Multi-dimensional time sequence photovoltaic grid-connected harmonic current prediction control method

Abstract

The invention discloses a multi-dimensional time sequence photovoltaic grid-connected harmonic current prediction control method, and particularly relates to the technical field of power system control. Grid-connected original multi-dimensional electrical signals are collected and decomposed into multi-dimensional frequency signals through Fourier transform, the characteristic components of an independent harmonic source are separated through the blind source separation and non-negative matrix factorization technology, and the frequency and amplitude characteristics of the characteristic components are extracted; then, a multi-dimensional time sequence prediction model is constructed through a recurrent neural network, dynamic behaviors of multiple harmonic sources are predicted at the same time, and frequency domain data are reconstructed and frequency aliasing errors are corrected in combination with high-frequency compensation and multi-scale analysis; and finally, inputting a prediction result into a photovoltaic inverter control system, dynamically generating a compensation current opposite to a target harmonic wave, realizing effective offset of the harmonic wave, accurately processing the multi-source harmonic wave problem in a complex power grid environment, and greatly improving the harmonic wave suppression capability, the electric energy quality and the system operation stability of a photovoltaic grid-connected system.

Description

Technical Field

[0001] The present invention relates to the technical field of power system control, and in particular to a multi-dimensional time-series photovoltaic grid-connected harmonic current prediction and control method. Background Art

[0002] Multi-dimensional time-series photovoltaic grid-connected harmonic current predictive control is an advanced control technology specifically used for photovoltaic power generation systems. The photovoltaic system converts direct current into alternating current through an inverter and connects it to the power grid, but in this process, harmonic currents may be generated, affecting the power supply quality of the power grid. Multi-dimensional time-series predictive control technology aims to suppress and optimize harmonic currents in photovoltaic power generation systems in advance by predicting their future behavior. This method uses the dynamic association between multi-dimensional variables (such as current, voltage, power factor, etc.) and time-series data, and accurately predicts and adjusts harmonic characteristics through mathematical modeling and optimization algorithms, thereby achieving stable operation of photovoltaic grid-connected systems.

[0003] Compared with traditional harmonic suppression technology, multi-dimensional time sequence predictive control has significant advantages. First, it can predict future harmonic change trends based on the current system status and historical data, making control decisions more forward-looking. Second, with the help of multi-dimensional data analysis, it can comprehensively consider the impact of multiple factors on harmonic currents, optimize control strategies, and improve prediction accuracy. In addition, by updating the prediction model in real time, multi-dimensional time sequence control can adapt to dynamic changes in the photovoltaic power generation environment, such as light intensity and temperature fluctuations.

[0004] The prior art has the following deficiencies:

[0005] There may be multiple independent harmonic sources in the power grid (such as industrial loads, grid-connected energy storage devices, etc.), which may be superimposed at different frequencies and amplitudes to form complex waveforms. If the model cannot distinguish the independent components of these harmonics, it may mistakenly regard the superposition effect as a single harmonic characteristic for prediction and control. And this superposition effect may produce frequency aliasing, making the input data received by the prediction model seem to conform to a certain error pattern, which in turn triggers incorrect control and causes system oscillation or overload. Summary of the invention

[0006] The purpose of the present invention is to provide a multi-dimensional sequential photovoltaic grid-connected harmonic current prediction and control method to solve the shortcomings of the background technology.

[0007] In order to achieve the above object, the present invention provides the following technical solution: a multi-dimensional sequential photovoltaic grid-connected harmonic current prediction and control method, comprising the following steps:

[0008] S1: collecting original multi-dimensional electrical signals from photovoltaic grid-connected points, wherein the original multi-dimensional electrical signals include current, voltage and frequency, and decomposing the original multi-dimensional electrical signals into multi-dimensional frequency signals by Fourier transform;

[0009] S2: Using blind source separation algorithm and non-negative matrix decomposition technology, the characteristic components of independent harmonic sources are separated from the multi-dimensional frequency signal, and the frequency and amplitude characteristics of each independent harmonic source are extracted to reduce the prediction deviation caused by the superposition effect;

[0010] S3: Based on the separated independent harmonic source characteristics, a time series prediction model is constructed using a recurrent neural network. Multiple harmonic sources are processed simultaneously through a multidimensional prediction model to predict the future characteristics of the overall harmonic source.

[0011] S4: Combined with the separated harmonic source data, high-frequency compensation and reconstruction are performed on the frequency domain data, and the frequency aliasing error is further corrected through multi-scale analysis to ensure the accuracy of the frequency separation results;

[0012] S5: Input the multi-dimensional prediction results into the control system of the photovoltaic inverter, dynamically adjust the inverter output current, and offset the target harmonic components in the power grid.

[0013] Preferably, in S1, the frequency analysis of the multi-dimensional time-domain electrical signals of the photovoltaic grid-connected point is performed by Fourier transform, and the discrete current signal, voltage signal and frequency signal data in the time domain are collected. Discrete Fourier transform is applied to each signal respectively to convert it from the time domain to the frequency domain, and the corresponding spectrum information, including frequency components and amplitude characteristics, is extracted. The spectrum data of the current, voltage and frequency signals are combined to form a multi-dimensional frequency matrix.

[0014] Preferably, in S2, the multidimensional frequency signal matrix is ​​set to be composed of a linear superposition of frequency components of multiple independent harmonic sources, and the signal matrix is ​​processed by blind source separation to whiten and eliminate the correlation between the signals, so that the components of each harmonic source are independent of each other;

[0015] In the process of blind source separation, the separation matrix is ​​solved by the optimization algorithm, and it is verified whether the separation result conforms to the harmonic characteristics, so as to obtain the frequency components of each independent harmonic source. The separation result is decomposed by non-negative matrix decomposition, and the non-negative constraint is introduced to ensure that the separated frequency and amplitude are positive. The weight matrix and basic component matrix of each independent harmonic source are obtained by NMF, and the main frequency characteristics and amplitude characteristics of the harmonic source are extracted; the main frequency and amplitude of each independent harmonic source are calculated to form the frequency and amplitude characteristic vector of the independent harmonic source.

[0016] Preferably, in S3, the independent feature vector T is extracted r =[f r ,A r],r=1,2,…,R; the time series characteristics of each independent harmonic source are expressed as: X r ={T r (t1),T r (t2),…,T r (t N )}; where T r (t i )=[f r (t i ),A r (t i )] represents the time t i The frequency and amplitude characteristics of the time are calculated, r = 1, 2, ..., R, which indicates the harmonic source number, N is the length of the time series, the input data is standardized, and a prediction model is constructed using a recursive neural network; the model input is a multidimensional characteristic time series, and the output is the predicted value at the future moment;

[0017] Input data ET for each time step r (t i ), which is the frequency and amplitude characteristics after standardization; a recursive structure is used to capture the long-term and short-term dependencies of the time series; the state update formula of the hidden layer is: h t =φ(W h h t-1 +W x x t +b h );h t is the hidden layer state, x t is the input feature vector, W h ,W x ,b h is the hidden layer parameter, φ(·) is the activation function; the output layer is used to predict the future frequency and amplitude characteristics, and the output is: The time series {T r (t1),T r (t2),…,T r (t N )} is input, and the RNN model predicts the future time t N+1 Features: For multiple harmonic sources r = 1, 2, ..., R, multiple sub-models are trained simultaneously, and the loss function is used to evaluate the error between the predicted result and the true value. The expression is: Where: T r (t i ) is the true value, is the predicted value; the time series of each harmonic source is segmented by sliding window to form a training sample set; the training set and validation set are divided: 80% of the data is divided into a training set and 20% into a validation set.

[0018] Preferably, in S3, for the current time series {T r (t1),T r (t2),…,T r (t N )} Input the model and predict the next time t N+1 Features: Based on the single-step prediction results, the predicted value As input to the next step, iteratively predict multiple time steps into the future: The predicted harmonic source characteristic matrix is: M represents the number of time steps for prediction; Contains the frequency and amplitude characteristics of the harmonic source r at future times.

[0019] Preferably, in S4, based on the obtained harmonic source frequency and amplitude data, a spectrum diagram is drawn, and the high-frequency component is compensated by an exponential model using the characteristic that the amplitude decays with the frequency to generate a compensated high-frequency harmonic amplitude, and the high-frequency harmonic signal is reconstructed using an inverse Fourier transform in combination with the phase information; multi-scale decomposition is performed using a wavelet transform to divide the signal into low-frequency and high-frequency components, and abnormal frequency components can be identified and corrected through multi-scale decomposition, retaining low-frequency information and high-frequency components, while removing aliasing parts; the corrected high-frequency and low-frequency harmonic signals are merged to form a complete harmonic signal.

[0020] Preferably, in S5, the multi-dimensional prediction result is input into the control system of the photovoltaic inverter, and the inverter output current is dynamically adjusted to offset the target harmonic component in the power grid, specifically:

[0021] For the rth independent harmonic source, the harmonic component at the future time t is expressed as: r (t) = A r sin(2πf r t+φ r );A r is the predicted harmonic amplitude, f r is the predicted harmonic frequency, φ r is the predicted harmonic phase; the target harmonic signal in the power grid is the superposition of independent harmonic components: The photovoltaic inverter needs to output a compensation current i that is equal to and opposite to the target harmonic. comp (t), so that the total harmonic current approaches zero: The inverter generates a discrete time signal through the control system T s : sampling period of the control system; n: index of the sampling time, the control system calculates the reference current i according to the compensation current formula ref [n]:i ref [n] = i comp [n]; iref [n] is the ideal current that the inverter needs to output.

[0022] Preferably, pulse width modulation (PWM) technology is used to adjust the duty cycle of the inverter switch device so that the actual output current tracks the reference current i ref [n], PWM signal duty cycle calculation formula: D[n]: PWM duty cycle; I max : The maximum output current of the inverter, real-time adjustment of output through closed-loop control: real-time measurement of actual output current i out [n], calculated error with reference value: e[n] = i ref [n]-i out [n]; Use the proportional-integral controller to generate a correction signal to adjust the output of the inverter: K p is the proportional gain; K i is the integral gain; u[n] is the control signal, used to correct the PWM duty cycle;

[0023] Correct the PWM duty cycle according to the control signal: D[n]=D[n-1]+u[n]; the compensation current i output by the inverter total (t), the total current at the grid connection point is expressed as: i total (t) = i harm (t)+i out (t); i out (t) = i comp (t) = -i harm Substituting (t), the total current becomes: i total (t) = i harm (t)-i harm (t) = 0; harmonic cancellation is achieved.

[0024] In the above technical solution, the technical effects and advantages provided by the present invention are:

[0025] 1. The present invention uses Fourier transform to perform spectrum decomposition on the original multi-dimensional electrical signals (current, voltage, frequency) of the photovoltaic grid-connected point, constructs a multi-dimensional frequency signal matrix, and realizes the extraction of harmonic characteristics. Combining blind source separation and non-negative matrix decomposition technology, the frequency and amplitude characteristics of independent harmonic sources are accurately separated to eliminate the interference of superposition effect; a time series prediction model is constructed using a recursive neural network to simultaneously process the dynamic characteristics of multiple harmonic sources and predict future harmonic behavior; frequency aliasing errors are corrected through high-frequency compensation and wavelet transform multi-scale analysis technology to ensure the accuracy of separation and prediction results; finally, the prediction results are input into the photovoltaic inverter control system to dynamically generate compensation currents to offset the target harmonics in the power grid.

[0026] 2. The present invention can accurately separate and predict the characteristics of multiple harmonic sources in a complex power grid environment, dynamically adjust the output current of the photovoltaic inverter, and achieve efficient suppression of harmonics and improvement of the power quality of the power grid. Through multi-dimensional data processing and intelligent prediction technology, this method greatly reduces the impact of harmonic superposition and frequency aliasing on system stability, ensures the reliability, flexibility and robustness of the photovoltaic grid-connected system, and provides important technical support for achieving high-quality operation of green energy grid-connected. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0028] Figure 1 The figure is a flow chart of the method of the present invention. DETAILED DESCRIPTION

[0029] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0030] For examples, see Figure 1 As shown, the multi-dimensional sequential photovoltaic grid-connected harmonic current prediction and control method described in this embodiment includes the following steps:

[0031] S1: collecting original multi-dimensional electrical signals from photovoltaic grid-connected points, wherein the original multi-dimensional electrical signals include current, voltage and frequency, and decomposing the original multi-dimensional electrical signals into multi-dimensional frequency signals by Fourier transform;

[0032] S2: Using blind source separation algorithm and non-negative matrix decomposition technology, the characteristic components of independent harmonic sources are separated from the multi-dimensional frequency signal, and the frequency and amplitude characteristics of each independent harmonic source are extracted to reduce the prediction deviation caused by the superposition effect;

[0033] S3: Based on the separated independent harmonic source characteristics, a time series prediction model is constructed using a recurrent neural network. Multiple harmonic sources are processed simultaneously through a multidimensional prediction model to predict the future characteristics of the overall harmonic source.

[0034] S4: Combined with the separated harmonic source data, high-frequency compensation and reconstruction are performed on the frequency domain data, and the frequency aliasing error is further corrected through multi-scale analysis to ensure the accuracy of the frequency separation results;

[0035] S5: Input the multi-dimensional prediction results into the control system of the photovoltaic inverter, dynamically adjust the inverter output current, and offset the target harmonic components in the power grid.

[0036] In S1, the original multi-dimensional electrical signal of the photovoltaic grid-connected point is collected, and the original multi-dimensional electrical signal includes current, voltage and frequency. The original multi-dimensional electrical signal is decomposed into a multi-dimensional frequency signal by Fourier transform, specifically:

[0037] Install a current transformer (CT) or Hall current sensor at the grid connection point to detect the grid current waveform in real time, covering the fundamental frequency and main harmonic frequency bands (such as 50Hz and its integer multiples). Install a voltage transformer (VT) or voltage divider at the grid connection point to collect the grid voltage signal to ensure that harmonic fluctuations are captured. Use a dedicated frequency measurement module (such as a frequency meter based on a phase-locked loop) to detect grid frequency fluctuations in real time.

[0038] Convert analog electrical signals to digital signals, and set the sampling frequency to more than twice the maximum frequency component of the target signal (Nyquist sampling theorem), 10kHz or higher is recommended. Use high-precision clock or GPS synchronization technology to ensure the consistency of current, voltage, and frequency data on the time axis. Use a high-speed data bus (such as SPI, CAN, or Ethernet) to transmit digital signals to the central processing unit (CPU) or FPGA for real-time processing. Store raw data in real time for subsequent analysis and modeling.

[0039] The collected current and voltage signals are discrete data in the time domain: current signal i[n] = i(n·T s ),n=0,1,2,…,N-1;voltage signal v[n]=v(n·T s ),n=0,1,2,…,N-1; where, f s is the sampling frequency; the frequency domain conversion of the time domain signal x(t) (such as current i(t) or voltage v(t)) is performed, and the expression of discrete Fourier transform (DFT) is: k=0,1,2,…,N-1; where X[k] is the spectrum value of the signal at the kth frequency component, and the frequency is x[n] is a discrete signal in the time domain, is a complex exponential basis function, used for frequency analysis;

[0040] Applying discrete Fourier transform to the current signal i[n], the expression is: k=0,1,2,…,N-1; where I[k] is the spectrum value of the frequency domain current signal at frequency index k;

[0041] Applying Fourier transform to the voltage signal v[n], the expression is: k=0,1,2,…,N-1; where V[k] is the spectrum value of the frequency domain voltage signal;

[0042] Applying Fourier transform to the frequency signal f[n], the expression is: k=0,1,2,…,N-1; where F[k] is the spectrum value of the frequency signal in the frequency domain;

[0043] The frequency domain results I[k], V[k], F[k] of the current, voltage and frequency signals are combined to form a multidimensional frequency signal matrix S[k]: S[k] = [I[k], V[k], F[k]]; S[k] is the eigenvector of the multidimensional frequency signal at frequency index k.

[0044] S2: The blind source separation algorithm and non-negative matrix decomposition technology are used to separate the characteristic components of independent harmonic sources from the multi-dimensional frequency signal, and the frequency and amplitude characteristics of each independent harmonic source are extracted to reduce the prediction deviation caused by the superposition effect.

[0045] Assume that the multidimensional frequency signal matrix S[k] is composed of the frequency components H[k] of M independent harmonic sources through a linear combination of a mixing matrix A: S[k] = A·H[k]; where: S[k]∈R N×M : Observation signal (frequency signal); H[k]∈R M ×M : Frequency component of independent harmonic source; A∈R N×M : Mixing matrix, which represents the linear superposition weights of each harmonic source. The goal of blind source separation is to solve H[k] and A from the observed signal S[k], so that each harmonic source Hi[k] is statistically independent of each other; the mixing matrix A maintains sparsity and non-negativity.

[0046] Preprocess S[k] to make its mean 0 and covariance matrix the unit matrix, eliminating the correlation between data: W = whitening matrix; maximization of independence: using the method of maximizing non-Gaussianity (such as the FastICA algorithm), the separation matrix W is found through iterative optimization to obtain statistically independent signal components: Separation result verification: Verify whether the frequency, amplitude, etc. of each signal in H[k] conform to the harmonic characteristics. Through blind source separation, the frequency signal component H[k] of each independent harmonic source is obtained to achieve preliminary separation. In order to further improve the physical interpretability of the separation results, non-negative matrix factorization (NMF) is used to decompose H[k]: H[k] = W H ·H′; where: W H ∈R M×M : Weight matrix, representing the frequency weight of the harmonic source; H′∈R M×M : Basic component matrix, including the main frequency characteristics of each harmonic source; R is the number of basic frequency components, usually full R <M。

[0047] Since both frequency and amplitude are non-negative, NMF adds non-negativity constraints to improve the physical meaning of the decomposition results: Non-negativity constraints: W H ≥0,H′≥0; optimization objective function: in The Frobenius norm of the matrix is ​​represented. The frequency and amplitude characteristics of independent harmonic sources are extracted by NMF, making the separation results more sparse and easier to interpret. For each independent harmonic source H′ r ∈H′, calculate its main frequency component f r :f r = argmax k |H′ r [k]|; where H′ r [k] represents the spectrum value of harmonic source r at frequency index k.

[0048] For each independent harmonic source H′ r , its amplitude characteristic is: A r =||H r ′||1; Through blind source separation and non-negative matrix decomposition, the frequency and amplitude characteristic vectors of multiple independent harmonic sources are obtained: characteristic vector T r =[f r ,A r ],r=1,2,…,R; these characteristic vectors can be input into the time series prediction model to provide a high-precision data base for subsequent prediction and control.

[0049] S3: Based on the separated characteristics of independent harmonic sources, a time series prediction model is constructed using a recursive neural network. Multiple harmonic sources are processed simultaneously through a multidimensional prediction model to predict the future characteristics of the overall harmonic source.

[0050] The independent feature vector T extracted from step S2 r =[f r ,A r ],r=1,2,…,R(frequency fr and amplitude A r ) represents the characteristics of each harmonic source, forming time series data. These data contain the dynamic change information of the harmonic source in time, and it is necessary to use recurrent neural network (RNN) to model its time series relationship and predict the frequency and amplitude characteristics in the future. Recurrent neural network (RNN) is an ideal tool for time series data prediction. It retains historical state information through the cyclic structure of the hidden layer and is suitable for capturing the dynamic characteristics in the time series.

[0051] The time series characteristics of each independent harmonic source are expressed as: X r ={T r (t1),T r (t2),…,T r (t N )}; where T r (t i )=[f r (t i ),A r (t i )] represents the time t i The frequency and amplitude characteristics of the time series are represented by r=1,2,…,R, which represents the harmonic source number and N is the length of the time series. The input data is standardized and a prediction model is constructed using a recurrent neural network (RNN). The model input is a multidimensional characteristic time series and the output is the predicted value at a future moment.

[0052] Input layer: input data ET at each time step r (t i ), are the frequency and amplitude characteristics after standardization;

[0053] Hidden layer: uses a recursive structure, such as LSTM or GRU, to capture the long-term and short-term dependencies of the time series. The state update formula of the hidden layer is: t =φ(W h h t-1 +W x x t +b h );h t is the hidden layer state, x t is the input feature vector, W h ,W x ,b h is the hidden layer parameter, φ(·) is the activation function.

[0054] Output layer: predict future frequency and amplitude characteristics, the output is:

[0055] The time series {T r (t1),Tr (t2),…,T r (t N )} is input, and the RNN model predicts the future time t N+1 Features: For multiple harmonic sources r = 1, 2, ..., R, multiple sub-models are trained simultaneously, or the data of all harmonic sources are concatenated into a multi-dimensional input matrix: X = [X1, X2, ..., X R ] All harmonic sources are processed by sharing the RNN network, which improves computational efficiency and modeling consistency.

[0056] The loss function is used to evaluate the error between the predicted result and the true value, and the commonly used loss function is the mean square error (MSE): Where: T r (t i ) is the true value, is the predicted value. Use stochastic gradient descent (SGD) or Adam optimization algorithm to adjust the model parameter W h ,W x ,b h To minimize the loss function.

[0057] The time series of each harmonic source is segmented by sliding windows to form a training sample set. Training set and validation set division: 80% of the data is divided into training set and 20% into validation set to prevent overfitting.

[0058] For the current time series {T r (t1),T r (t2),…,T r (t N )} Input the model and predict the next time t N+1 Features: Based on the single-step prediction results, the predicted value As input to the next step, iteratively predict multiple time steps into the future: The predicted harmonic source characteristic matrix is: M represents the number of time steps for prediction; Contains the frequency and amplitude characteristics of the harmonic source r at future times.

[0059] S4: Combined with the separated harmonic source data, high-frequency compensation and reconstruction are performed on the frequency domain data, and the frequency aliasing error is further corrected through multi-scale analysis to ensure the accuracy of the frequency separation results.

[0060] The obtained harmonic source frequency data f r [k] and amplitude A r[k], draw a spectrum: observe the attenuation characteristics of harmonic amplitude with frequency, and identify the frequency band that may be affected by high-frequency loss. The amplitude of high-frequency harmonic components usually decays exponentially with frequency, and estimate the ignored high-frequency components: in: is the harmonic amplitude after high-frequency compensation; α is the attenuation factor, which is obtained by fitting the harmonic frequency characteristic data; f cutoff is the Nyquist frequency of the sampling frequency.

[0061] Compensation steps: For f k >f cutoff The frequency components are compensated and the high-frequency amplitude is recalculated using the formula; the compensated frequency components are superimposed to form a complete high-frequency compensated harmonic spectrum.

[0062] Using the compensated spectrum and phase information φ r [k], reconstruct the high-frequency components into the time domain through inverse Fourier transform: is the harmonic source signal after high frequency compensation.

[0063] Using the harmonic separation results, we can find the frequency bands where aliasing may exist (such as f k Close to f cutoff The main features of aliasing detection are: sudden increase of harmonic components or abnormal frequency drift; inter-frequency harmonic relationships that do not conform to physical laws (such as non-integer frequency multiples).

[0064] Wavelet transform is used for multi-scale decomposition to separate the low-frequency and high-frequency components of the signal: Continuous wavelet transform: The harmonic source signal h r (t) Convert to the time-scale domain: Where: W(a,b) is the wavelet coefficient, which represents the signal component corresponding to scale a and time b; ψ * is the mother wavelet function (such as Morlet wavelet or Haar wavelet); a is the scale factor, which controls the frequency resolution; b is the translation factor, which controls the time resolution. The signal is divided into different frequency bands through multi-scale decomposition to identify and correct possible aliasing frequencies: retain the components of low-scale (high-frequency) signals; use high-scale (low-frequency) components to correct the aliasing parts. The low-frequency and intermediate-frequency signals after multi-scale correction are combined to obtain the complete harmonic source signal: By comparing the spectrum of the reconstructed signal with the separation result, it is verified whether the corrected signal conforms to the actual harmonic characteristics.

[0065] S5: Input the multi-dimensional prediction results into the control system of the photovoltaic inverter, dynamically adjust the inverter output current, and offset the target harmonic components in the power grid.

[0066] For the rth independent harmonic source, the harmonic component at the future time t is expressed as: r (t) = A r sin(2πf r t+φ r );A r is the predicted harmonic amplitude, f r is the predicted harmonic frequency, φ r is the predicted harmonic phase. The target harmonic signal in the power grid is the superposition of these independent harmonic components: The photovoltaic inverter needs to output a compensation current i that is equal to and opposite to the target harmonic. comp (t), so that the total harmonic current approaches zero: The inverter generates a discrete time signal through the control system T s : The sampling period of the control system; n: The index of the sampling time. The control system calculates the reference current i according to the compensation current formula ref [n]:i ref [n] = i comp [n]; i ref [n] is the ideal current that the inverter needs to output.

[0067] Pulse width modulation (PWM) technology is used to adjust the duty cycle of the inverter switching device so that the actual output current tracks the reference current i ref [n], PWM signal duty cycle calculation formula: D[n]: PWM duty cycle; I max : The maximum output current of the inverter. The actual inverter output current may deviate from the reference value due to system dynamics or external interference. The output is adjusted in real time through closed-loop control: Real-time measurement of the actual output current i out [n], calculated error with reference value: e[n] = i ref [n]-i out [n]; Use the proportional-integral (PI) controller to generate a correction signal to adjust the inverter output: K p is the proportional gain; K i is the integral gain; u[n] is the control signal, which is used to correct the PWM duty cycle.

[0068] Correct the PWM duty cycle according to the control signal: D[n]=D[n-1]+u[n]; the compensation current i output by the inverter total (t), the total current at the grid connection point is expressed as: i total (t) = iharm (t)+i out (t); i out (t) = i comp (t) = -i harm Substituting (t), the total current becomes: total (t) = i harm (t)-i harm (t) = 0; harmonic cancellation is achieved.

[0069] In this embodiment, first, the original multi-dimensional electrical signal (including current, voltage, and frequency) is collected at the photovoltaic grid-connected point, and Fourier transform is used to decompose it into a multi-dimensional frequency signal; then, the blind source separation algorithm and non-negative matrix decomposition technology are used to separate the characteristic components of the independent harmonic source from the frequency signal, and extract its frequency and amplitude characteristics to reduce the prediction deviation caused by the superposition effect. Subsequently, based on the separated harmonic source characteristics, a multi-dimensional time series prediction model is constructed using a recursive neural network to accurately predict the future behavior of each harmonic source. On this basis, combined with the data of the separated harmonic source, high-frequency compensation and reconstruction are performed on the frequency domain, and multi-scale analysis is used to correct the frequency aliasing error to ensure the accuracy of the frequency separation result. Finally, the prediction results are input into the photovoltaic inverter control system, the output current is dynamically adjusted, and a reverse harmonic compensation signal is generated to achieve effective offset of the target harmonics of the power grid, thereby improving the power quality and the stability of the grid-connected system.

[0070] The above formulas are all dimensionless and numerical calculations. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters in the formula are set by technicians in this field according to actual conditions.

[0071] The above embodiments can be implemented in whole or in part by software, hardware, firmware or any other combination. When implemented by software, the above embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the process or function described in the embodiment of the present application is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from one website site, computer, server or data center to another website site, computer, server or data center by wired or wireless (e.g., infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that contains one or more available media sets. The available medium can be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium. The semiconductor medium can be a solid-state hard disk.

[0072] It should be understood that the term "and / or" in this article is only a description of the association relationship of associated objects, indicating that there may be three relationships. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone, where A and B can be singular or plural. In addition, the character " / " in this article generally indicates that the objects associated with each other are in an "or" relationship, but it may also indicate an "and / or" relationship, which can be understood by referring to the context. A person of ordinary skill in the art can appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed in this article can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of this application.

[0073] The above description is only a specific implementation manner of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application.

Claims

1. A multi-dimensional sequential photovoltaic grid-connected harmonic current prediction and control method, characterized in that: The steps include: S1: collecting original multi-dimensional electrical signals from photovoltaic grid-connected points, wherein the original multi-dimensional electrical signals include current, voltage and frequency, and decomposing the original multi-dimensional electrical signals into multi-dimensional frequency signals by Fourier transform; S2: Using blind source separation algorithm and non-negative matrix decomposition technology, the characteristic components of independent harmonic sources are separated from the multi-dimensional frequency signal, and the frequency and amplitude characteristics of each independent harmonic source are extracted to reduce the prediction deviation caused by the superposition effect; S3: Based on the separated independent harmonic source characteristics, a time series prediction model is constructed using a recurrent neural network. Multiple harmonic sources are processed simultaneously through a multidimensional prediction model to predict the future characteristics of the overall harmonic source. S4: Combined with the separated harmonic source data, high-frequency compensation and reconstruction are performed on the frequency domain data, and the frequency aliasing error is further corrected through multi-scale analysis to ensure the accuracy of the frequency separation results; S5: Input the multi-dimensional prediction results into the control system of the photovoltaic inverter, dynamically adjust the inverter output current, and offset the target harmonic components in the power grid.

2. The multi-dimensional sequential photovoltaic grid-connected harmonic current prediction and control method according to claim 1, characterized in that: In S1, the frequency analysis of the multi-dimensional time-domain electrical signals of the photovoltaic grid-connected point is performed through Fourier transform, and the discrete current signal, voltage signal and frequency signal data in the time domain are collected. Discrete Fourier transform is applied to each signal to convert it from the time domain to the frequency domain, and the corresponding spectrum information, including frequency components and amplitude characteristics, is extracted. The spectrum data of the current, voltage and frequency signals are combined to form a multi-dimensional frequency matrix.

3. The multi-dimensional sequential photovoltaic grid-connected harmonic current prediction and control method according to claim 2, characterized in that: In S2, the multidimensional frequency signal matrix is ​​assumed to be composed of the linear superposition of frequency components of multiple independent harmonic sources, and the signal matrix is ​​processed by blind source separation to whiten and eliminate the correlation between signals, so that the components of each harmonic source are independent of each other; In the process of blind source separation, the separation matrix is ​​solved by the optimization algorithm, and it is verified whether the separation result conforms to the harmonic characteristics, so as to obtain the frequency components of each independent harmonic source. The separation result is decomposed by non-negative matrix decomposition, and the non-negative constraint is introduced to ensure that the separated frequency and amplitude are positive. The weight matrix and basic component matrix of each independent harmonic source are obtained by NMF, and the main frequency characteristics and amplitude characteristics of the harmonic source are extracted; the main frequency and amplitude of each independent harmonic source are calculated to form the frequency and amplitude characteristic vector of the independent harmonic source.

4. The multi-dimensional sequential photovoltaic grid-connected harmonic current prediction and control method according to claim 3, characterized in that: In S3, extract the independent feature vector T r =[f r ,A r ],r=1,2,…,R; the time series characteristics of each independent harmonic source are expressed as: X r ={T r (t1),T r (t2),…,T r (t N )}; where T r (t i )=[f r (t i ),A r (t i )] represents the time t i The frequency and amplitude characteristics of the time are calculated, r = 1, 2, ..., R, which indicates the harmonic source number, N is the length of the time series, the input data is standardized, and a prediction model is constructed using a recursive neural network; the model input is a multidimensional characteristic time series, and the output is the predicted value at the future moment; Input data ET for each time step r (t i ), which is the frequency and amplitude characteristics after standardization; a recursive structure is used to capture the long-term and short-term dependencies of the time series; the state update formula of the hidden layer is: h t =φ(W h h t-1 +W x x t +b h );h t is the hidden layer state, x t is the input feature vector, W h ,W x ,b h is the hidden layer parameter, φ(·) is the activation function; the output layer is used to predict the future frequency and amplitude characteristics, and the output is: The time series {T r (t1),T r (t2),…,T r (t N )} is input, and the RNN model predicts the future time t N+1 Features: For multiple harmonic sources r = 1, 2, ..., R, multiple sub-models are trained simultaneously, and the loss function is used to evaluate the error between the predicted result and the true value. The expression is: Where: T r (t i ) is the true value, is the predicted value; the time series of each harmonic source is segmented by sliding window to form a training sample set; the training set and validation set are divided: 80% of the data is divided into a training set and 20% into a validation set.

5. The multi-dimensional sequential photovoltaic grid-connected harmonic current prediction and control method according to claim 4, characterized in that: In S3, for the current time series {T r (t1),T r (t2),…,T r (t N )} Input the model and predict the next time t N+1 Features: Based on the single-step prediction results, the predicted value As input to the next step, iteratively predict multiple time steps into the future: The predicted harmonic source characteristic matrix is: M represents the number of time steps for prediction; Contains the frequency and amplitude characteristics of the harmonic source r at future times.

6. The multi-dimensional sequential photovoltaic grid-connected harmonic current prediction and control method according to claim 1, characterized in that: In S4, based on the obtained harmonic source frequency and amplitude data, a spectrum diagram is drawn, and the high-frequency component is compensated by an exponential model using the characteristic that the amplitude decays with the frequency, generating the compensated high-frequency harmonic amplitude, and the high-frequency harmonic signal is reconstructed by inverse Fourier transform in combination with the phase information; wavelet transform is used for multi-scale decomposition to divide the signal into low-frequency and high-frequency components. The abnormal frequency components can be identified and corrected through multi-scale decomposition, retaining the low-frequency information and high-frequency components, while removing the aliasing part; The corrected high-frequency and low-frequency harmonic signals are combined to form a complete harmonic signal.

7. The multi-dimensional sequential photovoltaic grid-connected harmonic current prediction and control method according to claim 1, characterized in that: In S5, the multi-dimensional prediction results are input into the control system of the photovoltaic inverter to dynamically adjust the inverter output current to offset the target harmonic components in the power grid, specifically: For the rth independent harmonic source, the harmonic component at the future time t is expressed as: r (t) = A r sin(2πf r t+φ r );A r is the predicted harmonic amplitude, f r is the predicted harmonic frequency, φ r is the predicted harmonic phase; the target harmonic signal in the power grid is the superposition of independent harmonic components: The photovoltaic inverter needs to output a compensation current i that is equal to and opposite to the target harmonic. comp (t), so that the total harmonic current approaches zero: The inverter generates a discrete time signal i through the control system comp [n]: T s : The sampling period of the control system; n: index of the sampling time, the control system calculates the reference current i according to the compensation current formula ref [n]:i ref [n] = i comp [n]; i ref [n] is the ideal current that the inverter needs to output.

8. The multi-dimensional sequential photovoltaic grid-connected harmonic current prediction and control method according to claim 7, characterized in that: Pulse width modulation (PWM) technology is used to adjust the duty cycle of the inverter switching device so that the actual output current tracks the reference current i ref [n], PWM signal duty cycle calculation formula: D[n]: PWM duty cycle; I max : The maximum output current of the inverter, real-time adjustment of output through closed-loop control: real-time measurement of actual output current i out [n], calculated error with reference value: e[n] = i ref [n]-i out [n]; Use the proportional-integral controller to generate a correction signal to adjust the output of the inverter: K p is the proportional gain; K i is the integral gain; u[n] is the control signal, used to correct the PWM duty cycle; Correct the PWM duty cycle according to the control signal: D[n]=D[n-1]+u[n]; the compensation current i output by the inverter total (t), the total current at the grid connection point is expressed as: i total (t) = i harm (t)+i out (t); i out (t) = i comp (t) = -i harm Substituting (t), the total current becomes: i total (t) = i harm (t)-i harm (t) = 0; harmonic cancellation is achieved.

Images