A power grid harmonic identification method and system based on deep learning

Abstract

The application relates to the field of electric digital processing, in particular to a power grid harmonic identification method and system based on deep learning, which comprises the following steps: inputting processed voltage data and current data into an LSTM model, outputting predicted data of voltage and current, obtaining voltage sequences and current sequences after processing, performing window division on the voltage sequences and the current sequences, obtaining local waveform differences in combination with amplitude spectra of corresponding windows, obtaining harmonic amplitude sequences by the amplitudes of the frequency corresponding to odd harmonics in the window sequence amplitude spectrum, obtaining harmonic amplitude change sequences after processing, obtaining harmonic similarity in combination with the local waveform differences, obtaining the harmonic contribution degree of a load feeder by using the harmonic amplitude sequences of different odd harmonics in the common coupling point voltage sequence to obtain the integrated harmonic similarity of the recognition significance degree, and sorting the harmonic contribution degrees and selecting a preset proportion of harmonic contribution degrees that meet preset conditions. The application has the effects of accurately locating a harmonic source and improving harmonic source identification accuracy.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of digital processing, in particular to a power grid harmonic identification method and system based on deep learning. BACKGROUND

[0002] Power grid harmonic identification refers to detecting, extracting and classifying harmonic components from voltage and current signals of the power grid through signal processing, data analysis or machine learning techniques, monitoring and analyzing the characteristics and sources of harmonic pollution in the power grid to maintain the stability and reliability of the power system. Harmonics are sinusoidal wave components with frequencies that are integer multiples of the fundamental frequency (usually 50Hz or 60Hz) in the power grid, introduced by nonlinear loads, power electronic devices (such as inverters, rectifiers) or new energy access (such as photovoltaic, wind power), causing signal distortion.

[0003] Chinese invention patent CN106383295B discloses a "power supply system harmonic source identification method and model", which identifies the harmonic source by calculating the harmonic power of the power supply system and each load at the corresponding time through real-time acquisition of the voltage and current full information of the power supply system power supply and all loads. However, it uses global waveform for harmonic identification, which may cause local harmonic information loss, making it impossible to identify the continuous change trend of different harmonic components, resulting in errors in harmonic identification and inaccurate harmonic source positioning. In addition, it identifies all odd and even harmonics, which may cause significant harmonic components to be smoothed by non-significant components, resulting in harmonic source positioning errors and low accuracy of existing harmonic source identification.

[0004] In addition, the current signal of the common coupling point is the superposition of all feeder currents, while the current of each load feeder is generated independently. The harmonic components of the superimposed current differ greatly from the simple addition of feeder current harmonics. The harmonic components on the common coupling point current signal are distorted, and the distortion cannot reflect the differences between the feeders. The harmonics of different feeder voltage signals will affect each other, resulting in small differences in harmonic components of different feeder voltage signals. The harmonic components on the load feeder voltage signal are distorted and cannot reflect the differences between the feeders, which may cause large errors in harmonic source identification. SUMMARY

[0005] In order to accurately locate the harmonic source and improve the accuracy of harmonic source identification, the present application provides a power grid harmonic identification method and system based on deep learning.

[0006] In the first aspect, the present application provides a power grid harmonic identification method based on deep learning, which adopts the following technical solution:

[0007] Calculate the harmonic contribution degree of each load feeder of the power grid to the common coupling point, and select the load corresponding to the load feeder whose harmonic contribution degree meets the preset condition as the identified harmonic source.

[0008] The calculation method of the harmonic contribution degree is:

[0009] The collected public coupling point voltage data is standardized to remove the dimension, input into the trained LSTM model, and output voltage prediction data. The processed public coupling point voltage sequence is obtained, and the load feeder current sequence is obtained in the same way.

[0010] The collected public coupling point voltage sequence is divided according to the preset window to obtain a plurality of voltage window sequences of the public coupling point. In the same way, a plurality of current window sequences of the load feeder are obtained.

[0011] The voltage window sequence of the public coupling point is transformed to obtain an amplitude spectrum, and the voltage amplitude sequence of the public coupling point is obtained by using the amplitude spectrum. In the same way, the current amplitude sequence of the load feeder is obtained. According to the voltage amplitude sequence of the public coupling point and the current amplitude sequence of the load feeder, the local waveform difference of the load feeder and the corresponding window sequence of the public coupling point is calculated.

[0012] For a frequency corresponding to an odd harmonic, the amplitude in the amplitude spectrum of the voltage window sequence of the public coupling point at the frequency is extracted to construct the amplitude sequence of the odd harmonic of the public coupling point voltage sequence. In the same way, the amplitude sequence of the odd harmonic of the load feeder current sequence is obtained.

[0013] Each data in the amplitude sequence of the odd harmonic of the public coupling point voltage sequence is processed to obtain the harmonic amplitude change sequence of the public coupling point voltage sequence. In the same way, the harmonic amplitude change sequence of the load feeder current sequence is obtained.

[0014] According to the amplitude sequence of the odd harmonic of the public coupling point voltage sequence, the amplitude sequence of the odd harmonic of the load feeder current sequence, the harmonic amplitude change sequence of the public coupling point voltage sequence, the harmonic amplitude change sequence of the load feeder current sequence, and the local waveform difference, the harmonic similarity of the load feeder and the public coupling point at the odd harmonic is calculated. According to the amplitude sequence of the odd harmonic of the public coupling point voltage sequence, the identification significance of the odd harmonic is calculated. According to the harmonic similarity and the identification significance, the harmonic contribution degree of the load feeder to the public coupling point is calculated.

[0015] Optionally, for the to-be-identified power grid, the voltage data of the public coupling point and the current data of the load feeder are obtained, and after processing the obtained data, the mean square error is selected as the loss function.

[0016] Optionally, the reciprocal of the fundamental frequency of the power grid current or voltage is used as the window length.

[0017] Optionally, the method for calculating the local waveform difference is: taking the DTW distance between the current window sequence of the load feeder and the voltage window sequence of the corresponding common coupling point as the first distance, taking the DTW distance between the amplitude sequence of the current window sequence of the load feeder and the voltage window sequence of the corresponding common coupling point as the second distance, and multiplying the first distance and the second distance to obtain the local waveform difference.

[0018] Optionally, the method for processing each data in the amplitude sequence of the odd harmonic of the common coupling point voltage sequence is: taking the first-order difference value of each data in the harmonic amplitude sequence of the voltage window sequence of the common coupling point and the data for ratio processing to obtain the harmonic amplitude variation sequence of the common coupling point voltage sequence.

[0019] Optionally, the method for calculating the harmonic similarity is: taking the absolute difference value of the same time data points of the amplitude sequence of the odd harmonic of the load feeder current sequence and the amplitude sequence of the same odd harmonic of the common coupling point voltage sequence as the first difference value, taking the absolute difference value of the corresponding data points of the harmonic amplitude variation sequence of the common coupling point voltage sequence and the harmonic amplitude variation sequence of the load feeder current sequence as the second difference value, multiplying the first difference value and the second difference value of the same odd harmonic and the local waveform difference as a first product, and performing negative correlation normalization on the sum of the products under all different odd harmonics to obtain the harmonic similarity.

[0020] Optionally, the method for calculating the recognition significance is: taking the ratio of the standard deviation of all data at the same data position in the amplitude sequence of the odd harmonic of the common coupling point voltage sequence and the average value of the standard deviations of the amplitude sequences of all odd harmonics of the common coupling point voltage sequence as the first ratio, taking the sum of the ratios of the amplitude sequence of the odd harmonic of the common coupling point voltage sequence at the same data position and the average value of the amplitude sequences of all odd harmonics of the common coupling point voltage sequence as the second ratio, and multiplying the first ratio and the second ratio to obtain the recognition significance.

[0021] Optionally, the method for calculating the harmonic contribution degree is: multiplying the normalized recognition significance of the same odd harmonic and the harmonic similarity to obtain a product value, and accumulating the product values of all different odd harmonics to obtain the harmonic contribution degree.

[0022] Optionally, the condition that meets the preset condition is: sorting the harmonic contribution degrees from large to small, selecting a preset proportion of the harmonic contribution degrees that meet the preset condition in the order from large to small; or sorting the harmonic contribution degrees from small to large, selecting a preset proportion of the harmonic contribution degrees that meet the preset condition in the order from small to large.

[0023] In a second aspect, the application provides a power grid harmonic identification system based on deep learning, which adopts the following technical scheme:

[0024] A power grid harmonic identification system based on deep learning includes a processor and a memory.

[0025] The beneficial effect is that the above-mentioned power grid harmonic identification system based on deep learning generates a computer program and is stored in the memory to be loaded and executed by the processor, so that the system is made according to the memory and the processor, and is convenient to use.

[0026] The present application has the following technical effects:

[0027] 1. It can more accurately analyze the waveform differences in the local window and the continuous change trend of different odd harmonics in the continuous sliding window, accurately evaluate the harmonic similarity between the load feeder and the point of common coupling for harmonic source positioning and identification. In addition, it can analyze the odd harmonic components and intensity of the point of common coupling for adaptive identification of harmonic identification to obtain the significance degree, avoid the smoothing of non-significant components to the significant harmonic components to cause harmonic source identification and positioning errors, and improve the identification accuracy of the harmonic source.

[0028] 2. The voltage data of the point of common coupling is collected using a voltage transformer, and the current data of the load feeder is collected using a current transformer, thereby reducing errors. BRIEF DESCRIPTION OF DRAWINGS

[0029] Figure 1 It is a flowchart of a power grid harmonic identification method based on deep learning. DETAILED DESCRIPTION

[0030] The embodiment of the present application discloses a power grid harmonic identification method based on deep learning, which comprises: calculating the harmonic contribution degree of each load feeder of the power grid to the point of common coupling, and selecting the load corresponding to the load feeder whose harmonic contribution degree meets the preset condition as the identified harmonic source; referring to Figure 1 , the calculation method of the harmonic contribution degree comprises steps S1-S9, which are as follows:

[0031] S1: The collected voltage data of the point of common coupling is standardized to remove the dimension, input into the trained LSTM (Long Short-Term Memory, long short-term memory) network model, output voltage prediction data, and obtain the voltage sequence of the point of common coupling after processing; the current sequence of the load feeder is obtained in the same way.

[0032] In one embodiment, for the power grid to be identified, the voltage data of the point of common coupling is collected using a voltage transformer, and the current data of the load feeder is collected using a current transformer. An exemplary collection frequency is 10 kHz.

[0033] The loss function of the LSTM model is a mean square error loss function. The construction and training of the LSTM model are prior art and will not be described herein. The collected and predicted voltage data are spliced to obtain a voltage sequence of the PCC, and the collected and predicted current data are spliced to obtain a current sequence of each load feeder.

[0034] S2: The collected PCC voltage sequence is divided according to a preset window to obtain a plurality of voltage window sequences of the PCC. Similarly, a plurality of current window sequences of the load feeder are obtained.

[0035] The inverse of the fundamental frequency of the grid current or voltage is taken as the window length, and the collected voltage sequence of the PCC and the current sequence of each load feeder are window-divided respectively, with a sliding step of 1, to obtain a plurality of voltage window sequences of the PCC and a plurality of current window sequences of the load feeder.

[0036] S3: The voltage window sequence of the PCC is transformed to obtain an amplitude spectrum, and the amplitude spectrum is used to obtain a voltage amplitude sequence of the PCC. Similarly, a current amplitude sequence of the load feeder is obtained.

[0037] Specifically, the voltage window sequence of the PCC is Fourier-transformed to obtain an amplitude spectrum; and the current window sequence of the load feeder is Fourier-transformed to obtain an amplitude spectrum.

[0038] S4: According to the voltage amplitude sequence of the PCC and the current amplitude sequence of the load feeder, a local waveform difference of the load feeder and the corresponding window sequence of the PCC is calculated.

[0039] In one embodiment, the calculation method of the local waveform difference is: taking a DTW (Dynamic Time Warping) distance between the current window sequence of the load feeder and the corresponding voltage window sequence of the PCC as a first distance, taking a DTW distance between the amplitude sequence of the current window sequence of the load feeder and the corresponding voltage window sequence of the PCC as a second distance, and multiplying the first distance and the second distance to obtain the local waveform difference; and the calculation formula is:

[0040]

[0041] wherein, represents the local waveform difference of the load feeder and the corresponding window sequence of the PCC, represents a DTW distance between the load feeder and the corresponding window sequence of the PCC, represents a DTW distance between the amplitude sequence of the load feeder and the corresponding window sequence of the PCC.

[0042] In one embodiment, due to the coordination of current and voltage harmonic changes in the power grid system, the voltage harmonic usually accompanies a similar change pattern, i.e., a similar waveform, when the current harmonic changes, so analyzing the waveforms between the corresponding window sequences of the point of common coupling and the load feeder is conducive to determining the correlation between the harmonic change of the point of common coupling and each load feeder.

[0043] The DTW distance tends to measure the waveform difference between two sequences, and the smaller the DTW distance is, the more similar the waveforms are and the more similar the change trends are. The window sequence and the amplitude sequence consider the local time-domain waveform difference and the local frequency-domain waveform difference of the point of common coupling and the load feeder, respectively.

[0044] S5: For a frequency corresponding to an odd harmonic, the amplitude of the frequency in the amplitude spectrum of the voltage window sequence of the point of common coupling is extracted to construct the amplitude sequence of the odd harmonic of the voltage sequence of the point of common coupling. Similarly, the amplitude sequence of the odd harmonic of the current sequence of the load feeder is obtained.

[0045] For the voltage window sequence or the current window sequence, one odd harmonic obtains one harmonic amplitude sequence.

[0046] S6: Each data in the amplitude sequence of the odd harmonic of the voltage sequence of the point of common coupling is processed to obtain the harmonic amplitude change sequence of the voltage sequence of the point of common coupling. Similarly, the harmonic amplitude change sequence of the current sequence of the load feeder is obtained.

[0047] The method for processing each data in the amplitude sequence of the odd harmonic of the voltage sequence of the point of common coupling is as follows:

[0048] S7: According to the amplitude sequence of the odd harmonic of the voltage sequence of the point of common coupling, the amplitude sequence of the odd harmonic of the current sequence of the load feeder, the harmonic amplitude change sequence of the voltage sequence of the point of common coupling, the harmonic amplitude change sequence of the current sequence of the load feeder, and the local waveform difference, the harmonic similarity of the load feeder and the point of common coupling at the odd harmonic is calculated.

[0049] The calculation method of the harmonic similarity is as follows: the absolute difference value of the same position data points of the amplitude sequence of the odd harmonic of the current sequence of the load feeder and the amplitude sequence of the same odd harmonic of the voltage sequence of the point of common coupling is taken as a first difference value, the absolute difference value of the corresponding data points of the harmonic amplitude change sequence of the voltage sequence of the point of common coupling and the harmonic amplitude change sequence of the current sequence of the load feeder at the same odd harmonic is taken as a second difference value, the first difference value and the second difference value of the same odd harmonic are multiplied by the local waveform difference to obtain a first product, and the harmonic similarity is obtained by negatively correlating and normalizing the products of all different odd harmonics; the calculation formula is as follows:

[0050]

[0051] wherein, This indicates that the load feeder and the common connection point are at the [number]th [number]. Harmonic similarity on odd harmonics Indicates the length of the harmonic amplitude sequence. The voltage sequence at the point of common coupling is represented in the first... The harmonic amplitude sequence on the odd harmonic One element, Indicates the load feeder current sequence in the th order. The harmonic amplitude sequence on the odd harmonic One element, The first harmonic amplitude variation sequence representing the voltage sequence at the point of common coupling. One element, Indicates the first The harmonic amplitude variation sequence of the load feeder current sequence is the first One element, This indicates that the common connection point and the load feeder are at the [number]th [number]. The harmonic amplitude sequence on the odd harmonic. Local waveform differences between window sequences corresponding to each element Indicated by An exponential function with base 0.

[0052] The smaller the difference in the harmonic amplitude sequence, the closer the amplitude values ​​of the common coupling point and the load feeder are on that odd harmonic, and the more similar the harmonics are. The smaller the difference in the harmonic amplitude variation sequence, the closer the amplitude variations of the common coupling point and the load feeder are on that odd harmonic, and the more similar the harmonics are. The smaller the difference in local waveforms, the more similar the waveforms are within the window sequence corresponding to that data.

[0053] Existing technologies perform segmented clustering or directly calculate similarity on the collected voltage or current signals. However, segmented clustering suffers from the drawback of neglecting the integrity and continuity of harmonic changes over continuous time. The global waveform cannot reflect the local harmonic trend and is easily dominated by the harmonic components at a certain moment, leading to errors in harmonic identification and failing to accurately locate the harmonic source.

[0054] S8: Calculate the significance of identifying odd harmonics based on the amplitude sequence of odd harmonics in the voltage sequence at the point of common coupling.

[0055] The first ratio is the ratio of the standard deviation of all data at the same data position in the amplitude sequence of the odd harmonics of the common junction voltage sequence to the average standard deviation of the amplitude sequence of the odd harmonics of the common junction voltage sequence. The second ratio is the sum of the ratios of the amplitude sequences of the odd harmonics of the common junction voltage sequence at the same data position to the average of the amplitude sequences of the odd harmonics of the common junction voltage sequence. The significance of identification is obtained by multiplying the first ratio and the second ratio.

[0056] To determine the significance of odd harmonic identification, calculate the standard deviation of the amplitude sequence of the odd harmonic of a common junction voltage sequence, and average the standard deviations of the amplitude sequences of the odd harmonics of all common junction voltage sequences. The calculation formula is as follows:

[0057]

[0058] in, Indicates the first The significance of identifying odd harmonics The voltage sequence at the point of common coupling is represented in the first... The standard deviation of all elements in the harmonic amplitude sequence on an odd harmonic. This represents the average of the standard deviations of all odd harmonics in the voltage sequence at the point of common coupling. Indicates the length of the harmonic amplitude sequence. The voltage sequence at the point of common coupling is represented in the first... The harmonic amplitude sequence on the odd harmonic One element, The harmonic amplitude sequence representing all odd harmonics of the voltage sequence at the point of common coupling is the first harmonic of the sequence. The average of the elements.

[0059] The greater the significance of identifying odd harmonics, the greater their volatility, indicating that the odd harmonics are more unstable and the common junction point is more affected, and therefore they should be given more attention.

[0060] In one embodiment, photovoltaic systems primarily generate the 3rd harmonic due to PWM (Pulse Width Modulation) and three-phase imbalance. Inverters such as electric arc furnaces, affected by 6-pulse rectification due to three-phase balance and filter suppression, primarily generate the 5th and 7th harmonics. Therefore, the significance of odd harmonics for harmonic identification varies depending on the specific situation. Thus, analyzing the energy values ​​and fluctuation levels of different odd harmonics at the point of common coupling to determine their significance is crucial for harmonic identification.

[0061] S9: Calculate the harmonic contribution of the load feeder to the common connection point based on harmonic similarity and identification significance.

[0062] After normalizing the significance of the identification of the same odd harmonic, the product of the harmonic similarity is taken as the first product value. The harmonic contribution is obtained by summing the product values ​​of all different odd harmonics. The calculation formula is as follows:

[0063]

[0064] in, This indicates the degree of harmonic contribution of a load feeder to the point of common coupling. indicates the number of odd harmonics, indicates the normalized identification significance of the th odd harmonic, indicates the harmonic similarity of the load feeder and the point of common coupling on the th odd harmonic.

[0065] The greater the harmonic contribution degree, the more the harmonics of the point of common coupling are mainly generated by the harmonic of the feeder, and the greater the possibility that the corresponding load of the feeder belongs to the harmonic source.

[0066] The harmonic contribution degrees are sorted from large to small, and a preset proportion of the harmonic contribution degrees that meet the preset condition are selected in the order from large to small, or the harmonic contribution degrees are sorted from small to large, and a preset proportion of the harmonic contribution degrees that meet the preset condition are selected in the order from small to large, and the technicians perform power grid maintenance and maintenance according to the load position corresponding to the harmonic source.

[0067] The embodiment of the application further discloses a power grid harmonic identification system based on deep learning, comprising a processor and a memory, and the memory stores computer program instructions, and when the computer program instructions are executed by the processor, a power grid harmonic identification method based on deep learning according to the application is realized.

[0068] The above system further comprises a communication bus and a communication interface and other components familiar to those skilled in the art, and the settings and functions thereof are known in the art, and therefore will not be described here.

[0069] In the present application, the aforementioned memory can be any tangible medium containing or storing a program, which can be used or combined with an instruction execution system, device or instrument. For example, the computer readable storage medium can be any appropriate magnetic storage medium or magneto-optical storage medium, such as resistive memory, dynamic random access memory, static random access memory, etc., or any other medium that can be used to store the required information and can be accessed by an application program, module or both. Any such computer storage medium can be part of the device or accessible or connectable to the device.

[0070] The above are the preferred embodiments of the present application, which do not limit the protection scope of the present application, therefore: any equivalent changes made on the basis of the structure, shape, principle of the present application shall be covered within the protection scope of the present application.

Claims

1. A deep learning-based method for identifying power grid harmonics, characterized in that, Including the following steps: Calculate the harmonic contribution of each load feeder of the power grid to the point of common coupling, and select the load corresponding to the load feeder whose harmonic contribution meets the preset conditions as the identified harmonic source. The method for calculating the harmonic contribution is as follows: The collected point of common coupling voltage data is standardized to remove dimensions, then input into the trained LSTM model to output voltage prediction data, which is then processed to obtain the point of common coupling voltage sequence; similarly, the load feeder current sequence is obtained. The collected common coupling point voltage sequence is divided according to the preset window to obtain multiple common coupling point voltage window sequences; similarly, multiple load feeder current window sequences are obtained. The voltage window sequence at the point of common coupling is transformed to obtain the amplitude spectrum, and the voltage amplitude sequence at the point of common coupling is obtained using the amplitude spectrum; similarly, the current amplitude sequence of the load feeder is obtained; based on the voltage amplitude sequence at the point of common coupling and the current amplitude sequence of the load feeder, the local waveform difference between the corresponding window sequences of the load feeder and the point of common coupling is calculated. For a frequency corresponding to an odd harmonic, the amplitude of that frequency within the amplitude spectrum of the voltage window sequence at the common coupling point is extracted to construct the amplitude sequence of the odd harmonics of the common coupling point voltage sequence; similarly, the amplitude sequence of the odd harmonics of the load feeder current sequence is obtained. Each data point in the amplitude sequence of the odd harmonics of the point of common coupling voltage sequence is processed to obtain the harmonic amplitude variation sequence of the point of common coupling voltage sequence; similarly, the harmonic amplitude variation sequence of the load feeder current sequence is obtained. Based on the amplitude sequences of odd harmonics in the common coupling point voltage sequence, the amplitude sequences of odd harmonics in the load feeder current sequence, the harmonic amplitude variation sequences of the common coupling point voltage sequence, the harmonic amplitude variation sequences of the load feeder current sequence, and the differences from local waveforms, the harmonic similarity between the load feeder and the common coupling point in odd harmonics is calculated. Based on the amplitude sequences of odd harmonics in the common coupling point voltage sequence, the significance of identifying odd harmonics is calculated. Based on the harmonic similarity and significance of identification, the harmonic contribution of the load feeder to the common coupling point is calculated.

2. The deep learning-based power grid harmonic identification method according to claim 1, characterized in that, For the power grid to be identified, the voltage data of the point of common coupling is obtained. After processing the obtained data, the mean square error is selected as the loss function.

3. The deep learning-based power grid harmonic identification method according to claim 1, characterized in that, Use the reciprocal of the grid current or voltage fundamental frequency as the window length.

4. The deep learning-based power grid harmonic identification method according to claim 1, characterized in that, The calculation method for the local waveform difference is as follows: the DTW distance between the current window sequence of the load feeder and the voltage window sequence of the corresponding common coupling point is taken as the first distance, the DTW distance between the amplitude sequence of the current window sequence of the load feeder and the voltage window sequence of the corresponding common coupling point is taken as the second distance, and the local waveform difference is obtained by multiplying the first distance and the second distance.

5. The deep learning-based power grid harmonic identification method according to claim 1, characterized in that, The method for processing each data point in the amplitude sequence of the odd harmonics of the voltage sequence at the common junction point is as follows: the first-order difference value of each data point in the harmonic amplitude sequence of the voltage window sequence at the common junction point is compared with the data point to obtain the harmonic amplitude variation sequence of the voltage sequence at the common junction point.

6. The deep learning-based power grid harmonic identification method according to claim 1, characterized in that, The method for calculating harmonic similarity is as follows: the absolute difference between the amplitude sequence of the odd harmonic of the load feeder current sequence and the amplitude sequence of the same odd harmonic of the common coupling point voltage sequence is taken as the first difference; the absolute difference between the harmonic amplitude change sequence of the common coupling point voltage sequence and the corresponding data point of the same odd harmonic of the load feeder current sequence is taken as the second difference; the first difference, the second difference and the local waveform difference of the same odd harmonic are multiplied as a first product; after accumulating the products under all different odd harmonics, negative correlation normalization is performed to obtain the harmonic similarity.

7. The deep learning-based power grid harmonic identification method according to claim 1, characterized in that, The method for calculating the significance level is as follows: the ratio of the standard deviation of all data at the same data position in the amplitude sequence of the odd harmonic of the common junction voltage sequence to the average standard deviation of the amplitude sequence of the odd harmonic of the common junction voltage sequence is taken as the first ratio; the sum of the ratios of the amplitude sequence of the odd harmonic of the common junction voltage sequence at the same data position to the average of the amplitude sequences of the odd harmonic of the common junction voltage sequence is taken as the second ratio; and the first ratio and the second ratio are multiplied to obtain the significance level.

8. The deep learning-based power grid harmonic identification method according to claim 1, characterized in that, The method for calculating the harmonic contribution is as follows: after normalizing the significance of the identification of the same odd harmonic, multiply it by the harmonic similarity as a first product value, and sum the product values ​​of all different odd harmonics to obtain the harmonic contribution.

9. The deep learning-based power grid harmonic identification method according to claim 1, characterized in that, The conditions that meet the preset criteria are: sorting the harmonic contributions from largest to smallest, and selecting a preset proportion of harmonic contributions that meet the preset criteria in descending order; or sorting the harmonic contributions from smallest to largest, and selecting a preset proportion of harmonic contributions that meet the preset criteria in ascending order.

10. A power grid harmonic identification system based on deep learning, characterized in that, include: A processor and a memory, the memory storing computer program instructions that, when executed by the processor, implement a deep learning-based power grid harmonic identification method according to any one of claims 1-9.

Images