Method and device for identifying power quality disturbance, equipment and storage medium

Abstract

The application relates to the technical field of power quality identification, and discloses a power quality disturbance identification method and device, equipment and a storage medium, the method comprising the following steps: acquiring a power quality disturbance signal; using a preset one-dimensional convolutional neural network and a one-dimensional residual neural network to extract features of the power quality disturbance signal, so as to obtain a target time feature vector of the power quality disturbance signal; using a preset multi-head attention mechanism to perform feature fusion processing on the target time feature vector, so as to obtain a fused signal feature; and using a preset classifier to classify the fused signal feature, so as to obtain a final classification result used for indicating a disturbance type of the power quality disturbance signal. The method can comprehensively integrate various feature information, effectively reduce the influence of noise on the classification result, improve the classification performance and generalization ability of the classifier, has better performance in terms of classification accuracy and noise resistance, improves disturbance identification accuracy, and provides effective support for power fault diagnosis in a power system.

Description

Technical Field

[0001] This invention relates to the field of power quality identification technology, and in particular to a method, device, equipment and storage medium for identifying power quality disturbances. Background Technology

[0002] Power quality disturbances (PQDs) include various types such as voltage sags, swells, interruptions, harmonics, and flicker. With the increasing electrification of power systems, the identification of power quality disturbances (PQDs) has become increasingly important.

[0003] Traditional power quality disturbance identification methods are mainly based on manual feature extraction and simple classifiers. When faced with complex and noisy actual power quality signals, they often perform poorly and cannot effectively distinguish noise from real disturbance features. Furthermore, traditional methods have limited ability to handle complex situations where multiple disturbances coexist (compound disturbances). Noise can mask the true characteristics of power quality disturbances, leading to misjudgments by traditional identification methods.

[0004] Therefore, the accuracy of disturbance identification needs to be improved. Summary of the Invention

[0005] The main objective of this invention is to provide a method, device, equipment, and storage medium for identifying power quality disturbances, which can solve the problem that the accuracy of disturbance identification needs to be improved in the prior art.

[0006] To achieve the above objectives, a first aspect of the present invention provides a method for identifying power quality disturbances, the method comprising:

[0007] Acquire power quality disturbance signals;

[0008] The power quality disturbance signal is subjected to feature extraction using a pre-defined one-dimensional convolutional neural network and a one-dimensional residual neural network to obtain the target time feature vector of the power quality disturbance signal;

[0009] The target temporal feature vector is fused using a preset multi-head attention mechanism to obtain fused signal features;

[0010] The fused signal features are classified using a preset classifier to obtain a final classification result, which is used to indicate the disturbance type of the power quality disturbance signal.

[0011] To achieve the above objectives, a second aspect of the present invention provides a power quality disturbance identification device, the device comprising:

[0012] Signal acquisition module: used to acquire power quality disturbance signals;

[0013] Feature extraction module: used to extract features from the power quality disturbance signal using a preset one-dimensional convolutional neural network and a one-dimensional residual neural network, to obtain the target time feature vector of the power quality disturbance signal;

[0014] Attention fusion module: used to perform feature fusion processing on the target temporal feature vector using a preset multi-head attention mechanism to obtain fused signal features;

[0015] Disturbance identification module: used to classify the features of the fused signal using a preset classifier to obtain a final classification result, which is used to indicate the disturbance type of the power quality disturbance signal.

[0016] To achieve the above objectives, a third aspect of the present invention provides a computer-readable storage medium storing a computer program, which, when executed by a processor, causes the processor to perform the steps of the method shown in the first aspect.

[0017] To achieve the above objectives, a fourth aspect of the present invention provides a computer device including a memory and a processor, the memory storing a computer program that, when executed by the processor, causes the processor to perform the steps of the method shown in the first aspect.

[0018] The embodiments of the present invention have the following beneficial effects:

[0019] This invention provides a method for identifying power quality disturbances. The method includes: acquiring a power quality disturbance signal; extracting features from the power quality disturbance signal using a pre-defined one-dimensional convolutional neural network and a one-dimensional residual neural network to obtain a target time feature vector of the power quality disturbance signal; performing feature fusion processing on the target time feature vector using a pre-defined multi-head attention mechanism to obtain fused signal features; and classifying the fused signal features using a pre-defined classifier to obtain a final classification result, which is used to indicate the disturbance type of the power quality disturbance signal. The proposed identification method can integrate multiple feature information, effectively reducing the impact of noise on the classification result, improving the classification performance and generalization ability of the classifier, and exhibiting better performance in terms of classification accuracy and noise resistance, thus improving the accuracy of disturbance identification and providing effective support for power fault diagnosis in power systems. Attached Figure Description

[0020] 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 some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0021] in:

[0022] Figure 1 This is a flowchart of a method for identifying power quality disturbances in an embodiment of the present invention;

[0023] Figure 2 This is a structural block diagram of a power quality disturbance identification system according to an embodiment of the present invention;

[0024] Figure 3 This is a block diagram of the model structure of a one-dimensional convolutional neural network and a one-dimensional residual neural network in an embodiment of the present invention;

[0025] Figure 4 This is a block diagram of a multi-head attention structure according to an embodiment of the present invention;

[0026] Figure 5 This is a structural block diagram of a power quality disturbance identification device according to an embodiment of the present invention;

[0027] Figure 6 This is a structural block diagram of a computer device in an embodiment of the present invention. Detailed Implementation

[0028] 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.

[0029] It should be noted that, addressing the issues of redundant feature extraction and low recognition accuracy in traditional power quality disturbance identification under strong noise environments, this invention proposes a noisy power quality disturbance identification method based on deep learning feature fusion. First, the power quality disturbance waveform signal under noisy conditions is acquired. Then, the noisy disturbance signal is denoised using an improved adaptive threshold and an improved threshold function algorithm. Next, a one-dimensional convolutional neural network and a one-dimensional residual neural network are used to extract features from the denoised power quality disturbance signal, and a multiplicative fusion method is used to concatenate the features, strengthening the spatial correlation of the features. Then, a multi-head attention mechanism is introduced to simultaneously capture and fuse features from different levels and angles, extracting and selecting suitable features. Finally, a classification module is used to classify the features to obtain the final classification result. This invention proposes a noisy power quality disturbance identification model that integrates multiple feature information, effectively reducing the impact of noise on the classification results and improving the model's classification performance and generalization ability. The proposed method exhibits better performance in terms of classification accuracy and noise resistance, providing effective support for power fault diagnosis in power systems.

[0030] Please see Figure 1 , Figure 1 This is a flowchart of a method for identifying power quality disturbances according to an embodiment of the present invention, as shown below. Figure 1 The method shown includes the following steps:

[0031] 101. Acquire power quality disturbance signals;

[0032] Understandably, in order to identify power quality disturbances, it is first necessary to collect power quality disturbance signals, which can also be noise-reduced signals.

[0033] For example, this identification method is applied to a power quality disturbance identification system, such as... Figure 2 As shown, Figure 2 This is a structural block diagram of a power quality disturbance identification system according to an embodiment of the present invention; Figure 2 The diagram illustrates the following: Data Input Module: Acquires the power quality disturbance waveform signal under noisy conditions and inputs it into the signal denoising module; Signal Denoising Module: Denoises the noisy disturbance signal using improved adaptive thresholding and improved threshold function algorithms; Inputs the denoised power quality disturbance signal into the feature extraction module; Feature Extraction Module: Processes the denoised power quality disturbance signal, using a one-dimensional convolutional neural network and a one-dimensional residual neural network to extract features, and uses a multiplication fusion method to concatenate features, strengthening the spatial correlation of features; Inputs the feature-extracted power quality disturbance signal into the feature fusion module; Feature Fusion Module: Introduces a multi-head attention mechanism to simultaneously capture and fuse features from different levels and angles, extracting and selecting suitable features; Inputs these features into the classification module; Classification Module: Uses a Softmax classifier to classify the features and obtain the final classification result.

[0034] In step 101, by setting Figure 2 The data input module and signal noise reduction module shown are implemented in step 102 by setting... Figure 2 The feature extraction module shown is implemented; step 103 is achieved by setting... Figure 2 The feature fusion module shown is implemented; step 104 is achieved by setting... Figure 2 The classification module shown is implemented.

[0035] Combination Figure 2The noise reduction method can be as follows: acquiring the power quality disturbance signal under noisy conditions; performing wavelet decomposition on the power quality disturbance signal to obtain the J-level wavelet coefficients of the power quality disturbance signal; using a preset improved adaptive threshold algorithm and improved threshold function algorithm to perform noise reduction processing on the wavelet coefficients to obtain denoised wavelet coefficients; and reconstructing the signal based on the denoised wavelet coefficients to obtain the denoised signal of the power quality disturbance signal. That is, this application uses an improved adaptive threshold algorithm and an improved threshold function algorithm to perform noise reduction processing on the power quality disturbance signal, reducing noise interference to the signal.

[0036] Specifically, the following information can be used as a reference for noise reduction methods:

[0037] S1. Acquire power quality disturbance signals under noisy environments;

[0038] S2. Perform wavelet decomposition on the power quality disturbance signal to obtain the J-layer wavelet coefficients of the power quality disturbance signal;

[0039] S3. The wavelet coefficients are denoised using a preset improved adaptive threshold algorithm and an improved threshold function algorithm to obtain denoised wavelet coefficients.

[0040] It should be noted that the power quality disturbance signal under noisy environment is acquired, the power quality disturbance signal is decomposed into wavelet coefficients to obtain the J-level wavelet coefficients of the power quality disturbance signal, and the wavelet coefficients are denoised using a preset improved adaptive threshold algorithm and improved threshold function algorithm to obtain denoised wavelet coefficients.

[0041] The power quality disturbance signal can be a power quality disturbance waveform signal. The wavelet coefficients of the signal are obtained through wavelet decomposition. Then, the wavelet threshold of the wavelet coefficients is obtained using an improved adaptive threshold algorithm. Finally, the wavelet threshold and wavelet coefficients are used for noise reduction to obtain the denoised wavelet coefficients.

[0042] In one feasible implementation, step S3 includes steps A01 to A03:

[0043] A01. Based on the J-layer wavelet coefficients and the preset standard deviation algorithm, determine the noise standard deviation. ;

[0044] It should be noted that a noise standard deviation is calculated for each layer. Specifically, the noise standard deviation is determined based on the J-layer wavelet coefficients and a preset standard deviation algorithm. .

[0045] A02. Determine the wavelet threshold corresponding to each layer of wavelet coefficients by using the peak-to-ratio correction factor of each layer of wavelet coefficients, the noise standard deviation, and the preset improved adaptive threshold algorithm respectively.

[0046] It should be noted that the threshold serves as the boundary between noise and the true signal in the decomposition of wavelet detail coefficients. A commonly used threshold is:

[0047] (1)

[0048] (2)

[0049] For wavelet threshold, The number of signal sampling points. These are the wavelet coefficients of each layer. The standard deviation of the noise is used to estimate the overall noise of the signal. Since the universal threshold is fixed, and the noise distribution is random, using the same threshold in other decomposition layers would lead to excessive removal of coefficients from the true signal. Therefore, improvements are made to address the shortcomings of this threshold setting.

[0050] A wavelet threshold is calculated for each layer. The wavelet threshold corresponding to the wavelet coefficients of each layer is determined by using the peak-to-sum ratio correction factor of the wavelet coefficients of each layer, the noise standard deviation, and a preset improved adaptive thresholding algorithm. .

[0051] For example, the improved adaptive threshold algorithm is as follows:

[0052] (3)

[0053] (4)

[0054] (5)

[0055] In the formula, The standard deviation of noise. This represents the wavelet coefficient in the k-th direction of the i-th level wavelet decomposition; This represents the median of the absolute values ​​of all wavelet coefficients. For the first j Layer wavelet threshold; The standard deviation of the noise. The number of signal sampling points. For the first Wavelet threshold of the layer For the first The noise standard deviation of the layer To represent the peak and ratio correction factor, It is the natural logarithm. For the first The ratio of peak value to sum value in the wavelet coefficients of the layer. For the first The length of the layer wavelet coefficients.

[0056] By introducing Estimating the noise standard deviation of wavelet coefficients layer by layer reduces the impact of noise on the meantime. Overall estimation. This formula reduces the value of the first-layer threshold while increasing the threshold values ​​of subsequent layers, thus more effectively preserving the wavelet coefficients of the true signal.

[0057] In one feasible implementation, before step A02, the method further includes: utilizing the lengths of the wavelet coefficients at each layer respectively. L j The ratio of peak value to sum value in wavelet coefficients P SRj Determine the peak and ratio correction factors for wavelet coefficients at each level. .

[0058] A03. Use the wavelet threshold and improved threshold function algorithm of each layer to denoise the wavelet coefficients and determine the denoised wavelet coefficients.

[0059] Then, the wavelet coefficients are denoised using the wavelet threshold and improved threshold function algorithm for each layer to determine the denoised wavelet coefficients. Specifically, the improved threshold function algorithm includes a first denoising algorithm, a second denoising algorithm, and a third denoising algorithm.

[0060] Specifically, step S3 includes the following steps:

[0061] B01. The adjustable parameters of wavelet coefficients at each level are determined using a preset adjustable parameter determination rule. The adjustable parameter determination rule includes at least the rule that the higher the number of decomposition levels, the lower the adjustable parameters.

[0062] For each layer of wavelet coefficients and wavelet threshold, the following processing is performed:

[0063] B02. If the wavelet coefficients are greater than or equal to the wavelet threshold, then the denoised wavelet coefficients are obtained by using the first denoising algorithm, the adjustable parameters, the wavelet coefficients, and the wavelet threshold.

[0064] B03. If the absolute value of the wavelet coefficients is less than the wavelet threshold, then the denoised wavelet coefficients are obtained by using the second denoising algorithm, the adjustable parameters, the wavelet coefficients, and the wavelet threshold.

[0065] B04. If the wavelet coefficients are less than or equal to the negative value of the wavelet threshold, then the denoised wavelet coefficients are obtained using the third denoising algorithm, the adjustable parameters, the wavelet coefficients, and the wavelet threshold.

[0066] By comparing the wavelet coefficients of each layer with the wavelet threshold of the layer, the denoising algorithm for the wavelet coefficients of that layer is determined, thereby obtaining the denoised wavelet coefficients of that layer and realizing adaptive denoising of the wavelet coefficients.

[0067] It should be noted that the noisy signal is defined as... It consists of pure signals and noise signals Composition, namely:

[0068] (6)

[0069] Traditional soft threshold function Defined as:

[0070] (7)

[0071] In the formula, It is a step function. The threshold is the wavelet threshold.

[0072] Traditional hard threshold function Defined as:

[0073] (8)

[0074] In order to have both and The advantages of this approach allow for the preservation of more detailed information after signal denoising. An improved threshold function algorithm is constructed using adjustable parameters. Achieve adaptive adjustment.

[0075] For example, the improved threshold function algorithm is as follows:

[0076] (9)

[0077] In the formula, These are wavelet coefficients. It is a wavelet threshold. It is an adjustable parameter. This is the first denoising algorithm; This is the second denoising algorithm; This is the third denoising algorithm; wavelet coefficients x The denoised wavelet coefficients.

[0078] according to and Establishing the energy distribution characteristics of each decomposition layer of wavelet transform Mathematical model:

[0079] (10)

[0080] In the formula: , The first In layer decomposition and The energy. The rest are It is possible to deduce the values ​​at each decomposition level. The value of is taken within the range [1, 11]. In the low-resolution layer of the wavelet transform, a larger value is chosen. The value causes the threshold function of this layer to be biased. This filters out most of the noise figure; in the high-resolution layer, a smaller value is selected. Value bias This better preserves information about local mutation points.

[0081] S4. Based on the denoised wavelet coefficients, the signal is reconstructed to obtain the denoised signal of the power quality disturbance signal.

[0082] Finally, the signal is reconstructed using the denoised wavelet coefficients, for example, by using inverse wavelet transform to restore the signal, thus obtaining the denoised signal of the power quality disturbance signal.

[0083] In this invention, the denoised signal is input into a deep learning model for power quality disturbance identification. Deep learning can automatically learn features from the original data, overcoming the limitations of traditional methods and simple machine learning-based methods, thus more effectively identifying noisy power quality disturbance signals. Therefore, this invention proposes a noisy power quality disturbance identification method based on deep learning feature fusion. This method further improves the disturbance identification accuracy, has a faster convergence speed, smaller fluctuation amplitude, strong noise resistance, and high recognition accuracy under different noise environments.

[0084] It is understood that the deep learning model shown in this application is a power quality disturbance identification model, which is a trained model. The training samples used for training this model include the correspondence between several power quality disturbance signals and disturbance type labels. The original deep learning model learns the relationship between signals and labels using these training samples until the deep learning model can output the correct label corresponding to the signal based on the signal, thus obtaining a power quality disturbance identification model that can identify the disturbance type of the signal. The specific training process is not described in detail, but can be referred to in the training process of existing deep learning models when performing classification tasks.

[0085] The deep learning model includes at least a one-dimensional convolutional neural network, a one-dimensional residual neural network, a multi-head attention network, and a classifier, as detailed below.

[0086] 102. Using a pre-defined one-dimensional convolutional neural network and a one-dimensional residual neural network, feature extraction is performed on the power quality disturbance signal to obtain the target time feature vector of the power quality disturbance signal;

[0087] After obtaining the power quality disturbance signal, a pre-defined one-dimensional convolutional neural network and a one-dimensional residual neural network can be used to extract features from the power quality disturbance signal, obtaining the target time feature vector of the power quality disturbance signal, thus obtaining preliminary disturbance features. Specifically, the denoised power quality disturbance signal is processed by using a one-dimensional convolutional neural network and a one-dimensional residual neural network to reduce the dimensionality of the power quality disturbance data and extract time feature vectors with high discriminative power quality disturbances (PQDs). and (See below) Figure 2 and Figure 3 , Figure 3 This is a block diagram of the model structure of a one-dimensional convolutional neural network and a one-dimensional residual neural network in an embodiment of the present invention; through Figure 2 and Figure 3 The feature extraction module shown performs feature extraction on the signal.

[0088] See Figure 2 and Figure 3 The step involves using a pre-defined one-dimensional convolutional neural network and a one-dimensional residual neural network to extract features from the power quality disturbance signal, obtaining the target time feature vector of the power quality disturbance signal, including C01 to C03:

[0089] C01. Based on the one-dimensional convolutional neural network, feature extraction is performed on the power quality disturbance signal to obtain the first time feature vector of the power quality disturbance signal;

[0090] In one feasible implementation, the one-dimensional convolutional neural network includes multiple convolutional layers and fully connected layers connected in series; then step C01 includes: inputting the power quality disturbance signal into the convolutional layers and fully connected layers, and processing it sequentially through the convolutional layers and fully connected layers to obtain a first time feature vector of the power quality disturbance signal.

[0091] C02. Based on the one-dimensional residual neural network, feature extraction is performed on the power quality disturbance signal to obtain the second time feature vector of the power quality disturbance signal;

[0092] In one feasible implementation, the one-dimensional residual neural network includes several convolutional layers, residual connection layers, and fully connected layers connected in series. Then, step C02 includes: inputting the power quality disturbance signal into the convolutional layer to obtain the signal features after convolution processing; inputting the power quality disturbance signal and the signal features after convolution processing into the residual connection layer to obtain the signal features after residual connection; and inputting the signal features after residual connection into the fully connected layer to obtain a second time feature vector.

[0093] C03. Using the first time feature vector and the second time feature vector, feature splicing is performed to obtain the target time feature vector of the power quality disturbance signal.

[0094] It should be noted that the feature extraction module uses multiplication fusion (with... Feature concatenation is performed using an element-wise multiplication method with exponentially quantified features as the base. The formula for multiplicative feature concatenation is:

[0095] (11)

[0096] In the formula: for The amount; for The components are: B is the first-time feature vector, and A is the second-time feature vector. This is the target time feature vector.

[0097] The advantage of multiplicative fusion lies in its ability to comprehensively consider the interaction relationships between each pair of feature elements. Multiplicative fusion exhibits good robustness when dealing with outliers (especially extremely large or small values) because it does not easily lead to rapid numerical growth during multiplication.

[0098] 103. The target temporal feature vector is subjected to feature fusion processing using a preset multi-head attention mechanism to obtain fused signal features;

[0099] Specifically, step 103 includes: utilizing the multi-head attention mechanism to simultaneously capture and fuse the features of the target temporal feature vector at different levels and angles, thereby obtaining fused signal features. (See also...) Figure 2 and Figure 4 , Figure 4 This is a block diagram of a multi-head attention structure in an embodiment of the present invention.

[0100] pass Figure 2 and Figure 4 The feature fusion module shown introduces a multi-head attention mechanism to simultaneously capture and fuse features from different levels and angles, extract and filter suitable features, and input them into the classification module.

[0101] It should be noted that multi-head attention is an extended form of attention mechanism. By introducing multiple independent attention heads, it simultaneously captures features from different levels and perspectives, thereby improving the model's expressive power and generalization ability. Based on the deep learning feature fusion module, a soft attention mechanism is selected. The calculation process involves calculating attention weights and performing a weighted average, as follows:

[0102] Assume there is a set of PQDs feature data At the same time, a query vector is given. Through the scoring function Calculate each input and The correlation between PQDs is determined, and then the correlation scores are normalized using the Softmax function to obtain the attention distribution corresponding to PQDs. Finally, the input data is weighted and summed using the attention distribution to obtain the output result. The calculation formula is:

[0103] (12)

[0104] (13)

[0105] Attention scoring function , is the dimension of the input vector.

[0106] Multi-head attention (MA), compared to conventional attention mechanisms, allows the output of the attention layer to contain representational information from different subspaces, thereby enhancing the model's expressive power. It utilizes different query vectors... Multi-head attention structures focus on different parts of the input information to analyze it from various perspectives. Figure 4 As shown.

[0107] It mainly consists of three steps: First, input the extracted features. And then Perform a linear transformation, to Mapped to query space respectively Key space Sum value space Then, the scaled dot product and the Softmax function are used to calculate each attention distribution, and the attention distributions are weighted and summed to obtain the corresponding output. Finally, the multiple output results are concatenated using a splicing method. The formulas are shown in equations (14) to (18).

[0108] (14)

[0109] (15)

[0110] (16)

[0111] (17)

[0112] (18)

[0113] In the formula, C represents the fused signal characteristics. It is the i-th query vector q The weighted summation result over all keys; , , respectively query space Q Key space K Value space V The linear transformation parameters; A matrix consisting of the dimensions of each key; is a vector of elements in the key space; It is a vector of elements in the value space; This is a transpose transformation; q For query vector, The number of query vectors; For feature concatenation function; H is the dimension after linear transformation; H is the target time feature vector. ; It is the softmax function; Used to calculate the i query vectors and the j Key vectors The correlation between them.

[0114] 104. Classify the features of the fused signal using a preset classifier to obtain a final classification result, which is used to indicate the disturbance type of the power quality disturbance signal.

[0115] Finally, the fused signal feature C is input into a classifier for classification to identify the type of disturbance, i.e., setting... Figure 2 The classification module shown is responsible for receiving fused features and classifying them. It uses a Softmax classifier to classify the features and obtain the final classification result.

[0116] This invention proposes a noisy power quality disturbance identification model that integrates multiple feature information, which can effectively reduce the impact of noise on classification results and improve the model's classification performance and generalization ability.

[0117] The proposed method has better performance in terms of classification accuracy and noise resistance, providing effective support for power fault diagnosis in power systems.

[0118] This invention provides a method for identifying power quality disturbances. The method includes: acquiring a power quality disturbance signal; extracting features from the power quality disturbance signal using a pre-defined one-dimensional convolutional neural network and a one-dimensional residual neural network to obtain a target time feature vector of the power quality disturbance signal; performing feature fusion processing on the target time feature vector using a pre-defined multi-head attention mechanism to obtain fused signal features; and classifying the fused signal features using a pre-defined classifier to obtain a final classification result, which is used to indicate the disturbance type of the power quality disturbance signal. The proposed identification method can integrate multiple feature information, effectively reducing the impact of noise on the classification result, improving the classification performance and generalization ability of the classifier, and exhibiting better performance in terms of classification accuracy and noise resistance, thus improving the accuracy of disturbance identification and providing effective support for power fault diagnosis in power systems.

[0119] Please see Figure 5 , Figure 5 This is a structural block diagram of a power quality disturbance identification device according to an embodiment of the present invention, as shown below. Figure 5 The apparatus shown includes:

[0120] Signal acquisition module 501: used to acquire power quality disturbance signals;

[0121] Feature extraction module 502: used to extract features from the power quality disturbance signal using a preset one-dimensional convolutional neural network and a one-dimensional residual neural network, to obtain the target time feature vector of the power quality disturbance signal;

[0122] Attention fusion module 503: used to perform feature fusion processing on the target time feature vector using a preset multi-head attention mechanism to obtain fused signal features;

[0123] Disturbance identification module 504: used to classify the features of the fused signal using a preset classifier to obtain a final classification result, the final classification result being used to indicate the disturbance type of the power quality disturbance signal.

[0124] It should be noted that, Figure 5 The contents of each module in the identification device shown are the same as those in the identification device shown. Figure 1 The steps in the identification method shown are similar, and will not be repeated here to avoid repetition. For details, please refer to [link / reference needed]. Figure 1 The content of each step in the identification method shown.

[0125] This invention provides a device for identifying power quality disturbances. The device includes: a signal acquisition module for acquiring power quality disturbance signals; a feature extraction module for extracting features from the power quality disturbance signals using a preset one-dimensional convolutional neural network and a one-dimensional residual neural network to obtain a target time feature vector of the power quality disturbance signals; an attention fusion module for performing feature fusion processing on the target time feature vector using a preset multi-head attention mechanism to obtain fused signal features; and a disturbance identification module for classifying the fused signal features using a preset classifier to obtain a final classification result, wherein the final classification result indicates the disturbance type of the power quality disturbance signal. The proposed identification method can integrate multiple feature information, effectively reducing the impact of noise on the classification results, improving the classification performance and generalization ability of the classifier, and exhibiting better performance in terms of classification accuracy and noise resistance, thus improving the accuracy of disturbance identification and providing effective support for power fault diagnosis in power systems.

[0126] Figure 6 An internal structural diagram of a computer device in one embodiment is shown. This computer device can specifically be a terminal or a server. Figure 6 As shown, the computer device includes a processor, memory, and a network interface connected via a system bus. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system and may also store a computer program, which, when executed by the processor, causes the processor to perform the aforementioned methods. The internal memory may also store a computer program, which, when executed by the processor, causes the processor to perform the aforementioned methods. Those skilled in the art will understand that… Figure 6 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0127] In one embodiment, a computer device is provided, including a memory and a processor, the memory storing a computer program that, when executed by the processor, causes the processor to perform actions such as... Figure 1 The steps of the method shown.

[0128] In one embodiment, a computer-readable storage medium is provided storing a computer program that, when executed by a processor, causes the processor to perform the following actions: Figure 1 The steps of the method shown.

[0129] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments described above. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and RAMbus dynamic RAM (RDRAM), etc.

[0130] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0131] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this patent application should be determined by the appended claims.

Claims

1. A method for identifying power quality disturbances, characterized in that, The method includes: Acquire power quality disturbance signals; The power quality disturbance signal is subjected to feature extraction using a pre-defined one-dimensional convolutional neural network and a one-dimensional residual neural network to obtain the target time feature vector of the power quality disturbance signal; The target temporal feature vector is fused using a preset multi-head attention mechanism to obtain fused signal features; The fused signal features are classified using a preset classifier to obtain a final classification result, which is used to indicate the type of disturbance in the power quality disturbance signal. Wherein, the power quality disturbance signal includes a noise-reduced signal, then acquiring the power quality disturbance signal includes: Acquire power quality disturbance signals under noisy environments; The power quality disturbance signal is decomposed using wavelet decomposition to obtain the J-level wavelet coefficients of the power quality disturbance signal; The wavelet coefficients are denoised using a preset improved adaptive threshold algorithm and an improved threshold function algorithm to obtain denoised wavelet coefficients. The signal is reconstructed based on the denoised wavelet coefficients to obtain the denoised signal of the power quality disturbance signal; The step of using a preset improved adaptive threshold algorithm and an improved threshold function algorithm to perform denoising processing on the wavelet coefficients to obtain denoised wavelet coefficients includes: Based on the J-layer wavelet coefficients and the preset standard deviation algorithm, the noise standard deviation is determined; The wavelet threshold corresponding to each layer of wavelet coefficients is determined by using the peak-to-ratio correction factor of each layer of wavelet coefficients, the noise standard deviation, and the preset improved adaptive threshold algorithm. The wavelet coefficients are denoised using the wavelet threshold and improved threshold function algorithm for each layer to determine the denoised wavelet coefficients; The step of determining the wavelet threshold corresponding to each layer of wavelet coefficients by utilizing the peak-to-ratio correction factor of each layer of wavelet coefficients, the noise standard deviation, and a preset improved adaptive threshold algorithm, further includes: The peak-to-sum ratio correction factor for each layer of wavelet coefficients is determined by using the length of each layer of wavelet coefficients and the ratio of peak value to sum value in the wavelet coefficients. The improved threshold function algorithm includes a first denoising algorithm, a second denoising algorithm, and a third denoising algorithm. The step of using the wavelet threshold and the improved threshold function algorithm at each layer to denoise the wavelet coefficients and determine the denoised wavelet coefficients includes: The adjustable parameters of the wavelet coefficients at each level are determined using a preset adjustable parameter determination rule. The adjustable parameter determination rule includes at least the rule that the higher the number of decomposition levels, the lower the adjustable parameters. For each layer of wavelet coefficients and wavelet threshold, the following processing is performed: If the wavelet coefficients are greater than or equal to the wavelet threshold, then the denoised wavelet coefficients are obtained using the first denoising algorithm, the adjustable parameters, the wavelet coefficients, and the wavelet threshold. If the absolute value of the wavelet coefficient is less than the wavelet threshold, then the denoised wavelet coefficient is obtained by using the second denoising algorithm, the adjustable parameter, the wavelet coefficient, and the wavelet threshold. If the wavelet coefficients are less than or equal to the negative value of the wavelet threshold, then the denoised wavelet coefficients are obtained using the third denoising algorithm, the adjustable parameters, the wavelet coefficients, and the wavelet threshold.

2. The identification method according to claim 1, characterized in that, The step of extracting features from the power quality disturbance signal using a pre-defined one-dimensional convolutional neural network and a one-dimensional residual neural network to obtain the target time feature vector of the power quality disturbance signal includes: Based on the one-dimensional convolutional neural network, feature extraction is performed on the power quality disturbance signal to obtain the first time feature vector of the power quality disturbance signal; Based on the one-dimensional residual neural network, feature extraction is performed on the power quality disturbance signal to obtain the second time feature vector of the power quality disturbance signal; The target time feature vector of the power quality disturbance signal is obtained by concatenating the first time feature vector and the second time feature vector.

3. The identification method according to claim 2, characterized in that, The one-dimensional convolutional neural network includes multiple convolutional layers and fully connected layers connected in series; then, the feature extraction of the power quality disturbance signal based on the one-dimensional convolutional neural network to obtain the first time feature vector of the power quality disturbance signal includes: The power quality disturbance signal is input into the convolutional layer and the fully connected layer, and processed sequentially by the convolutional layer and the fully connected layer to obtain the first time feature vector of the power quality disturbance signal.

4. The identification method according to claim 2, characterized in that, The one-dimensional residual neural network includes several convolutional layers, residual connection layers, and fully connected layers connected in series. The step of extracting features from the power quality disturbance signal based on the one-dimensional residual neural network to obtain the second time feature vector of the power quality disturbance signal includes: The power quality perturbation signal is input into the convolutional layer to obtain the signal features after convolution processing; The power quality disturbance signal and the signal features after convolution are input into the residual connection layer to obtain the signal features after residual connection. The signal features after residual connection are input into the fully connected layer to obtain the second time feature vector.

5. The identification method according to claim 2, characterized in that, The target time feature vector is as follows: ； In the formula, B is the first time-time feature vector, and A is the second time-time feature vector. for A The amount; for B The amount; This is the target time feature vector.

6. The identification method according to claim 1, characterized in that, The step of using a preset multi-head attention mechanism to perform feature fusion processing on the target temporal feature vector to obtain fused signal features includes: By utilizing the multi-head attention mechanism, the features of the target temporal feature vector at different levels and angles are captured simultaneously and fused together to obtain fused signal features.

7. The identification method according to claim 6, characterized in that, The characteristics of the fused signal are as follows: In the formula, C represents the fused signal characteristics. It is the i-th query vector q The weighted summation result over all keys; respectively query space Q Key space K Value space V The linear transformation parameters; M k A matrix consisting of the dimensions of each key; k is a vector of elements in the key space; v It is a vector of elements in the value space; T This is a transpose transformation; q For query vector, The number of query vectors; For feature concatenation function; H is the dimension after linear transformation; H is the target time feature vector. ; It is the softmax function; Used to calculate the i query vectors q i and the j Key vectors k j The correlation between them.

8. A device for identifying power quality disturbances, characterized in that, The method of any one of claims 1 to 7 is applied to the apparatus, the apparatus comprising: Signal acquisition module: used to acquire power quality disturbance signals; Feature extraction module: used to extract features from the power quality disturbance signal using a preset one-dimensional convolutional neural network and a one-dimensional residual neural network, to obtain the target time feature vector of the power quality disturbance signal; Attention fusion module: used to perform feature fusion processing on the target temporal feature vector using a preset multi-head attention mechanism to obtain fused signal features; Disturbance identification module: used to classify the features of the fused signal using a preset classifier to obtain a final classification result, which is used to indicate the disturbance type of the power quality disturbance signal.

9. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it causes the processor to perform the steps of the method as described in any one of claims 1 to 7.

10. A computer device, comprising a memory and a processor, characterized in that, The memory stores a computer program that, when executed by the processor, causes the processor to perform the steps of the method as described in any one of claims 1 to 7.

Images