Dense signal component identification method and system for distributed photovoltaic grid connection

Abstract

The invention discloses a dense signal component identification method and system for distributed photovoltaic grid connection. The method comprises the following steps: S1, establishing a model of dense signals and contrast signals; s2, respectively generating w groups of dense signals and contrast signals; s3, windowing the dense signals and the contrast signals and completing time-frequency domain conversion of the signals; s4, inputting the dense signals converted in the step S3 and the contrast signals into a BP neural network for training, and judging the number M of components contained in the dense frequency band of the unknown to-be-detected signal according to an output classification result; s5, sampling a to-be-measured signal, performing discrete Fourier transform, and determining an initial estimation frequency corresponding to the non-dense frequency band component according to the quasi-peak value of the frequency domain information; and S6, solving based on the BP neural network to obtain the frequency value of each component of the to-be-detected signal. According to the method, robustness classification and frequency parameter estimation of dense signal components are realized through collaborative optimization of the frequency domain features and the neural network.

Description

Technical Field

[0001] The present invention relates to the technical field of signal recognition, and particularly to a method and system for identifying dense signal components for distributed photovoltaic grid connection. Background Art

[0002] With the large-scale access of high-proportion new energy and power electronic devices to the power grid, the interharmonic components in various frequency bands of the power grid signal have increased significantly. The occurrence frequency bands of interharmonics are randomly distributed. When the interharmonics are close to the fundamental or harmonic frequencies, the signal spectrum components overlap with each other, and it is difficult to distinguish the number and components of the signals, seriously affecting the estimation accuracy of key parameters such as the frequency of the signal components. This situation is more serious near the fundamental frequency. Since the fundamental amplitude is usually more than ten times that of the interharmonics, the spectral leakage value of the fundamental wave has the risk of submerging the adjacent interharmonic components, posing challenges to power quality monitoring and harmonic and interharmonic governance.

[0003] The spectral analysis method based on the discrete Fourier transform (DFT) dominates in the detection of power system signals, but its inherent defects are particularly prominent in the scenario of dense components. When the interval between signal components is less than three times the frequency resolution, the main lobe overlap of the DFT spectrum will cause multiple components to be misjudged as a single component, and its side lobe interference will make it difficult to distinguish adjacent components and components with larger amplitudes.

[0004] For the problem of identifying dense signal components based on frequency domain information, the existing methods include: spectral quasi-peak extraction, spectral peak iterative correction, superposition of adaptive window functions, etc. However, when the frequency interval between signal components is small and there is strong noise interference, there are still problems such as low recognition accuracy and poor anti-noise performance, which are difficult to meet the requirements of the new power system for refined analysis of dense signals. Therefore, how to identify the dense signal components of the power system based on frequency domain information is an urgent problem to be solved at present. Summary of the Invention

[0005] The purpose of the present invention is to provide a method and system for identifying dense signal components for distributed photovoltaic grid connection, which provides core technical support for the precise governance of harmonics and interharmonics in a complex power grid environment through the collaborative optimization of frequency domain information and BP neural network.

[0006] To achieve the above purpose, the present invention provides the following technical solutions:

[0007] On the one hand, the present invention provides a method for identifying dense signal components for distributed photovoltaic grid connection, including the following steps:

[0008] S1. Original signal modeling: Sampling the dense signal and the reference signal, and establishing models of the dense signal and the reference signal;

[0009] S2. Dataset Generation: Generate w groups of dense signals and w groups of control signals, and set the label matrix corresponding to the dense signals and the label matrix corresponding to the control signals;

[0010] S3. Time-Frequency Domain Conversion: Window the dense signals and the control signals, and complete the conversion of the signals from the time domain to the frequency domain through the discrete Fourier transform;

[0011] S4. Construction and Training of the Pattern Recognition Network: Input the dense signals and the control signals after the time-frequency domain conversion in S3 into the BP neural network for training. According to the output classification result, judge the number M of components contained in the dense frequency band of the unknown signal to be measured, and the recognition result is M = 1 or M = 2;

[0012] S5. Preprocessing of the Signal to be Measured: Sample the signal to be measured and perform the discrete Fourier transform, and determine the initial estimated frequency corresponding to the non-dense frequency band components according to the quasi-peak value of the frequency domain information;

[0013] S6. Parameter Solving Based on the BP Neural Network: Set the activation function of the hidden layer and solve the frequency values of each component of the signal to be measured.

[0014] In some embodiments, the model of the control signal is:

[0015]

[0016] The model of the dense signal is:

[0017]

[0018] where f u = f v + Δf·(l + rand);

[0019] In the formula, A v is the amplitude of the control component; A u is the amplitude of the dense component; f v is the frequency of the control component; f u is the frequency of the dense component; is the phase of the control component; is the phase of the dense component; s noise is the Gaussian white noise; t s is the sampling period, t s = 1 / f s , f s is the sampling frequency; n = 0, 1, 2,..., N - 1, N is the total number of sampling points; Δf = 1 / (t s N) is the frequency resolution; l is the minimum spectral spacing, taking a positive real number; rand is a random number between (0, 1).

[0020] In some embodiments, S2 includes the following steps:

[0021] Obtain the frequency f of the control component with a fixed value v , and obtain the frequency f of w groups of dense components u and w groups of Gaussian white noise s noise , and generate w groups of dense signals and w groups of control signals;

[0022] Set the label matrix Γ corresponding to the dense signal and the label matrix Γ corresponding to the control signal ′ as a 2×w order matrix;

[0023] wherein, any column value of Γ is [0,1] ′ ; Γ ′ any column value of is [1,0] ′ .

[0024] In some embodiments, in S3, the frequency-domain signal after windowing the dense signal is:

[0025]

[0026] The frequency-domain signal after windowing the control signal is:

[0027]

[0028] wherein, W(·) is the spectrum function after applying the Hanning window; k = 0, 1, 2,..., K are the spectral lines corresponding to the respective frequencies of the frequency-domain signal, K is the maximum measurable frequency value, and according to the Nyquist sampling theorem, K = ceil(f s / 2); A v is the amplitude of the control component; A u is the amplitude of the dense component; j is the imaginary unit; f v is the frequency of the control component; f u is the frequency of the dense component; is the phase of the control component; is the phase of the dense component; Δf is the frequency resolution.

[0029] In some embodiments, in S5, the discrete sequence obtained by sampling the signal to be measured is:

[0030]

[0031] wherein, D is the number of components contained in the non-dense frequency band of the signal to be measured; M is the number of components contained in the dense frequency band of the signal to be measured; A i is the amplitude corresponding to the i-th component in the signal to be measured; f i is the frequency corresponding to the i-th component in the signal to be measured; is the phase corresponding to the i-th component in the signal to be measured; snoise is Gaussian white noise; t s is the sampling period, t s = 1 / f s , f s is the sampling frequency; n = 0, 1, 2, ..., N-1, where N is the total number of sampling points;

[0032] For the dense frequency band of the signal to be measured, when M = 1, the local quasi-peak value in the frequency domain is taken as the initial estimated frequency, and when M = 2, the front and back spectra of the quasi-peak value in the frequency domain are taken as the initial estimated frequency.

[0033] In some embodiments, S6 includes the following steps:

[0034] Establish a BP neural network. The total number of layers of the neural network is set to 3, the number of neurons in the hidden layer is set to 2×(D+M), and the activation function is a combination of sine and cosine; D is the number of components contained in the non-dense frequency band of the signal to be measured; M is the number of components contained in the dense frequency band of the signal to be measured;

[0035] Gradually optimize the network weights and biases through forward propagation, error calculation, and backpropagation until the error is less than the preset accuracy or the maximum number of iterations is reached;

[0036] Obtain the estimation matrices of amplitude, phase, and frequency and Use the estimation matrix of frequency to obtain the frequency values of each component.

[0037] In some embodiments, the activation function of the hidden layer is:

[0038] δ = cos(bt s n)·ω1 + sin(bt s n)·ω2;

[0039] In the formula, t s is the sampling period, t s = 1 / f s , f s is the sampling frequency; n = 0, 1, 2, ..., N-1, where N is the total number of sampling points; ω1, ω2, and b are the initial estimation matrices of amplitude, phase, and frequency respectively.

[0040] On the other hand, the present invention provides a dense signal component identification system for distributed photovoltaic grid connection, using the above method, including:

[0041] Original signal modeling module: used to sample the dense signal and the reference signal, and establish the models of the dense signal and the reference signal;

[0042] Dataset generation module: used to generate w groups of dense signals and w groups of reference signals;

[0043] Time-frequency domain conversion module: By windowing the dense signal and the reference signal, and using the discrete Fourier transform to complete the conversion of the signal from the time domain to the frequency domain;

[0044] Pattern recognition network construction and training module: Input the dense signal and the reference signal converted by the time-frequency domain conversion module into the BP neural network for training. According to the output classification result, judge the number M of components contained in the dense frequency band of the unknown signal to be measured, and the recognition result is M = 1 or M = 2;

[0045] Preprocessing module for the signal to be measured: Used to sample the signal to be measured and perform discrete Fourier transform, and determine the initial estimated frequency corresponding to the non-dense frequency band components according to the quasi-peak value of the frequency domain information;

[0046] BP neural network parameter solving module: By setting the activation function of the hidden layer of the BP neural network, solve the frequency values of each component of the signal to be measured.

[0047] Compared with the prior art, the present invention has the following beneficial effects:

[0048] The multi-dimensional feature extraction method of the present invention based on spectral features, leakage energy distribution characteristics and frequency point correlation strengthens the modeling ability of the neural network for the dense component boundary and superposition effect; through the collaborative optimization of frequency domain features and neural networks, robust classification of dense signal components and frequency parameter estimation are realized. Brief description of the drawings

[0049] Figure 1 It is a schematic diagram of the overall process of the present invention;

[0050] Figure 2 It is a schematic diagram of the spectrum of the dense signal in Embodiment 1 of the present invention;

[0051] Figure 3 It is a schematic diagram of the recognition effect of the number of dense frequency band components in Embodiment 1 of the present invention. Among them, (a) is the recognition result of the confusion matrix of the training data, (b) is the recognition result of the confusion matrix of the verification data, (c) is the recognition result of the confusion matrix of the test data, and (d) is the recognition result of the confusion matrix of all data. Detailed implementation manners

[0052] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments; based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0053] Embodiment 1:

[0054] Please refer to Figure 1 , a method for identifying intensive signal components for distributed photovoltaic grid connection, comprising the following steps:

[0055] S1. Original signal modeling;

[0056] Sample the intensive signal and the reference signal, and establish models of the intensive signal and the reference signal.

[0057] Set the sampling period as t s , and the total number of sampling points as N. Take the reference signal x ′ (t) as a signal containing only a single frequency component, and denote the frequency of the signal with the single frequency component as f v ; Add a component with a frequency of f ′ to the reference signal x u (t) to obtain the intensive signal x(t). The value of the frequency f u of the intensive component is as follows:

[0058] f u = f v + Δf·(l + rand)

[0059] In the formula, Δf is the frequency resolution, Δf = 1 / (t s N) is the frequency resolution; l is the minimum spectral spacing, taking a positive real number; rand is a random number between (0,1).

[0060] Sample the reference signal x ′ (t) and the intensive signal x(t) respectively to obtain discrete sequences x ′ (n) and x(n). Define the discrete sequence models of the reference signal and the intensive signal x ′ (n), x(n) as follows:

[0061]

[0062] S2. Dataset generation;

[0063] Generate w groups of intensive signals and w groups of reference signals;

[0064] Take the frequency f v of the reference component with a fixed value (determined according to the grid harmonic characteristics and the possible frequency range of the intensive signal after distributed photovoltaic grid connection, generally in the range of [100,500]) u , the frequencies f noise of w groups of intensive components and w groups of Gaussian white noise s ′ to generate w groups of intensive signals x(n) and w groups of reference signals x

[0065] Set up the label matrix Γ corresponding to the dense signal x(t) and the label matrix Γ corresponding to the control signal ′ 。The label matrices Γ and Γ ′ are both 2×w order matrices, and any column value of Γ is [0,1] ′ ; Γ ′ Any column value of is [1,0] ′ 。

[0066] S3. Time-frequency domain conversion;

[0067] Window the dense signal and the control signal, and complete the conversion of the signal from the time domain to the frequency domain through the discrete Fourier transform.

[0068] Window the discrete sequence x(n) of the dense signal and the discrete sequence x ′ (n) of the control signal, and complete the time-frequency domain conversion of the signal through the discrete Fourier (DFT) transform. The frequency domain signal of the control signal after windowing is X′ w (k):

[0069]

[0070] The frequency domain signal of the dense signal after windowing is X w (k):

[0071]

[0072] S4. Construction and training of the pattern recognition network;

[0073] Merge the w groups of frequency domain signals X w (k) and X′ w (k) into the dense signal data matrix Φ and the control signal data matrix Φ ′ respectively. The dense signal data matrix Φ and the control signal data matrix Φ ′ are both N×w order matrices and correspond to the label matrices Γ and Γ ′ Randomly divide the dense signal data matrix Φ and the control signal data matrix Φ according to the ratio of 70% for training data, 15% for validation data, and 15% for test data ′ 。

[0074] Construct a pattern recognition neural network with 3 hidden layer nodes, and use the training data and the target feature data to train the network. According to the output classification result, the number M of components contained in the dense frequency band of the unknown signal to be measured can be judged, and the recognition result is M = 1 or M = 2.

[0075] S5. Pretreatment of the signal to be measured;

[0076] Let the number of components contained in the signal y(t) to be measured be D + M, where D is the number of components contained in the non-dense frequency band of the signal y(t) to be measured, and M is the number of components contained in the dense frequency band of the signal y(t) to be measured. Sample the signal y(t) to be measured to obtain a discrete sequence y(n).

[0077] Define its signal model as:

[0078]

[0079] Perform a discrete Fourier transform on the signal. Determine the initial estimated frequencies corresponding to the components in the non-dense frequency band according to the quasi-peak values of the frequency-domain information; for the dense frequency band, when M = 1, take the local quasi-peak value in the frequency domain as the initial estimated frequency, and when M = 2, take the front and back spectra of the quasi-peak value in the frequency domain as the initial estimated frequencies.

[0080] S6. Solve the parameters based on the BP neural network;

[0081] Build a BP neural network. The total number of layers of the neural network is set to 3, and the number of neurons in the hidden layer is set to 2×(D + M), where D is the number of components contained in the non-dense frequency band of the signal to be measured; M is the number of components contained in the dense frequency band of the signal to be measured. The activation function δ is a combination of sine and cosine:

[0082] δ = cos(bt s n)·ω1 + sin(bt s n)·ω2;

[0083] The relevant parameters are as follows: in the established neural network, the learning rates of amplitude, phase, and frequency are l1, l2, and λ respectively; the initial estimated matrices ω1, ω2, and b of amplitude, phase, and frequency, where ω1 and ω2 take random values in (0, 1).

[0084] Gradually optimize the network weights and biases through forward propagation, error calculation, and backpropagation until the error is less than the preset accuracy or the maximum number of iterations is reached.

[0085] The estimated matrices of amplitude, phase, and frequency can be obtained through this BP neural network and Then the frequency f of each component can be obtained through the following formula:

[0086]

[0087] In a specific embodiment,

[0088] Take the sampling frequency f s as 5 kHz, and the number of sampling points is taken as N = 200. At this time, the frequency resolution Δf is 25 Hz. The component parameters corresponding to the signal to be measured are shown in Table 1, and 40 dB Gaussian white noise is added to the signal to be measured.

[0089] Table 1 Parameter Table of Dense Signal Components

[0090]

[0091] As Figure 2 shown in the dense signal spectrogram, the number of quasi-peak values of signals with peak values greater than 0.05 is 4, which does not match the actual number of signal components. It is difficult to obtain effective detection results through iterative correction of spectral quasi-peak values or spectral peak values.

[0092] This embodiment can illustrate the beneficial effects of the present invention from the perspective of the recognition ability of the number of dense frequency band signal components and the solution results of dense signal parameters.

[0093] As Figure 3 shown, it is the recognition effect of the number of dense frequency band components in the fundamental wave (amplitude ratio 10:1) frequency band, and the recognition rate is 100% at this time. Figure 3 (a)-(d) respectively represent the recognition results of the number of dense frequency band components based on training data, validation data, test data, and total generated data. Among them, the green area gives the number of data with consistent predicted labels and true labels and the percentage of the total data, and the red area gives the number of data with inconsistent predicted labels and true labels and the percentage of the total data. Taking Figure 3 (a) as an example, among 280 randomly selected training data, this embodiment achieved the recognition of 140 reference signals (accounting for 50%) and 140 dense signals (accounting for 50%), and the recognition accuracy rate was 100%.

[0094] For the established neural network, its parameter selection is as follows: the minimum frequency spacing multiple l = 0.4, the number of data groups taken w = 200, the number of components contained in the non-dense frequency band D = 2, the amplitude learning rate l1 = 0.1, the phase learning rate l2 = 0.1, and the frequency learning rate λ = 0.04. According to the recognition results of the pattern recognition network and the spectral information, the initial estimation matrix of the frequency is set as:

[0095] b = [50·2π, 75·2π, 250·2π, 275·2π, 450·2π, 575·2π];

[0096] The detection results are shown in Table 2.

[0097] Table 2 Detection Result Table of Dense Signal Component Frequencies

[0098]

[0099] As shown in Table 2, this embodiment can accurately identify each component in the dense signal and obtain effective and accurate detection results. The detection results can be used to evaluate the power quality (such as power factor, waveform distortion, etc.) after the distributed photovoltaic inverter is connected to the grid, help adjust the inverter control strategy, reduce harmonic pollution and voltage imbalance problems, and ensure that the grid-connected electric energy meets the national standards.

[0100] Embodiment 2

[0101] A dense signal component identification system for distributed photovoltaic grid connection, using the above method, includes:

[0102] Original signal modeling module: used to sample the dense signal and the reference signal, and establish models of the dense signal and the reference signal;

[0103] Dataset generation module: used to generate w groups of dense signals and w groups of reference signals;

[0104] Time-frequency domain conversion module: by windowing the dense signal and the reference signal, use the discrete Fourier transform to complete the conversion of the signal from the time domain to the frequency domain;

[0105] Pattern recognition network construction and training module: input the dense signal and the reference signal converted by the time-frequency domain conversion module into the BP neural network for training, and judge the number M of components contained in the dense frequency band of the unknown signal to be measured according to the output classification result, and the recognition result is M = 1 or M = 2;

[0106] Preprocessing module for the signal to be measured: used to sample the signal to be measured and perform discrete Fourier transform, and determine the initial estimated frequency corresponding to the non-dense frequency band components according to the quasi-peak value of the frequency domain information;

[0107] BP neural network parameter solving module: by setting the activation function of the hidden layer of the BP neural network, solve the frequency values of each component of the signal to be measured.

[0108] A power system dense signal component identification system based on frequency domain information of the present invention can be installed in a computer device. The computer device includes a processor, a memory, and a computer program stored in the memory and executable on the processor, such as a power system dense signal component identification program based on frequency domain information. Among them, the memory includes at least one type of readable storage medium, and the readable storage medium includes flash memory, mobile hard disk, multimedia card, card-type memory (such as: SD or DX memory, etc.), magnetic memory, magnetic disk, optical disc, etc. The processor is the control core of the electronic device, connects various components of the entire computer device through various interfaces and lines, and executes various functions of the computer device and processes data by running or executing programs or modules stored in the memory, and calling data stored in the memory.

[0109] As used herein, the module refers to a series of computer program segments that can be executed by a processor of a computer device and can perform fixed functions, and are stored in the memory of the computer device.

[0110] Those skilled in the art will readily conceive of other embodiments of the present application after considering the specification and practicing the disclosure herein. The present application is intended to cover any variations, uses, or adaptations of the present application, which follow the general principles of the present application and include known common general knowledge or conventional technical means in the technical field not disclosed in the present application.

Claims

1. A method for identifying intensive signal components for distributed photovoltaic grid connection, characterized in that It includes the following steps: S1. Sample the dense signal and the reference signal, and establish models for the dense signal and the reference signal; S2. Generate w groups of dense signals and w groups of reference signals; S3. Window the dense signal and the reference signal, and complete the conversion of the signal from the time domain to the frequency domain through discrete Fourier transform; S4. Input the dense signal and the reference signal after the time-frequency domain conversion in S3 into a BP neural network for training. According to the output classification result, judge the number M of components contained in the dense frequency band of the unknown signal to be measured, and the recognition result is M = 1 or M = 2; S5. Sample the signal to be measured and perform discrete Fourier transform, and determine the initial estimated frequency corresponding to the non-dense frequency band components according to the quasi-peak value of the frequency domain information; S6. Set the activation function of the hidden layer of the BP neural network, and solve to obtain the frequency values of each component of the signal to be measured.

2. The intensive signal component identification method for distributed photovoltaic grid connection according to claim 1, wherein The model of the reference signal is: The model of the dense signal is: Wherein, f u = f v + Δf·(l + rand); Where, A v is the amplitude of the reference component; A u is the amplitude of the dense component; f v is the frequency of the reference component; f u is the frequency of the dense component; is the phase of the reference component; is the phase of the dense component; s noise is Gaussian white noise; t s is the sampling period, t s = 1 / f s , f s is the sampling frequency; n = 0, 1, 2, ..., N-1, where N is the total number of sampling points; Δf = 1 / (t s N) is the frequency resolution; l is the minimum spectral spacing, taking a positive real number; rand is a random number between (0, 1).

3. A method for identifying intensive signal components for distributed photovoltaic grid connection according to claim 1, characterized in that, S2 includes the following steps: Take the frequency f of the reference component v as a fixed value, and take the frequencies f of w groups of dense components u and w groups of Gaussian white noise s noise to generate w groups of dense signals and w groups of reference signals; Set the label matrix Γ corresponding to the dense signal and the label matrix Γ corresponding to the control signal ′ as a 2×w order matrix; In the formula, any column value of Γ is in the range of [0, 1] ′ ; Γ ′ Any column value of ′ is in the range of [1, 0] 4. A method for identifying intensive signal components for distributed photovoltaic grid connection according to claim 1, characterized in that, In S3, the frequency domain signal of the dense signal after windowing is: The frequency domain signal of the reference signal after windowing is: where \(W(\cdot)\) is the spectrum function after applying the Hanning window; \(k = 0, 1, 2, \cdots, K\) are the spectral lines corresponding to the frequencies of the frequency-domain signal, \(K\) is the maximum measurable frequency value, and according to the Nyquist sampling theorem, \(K=\lceil f s / 2\rceil\); \(A v \) is the amplitude of the reference component; \(A u \) is the amplitude of the dense component; \(j\) is the imaginary unit; \) is the phase of the reference component; \) is the phase of the dense component; \(f v \) is the frequency of the reference component; \(f u \) is the frequency of the dense component; \(\Delta f\) is the frequency resolution.

5. A method for identifying intensive signal components for distributed photovoltaic grid connection according to claim 1, characterized in that In S5, the discrete sequence obtained by sampling the signal to be measured is: Where D is the number of components contained in the non-dense frequency band of the signal to be measured; M is the number of components contained in the dense frequency band of the signal to be measured; A i is the amplitude corresponding to the i-th component in the signal to be measured; f i is the frequency corresponding to the i-th component in the signal to be measured; is the phase corresponding to the i-th component in the signal to be measured; s noise is Gaussian white noise; t s is the sampling period, t s = 1 / f s where f s is the sampling frequency; n = 0, 1, 2,..., N - 1, and N is the total number of sampling points; For the dense frequency band of the signal to be measured, when M = 1, take the local quasi-peak value of the frequency domain as the initial estimated frequency, and when M = 2, take the front and back spectra of the quasi-peak value of the frequency domain as the initial estimated frequency.

6. A method for identifying dense signal components for distributed photovoltaic grid connection according to claim 1, characterized in that S6 includes the following steps: Build a BP neural network, set the total number of layers of the neural network to 3, set the number of neurons in the hidden layer to 2×(D + M), and the activation function is a combination of sine and cosine; D is the number of components contained in the non-dense frequency band of the signal to be measured; M is the number of components contained in the dense frequency band of the signal to be measured; Gradually optimize the network weights and biases through forward propagation, error calculation, and backpropagation until the error is less than the preset accuracy or the maximum number of iterations is reached; Obtain the estimated matrices of amplitude, phase and frequency and Using the estimated matrix of frequency Obtain the frequency values of each component.

7. A method for identifying intensive signal components for distributed photovoltaic grid connection according to claim 6, characterized in that The activation function of the hidden layer is: δ = cos(bt s n)·ω1 + sin(bt s n)·ω2; where t s is the sampling period, and t s = 1 / f s , f s is the sampling frequency; n = 0, 1, 2, ..., N - 1, N is the total number of sampling points; ω1, ω2, and b are the initial estimation matrices of amplitude, phase, and frequency, respectively.

8. A dense signal component identification system for distributed photovoltaic grid connection, using the method according to any one of claims 1-7, characterized in that It includes: Original signal modeling module: used to sample the dense signal and the reference signal, and establish models for the dense signal and the reference signal; Data set generation module: used to generate w groups of dense signals and w groups of reference signals; Time-frequency domain conversion module: complete the conversion of the signal from the time domain to the frequency domain by windowing the dense signal and the reference signal and using discrete Fourier transform; Pattern recognition network construction and training module: input the dense signal and the reference signal after conversion by the time-frequency domain conversion module into a BP neural network for training. According to the output classification result, judge the number M of components contained in the dense frequency band of the unknown signal to be measured, and the recognition result is M = 1 or M = 2; Preprocessing module of the signal to be measured: used to sample the signal to be measured and perform discrete Fourier transform, and determine the initial estimated frequency corresponding to the non-dense frequency band components according to the quasi-peak value of the frequency domain information; BP neural network parameter solving module: solve to obtain the frequency values of each component of the signal to be measured by setting the activation function of the hidden layer of the BP neural network.

Images