Electric energy quality disturbance identification method and device, electronic equipment and storage medium

Abstract

The embodiment of the invention discloses a power quality disturbance identification method and device, electronic equipment and a storage medium, the method is applied to a power quality disturbance identification system, a time-frequency information extraction module of the system adopts an adaptive incomplete S transformation algorithm, and a shallow double-path neural network module of the system adopts a double-path structure; the method comprises the following steps: acquiring an electric energy quality disturbance signal in a preset time period; determining time-frequency information of the signal through an adaptive incomplete S transformation algorithm; the time-frequency information is processed in parallel through a double-path structure, and target disturbance characteristics of the signal are obtained; determining a disturbance category of the signal based on the target disturbance characteristic to obtain a plurality of disturbance categories, each disturbance category corresponding to a disturbance probability; determining a loss value of each disturbance category based on the plurality of disturbance categories and the disturbance probability corresponding to each disturbance category to obtain a plurality of loss values; and taking the disturbance category corresponding to the minimum value in the plurality of loss values as a target disturbance category of the power quality disturbance signal.

Description

Technical Field

[0001] The present invention relates to the field of power quality technology, and in particular to a power quality disturbance identification method, device, electronic equipment and storage medium. Background Art

[0002] Modern industrial users' requirements for power quality are mainly reflected in three aspects: high reliability, rapid recovery capability and customized services. Among them, rapid recovery capability requires that the power system can quickly detect and respond to power quality disturbances, and provide necessary protection and recovery measures to reduce production interruption time and economic losses. Therefore, identifying power quality disturbances is a very important step in power grid operation.

[0003] However, when the power grid processes a large amount of data, the real-time power quality detection system has low computational efficiency for various disturbance types, which affects the real-time performance of power quality detection.

[0004] Therefore, there is an urgent need for a power quality disturbance identification method, device, electronic device and storage medium to improve the calculation efficiency when detecting power quality. Summary of the invention

[0005] In order to solve the above problems, the embodiments of the present invention provide a method, device, electronic device and storage medium for identifying power quality disturbances, which can improve the calculation efficiency when detecting power quality.

[0006] In a first aspect, an embodiment of the present invention provides a method for identifying power quality disturbances, the method being applied to a power quality disturbance identification system, the system comprising a time-frequency information extraction module and a shallow dual-path neural network module, the time-frequency information extraction module using an adaptive incomplete S-transformation algorithm to extract time-frequency information, the shallow dual-path neural network module using a dual-path structure; the method comprising:

[0007] Obtaining a power quality disturbance signal within a preset time period;

[0008] Determining the time-frequency information of the power quality disturbance signal by using the adaptive incomplete S-transform algorithm;

[0009] Processing the time-frequency information in parallel through the dual-path structure to obtain target disturbance characteristics of the power quality disturbance signal;

[0010] Determining the disturbance category of the power quality disturbance signal based on the target disturbance feature to obtain a plurality of disturbance categories, each disturbance category corresponding to a disturbance probability;

[0011] Determine a loss value of each disturbance category based on the multiple disturbance categories and the disturbance probability corresponding to each disturbance category to obtain multiple loss values;

[0012] The disturbance category corresponding to the minimum value among the multiple loss values ​​is used as the target disturbance category of the power quality disturbance signal.

[0013] In a second aspect, an embodiment of the present invention provides a power quality disturbance identification device, the device is applied to a power quality disturbance identification system, the system includes a time-frequency information extraction module and a shallow dual-path neural network module, the time-frequency information extraction module uses an adaptive incomplete S-transformation algorithm to extract time-frequency information, and the shallow dual-path neural network module uses a dual-path structure; the device includes an acquisition unit and a processing unit;

[0014] The acquisition unit is used to acquire the power quality disturbance signal within a preset time period;

[0015] The processing unit is used to determine the time-frequency information of the power quality disturbance signal by using the adaptive incomplete S-transform algorithm;

[0016] Processing the time-frequency information in parallel through the dual-path structure to obtain target disturbance characteristics of the power quality disturbance signal;

[0017] Determining the disturbance category of the power quality disturbance signal based on the target disturbance feature to obtain a plurality of disturbance categories, each disturbance category corresponding to a disturbance probability;

[0018] Determine a loss value of each disturbance category based on the multiple disturbance categories and the disturbance probability corresponding to each disturbance category to obtain multiple loss values;

[0019] The disturbance category corresponding to the minimum value among the multiple loss values ​​is used as the target disturbance category of the power quality disturbance signal.

[0020] In a third aspect, an embodiment of the present invention provides an electronic device, comprising a processor and a memory, wherein the processor is connected to the memory, the memory is used to store a computer program, and the processor is used to execute the computer program stored in the memory, so that the electronic device performs the method described in the first aspect.

[0021] In a fourth aspect, an embodiment of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and the computer program is executed by a processor to implement the method as described in the first aspect.

[0022] In a fifth aspect, an embodiment of the present application provides a computer program product, which includes a non-transitory computer-readable storage medium storing a computer program, and the computer is operable to cause the computer to execute the method described in the first aspect.

[0023] Implementing the embodiments of the present application has the following beneficial effects:

[0024] In the implementation mode of the present application, the power quality disturbance signal within the preset time period is first obtained, and then the time-frequency information of the power quality disturbance signal is determined by an adaptive incomplete S transform algorithm, and the time-frequency information is processed in parallel by a dual-path structure to obtain the target disturbance feature of the power quality disturbance signal. Next, the disturbance category of the power quality disturbance signal is determined based on the target disturbance feature to obtain multiple disturbance categories, each disturbance category corresponds to a disturbance probability, and then the loss value of each disturbance category is determined based on the multiple disturbance categories and the disturbance probability corresponding to each disturbance category to obtain multiple loss values, and the disturbance category corresponding to the minimum value of the multiple loss values ​​is used as the target disturbance category of the power quality disturbance signal. Thus, the time-frequency information extraction is realized by an adaptive incomplete S transform algorithm, and the target disturbance feature of the power quality disturbance signal is obtained by processing the time-frequency information in parallel by a dual-path structure, and the target disturbance category is determined based on the target disturbance feature, which can improve the calculation efficiency when detecting power quality. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the background technology, the drawings required for use in the embodiments of the present invention or the background technology will be described below. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0026] Figure 1 It is a schematic diagram of the architecture of a power quality disturbance identification system provided in an embodiment of the present application;

[0027] Figure 2 is a flow chart of a method for identifying power quality disturbances provided in an embodiment of the present application;

[0028] Figure 3 It is a schematic diagram of the structure of a shallow dual-path neural network provided in an embodiment of the present application;

[0029] Figure 4 It is a structural schematic diagram of a power quality disturbance identification device provided in an embodiment of the present application;

[0030] Figure 5 It is a structural schematic diagram of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0031] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0032] The terms "first", "second", "third" and "fourth" etc. in the specification and claims of the present application and the drawings are used to distinguish different objects, rather than to describe a specific order. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or modules is not limited to the listed steps or modules, but optionally includes steps or modules that are not listed, or optionally includes other steps or modules inherent to these processes, methods, products or devices.

[0033] Reference to "embodiments" herein means that a particular feature, result, or characteristic described in conjunction with the embodiments may be included in at least one embodiment of the present application. The appearance of the phrase in various locations in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0034] The power quality disturbance identification method provided in the embodiment of the present application is applied to the power quality disturbance identification system, which includes a time-frequency information extraction module and a shallow dual-path neural network module, wherein the time-frequency information extraction module uses an adaptive incomplete S-transformation algorithm to realize time-frequency information extraction, and the shallow dual-path neural network module uses a dual-path structure. Figure 1 , Figure 1 This is a schematic diagram of the architecture of a power quality disturbance identification system provided by an embodiment of the present application. In this system, the frequency domain information in the power disturbance signal is extracted by the time-frequency information extraction module, and then the power quality disturbance is effectively identified by the shallow dual-path neural network module, thereby obtaining categories such as temporary rise and temporary drop. Figure 1 The time-frequency information extracted from the power quality disturbance signal by the time-frequency information extraction module is shown in the adaptive incomplete S-transform time-frequency domain amplitude curve in the figure. When the shallow dual-path neural network module uses a dual-path structure to process the time-frequency information, the time-frequency information of the two paths can be processed, integrated and classified by the following formula: Among them, h t is the time-frequency information, x kis the time-frequency information h of one of the paths in the dual-path structure t The processing result, y k is the time-frequency information h of the other path in the dual-path structure t The processing result, r k For x k and k The integration result, h k For r k The classification results.

[0035] In a possible embodiment, the power quality problems that may occur in the power grid include transient disturbance signals and steady-state disturbance signals. Transient disturbance signals include voltage sag, voltage swell, voltage interruption and transient pulses. Steady-state disturbance signals include harmonics, interharmonics, voltage fluctuations and flicker, etc. Among them, voltage sag refers to a temporary decrease in the effective value of the power supply voltage, which usually drops to less than 90% of the rated voltage and lasts from half a cycle to 1 minute. It is mainly caused by short-circuit faults in the power grid, starting of large motors, transformer tap switching, etc., and may cause some equipment to stop working or fail. Voltage swell refers to a temporary increase in the effective value of the power supply voltage, which exceeds 110%-180% of the rated voltage and lasts for half a cycle to 1 minute. It is mainly caused by the input of capacitive loads in the power grid, the recovery process after the power system fault, lightning strikes, etc., and may cause accelerated aging of the insulation of electrical equipment. For some voltage-sensitive electronic equipment, such as computer power supplies, precision instruments, etc., it may cause equipment damage or abnormal operation. Voltage interruption refers to the complete disappearance of the power supply voltage for a period of time, usually more than 1 minute. It is mainly caused by power grid failures and planned power outages, which will cause most electrical equipment to stop working. For some equipment without backup power, it may cause serious consequences such as data loss and equipment damage. Transient pulses are extremely short-duration voltage or current pulses with high amplitudes. They are mainly caused by switching operations of electrical equipment, lightning induction, electrostatic discharge, etc., and may damage semiconductor components of electronic equipment because these components are sensitive to overvoltage pulses. Harmonics refer to sinusoidal components with frequencies that are integer multiples of the fundamental frequency. They are mainly caused by nonlinear loads in the power system, such as rectifiers, inverters, arc furnaces, etc., which will increase the loss of equipment and cause equipment heating. For communication systems, harmonics may generate electromagnetic interference and affect communication quality. Interharmonics refer to sinusoidal components with frequencies that are non-integer multiples of the fundamental frequency. They are mainly generated by equipment such as variable frequency speed regulation devices and induction motors, which will cause problems such as light flickering and motor torque pulsation. Voltage fluctuation refers to a series of continuous changes in the effective value of voltage, usually characterized by changes in the root mean square value of voltage. Flicker refers to a subjective visual reaction of the human eye to the flickering sensation of lights, which is mainly caused by voltage fluctuations, mainly caused by frequent starting and stopping of loads with large fluctuations, switching of capacitor banks, etc., which will affect the normal lighting of lighting equipment. For some voltage-sensitive equipment, such as precision instruments, electronic equipment, etc., voltage fluctuations may affect their working performance.

[0036] See also Figure 2 , Figure 2 is a flow chart of a method for identifying power quality disturbances provided in an embodiment of the present application, such as Figure 2 As shown, the power quality disturbance identification method provided in the embodiment of the present application includes but is not limited to the following steps:

[0037] Step S101: obtaining a power quality disturbance signal within a preset time period;

[0038] Step S102: determining the time-frequency information of the power quality disturbance signal by using an adaptive incomplete S-transform algorithm;

[0039] Step S103: Processing the time-frequency information in parallel through a dual-path structure to obtain target disturbance characteristics of the power quality disturbance signal;

[0040] Step S104: determining the disturbance category of the power quality disturbance signal based on the target disturbance feature, obtaining multiple disturbance categories, each disturbance category corresponding to a disturbance probability;

[0041] Step S105: determining a loss value of each disturbance category based on multiple disturbance categories and a disturbance probability corresponding to each disturbance category, to obtain multiple loss values;

[0042] Step S106: taking the disturbance category corresponding to the minimum value among the multiple loss values ​​as the target disturbance category of the power quality disturbance signal.

[0043] In a possible embodiment, the power quality disturbance signal within a preset time period can be obtained by a power system monitoring device. The monitoring device can be a voltage transformer and a current transformer. The voltage transformer and the current transformer can convert high voltage and high current signals into low voltage and low current signals suitable for measurement and processing. These transformers are installed at substations or user access points to obtain grid voltage and current signals in real time. The monitoring device can also be a smart meter or a power quality monitor. The smart meter has a built-in data acquisition module that can collect voltage and current signals at a certain sampling frequency. It can not only record the user's power consumption, but also obtain power quality related information such as voltage and current. Smart meters installed at residential areas or industrial and commercial user ends can obtain a large amount of terminal power quality data. The power quality monitor is used to monitor power quality, and can simultaneously measure multiple power quality parameters, such as voltage effective value, current effective value, harmonic content, three-phase imbalance, frequency deviation, etc. It has a higher sampling frequency and more accurate measurement function, and can comprehensively monitor power quality disturbances. In addition, historical data can also be obtained through a data recording device as a power quality disturbance signal. The data recording device can be a fault recorder, which is used to record the change process of electrical quantities during power system faults, such as waveform data of voltage, current and other signals before, during and after the fault.

[0044] In a possible embodiment, before obtaining the power quality disturbance signal within a preset time period, the type, accuracy, range, monitoring location and data acquisition frequency of the monitoring equipment are determined. For high-voltage power grids, electromagnetic or capacitive voltage transformers are used, and for low-voltage power grids, small voltage sensors are used. The monitoring location needs to accurately reflect the voltage situation at the monitoring point, and is generally set at the busbar of the substation, the incoming line end of important users, etc. The monitoring locations of the voltage transformer and the current transformer correspond. When determining the data acquisition frequency, the appropriate data acquisition frequency is determined according to the characteristics of the power quality problem. For transient problems such as voltage sags and swells, a higher acquisition frequency is determined. For problems such as harmonics and flicker, the acquisition frequency can be relatively low.

[0045] In a possible embodiment, the time-frequency information of the power quality disturbance signal is determined by an adaptive incomplete S transform algorithm, wherein the adaptive incomplete S transform algorithm is mainly used for time-frequency analysis, and the time-frequency information can be in various forms according to the specific implementation method and application requirements. The time-frequency information can be a time-frequency matrix, which can be regarded as a feature map, wherein the row represents time, the column represents frequency, and the elements in the matrix represent the amplitude or other relevant information of the signal at a specific time and frequency. The time-frequency information can also be further processed frequency domain information, such as only extracting the main frequency components and their corresponding amplitude lists, or energy distribution data according to frequency interval statistics. If it is necessary to intuitively display the changes of the signal in the two dimensions of time and frequency, the output time-frequency information is a time-frequency matrix, that is, a feature map. For certain key parameters in the frequency domain, such as the amplitude of a specific harmonic, the size of the frequency deviation, etc., the output time-frequency information can be filtered and refined frequency domain data, rather than a complete feature map. For example, when it is only necessary to determine whether there is harmonic interference of a specific frequency, the adaptive incomplete S transform can only output the amplitude information of these harmonic frequency points.

[0046] In a possible embodiment, for discrete signals, the time-frequency coefficients are calculated according to the discrete formula of the adaptive incomplete S transform. Specifically, at each time point and frequency point, the signal is multiplied by the Gaussian window function, and then the product is discretely Fourier transformed to obtain the time-frequency coefficients. In the calculation process, the window size and shape are adaptively adjusted according to the frequency of the signal. Generally speaking, for the low-frequency part, the window is wider to obtain better frequency resolution, which can more accurately analyze the frequency components of the low-frequency signal. For the high-frequency part, the window is narrower, providing better time resolution, which is conducive to capturing the rapid changes of high-frequency signals such as transient pulses. In the process of calculating the time-frequency coefficients, incomplete transformation can also be performed according to preset rules. For example, by truncating or ignoring some transformation parts that are not very important for analyzing power quality disturbances, or using different weights for the transformation results of different parts, the amount of calculation can be reduced and key information can be highlighted. For example, the weak noise component of the high-frequency part can be appropriately ignored in the incomplete transformation, while a higher weight is determined for the main harmonic and transient pulse components. Then, the calculated time-frequency coefficients are combined into a time-frequency matrix, where the rows of the matrix correspond to different time points, the columns correspond to different frequency points, and each element in the time-frequency matrix represents the amplitude or energy of the signal at time and frequency. In addition to the time-frequency matrix, the main frequency domain features can also be extracted from the time-frequency coefficients. For example, determine the fundamental frequency and its amplitude, the frequency and amplitude of each harmonic, the main frequency range and energy of the transient pulse, etc. These main frequency domain features can be obtained by searching for spectrum peaks in the time-frequency matrix, statistical energy distribution, etc. For example, the frequency of the harmonic is determined by finding the frequency point with a larger amplitude in the time-frequency matrix, and the energy of the harmonic is calculated by integrating or summing the amplitudes of the corresponding frequency points.

[0047] In a possible embodiment, the output of the adaptive incomplete S transform can be visualized by using a color map instead of a grayscale map. By using different color channels to represent different features or value ranges, multiple information can be presented simultaneously. For example, in the visualization of the time-frequency matrix, the red channel can be used to represent the amplitude, the green channel to represent the frequency change rate, and the blue channel to represent the phase information of the signal. In this way, the color of a pixel can comprehensively reflect information of multiple dimensions. Compared with a grayscale map, the change of color is more intuitive and can more keenly capture the gradient of the change of the amplitude.

[0048] In a possible embodiment, the time-frequency information is processed in parallel by a dual-path structure to obtain the target disturbance feature of the power quality disturbance signal. The shallow dual-path neural network can process a variety of data types, depending on the application scenario. The shallow dual-path neural network can process the feature map obtained from the adaptive incomplete S transform, and can also process other forms of data. For example, it can receive a preprocessed frequency domain feature vector, which may be composed of the main frequency domain features extracted by the adaptive incomplete S transform, such as frequency peak, harmonic content and other values, or it can also process the vector after the fusion of time domain features and frequency domain features, where the time domain feature can be the peak value, mean value, etc. of the signal, and the frequency domain feature is the refined frequency domain information mentioned above. The network structure of the shallow dual-path neural network is adaptable. According to the type of input data, the structure of the shallow dual-path neural network can also be adjusted. If the input is a feature map, the network may include a convolution layer to extract the features in the map. If the input is a feature vector, the network mainly uses a fully connected layer to process the data. For example, when the input is a frequency domain feature vector, the network can classify these features through multiple layers of fully connected layers to determine the relationship between different frequency domain feature combinations and power quality disturbance types.

[0049] In a possible embodiment, for time-frequency information, one path in the dual-path structure can process the time domain information in the time-frequency information, and the other path processes the frequency domain information in the time-frequency information. If the power quality disturbance signal contains a variety of complex frequency domain features, both paths in the dual-path structure can be used to process the frequency domain information to mine the frequency domain features from different angles. For example, in an electric power system environment with both complex harmonics and frequent transient pulses, the two paths can process the frequency domain features of harmonics and transient pulses respectively to better identify and classify power quality disturbances. One path can extract the harmonic features of the low-frequency band, such as the amplitude and phase of the fundamental and harmonic waves, and the other path can analyze the spectrum distribution, energy proportion and other characteristics of special frequency components such as transient pulses in the high-frequency band. The two paths can use different algorithms or techniques to process frequency domain information. For example, one path uses spectrum analysis methods, such as amplitude statistics after fast Fourier transform, to obtain the intensity of the main frequency components. The other path can use time-frequency analysis methods such as wavelet transform to decompose in the frequency domain and extract frequency domain features at different scales, such as the frequency domain range information corresponding to each layer of coefficients after wavelet decomposition, so as to perform a deeper analysis of the frequency domain signal.

[0050] In a possible embodiment, the disturbance category of the power quality disturbance signal is determined based on the target disturbance feature to obtain multiple disturbance categories, each disturbance category corresponds to a disturbance probability, and the loss value of each disturbance category is determined based on the multiple disturbance categories and the disturbance probability corresponding to each disturbance category to obtain multiple loss values, and the disturbance category corresponding to the minimum value of the multiple loss values ​​is used as the target disturbance category of the power quality disturbance signal.

[0051] In a possible embodiment, voltage sag and voltage swell are mainly manifested as a sudden drop or rise in voltage amplitude. The start and end time, duration and amplitude of the voltage amplitude change can be directly determined through the time domain waveforms of the two. Although voltage sag and voltage swell are mainly time domain phenomena, auxiliary judgment can be made through frequency domain information. For example, voltage sag and voltage swell may cause changes in the harmonic characteristics of the system. When the voltage changes suddenly, due to the characteristics of the nonlinear elements of the power system, the harmonic content may increase or decrease. By analyzing the frequency domain information, the changes in harmonic amplitude and phase can be observed, so as to further understand the impact of voltage sudden changes on the system. For flicker, flicker is manifested as rapid fluctuations in voltage amplitude in the time domain. This fluctuation can cause flickering of loads such as lights. Through the time domain waveform, it can be determined that the voltage amplitude of the flicker presents periodic or non-periodic rapid changes. Flicker usually contains multiple frequency components, which may be caused by different fluctuation sources. By analyzing the frequency domain information, such as performing spectrum decomposition, the main frequency components and amplitudes of the flicker can be determined. For harmonics, in the frequency domain, harmonics are sinusoidal wave components with frequencies that are integer multiples of the fundamental frequency. By performing frequency domain analysis on the signal, the frequency, amplitude, and phase of each harmonic can be identified. Although harmonics are mainly analyzed through the frequency domain, time domain information can assist in determining the impact of harmonics on the waveform. When harmonics exist, the time domain waveform will be distorted and is no longer a simple sine wave. By determining the degree of distortion, symmetry, and other characteristics of the waveform, it can be determined whether there are harmonics and the general situation of the harmonics.

[0052] In the embodiment of the present application, the proposed power quality disturbance identification method is based on an adaptive incomplete S transform and a shallow dual-path neural network, wherein the adaptive incomplete S transform is a signal processing technology that improves the efficiency of feature extraction in the identification of power quality composite disturbances through an adaptive window incomplete S transform. This transformation method can adaptively adjust the window function according to the characteristics of the signal, thereby more accurately extracting the characteristics of the signal. For example, in the identification of power quality disturbances, the adaptive incomplete S transform can effectively extract the characteristics of the disturbance signal, thereby improving the recognition ability and learning speed of the classifier. The shallow dual-path neural network is a neural network structure that combines the ideas of residual group convolution and dense connection. Through the dual-path structure, the shallow dual-path neural network can reuse features using the skip connection of the residual network and continuously explore new features using dense connection paths. This structure performs well in tasks such as image classification, and can significantly reduce model size, computational cost, and memory consumption, while improving training speed and model performance. Combining these two technologies, signal data can be effectively processed, thereby effectively identifying and classifying. The adaptive incomplete S transform and shallow dual-path neural network are combined to improve the efficiency and accuracy of power quality disturbance identification. The adaptive incomplete S transform reduces the computational complexity by optimizing the transformation parameters, while retaining the key time-frequency features, which helps to extract more accurate disturbance features. The shallow dual-path neural network processes the time domain and frequency domain information respectively through two parallel paths, enhancing the model's feature learning ability and generalization performance, enabling the identification system to respond to various power quality disturbances more quickly and accurately, and improving the performance of real-time monitoring.

[0053] Optionally, the adaptive incomplete S transform algorithm includes a window function, which is a function that has a non-zero value in a finite interval and is zero outside the interval, and is used to intercept a section of the signal for analysis in time-frequency analysis. Window functions of different shapes have different characteristics in the time domain and the frequency domain. Step S102, determining the time-frequency information of the power quality disturbance signal by the adaptive incomplete S transform algorithm, may include the following steps:

[0054] Step S201: obtaining a frequency range of a power quality disturbance signal;

[0055] Step S202: determining an initial scale parameter range of a window function based on a frequency range; the window function is used to determine a calculation region of an adaptive incomplete S-transform algorithm on a power quality disturbance signal;

[0056] Step S203: determining a target indicator set and a target weight group; the target indicator set includes a time-frequency resolution trade-off loss indicator, a key frequency component detection accuracy indicator, and a computational cost indicator; the target weight group includes a first weight, a second weight, and a third weight; each indicator in the target indicator set corresponds to a corresponding weight in the target weight group in a one-to-one correspondence;

[0057] Step S204: Calculate based on the preset first calculation formula, target indicator set and target weight group to obtain a calculation result; the first calculation formula is: F=ω1L+ω2P+ω3C, where L, P and C are the time-frequency resolution trade-off loss indicator, the key frequency component detection accuracy indicator and the calculation cost indicator respectively, ω1, ω2 and ω3 are the first weight, the second weight and the third weight respectively, and F is the calculation result;

[0058] Step S205: determining the target scale parameter within the initial scale parameter range according to the calculation result;

[0059] Step S206: Determine the time-frequency information based on the adaptive incomplete S-transform algorithm, the target scale parameter and the power quality disturbance signal.

[0060] In a possible embodiment, the scale parameter is the width of the window function. When the scale parameter is large, the Gaussian window function is wide, and the coverage of the signal on the time axis is large, which can better capture the frequency characteristics of the signal in a long time. When the scale parameter is small, the Gaussian window function is narrower, and it focuses more on analyzing the transient changes of the signal in a short time, and has a stronger ability to capture high-frequency signals. Determining the target scale parameter of the window function can optimize the time-frequency resolution and improve the calculation efficiency. When processing power quality disturbance signals, the signal contains multiple frequency components. For low-frequency components such as fundamental waves and harmonics, the low-frequency signal changes relatively slowly over a long time range. The adaptive adjustment strategy of the adaptive incomplete S transform can make the window wider, and the wide window can provide better frequency resolution, so as to more accurately analyze the frequency, amplitude and phase information of these low-frequency components. For high-frequency components such as transient pulses, the high-frequency signal changes rapidly. The adaptive strategy of the adaptive incomplete S transform makes the window narrower, and the narrow window can provide higher time resolution, accurately capturing the key time points such as the start time, peak time and end time of the transient pulse. For frequency regions or signal components that have little impact on the overall analysis, the amount of calculation can be reduced by adjusting the scale parameters of the window. If a certain high-frequency band is determined to be mainly noise, and the current analysis task is to detect specific harmonic disturbances, then there is no need to analyze the frequency band in detail in the current analysis task. The adaptive strategy can make the window use a simpler calculation method in this frequency band or directly ignore part of the calculation, thereby saving computing resources and improving processing speed. For example, when analyzing transient fault signals in power systems, for the fault initiation stage that may contain high-frequency components, a smaller scale parameter is required to highlight the high-frequency information, while in the steady-state recovery stage after the fault, a larger scale parameter is helpful for analyzing low-frequency components such as the fundamental frequency and its harmonics.

[0061] In a possible embodiment, the frequency range of the power quality disturbance signal is obtained, and the initial scale parameter range corresponding to the window function in the frequency range is determined based on the frequency range and the correspondence between the frequency and the scale parameter. Frequency resolution refers to the minimum frequency interval that can distinguish two adjacent frequency components in spectrum analysis. It reflects the degree of refinement of the signal frequency component analysis in the frequency domain. The frequency resolution is proportional to the scale parameter, and a larger scale parameter corresponds to a higher frequency resolution. Time resolution refers to the minimum time interval that can distinguish two adjacent time events in the time domain. It reflects the accuracy of the analysis of the signal time change characteristics. The time resolution is inversely proportional to the scale parameter, and a smaller scale parameter corresponds to a higher time resolution.

[0062] In a possible embodiment, a target indicator set and a target weight group are determined, wherein the target indicator set includes a time-frequency resolution trade-off loss indicator, a key frequency component detection accuracy indicator, and a computational cost indicator, and the target weight group includes a first weight, a second weight, and a third weight, and each indicator in the target indicator set corresponds to the corresponding weight in the target weight group. The time-frequency resolution trade-off loss indicator involves a trade-off relationship between time resolution and frequency resolution. When the window function is narrow, the range of the signal intercepted in time is small, which can better capture the rapid changes of the signal in time, that is, the time resolution is high, but at the same time, due to the narrow window function, the frequency information contained is limited, resulting in a low frequency resolution. On the contrary, when the window function is wide, the frequency information contained is richer and the frequency resolution is high, but the time resolution will be reduced because the wider window function will blur the rapid changes of the signal in time. Therefore, in order to evaluate the degree of performance degradation caused by this trade-off relationship in the time-frequency analysis process, the time-frequency resolution trade-off loss is introduced as a quantitative indicator to measure the loss caused by the inability of time resolution and frequency resolution to achieve the optimal at the same time. Through the adaptive adjustment of the adaptive incomplete S transform, for fast-changing high-frequency transient signals, high time resolution is determined to determine the mutation moment of the signal, and for slowly changing low-frequency harmonics, high frequency resolution is determined to determine the harmonic components.

[0063] In a possible embodiment, the calculation is performed based on the preset first calculation formula, the target index set and the target weight group to obtain the calculation result; the first calculation formula is: F = ω1L + ω2P + ω3C, where L, P and C are the time-frequency resolution trade-off loss index, the key frequency component detection accuracy index and the calculation cost index, respectively, ω1, ω2 and ω3 are the first weight, the second weight and the third weight, respectively, and F is the calculation result, while pursuing to minimize the time-frequency resolution trade-off loss, maximize the detection accuracy of the key frequency component and minimize the calculation cost. Then, according to the calculation result, the target scale parameter is determined within the initial scale parameter range, and the time-frequency information is determined based on the adaptive incomplete S transform algorithm, the target scale parameter and the power quality disturbance signal.

[0064] In a possible embodiment, in the field of signal analysis, the Fourier transform is not effective in processing sudden change signals. The short-time Fourier transform may cause decomposition errors when processing long-period signals due to the fixed time window. Although the wavelet transform has good multi-resolution analysis capabilities, the selection of its mother wavelet and scaling function is more difficult. In contrast, the S transform combines the advantages of the short-time Fourier transform and the wavelet transform, can adaptively adjust the resolution, and its inverse transform is lossless and reversible. As a phase correction of the continuous wavelet transform, the S transform avoids the selection of the wavelet basis and ensures the accuracy of the data processing results. It inherits the respective advantages of the wavelet transform and the Fourier transform, and can extract the time-frequency characteristics of the signal. Compared with the wavelet transform, the S transform provides richer spectral information and is suitable for applications such as signal separation. For continuous signals, the definition of the S transform is shown in formula (1):

[0065]

[0066] Among them, S(τ, f) is the S transform of the signal x(t), ω(t-τ, f) is the window function used by the S transform, that is, the Gaussian window, t is the time variable, τ is the time shift parameter controlling the Gaussian window, and f represents the frequency. The window function can be expressed by formula (2):

[0067]

[0068] The Gaussian window is a frequency window, and its width and height can change with the frequency. Considering that in actual situations, the sampling signal of the sampling device is discrete, the formula of discrete S transform can be expressed by formula (3) for discrete signals:

[0069]

[0070] Where T is the sampling time interval; N is the number of sampling points; k, i, m, n = 0, 1, ..., N-1; W GS(m, n) is a Gaussian window function. The Fourier transform in formula (3) can be expressed by formula (4):

[0071]

[0072] In the discrete Fourier transform, F(k) represents the signal or spectrum of a signal in the frequency domain. Specifically, F(k) is the representation of the input signal x(n) in the frequency domain, where k is the frequency index. In the discrete Fourier transform, x(n) represents the signal or samples of a signal in the time domain. Specifically, x(n) is the representation of the input signal in the time domain, where n is the time or sample index. This formula represents the process of converting the time domain signal x(n) to the frequency domain.

[0073] In the embodiment of the present application, in terms of time-frequency analysis, the S transform can provide the time and frequency information of the signal at the same time, and is suitable for analyzing the time-frequency characteristics of non-stationary signals. Compared with the short-time Fourier transform, the Gaussian window width of the S transform varies with the frequency, which can better adapt to the analysis requirements of different frequency components. In terms of adaptive resolution, the Gaussian window width of the S transform varies with the frequency, the window is wider at low frequencies, and the window is narrower at high frequencies, thereby providing better frequency resolution in the low-frequency part and better time resolution in the high-frequency part. This adaptive resolution adjustment makes the S transform more flexible when processing signals with different frequency components. In terms of lossless reversibility, the inverse transform of the S transform is lossless, which means that the original signal can be completely restored from the result of the S transform. This feature is very important in signal processing, especially in situations where multiple transformations and inverse transformations of the signal are required.

[0074] In the embodiment of the present application, the adaptive incomplete S transform is mainly used to analyze and process the fault traveling wave signal. By performing S transform on the wave head of the fault traveling wave, the wave head of the fault traveling wave can be clearly identified to avoid the interference of noise on the wave head identification. The adaptive incomplete S transform is expressed by formula (5):

[0075]

[0076] Where S(x, f) is the adaptive incomplete S transform of x(t), x(t) is the time-frequency information, φ(t, f) is the window function, and φ * (t, f) is the conjugate of the window function. Because the window function uses frequency f as a variable, φ(t, f) has adaptive capabilities and can change with changes in frequency.

[0077] In the embodiment of the present application, the adaptive incomplete S transform technology can be used to extract the characteristics of the fault type. By performing an S transform on the fault signal, the characteristics of the signal at different frequencies can be analyzed, thereby extracting the characteristics related to the fault type, which is very important for power system fault detection and classification, because accurate feature extraction can help distinguish different types of faults. The adaptive S transform can effectively suppress noise and improve the signal-to-noise ratio of the signal through the selection and adjustment of the Gaussian window function, which is very critical for extracting useful signals in a noisy environment. By suppressing noise, the characteristics of the fault signal can be more clearly extracted, thereby improving the accuracy of fault type identification. By selecting a specific frequency point for S transform, the characteristics of the signal at a specific frequency can be targeted, which is very helpful for analyzing the frequency components and characteristics of the signal. Through frequency selective analysis, frequency characteristics related to the fault type can be extracted, thereby achieving accurate classification of the fault type. The adaptive S transform can strike a balance between computational efficiency and signal processing accuracy through the combination of discrete Fourier transform and Gaussian window function, and is suitable for real-time signal processing applications. Efficient signal processing methods can ensure that the characteristics of the fault type are quickly extracted in real-time applications, thereby achieving fast and accurate fault detection and classification.

[0078] Optionally, step S205, determining the target scale parameter within the initial scale parameter range according to the calculation result, may include the following steps:

[0079] Step S301: determining characteristic parameters of a power quality disturbance signal, the characteristic parameters including instantaneous frequency or instantaneous amplitude;

[0080] Step S302: determining the change rate of the characteristic parameter within a preset time period; the change rate is used to reflect the severity of the change of the power quality disturbance signal within the preset time period;

[0081] Step S303: determining a scale adjustment parameter range within the initial scale parameter range according to the change rate;

[0082] Step S304: determining a target scale parameter within the scale adjustment parameter range based on the calculation result.

[0083] In a possible embodiment, characteristic parameters of a power quality disturbance signal are determined, the characteristic parameters include instantaneous frequency or instantaneous amplitude, and the rate of change of the characteristic parameters within a preset time period is determined, wherein the rate of change is used to reflect the severity of the change of the power quality disturbance signal within the preset time period, and the scale adjustment parameter range is determined in the initial scale parameter range according to the rate of change, and the target scale parameter is determined in the scale adjustment parameter range based on the calculation result. Among them, the scale adjustment parameter range is determined according to the calculated characteristic parameter change rate. If the rate of change is large, it means that the signal changes violently and requires a higher time resolution. At this time, the scale parameter range should be narrowed so that the analysis method focuses more on time localization. When determining the target scale parameter in the scale adjustment parameter range, within the determined scale adjustment parameter range, each possible scale parameter is analyzed and calculated, and the calculation result must be the minimum value at the same time. In the embodiment of the present application, the scale adjustment parameter range is determined according to the characteristic parameter change rate, avoiding comprehensive calculation within the entire initial scale parameter range. By determining the target scale parameter in the narrowed scale adjustment parameter range in a targeted manner, unnecessary calculation amount is reduced and calculation efficiency is improved, especially when processing power quality data of a long time series, the calculation time can be significantly shortened.

[0084] Optionally, the dual-path structure includes a first preset path and a second preset path. Step S103, in which the time-frequency information is processed in parallel by the dual-path structure to obtain a target disturbance feature of the power quality disturbance signal, may include the following steps:

[0085] Step S401: Determine the initial dimension parameters of time-frequency information;

[0086] Step S402: determining a first dimension parameter of the time-frequency information in the first preset path and a second dimension parameter of the time-frequency information in the second preset path based on the initial dimension parameter;

[0087] Step S403: splitting the time-frequency information based on the first dimension parameter and the second dimension parameter to obtain the first time-frequency information and the second time-frequency information;

[0088] Step S404: determining a first path weight of the first preset path and a second path weight of the second preset path based on the first time-frequency information and the second time-frequency information;

[0089] Step S405: Convolve the first time-frequency information and the second time-frequency information based on the first path weight and the second path weight respectively to obtain a first disturbance feature and a second disturbance feature;

[0090] Step S406: Integrate the first disturbance feature and the second disturbance feature to obtain a target disturbance feature.

[0091] In a possible embodiment, the dual-path structure is composed of two independent processing paths, namely a first preset path and a second preset path. The two paths can process the input data at the same time. One path can specifically process time domain information, and the other path processes frequency domain information. The respective processing results are merged in the output layer or the intermediate fusion layer of the network to obtain the final output. Each path contains multiple neural network layers, such as convolutional layers, pooling layers, and fully connected layers, wherein the convolutional layers are used to extract local features, the pooling layers are used to reduce data dimensions and extract main features, and the fully connected layers are used to integrate feature information. These layers can be configured with different parameters in different paths according to the type of features being processed. For example, in the time domain path, the convolution kernel size and step size of the convolution layer are set according to the characteristics of the time series data to better determine the changing characteristics of the signal on the time axis, while in the frequency domain path, these parameters are optimized according to the extraction and processing of frequency information.

[0092] In a possible embodiment, the initial dimension parameters of the time-frequency information are determined, and the first dimension parameters of the time-frequency information in the first preset path and the second dimension parameters in the second preset path are determined based on the initial dimension parameters, and the sum of the first dimension parameters and the second dimension parameters is the initial dimension parameters. The time-frequency information is split based on the first dimension parameters and the second dimension parameters to obtain the first time-frequency information and the second time-frequency information. When the time-frequency information enters the dual-path neural network, the time-frequency information will be divided into two parts, the first time-frequency information and the second time-frequency information, and enter the first preset path and the second preset path respectively. In this process, the splitting method of the time-frequency information depends on the design of the network and the type of input data. For example, if the input is a time-frequency matrix, the time domain path receives information related to the time series in the matrix, such as the comprehensive amplitude information at each time point, and the frequency domain path receives information related to the frequency component, such as the amplitude distribution at each frequency point, then the first dimension parameter and the second dimension parameter can be determined according to the dimension related to the time series and the dimension related to the frequency component in the initial dimension parameters of the time-frequency information.

[0093] In a possible embodiment, initial dimensional parameters are determined based on the time-frequency matrix in the time-frequency information, and first dimensional parameters and second dimensional parameters are determined based on the separation target of the frequency range, such as separation based on low frequency and high frequency, separation based on harmonic orders, separation based on steady-state and transient characteristics, or separation based on different devices or interference sources.

[0094] In a possible embodiment, the first path weight of the first preset path and the second path weight of the second preset path are determined based on the first time-frequency information and the second time-frequency information, and the first time-frequency information and the second time-frequency information are convolved based on the first path weight and the second path weight, respectively, to obtain the first disturbance feature and the second disturbance feature, and the first disturbance feature and the second disturbance feature are integrated to obtain the target disturbance feature. The two paths process their respective information independently, and in the time domain path, the time domain features are gradually extracted through a series of neural network layers. For example, for a signal whose voltage amplitude changes with time, the time domain path extracts the rate of change of the voltage amplitude, the time interval of the peak, the duration of the voltage sag or swell, and other features, and in the frequency domain path, frequency domain features are extracted, such as the fundamental frequency, the frequency and amplitude of each harmonic, the frequency range of the transient pulse, etc. After being independently processed by each path, the first disturbance feature and the second disturbance feature extracted by the two paths are merged in the fusion layer. When integrating the first perturbation feature and the second perturbation feature, a simple splicing method can be used, such as directly splicing the feature vectors of the two paths together, or a weighted summation method can be used, such as assigning different weights to the two paths according to their importance for summing them up, or an integration method based on an attention mechanism can be used, such as dynamically allocating weights of the two paths according to the characteristics of the input data.

[0095] In a possible embodiment, the weight of the first path and the weight of the second path are dynamically adjusted according to the time-frequency characteristics of the time-frequency information. If the time-frequency information shows obvious frequency domain characteristics at a certain moment, such as an increase in harmonic content, the weight of the frequency domain path can be increased, so that the network focuses more on learning and extracting features from the frequency domain perspective to better identify such frequency-domain-based disturbances. On the contrary, if the time domain characteristics of the signal are more prominent, such as a sudden change in voltage amplitude, the weight of the time domain path is increased. For example, when a disturbance such as a sudden rise or drop in voltage is detected, which is mainly reflected in the time domain, enhancing the weight of the time domain path helps to identify such anomalies more quickly and accurately.

[0096] In a possible embodiment, the number of convolution layers of the first preset path and the second preset path is determined according to the first path weight and the second path weight, and the path with a high weight has more convolution layers, and the path with a low weight has fewer convolution layers. By grouping convolution, it is possible to effectively utilize the feature differences between different groups while reducing the amount of calculation, and the residual connection enables the network to reuse the features of the previous layer.

[0097] In a possible embodiment, for a shallow dual-path neural network, first, the input feature map is split into two parts, each of which is convolved, and then spliced ​​together. This splitting strategy helps to reduce the amount of calculation and the amount of parameters. Then, the spliced ​​feature map is separated in the channel dimension through the feature separation module, and the separated feature map is grouped and convolved to further reduce the amount of parameters. Subsequently, the separated feature map is mapped to a smaller dimension to reduce the computational complexity. In the dual-path neural network, one path is set to a positive weight, and the other path can be set to a negative weight, zero weight, or a weight close to zero to highlight the difference between the two paths. The negative weight path can suppress some unnecessary features or compensate for the positive weight path, and the feature map of the positive weight is deeply convolved to extract richer feature information. Finally, the feature maps output by the two paths are spliced ​​together as the final output of this stage. This splicing method can comprehensively utilize the feature information of the two paths and improve the expression ability of the model. Through operations such as feature map splitting, channel separation, group convolution, dimension compression, and deep convolution, the dual-path network model achieves a good balance between computational efficiency and model performance. After feature extraction, the dual-path network usually fuses the features of the two paths and then makes the final decision through the classification layer. Specifically, the two paths can be based on different types of convolutional neural networks, and the representations of the first perturbation feature and the second perturbation feature of the two paths are F and F, respectively. ResNet and F DenseNet , then the target disturbance feature F mix =[F ResNet , F DenseNet ], where F mix It represents the target disturbance characteristics after the integration of the two paths, which can be expressed in the form of a vector. ResNet and F DenseNet The concatenation operation is performed and then the fused features are classified through the fully connected layer.

[0098] For example, see Figure 3 , Figure 3 is a schematic diagram of the structure of a shallow dual-path neural network provided in an embodiment of the present application, such as Figure 3As shown in the figure, d represents the dimension, and FS is the feature separation module. The initial dimension parameter is 256, which is consistent with the size of the input feature map. The first dimension parameter of the first preset path is 64, and the second dimension parameter of the second preset path is 192. First, feature selection (FS) is performed, and the input and output feature map sizes remain unchanged, still 256. Then, the input feature map size is changed to 64 and 192 respectively through a 1×1 convolution layer, and then the 192-size feature map is further reduced to 48 through a 1×1 convolution layer, and then the 48-size feature map is kept unchanged through a 3×3 convolution layer. After that, the 48-size feature map is restored to 192 through a 1×1 convolution layer, and then the 192-size feature map is processed through multiple repeated blocks to keep the size unchanged. Finally, the 64-size and 192-size feature maps are merged to obtain the final output feature map size 256. Among them, the feature separation module can divide the frequency axis in the time-frequency matrix according to certain rules, such as into low frequency bands and high frequency bands, or according to the typical frequency ranges corresponding to different power quality disturbance characteristics. Through these steps, the network can effectively process and transform the input feature map, realize feature extraction and fusion, and provide high-quality feature representation for subsequent tasks.

[0099] In the embodiment of the present application, by processing the time domain and frequency domain information respectively through two paths, the features in the data can be extracted more comprehensively. In the analysis of power quality disturbance signals, both the time domain features and the frequency domain features contain important information. The dual-path neural network can capture these changes in the time domain and frequency domain at the same time, thereby improving the feature extraction capability of the power quality disturbance signal. The dual-path structure enables the network to better adapt to different types of data and tasks. For different power quality disturbance signals, their performance in the time domain and frequency domain may be different. The dual-path neural network can flexibly adjust the processing methods and weights of the two paths according to the specific characteristics of the signal, so as to better adapt to various disturbance situations and improve the versatility and flexibility of the model. In addition, by determining the weights of the dual paths, when facing different power system environments and various complex disturbance situations, the parameters of the neural network can be dynamically adjusted according to the new data features, which makes it not overly dependent on fixed parameter settings and can better adapt to new sample data. For example, when a new power electronic device is connected to the power system, a new disturbance feature is generated. The path weight can be adjusted according to these new features, and these changes can be quickly learned and adapted to, thereby improving the generalization ability in different scenarios and reducing overfitting of specific data sets.

[0100] Optionally, step S104, determining the disturbance category of the power quality disturbance signal based on the target disturbance feature to obtain multiple disturbance categories, each disturbance category corresponding to a disturbance probability, may include the following steps:

[0101] Step S501: Acquire historical power quality disturbance signals;

[0102] Step S502: determining the disturbance category of the power quality disturbance signal based on the historical power quality disturbance signal and the target disturbance feature, and obtaining multiple disturbance categories;

[0103] Step S503: determining a reference disturbance probability of each disturbance category in a plurality of disturbance categories according to historical power quality disturbance signals, and obtaining a plurality of reference disturbance probabilities;

[0104] Step S504: determining a calculation priority of each disturbance category in a plurality of disturbance categories based on a plurality of reference disturbance probabilities, to obtain a plurality of calculation priorities;

[0105] Step S505: determining a weight matrix and a bias vector of each disturbance category in a plurality of disturbance categories based on historical power quality disturbance signals, and obtaining a plurality of weight matrices and a plurality of bias vectors;

[0106] Step S506: determining the power quality disturbance signal to perform calculations based on multiple calculation priorities, a preset probability calculation function, target disturbance characteristics, multiple weight matrices and multiple bias vectors to obtain a disturbance probability corresponding to each disturbance category.

[0107] In a possible embodiment, a historical power quality disturbance signal is obtained, and the disturbance category of the power quality disturbance signal is determined based on the historical power quality disturbance signal and the target disturbance feature to obtain multiple disturbance categories. A reference disturbance probability of each disturbance category in the multiple disturbance categories is determined based on the historical power quality disturbance signal to obtain multiple reference disturbance probabilities. A calculation priority of each disturbance category in the multiple disturbance categories is determined based on the multiple reference disturbance probabilities to obtain multiple calculation priorities. A weight matrix and a bias vector of each disturbance category in the multiple disturbance categories are determined based on the historical power quality disturbance signal to obtain multiple weight matrices and multiple bias vectors. Based on multiple calculation priorities, a preset probability calculation function, a target disturbance feature, multiple weight matrices and multiple bias vectors, the power quality disturbance signal is determined to be calculated in order from high to low priority to obtain the disturbance probability corresponding to each disturbance category. When determining the disturbance probability, the determination method can be represented by formula (6):

[0108] Y = softmax(W·F mix +b)..............(6)

[0109] Among them, W and b are the weight matrix and bias vector of the fully connected layer respectively, Y is the disturbance probability vector, which indicates the probability that the output is a certain type of power quality disturbance, and the softmax function is an activation function, which is mainly used in the output layer in multi-classification problems. Its function is to convert a real number vector into a probability distribution so that the value of each element is between [0, 1] and the sum of all elements is 1.

[0110] Optionally, step S101, obtaining a power quality disturbance signal within a preset time period, may include the following steps:

[0111] Step S601: determining m initial sampling time periods according to historical power quality disturbance signals; m is a positive integer;

[0112] Step S602: determining an initial sampling frequency of each of m initial sampling time periods according to a historical power quality disturbance signal, to obtain m initial sampling frequencies;

[0113] Step S603: sampling the initial electric energy signal based on m initial sampling frequencies to obtain a first sampling result;

[0114] Step S604: predicting a first disturbance trend of the initial power signal based on the historical power quality disturbance signal and the first sampling result;

[0115] Step S605: if the first disturbance trend is consistent with the preset disturbance trend, determine the target sampling frequency according to the first disturbance trend;

[0116] Step S606: sampling the initial electric energy signal based on the target sampling frequency to obtain a power quality disturbance signal of a preset time period.

[0117] In a possible embodiment, before obtaining the power quality disturbance signal, the monitoring target and scope are determined. For different monitoring targets, the focus of power quality issues is different. In the substation, the main focus is on the bus voltage quality, the power quality parameters of the incoming and outgoing lines, etc., and on the industrial user side, the focus is on the disturbance signals such as voltage sag and harmonics that affect the production equipment. And determine the grid topology and electrical parameters. The grid topology can be determined by obtaining the connection mode of the grid, such as radial, ring or grid type, etc. Different topologies have different propagation characteristics when power quality disturbance occurs. Electrical parameters can be basic parameters such as rated voltage, rated current, and frequency of the grid. These parameters are the basis for judging whether the power quality is normal. Load characteristics and load distribution can also be determined. The types of loads carried by the grid include linear loads, such as incandescent lamps, resistance furnaces, etc., and nonlinear loads, such as rectifiers, inverters, arc furnaces, etc. Nonlinear loads are one of the main causes of power quality disturbances such as harmonics and voltage fluctuations. Determine the working principle and electrical characteristics of different types of loads. For example, during the smelting process of an arc furnace, the drastic change in the arc current will cause fluctuations and flickering in the grid voltage, while the switching power supply in electronic equipment such as computers will generate high-order harmonics. Determine the load distribution in different areas and time periods to predict the probability and severity of power quality disturbances. For example, in commercial areas, the load during the day is mainly the electrical equipment of commercial facilities, and the load may decrease at night. In industrial concentrated areas, the load distribution is relatively stable, but there may be large changes when the production shift of the enterprise is adjusted. The start and stop of large loads may cause disturbances such as voltage sag or surge.

[0118] In a possible embodiment, m initial sampling time periods are determined according to historical power quality disturbance signals, where m is a positive integer. The initial sampling frequency of each of the m initial sampling time periods is determined according to the historical power quality disturbance signals to obtain m initial sampling frequencies. For example, multiple initial sampling time periods are determined according to the time when disturbances occur in historical power quality disturbance signals, and multiple initial sampling frequencies are determined according to the frequency of disturbances in historical power quality disturbance signals. Then, the initial power signal is sampled based on the m initial sampling frequencies to obtain a first sampling result. The first disturbance trend of the initial power signal is predicted based on the historical power quality disturbance signal and the first sampling result. The preset disturbance trend may be one or more of the trends such as voltage swell, voltage sag, harmonics, etc. If the first disturbance trend is consistent with the preset disturbance trend, the target sampling frequency is determined according to the first disturbance trend, and the initial power signal is sampled based on the target sampling frequency to obtain the power quality disturbance signal of the preset time period.

[0119] Optionally, step S105, determining the loss value of each disturbance category based on the multiple disturbance categories and the disturbance probability corresponding to each disturbance category to obtain multiple loss values, may include the following steps:

[0120] Step S701: determining the number of categories of multiple disturbance categories;

[0121] Step S702: determining a disturbance code of each disturbance category in a plurality of disturbance categories based on the number of categories, to obtain a plurality of disturbance codes;

[0122] Step S703: determining a loss value of each disturbance category in a plurality of disturbance categories according to a plurality of disturbance codes and a disturbance probability corresponding to each disturbance category, to obtain a plurality of loss values.

[0123] In one possible embodiment, assume that the true label is y, where y is a one-hot encoded vector. One-hot encoding is a technique for converting categorical variables into binary vector representations. For a categorical variable with C perturbation categories, the category label y of each sample can be represented as an integer y∈{1, 2, ..., C}. One-hot encoding converts this integer label into a binary vector of length C, where the length of the vector is C, and only one element is 1, and the rest are 0. Specifically, the i-th element is 1 if and only if the category label of the sample is i. This encoding method ensures that each category is uniquely mapped to a binary vector, thereby avoiding the order relationship between categories and enabling the model to correctly process categorical data. y represents the true category of the sample. The cross entropy loss function is defined as shown in formula (7):

[0124]

[0125] Among them, y i is the i-th element of the true label, p i is the i-th element of the predicted probability. The cross entropy loss function measures the difference between the probability distribution predicted by the model and the probability distribution of the true label. The cross entropy loss function evaluates the prediction effect of the model by calculating the difference between the probability distribution predicted by the model and the one-hot encoding of the true label. The greater the difference, the greater the loss value, and vice versa. During the training process, the cross entropy loss function is used as an optimization objective to adjust the model parameters by minimizing the loss function so that the model's prediction results are as close to the true label as possible. Since it can handle probability distributions between multiple categories, the cross entropy loss function is particularly suitable for multi-classification problems.

[0126] In the identification of power quality disturbances, taking the classification problem of voltage sag and swell as an example, assume that there are C possible categories, for example, C = 2, representing sag and swell respectively. The true label y is a one-hot encoded vector, and the model's predicted output is Is a probability distribution vector. The cross entropy loss function can help evaluate the prediction effect of the model. The greater the difference, the greater the loss value, and vice versa.

[0127] Specifically, if the true label is a temporary drop, that is, y = [1, 0], and the probability distribution predicted by the model is Then the calculation of the cross entropy loss function is shown in formula (8):

[0128]

[0129] If the true label is a temporary rise, that is, y = [0, 1], and the probability distribution predicted by the model is Then the calculation of the cross entropy loss function is shown in formula (9):

[0130]

[0131] By calculating the cross entropy loss function, we can determine the performance of the model in the voltage sag and swell classification task. The smaller the loss value, the closer the model's prediction result is to the true label, and the better the model's classification effect is. Because -log(0.8) is smaller than -log(0.7), the judgment result in the above example is a sag.

[0132] In summary, in the implementation of the present application, the power quality disturbance signal within the preset time period is first obtained, and then the time-frequency information of the power quality disturbance signal is determined by the adaptive incomplete S transform algorithm, and the time-frequency information is processed in parallel by the dual-path structure to obtain the target disturbance feature of the power quality disturbance signal. Next, the disturbance category of the power quality disturbance signal is determined based on the target disturbance feature to obtain multiple disturbance categories, each disturbance category corresponds to a disturbance probability, and then, the loss value of each disturbance category is determined based on the multiple disturbance categories and the disturbance probability corresponding to each disturbance category to obtain multiple loss values, and the disturbance category corresponding to the minimum value of the multiple loss values ​​is used as the target disturbance category of the power quality disturbance signal. Thus, the time-frequency information extraction is realized by the adaptive incomplete S transform algorithm, and the target disturbance feature of the power quality disturbance signal is obtained by processing the time-frequency information in parallel by the dual-path structure, and the target disturbance category is determined based on the target disturbance feature, which can improve the computational efficiency when detecting power quality. Based on the method of adaptive incomplete S transform and shallow dual-path neural network, the transformation parameters can be dynamically adjusted according to the characteristics of the signal through the adaptive incomplete S transform to reduce unnecessary calculations. The dual-path design can process different types of features in parallel and improve computational efficiency. The adaptive incomplete S-transform can automatically extract the key features of the signal and reduce the reliance on manual feature selection. The adaptive incomplete S-transform can adapt to different types of signals and disturbances and improve the ability to recognize unseen disturbance types. The shallow dual-path design can process different types of features at the same time, reducing the complexity of the model, reducing the risk of overfitting, and enhancing the generalization ability of the model.

[0133] The method of the embodiment of the present invention is described in detail above, and the device of the embodiment of the present invention is provided below.

[0134] See also Figure 4 , Figure 4 It is a structural diagram of a power quality disturbance identification device provided in an embodiment of the present application. The power quality disturbance identification device 800 is applied to a power quality disturbance identification system, and the system includes a time-frequency information extraction module and a shallow dual-path neural network module. The time-frequency information extraction module uses an adaptive incomplete S-transformation algorithm to realize time-frequency information extraction, and the shallow dual-path neural network module uses a dual-path structure; the power quality disturbance identification device 800 includes an acquisition unit 801 and a processing unit 802;

[0135] An acquisition unit 801 is used to acquire a power quality disturbance signal within a preset time period;

[0136] The processing unit 802 is used to determine the time-frequency information of the power quality disturbance signal by using an adaptive incomplete S-transform algorithm;

[0137] The target disturbance characteristics of the power quality disturbance signal are obtained by processing the time-frequency information in parallel through a dual-path structure;

[0138] Determine the disturbance category of the power quality disturbance signal based on the target disturbance feature, and obtain multiple disturbance categories, each disturbance category corresponding to a disturbance probability;

[0139] Determine a loss value of each disturbance category based on multiple disturbance categories and a disturbance probability corresponding to each disturbance category to obtain multiple loss values;

[0140] The disturbance category corresponding to the minimum value among the multiple loss values ​​is used as the target disturbance category of the power quality disturbance signal.

[0141] In some possible embodiments, the adaptive incomplete S transform algorithm includes a window function. In determining the time-frequency information of the power quality disturbance signal by the adaptive incomplete S transform algorithm, the processing unit 802 is specifically configured to:

[0142] Obtain the frequency range of power quality disturbance signals;

[0143] An initial scale parameter range of the window function is determined based on the frequency range; the window function is used to determine the calculation area of ​​the adaptive incomplete S transform algorithm on the power quality disturbance signal;

[0144] Determine a target indicator set and a target weight group; the target indicator set includes a time-frequency resolution trade-off loss indicator, a key frequency component detection accuracy indicator, and a computational cost indicator; the target weight group includes a first weight, a second weight, and a third weight; each indicator in the target indicator set corresponds to a corresponding weight in the target weight group;

[0145] The calculation is performed based on the preset first calculation formula, the target indicator set and the target weight group to obtain the calculation result; the first calculation formula is: F=ω1L+ω2P+ω3C, wherein L, P and C are the time-frequency resolution trade-off loss indicator, the key frequency component detection accuracy indicator and the calculation cost indicator respectively, ω1, ω2 and ω3 are the first weight, the second weight and the third weight respectively, and F is the calculation result;

[0146] Determine the target scale parameter within the initial scale parameter range according to the calculation result;

[0147] The time-frequency information is determined based on an adaptive incomplete S-transform algorithm, target scale parameters and power quality disturbance signals.

[0148] In some possible embodiments, in determining the target scale parameter within the initial scale parameter range according to the calculation result, the processing unit 802 is specifically configured to:

[0149] Determine characteristic parameters of the power quality disturbance signal, the characteristic parameters including instantaneous frequency or instantaneous amplitude;

[0150] Determine the rate of change of the characteristic parameter within a preset time period; the rate of change is used to reflect the severity of the change of the power quality disturbance signal within the preset time period;

[0151] Determining a scale adjustment parameter range within an initial scale parameter range according to the rate of change;

[0152] A target scale parameter is determined within the scale adjustment parameter range based on the calculation result.

[0153] In some possible embodiments, the dual-path structure includes a first preset path and a second preset path. In terms of obtaining a target disturbance feature of a power quality disturbance signal by processing time-frequency information in parallel through the dual-path structure, the processing unit 802 is specifically configured to:

[0154] Determine the initial dimension parameters of time-frequency information;

[0155] Determine, based on the initial dimension parameters, a first dimension parameter of the time-frequency information in the first preset path and a second dimension parameter of the second preset path;

[0156] Splitting the time-frequency information based on the first dimension parameter and the second dimension parameter to obtain the first time-frequency information and the second time-frequency information;

[0157] Determine a first path weight of a first preset path and a second path weight of a second preset path based on the first time-frequency information and the second time-frequency information;

[0158] Convolving the first time-frequency information and the second time-frequency information based on the first path weight and the second path weight respectively to obtain a first disturbance feature and a second disturbance feature;

[0159] The first disturbance feature and the second disturbance feature are integrated to obtain the target disturbance feature.

[0160] In some possible embodiments, in determining the disturbance category of the power quality disturbance signal based on the target disturbance feature, obtaining multiple disturbance categories, each disturbance category corresponding to a disturbance probability, the processing unit 802 is specifically used to:

[0161] Obtain historical power quality disturbance signals;

[0162] Determine the disturbance category of the power quality disturbance signal based on the historical power quality disturbance signal and the target disturbance feature, and obtain multiple disturbance categories;

[0163] Determining a reference disturbance probability of each disturbance category in a plurality of disturbance categories according to historical power quality disturbance signals, and obtaining a plurality of reference disturbance probabilities;

[0164] Determining a calculation priority of each disturbance category in a plurality of disturbance categories based on a plurality of reference disturbance probabilities to obtain a plurality of calculation priorities;

[0165] Determine a weight matrix and a bias vector for each disturbance category in a plurality of disturbance categories based on historical power quality disturbance signals, and obtain a plurality of weight matrices and a plurality of bias vectors;

[0166] Based on multiple calculation priorities, a preset probability calculation function, a target disturbance feature, multiple weight matrices and multiple bias vectors, a power quality disturbance signal is determined to perform calculations to obtain a disturbance probability corresponding to each disturbance category.

[0167] In some possible embodiments, in terms of acquiring the power quality disturbance signal within a preset time period, the acquiring unit 801 is specifically configured to:

[0168] Determine m initial sampling time periods according to historical power quality disturbance signals; m is a positive integer;

[0169] Determine the initial sampling frequency of each of the m initial sampling time periods according to the historical power quality disturbance signal, and obtain m initial sampling frequencies;

[0170] Sampling the initial electric energy signal based on m initial sampling frequencies to obtain a first sampling result;

[0171] Predicting a first disturbance trend of an initial power signal based on a historical power quality disturbance signal and a first sampling result;

[0172] If the first disturbance trend is consistent with the preset disturbance trend, determining the target sampling frequency according to the first disturbance trend;

[0173] The initial electric energy signal is sampled based on the target sampling frequency to obtain a power quality disturbance signal of a preset time period.

[0174] In some possible embodiments, in determining the loss value of each disturbance category based on multiple disturbance categories and the disturbance probability corresponding to each disturbance category to obtain multiple loss values, the processing unit 802 is specifically configured to:

[0175] Determining the number of categories for multiple perturbation categories;

[0176] Determining a disturbance code for each disturbance category in a plurality of disturbance categories based on the number of categories to obtain a plurality of disturbance codes;

[0177] A loss value of each disturbance category in the multiple disturbance categories is determined according to the multiple disturbance codes and the disturbance probability corresponding to each disturbance category to obtain multiple loss values.

[0178] See also Figure 5 , Figure 5Schematic diagram of the structure of an electronic device provided in an embodiment of the present application. Figure 5 As shown, the electronic device 900 includes a transceiver 901, a processor 902 and a memory 903, which are connected via a bus 904. The memory 903 is used to store computer programs and data, and can transmit the data stored in the memory 903 to the processor 902. The electronic device can be the above-mentioned power quality disturbance identification device, and the processor 902 can be the above-mentioned acquisition unit 801 and processing unit 802.

[0179] The processor 902 is used to read the computer program in the memory 903 and perform the following operations:

[0180] Obtaining a power quality disturbance signal within a preset time period;

[0181] Determine the time-frequency information of power quality disturbance signal through adaptive incomplete S-transform algorithm;

[0182] The target disturbance characteristics of the power quality disturbance signal are obtained by processing the time-frequency information in parallel through a dual-path structure;

[0183] Determine the disturbance category of the power quality disturbance signal based on the target disturbance feature, and obtain multiple disturbance categories, each disturbance category corresponding to a disturbance probability;

[0184] Determine a loss value of each disturbance category based on multiple disturbance categories and a disturbance probability corresponding to each disturbance category to obtain multiple loss values;

[0185] The disturbance category corresponding to the minimum value among the multiple loss values ​​is used as the target disturbance category of the power quality disturbance signal.

[0186] In some possible embodiments, the adaptive incomplete S-transform algorithm includes a window function. In determining the time-frequency information of the power quality disturbance signal by using the adaptive incomplete S-transform algorithm, the processor 902 is specifically configured to perform the following operations:

[0187] Obtain the frequency range of power quality disturbance signals;

[0188] An initial scale parameter range of the window function is determined based on the frequency range; the window function is used to determine the calculation area of ​​the adaptive incomplete S transform algorithm on the power quality disturbance signal;

[0189] Determine a target indicator set and a target weight group; the target indicator set includes a time-frequency resolution trade-off loss indicator, a key frequency component detection accuracy indicator, and a computational cost indicator; the target weight group includes a first weight, a second weight, and a third weight; each indicator in the target indicator set corresponds to a corresponding weight in the target weight group;

[0190] The calculation is performed based on the preset first calculation formula, the target indicator set and the target weight group to obtain the calculation result; the first calculation formula is: F=ω1L+ω2P+ω3C, wherein L, P and C are the time-frequency resolution trade-off loss indicator, the key frequency component detection accuracy indicator and the calculation cost indicator respectively, ω1, ω2 and ω3 are the first weight, the second weight and the third weight respectively, and F is the calculation result;

[0191] Determine the target scale parameter within the initial scale parameter range according to the calculation result;

[0192] The time-frequency information is determined based on an adaptive incomplete S-transform algorithm, target scale parameters and power quality disturbance signals.

[0193] In some possible embodiments, in determining the target scale parameter within the initial scale parameter range according to the calculation result, the processor 902 is specifically configured to perform the following operations:

[0194] Determine characteristic parameters of the power quality disturbance signal, the characteristic parameters including instantaneous frequency or instantaneous amplitude;

[0195] Determine the rate of change of the characteristic parameter within a preset time period; the rate of change is used to reflect the severity of the change of the power quality disturbance signal within the preset time period;

[0196] Determining a scale adjustment parameter range within an initial scale parameter range according to the rate of change;

[0197] A target scale parameter is determined within the scale adjustment parameter range based on the calculation result.

[0198] In some possible embodiments, the dual-path structure includes a first preset path and a second preset path. In terms of obtaining a target disturbance feature of a power quality disturbance signal by processing time-frequency information in parallel through the dual-path structure, the processor 902 is specifically configured to perform the following operations:

[0199] Determine the initial dimension parameters of time-frequency information;

[0200] Determine, based on the initial dimension parameters, a first dimension parameter of the time-frequency information in the first preset path and a second dimension parameter of the second preset path;

[0201] Splitting the time-frequency information based on the first dimension parameter and the second dimension parameter to obtain the first time-frequency information and the second time-frequency information;

[0202] Determine a first path weight of a first preset path and a second path weight of a second preset path based on the first time-frequency information and the second time-frequency information;

[0203] Convolving the first time-frequency information and the second time-frequency information based on the first path weight and the second path weight respectively to obtain a first disturbance feature and a second disturbance feature;

[0204] The first disturbance feature and the second disturbance feature are integrated to obtain the target disturbance feature.

[0205] In some possible embodiments, in determining the disturbance category of the power quality disturbance signal based on the target disturbance feature, obtaining multiple disturbance categories, each disturbance category corresponding to a disturbance probability, the processor 902 is specifically configured to perform the following operations:

[0206] Obtain historical power quality disturbance signals;

[0207] Determine the disturbance category of the power quality disturbance signal based on the historical power quality disturbance signal and the target disturbance feature, and obtain multiple disturbance categories;

[0208] Determining a reference disturbance probability of each disturbance category in a plurality of disturbance categories according to historical power quality disturbance signals, and obtaining a plurality of reference disturbance probabilities;

[0209] Determining a calculation priority of each disturbance category in a plurality of disturbance categories based on a plurality of reference disturbance probabilities to obtain a plurality of calculation priorities;

[0210] Determine a weight matrix and a bias vector for each disturbance category in a plurality of disturbance categories based on historical power quality disturbance signals, and obtain a plurality of weight matrices and a plurality of bias vectors;

[0211] Based on multiple calculation priorities, a preset probability calculation function, a target disturbance feature, multiple weight matrices and multiple bias vectors, a power quality disturbance signal is determined to perform calculations to obtain a disturbance probability corresponding to each disturbance category.

[0212] In some possible embodiments, in terms of obtaining the power quality disturbance signal within a preset time period, the processor 902 is specifically configured to perform the following operations:

[0213] Determine m initial sampling time periods according to historical power quality disturbance signals; m is a positive integer;

[0214] Determine the initial sampling frequency of each of the m initial sampling time periods according to the historical power quality disturbance signal, and obtain m initial sampling frequencies;

[0215] Sampling the initial electric energy signal based on m initial sampling frequencies to obtain a first sampling result;

[0216] Predicting a first disturbance trend of an initial power signal based on a historical power quality disturbance signal and a first sampling result;

[0217] If the first disturbance trend is consistent with the preset disturbance trend, determining the target sampling frequency according to the first disturbance trend;

[0218] The initial electric energy signal is sampled based on the target sampling frequency to obtain a power quality disturbance signal of a preset time period.

[0219] In some possible embodiments, in determining the loss value of each disturbance category based on multiple disturbance categories and the disturbance probability corresponding to each disturbance category to obtain multiple loss values, the processor 902 is specifically configured to perform the following operations:

[0220] Determining the number of categories for multiple perturbation categories;

[0221] Determining a disturbance code for each disturbance category in a plurality of disturbance categories based on the number of categories to obtain a plurality of disturbance codes;

[0222] A loss value of each disturbance category in the multiple disturbance categories is determined according to the multiple disturbance codes and the disturbance probability corresponding to each disturbance category to obtain multiple loss values.

[0223] An embodiment of the present application also provides a computer-readable storage medium, which stores a computer program. The computer program is executed by a processor to implement part or all of the steps of any power quality disturbance identification method recorded in the above method embodiments.

[0224] An embodiment of the present application also provides a computer program product, which includes a non-transitory computer-readable storage medium storing a computer program, and the computer program is operable to cause a computer to execute part or all of the steps of any one of the power quality disturbance identification methods recorded in the above method embodiments.

[0225] It should be noted that, for the aforementioned method embodiments, for the sake of simplicity, they are all expressed as a series of action combinations, but those skilled in the art should be aware that the present application is not limited by the described order of actions, because according to the present application, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in the specification are all optional embodiments, and the actions and modules involved are not necessarily required by the present application.

[0226] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0227] In the several embodiments provided in the present application, it should be understood that the disclosed device can be implemented in other ways. For example, the device embodiments described above are only schematic, such as the division of the modules, which is only a logical function division. There may be other division methods in actual implementation, such as multiple modules or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, and the indirect coupling or communication connection of the device or module can be electrical or other forms.

[0228] The modules described as separate components may or may not be physically separated, and the components shown as modules may or may not be physical modules, that is, they may be located in one place or distributed on multiple network modules. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0229] In addition, each functional module in each embodiment of the present application can be integrated into a processing module, or each module can exist physically separately, or two or more modules can be integrated into one module. The above integrated modules can be implemented in the form of hardware or software program modules.

[0230] If the integrated module is implemented in the form of a software program module and sold or used as an independent product, it can be stored in a computer-readable memory. Based on this understanding, the technical solution of the present application, in essence, or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product, which is stored in a memory and includes several instructions for a computer device (which can be a personal computer, a server or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present application. The aforementioned memory includes: U disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), mobile hard disk, disk or optical disk and other media that can store program codes.

[0231] The embodiments of the present application are introduced in detail above. Specific examples are used in this article to illustrate the principles and implementation methods of the present application. The description of the above embodiments is only used to help understand the method and core idea of ​​the present application. At the same time, for general technical personnel in this field, according to the idea of ​​the present application, there will be changes in the specific implementation method and application scope. In summary, the content of this specification should not be understood as a limitation on the present application.

Claims

1. A method for identifying power quality disturbances, characterized in that: Applied to a power quality disturbance identification system, the system comprises a time-frequency information extraction module and a shallow dual-path neural network module, the time-frequency information extraction module uses an adaptive incomplete S-transformation algorithm to extract time-frequency information, and the shallow dual-path neural network module uses a dual-path structure; the method comprises: Obtaining a power quality disturbance signal within a preset time period; Determining the time-frequency information of the power quality disturbance signal by using the adaptive incomplete S-transform algorithm; Processing the time-frequency information in parallel through the dual-path structure to obtain target disturbance characteristics of the power quality disturbance signal; Determining the disturbance category of the power quality disturbance signal based on the target disturbance feature to obtain a plurality of disturbance categories, each disturbance category corresponding to a disturbance probability; Determine a loss value of each disturbance category based on the multiple disturbance categories and the disturbance probability corresponding to each disturbance category to obtain multiple loss values; The disturbance category corresponding to the minimum value among the multiple loss values ​​is used as the target disturbance category of the power quality disturbance signal.

2. The method according to claim 1, characterized in that The adaptive incomplete S-transform algorithm includes a window function, and determining the time-frequency information of the power quality disturbance signal by the adaptive incomplete S-transform algorithm includes: Acquire the frequency range of the power quality disturbance signal; Determining an initial scale parameter range of the window function based on the frequency range; the window function is used to determine a calculation area of ​​the adaptive incomplete S transform algorithm on the power quality disturbance signal; Determine a target indicator set and a target weight group; the target indicator set includes a time-frequency resolution trade-off loss indicator, a key frequency component detection accuracy indicator, and a computational cost indicator; the target weight group includes a first weight, a second weight, and a third weight; each indicator in the target indicator set corresponds to a corresponding weight in the target weight group in a one-to-one correspondence; Calculation is performed based on a preset first calculation formula, the target indicator set and the target weight group to obtain a calculation result; the first calculation formula is: F=ω1L+ω2P+ω3C, wherein L, P and C are the time-frequency resolution trade-off loss indicator, the key frequency component detection accuracy indicator and the calculation cost indicator, respectively, ω1, ω2 and ω3 are the first weight, the second weight and the third weight, respectively, and F is the calculation result; Determining a target scale parameter within the initial scale parameter range according to the calculation result; The time-frequency information is determined based on the adaptive incomplete S-transform algorithm, the target scale parameter and the power quality disturbance signal.

3. The method according to claim 2, characterized in that Determining the target scale parameter within the initial scale parameter range according to the calculation result includes: Determining characteristic parameters of the power quality disturbance signal, wherein the characteristic parameters include instantaneous frequency or instantaneous amplitude; Determine the change rate of the characteristic parameter within the preset time period; the change rate is used to reflect the severity of the change of the power quality disturbance signal within the preset time period; Determining a scale adjustment parameter range within the initial scale parameter range according to the change rate; The target scale parameter is determined within the scale adjustment parameter range based on the calculation result.

4. The method according to any one of claims 1 to 3, characterized in that: The dual-path structure includes a first preset path and a second preset path, and the time-frequency information is processed in parallel by the dual-path structure to obtain the target disturbance feature of the power quality disturbance signal, including: Determining initial dimension parameters of the time-frequency information; Determine, based on the initial dimension parameter, a first dimension parameter of the time-frequency information in the first preset path and a second dimension parameter of the time-frequency information in the second preset path; Splitting the time-frequency information based on the first dimension parameter and the second dimension parameter to obtain first time-frequency information and second time-frequency information; Determine a first path weight of the first preset path and a second path weight of the second preset path based on the first time-frequency information and the second time-frequency information; Convolving the first time-frequency information and the second time-frequency information based on the first path weight and the second path weight respectively to obtain a first disturbance feature and a second disturbance feature; The first disturbance feature and the second disturbance feature are integrated to obtain the target disturbance feature.

5. The method according to any one of claims 1 to 3, characterized in that: The determining the disturbance category of the power quality disturbance signal based on the target disturbance feature to obtain a plurality of disturbance categories, each disturbance category corresponding to a disturbance probability, includes: Obtain historical power quality disturbance signals; Determining the disturbance category of the power quality disturbance signal based on the historical power quality disturbance signal and the target disturbance feature to obtain the multiple disturbance categories; Determining a reference disturbance probability of each disturbance category in the multiple disturbance categories according to the historical power quality disturbance signal to obtain multiple reference disturbance probabilities; Determining a calculation priority of each disturbance category in the plurality of disturbance categories based on the plurality of reference disturbance probabilities to obtain a plurality of calculation priorities; Determining a weight matrix and a bias vector for each disturbance category in the plurality of disturbance categories based on the historical power quality disturbance signal, to obtain a plurality of weight matrices and a plurality of bias vectors; Based on the multiple calculation priorities, the preset probability calculation function, the target disturbance characteristics, the multiple weight matrices and the multiple bias vectors, the power quality disturbance signal is determined to be operated to obtain the disturbance probability corresponding to each disturbance category.

6. The method according to claim 5, characterized in that The obtaining of the power quality disturbance signal within a preset time period includes: Determine m initial sampling time periods according to the historical power quality disturbance signal; m is a positive integer; Determine the initial sampling frequency of each of the m initial sampling time periods according to the historical power quality disturbance signal to obtain m initial sampling frequencies; Sampling the initial electric energy signal based on the m initial sampling frequencies to obtain a first sampling result; Predicting a first disturbance trend of the initial power signal based on the historical power quality disturbance signal and the first sampling result; If the first disturbance trend is consistent with a preset disturbance trend, determining a target sampling frequency according to the first disturbance trend; The initial electric energy signal is sampled based on the target sampling frequency to obtain the power quality disturbance signal of the preset time period.

7. The method according to any one of claims 1 to 3, characterized in that: The determining the loss value of each disturbance category based on the multiple disturbance categories and the disturbance probability corresponding to each disturbance category to obtain multiple loss values ​​includes: determining the number of categories of the plurality of disturbance categories; Determining a disturbance code for each disturbance category in the plurality of disturbance categories based on the number of categories to obtain a plurality of disturbance codes; The loss value of each disturbance category in the plurality of disturbance categories is determined according to the plurality of disturbance codes and the disturbance probability corresponding to each disturbance category to obtain the plurality of loss values.

8. A power quality disturbance identification device, characterized in that: Applied to the power quality disturbance identification system, the system includes a time-frequency information extraction module and a shallow dual-path neural network module, the time-frequency information extraction module uses an adaptive incomplete S-transformation algorithm to realize time-frequency information extraction, and the shallow dual-path neural network module uses a dual-path structure; the device includes an acquisition unit and a processing unit; The acquisition unit is used to acquire the power quality disturbance signal within a preset time period; The processing unit is used to determine the time-frequency information of the power quality disturbance signal by using the adaptive incomplete S-transform algorithm; Processing the time-frequency information in parallel through the dual-path structure to obtain target disturbance characteristics of the power quality disturbance signal; Determining the disturbance category of the power quality disturbance signal based on the target disturbance feature to obtain a plurality of disturbance categories, each disturbance category corresponding to a disturbance probability; Determine a loss value of each disturbance category based on the multiple disturbance categories and the disturbance probability corresponding to each disturbance category to obtain multiple loss values; The disturbance category corresponding to the minimum value among the multiple loss values ​​is used as the target disturbance category of the power quality disturbance signal.

9. An electronic device, characterized in that: include: A processor and a memory, wherein the processor is connected to the memory, the memory is used to store a computer program, and the processor is used to execute the computer program stored in the memory, so that the electronic device executes the method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, wherein the computer program includes program instructions, and when the program instructions are executed by a processor, the processor is caused to perform the method according to any one of claims 1 to 7.

Images