed the energy is. Energy entropy is expressed as:

[0046] in, Let represent energy entropy; 'e' is a minimum value to avoid logarithmic infinity. Energy entropy is used to measure the uncertainty of energy distribution. For signals dominated by a single mode, When the energy is close to 0 and uniformly distributed across multiple modes, Larger; The residual energy ratio is expressed as:

[0047] in, The residual energy ratio indicates whether SVMD decomposition has adequately extracted the major components. For flashes with distinct characteristics, It should be relatively small, if A larger value may mean that the signal contains components that are difficult to decompose or that the noise level is high. Based on the center frequency of each local vibration mode The mean, median, range, and standard deviation of the center frequency are calculated as frequency characteristics; Based on a preset screening strategy, some intrinsic modes are selected to calculate the corresponding time-domain statistical features. These time-domain statistical features include extreme value features, higher-order statistics, amplitude indices, and zero-crossing rates. Under different flicker sources, the parameter combinations that dominate the IMF can reflect the differences in waveform morphology.

[0048] Furthermore, a screening strategy is set based on the energy proportion and center frequency distribution of each intrinsic mode, specifically as follows: 1) Select intrinsic modes whose energy percentage is higher than a preset threshold; 2) Select the eigenmodes that rank in the top M positions according to their energy percentage, where M is a positive integer greater than or equal to 1; 3) The intrinsic mode whose center frequency is located within the main operating frequency band of voltage flicker.

[0049] Furthermore, the maximum, minimum, and peak-to-peak values ​​of the selected intrinsic modes are calculated as extreme value features; the skewness, kurtosis, and other statistics of the sample sequences of the selected intrinsic modes are calculated as higher-order statistics; the root mean square value, mean absolute value, and other amplitude correlation quantities of the selected intrinsic modes are calculated as amplitude indicators; and the number of sign changes between adjacent sampling points of the selected intrinsic modes is used as the zero-crossing rate.

[0050] Furthermore, the HistGBT classifier is represented as:

[0051] in, This represents the prediction result of the m-th iteration; Indicates the learning rate; denoted as the base learner obtained in the m-th training round; the HistGBT classifier is based on an additive model, and progressively stacks decision trees to approximate the true objective function. The optimization objective is to minimize the overall risk of the sample under a given loss function.

[0052] Furthermore, the HistGBT classifier is trained using log loss L, as follows:

[0053] in, Indicates sample Indication function under voltage flicker type c; This represents the predicted probability, i.e., the softmax probability; C represents the total number of voltage flicker types.

[0054] Furthermore, the HistGBT classifier divides continuous features into a fixed number of histogram buckets and accumulates gradients and Hessian quantities at the bucket level, thereby reducing computational complexity from... Reduced to Let the cumulative amount of the left and right subsets corresponding to a certain candidate split point be... and The nodes before the split form the left subset of the candidate split points, G L H represents the gradient of the left subset. L G represents the Hessian quantity of the left subset, and the nodes after the split form the right subset of the candidate split point. R H represents the gradient of the right subset. R The Hessian quantity of the right subset is represented by the parent node. G P H represents the gradient of the parent node. P Let the Hessian quantity of the parent node be represented, then the split gain is expressed as:

[0055] in, Indicates classification gain; For regularization parameters, To penalize the complexity of leaf nodes, the threshold is calculated by traversing all histogram buckets. The HistGBT classifier can efficiently select the optimal split point. The gradient G is the accumulated value of the first derivative of the sample loss function with respect to the classifier's predicted value, used to characterize the first-order change direction of the classifier's prediction error; the Hessian quantity H is the accumulated value of the second derivative of the sample loss function with respect to the classifier's predicted value, used to reflect the curvature information of the loss function.

[0056] Furthermore, the voltage flicker types include sinusoidal wave modulation flicker, rectangular wave modulation flicker, and interharmonic modulation flicker. Historical voltage flicker signals corresponding to these three types are collected, and the corresponding signal features are extracted through S2 and S3 to form a training set. The HistGBT classifier is trained using the training set. The training process is as follows: The signal features corresponding to all intrinsic modes in the training set are concatenated to obtain a unified feature vector. The uniform feature vector is automatically discretized into histogram buckets according to fixed intervals; Histogram accumulation is performed based on the gradient of the histogram bucket and the Hessian quantity to calculate the split gain; Based on the gradient boosting mechanism, the split point is selected iteratively according to the split gain, a gradient boosting tree is constructed, and the feature patterns corresponding to the three voltage flicker types are learned layer by layer to predict the probability distribution of the three voltage flicker types.

[0057] In one specific embodiment, the training process of the HistGBT classifier includes the following steps.

[0058] 1) Feature binning: Divide continuous features into B bins according to fixed intervals, record the sample mapping, and match the divided features in the sample with the bins; 2) Initialization: Using a constant model Minimize the initial loss; 3) Iterative optimization: In each round, histogram accumulation is performed using sample gradients and Hessian quantities. Based on the feature binning results, the gradients and Hessian quantities of all samples falling into each histogram bin are accumulated. The splitting gain of all possible splitting points is calculated based on the accumulated values, and the optimal splitting point is selected. 4) Leaf Update: After selecting the optimal split point and determining the leaf node structure, the leaf node parameter update phase begins. Each leaf node is assigned a weight to update the overall model; the weights are represented as follows: ; 5) Termination condition: When the maximum number of iterations is reached or the performance on the validation set no longer improves, training is stopped and the HistGBT classifier is obtained.

[0059] The feature vector extracted from the acquired signal is used as input to train the HistGBT classifier, which can distinguish three types of photovoltaic voltage flicker. The classifier outputs the probability distribution of the photovoltaic voltage flicker signal to be identified, and finally assigns the category label corresponding to the highest probability, thus realizing the task of identifying photovoltaic voltage flicker signals.

[0060] In one specific embodiment, such as Figure 2 The photovoltaic power generation system shown, combined with its typical operating conditions, has voltage flicker categorized into three types: sinusoidal-modulated flicker (SiMF), square-modulated flicker (SqMF), and interharmonic-modulated flicker (IhMF). These three types of flicker not only cover the most common disturbance modes in photovoltaic operation, but also have good representativeness and distinguishability.

[0061] (1) In the power distribution network, when a large area of ​​cloud cover passes over a photovoltaic power station at a relatively stable speed, the photovoltaic output will exhibit periodic fluctuations. This slow, approximately sinusoidal wave action affects the voltage fundamental wave, causing its amplitude to be periodically modulated under the action of low-frequency signals, forming sinusoidal wave modulated flicker; sinusoidal wave modulated flicker is represented as:

[0062]

[0063] in, This represents the voltage of the sinusoidal modulation flicker at time t; The fundamental frequency of the voltage. m is the modulation frequency, and m is the modulation intensities; the spectrum of SiMF contains a fundamental frequency component. and its symmetrical double-sided component The energy distribution is concentrated, and the sideband amplitude is proportional to the modulation degree m, which reflects the typical symmetrical double-sideband characteristics. On the user side, SiMF manifests as regular flickering of the light source, and its flickering depth is controlled by the modulation degree. When m is large, the flicker level may exceed the standard limit, which will have a negative impact on the lighting comfort of residents and the operational stability of equipment.

[0064] (2) In the scenario of rapid cloud shading, the photovoltaic array experiences a large-scale power drop and recovery in a short period of time. The inverter output power exhibits quasi-rectangular wave characteristics. This power fluctuation will be directly coupled to the voltage port to form rectangular wave modulation flicker. Rectangular wave modulation flicker is represented as:

[0065]

[0066]

[0067] in, The voltage of the rectangular wave modulated flicker at time t is represented by m, where m is the modulation index. D is the modulation period; D is the duty cycle; The k-th harmonic phase term is determined by the duty cycle D, characterizing the effect of duty cycle variation on harmonic phase characteristics; SqMF is represented as the power frequency component in the frequency domain. and distributed in The multi-order sidebands are predominantly odd-order, decaying with increasing order; when the duty cycle... Even-order sidebands may also appear. Compared with sinusoidal modulation, the energy distribution of this type of flicker is more discrete and the spectral structure is more complex. The multi-order sideband characteristics of the SqMF frequency domain reflect the abrupt change process of the voltage envelope. Therefore, this type of flicker will cause drastic changes in voltage fluctuations and has a more serious impact on the stability of the grid voltage. In actual scenarios, it may cause the inverter group to frequently disconnect from the grid or restart, endangering the reliability of the system.

[0068] (3) During the grid-connected operation of the photovoltaic inverter, factors such as resonance in the control loop, carrier modulation subwave, and harmonic injection may introduce interharmonic currents with frequencies near the power frequency. When this current is applied to the voltage port through the grid impedance, it forms interharmonic modulation flicker. Interharmonic modulation flicker is represented as:

[0069]

[0070]

[0071] in, This represents the voltage of the harmonic modulation flicker at time t; It is the interharmonic frequency; Aspect ratio; Indicates the beat frequency; The envelope amplitude of the harmonic modulation flicker at time t is represented; IhMF is essentially the superposition of the power frequency and a non-integer harmonic sine wave, and its envelope exhibits a beat frequency effect; in the frequency domain, IhMF manifests as the power frequency component. With isolated interharmonic peaks Furthermore, it may be accompanied by a beat frequency envelope, characterized by asymmetric sidebands and isolated peaks, distinguishing it from the regular structure of the previous two types of flicker. The isolated spectral components of IhMF may resonate with the system's natural frequency, thereby causing grid oscillations and instability. On the user side, this type of flicker often causes lighting flicker with complex frequency characteristics, making it difficult to monitor effectively using traditional single-index methods.

[0072] It can be seen that SiMF, SqMF and IhMF correspond to periodic sinusoidal modulation, abrupt rectangular modulation and interharmonic superposition effect in photovoltaic access scenarios, respectively, and show obvious differences in their formation mechanism, time domain performance and frequency domain distribution. SiMF has symmetrical double-sideband characteristics with concentrated energy, SqMF exhibits multi-order sideband and envelope abrupt characteristics, while IhMF exhibits isolated spectral components and beat frequency effect.

[0073] In one specific embodiment, a mathematical model of typical voltage flicker caused by photovoltaics is constructed at the mechanistic level, a grid-connected photovoltaic power generation system is established, the voltage phasor at the grid connection point is determined, and three types of voltage flicker signals are collected based on the constructed grid-connected photovoltaic power generation system for feature analysis to train a classifier, and the voltage flicker signal to be identified is collected for detection; including the following steps: S1: Construct an equivalent circuit model of the photovoltaic module, such as Figure 3 As shown, based on the current equation of a single diode and the equivalent circuit model of a photovoltaic module, the basic current equation of the photovoltaic module is established, expressed as:

[0074]

[0075] Among them, I ph Photocurrent; I o V is the reverse saturation current of the diode; V is the output voltage; I is the output current; A is the diode's corrected quality factor; R s R is the equivalent series resistance of the photovoltaic module; p q is the equivalent parallel resistance; n is the diode quality factor; k is the Boltzmann constant; T is the absolute temperature of the photovoltaic module backsheet; q is the electron charge constant. S2: Construct the single-phase equivalent circuit of the grid-connected photovoltaic power generation system, such as Figure 4 As shown, it consists of three parts: a DC circuit, an inverter bridge, and an AC circuit. PV and I PV These represent the output voltage and output current of the photovoltaic array, respectively; C is the DC-side supporting capacitor; and M and For the inverter bridge amplitude modulation ratio and phase shift angle, U i I is the AC voltage vector output by the inverter bridge. i It is the AC current vector output by the inverter bridge, L f and C f It consists of the single-phase inductor and capacitor of the filter. It's the fundamental angular frequency, used in step-up transformers. Equivalent circuit representation, R T X T G T BT and BT represent the transformer's resistance, reactance, conductance, and susceptance, respectively. Ug I is the voltage phasor at the grid connection point. g The output current of the grid-connected photovoltaic power generation system is given. The inverter bridge is defined as an ideal three-phase bridge inverter circuit. Using sinusoidal pulse width modulation, the AC voltage vector output by the inverter bridge is expressed as: ; S3: Substitute the output voltage from the basic current equation of the photovoltaic module as the output voltage of the photovoltaic array, and use it as the AC voltage vector output by the inverter bridge to calculate the AC voltage vector output by the inverter bridge, thus obtaining the grid connection point voltage phasor U. g , represented as:

[0076] Among them, U i is the AC voltage vector output by the inverter bridge; k is the transformer turns ratio, which can be combined with the above. The equivalent circuit of the type was solved; the voltage phasor at the grid connection point established a quantitative relationship between the physical operating state of the photovoltaic inverter side and the voltage flicker phenomenon on the grid side, providing a unified signal source and analysis object for the construction of three types of voltage flicker and the identification of flicker sources.

[0077] On the other hand, the identification device set up by the voltage flicker source identification method in the distribution network based on SVMD-HistGBT includes: The signal acquisition module acquires the voltage flicker signal to be identified; The mode decomposition module decomposes the voltage flicker signal to be identified based on the successive variational mode decomposition method to obtain several intrinsic modes (IMFs). The feature extraction module extracts corresponding signal features from several intrinsic mode functions (IMFs). The signal recognition module deploys a pre-trained HistGBT classifier to identify signal features and output the voltage flicker type.

[0078] On the other hand, in a specific embodiment, the effectiveness of the voltage flicker source identification method in the distribution network based on SVMD-HistGBT is experimentally verified. A simulation platform with multi-point photovoltaic access is constructed based on a typical 33-node distribution network, and its network topology is as follows. Figure 5 As shown, the node distribution and photovoltaic (PV) grid connection of the network are illustrated. Nodes 11, 16, 21, 26, and 31 are selected for distributed PV grid connection to simulate the impact of high-penetration distributed power sources on voltage operation. Further results can be obtained as follows: Figure 6 The node voltage distribution curve shown has the node number on the horizontal axis and the voltage amplitude on the vertical axis.

[0079] Depend on Figure 6It can be seen that the voltage of all nodes is within their respective allowable normal fluctuation range. This indicates that under steady-state conditions of normal photovoltaic output and no sudden load changes, the constructed network model can maintain good voltage quality, providing a reliable benchmark for subsequent superposition of different operating conditions. The slight increase in voltage at the photovoltaic nodes is due to the active power injected by the photovoltaic grid connection. The voltage near the PV nodes is slightly higher than that of adjacent nodes. This phenomenon is consistent with actual laws and also confirms the rationality of the photovoltaic capacity and its voltage support effect in the model.

[0080] Based on this steady-state network, voltage flicker under common photovoltaic access conditions such as SiMF, SqMF, and IhMF will be superimposed on it. Then, SVMD decomposition, IMF feature extraction, and HistGBT classification training will be performed on the sampled signals to achieve rapid and accurate identification of different voltage flicker sources. For each generated voltage signal sample, 75% will be used as the training set and 25% as the test set.

[0081] The voltage flicker waveforms under three typical photovoltaic access conditions obtained based on the established signal generator are as follows: Figure 7 As shown, the sampling frequency is 2kHz and the duration is 0.8s. The horizontal axis represents time and the vertical axis represents voltage. Figure 7 In the middle (a), the main frequency of SiMF is a sine wave with a power frequency of 50Hz. The voltage amplitude fluctuates slowly with the low-frequency sine signal. When a high proportion of photovoltaics are connected to the distribution network, if the cloud covers the photovoltaic power station at an approximately stable speed, the irradiance will show an approximately sinusoidal slow fluctuation, which will cause the photovoltaic output voltage to change periodically. The waveform shows "envelope sine modulation" and the envelope change is smooth. Figure 7 The SqMF voltage fundamental wave shown in (b) experiences a sudden drop in amplitude over a specific period of time, followed by recovery, forming a square or quasi-rectangular envelope modulation. This corresponds to the rapid cloud shading effect or the shutdown or restart of some photovoltaic inverters due to protection strategies, resulting in a sudden voltage fluctuation. The waveform envelope is rectangular, and the amplitude suddenly drops and then recovers, exhibiting a non-smooth fluctuation process. Figure 7 As shown in (c), the main voltage fluctuation of IhMF remains at the power frequency of 50Hz, but due to the influence of interharmonic injection, additional high-frequency modulation components are superimposed. During the grid connection of the photovoltaic inverter, if there is harmonic oscillation or interharmonic current injection in the control loop, the voltage will generate interharmonic components. The waveform still appears approximately continuous on the surface, but periodic voltage distortion is visible, manifested as uneven modulation depth. The above simulation results verify the rationality of the established photovoltaic voltage fluctuation model.

[0082] (1) The waveforms of the three typical photovoltaic voltage flicker obtained from the above simulation were decomposed by SVMD, and the results are as follows: Figure 8 As shown, the horizontal axis represents the sampling points, and the vertical axis represents the voltage amplitude. Figure 8 In the middle (a), the decomposition results of SiMF are shown. Figure 8In the middle (b), the result of SqMF decomposition is shown. Figure 8 (c) represents the IhMF decomposition result. SVMD can effectively decompose three types of voltage flicker into five narrow-band IMFs and one residual term, preserving the fundamental frequency while isolating the perturbation components by frequency band. Specifically, the residual signal amplitude of SiMF is low and exhibits some fluctuation; the residual signal amplitude of SqMF is high; and the residual signal amplitude of IhMF is also low, mainly consisting of noise and undecomposed small-amplitude perturbations. These three results demonstrate that SVMD decomposition can accurately capture the characteristic frequency components of different flicker sources, providing a reliable basis for subsequent construction of feature vectors based on frequency domain energy spectra and for classification and identification.

[0083] (2) Based on the above SVMD decomposition results, the power spectral density and center frequency distribution of each mode can be further obtained, wherein the power spectral density of each mode is as follows: Figure 9 As shown, the horizontal axis represents frequency, and the vertical axis represents power spectral density. Figure 9 In the diagram (a), the power spectral density of each mode of SiMF is represented. Figure 9 In the diagram, (b) represents the power spectral density of the SqMF mode. Figure 9 In the table, (c) represents the power spectral density of the IhMF mode, and the center frequency statistics are shown in Table 1.

[0084] Table 1. Statistical data on the center frequencies of each mode (unit: Hz)

[0085] Combination Figure 9 As shown in Table 1, SiMF exhibits a distinct main peak near the power frequency, along with symmetrical sideband components at 44Hz and 56Hz, consistent with the dual-sideband characteristics of modulation theory. In contrast, SqMF, in addition to the main peak at the power frequency, also forms multiple sidebands at frequencies such as 53Hz, 47Hz, 54Hz, and 46Hz, which is consistent with the Fourier expansion of the square wave envelope, reflecting the harmonic enhancement effect caused by its abrupt change. As for IhMF, isolated non-integer octave peaks appear outside the power frequency components, exhibiting typical beat frequency effects and complex flicker characteristics.

[0086] (3) Based on the above mode decomposition results, the three classifiers RF, ET, and HistGBT are compared. First, a confusion matrix analysis is performed on the three classifiers. The analysis results are as follows: Figure 10 As shown, the horizontal axis represents the predicted value, and the vertical axis represents the actual value. Figure 10 In the middle (a), the results of the RF analysis are shown. Figure 10 (b) represents the ET analysis results. Figure 10 (c) represents the HistGBT analysis results. Figure 10It can be seen that HistGBT has the best classification performance, showing high robustness in the identification of all three types of flicker, especially in the identification of SiMF and IhMF, indicating that HistGBT can effectively capture discriminative information in the feature space. RF has the second best accuracy, but still has some misclassification in distinguishing between SiMF and SqMF, mainly because the similarity of their temporal fluctuation features leads to blurred class boundaries. ET has a higher probability of misclassification between SiMF and SqMF, often misclassifying SqMF as SiMF, indicating that ET's reliance on distance metrics in high-dimensional feature space makes it difficult to effectively distinguish similar feature samples. The overall performance comparison of the three classifiers is shown in Table 2 below. As can be seen from Table 2, HistGBT has the best overall performance among the three, while RF and ET can be used as backup models for fast coarse classification.

[0087] Table 2 Comparison of the overall performance of the three classifiers

[0088] like Figure 11 The image shows the training and learning curves of HistGBT. Figure 11 It can be seen that in the early stages of HistGBT learning, there may be overfitting. However, as the number of samples increases, it will gradually enter the effective learning stage and eventually stabilize, which reflects the stability of the HistGBT classifier.

[0089] On the other hand, in one specific embodiment, an identification system is formed by embedding a voltage flicker source identification device in a power quality monitoring device within a power distribution network based on SVMD-HistGBT. Figure 12 As shown, the voltage flicker source identification device in the distribution network based on SVMD-HistGBT, as the voltage flicker source identification module, is the core processing unit of the entire system, responsible for implementing the voltage flicker identification function based on the SVMD-HistGBT algorithm. The system's workflow is as follows: S1: The AC board and signal circuit acquire three-phase voltage and current signals from the bus in real time, and input them as voltage flicker signals to be identified to the voltage flicker source identification module. S2: The voltage flicker source identification module sequentially performs successive variational mode decomposition (SVMD), feature extraction, and HistGBT classification to determine the flicker type and identify whether it belongs to sinusoidal wave modulation, rectangular wave modulation, or interharmonic modulation flicker, and obtain the identification result. S3: The identification result is controlled by the relay output (DO1–DO11), which can trigger the corresponding alarm output (Alarm1–Alarm3) or drive the external actuator according to the identified flicker type. Furthermore, S2 can simultaneously receive external status signals (such as circuit breaker position, protection action, manual intervention command, etc.) through digital inputs (DI0–DI13). These signals can serve as auxiliary criteria or triggering conditions for the identification process, enabling linkage with field equipment.

[0090] Furthermore, the system uploads the identification results, real-time data, and alarm information to the monitoring system through communication interfaces such as Ethernet and RS-485, realizing complete voltage flicker monitoring, identification, and control.

[0091] Furthermore, the detailed system settings are as follows: 1. Operating power supply range: 88~286VAC / DC, 47~440Hz; power consumption: <13W.

[0092] 2. Bus voltage 1 / Bus voltage 2 converts high voltage (such as 10kV, 35kV, etc.) to low voltage (such as 100V or 220V) proportionally for use by measuring instruments or protection devices.

[0093] 3. Line current 1 / line current 2: This converts large currents (such as hundreds of amperes) into smaller currents (such as 5A or 1A) to facilitate measurement and protection.

[0094] 4. Digital input (DI), used for signal reception.

[0095] 5. Digital output (DO), used for signal output.

[0096] 6. Voltage flicker source identification module, controller implementation, and detection of power flicker accuracy.

[0097] 7. The RS-485 interface is connected to P3 (RS-485) and P4 (RS-485) ports, and the terminals are marked as D+, D-, and SH. It uses a dedicated 485 isolation chip for isolation and has a protection circuit to prevent common-mode and differential-mode voltage interference, lightning strikes, and damage to the communication port due to incorrect wiring.

[0098] 8. Use GPS time synchronization cable for GPS communication.

[0099] 9. Ethernet enables information exchange between the motherboard and the high-precision voltage flicker parameter detection module.

[0100] 10. TCP / IP and Modbus are used to implement the information exchange protocol between the detection module and the outside world.

[0101] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to the method section.

[0102] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for identifying voltage flicker sources in a distribution network based on SVMD-HistGBT, characterized in that, Includes the following steps: Step 1: Acquire the voltage flicker signal to be identified; Step 2: Decompose the voltage flicker signal to be identified based on the successive variational mode decomposition method to obtain several intrinsic modes (IMFs); Step 3: Extract the corresponding signal features from several intrinsic mode functions (IMFs); Step 4: Input the signal features into the HistGBT classifier for identification to obtain the voltage flicker type.

2. The method for identifying voltage flicker sources in a distribution network based on SVMD-HistGBT as described in claim 1, characterized in that, Step 2, the process of decomposing the voltage flicker signal to be identified based on the successive variational mode decomposition method, is as follows: Step 21: Set the number of modes K and penalty parameters ; Step 22: Initialize residuals u(t) represents the voltage flicker signal to be identified; Step 23: Extract the i-th eigenmode from the voltage flicker signal u(t) to be identified. , This represents the i-th intrinsic mode component extracted in the current round. This represents the residual corresponding to the eigenmode extracted in the previous round; Step 24: Calculate the new residuals ; Step 25: If If i=K, then output all extracted eigenmodes; otherwise, let the current eigenmode... , i = i + 1, and return to step 23 to extract the next round of intrinsic modes.

3. The method for identifying voltage flicker sources in a distribution network based on SVMD-HistGBT as described in claim 1, characterized in that, The voltage flicker signal to be identified is decomposed and represented based on the successive variational mode decomposition method as follows: in, This represents the actual number of intrinsic modes extracted. ; This represents the residual of the voltage flicker signal to be identified at time t; This represents the k-th intrinsic mode corresponding to the voltage flicker signal to be identified at time t.

4. The method for identifying voltage flicker sources in a distribution network based on SVMD-HistGBT as described in claim 1, characterized in that, The signal features extracted from several intrinsic mode functions (IMFs) in step 3 include energy distribution features, frequency features, and time-domain statistical features; the energy distribution features include energy concentration, energy entropy, and residual energy ratio.

5. The method for identifying voltage flicker sources in a distribution network based on SVMD-HistGBT as described in claim 1, characterized in that, The HistGBT classifier is represented as: in, This represents the prediction result of the m-th iteration; Indicates the learning rate; This represents the base learner obtained in the m-th training round.

6. The method for identifying voltage flicker sources in a distribution network based on SVMD-HistGBT as described in claim 1, characterized in that, The HistGBT classifier is trained using log loss L, as follows: in, Indicates sample Indication function under voltage flicker type c; denoted by , indicating the predicted probability; C represents the total number of voltage flicker types.

7. The method for identifying voltage flicker sources in a distribution network based on SVMD-HistGBT as described in claim 1, characterized in that, Voltage flicker types include sinusoidal wave modulation flicker, rectangular wave modulation flicker, and interharmonic modulation flicker. Historical voltage flicker signals corresponding to the three voltage flicker types are collected, and the corresponding signal features are extracted through steps 2 and 3 to form a training set. The HistGBT classifier is trained using the training set.

8. The method for identifying voltage flicker sources in a distribution network based on SVMD-HistGBT as described in claim 7, characterized in that, The training process for the HistGBT classifier is as follows: The signal features corresponding to all intrinsic modes in the training set are concatenated to obtain a unified feature vector. Discretize the uniform feature vector into histogram buckets according to fixed intervals; Histogram accumulation is performed based on the gradient of the histogram bucket and the Hessian quantity to calculate the split gain; Based on the gradient boosting mechanism, the split point is selected iteratively according to the split gain, a gradient boosting tree is constructed, the feature patterns corresponding to the three voltage flicker types are learned, and the probability distribution of the three voltage flicker types is predicted.

9. The method for identifying voltage flicker sources in a distribution network based on SVMD-HistGBT as described in claim 8, characterized in that, The splitting gain is calculated by accumulating the bucket-level gradient G and Hessian quantity H based on the histogram buckets, and is expressed as: in, Indicates classification gain; For regularization parameters, Penalize the complexity of leaf nodes; G represents the cumulative left subset of a candidate split point. Nodes before the split constitute the left subset of the candidate split point. L H represents the gradient of the left subset. L The Hessian quantity represents the left subset; G represents the cumulative right subset of a candidate split point. The nodes after the split constitute the right subset of the candidate split point. R H represents the gradient of the right subset. R The Hessian quantity represents the right subset; G represents the cumulative amount of the parent node. P H represents the gradient of the parent node. P This represents the Hessian quantity of the parent node.

10. A device for identifying voltage flicker sources in a distribution network based on SVMD-HistGBT, characterized in that, The method for identifying voltage flicker sources in a distribution network based on SVMD-HistGBT, as described in any one of claims 1-9, includes: The signal acquisition module acquires the voltage flicker signal to be identified; The mode decomposition module decomposes the voltage flicker signal to be identified based on the successive variational mode decomposition method to obtain several intrinsic modes (IMFs). The feature extraction module extracts corresponding signal features from several intrinsic mode functions (IMFs). The signal recognition module deploys a pre-trained HistGBT classifier to identify signal features and output the voltage flicker type.

Images