A method for identifying power quality disturbance characteristics of a distributed photovoltaic interference power supply

Abstract

The application relates to the technical field of power quality monitoring and discloses a method for identifying the disturbance characteristics of the power quality of a distributed photovoltaic interference power supply, which comprises the following steps: S1, data acquisition and preprocessing, acquiring the power quality signals of a photovoltaic system and carrying out denoising, normalization and time alignment; S2, adaptive decomposition of the disturbance signals based on improved variational mode decomposition, and automatically selecting the mode number by using information geometry optimization. The improved variational mode decomposition method is combined with information geometry optimization and Bayesian optimization to realize adaptive mode decomposition of the disturbance signals. The improved variational mode decomposition can automatically determine the optimal mode number and the penalty factor, avoids the decomposition error caused by manually setting parameters in traditional variational mode decomposition, and has higher analytical capacity for complex disturbance signals compared with the fixed parameter variational mode decomposition method in the prior art, so that the disturbance modes in the photovoltaic system can be accurately decomposed.

Description

Technical Field

[0001] This invention relates to the field of power quality monitoring technology, specifically a method for identifying power quality disturbance characteristics of distributed photovoltaic interference power sources. Background Technology

[0002] With the widespread application of distributed photovoltaic (PV) systems, power quality issues are becoming increasingly prominent. Especially during grid connection, PV power sources generate various disturbances, such as voltage fluctuations, frequency variations, and harmonic distortion, which can affect power quality and lead to system instability. Therefore, accurately identifying and providing early warning of disturbance characteristics is crucial for ensuring grid security and improving the operational reliability of the power system.

[0003] During operation, the current system often experiences disturbances such as low-frequency harmonics, short-term fluctuations, and instantaneous drops. Existing technologies mainly rely on fixed parameter decomposition and traditional statistical models, which cannot take into account the complex signal characteristics and the influence of meteorological factors, thus limiting the accuracy and reliability of monitoring and early warning.

[0004] Most current solutions use fixed-parameter variational mode decomposition, where parameters need to be set manually and cannot be dynamically adjusted. This makes it difficult to cope with the diversity of disturbance signals and results in unstable decomposition performance.

[0005] Existing technologies rely heavily on traditional Fourier transform or empirical mode decomposition when modeling perturbation signals, which fail to capture the nonlinear chaotic characteristics of the signals, lack physical connotations, and cannot accurately identify short-time composite perturbations.

[0006] Common forecasting methods rely solely on statistical regression or single neural network models, neglecting the impact of environmental and meteorological data on disturbances. This limits the accuracy of both short-term and long-term forecasts and results in delayed early warning responses.

[0007] Therefore, those skilled in the art provide a method for identifying the power quality disturbance characteristics of distributed photovoltaic interference power sources to solve the problems mentioned above. Summary of the Invention

[0008] To address the shortcomings of existing technologies, this invention provides a method for identifying power quality disturbance characteristics of distributed photovoltaic interference power sources, thereby solving the problems mentioned in the background section.

[0009] To achieve the above objectives, the present invention provides the following technical solution: a method for identifying power quality disturbance characteristics of distributed photovoltaic interference power sources, comprising:

[0010] Step S1, Data Acquisition and Preprocessing: Acquire the power quality signal of the photovoltaic system and perform noise reduction, normalization and time alignment; Step S2: Based on the improved variational mode decomposition, the perturbation signal is adaptively decomposed, the number of modes is automatically selected using information geometry optimization, and the penalty factor is selected through Bayesian optimization. Step S3: Model the disturbance characteristics based on the Hamiltonian dynamics model, construct the Hamiltonian system, and calculate the phase space dynamic equation of the disturbance signal. Step S4: Calculate the Lyapunov exponent for the Hamiltonian dynamic equation established in step S3, analyze the nonlinear characteristics of the disturbance signal based on the Lyapunov exponent, calculate the maximum Lyapunov exponent of the disturbance signal and determine its chaotic characteristics. Step S5: Combining the disturbance prediction with meteorological data, the ARIMA model is used for short-term disturbance prediction, and the Transformer-LSTM hybrid model is used for long-term disturbance prediction. Step S6: Output and application of results. Output the disturbance identification results and provide real-time early warning.

[0011] Preferably, in step S1, data acquisition and preprocessing further includes: Step 1.1, Acquire power quality signals: When collecting power quality signals, data acquisition equipment is used to record the three-phase voltage output by the photovoltaic system. and three-phase current And simultaneously record the power quality parameters on the grid side; Let the sampling frequency be If the sampling duration is T, then the discrete form of the acquired signal can be expressed as: , in, Let be the discrete value of the voltage of the nth phase at the kth sampling point. Let n be the discrete value of the nth phase current at the kth sampling point. =1 / The sampling period is = This represents the total number of sampling points; Step 1.2, Signal Denoising: Wavelet transform is used to decompose the signal at multiple scales to remove high-frequency noise components; Let the number of wavelet decomposition levels be J, and the mother wavelet function be... The wavelet transform coefficients of the signal are then expressed as: , in, For signal Wavelet transform coefficients at scale a and location b For the mother wavelet function, Let be the complex conjugate of the mother wavelet function, a be the scaling factor, and b be the translation factor; The noise reduction process includes the following steps: Calculate the wavelet transform coefficients of the signal ; Set threshold Soft threshold filtering is applied to the high-frequency wavelet coefficients: , in, The threshold for the j-th level wavelet decomposition is... These are the update coefficients after soft thresholding. Wavelet reconstruction is performed to obtain the denoised signal. ; Step 1.3, Signal Normalization: The signal is mapped to the interval [1, 2] using the min-max normalization method: , , in, The original voltage signal, This is the original current signal.

[0012] This is the minimum value of the voltage signal. This represents the maximum value of the voltage signal. This represents the minimum value of the current signal. This represents the maximum value of the current signal. The normalized voltage signal. This is the normalized current signal; Step 1.4, Signal Time Alignment: The time delay between signals is calculated using the cross-correlation function. The formula for the cross-correlation function is as follows: , in, voltage signal and current signal The cross-correlation function values ​​between them Let k be the discrete value of the a-th phase voltage signal. Let k be the discrete value of the a-th phase current signal. For time delay; Find the value corresponding to the maximum cross-correlation coefficient. Furthermore, the signals are adjusted accordingly to ensure that all signals are aligned to the same time base.

[0013] Preferably, in step S2, the adaptive decomposition of the perturbation signal based on the improved variational mode decomposition further includes: Step 2.1, Mathematical modeling of variational mode decomposition: Let the preprocessed disturbance signal be represented as : , in, For the k-th modal component, This is the preprocessed perturbation signal. The target number of modes; Modal components At a specific center frequency The vicinity has limited bandwidth; Variational mode decomposition achieves signal decomposition by solving the following variational optimization problem: , in, For modal components The signal after Hilbert transform, where j is the imaginary unit. The center frequency of the modal component, For time derivative operators, For the i-th modal component; Using the Lagrange multiplier method to introduce constraints, construct the Lagrange function: , in, Let Lagrangian function be the variational mode decomposition function. As a penalty factor, For Lagrange multipliers, This is an inner product operation; Step 2.2, Information geometry optimization automatically selects the number of modes: Based on the information geometry optimization method, the spectral entropy of each modal component is maximized. Calculate the optimal number of modes: , in, Modal components In frequency Normalized energy distribution at: , Let the total information entropy As the objective function: , Iterate through the different modal numbers K to find the one that makes Maximum number of optimal modes : , in, Let be the spectral entropy of the k-th modal component. Modal components In frequency Normalized energy distribution at the location, Modal components In frequency Fourier transform at the point, For frequency points, For information entropy, The optimal number of modes; Step 2.3, Bayesian optimization to determine the penalty factor: A Bayesian optimization method is used to determine the penalty factor based on the reconstruction error of the decomposed signal. As the objective function: , Using Gaussian process regression model to predict different under the value And select the optimal penalty factor based on the expected improvement criterion: , in, For signal reconstruction error, As a penalty factor, This is the reconstructed signal of all modal components after variational mode decomposition. To calculate the penalty factor that minimizes the reconstruction error ; Step 2.4, calculate the decomposed disturbance signal: Based on certainty and The preprocessed disturbance signal is then subjected to final variational mode decomposition to obtain... Group modal components: , Among them, each These represent different frequency band components in the photovoltaic disturbance signal; Thus, the adaptive decomposition of the perturbation signal based on the improved variational mode decomposition is completed, and the decomposed perturbation signal is obtained. , This will be used in step S3 to construct the Hamiltonian dynamics model.

[0014] Preferably, in step S3, the perturbation feature modeling based on the Hamiltonian dynamics model further includes: After the adaptive decomposition of the disturbance signal is completed in step S2, the following is obtained: Group modal components The modal components represent disturbance signals in different frequency bands in the photovoltaic system and will be used to construct the Hamiltonian dynamics model in step S3. Step 3.1, Hamiltonian dynamics modeling: The phase space of the disturbance signal is set as ,in, and Represents modal components Position and momentum; , in, Modal components quality It is the potential energy function; For the disturbance signal, momentum With position Satisfy the following Hamiltonian equations: , , in, = and = Indicates speed and force; Step 3.2, construct the phase space model of the perturbation signal: By analyzing each modal component Numerical integration of Hamilton's equations yields the trajectory in phase space. Assuming each modal component Potential energy function For simple second potential energy: , in, Modal components The elastic constant; By solving the Hamiltonian equations, the dynamic equations in phase space are obtained: , , in, Represents modal components Position and momentum in phase space; Step 3.3, calculate the phase space dynamic equation of the disturbance signal: The classic fourth-order Runge-Kutta algorithm is used to numerically integrate the Hamiltonian equation, with a set time step. Through iterative updates and Value: , , Step 3.4: Perform Hamiltonian dynamics modeling analysis based on the disturbance signal: After step S3 is completed, the dynamic characteristics of the disturbance signal have been modeled using the Hamiltonian dynamics model. Then, step S4 is performed to further analyze the nonlinear characteristics of the disturbance signal using the Lyapunov exponent.

[0015] Preferably, in step S4, the analysis of the nonlinear characteristics of the disturbance signal based on the Lyapunov exponent further includes: In step S3, the Hamiltonian dynamic model of the perturbation signal has been constructed, and the phase space dynamic equation of the perturbation signal has been obtained. In order to further analyze the nonlinear characteristics and chaotic features of the perturbation signal, in step S4, the Lyapunov exponent needs to be calculated, and the chaotic characteristics of the perturbation signal are judged based on the maximum Lyapunov exponent. Step 4.1, Mathematical definition of the Lyapunov exponent: In phase space, suppose the phase trajectory of the disturbance signal is described by Hamiltonian dynamics equations, and the state vector is: , in, and Modal components phase space coordinates, The optimal number of modal components; The time evolution of the perturbation signal in phase space is described by the following Hamiltonian dynamic equation: , in, The system's evolution equation; At the initial moment Select adjacent trajectory points in phase space and The initial small perturbation is: , The evolution after time t satisfies: , in, The Lyapunov exponent describes the exponential growth rate of perturbations between adjacent trajectories. The rate of change of distance between trajectories is given by the following formula: , Step 4.2, calculate the maximum Lyapunov exponent: Calculating the maximum Lyapunov exponent requires numerical simulation of the phase space trajectory. A small perturbation method is used to track the divergence of adjacent trajectories. The specific calculation steps are as follows: Step 4.2.1, in phase space trajectory In the middle, select the initial point And select an initial small perturbation within the neighborhood. To satisfy: , in, The initial disturbance amplitude; Step 4.2.2: Solve the Hamiltonian dynamic equations using numerical integration to calculate the trajectory after the disturbance. Compared with the original trajectory Distance between: , in, It reflects the trend of the disturbance trajectory over time.

[0016] Step 4.2.3: Plot the graph in logarithmic coordinates. The curve varies with time t, and the slope is solved using linear regression. : , in, For time steps; Step 4.3, determine the chaotic characteristics based on the maximum Lyapunov exponent: According to the maximum Lyapunov index The magnitude of the value is used to determine the chaotic characteristics of the disturbance signal: like If the value is greater than 0, the disturbance signal exhibits chaotic characteristics. like =0, then the system belongs to quasi-periodic motion; like If the value is less than 0, then the system has an attractor. Step 4.4, calculate multiple sets of Lyapunov indices: Except for the maximum Lyapunov index In addition, calculate other Lyapunov indices. To further analyze the stability of the perturbation signal, the complete Lyapunov exponent spectrum was obtained. It consists of multiple sets of indices: , in, The optimal number of modes; Using the QR decomposition method, the Lyapunov exponent spectrum is calculated. First, the Jacobian matrix of the perturbation equation is constructed. : Then, by tracking the exponential divergence rate of the orthogonal basis through QR decomposition, all Lyapunov exponents were calculated. Step 4.5, Result Output and Next Prediction: After completing the Lyapunov exponent calculation, the nonlinear characteristics of the disturbance signal are obtained: like If the value is greater than 0, it indicates that the disturbance signal has chaotic characteristics, and a prediction model adapted to chaotic characteristics is adopted.

[0017] like If ≤0, the disturbance signal exhibits periodic and convergent behavior, and a traditional time series prediction model is used; At this point, step S4 is complete, and the Lyapunov exponent of the disturbance signal has been successfully extracted and its nonlinear characteristics have been analyzed. Next, we proceed to step S5, which involves making short-term and long-term predictions of the disturbance signal based on meteorological data.

[0018] Preferably, in step S5, the disturbance prediction based on meteorological data further includes: In step S4, the nonlinear characteristics of the disturbance signal are analyzed and chaotic behavior is judged by calculating the Lyapunov exponent. In step S5, the disturbance signal is predicted in the short and long term by combining the meteorological data of the photovoltaic system. Step 5.1, Short-term disturbance prediction based on the ARIMA model: Assume the time series of the disturbance signal is as follows: ,in, Let N be the disturbance signal value with time step N, where N is the total number of observation time points; Step 5.1.1, Establishing the ARIMA model: The ARIMA model consists of parameters. Decide: p is the autoregressive order, representing the influence of the value at time p on the current value; d is the difference order, which represents the number of differences required to stabilize the time series; q is the moving average order, representing the influence of the first q error terms on the current value; The mathematical expression for the ARIMA model is: , in, For the shift operator, For the autoregressive part, is a polynomial. For the polynomial of the moving average part, This is the white noise term; Step 5.1.2: Optimize and select the best option using AIC (Akaike Information Criterion) and BIC (Bayesian Information Criterion). : , , Where L is the log-likelihood function of the model. N represents the number of model parameters, and N represents the total number of observation time points. Optimal parameters Determined by the following optimization objectives: , Step 5.1.3, Predict the disturbance signal: Based on the optimal ARIMA model, predict the short-term disturbance signal: , Where h is the prediction step size, For the predicted future time steps The disturbance signal value, These are the autoregressive coefficients. This is the moving average coefficient.

[0019] Preferably, in step S5, the disturbance prediction based on meteorological data further includes: Step 5.2, Long-term perturbation prediction is based on a Transformer-LSTM hybrid model: Due to the chaotic nature of the perturbation signal, a deep learning model that can capture long-term dependencies is adopted, namely the Transformer-LSTM hybrid model. Step 5.2.1, LSTM network modeling. The LSTM structure consists of an input gate, a forget gate, and an output gate. The calculation formula is as follows: Forgotten Gate:

[0020] in, For the current input, This is the hidden state from the previous moment. and For the weights and biases of the forget gate, For the Sigmoid function; Input Gate: , , in, The output of the input gate, Here is the weight matrix of the input gate. Let be the bias vector of the input gate. Candidate cell state, This is the weight matrix for the candidate cell states. is the bias vector for the candidate cell state; Cell status update: , in, The current cell state, This represents the cell state at the previous moment; Output gate: , , in, For the output of the output gate, This is the weight matrix of the output gate. This is the bias vector for the output gate. This is the final output of the LSTM; Step 5.2.2, Transformer processes global features: The Transformer uses a self-attention mechanism to extract global features. Given an input perturbation signal sequence S, define the query Q, key K, and value V: , in, , , It is a linear transformation matrix; Calculate attention weights:

[0021] in, The dimension of the key vector; Step 5.2.3, Predict the perturbation signal: Extract long-term dependency features using Transformer, and predict using LSTM: , in, for Extracted features This is the LSTM prediction function.

[0022] Preferably, in step S5, the disturbance prediction based on meteorological data further includes: Step 5.3: Perform disturbance correction based on meteorological data: Disturbance signals from photovoltaic systems are affected by meteorological factors, including: Solar radiation intensity ,temperature Wind speed and cloud cover ; Constructing a perturbation correction model: , in, , , , As the weight of meteorological factors, These are the predicted values ​​after correction from meteorological data. This is the original predicted value; Step 5.4, Results Output and Application: Short-term forecasts are used for real-time monitoring of photovoltaic systems; Long-term forecasting is used for power grid dispatching; The prediction results will be used for real-time early warning in step S6 to prevent photovoltaic disturbances from affecting the power grid; at this point, step S5 is completed, and the short-term and long-term evolution trends of the disturbance signal have been successfully predicted.

[0023] A terminal device includes a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the computer program is configured to perform a method for identifying power quality disturbance characteristics of a distributed photovoltaic interference power source.

[0024] A storage medium storing a computer program, which, when executed by a processor, implements a method for identifying power quality disturbance characteristics of distributed photovoltaic interference power sources.

[0025] This invention provides a method for identifying power quality disturbance characteristics of distributed photovoltaic (PV) interference power sources. It has the following beneficial effects: 1. This invention employs an improved variational mode decomposition method, combining information geometry optimization and Bayesian optimization, to achieve adaptive mode decomposition of perturbation signals. The improved variational mode decomposition can automatically determine the optimal number of modes and penalty factor, avoiding the decomposition error caused by manually setting parameters in traditional variational mode decomposition. Compared with the fixed parameter variational mode decomposition method in the prior art, this invention improves the analytical capability of complex perturbation signals, enabling accurate decomposition of perturbation modes in photovoltaic systems.

[0026] 2. This invention employs Hamiltonian dynamics modeling to model and analyze photovoltaic disturbance signals from the perspective of energy conservation. Furthermore, it combines Lyapunov exponent calculation to extract the chaotic characteristics of the signal, enabling accurate identification of short-term composite disturbances and effective differentiation between disturbances and systemic disturbances. Compared with Fourier analysis or empirical mode decomposition in existing technologies, the dynamics modeling method of this invention can intuitively reveal the physical characteristics of disturbance signals, improving the interpretability and reliability of power quality monitoring.

[0027] 3. This invention combines meteorological data from photovoltaic systems with a hybrid prediction model to achieve short-term and long-term predictions of disturbance signals. The ARIMA model is suitable for short-term trend prediction, while the Transformer-LSTM can capture long-term evolution patterns. The combination of the two can effectively improve the accuracy of disturbance warnings. Compared with existing prediction methods based on statistical regression or single LSTM, this invention can accurately predict power quality disturbances caused by light fluctuations and meteorological changes, enabling efficient intelligent scheduling of photovoltaic systems. Attached Figure Description

[0028] Figure 1 This is a flowchart of the present invention. Detailed Implementation

[0029] To enable those skilled in the art to understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. Other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort should fall within the scope of protection of the present invention.

[0030] The present invention will now be described in detail with reference to the accompanying drawings: Example: Please see the appendix Figure 1 This invention provides a method for identifying power quality disturbance characteristics of distributed photovoltaic interference power sources, comprising: Step S1, Data Acquisition and Preprocessing: Acquire the power quality signal of the photovoltaic system and perform noise reduction, normalization and time alignment; Step 1.1, Acquire power quality signals: When collecting power quality signals, data acquisition equipment is used to record the three-phase voltage output by the photovoltaic system. and three-phase current And simultaneously record the power quality parameters on the grid side; Let the sampling frequency be If the sampling duration is T, then the discrete form of the acquired signal can be expressed as: , in, Let be the discrete value of the voltage of the nth phase at the kth sampling point. Let n be the discrete value of the nth phase current at the kth sampling point. =1 / The sampling period is = This represents the total number of sampling points; Step 1.2, Signal Denoising: Wavelet transform is used to decompose the signal at multiple scales to remove high-frequency noise components; Let the number of wavelet decomposition levels be J, and the mother wavelet function be... The wavelet transform coefficients of the signal are then expressed as: , in, For signal Wavelet transform coefficients at scale a and location b For the mother wavelet function, Let be the complex conjugate of the mother wavelet function, a be the scaling factor, and b be the translation factor; The noise reduction process includes the following steps: Calculate the wavelet transform coefficients of the signal ; Set threshold Soft threshold filtering is applied to the high-frequency wavelet coefficients: , in, The threshold for the j-th level wavelet decomposition is... These are the update coefficients after soft thresholding. Wavelet reconstruction is performed to obtain the denoised signal. ; Step 1.3, Signal Normalization: The signal is mapped to the interval [1, 2] using the min-max normalization method: , , in, The original voltage signal, The original current signal, This is the minimum value of the voltage signal. This represents the maximum value of the voltage signal. This represents the minimum value of the current signal. This represents the maximum value of the current signal. The normalized voltage signal. This is the normalized current signal; Step 1.4, Signal Time Alignment: The time delay between signals is calculated using the cross-correlation function. The formula for the cross-correlation function is as follows: , in, voltage signal and current signal The cross-correlation function values ​​between them Let k be the discrete value of the a-th phase voltage signal. Let k be the discrete value of the a-th phase current signal. For time delay; Find the value corresponding to the maximum cross-correlation coefficient. Furthermore, the signals are adjusted accordingly to ensure that all signals are aligned to the same time base. Step S2: Based on the improved variational mode decomposition, the perturbation signal is adaptively decomposed, the number of modes is automatically selected using information geometry optimization, and the penalty factor is selected through Bayesian optimization. Step 2.1, Mathematical modeling of variational mode decomposition: Let the preprocessed disturbance signal be represented as : , in, For the k-th modal component, This is the preprocessed perturbation signal. The target number of modes; Modal components At a specific center frequency The vicinity has limited bandwidth; Variational mode decomposition achieves signal decomposition by solving the following variational optimization problem: , in, For modal components The signal after Hilbert transform, where j is the imaginary unit. The center frequency of the modal component, For time derivative operators, For the i-th modal component; Using the Lagrange multiplier method to introduce constraints, construct the Lagrange function: , in, Let Lagrangian function be the variational mode decomposition function. As a penalty factor, For Lagrange multipliers, This is an inner product operation; Step 2.2, Information geometry optimization automatically selects the number of modes: Based on the information geometry optimization method, the spectral entropy of each modal component is maximized. Calculate the optimal number of modes: , in, Modal components In frequency Normalized energy distribution at: , Let the total information entropy As the objective function: , Iterate through the different modal numbers K to find the one that makes Maximum number of optimal modes : , in, Let be the spectral entropy of the k-th modal component. Modal components In frequency Normalized energy distribution at the location, Modal components In frequency Fourier transform at the point, For frequency points, For information entropy, The optimal number of modes; Step 2.3, Bayesian optimization to determine the penalty factor: A Bayesian optimization method is used to determine the penalty factor based on the reconstruction error of the decomposed signal. As the objective function: , Using Gaussian process regression model to predict different under the value And select the optimal penalty factor based on the expected improvement criterion: , in, For signal reconstruction error, As a penalty factor, This is the reconstructed signal of all modal components after variational mode decomposition. To calculate the penalty factor that minimizes the reconstruction error ; Step 2.4, calculate the decomposed disturbance signal: Based on certainty and The preprocessed disturbance signal is then subjected to final variational mode decomposition to obtain... Group modal components: , Among them, each These represent different frequency band components in the photovoltaic disturbance signal; Thus, the adaptive decomposition of the perturbation signal based on the improved variational mode decomposition is completed, and the decomposed perturbation signal is obtained. , This will be used in step S3 to construct the Hamiltonian dynamics model; Step S3: Model the disturbance characteristics based on the Hamiltonian dynamics model, construct the Hamiltonian system, and calculate the phase space dynamic equation of the disturbance signal. Step 3.1, Hamiltonian dynamics modeling: The phase space of the disturbance signal is set as ,in, and Represents modal components Position and momentum; , in, Modal components quality It is the potential energy function; For the disturbance signal, momentum With position Satisfy the following Hamiltonian equations: , , in, = and = Indicates speed and force; Step 3.2, construct the phase space model of the perturbation signal: By analyzing each modal component Numerical integration of Hamilton's equations yields the trajectory in phase space. Assuming each modal component Potential energy function For simple second potential energy: , in, Modal components The elastic constant; By solving the Hamiltonian equations, the dynamic equations in phase space are obtained: , , in, Represents modal components Position and momentum in phase space; Step 3.3, calculate the phase space dynamic equation of the disturbance signal: The classic fourth-order Runge-Kutta algorithm is used to numerically integrate the Hamiltonian equation, with a set time step. Through iterative updates and Value: , , Step 3.4: Perform Hamiltonian dynamics modeling analysis based on the disturbance signal: After step S3 is completed, the dynamic characteristics of the disturbance signal have been modeled using the Hamiltonian dynamics model. Then, step S4 is performed to further analyze the nonlinear characteristics of the disturbance signal using the Lyapunov exponent. Step S4: Calculate the Lyapunov exponent for the Hamiltonian dynamic equation established in step S3, analyze the nonlinear characteristics of the disturbance signal based on the Lyapunov exponent, calculate the maximum Lyapunov exponent of the disturbance signal and determine its chaotic characteristics. Step 4.1, Mathematical definition of the Lyapunov exponent: In phase space, suppose the phase trajectory of the disturbance signal is described by Hamiltonian dynamics equations, and the state vector is: , in, and Modal components phase space coordinates, The optimal number of modal components; The time evolution of the perturbation signal in phase space is described by the following Hamiltonian dynamic equation: , in, The system's evolution equation; At the initial moment Select adjacent trajectory points in phase space and The initial small perturbation is: , The evolution after time t satisfies: , in, The Lyapunov exponent describes the exponential growth rate of perturbations between adjacent trajectories. The rate of change of distance between trajectories is given by the following formula: , Step 4.2, calculate the maximum Lyapunov exponent: Calculating the maximum Lyapunov exponent requires numerical simulation of the phase space trajectory. A small perturbation method is used to track the divergence of adjacent trajectories. The specific calculation steps are as follows: Step 4.2.1, in phase space trajectory In the middle, select the initial point And select an initial small perturbation within the neighborhood. To satisfy: , in, The initial disturbance amplitude; Step 4.2.2: Solve the Hamiltonian dynamic equations using numerical integration to calculate the trajectory after the disturbance. Compared with the original trajectory Distance between: , in, It reflects the trend of the disturbance trajectory over time.

[0031] Step 4.2.3: Plot the graph in logarithmic coordinates. The curve varies with time t, and the slope is solved using linear regression. : , in, For time steps; Step 4.3, determine the chaotic characteristics based on the maximum Lyapunov exponent: According to the maximum Lyapunov index The magnitude of the value is used to determine the chaotic characteristics of the disturbance signal: like If the value is greater than 0, the disturbance signal exhibits chaotic characteristics. like =0, then the system belongs to quasi-periodic motion; like If the value is less than 0, then the system has an attractor. Step 4.4, calculate multiple sets of Lyapunov indices: Except for the maximum Lyapunov index In addition, calculate other Lyapunov indices. To further analyze the stability of the perturbation signal, the complete Lyapunov exponent spectrum was obtained. It consists of multiple sets of indices: , in, The optimal number of modes; Using the QR decomposition method, the Lyapunov exponent spectrum is calculated. First, the Jacobian matrix of the perturbation equation is constructed. : Then, by tracking the exponential divergence rate of the orthogonal basis through QR decomposition, all Lyapunov exponents were calculated. Step 4.5, Result Output and Next Prediction: After completing the Lyapunov exponent calculation, the nonlinear characteristics of the disturbance signal are obtained: like If the value is greater than 0, it indicates that the disturbance signal has chaotic characteristics, and a prediction model adapted to chaotic characteristics is adopted.

[0032] like If ≤0, the disturbance signal exhibits periodic and convergent behavior, and a traditional time series prediction model is used; At this point, step S4 is complete, the Lyapunov exponent of the disturbance signal has been successfully extracted, and the nonlinear characteristics have been analyzed. Next, we proceed to step S5, which involves making short-term and long-term predictions of the disturbance signal based on meteorological data. Step S5: Combining the disturbance prediction with meteorological data, the ARIMA model is used for short-term disturbance prediction, and the Transformer-LSTM hybrid model is used for long-term disturbance prediction. Step 5.1, Short-term disturbance prediction based on the ARIMA model: Assume the time series of the disturbance signal is as follows: ,in, Let N be the disturbance signal value with time step N, where N is the total number of observation time points; Step 5.1.1, Establishing the ARIMA model: The ARIMA model consists of parameters. Decide: p is the autoregressive order, representing the influence of the value at time p on the current value; d is the difference order, which represents the number of differences required to stabilize the time series; q is the moving average order, representing the influence of the first q error terms on the current value; The mathematical expression for the ARIMA model is: , in, For the shift operator, For the autoregressive part, is a polynomial. For the polynomial of the moving average part, This is the white noise term; Step 5.1.2: Optimize and select the best option using AIC (Akaike Information Criterion) and BIC (Bayesian Information Criterion). : , , Where L is the log-likelihood function of the model. N represents the number of model parameters, and N represents the total number of observation time points. Optimal parameters Determined by the following optimization objectives: , Step 5.1.3, Predict the disturbance signal: Based on the optimal ARIMA model, predict the short-term disturbance signal: , Where h is the prediction step size, For the predicted future time steps The disturbance signal value, These are the autoregressive coefficients. The moving average coefficient; Step 5.2, Long-term perturbation prediction is based on a Transformer-LSTM hybrid model: Due to the chaotic nature of the perturbation signal, a deep learning model that can capture long-term dependencies is adopted, namely the Transformer-LSTM hybrid model. Step 5.2.1, LSTM network modeling. The LSTM structure consists of an input gate, a forget gate, and an output gate. The calculation formula is as follows: Forgotten Gate:

[0033] in, For the current input, This is the hidden state from the previous moment. and For the weights and biases of the forget gate, For the Sigmoid function; Input Gate: , , in, The output of the input gate, Here is the weight matrix of the input gate. Let be the bias vector of the input gate. Candidate cell state, This is the weight matrix for the candidate cell states. is the bias vector for the candidate cell state; Cell status update: , in, The current cell state, This represents the cell state at the previous moment; Output gate: , , in, For the output of the output gate, This is the weight matrix of the output gate. This is the bias vector for the output gate. This is the final output of the LSTM; Step 5.2.2, Transformer processes global features: The Transformer uses a self-attention mechanism to extract global features. Given an input perturbation signal sequence S, define the query Q, key K, and value V: , in, , , It is a linear transformation matrix; Calculate attention weights:

[0034] in, The dimension of the key vector; Step 5.2.3, Predict the perturbation signal: Extract long-term dependency features using Transformer, and predict using LSTM: , in, for Extracted features For LSTM prediction functions; Step 5.3: Perform disturbance correction based on meteorological data: Disturbance signals from photovoltaic systems are affected by meteorological factors, including: Solar radiation intensity ,temperature Wind speed and cloud cover ; Constructing a perturbation correction model: , in, , , , As the weight of meteorological factors, These are the predicted values ​​after correction from meteorological data. This is the original predicted value; Step 5.4, Results Output and Application: Short-term forecasts are used for real-time monitoring of photovoltaic systems; Long-term forecasting is used for power grid dispatching; The prediction results will be used for real-time early warning in step S6 to prevent photovoltaic disturbances from affecting the power grid; at this point, step S5 is completed, and the short-term and long-term evolution trends of the disturbance signal have been successfully predicted. Step S6: Output and application of results. Output the disturbance identification results and provide real-time early warning.

[0035] This embodiment addresses the common low-frequency harmonics and short-term fluctuations in photovoltaic systems by employing a multi-stage processing flow. The technical solution of this invention starts with data acquisition and preprocessing, performs adaptive signal decomposition through improved variational mode decomposition, extracts nonlinear features using Hamiltonian dynamics and Lyapunov exponents, and finally constructs a hybrid prediction model by combining meteorological data.

[0036] Step S1, Data Acquisition and Preprocessing: Data acquisition uses a dedicated data acquisition device to simultaneously record the three-phase voltage, current, and grid-side parameters output by the photovoltaic system; the sampling frequency and sampling duration are set to form a discrete signal sequence; signal denoising uses wavelet transform for multi-scale decomposition to remove high-frequency noise; In the process of signal denoising, the number of decomposition layers J and the mother wavelet function are set, soft threshold filtering is performed on the coefficients of each layer, and then the signal is recovered by wavelet reconstruction. Signal normalization and time alignment utilize the minimum-maximum normalization method to map the original data to a predetermined interval. Then, the time delay is found by calculating the cross-correlation function, and the signals of each channel are aligned to ensure synchronization.

[0037] Step S2, Adaptive decomposition based on improved variational mode decomposition: An improved variational mode decomposition model is established to construct a mathematical model for the preprocessed perturbation signal. The model assumes that each modal component satisfies the finite bandwidth characteristic. Automatic parameter selection utilizes information geometry optimization to automatically select the optimal number of modes K based on maximizing spectral entropy. Through Bayesian optimization, the penalty factor α is dynamically determined with reconstruction error as the objective function. Decomposition calculations decompose the signal into several modes, each reflecting perturbations in different frequency bands. During numerical calculations, an iterative method is used to update the Lagrange multipliers to ensure that the constraints are met. After using the improved variational mode decomposition, no manual parameter fixing is required, ensuring that the decomposition results closely match the actual changes in the signal and improving the accuracy of perturbation pattern recognition.

[0038] Step S3, Hamiltonian dynamics model construction: Phase space construction constructs position and momentum variables for each modal component, establishes corresponding Hamiltonian functions, and simply uses quadratic potential energy to describe vibration characteristics; numerical integration and dynamic equations utilize the fourth-order Runge-Kutta algorithm to numerically integrate the Hamiltonian equations, obtaining the motion trajectory of each mode in phase space, truly reflecting the energy transfer of disturbance, and through the Hamiltonian model, analyzing the disturbance signal from the perspective of energy conservation, intuitively revealing the inherent physical mechanism of the signal.

[0039] Step S4, Lyapunov exponent calculation and chaotic feature extraction: The initial perturbation setting selects a nearby point in the phase space as the initial perturbation to ensure that the perturbation is small; after numerical integration of the exponent calculation, the divergence rate of adjacent trajectories is tracked to calculate the maximum Lyapunov exponent, and the Lyapunov exponent spectrum is obtained simultaneously using the QR decomposition method; characteristic judgment: if the maximum Lyapunov exponent is greater than zero, it proves that the perturbation signal has chaotic characteristics; to achieve a deep interpretation of the nonlinear behavior of the perturbation signal, improve the physical interpretability of the signal and the reliability of monitoring.

[0040] Step S5, Disturbance prediction based on meteorological data: Short-term forecasting uses the ARIMA model to model the time-series characteristics of the disturbance signal, and uses the AIC and BIC criteria to determine the optimal model parameters to achieve short-term trend forecasting. Long-term prediction constructs a Transformer-LSTM hybrid model, using Transformer to extract global features and LSTM to capture long-term dependencies, which can effectively grasp the long-term evolution of perturbations. Meteorological data is collected synchronously and integrated to build a correction model, which corrects the forecast results, takes into account the influence of the external environment, and combines meteorological data to make the forecast model closer to reality, thus significantly improving the accuracy of early warning.

[0041] Step S6, Result Output and Application: The disturbance identification results integrate the decomposition results, dynamic model data and predicted values ​​output from each step to form a real-time disturbance early warning signal; Application scenarios include real-time monitoring of photovoltaic systems; power grid dispatch centers can respond quickly based on early warning signals to ensure power grid stability.

[0042] A terminal device includes a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the computer program is configured to perform a method for identifying power quality disturbance characteristics of a distributed photovoltaic interference power source.

[0043] A storage medium storing a computer program, which, when executed by a processor, enables a method for identifying power quality disturbance characteristics of a distributed photovoltaic interference power source.

[0044] The terminal equipment automatically collects multiple power quality signals, performs signal preprocessing, denoising, normalization and time alignment, and uses improved variational mode decomposition to adaptively decompose the disturbance signal, constructs Hamiltonian dynamics model to calculate Lyapunov exponent, extracts nonlinear features of the signal, and combines meteorological data to make short-term and long-term disturbance predictions. The terminal equipment has a simple structure and is easy to integrate. It can process and feedback disturbance information in real time on site, reduce the false judgment rate, and ensure the accuracy of monitoring.

[0045] The storage medium can be a non-volatile medium such as solid-state memory or flash memory; the computer program contains a complete signal acquisition, preprocessing, decomposition, modeling and prediction process; it is easy to port to different hardware platforms, ensuring technology reproducibility and system upgrades.

[0046] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for identifying power quality disturbance characteristics of distributed photovoltaic interference power sources, characterized in that, include: Step S1, Data Acquisition and Preprocessing: Acquire the power quality signal of the photovoltaic system and perform noise reduction, normalization and time alignment; Step S2: Based on the improved variational mode decomposition, the perturbation signal is adaptively decomposed, the number of modes is automatically selected using information geometry optimization, and the penalty factor is selected through Bayesian optimization. Among them, information geometry optimization automatically selects the number of modes; Based on the information geometry optimization method, the optimal number of modes is calculated by maximizing the spectral entropy of each modal component; Bayesian optimization to determine the penalty factor: A Bayesian optimization method is used to determine the reconstruction error of the decomposed signal. As the objective function; Predicting different penalty factors using a Gaussian process regression model under the value Furthermore, the optimal penalty factor is selected based on the expected improvement criterion; Step S3: Model the disturbance characteristics based on the Hamiltonian dynamics model, construct the Hamiltonian system, and calculate the phase space dynamic equation of the disturbance signal. The disturbance feature modeling based on the Hamiltonian dynamics model further includes: after completing the adaptive decomposition of the disturbance signal in step S2, obtaining... Group modal components Modal components represent disturbance signals in different frequency bands in the photovoltaic system; Hamiltonian dynamics modeling: The phase space of the disturbance signal is set as ,in, and Represents modal components Position and momentum; ， in, Modal components quality It is the potential energy function; Constructing a phase space model of the perturbation signal: By analyzing each modal component Numerical integration of Hamilton's equations yields the trajectory in phase space. Assuming each modal component Potential energy function For simple second potential energy: ， in, Modal components The elastic constant; By solving the Hamiltonian equations, the dynamic equations in phase space are obtained; Calculate the phase space dynamic equation of the disturbance signal: The classic fourth-order Runge-Kutta algorithm is used to numerically integrate the Hamiltonian equation, with a set time step. Through iterative updates and The value; Hamiltonian dynamics modeling and analysis were performed in conjunction with disturbance signals; Step S4: Calculate the Lyapunov exponent for the Hamiltonian dynamic equation established in step S3, analyze the nonlinear characteristics of the disturbance signal based on the Lyapunov exponent, calculate the maximum Lyapunov exponent of the disturbance signal and determine its chaotic characteristics. Among them, if the maximum Lyapunov index If the value is greater than 0, it indicates that the disturbance signal has chaotic characteristics, and a prediction model adapted to chaotic characteristics is adopted. If the maximum Lyapunov index If ≤0, the disturbance signal exhibits periodic and convergent behavior, and a traditional time series prediction model is used; Step S5: Combining the disturbance prediction with meteorological data, the ARIMA model is used for short-term disturbance prediction, and the Transformer-LSTM hybrid model is used for long-term disturbance prediction. Step S6: Output and application of results. Output the disturbance identification results and provide real-time early warning.

2. The method for identifying power quality disturbance characteristics of a distributed photovoltaic interference power source according to claim 1, characterized in that, In step S1, data acquisition and preprocessing further include: Step 1.1, Acquire power quality signals: When collecting power quality signals, data acquisition equipment is used to record the three-phase voltage output by the photovoltaic system. and three-phase current And simultaneously record the power quality parameters on the grid side; Let the sampling frequency be If the sampling duration is T, then the discrete form of the acquired signal can be expressed as: ， in, Let be the discrete value of the voltage of the nth phase at the kth sampling point. Let n be the discrete value of the nth phase current at the kth sampling point. =1 / The sampling period is = This represents the total number of sampling points; Step 1.2, Signal Denoising: Wavelet transform is used to decompose the signal at multiple scales to remove high-frequency noise components; Let the number of wavelet decomposition levels be J, and the mother wavelet function be... The wavelet transform coefficients of the signal are then expressed as: ， in, For signal Wavelet transform coefficients at scale a and location b For the mother wavelet function, Let be the complex conjugate of the mother wavelet function, a be the scaling factor, and b be the translation factor; The noise reduction process includes the following steps: Calculate the wavelet transform coefficients of the signal ; Set threshold Soft threshold filtering is applied to the high-frequency wavelet coefficients: ， in, The threshold for the j-th level wavelet decomposition is... These are the update coefficients after soft thresholding. Wavelet reconstruction is performed to obtain the denoised signal. ; Step 1.3, Signal Normalization: The signal is mapped to the interval [1, 2] using the min-max normalization method: ， ， in, The original voltage signal, The original current signal, This is the minimum value of the voltage signal. This represents the maximum value of the voltage signal. This represents the minimum value of the current signal. This represents the maximum value of the current signal. The normalized voltage signal. This is the normalized current signal; Step 1.4, Signal Time Alignment: The time delay between signals is calculated using the cross-correlation function. The formula for the cross-correlation function is as follows: ， in, voltage signal and current signal The cross-correlation function values ​​between them Let k be the discrete value of the a-th phase voltage signal. Let k be the discrete value of the a-th phase current signal. For time delay; Find the value corresponding to the maximum cross-correlation coefficient. Furthermore, the signals are adjusted accordingly to ensure that all signals are aligned to the same time base.

3. The method for identifying power quality disturbance characteristics of a distributed photovoltaic interference power source according to claim 1, characterized in that, In step S2, the adaptive decomposition of the perturbation signal based on the improved variational mode decomposition further includes: Step 2.1, Mathematical modeling of variational mode decomposition: Let the preprocessed disturbance signal be represented as : ， in, For the k-th modal component, This is the preprocessed perturbation signal. The target number of modes; Modal components At a specific center frequency The vicinity has limited bandwidth; Variational mode decomposition achieves signal decomposition by solving the following variational optimization problem: ， in, For modal components The signal after Hilbert transform, where j is the imaginary unit. The center frequency of the modal component, For time derivative operators, For the i-th modal component; Using the Lagrange multiplier method to introduce constraints, construct the Lagrange function: ， in, Let Lagrangian function be the variational mode decomposition function. As a penalty factor, For Lagrange multipliers, This is an inner product operation; Step 2.2, Information geometry optimization automatically selects the number of modes: Based on the information geometry optimization method, the spectral entropy of each modal component is maximized. Calculate the optimal number of modes: ， in, Modal components In frequency Normalized energy distribution at: ， Let the total information entropy As the objective function: ， Iterate through the different modal numbers K to find the one that makes Maximum number of optimal modes : ， in, Let be the spectral entropy of the k-th modal component. Modal components In frequency Normalized energy distribution at the location, Modal components In frequency Fourier transform at the point, For frequency points, For information entropy, The optimal number of modes; Step 2.3, Bayesian optimization to determine the penalty factor: A Bayesian optimization method is used to determine the penalty factor based on the reconstruction error of the decomposed signal. As the objective function: ， Predicting different penalty factors using a Gaussian process regression model under the value And select the optimal penalty factor based on the expected improvement criterion: ， in, For signal reconstruction error, As a penalty factor, This is the reconstructed signal of all modal components after variational mode decomposition. To calculate the penalty factor that minimizes the reconstruction error ; Step 2.4, calculate the decomposed disturbance signal: Based on certainty and The preprocessed disturbance signal is then subjected to final variational mode decomposition to obtain... Group modal components: ， Among them, each These represent different frequency band components in the photovoltaic disturbance signal; Thus, the adaptive decomposition of the perturbation signal based on the improved variational mode decomposition is completed, and the decomposed perturbation signal is obtained. , This will be used in step S3 to construct the Hamiltonian dynamics model.

4. The method for identifying power quality disturbance characteristics of a distributed photovoltaic interference power source according to claim 1, characterized in that, For the disturbance signal, momentum With position Satisfy the following Hamiltonian equations: ， ， in, = and = Indicates speed and force; By solving the Hamiltonian equations, the dynamic equations in phase space are obtained: ， ， in, Represents modal components Position and momentum in phase space; Calculate the phase space dynamic equation of the disturbance signal: The classic fourth-order Runge-Kutta algorithm is used to numerically integrate the Hamiltonian equation, with a set time step. Through iterative updates and The value; ， ， After step S3 is completed, the dynamic characteristics of the disturbance signal have been modeled using the Hamiltonian dynamics model. Then, step S4 is performed to further analyze the nonlinear characteristics of the disturbance signal using the Lyapunov exponent.

5. The method for identifying power quality disturbance characteristics of a distributed photovoltaic interference power source according to claim 1, characterized in that, In step S4, the analysis of the nonlinear characteristics of the disturbance signal based on the Lyapunov exponent further includes: In step S3, the Hamiltonian dynamic model of the perturbation signal has been constructed, and the phase space dynamic equation of the perturbation signal has been obtained. In order to further analyze the nonlinear characteristics and chaotic features of the perturbation signal, in step S4, the Lyapunov exponent needs to be calculated, and the chaotic characteristics of the perturbation signal are judged based on the maximum Lyapunov exponent. Step 4.1, Mathematical definition of the Lyapunov exponent: In phase space, suppose the phase trajectory of the disturbance signal is described by Hamiltonian dynamics equations, and the state vector is: , in, and Modal components phase space coordinates, The optimal number of modal components; The time evolution of the perturbation signal in phase space is described by the following Hamiltonian dynamic equation: ， in, The system's evolution equation; At the initial moment Select adjacent trajectory points in phase space and The initial small perturbation is: , The evolution after time t satisfies: , in, The Lyapunov exponent describes the exponential growth rate of perturbations between adjacent trajectories. The rate of change of distance between trajectories is given by the following formula: , Step 4.2, calculate the maximum Lyapunov exponent: Calculating the maximum Lyapunov exponent requires numerical simulation of the phase space trajectory. A small perturbation method is used to track the divergence of adjacent trajectories. The specific calculation steps are as follows: Step 4.2.1, in phase space trajectory In the middle, select the initial point And select an initial small perturbation within the neighborhood. To satisfy: , in, The initial disturbance amplitude; Step 4.2.2: Solve the Hamiltonian dynamic equations using numerical integration to calculate the trajectory after the disturbance. Compared with the original trajectory Distance between: , in, Reflects the trend of change in the disturbance trajectory over time; Step 4.2.3: Plot the graph in logarithmic coordinates. The curve varies with time t, and the slope is solved using linear regression. : , in, For time steps; Step 4.3, determine the chaotic characteristics based on the maximum Lyapunov exponent: According to the maximum Lyapunov index The magnitude of the value is used to determine the chaotic characteristics of the disturbance signal: like If the value is greater than 0, the disturbance signal exhibits chaotic characteristics. like =0, then the system belongs to quasi-periodic motion; like If the value is less than 0, then the system has an attractor. Step 4.4, calculate multiple sets of Lyapunov indices: Except for the maximum Lyapunov index In addition, calculate other Lyapunov indices. To further analyze the stability of the perturbation signal, the complete Lyapunov exponent spectrum was obtained. It consists of multiple sets of indices: ， in, The optimal number of modes; Using the QR decomposition method, the Lyapunov exponent spectrum is calculated. First, the Jacobian matrix of the perturbation equation is constructed. : Then, by tracking the exponential divergence rate of the orthogonal basis through QR decomposition, all Lyapunov exponents were calculated. Step 4.5, Result Output and Next Prediction: After completing the Lyapunov exponent calculation, the nonlinear characteristics of the disturbance signal are obtained: If the maximum Lyapunov index If the value is greater than 0, it indicates that the disturbance signal has chaotic characteristics, and a prediction model adapted to chaotic characteristics is adopted. If the maximum Lyapunov index If ≤0, the disturbance signal exhibits periodic and convergent behavior, and a traditional time series prediction model is used; At this point, step S4 is complete, and the Lyapunov exponent of the disturbance signal has been successfully extracted and its nonlinear characteristics have been analyzed. Next, we proceed to step S5, which involves making short-term and long-term predictions of the disturbance signal based on meteorological data.

6. The method for identifying power quality disturbance characteristics of a distributed photovoltaic interference power source according to claim 1, characterized in that, In step S5, the disturbance prediction based on meteorological data further includes: In step S4, the nonlinear characteristics of the disturbance signal are analyzed and chaotic behavior is judged by calculating the Lyapunov exponent. In step S5, the disturbance signal is predicted in the short and long term by combining the meteorological data of the photovoltaic system. Step 5.1, Short-term disturbance prediction based on the ARIMA model: Assume the time series of the disturbance signal is as follows: ,in, Let N be the disturbance signal value with time step N, where N is the total number of observation time points; Step 5.1.1, Establishing the ARIMA model: The ARIMA model consists of parameters. Decide: p is the autoregressive order, representing the influence of the value at time p on the current value; d is the difference order, which represents the number of differences required to stabilize the time series; q is the moving average order, representing the influence of the first q error terms on the current value; The mathematical expression for the ARIMA model is: ， in, For the shift operator, For the autoregressive part, is a polynomial. For the polynomial of the moving average part, This is the white noise term; Step 5.1.2: Optimize and select the best option using AIC (Akaike Information Criterion) and BIC (Bayesian Information Criterion). : ， ， Where L is the log-likelihood function of the model. N represents the number of model parameters, and N represents the total number of observation time points. Optimal parameters Determined by the following optimization objectives: ， Step 5.1.3, Predict the disturbance signal: Based on the optimal ARIMA model, predict the short-term disturbance signal: ， Where h is the prediction step size, For the predicted future time steps The disturbance signal value, These are the autoregressive coefficients. This is the moving average coefficient.

7. The method for identifying power quality disturbance characteristics of a distributed photovoltaic interference power source according to claim 1, characterized in that, In step S5, the disturbance prediction based on meteorological data further includes: Step 5.2, Long-term perturbation prediction is based on a Transformer-LSTM hybrid model: Due to the chaotic nature of the perturbation signal, a deep learning model that can capture long-term dependencies is adopted, namely the Transformer-LSTM hybrid model. Step 5.2.1, LSTM network modeling. The LSTM structure consists of an input gate, a forget gate, and an output gate. The calculation formula is as follows: Forgotten Gate: in, For the current input, This is the hidden state from the previous moment. and For the weights and biases of the forget gate, For the Sigmoid function; Input Gate: , ， in, The output of the input gate, Here is the weight matrix of the input gate. Let be the bias vector of the input gate. Candidate cell state, This is the weight matrix for the candidate cell states. is the bias vector for the candidate cell state; Cell status update: , in, The current cell state, This represents the cell state at the previous moment; Output gate: , ， in, For the output of the output gate, This is the weight matrix of the output gate. This is the bias vector for the output gate. This is the final output of the LSTM; Step 5.2.2, Transformer processes global features: The Transformer uses a self-attention mechanism to extract global features. Given an input perturbation signal sequence S, define the query Q, key K, and value V: ， in, , , It is a linear transformation matrix; Calculate attention weights: in, The dimension of the key vector; Step 5.2.3, Predict the perturbation signal: Extract long-term dependency features using Transformer, and predict using LSTM: ， in, for Extracted features This is the LSTM prediction function.

8. The method for identifying power quality disturbance characteristics of a distributed photovoltaic interference power source according to claim 1, characterized in that, In step S5, the disturbance prediction based on meteorological data further includes: Step 5.3: Perform disturbance correction based on meteorological data: Disturbance signals from photovoltaic systems are affected by meteorological factors, including: Solar radiation intensity ,temperature Wind speed and cloud cover ; Constructing a perturbation correction model: ， in, , , , As the weight of meteorological factors, These are the predicted values ​​after correction from meteorological data. This is the original predicted value; Step 5.4, Results Output and Application: Short-term forecasts are used for real-time monitoring of photovoltaic systems; Long-term forecasting is used for power grid dispatching; The prediction results will be used for real-time early warning in step S6 to prevent photovoltaic disturbances from affecting the power grid; at this point, step S5 is completed, and the short-term and long-term evolution trends of the disturbance signal have been successfully predicted.

9. A terminal device, characterized in that, The system includes a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the computer program is configured to perform the method for identifying the power quality disturbance characteristics of a distributed photovoltaic interference power source as described in any one of claims 1-8.

10. A storage medium, characterized in that, The storage medium stores a computer program, which, when executed by a processor, implements the method for identifying the power quality disturbance characteristics of a distributed photovoltaic interference power source as described in any one of claims 1-8.

Images