Voltage characteristic analysis method considering line impedance under new energy high-penetration power grid

Abstract

The application discloses a voltage characteristic analysis method considering line impedance under a new energy high-permeation power grid, which comprises the following steps: collecting real-time voltage signals and current signals, and pre-processing the signals; performing multi-dimensional predetermined harmonic extraction to obtain multi-dimensional harmonic components; calculating a complex impedance estimation value; calling a pre-configured credibility evaluation model to obtain a line inductance estimation value; obtaining system short-circuit capacity parameters and calculating a power grid voltage variation; performing time series analysis and multi-scale decomposition to obtain fluctuation trends and periodic characteristics; adopting fuzzy logic and a predictive analysis method to evaluate an out-of-limit risk level and issue a warning; constructing an analytic hierarchy process model, calculating the weight of a stability index, adopting a fuzzy comprehensive evaluation method to comprehensively evaluate stability, and obtaining a stability evaluation result. The application can effectively and accurately analyze the voltage characteristics of the new energy high-permeation power grid, and provides comprehensive, accurate and real-time decision support for power grid operators.

Description

Technical Field

[0001] The present invention belongs to the field of new energy station grid connection, and in particular to a voltage characteristic analysis method considering line impedance in a new energy high-penetration power grid. Background Art

[0002] Grid voltage fluctuations and over-limit issues are prominent challenges faced during the integration of high-penetration renewable energy. On the one hand, the output power of photovoltaic and wind farms varies with environmental factors such as location, sunlight intensity, and wind speed. This fluctuation inevitably causes system voltage instability. On the other hand, the main areas connected to photovoltaic farms often have low electricity demand. Therefore, photovoltaic farms must transmit power over long distances to higher-loaded areas via high-voltage transmission lines. During this transmission process, the inherent uncertainty of sunlight and the long transmission distances can affect the reactive power balance of the system, causing significant fluctuations in the busbar voltage that transmits photovoltaic power, which is detrimental to system voltage stability.

[0003] In practical engineering, grid impedance is a dynamic variable that changes with the addition of transformers and transmission lines, and its range of variation is uncertain, leading to variations in the system's grid voltage. To ensure stable power system operation, accurate measurement and assessment of line impedance are required, along with appropriate control measures to maintain the system's electrical characteristics.

[0004] Grid impedance is an unknown element in grid-connected systems with high renewable energy penetration. Using online impedance identification technology to measure the grid's equivalent impedance in real time can effectively improve the adaptability of these systems to the grid. Research on grid impedance measurement has attracted considerable attention. Initially, research on impedance measurement methods focused on offline measurement or the addition of additional hardware circuits. However, with the development of distributed generation systems, these offline or hardware-based measurement methods have become less applicable. Subsequently, online measurement methods have been proposed. These can be categorized as active or passive, depending on whether or not a disturbance signal is injected. Active detection methods include injecting non-characteristic harmonics, applying power disturbances, and inducing LCL resonance. These measurement methods require injecting a disturbance signal into the system and analyzing and calculating the grid's equivalent impedance. However, the injection of the disturbance signal can affect the system's power quality, and research on how to reduce or offset this impact on grid-connected power quality is still lacking. Furthermore, the accuracy of these measurement methods relies heavily on accurately extracting the disturbance signal component, resulting in low measurement accuracy.

[0005] In summary, further research and innovation are needed to effectively coordinate existing resources to overcome the adverse effects of high-penetration photovoltaic and wind power cluster access on the power system, and to achieve the goal of improving the power quality of the power system and ensuring safety and stability. Summary of the Invention

[0006] The purpose of the invention is to provide a voltage characteristic analysis method considering line impedance in a new energy high penetration power grid to solve the above-mentioned problems existing in the prior art.

[0007] The technical solution is a voltage characteristic analysis method considering line impedance in a high-penetration renewable energy grid, including the following steps:

[0008] S1. Collect real-time voltage and current signals of a three-phase AC power grid, perform preprocessing and multi-dimensional predetermined subharmonic extraction on them, and obtain multi-dimensional harmonic components;

[0009] S2. Based on the multi-dimensional harmonic components, a complex impedance estimation value is calculated; a pre-configured fuzzy logic-based credibility assessment model is called to obtain a line inductance estimation value;

[0010] S3. Obtain system short-circuit capacity parameters and, combined with estimated line inductance, calculate grid voltage changes. Perform time series analysis and multi-scale decomposition to determine fluctuation trends and periodic characteristics. Use fuzzy logic and predictive analysis methods to assess over-limit risk levels and issue warnings.

[0011] S4. Construct a system state space model to obtain stability indicators; construct a hierarchical analysis model to calculate the weights of the stability indicators, use a fuzzy comprehensive evaluation method to conduct a comprehensive stability assessment, and obtain the results of the stability assessment.

[0012] S1. Collecting real-time voltage and current signals of a three-phase AC power grid and preprocessing them to obtain preprocessed voltage and current signals; performing multi-dimensional predetermined subharmonic extraction based on the preprocessed voltage and current signals to obtain multi-dimensional harmonic components;

[0013] S2. Based on the multi-dimensional harmonic components, a multi-model fusion algorithm based on Bayesian estimation is used, combined with a dynamic weight adjustment strategy, to calculate the complex impedance estimate. Based on the complex impedance estimate, a pre-configured credibility assessment model based on fuzzy logic is called to obtain the line inductance estimate.

[0014] S3. Obtain system short-circuit capacity parameters. Based on these system short-circuit capacity parameters and estimated line inductance, calculate the grid voltage change. Perform time series analysis and multi-scale decomposition based on the grid voltage change to determine fluctuation trends and periodicity characteristics. Based on these fluctuation trends and periodicity characteristics, use fuzzy logic and predictive analysis methods to assess the level of over-limit risk and issue an early warning.

[0015] S4. Based on the preprocessed voltage and current signals, line inductance estimation values, and grid voltage changes, a system state space model that considers multiple dynamic characteristics is constructed to obtain stability indicators. A hierarchical analysis model is constructed to calculate the weights of the stability indicators. Based on the weights of the stability indicators, a fuzzy comprehensive evaluation method is used to perform a comprehensive stability assessment to obtain the results of the stability assessment.

[0016] Beneficial effects: The present invention can effectively and accurately analyze the voltage characteristics of a new energy high-penetration power grid, provide comprehensive, precise, and real-time decision-making support for power grid operators, improve measurement accuracy and power system power quality, and can also promptly detect potential voltage problems and stability risks, and provide targeted optimization suggestions. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 Flowchart of the present invention.

[0018] Figure 2 This is a flow chart of step S1 of the present invention.

[0019] Figure 3 This is a flow chart of step S2 of the present invention.

[0020] Figure 4 This is a flow chart of step S3 of the present invention.

[0021] Figure 5 This is a flow chart of step S4 of the present invention.

[0022] Figure 6 This is the main circuit of the high-penetration new energy grid-connected system according to an embodiment of the present invention.

[0023] Figure 7 This is the grid-connected equivalent circuit diagram of a high-penetration new energy unit according to an embodiment of the present invention.

[0024] Figure 8 FIG. 4 is a schematic diagram of a method for extracting specific subharmonic components according to an embodiment of the present invention.

[0025] Figure 9 2 is a diagram of an online inductance identification solution according to an embodiment of the present invention.

[0026] Figure 10 This is a grid-connected equivalent circuit diagram of a high-penetration new energy unit according to an embodiment of the present invention. DETAILED DESCRIPTION

[0027] The following describes the present application in more detail with reference to specific embodiments. Figure 1 As shown, this application proposes a voltage characteristic analysis method considering line impedance in a new energy high penetration power grid, including the following steps:

[0028] S1, collect real-time voltage signals and current signals of a three-phase alternating current power grid, and pre-process the same to obtain pre-processed voltage and current signals; based on the pre-processed voltage and current signals, perform multi-dimensional predetermined harmonic extraction to obtain multi-dimensional harmonic components;

[0029] S2, based on the multi-dimensional harmonic components, use a multi-model fusion algorithm based on Bayesian estimation combined with a dynamic weight adjustment strategy to calculate a complex impedance estimation value; based on the complex impedance estimation value, call a pre-configured fuzzy logic-based credibility evaluation model to obtain a line inductance estimation value;

[0030] S3, obtain a system short-circuit capacity parameter, based on the system short-circuit capacity parameter and the line inductance estimation value, calculate a power grid voltage variation; based on the power grid voltage variation, perform time series analysis and multi-scale decomposition to obtain fluctuation trend and periodic characteristics; based on the fluctuation trend and periodic characteristics, use fuzzy logic and predictive analysis methods to evaluate the risk level of over-limit and issue a warning;

[0031] S4, based on the pre-processed voltage and current signals, the line inductance estimation value and the power grid voltage variation, construct a system state space model considering multiple dynamic characteristics to obtain a stability index; construct an analytic hierarchy process model to calculate the weight of the stability index, based on the weight of the stability index, use a fuzzy comprehensive evaluation method to perform comprehensive stability evaluation to obtain the result of the stability evaluation.

[0032] As shown in the figure, according to one aspect of the present application, step S1 is further as follows: Figure 2

[0033] S11, obtain real-time voltage signals of a three-phase alternating current power grid through a high-precision voltage sensor, and use an adaptive sampling rate algorithm and a high-precision analog-to-digital converter to process the same to obtain a discrete-time voltage sequence;

[0034] S12, collect real-time current signals of a three-phase alternating current power grid through a Hall effect current sensor, and use an adaptive sampling rate algorithm and a high-precision analog-to-digital converter to process the same to obtain a discrete-time current sequence;

[0035] S13, based on the discrete-time voltage sequence and the discrete-time current sequence, use a multi-objective optimization algorithm to perform intelligent phase selection to output optimal single-phase voltage and current sequences;

[0036] S14, use an adaptive notch filter and a machine learning optimization algorithm to filter out fundamental components from the optimal single-phase voltage and current sequences to obtain voltage and current sequences with fundamental components removed;

[0037] ​S15. Based on the voltage and current sequence with the fundamental component removed, a hybrid algorithm of wavelet packet transform and empirical mode decomposition is adopted, combined with adaptive threshold technology, to perform multi-dimensional predetermined subharmonic extraction to obtain multi-dimensional harmonic components.

[0038] According to one aspect of the present application, step S13 is further as follows:

[0039] S131. Perform fast Fourier transform based on the discrete-time voltage sequence and the discrete-time current sequence to obtain frequency spectrum information of the voltage and current signals;

[0040] S132. Calculate the signal-to-noise ratio and total harmonic distortion of each phase based on the frequency spectrum information of the voltage and current signals; and evaluate the phase balance among the three phases based on the signal-to-noise ratio and total harmonic distortion of each phase.

[0041] S133. Based on the phase balance degree, a multi-objective optimization problem is constructed and solved using a non-dominated sorting genetic algorithm to obtain a Pareto optimal solution set.

[0042] S134. Based on the Pareto optimal solution set, determine the optimal single-phase voltage and current sequence.

[0043] According to one aspect of the present application, step S14 is further as follows:

[0044] S141. Based on the optimal single-phase voltage and current sequence, a recursive least squares algorithm is used to dynamically estimate the actual grid frequency.

[0045] S142, initializing the notch filter, and adjusting the center frequency of the notch filter based on the estimated actual grid frequency to obtain an adjusted notch filter;

[0046] S143, calling a preconfigured support vector regression model to optimize the adjusted notch filter to obtain an optimized notch filter, repeating step S143 until a preset performance indicator is achieved to obtain a final optimized notch filter;

[0047] S144. Use the finally optimized notch filter to perform fundamental wave filtering on the optimal single-phase voltage and current sequence to obtain a voltage and current sequence with the fundamental wave component removed.

[0048] According to one aspect of the present application, step S15 is further as follows:

[0049] S151. Perform wavelet packet decomposition on the voltage and current sequence with the fundamental component removed to obtain coefficients of different frequency bands.

[0050] S152. Performing empirical mode decomposition on the coefficients of each frequency band to obtain a plurality of eigenmode functions; calculating the instantaneous frequency of each eigenmode function using a Hilbert transform; identifying an eigenmode function containing a predetermined subharmonic based on the instantaneous frequency information to obtain a selected eigenmode function;

[0051] S153, denoising the selected intrinsic mode function using an adaptive soft threshold method to obtain a denoised intrinsic mode function;

[0052] S154. Reconstruct the harmonic signal based on the denoised intrinsic mode function to obtain multi-dimensional harmonic components.

[0053] In one embodiment of the present application, a high-precision voltage sensor is used to obtain the real-time voltage signal of a three-phase AC power grid. An adaptive sampling rate algorithm is used to dynamically adjust the sampling frequency to accommodate small changes in the power grid frequency. A high-precision analog-to-digital converter is used to convert the analog signal into a digital signal to obtain a discrete-time voltage sequence. Specifically:

[0054] Deploy high-precision voltage sensors at appropriate locations on the three-phase power grid to obtain real-time analog voltage signals Va(t), Vb(t), and Vc(t). Implement an adaptive sampling rate algorithm to dynamically adjust the sampling frequency to adapt to slight changes in the power grid frequency:

[0055] fs[n]=f0+Δf[n];

[0056] Where fs[n] is the sampling frequency of the nth sampling period, f0 is the reference sampling frequency (e.g., 10 kHz), and Δf[n] is the frequency adjustment amount based on the grid frequency estimate.

[0057] Use a high-precision analog-to-digital converter to convert the analog signal into a digital signal to obtain a discrete time voltage series:

[0058] Va[n]=Va(nTs[n]); Vb[n]=Vb(nTs[n]); Vc[n]=Vc(nTs[n]);

[0059] Where Ts[n] = 1 / fs[n] is the sampling interval of the nth sampling cycle, and the discrete time voltage sequence Va[n], Vb[n], and Vc[n] obtained by sampling are stored for subsequent processing.

[0060] A Hall-effect current sensor is used to collect real-time current signals from a three-phase AC power grid. An adaptive sampling rate algorithm is also used to dynamically adjust the sampling frequency to maintain synchronization with voltage sampling. A high-precision analog-to-digital converter is used to convert the analog signal into a digital signal to obtain a discrete-time current sequence. Specifically:

[0061] Install Hall effect current sensors at appropriate locations on the three-phase power grid to obtain real-time analog current signals Ia(t), Ib(t), and Ic(t). Use the same adaptive sampling rate algorithm as voltage sampling to ensure synchronization between current and voltage sampling:

[0062] Use the sampling frequency fs[n] as the sampling frequency for current sampling. Use a high-precision analog-to-digital converter to convert the analog signal into a digital signal to obtain a discrete time current sequence:

[0063] Ia[n]=Ia(nTs[n]); Ib[n]=Ib(nTs[n]); Ic[n]=Ic(nTs[n]);

[0064] Ts[n] uses the same sampling interval as the voltage sampling, and the sampled discrete time current sequence Ia[n], Ib[n], Ic[n] is stored for subsequent processing.

[0065] A multi-objective optimization algorithm is used to perform intelligent phase selection on the sampled three-phase voltage and current sequence, outputting the optimal single-phase voltage and current sequence. The optimization objectives include maximizing the signal-to-noise ratio, minimizing the harmonic content, and achieving phase balance. Specifically:

[0066] Perform a Fast Fourier Transform (FFT) on each phase voltage and current signal to obtain spectral information. Calculate the signal-to-noise ratio (SNR) and total harmonic distortion (THD) for each phase. Evaluate the phase balance between the three phases. Construct a multi-objective optimization problem and define the objective function, including maximizing SNR, minimizing THD, and minimizing phase imbalance:

[0067] f1(X)=-SNR(X); f2(X)=THD(X); f3(X)=PhaseImbalance(X);

[0068] Where X = [Va[n], Vb[n], Vc[n], Ia[n], Ib[n], Ic[n]] is the input sequence, f1(X) is to maximize SNR, f2(X) is to minimize THD, f3(X) is to minimize phase imbalance, and PhaseImbalance() represents phase imbalance. The non-dominated sorting genetic algorithm II (NSGA-II) is used to solve the multi-objective optimization problem: min F(X) = [f1(X), f2(X), f3(X)]; stX∈Ω, where Ω is the feasible solution set; the best solution is selected from the Pareto optimal solution set P obtained by NSGA-II, and the optimal phase is determined: X* = argmin(w1f1(X)+w2f2(X)+w3f3(X)), X∈P; where w1, w2, w3 are weight coefficients, and argmin() represents the variable value that minimizes the objective function; the voltage and current sequences Vopt[n], Iopt[n] of the selected phase are output, that is, the optimal single-phase voltage and current sequences corresponding to X*.

[0069] An adaptive notch filter and a machine learning optimization algorithm are used to filter the fundamental component of a selected single-phase voltage and current sequence, resulting in a voltage and current sequence with the fundamental component removed. The adaptive notch filter can dynamically adjust its center frequency to accommodate small changes in the grid frequency. The machine learning algorithm is used to optimize the filter parameters and improve the filtering effect. Specifically:

[0070] Initialize the adaptive notch filter and set the center frequency to the nominal grid frequency (such as 50Hz or 60Hz). The filter transfer function is:

[0071] H(z)=(1-2cos(ω0)z -1 +z -2 ) / (1-2rcos(ω0)z -1 +r 2 z -2 );

[0072] Wherein, ω0 = 2πf0 / fs, f0 is the estimated grid frequency, fs is the sampling frequency, r is the bandwidth control parameter, and z represents a complex variable.

[0073] The actual grid frequency is dynamically estimated using the Recursive Least Squares (RLS) algorithm:

[0074] ε[n]=y[n]-φ[n] T θ[n-1];

[0075] θ[n]=θ[n-1]+K[n]ε[n];

[0076] K[n]=P[n-1]φ[n] / (λ+φ[n] TP[n-1]φ[n]);

[0077] P[n]=(IK[n]φ[n] T )P[n-1] / λ;

[0078] Where ε[n] is the error term, representing the difference between the current input signal and the predicted signal, y[n] is the current input signal, φ[n] is the regression vector, θ[n] is the current parameter vector, θ[n-1] is the parameter vector at the previous moment, λ is the forgetting factor, K[n] is the gain vector, P[n] is the current covariance matrix, P[n-1] is the covariance matrix at the previous moment, I represents the identity matrix, and n represents the time step. Based on the estimated frequency, the center frequency of the notch filter is adjusted. The notch filter is applied to filter the voltage and current series. The filter bandwidth and stopband attenuation are optimized using the support vector regression (SVR) model:

[0079] min(1 / 2)||w|| 2 +CΣ(ξi+ξi*);

[0080] styi-(w T xi+b)≤ε+ξi(w T xi+b)-yi≤ε+ξi*ξi, ξi*≥0;

[0081] Where w and b are SVR model parameters, ε is the accuracy parameter, C is the penalty factor, ξi and ξi* are slack variables, yi is the actual output value of the i-th sample, and xi is the input eigenvector of the i-th sample. The optimization process is iterated until the preset performance index is achieved, and the voltage and current sequences Vf[n] and If[n] are output after removing the fundamental component.

[0082] A hybrid algorithm of wavelet packet transform and empirical mode decomposition, combined with adaptive threshold technology, is used to extract multi-dimensional specific subharmonics from the filtered voltage and current series. This can effectively separate different frequency components and has good processing capabilities for non-stationary signals. Specifically:

[0083] Perform wavelet packet decomposition on the filtered signal to obtain coefficients of different frequency bands:

[0084] c j，k [n]=Σh[m-2n]c j-1，(k / 2) [m]; d j，k [n]=Σg[m-2n]c j-1，(k / 2) [m];

[0085] Among them, c j，k and d j，kare the approximate coefficient and detail coefficient of the kth node in the jth layer, respectively. h and g are low-pass and high-pass filters, respectively. m represents the index variable in the convolution operation, which is used to traverse the filter coefficients. n represents the time step. The coefficients of each frequency band are subjected to empirical mode decomposition (EMD) to obtain multiple intrinsic mode functions (IMFs): x(t) = Σc i (t)+r(t); where c i (t) is the intrinsic mode function and r(t) is the residual. The instantaneous frequency of each IMF is calculated using the Hilbert transform:

[0086] H[c i (t)]=(1 / π)∫(c i (τ) / (t-τ))dτ;f i (t) = (1 / 2π)(dθ i (t) / dt);

[0087] Among them, θ i (t) = arctan(H[c i (t)] / c i (t)) represents the instantaneous phase, H[c i (t)] represents the Hilbert transform of the eigenmode function, τ represents the integral variable, t represents the time point, d represents the differential operator, and f i (t) represents the instantaneous frequency, and arctan() represents the inverse tangent function. Based on the instantaneous frequency information, the IMF containing specific subharmonics is identified. The selected IMF is denoised using the adaptive soft threshold method:

[0088] η(x)=sign(x)(|x|-λ*)+λ*=σsqrt(2log(N));

[0089] Where σ is the noise standard deviation estimate, N is the signal length, x represents the input intrinsic mode function, sign() represents the sign function, and λ* represents the threshold. Reconstruct the specific subharmonic signal and output the extracted kth harmonic voltage and current sequence Vh,k[n],Ih,k[n].

[0090] In another embodiment of the present application, a high-precision voltage sensor is used to acquire real-time voltage signals from a three-phase AC power grid. An adaptive sampling rate algorithm is used to dynamically adjust the sampling frequency. Specifically, the sampling rate is initialized to a default value (e.g., 10kHz); the rate of change of the signal at the last N sampling points is calculated; if the rate of change exceeds a preset threshold, the sampling rate is increased; if it falls below the threshold, the sampling rate is decreased; the sampling rate adjustment range is limited to between 5kHz and 20kHz to balance accuracy and data volume; and the sampling rate is reassessed every 100ms.

[0091] The sampled analog signals are input to a 24-bit high-precision analog-to-digital converter to obtain discrete-time sequences ua(k), ub(k), uc(k), where k is the sampling index. These sequences are stored in a high-speed SRAM buffer.

[0092] Real-time current signals of the three-phase AC power grid are obtained using Hall effect current sensors. The same adaptive sampling rate algorithm described above is used. The sampled signals are converted to discrete-time sequences ia(k), ib(k), ic(k) by a 24-bit high-precision analog-to-digital converter, and these sequences are stored in a high-speed SRAM buffer.

[0093] The obtained discrete-time sequences ua(k), ub(k), uc(k) and ia(k), ib(k), ic(k) are read from the high-speed SRAM buffer. A newly developed multi-objective optimization algorithm is used to select the optimal single-phase, which is: calculate the signal-to-noise ratio (SNR) of each phase; evaluate the phase balance degree of each phase, calculate the deviation from the ideal 120° phase difference; use the Fast Fourier Transform (FFT) to calculate the total harmonic distortion (THD) of each phase; normalize the SNR, phase balance degree and THD to the range of 0-1; use the weighted sum method to calculate the comprehensive score of each phase, the weight can be adjusted according to the specific application; select the phase with the highest score as the optimal phase. Output the selected single-phase voltage sequence u(k) and single-phase current sequence i(k), and store them in the data processing buffer.

[0094] The obtained single-phase voltage sequence u(k) and single-phase current sequence i(k) are read from the data processing buffer. A newly developed adaptive notch filter is used, and the filter center frequency is initialized to the nominal grid frequency (such as 50Hz); use the phase-locked loop (PLL) technology to estimate the actual grid frequency in real time; according to the estimated frequency, dynamically adjust the center frequency of the filter; use the adaptive algorithm (such as LMS algorithm) to adjust the filter bandwidth to balance between filtering effect and transient response; apply the adjusted filter to the input signals u(k) and i(k); update the filter parameters every 10ms.

[0095] Use machine learning algorithms to dynamically optimize filter parameters: collect the filtering effect data of the last 1 second, including the signal energy ratio before and after filtering, phase response, etc.; input these data into the pre-trained neural network model; the neural network outputs the optimized filter parameters, such as Q factor, bandwidth, etc.; apply the optimized parameters to the filter; this optimization process is performed every 1 second. After processing, the voltage sequence u'(k) and current sequence i'(k) without the fundamental component are obtained, and they are stored in the data processing buffer.

[0096] The voltage sequence u'(k) and current sequence i'(k) are read from the data processing buffer. A newly developed hybrid algorithm combining wavelet packet transform and empirical mode decomposition is used. Specifically, the voltage sequence u'(k) and current sequence i'(k) are subjected to wavelet packet decomposition using a Daubechies wavelet basis with a decomposition order of 5. Each wavelet packet coefficient is subjected to empirical mode decomposition (EMD) to obtain a series of intrinsic mode functions (IMFs). The instantaneous frequency and amplitude of each IMF are calculated. The IMFs are grouped according to the instantaneous frequency, corresponding to harmonics of different orders. Within each harmonic group, the instantaneous amplitude and phase are calculated using the Hilbert transform.

[0097] Adaptive threshold technology is used to dynamically determine the accuracy of harmonic extraction: calculate the total energy of the signal; set the initial threshold to 1% of the total energy; traverse each harmonic from high to low, and accumulate the energy until the threshold is reached; if the accumulated harmonic number is too high (>20), increase the threshold; if it is too low (<5), lower the threshold; repeat the above steps until the appropriate number of harmonics to be extracted is found (usually 5-20). Get the multi-dimensional specific harmonic voltage component u sh (k) and the current component i sh (k). The voltage component u sh (k) and the current component i sh (k) Store in the data analysis buffer for use in subsequent steps.

[0098] This embodiment achieves high-precision, real-time, and adaptive data acquisition and preprocessing, improving the accuracy and reliability of subsequent analysis. High-precision voltage sensors and Hall-effect current sensors, combined with an adaptive sampling rate algorithm, enable precise acquisition of real-time voltage and current signals from a three-phase AC power grid. This adaptive sampling technology dynamically adjusts the sampling frequency, effectively addressing minor fluctuations in the grid frequency and ensuring that the sampled data maintains optimal quality. The use of a high-precision analog-to-digital converter further reduces quantization error during the digitization process, providing high-quality discrete-time series for subsequent analysis. A multi-objective optimization algorithm is introduced for intelligent phase selection, not only maximizing the signal-to-noise ratio and minimizing harmonic content, but also incorporating phase balance into the optimization objectives. This maximizes the reflection of the system's true state while ensuring data quality. The non-dominated sorting genetic algorithm II (NSGA-II) solves the multi-objective optimization problem, effectively balancing multiple competing objectives and outputting the optimal single-phase voltage and current series. An adaptive notch filter and a machine learning optimization algorithm are used to filter the fundamental wave of the selected series, dynamically adjusting the filter parameters to adapt to minor changes in the grid frequency. Machine learning algorithms (such as support vector regression) are used to optimize filter performance, improving filtering effectiveness. Combining a hybrid algorithm of wavelet packet transform and empirical mode decomposition, coupled with adaptive thresholding technology, enables multi-dimensional, specific subharmonic extraction from filtered sequences. This method not only effectively separates different frequency components but also handles non-stationary signals, improving the accuracy and robustness of harmonic extraction. This embodiment enhances data quality, laying a solid foundation for subsequent impedance detection, voltage fluctuation analysis, and stability assessment, and enhancing the reliability and accuracy of the entire analysis process.

[0099] like Figure 3 As shown, according to one aspect of the present application, step S2 is further:

[0100] S21. Based on the multi-dimensional harmonic components, a harmonic importance evaluation algorithm based on information entropy and dynamic programming is used to perform optimization selection and output the optimal harmonic voltage and current components;

[0101] S22. Adopting an adaptive piecewise Fourier transform algorithm and intelligent denoising technology based on deep learning, the optimal harmonic voltage and current components are converted into frequency domain and noise is suppressed to obtain a complex voltage and current spectrum.

[0102] S23. Based on the voltage and current spectra, a multi-model fusion algorithm of least squares method, total least squares method and Bayesian estimation is used, combined with a dynamic weight adjustment strategy, to calculate the complex impedance estimate;

[0103] S24. Based on the complex impedance estimate, frequency response analysis and equivalent circuit fitting using a genetic algorithm are used, combined with impedance characteristic classification using a support vector machine to perform impedance decomposition and characteristic analysis, and output equivalent circuit parameters, classification results, and preliminary line inductance estimates.

[0104] S25. Based on the equivalent circuit parameters and classification results, a multi-scale Kalman filter algorithm is used to smooth the preliminary line inductance estimation value, and a pre-configured fuzzy logic-based credibility assessment model is called to output the optimized line inductance estimation value and its credibility index.

[0105] In one embodiment of the present application, a harmonic importance evaluation algorithm based on information entropy and dynamic programming is used to optimize the selection of multi-dimensional harmonic components and output the optimal harmonic voltage and current components. The energy, frequency characteristics, and importance of the harmonics to the system are taken into account to ensure that the most representative and informative harmonic components are selected for subsequent analysis. Specifically:

[0106] Perform a preliminary analysis of the multi-dimensional harmonic components and calculate the energy of each harmonic component: E k =Σ(|Vh,k[n]| 2 +|Ih,k[n]| 2 ); where E k is the energy of the kth harmonic, Vh,k[n] and Ih,k[n] are the voltage and current sequences of the kth harmonic respectively. Calculate the information entropy of each harmonic component: H k =-Σ(p i *log(p i )); where p i is the normalized energy distribution of the kth harmonic in a certain time window. Define the harmonic importance index: I k =w1*(E k / E total )+w2*(H k / H total )+w3*f(k); where w1, w2, w3 are weight coefficients, E total and H total are the total energy and total information entropy of all harmonics, respectively, and f(k) is a function that takes into account the order of harmonics (e.g., giving higher weight to low-order harmonics). Use a dynamic programming algorithm to select the optimal harmonic subset: define the state S(i, j) as the maximum sum of the importance of the j among the first i harmonics. Recursive relation: S(i, j) = max(S(i-1, j), S(i-1, j-1) + I i ; Among them I i Represents the importance index of the i-th harmonic; boundary conditions: S(i,0)=0,; S(0,j)=0. Backtrack to obtain the optimal harmonic subset and output the selected harmonic voltage and current component Vhopt [n, k], Ih opt [n, k].

[0107] Adaptive piecewise Fourier transform algorithm and deep learning-based intelligent denoising technology are used to perform frequency domain conversion and noise suppression on the selected harmonic components, obtaining the voltage and current spectrum in complex form. This can effectively process non-stationary signals and minimize noise while retaining useful information. Specifically, the adaptive piecewise Fourier transform (STFT) is applied to the selected harmonic components: X(n, ω) = Σx[t]w[tn]e -jωt ; where w[tn] is a time-varying window function, the window length is adaptively adjusted according to the local characteristics of the signal, ω represents the frequency variable, j represents the imaginary unit, and x[] represents the value of the discrete time signal. Construct a denoising autoencoder based on a convolutional neural network (CNN): The encoder is: h = f(W e *X+b e ); the decoder is: X'=f(W d *h+b d ); where X is the STFT result, W e , W d is the convolution kernel, b e , b d is the bias, and f() is the activation function. Train the denoising autoencoder and use the mean square error (MSE) as the loss function: L = Σ(XX') 2 Apply the trained denoising autoencoder to the STFT result to get the denoised spectrum: X denoised =Denoiser(X), where Denoiser represents the denoising autoencoder function. Convert the denoised spectrum into a complex voltage and current spectrum: V(f) = |X denoised _ v |*exp(j*angle(X denoised_v )); I(f)=|X denoised _ i |*exp(j*angle(X denoised_i )), where exp() represents the exponential function, X denoised _ v represents the voltage spectrum after denoising, X denoised _ i represents the current spectrum after denoising, angle() represents the phase angle function, and f represents the frequency variable.

[0108] The multi-model fusion algorithm of least squares, total least squares and Bayesian estimation is used, combined with a dynamic weight adjustment strategy, to calculate the complex impedance estimate. This can combine the advantages of different estimation algorithms and improve the accuracy and robustness of impedance estimation. Specifically: Impedance is estimated using the least squares method (LS): Z LS =(V H *V) -1 *V H *I, where V and I are the frequency domain voltage and current vectors, and H represents the conjugate transpose. The total least squares (TLS) method is used to estimate the impedance: [U, S, V*] = svd([V, I]); Z TLS =-V(1:end-1,end) / V(end,end), where U represents the left singular matrix, S represents the singular value matrix, V* represents the right singular matrix, svd() represents the singular value decomposition function, and Z TLS Represents the impedance estimated by the total least squares method, and end represents the last element index of the matrix. Implement Bayesian estimation: Assume that the prior distribution p(Z) is Gaussian distribution and the posterior distribution p(Z|V, I)∝p(V, I|Z)*p(Z); use the Markov Chain Monte Carlo (MCMC) method to sample and obtain the impedance Z Bayes Dynamic weight fusion: Z = w1*Z LS +w2*Z TLS +w3*Z Bayes The weights w1, w2, and w3 are dynamically adjusted based on the estimation error of each method: wi = exp(-λ*εi) / Σexp(-λ*εj), where εi is the estimation error of the corresponding method, λ is the temperature parameter, and the fused complex impedance estimate Z is output.

[0109] Based on frequency response analysis and equivalent circuit fitting of genetic algorithm, combined with impedance characteristic classification of support vector machine, impedance decomposition and characteristic analysis are performed, and equivalent circuit parameters and classification results are output. This can deeply understand the physical characteristics of impedance and provide an important basis for subsequent power grid analysis. Specifically: Construct a candidate equivalent circuit model library, including various combinations such as RL series, RL parallel, RLC, etc. Use genetic algorithm to optimize equivalent circuit parameters: Fitness function: f = 1 / Σ|Z measured (ω)-Z model (ω)| 2 ; where Z measured represents the measured impedance, Z model represents the impedance of the model, ω represents the angular frequency, and the chromosome encoding is: [R1, L1, C1, R2, L2, C2, ...]; selection, crossover, mutation operations, and iterative optimization are performed. The optimal equivalent circuit model and parameters are selected. Impedance characteristics such as amplitude, phase angle, and resonant frequency are extracted: |Z| = sqrt(R 2+(ωL-1 / (ωC)) 2 ); θ = arctan((ωL-1 / (ωC)) / R), where |Z| represents the impedance amplitude, R represents the resistance, L represents the inductance, C represents the capacitance, and θ represents the phase angle of the impedance; use support vector machine (SVM) to classify impedance characteristics: train the SVM classifier: y = sign(Σ(αi*yi*K(xi,x)+b)), where K(xi,x) is the kernel function, such as the RBF kernel: K(xi,x) = exp(-γ||xi-x|| 2 ), αi represents the Lagrange multiplier, yi represents the label of the training sample, xi represents the feature vector of the training sample, x represents the feature vector of the sample to be classified, b represents the bias term, and γ represents the parameter of the kernel function. New samples are classified to obtain impedance characteristic categories, and the equivalent circuit parameters and impedance characteristic classification results are output.

[0110] A multi-scale Kalman filter algorithm is used to smooth the line inductance estimate, and a fuzzy logic-based credibility assessment model is used to output the optimized line inductance estimate and its credibility index. This can improve the accuracy and reliability of inductance estimation and provide more reliable parameters for power grid analysis. Specifically: Initialize the multi-scale Kalman filter: State equation: x (k+1) =F*x (k) +w (k) ; Observation equation: z (k) =H*x (k) +v (k) ; where x is the inductance value, z is the observation value, F is the state transfer matrix, H is the observation matrix, w and v are process noise and observation noise respectively. Perform multi-scale Kalman filtering: Prediction: x (k|k-1) =F*x (k-1|k-1) ;P (k|k-1) =F*P (k-1|k-1) *F'+Q; Update: K k =P (k|k-1) *H' / (H*P (k|k-1) *H'+R);x (k|k) =x (k|k-1) +K k *(z k -H*x (k|k-1) );P (k|k) =(IK k *H)*P (k|k-1); Where P is the covariance matrix, K is the Kalman gain, Q and R are the covariance matrices of process noise and observation noise respectively, F' represents the transposed matrix of the state transition matrix, and H' represents the transposed matrix of the observation matrix. Construct a fuzzy logic credibility evaluation system: Input variables: measurement noise level, estimated value change rate, historical data consistency; Fuzzy rules: IF-THEN rule set, such as: IF (low noise) AND (small change rate) AND (high consistency) THEN (high credibility). Fuzzy reasoning: Fuzzification: Map input to fuzzy set; Rule evaluation: Apply fuzzy rules; Aggregation: Merge rule outputs; Defuzzification: Obtain a clear credibility indicator. Output the optimized line inductance estimate L est and the corresponding credibility index C index .

[0111] In another embodiment of the present application, the multi-dimensional specific subharmonic voltage component u is read from the data analysis buffer. sh (k) and the current component i sh (k). Use the newly developed harmonic importance evaluation algorithm to calculate the information entropy of each harmonic component: normalize the harmonic component to the interval [0, 1]; divide the interval into N equal parts (such as N = 20), and calculate the probability distribution of the signal in each small area; use the Shannon entropy formula to calculate the information entropy. Calculate the signal-to-noise ratio (SNR) of each harmonic. Evaluate the stability of each harmonic: divide the signal into M segments (such as M = 10); calculate the average amplitude and phase of each segment; calculate the coefficient of variation between the M segments as a stability indicator. Combine the information entropy, SNR and stability to calculate the importance score of each harmonic.

[0112] The optimal harmonic combination is selected using a dynamic programming method: the objective function is set to maximize the sum of importance scores while limiting the number of harmonics to be selected; the state is defined as the set of selected harmonics, and the action is to select or not select a certain harmonic; starting from the lower harmonics, the optimal solution is constructed one by one: if selecting the current harmonic can improve the objective function value, it is selected; otherwise, the harmonic is not selected. The selection process ends when the preset number of harmonics is reached or the importance score no longer increases significantly. The optimal harmonic voltage component u is output. sh_opt (k) and the current component i sh_opt (k), and store it in the harmonic analysis buffer.

[0113] Read the optimized harmonic voltage component u from the harmonic analysis buffer sh_opt (k) and the current component i sh_opt(k). Using a newly developed adaptive piecewise Fourier transform algorithm, the window size is initialized to 1 / 4 of the signal length. The signal is subjected to a short-time Fourier transform (STFT). The time-frequency characteristics of the STFT results are analyzed: the spectral entropy at each time point is calculated. Time points with drastic changes in spectral entropy are identified as segmentation points. Based on the segmentation points, the signal is divided into several segments. A Hanning window of different sizes is applied to each signal segment, with the window size adaptively adjusted based on the frequency characteristics of the segment. The FFT is performed on each windowed signal segment.

[0114] Using intelligent denoising technology based on deep learning: The FFT results are fed into a pre-trained convolutional neural network (CNN). The CNN structure includes three convolutional layers, each followed by batch normalization and ReLU activation functions; two fully connected layers, the last of which uses a Sigmoid activation function. The CNN outputs a confidence score (between 0 and 1) for each frequency component; the original spectrum is multiplied by the confidence score to obtain the denoised spectrum. The processed signal is then converted to the frequency domain to obtain the complex voltage spectrum U. sh (f) and current spectrum I sh (f). The voltage spectrum U sh (f) and current spectrum I sh (f) Store in frequency domain data buffer.

[0115] Read the voltage spectrum U from the frequency domain data buffer sh (f) and current spectrum I sh (f) Three different impedance estimation methods are applied: a) Least Squares (LS) method: Construct an overdetermined system of equations AZ=B, where A contains the current spectrum and B contains the voltage spectrum. Use QR decomposition to solve Z=(A T A) -1 A T B. b) Total Least Squares (TLS): Construct the augmented matrix [AB]; perform singular value decomposition (SVD) on the augmented matrix; use the SVD results to calculate Z. c) Bayesian Estimation: Assume a Gaussian prior distribution; use the Markov Chain Monte Carlo (MCMC) method to estimate the posterior distribution; take the mean of the posterior distribution as the estimated value. Perform a weighted fusion of the results of the three methods: initialize the weights to an equal value (1 / 3); calculate the average deviation of each method's estimate from the other two methods; dynamically adjust the weights using the softmax function based on the deviation; and calculate the weighted average using the adjusted weights. Obtain the complex impedance estimate Z(f). Store the complex impedance estimate Z(f) in the impedance analysis buffer.

[0116] Read the complex impedance estimate Z(f) from the impedance analysis buffer. Perform frequency response analysis on the complex impedance estimate Z(f): Plot Bode and Nyquist plots; Identify key frequency points, such as cutoff frequency and resonant frequency. Use a genetic algorithm to fit an equivalent circuit model: Define a library of candidate models, including various combinations such as RL series, RC parallel, and RLC series-parallel. Use a genetic algorithm to optimize the parameters of each model to minimize the error relative to the actual impedance. Select the model with the smallest fitting error and moderate complexity as the final equivalent circuit. Based on the equivalent circuit model, extract the equivalent resistance R, equivalent inductance L*, and equivalent capacitance C (if any). Calculate the impedance modulus |Z| and the impedance angle θ.

[0117] Impedance classification is performed using machine learning. A feature vector is constructed, including the equivalent resistance R, equivalent inductance L*, equivalent capacitance C, impedance modulus |Z|, impedance angle θ, and their frequency derivatives. A support vector machine (SVM) is used for multi-class classification, classifying the impedance characteristics as inductive, capacitive, resistive, or mixed. The classification results and confidence levels are output. The equivalent resistance R, equivalent inductance L*, equivalent capacitance C, impedance modulus |Z|, impedance angle θ, and classification results are stored in a parameter feature buffer.

[0118] The equivalent inductance L* is read from the parameter characteristic buffer. L* is smoothed using a newly developed multi-scale Kalman filter algorithm. Specifically, three time-scale Kalman filters are constructed: a fast filter with a sampling period of 0.1 seconds to capture rapid changes; a medium filter with a sampling period of 1 second to smooth short-term fluctuations; and a slow filter with a sampling period of 10 seconds to track long-term trends. For each filter, the current state is predicted based on the previous state; the state estimate is updated based on the measured values. The outputs of the three filters are fused: the short-term variance of each filter output is calculated; and a weighted average is performed using the inverse of the variance as the weight.

[0119] A credibility assessment model based on fuzzy logic is used: input variables are defined: signal-to-noise ratio, consistency index, and historical data deviation. A fuzzy rule set is constructed. For example, if the signal-to-noise ratio is high, the consistency is good, and the historical deviation is small, the credibility is high; if the signal-to-noise ratio is low, the consistency is poor, or the historical deviation is large, the credibility is low. The credibility index is calculated using the Mamdani reasoning system. Defuzzification is performed using the center of gravity method to obtain the final credibility index CI (0-100). The optimized line inductance estimate L* is output. opt and its reliability index CI. The estimated value of line inductance L* opt The CI and its credibility index are stored in the result output buffer for use in subsequent steps.

[0120] This embodiment achieves high-precision, real-time impedance detection and line inductance identification, improving the accuracy and reliability of grid parameter estimation. A harmonic importance assessment algorithm based on information entropy and dynamic programming is adopted, introducing information theory into harmonic analysis. By calculating the information entropy and energy contribution of each harmonic component and combining it with a dynamic programming algorithm to optimize the selection process, intelligent screening of optimal harmonic voltage and current components is achieved. This not only improves the accuracy of subsequent impedance estimation but also reduces the computational effort, improving algorithm efficiency. By combining adaptive piecewise Fourier transform and deep learning-based intelligent denoising technology, frequency domain conversion and noise suppression are performed on the preferred harmonic components. The adaptive piecewise Fourier transform can dynamically adjust the window length based on the non-stationary characteristics of the signal, effectively improving the resolution and accuracy of spectrum analysis. Deep learning denoising models (such as convolutional neural network-based autoencoders) can learn complex noise patterns, achieve intelligent and efficient noise suppression, and improve the signal-to-noise ratio of frequency domain data. A multi-model fusion algorithm combining least squares, total least squares, and Bayesian estimation is proposed, combined with a dynamic weight adjustment strategy to achieve high-precision complex impedance estimation. This multi-model fusion approach leverages the strengths of various estimation algorithms, effectively addressing estimation deviations under different operating conditions through dynamic weight adjustment, and improving the robustness and accuracy of impedance estimation. Equivalent circuit fitting based on frequency response analysis and genetic algorithms, combined with impedance characteristic classification using a support vector machine, enables in-depth impedance decomposition and characteristic analysis. It not only accurately identifies equivalent circuit parameters but also intelligently classifies impedance characteristics, providing rich information support for subsequent power grid analysis and control. A multi-scale Kalman filter algorithm is used to smooth the line inductance estimate, and combined with a fuzzy logic-based credibility assessment model, highly reliable real-time line inductance estimation is achieved. Multi-scale Kalman filtering effectively handles disturbances of varying time scales, improving the stability of the estimate, while fuzzy logic credibility assessment provides a quantitative reliability indicator for the estimation result, providing an important basis for subsequent decision-making. This embodiment improves the accuracy and reliability of impedance detection and line inductance identification, provides an accurate parameter basis for power grid voltage characteristic analysis, and enhances the performance and applicability of the entire analysis system.

[0121] like Figure 4 As shown, according to one aspect of the present application, step S3 is further:

[0122] S31, obtaining a system short-circuit capacity parameter and an initial grid parameter, and dynamically updating the system short-circuit capacity parameter based on the initial grid parameter using a distributed sensor network and a recursive least squares method to obtain an updated system parameter;

[0123] S32. Calculate the new energy penetration rate based on the updated system parameters, use the exponentially weighted moving average and adaptive smoothing factor methods to process the time-varying characteristics of the new energy penetration rate, and obtain the optimized new energy penetration rate;

[0124] S33. Obtain real-time frequency measurement and temperature compensation, and calculate the line inductive reactance taking into account various factors based on the real-time frequency measurement, temperature compensation, and line inductance estimation;

[0125] S34. Based on the updated system parameters, the optimized renewable energy penetration rate, and the line inductance, the Monte Carlo method is used to calculate the grid voltage change and its probability distribution.

[0126] S35. Based on the statistical information of the grid voltage variation and in combination with Kalman filtering, calculate the point estimate, interval estimate, and voltage fluctuation rate of the real-time grid voltage;

[0127] S36, based on point estimation and interval estimation, uses pre-configured ARIMA model and wavelet analysis to perform time series analysis and multi-scale decomposition of voltage fluctuation rate, and outputs fluctuation trend and periodic characteristics;

[0128] S37. Based on the fluctuation trend and periodic characteristics, fuzzy logic and predictive analysis methods are used to make out-of-limit judgments; based on the judgment results, the out-of-limit risk level is evaluated and an early warning is issued.

[0129] According to one aspect of the present application, step S3 further includes:

[0130] S38. Based on the grid voltage variation, point estimation, interval estimation, and voltage fluctuation rate, linear regression, LSTM neural network, and FFT analysis are used to calculate the short-term change rate, predict future trends, and perform periodic analysis to obtain the analysis results.

[0131] S39. Based on the analysis results, data visualization technology and machine learning algorithms are used to generate a comprehensive analysis report containing risk assessment and action recommendations.

[0132] In one embodiment of the present application, the initial grid parameters are read from the system configuration database, and key parameters such as the system short-circuit capacity are dynamically updated using a distributed sensor network and recursive least squares method. This ensures that subsequent analysis is based on the latest system status, improving the accuracy and real-time nature of the analysis results. Specifically, the initial grid parameters, including rated voltage, rated frequency, initial short-circuit capacity, etc., are read from the system configuration database. A distributed sensor network is deployed to collect data such as voltage, current, and power in real time. The system short-circuit capacity is dynamically updated using the recursive least squares method (RLS):

[0133] θ(k)=θ(k-1)+K(k)[y(k)-φ T(k)θ(k-1)];

[0134] K(k)=P(k-1)φ(k)[λ+φ T (k)P(k-1)φ(k)] -1 ;

[0135] P(k)=[IK(k)φ T (k)]P(k-1) / λ;

[0136] Where θ(k) is the parameter vector of the current moment k (including short-circuit capacity), θ(k-1) is the parameter vector of the previous moment k-1, K(k) is the Kalman gain at the current moment, y(k) is the current measurement value, φ T (k) represents the transpose of the current regression vector, λ is the forgetting factor, P(k-1) is the covariance matrix at the previous moment, φ(k) represents the current regression vector, P(k) is the current covariance matrix, and I represents the identity matrix. Update other key parameters, such as system impedance and power factor. Store the updated parameters in a real-time database for subsequent analysis.

[0137] Based on the updated system parameters, the renewable energy penetration rate is calculated, and its time-varying characteristics are processed using the exponentially weighted moving average and adaptive smoothing factor methods. This can accurately reflect the proportion of renewable energy in the power grid and its dynamic changes, providing an important basis for subsequent analysis. Specifically: Calculate the instantaneous renewable energy penetration rate: PR(t) = P renewable (t) / P otal (t), where P renewable is the power generated by renewable energy, P total is the total power generation. Exponentially weighted moving average (EWMA) is used to process time-varying characteristics: EWMA(t) = α*PR(t) + (1-α)*EWMA(t-1), where α is the smoothing factor. Adaptive smoothing factor method is used: α(t) = |PR(t)-EWMA(t-1)| / (β+|PR(t)-EWMA(t-1)|), where β is the adjustment parameter that controls the adaptive speed. Calculate long-term trends and short-term fluctuations: Trend(t) = EWMA long (t); Fluctuation(t)=EWMA short (t)-EWMA long (t), where EWMA long and EWMA short Use larger and smaller smoothing factors respectively. Output the processed new energy penetration time series and its trend and fluctuation characteristics.

[0138] By combining the line inductance value identified online, real-time frequency measurement, and temperature compensation, the line inductance is calculated taking into account various factors. This can accurately reflect the real-time status of line parameters and provide reliable basic data for voltage fluctuation analysis. Specifically: Get the line inductance value L identified online est ; Use zero-crossing detection or phase-locked loop (PLL) technology to measure the system frequency f in real time; measure the line temperature T and perform temperature compensation: L(T) = L est *(1+α L *(TT ref )), where α L is the inductance temperature coefficient, T ref is the reference temperature. Calculate the line inductance: X L =2πfL(T); Considering the influence of skin effect and proximity effect: X L_eff =X L *k kin *k proximity， where k skin and k proximity are the correction coefficients for skin effect and proximity effect respectively. Output the line inductance X considering various factors. L_eff .

[0139] Using the updated system parameters, renewable energy penetration, and line inductance, combined with the Monte Carlo method, the grid voltage change and its probability distribution are calculated. This fully accounts for system uncertainty and provides statistical characteristics of voltage variation. Specifically, the probability distribution of input variables, such as load variation and renewable energy output, is defined. Multiple simulations are performed using the Monte Carlo method: for i = 1 to N, random input samples are generated; the voltage change is calculated as ΔV = f(system parameters, renewable energy penetration, line inductance, random input); and the results are recorded.

[0140] Statistical analysis of simulation results: Calculation of mean: μ ΔV =ΣΔV i / N, N is a natural number greater than 1, calculate the standard deviation: σ ΔV =sqrt(Σ(ΔV i -μ ΔV ) 2 / (N-1)); Fitting probability distribution: Use kernel density estimation or parameter fitting method to obtain the probability density function p(ΔV) of ΔV, the statistical characteristics of the output voltage change and the probability distribution function.

[0141] Based on the statistical information of voltage variation, combined with Kalman filtering, the point estimate and interval estimate of the real-time grid voltage are calculated. This can provide the optimal estimate of the voltage and its uncertainty range, providing a basis for grid operation decision-making. Specifically: Construct a state space model: x(k+1)=x(k)+ΔV(k)+w(k); z(k)=x(k)+v(k), where x is the voltage state, ΔV is the voltage variation, w and v are process noise and measurement noise respectively. Initialize the Kalman filter parameters. Execute Kalman filtering: Prediction: x (k|k-1) =x (k-1|k-1) +μ ΔV ;P (k|k-1) =P (k-1|k-1) +Q; Update: K k =P (k|k-1) / (P (k|k-1) +R);x (k|k) =x (k|k-1) +K k (z k -x (k|k-1) );P (k|k) =(1-K k )P (k|k-1) ; Calculate point estimate: V est =x (k|k)； Calculate interval estimate: V lower =V est -1.96*sqrt(P (k|k) );V upper =V est +1.96*sqrt(P (k|k) ); point estimate of output voltage V est and 95% confidence interval [V lower , V upper ].

[0142] Using the Autoregressive Integrated Moving Average (ARIMA) model and wavelet analysis, the voltage fluctuation rate is analyzed in time series and decomposed at multiple scales to output fluctuation trends and periodic characteristics. This can deeply reveal the time and frequency characteristics of voltage fluctuations, providing important reference for power grid operation and planning. Specifically: Calculate the voltage fluctuation rate time series: FR(t) = (V(t) - V avg ) / V avg

[0143] , where V(t) is the voltage at time t, V avg is the average voltage. Apply ARIMA model to time series analysis:

[0144] (1-φ1B-...-φ p B p )(1-B) d Xt =(1+θ1B+...+θ q B q )ε t ;

[0145] Where B is the lag operator, d is the difference order, φ and θ are model parameters, ε t is white noise.

[0146] Wavelet analysis is used for multiscale decomposition: W(a, b) = (1 / sqrt(a))∫FR(t)*ψ((tb) / a)dt, where ψ is the wavelet basis function, a is the scale parameter, and b is the shift parameter. Leveraging the long-term prediction results of the ARIMA model, spectral analysis of the wavelet coefficients identifies the main cyclical components, predicting the trend of output voltage fluctuations and the main cyclical characteristics.

[0147] Combining fuzzy logic and predictive analysis, voltage fluctuations are judged as exceeding limits, the risk level is assessed, and early warnings are issued. This allows for timely identification of potential voltage issues and provides decision support for grid operators. Specifically, fuzzy sets are defined: voltage deviation: {negatively large, negatively small, normal, positively small, positively large}; rate of change: {rapid drop, slow drop, stable, slow rise, rapid rise}; duration: {short, medium, long}. A fuzzy rule base is constructed: IF (voltage deviation IS positively large) AND (rate of change IS rapidly rising) AND (duration IS medium) THEN (risk IS high). Fuzzy reasoning: fuzzify input; apply fuzzy rules; aggregate output; and defuzzify. Use the ARIMA model to predict future voltage trends. Determine the risk level by combining current fuzzy reasoning results with the predicted trend. Generate early warning information of the corresponding level based on the risk level.

[0148] Using linear regression, LSTM neural network and FFT analysis, the voltage data is used to calculate the short-term rate of change, predict future trends and perform periodic analysis. This can fully characterize the voltage variation characteristics and provide multi-dimensional analysis results for power grid operation and planning. Specifically: Short-term rate of change calculation (linear regression): dV / dt = (ΣV i T i -nV "T") / (ΣT i 2 -nT" 2 ), where V i is the voltage sampling value, T i is the sampling time, V' is the average value of the voltage sampling value, and T' is the average value of the sampling time. Future trend prediction (LSTM neural network): Construct LSTM network:

[0149] f t =σ(W f ·[h t-1 , xt ]+b f );

[0150] i t =σ(W i ·[h t-1 , x t ]+b i );

[0151] C' t =tanh(W C ·[h t-1 , x t ]+b C );

[0152] C t =f t *C t-1 +i t *C't;

[0153] o t =σ(W o ·[h t-1 , x t ]+b o );

[0154] h t =o t *tanh(C t );

[0155] where f t Represents the output of the forget gate, i t represents the output of the input gate, C' t Represents the candidate memory cell state, C t Indicates the current memory unit state, o t represents the output of the output gate, h t represents the current hidden state, σ represents the activation function, W f Represents the weight matrix of the forget gate, h t-1 represents the hidden state at the previous moment, x t represents the input at the current moment, b f Represents the bias of the forget gate, W i represents the weight matrix of the input gate, b i Represents the bias of the input gate, tanh() represents the hyperbolic tangent activation function, W C The weight matrix representing the candidate memory cell state, b C represents the bias of the candidate memory cell state, C t-1 Indicates the state of the memory unit at the previous moment, W o represents the weight matrix of the output gate, b oRepresents the bias of the output gate; train the LSTM network and make predictions. Perform periodic analysis (FFT): X(k) = Σx(n)*e -j2πkn / N , k = 0, 1, ..., N-1, where x(n) is the time-domain voltage series and X(k) is the frequency-domain representation. Comprehensive analysis results: Calculate short-term rate-of-change statistical characteristics; generate a 24-hour voltage trend forecast; identify the main periodic components and their amplitudes; and output a comprehensive analysis report of voltage variation characteristics.

[0156] Integrate the results of various analyses, use data visualization technology and machine learning algorithms to generate a comprehensive analysis report that includes risk assessment and operational recommendations. Convert complex technical analysis results into intuitive and actionable decision support information. Specifically: collect the analysis results of each step above, use line charts to display voltage trends and forecasts; use heat maps to display the distribution of voltage fluctuation risks; use radar charts to represent system parameter changes. Use principal component analysis (PCA) to reduce dimensionality and extract the most representative indicators. Use the random forest algorithm for comprehensive risk assessment: construct a set of decision trees; each tree independently predicts the risk level; and obtain the final risk rating through a voting mechanism. Generate operational recommendations based on the rule base and machine learning model, fill the analysis results, visualization charts, risk assessments, and operational recommendations into the predefined report template, and output the final comprehensive analysis report.

[0157] In another embodiment of the present application, the following parameters are read from the system configuration database: Us: grid rated voltage (V); Ssc: system short-circuit capacity (VA); Sn: actual injection capacity of new energy units (VA); φ: power factor angle (rad). These parameters are dynamically updated using an adaptive parameter estimation algorithm: a distributed sensor network is deployed to collect voltage, current, and power data in real time; Ssc is estimated using recursive least squares (RLS): a state equation is constructed: Ssc(k) = Ssc(k-1) + w(k); a measurement equation: Us(k) = Us(k-1)*(1-Sn / Ssc(k)) + v(k), where w(k) and v(k) are process noise and measurement noise; the estimated value of the system short-circuit capacity Ssc is recursively updated. The power factor angle φ is updated using phase measurement unit (PMU) data. The parameter estimate is updated every 5 minutes. The updated parameters Us, Ssc, Sn, and φ are stored in the system parameter buffer.

[0158] Read the actual injection capacity Sn of the new energy unit and the system short-circuit capacity Ssc from the system parameter buffer. Calculate the new energy penetration rate k: the basic calculation is k = Sn / Ssc; consider the time-varying characteristics: use the exponentially weighted moving average (EWMA) to smooth the k value; k smooth (t) = α*k(t) + (1-α)*k smooth(t-1), where α is the smoothing factor, initially set to 0.1; adaptively adjust α: if |k(t)-k smooth (t-1)|>threshold, increase α; otherwise, decrease α; the range of α is limited to [0.05, 0.3], and threshold represents the preset threshold. smooth Store in the permeability buffer.

[0159] Read the line inductance estimate L* from the result output buffer opt , read the system frequency f from the system configuration database. Calculate the line inductance X: The basic calculation is X = 2πfL* opt ; Consider frequency changes: Use frequency measurement device to monitor f in real time; If |ff nominal |>0.1Hz, recalculate X, where f nominal Indicates the rated frequency of the system; Consider the temperature effect: use a temperature sensor to monitor the line temperature T; adjust X according to the temperature coefficient β: X adjusted =X*(1+β*(TT reference )), where T reference Indicates the reference temperature, which is used to calculate the effect of temperature on line inductance. adjusted Store in the line parameter buffer.

[0160] Read the grid rated voltage Us and power factor angle φ from the system parameter buffer, and read the smoothed k value k from the penetration buffer smooth , read the line inductance X from the line parameter buffer adjusted , calculate the voltage change ΔU: the basic formula is ΔU=Us*k smooth *(X adjusted *cos(φ)+R*sin(φ)) / sqrt(R 2 +X adjusted 2 ); where R is the resistance; consider the influence of resistance R: use adaptive estimation algorithm to estimate R / X ratio; if R / X>0.1, keep R; otherwise, ignore R; introduce probability distribution: assume k smooth , X adjusted , φ follows a normal distribution, and the Monte Carlo method is used to generate multiple sets of parameters. The probability distribution and confidence interval of the voltage change ΔU are calculated. The mean, standard deviation, and 95% confidence interval of the voltage change ΔU are stored in the voltage change buffer.

[0161] Read the grid rated voltage Us from the system parameter buffer and the statistical information of the voltage change ΔU from the voltage change buffer. Calculate the real-time grid voltage U: Point estimate: U=Us+ΔU mean , where ΔU meanRepresents the mean value of voltage change; interval estimation: [U lower， U upper ]=[Us+ΔU 5percentile , Us+ΔU 95percentile ], where ΔU 5percentile Indicates the 5th percentile of the voltage change, indicating that in 95% of cases, the voltage change will be greater than this value, ΔU 95percentile Indicates the 95th percentile of the voltage change, indicating that in 95% of cases, the voltage change will be less than this value; use Kalman filtering to smooth the grid voltage U: state equation: U(k)=U(k-1)+w(k); observation equation: U measured (k) = U(k) + v(k); recursively update the estimated value of voltage U. Store the point estimate and interval estimate of voltage U in the real-time voltage buffer.

[0162] The voltage U is read from the real-time voltage buffer, and the grid rated voltage Us is read from the system parameter buffer. The voltage fluctuation rate ΔU% is calculated using the basic formula ΔU% = (U - Us) / Us * 100%. Time series analysis is performed using the Autoregressive Integrated Moving Average (ARIMA) model to analyze the time series of the voltage fluctuation rate ΔU%, extracting trend and cyclical components. Wavelet decomposition is performed on the voltage fluctuation rate ΔU%, separating fluctuations at different time scales and analyzing the energy distribution of fluctuations at each scale. The current value, trend, cyclical periodicity, and multi-scale decomposition results of the voltage fluctuation rate ΔU% are stored in the fluctuation analysis buffer.

[0163] The voltage fluctuation rate ΔU% is read from the fluctuation rate analysis buffer, and the preset voltage fluctuation limit ΔUlimit% is read from the system configuration database. Voltage over-limit judgment is performed: the basic judgment is to compare |ΔU%| with ΔUlimit%. Fuzzy logic judgment uses input variables: ΔU%, ΔU% change rate, and duration. Fuzzy rules, such as "If ΔU% is close to the limit, the rate of change is large, and the duration is long, the over-limit risk is high," are used to output the over-limit risk level (low, medium, or high). Predictive judgment uses the ARIMA model to predict the short-term future ΔU%. If the predicted value exceeds ΔUlimit%, an alert is issued. The over-limit judgment result and risk level are stored in the over-limit analysis buffer.

[0164] Read historical voltage data U(t) and current voltage U from the real-time voltage buffer. Perform voltage fluctuation trend analysis: Calculate the short-term (5-minute) voltage rate of change using linear regression to calculate the slope; apply an adaptive threshold to detect mutation points. Perform time series forecasting: Use a long short-term memory (LSTM) network to predict voltage trends for the next hour; input features include historical voltage, load forecasts, weather data, etc.; regularly retrain the model with new data. Perform periodicity analysis: Use a fast Fourier transform (FFT) to analyze the spectrum of voltage data; identify major periodic components (such as daily and weekly cycles). Store the short-term rate of change, forecast trend, and periodicity analysis results in the trend analysis buffer.

[0165] Analysis results are read from each buffer to generate a comprehensive analysis report, summarizing voltage values, fluctuations, limit violations, and trend forecasts, and using data visualization techniques to generate charts. A comprehensive risk assessment is performed based on limit violations, trend forecasts, and system parameters, and risk classification is performed using a decision tree algorithm. Action recommendations are generated based on predefined rules and machine learning models, taking into account system constraints and economic considerations. A template engine is used to generate structured reports, supporting multiple output formats (PDF, HTML, etc.). Generated reports are stored in a report database and displayed through the user interface.

[0166] The embodiment realizes comprehensive, accurate and real-time power grid voltage fluctuation analysis, and improves the accuracy and timeliness of power grid operation state evaluation and prediction. The distributed sensor network and recursive least squares method are used to realize dynamic updating of key parameters such as system short-circuit capacity. Not only the real-time and accuracy of parameter estimation are improved, but also the dynamic changes of power grid topology and operation state are effectively dealt with, providing reliable basic data for subsequent analysis. The exponential weighted moving average and adaptive smoothing factor method are introduced to deal with the time-varying characteristics of new energy penetration rate, smooth the volatility of new energy output, and retain the trend information, providing more stable and reliable input for voltage fluctuation analysis. Combined with the online identification of line inductance value, real-time frequency measurement and temperature compensation, the line inductance considering various factors is calculated. The accuracy of line parameter estimation is improved, and more accurate basis is provided for voltage change calculation. The Monte Carlo method is used to combine the updated system parameters, new energy penetration rate and line inductance to realize the calculation of power grid voltage change and its probability distribution. Not only the uncertainty of the system is comprehensively considered, but also rich statistical information is provided, providing a reliable probability model for voltage evaluation and prediction. Based on the statistical information of voltage change, combined with Kalman filtering, point estimation and interval estimation of real-time power grid voltage are realized. Not only the optimal estimation value is provided, but also the uncertainty range of estimation is given, providing more comprehensive information support for power grid operation decision. Combined with ARIMA model and wavelet analysis, time series analysis and multi-scale decomposition are performed on voltage fluctuation rate. The long-term trend and short-term fluctuation characteristics of voltage fluctuation can be captured at the same time, providing in-depth insight for power grid planning and operation. Combined with fuzzy logic and predictive analysis, intelligent evaluation and early warning of voltage out-of-limit risk are realized. Not only potential voltage problems can be identified in time, but also quantitative evaluation of risk level is provided, providing strong decision support for operation personnel. Combined with linear regression, LSTM neural network and FFT analysis, short-term change rate calculation, future trend prediction and periodic analysis of voltage data are realized. The change characteristics of voltage are comprehensively described, providing rich analysis results for power grid operation and planning. By integrating various analysis results, data visualization technology and machine learning algorithm are used to generate a comprehensive analysis report containing risk evaluation and operation suggestions. The complex technical analysis results are converted into intuitive and operable decision support information, improving the practicality and understandability of the analysis results. The embodiment realizes comprehensive, accurate and real-time analysis of power grid voltage fluctuation, improves the ability of power grid operation state evaluation and prediction, and provides strong technical support for safe, stable and economic operation of power grid.

[0167] As Figure 5 shown, according to one aspect of the present application, step S4 further comprises:

[0168] S41, based on the pre-processed voltage and current signals, the line inductance estimation value and the power grid voltage variation, a system state space model considering multiple dynamic characteristics is constructed, the eigenvalues of the system state space model are solved by using a parallel QR algorithm; based on the eigenvalues, small signal stability is evaluated;

[0169] S42, based on the parameters of the system state space model, an implicit Runge-Kutta method with adaptive step size is used to perform time domain simulation for each pre-defined fault scenario; based on the results of the time domain simulation, a sliding window algorithm and a machine learning model are used to analyze transient stability;

[0170] S43, based on the small signal stability, a continuous power flow method and a singular value decomposition technique are used to calculate PV / QV curves and determine voltage collapse points; based on the voltage collapse points and the current operating point, voltage stability margin is calculated, and a predicted sequence of the voltage stability margin is obtained by using an artificial neural network prediction;

[0171] S44, based on the small signal stability, the transient stability and the predicted sequence, an analytic hierarchy process model is constructed, the weights of each stability index are calculated, based on the weights of the stability indexes, a fuzzy comprehensive evaluation method is used for comprehensive stability evaluation, and the results of the stability evaluation are obtained;

[0172] S45, based on the results of the stability evaluation, a pre-configured Bayesian network and a deep learning model are used for risk assessment and stability prediction, and risk assessment results and future stability prediction sequences are generated.

[0173] In an embodiment of the present application, a system state space model considering multiple dynamic characteristics is constructed, the eigenvalues are solved by using a parallel QR algorithm, and the dominant mode is identified by using a clustering algorithm to evaluate the small signal stability. In this way, the dynamic characteristics of the system under small disturbance can be revealed, which provides an important basis for stability evaluation. Specifically:

[0174] A state space model is constructed: dx / dt=Ax+Bu; y=Cx+Du; where x is a state vector, u is an input vector, y is an output vector, A, B, C, D are system matrices. Considering multiple dynamic characteristics such as synchronous generators, new energy interfaces, load characteristics, etc., the A matrix is expanded. The eigenvalues are solved by using a parallel QR algorithm: the A matrix is decomposed into an upper Hessenberg form: A=QHQ T , where QHQ is the product of the orthogonal matrix Q and the upper Hessenberg matrix H in QR decomposition, representing the decomposition form of the A matrix; QR iteration is performed in parallel: H k+1 =Q k T H k Q k , where Q krepresents the orthogonal matrix in QR decomposition, H k+1 represents the upper Hessenberg matrix after QR iteration, H k represents the upper Hessenberg matrix in the QR iteration; extract the diagonal elements as eigenvalues ​​λ i Apply K-means clustering algorithm to identify dominant patterns: represent eigenvalues ​​on the complex plane; perform K-means clustering: minΣΣ||λ j -μ i || 2 ; where μ i represents the cluster center in the K-means clustering algorithm, λ j Represent eigenvalues; identify the cluster containing the eigenvalue with the largest real part as the dominant mode. Assess small-signal stability: If the real part of all eigenvalues ​​is negative, the system is small-signal stable. Calculate the damping ratio: ζi = -Re(λi) / |λi|; determine whether the damping ratio of the critical mode meets the requirements (e.g., ζ>0.05). Output small-signal stability assessment results, including the dominant mode eigenvalue, damping ratio, and stability margin.

[0175] The implicit Runge-Kutta method with adaptive step size is used for time domain simulation, combined with sliding window algorithm and machine learning model to analyze transient stability and calculate critical fault clearing time. This can evaluate the dynamic behavior of the system under large disturbances and provide a safety margin for system operation. Specifically: establish a system differential algebraic equation (DAE) model: dx / dt = f(x, y, t); 0 = g(x, y, t), where x is the dynamic state variable, y is the algebraic variable, f() represents the differential equation of the dynamic state variable, and g() represents the algebraic equation; implement the implicit Runge-Kutta method with adaptive step size: x n+1 =x n +hΣb i k i ;k i =f(t n +c i h, x n +hΣa ij k j ,y n+1 ), where h is the step size, a ij 、b i 、c i is the Butcher table coefficient, k i Represents the intermediate variable in the Runge-Kutta method, indicating the state increment at the current step size, k j Represents the intermediate variable in the Runge-Kutta method, indicating the state increment at the previous moment; adaptive step size control: h n+1=h n *min(f max ,max(f min , (tol / err) 1 / p ), where tol is the tolerance, err is the estimation error, p is the method order, and f max Represents the maximum step size adjustment factor in adaptive step size control, f min Represents the minimum step size adjustment factor in adaptive step size control. Perform time-domain simulation: Set initial conditions and fault scenarios; Iteratively solve the DAE system; Record the time series of key variables (such as the generator rotor angle). Analyze transient stability using a sliding window algorithm: Define the sliding window size W and step size S; Calculate stability indicators (such as maximum angle deviation) for each window; If the indicator exceeds a threshold, it is considered unstable. Apply machine learning models (such as support vector machines (SVMs)) to predict stability: Feature extraction: Extract feature vectors from the time series; Train the SVM: min(1 / 2)||w|| 2 +CΣξi; styi(wφ(xi)+b)≥1-ξi, ξi≥0; use the trained SVM to quickly evaluate the stability of new scenarios. Calculate the critical fault clearing time (CCT): Use a binary search method to search for the CCT; perform simulation and stability judgment for each candidate clearing time; perform the analysis to a preset accuracy (e.g., 0.01s). Output transient stability analysis results, including the critical fault clearing time and stability assessment for key scenarios.

[0176] The continuation power flow method and singular value decomposition techniques are used to calculate the PV / QV curve and determine the voltage collapse point. Artificial neural networks are then used to predict short-term voltage stability margin trends. This allows for the assessment of system voltage stability and provides early warning information. The continuation power flow algorithm is implemented by modifying the standard power flow equation and introducing the load parameter λ: f(x, λ) = 0; the prediction step is [Δx, Δλ]. T =s*[dx / dλ,1] T Correction step, solve the augmented Jacobian equations. Calculate the PV / QV curve: gradually increase λ and solve the power flow; record the trajectory of the key bus voltage V as P or Q changes. Use singular value decomposition (SVD) to determine the voltage collapse point: perform SVD on the Jacobian matrix J: J = UΣV T ; Monitor the minimum singular value σ min , when σ min When it approaches zero, it is determined to be the voltage collapse point. Calculate the voltage stability margin: VSM=(λ critical -λ current ) / λ critical *100%, where λ critical represents the critical load parameter, λ currentRepresent current load parameters; construct an artificial neural network (ANN) to predict short-term voltage stability margin: Input layer: system characteristics (such as load level, generator output, etc.); Hidden layer: uses the ReLU activation function; Output layer: predicted VSM; Training algorithm: Backpropagation + Adam optimizer. Use the trained ANN to predict VSM trends for the next 24 hours. Output voltage stability analysis results, including PV / QV curves, current VSM, and predicted trends.

[0177] Based on the hierarchical analysis model and fuzzy comprehensive evaluation method, multi-criteria decision-making is carried out for various stability indicators, and Bayesian networks and deep learning models are used for risk assessment and stability prediction. Various stability analysis results are integrated to provide a comprehensive system stability assessment. Specifically: Construct a hierarchical analysis model (AHP): define the target layer: comprehensive system stability; the criterion layer: small signal stability, transient stability, voltage stability; the indicator layer: various specific indicators (such as damping ratio, CCT, VSM, etc.). Calculate weights: Construct the judgment matrix A: a* ij Indicates the importance of i relative to j; solve the eigenvalue equation: AW = λ max W; normalize the feature vector to obtain the weight vector W. Perform consistency test: CI = (λ max -n) / (n-1); CR = CI / RI < 0.1, where CI represents the consistency index, CR represents the consistency ratio, and RI represents the random consistency index. Applying fuzzy comprehensive evaluation: Establishing the factor set U* and the comment set V*, constructing the fuzzy relationship matrix R*; and calculating the comprehensive evaluation result: B = W·R*. Constructing a Bayesian network for risk assessment: Defining nodes as stability indicators and system states; Setting up a conditional probability table (CPT); and Using a variational inference algorithm to calculate posterior probabilities. Implementing a deep learning model for stability prediction: Using a long short-term memory network (LSTM) to capture time series features; Incorporating a feedforward neural network to process static features; The output layer provides stability prediction results. Model training and validation: Using historical datasets to train the deep learning model; K-fold cross-validation to evaluate model performance. Generating a comprehensive stability assessment report: Summarizing various stability indicators; Determining a comprehensive stability rating; Providing risk assessment results and forecast trends; and Generating visualization charts to display key information. Outputting a final comprehensive stability assessment report, including current stability status, potential risks, and future trend forecasts.

[0178] In another embodiment of the present application, the line inductance estimate value L*opt and resistance R are read from the result output buffer, and the system parameters are read from the system parameter buffer. Small signal stability analysis is performed: a system state space model is constructed, considering the dynamic characteristics of generators, excitation systems, governors, etc., and the model parameters are updated in real time using an adaptive identification algorithm. The system eigenvalues are calculated, the characteristic equation is solved using the QR algorithm, and parallel computation is used to improve efficiency. The eigenvalues are analyzed, the damping ratio and oscillation frequency are calculated, and the dominant mode is identified using a clustering algorithm. Stability margin evaluation is performed, the distance of the key eigenvalues to the imaginary axis is calculated, and the uncertainty of the stability margin is evaluated using a probabilistic method. The damping ratio, oscillation frequency, and stability margin are stored in the small signal analysis buffer.

[0179] The system dynamic model and fault scenario settings are read from the system configuration database. Transient stability analysis is performed: time domain simulation is performed, the implicit Runge-Kutta method is used to solve the differential algebraic equations, and the adaptive step size algorithm is used to improve computational efficiency; key parameter analysis is performed, the generator rotor angle, frequency, voltage, etc. are tracked, and the stability index is calculated in real time using a sliding window algorithm; the critical fault clearance time is calculated, the bisection method is used to search for the critical point, and a machine learning model (such as random forest) is used to predict the critical clearance time of different faults; transient stability margin evaluation is performed, the stability margin is calculated based on the energy function method, the system uncertainty is considered, and the probability distribution of the margin is evaluated using the Monte Carlo method. The critical fault clearance time and transient stability margin are stored in the transient analysis buffer.

[0180] The load power, generator output power, and line parameters are read from the system parameter buffer. Voltage stability analysis is performed: PV and QV curves are calculated using the continuation power flow method, and the prediction correction technique is used to improve computational efficiency; singular value decomposition (SVD) method is used to identify the collapse point, and fuzzy inference system is applied to evaluate the collapse risk; the maximum transferable power is determined based on the PV curve, and sensitivity analysis is used to evaluate the impact of different factors on the maximum transferable power; the L index is used to evaluate the node voltage stability margin, and an artificial neural network is used to predict the short-term voltage stability margin trend. The voltage stability margin and maximum transferable power are stored in the voltage stability buffer.

[0181] Read the analysis results from the small signal analysis buffer, transient analysis buffer, and voltage stability buffer. Conduct a comprehensive stability assessment: multi-criteria decision-making, establish a hierarchical analysis model, including small signal stability, transient stability, and voltage stability, and use the fuzzy comprehensive evaluation method to calculate the overall stability index; build a Bayesian network model to evaluate the probability of various instabilities, and use Monte Carlo simulation to evaluate system behavior under extreme conditions; develop a time series prediction model based on deep learning to predict stability trends in the short term (such as within 1 hour) in the future; use multi-dimensional data visualization technology to display stability analysis results, develop interactive dashboards, and support drill-down analysis. Generate a comprehensive system stability report, including overall stability evaluation, risk analysis, prediction results, and improvement suggestions. Store the report in the stability analysis database and display it through the user interface.

[0182] This embodiment achieves comprehensive, in-depth, and efficient system stability analysis, improving the accuracy, real-time performance, and reliability of grid stability assessment and prediction. In small-signal stability analysis, a system state-space model that considers multiple dynamic characteristics is constructed, and a parallel QR algorithm is used to efficiently solve eigenvalues. This not only improves the accuracy of the model, but also improves computational efficiency. By identifying dominant modes through a clustering algorithm, the analysis process is further simplified, making small-signal stability assessment of large-scale systems feasible and efficient. In transient stability analysis, an implicit Runge-Kutta method with adaptive step size, a sliding window algorithm, and a machine learning model are combined. The adaptive step size technology improves the efficiency and accuracy of time-domain simulation, the sliding window algorithm enables real-time analysis of simulation results, and machine learning models (such as support vector machines) accelerate the stability judgment and critical fault clearing time calculation process. This not only improves the accuracy of the analysis, but also reduces computational time, enabling rapid assessment of a large number of scenarios. In voltage stability analysis, the continuation power flow method and singular value decomposition technology are combined with artificial neural networks to predict short-term voltage stability margin trends. The continuous power flow method accurately tracks the PV / QV curve, while singular value decomposition technology provides a powerful tool for precisely locating voltage collapse points. The introduction of artificial neural networks enables rapid prediction of voltage stability margins, providing operators with timely early warning information. This not only improves the accuracy of voltage stability analysis but also effectively predicts future trends. The comprehensive stability assessment combines the analytic hierarchy process (AHP) model, the fuzzy comprehensive evaluation method, a Bayesian network, and a deep learning model. The AHP model and the fuzzy comprehensive evaluation method enable a scientific trade-off and comprehensive assessment of various stability indicators, while the Bayesian network provides a robust risk assessment framework. Deep learning models (such as LSTM networks) enable long-term predictions of system stability. This comprehensive approach not only comprehensively considers various stability factors but also provides reliable risk assessments and future trend forecasts, providing strong support for long-term planning and short-term operational decisions for the power grid. This embodiment achieves a comprehensive, in-depth, and efficient analysis of power grid system stability, improving the accuracy, real-time nature, and reliability of stability assessment and prediction. It not only provides a solid technical guarantee for the safe and stable operation of the power grid, but also provides an important decision-making basis for the formulation of power grid stability control strategies in the context of large-scale access to new energy, which is of great significance to improving the operating efficiency and reliability of the entire power system.

[0183] According to another aspect of the present application, a voltage characteristic analysis method considering line impedance in a new energy high penetration power grid includes the following steps:

[0184] Step S1, sampling: sampling the grid voltage to obtain the three-phase voltage u of a, b, and c pcca(k) 、u pccb(k) 、u pccc(k); Sampling the grid current to obtain the three-phase currents a, b, and c i a(k) 、i b(k) 、i c(k) ;

[0185] Step S2 / Impedance detection method: Since the parameters of the three-phase AC circuit on the grid side set in this paper are exactly the same, the specific subharmonic component u of the grid voltage and grid current can be obtained by selecting any one of the phases through the trap filter. pcc_sh(k) 、i sh(k) , and then calculate the value of L* according to the formula;

[0186] Step S3, grid voltage fluctuation analysis method: based on the line inductance L* calculated by the impedance detection method, update the voltage fluctuation of the high-penetration new energy grid in real time;

[0187] In one embodiment of the present application, Figure 6 As shown in the figure, N new energy units are connected to the grid through the grid-side inverter and line impedance. dc1 、V dc2 ,…,V dcN is the DC voltage of the grid-side inverter of each new energy unit, L f，1 , L f，2 ,…,L f，N is the filter inductance of the grid-side inverter of each new energy unit, U PCC，1 、U PCC，2 ,…,U PCC，N is the voltage amplitude of each new energy unit grid connection point, P1+jQ1, P2+jQ2, ..., P N +jQ N are the active power and reactive power generated by each new energy unit, i1, i2, ..., i N The output current of each new energy unit, R1, R2, ..., R N is the line resistance from the grid connection point of each new energy unit to the grid side, L*1, L*2, ..., L* N It is the line inductance from the grid connection point of each new energy unit to the grid side.

[0188] In actual engineering, the grid impedance Z is a dynamic quantity that changes with the connection of transformers and transmission lines, and its range of change is also uncertain, which will cause the grid voltage of the system to change. In order to ensure the stable operation of the power system, it is necessary to accurately measure and evaluate the line impedance, and maintain the electrical characteristics of the system through appropriate control measures. At present, most of the research on grid impedance detection uses ordinary harmonic injection methods, which will affect the grid current quality of the system to a certain extent. Combined with the analysis of the current status of the research on impedance online identification technology, this embodiment uses the method of the system's natural frequency subharmonics to realize the identification of the grid impedance. Since no interference is introduced into the control system, it will not affect the control effect of the current, and has good research significance. Because the embodiment is a grid connection point voltage u pcc And the grid current i is sampled, so the grid-side inductance L* needs to be measured, and the grid-connected point voltage u pcc and the grid-connected current i to identify L*, the following analysis is required: Since the inverter side filter L f The main purpose is to attenuate high-frequency signal components. Therefore, it can be considered that the low-order harmonic components in the inverter output current are approximately equal to the low-order harmonic components in the grid-connected current. Therefore, the low-order harmonic components in the sampled signal are considered to be used to identify the grid impedance. Considering that the inductive reactance value L* of the inductor is much larger than the impedance value R in actual work, the influence of the inductor parasitic resistance R can be ignored. The equivalent circuit diagram of the high-penetration new energy unit grid-connected is shown in the figure below. Figure 7 As shown, the value of L* can be estimated by the following formula:

[0189]

[0190] Where u pcc_sh is the specific low-frequency harmonic component of the grid-connected point voltage, i sh is the specific low-frequency harmonic component of the grid-connected current, ω res is the angular frequency of a specific sub-low-frequency harmonic. Since the parameters of the three-phase AC circuit on the grid side set in this embodiment are completely consistent, any one phase can be selected for estimation. When performing online impedance identification, it is necessary to extract the specific sub-harmonic components in the sampled signal. Among them, the notch filter is often used to extract certain specific frequency signals, so the notch filter can be used to process the sampled signal. Because the fundamental component in the actual sampled signal accounts for a large proportion, in order to reduce its influence on the specific sub-low-frequency component, it is considered to first attenuate the fundamental signal in the sampled signal, and then extract the signal of the specific sub-frequency. The specific extraction method is as follows: Figure 8 shown.

[0191] According to the single-phase grid connection point voltage and current u obtained by sampling pcc(k) ,i (k)The signal u′ after attenuating the fundamental frequency component is obtained by a 50Hz notch filter pcc(k) , i′ (k) , then u′ pcc(k) with i′ (k) The specific subharmonic components u of the grid-connected voltage and grid-connected current are obtained by subtracting the signals obtained by the specific subfrequency notch filter. pcc_sh(k) ,i sh(k) , where the notch filter expression is:

[0192]

[0193] Where, ζ z and p The damping ratios of the conjugate zeros and poles are 0.01 and 0.1 respectively, ω res is the specific subharmonic angular frequency. Based on the above analysis, the inductance L* online identification scheme diagram can be obtained, as shown in Figure 9 shown.

[0194] like Figure 10 As shown, the grid voltage fluctuation is analyzed through the grid-connected equivalent circuit diagram of a single new energy unit, U PCC ∠□ is the voltage at the grid connection point of the new energy unit; U S ∠0 is the grid bus voltage in the power system. In fact, the grid bus voltage will change due to various factors; Z=R+jX is the equivalent impedance of the line from the grid connection point of the new energy unit to the grid side; I is the line current from the grid connection point of the new energy unit to the grid side; P and Q are the active power and reactive power injected into the grid by the new energy unit, respectively. From the equivalent circuit diagram, we can see that the voltage at the grid connection point of the new energy unit and grid bus voltage The relationship between them is:

[0195] The line current I is used to represent the change in the voltage at the grid connection point of the new energy unit and the grid bus voltage. The grid bus voltage

[0196]

[0197] Where |Z| is the line impedance modulus, θ is the line impedance angle, and |I| is the line current modulus. is the power factor angle.

[0198] Let S = U 2 PCC / |Z| is the short-circuit capacity of the power system, then |Z|=U 2 PCC / S, the capacity of the new energy unit actually injected into the grid S nThe relationship with the injected current |I| is |I|=S n / U PCC , put it into the above formula, and we get:

[0199]

[0200] K=S n / S is the penetration rate of new energy. When the full capacity of high-penetration new energy is injected into the grid, the grid voltage changes the most. That is:

[0201]

[0202] Therefore, the L* obtained by combining the impedance online identification method described above can be used to detect the grid voltage fluctuation in real time.

[0203]

[0204] Substitute the calculated online identification L* into the above formula, and the grid voltage Expressed as:

[0205]

[0206] That is, when the high-penetration new energy load changes, the grid voltage fluctuation can be analyzed online.

[0207] This application integrates four key steps: data sampling and preprocessing, impedance detection and online identification of line inductance, grid voltage fluctuation analysis, and system stability analysis. This achieves a comprehensive, accurate, and real-time analysis of the voltage characteristics of the line impedance under a high-penetration new energy grid, thereby improving the ability to evaluate, predict, and control the stability of the grid's operating status.

[0208] During the data sampling and preprocessing phase, high-precision sensors, adaptive sampling algorithms, and advanced signal processing techniques were employed to provide high-quality, highly reliable basic data for subsequent analysis. A multi-objective optimization algorithm enabled intelligent phase selection, while a hybrid algorithm combining wavelet packet transform and empirical mode decomposition ensured the accuracy of harmonic extraction, improving data quality and laying a solid foundation for the entire analysis process.

[0209] In the impedance detection and line inductance online identification stages, high-precision, real-time impedance estimation and inductance identification are achieved by combining advanced technologies such as information entropy, dynamic programming, deep learning denoising, and multi-model fusion. This not only improves the accuracy of parameter estimation but also enhances the system's adaptability to dynamic grid changes, providing reliable parameter support for voltage characteristic analysis.

[0210] In terms of grid voltage fluctuation analysis, the application integrates distributed sensing networks, adaptive filtering, Monte Carlo simulation, time series analysis and other advanced technologies to achieve comprehensive and multi-dimensional analysis of voltage fluctuations. Not only can it accurately assess the current grid state, but also effectively predict future trends, providing comprehensive information support for grid operation decisions.

[0211] In system stability analysis, by combining parallel algorithms, machine learning, continuous power flow method and deep learning technology, comprehensive assessment of small signal stability, transient stability and voltage stability is achieved. This multi-angle and multi-method stability analysis not only improves the accuracy and efficiency of the assessment, but also provides a reliable theoretical basis for grid stability control in the context of large-scale new energy integration.

[0212] Through the close integration and data flow between each step, the application forms a closed-loop analysis system. From raw data collection to final stability assessment, each step is based on the output of the previous step and provides the necessary input for the subsequent step, achieving efficient use of information and continuity of analysis. Not only does it improve the accuracy and reliability of the overall analysis, but also enhances the real-time and adaptability of the analysis. The application can effectively and accurately analyze the voltage characteristics of new energy high-penetration power grids, providing comprehensive, accurate and real-time decision support for grid operators. It not only can timely identify potential voltage problems and stability risks, but also can provide targeted optimization suggestions, thereby improving the efficiency, safety and reliability of the grid operation. This has important practical significance and long-term strategic value for coping with the challenges brought by large-scale new energy integration and realizing the intelligent and low-carbon transformation of the grid.

[0213] The above describes the preferred embodiments of the application, but the application is not limited to the specific details of the above embodiments. Within the technical concept of the application, various equivalent transformations of the technical solutions of the application can be made, and these equivalent transformations all belong to the protection scope of the application.

Claims

1. A voltage characteristic analysis method considering line impedance in a high-penetration renewable energy grid is characterized by: The steps include: S1. Collect real-time voltage and current signals of a three-phase AC power grid, perform preprocessing and multi-dimensional predetermined subharmonic extraction on them, and obtain multi-dimensional harmonic components; S2. Based on the multi-dimensional harmonic components, a complex impedance estimation value is calculated; a pre-configured fuzzy logic-based credibility assessment model is called to obtain a line inductance estimation value; S3. Obtain the system short-circuit capacity parameters and, combined with the estimated line inductance, calculate the grid voltage change. Perform time series analysis and multi-scale decomposition to obtain fluctuation trends and periodic characteristics. Use fuzzy logic and predictive analysis methods to assess the risk level of exceeding the limit and issue early warnings; S4. Construct a system state space model and obtain stability indicators; A hierarchical analysis model is constructed, the weights of stability indicators are calculated, and a fuzzy comprehensive evaluation method is used to conduct a comprehensive stability assessment to obtain the results of the stability assessment.

2. The voltage characteristic analysis method considering line impedance in a new energy high penetration power grid according to claim 1 is characterized in that: Step S1 is further as follows: S11, obtaining a real-time voltage signal of a three-phase AC power grid through a high-precision voltage sensor, processing it using an adaptive sampling rate algorithm and a high-precision analog-to-digital converter to obtain a discrete-time voltage sequence; S12, collecting real-time current signals of the three-phase AC power grid through a Hall effect current sensor, and using an adaptive sampling rate algorithm and high-precision analog-to-digital conversion processing to obtain a discrete-time current sequence; S13, based on discrete time voltage sequence and discrete time current sequence, adopt multi-objective optimization algorithm to perform intelligent phase selection and output optimal single-phase voltage and current sequence; S14, using an adaptive notch filter and a machine learning optimization algorithm to filter the fundamental wave of the optimal single-phase voltage and current sequence to obtain a voltage and current sequence with the fundamental wave component removed; S15. Based on the voltage and current sequence with the fundamental component removed, a hybrid algorithm of wavelet packet transform and empirical mode decomposition is adopted, combined with adaptive threshold technology, to perform multi-dimensional predetermined subharmonic extraction to obtain multi-dimensional harmonic components.

3. The voltage characteristic analysis method considering line impedance in a new energy high penetration power grid according to claim 2 is characterized in that: Step S2 is further as follows: S21. Based on the multi-dimensional harmonic components, a harmonic importance evaluation algorithm based on information entropy and dynamic programming is used to perform optimization selection and output the optimal harmonic voltage and current components; S22. Adopting an adaptive piecewise Fourier transform algorithm and intelligent denoising technology based on deep learning, the optimal harmonic voltage and current components are converted into frequency domain and noise is suppressed to obtain a complex voltage and current spectrum. S23. Based on the voltage and current spectra, a multi-model fusion algorithm using least squares method, total least squares method and Bayesian estimation is used, combined with a dynamic weight adjustment strategy, to calculate the complex impedance estimate; S24. Based on the complex impedance estimate, frequency response analysis and equivalent circuit fitting using a genetic algorithm are used, combined with impedance characteristic classification using a support vector machine to perform impedance decomposition and characteristic analysis, and output equivalent circuit parameters, classification results, and preliminary line inductance estimates. S25. Based on the equivalent circuit parameters and classification results, a multi-scale Kalman filter algorithm is used to smooth the preliminary line inductance estimation value, and a pre-configured fuzzy logic-based credibility assessment model is called to output the optimized line inductance estimation value and its credibility index.

4. The voltage characteristic analysis method considering line impedance in a new energy high penetration power grid according to claim 3 is characterized in that: Step S3 is further as follows: S31, obtaining a system short-circuit capacity parameter and an initial grid parameter, and dynamically updating the system short-circuit capacity parameter based on the initial grid parameter using a distributed sensor network and a recursive least squares method to obtain an updated system parameter; S32. Calculate the new energy penetration rate based on the updated system parameters, use the exponentially weighted moving average and adaptive smoothing factor methods to process the time-varying characteristics of the new energy penetration rate, and obtain the optimized new energy penetration rate; S33. Obtain real-time frequency measurement and temperature compensation, and calculate the line inductive reactance taking into account various factors based on the real-time frequency measurement, temperature compensation, and line inductance estimation; S34. Based on the updated system parameters, the optimized renewable energy penetration rate, and the line inductance, the Monte Carlo method is used to calculate the grid voltage change and its probability distribution. S35. Based on the statistical information of the grid voltage variation and in combination with Kalman filtering, calculate the point estimate, interval estimate, and voltage fluctuation rate of the real-time grid voltage; S36, based on point estimation and interval estimation, uses pre-configured ARIMA model and wavelet analysis to perform time series analysis and multi-scale decomposition of voltage fluctuation rate, and outputs fluctuation trend and periodic characteristics; S37. Based on the fluctuation trend and periodic characteristics, fuzzy logic and predictive analysis methods are used to make out-of-limit judgments; based on the judgment results, the out-of-limit risk level is evaluated and an early warning is issued.

5. The voltage characteristic analysis method considering line impedance in a new energy high penetration power grid according to claim 4 is characterized in that: Step S4 is further as follows: S41. Based on the preprocessed voltage and current signals, the estimated line inductance, and the grid voltage variation, a system state space model considering various dynamic characteristics is constructed, and the eigenvalues ​​of the system state space model are solved using a parallel QR algorithm. Evaluate small signal stability based on eigenvalues; S42. Based on the parameters of the system state space model, an implicit Runge-Kutta method with adaptive step size is used to perform time domain simulation on each predefined fault scenario; Based on the results of time-domain simulation, a sliding window algorithm and a machine learning model are used to analyze transient stability. S43. Based on small signal stability, the continuation power flow method and singular value decomposition technique are used to calculate the PV / QV curve and determine the voltage collapse point; Based on the voltage collapse point and the current operating point, the voltage stability margin is calculated and the prediction sequence of the voltage stability margin is obtained using artificial neural network prediction; S44. Based on small signal stability, transient stability and prediction sequence, a hierarchical analysis model is constructed to calculate the weight of each stability index. Based on the weight of the stability index, a fuzzy comprehensive evaluation method is used to conduct a comprehensive stability assessment to obtain the stability assessment results. S45. Based on the results of the stability assessment, a pre-configured Bayesian network and deep learning model are used to perform risk assessment and stability prediction, generating risk assessment results and future stability prediction sequences.

6. The voltage characteristic analysis method considering line impedance in a new energy high penetration power grid according to claim 5 is characterized in that: Step S3 further includes: S38. Based on the grid voltage variation, point estimation, interval estimation, and voltage fluctuation rate, linear regression, LSTM neural network, and FFT analysis are used to calculate the short-term change rate, predict future trends, and perform periodic analysis to obtain the analysis results. S39. Based on the analysis results, data visualization technology and machine learning algorithms are used to generate a comprehensive analysis report containing risk assessment and action recommendations.

7. The voltage characteristic analysis method considering line impedance in a new energy high penetration power grid according to claim 5 is characterized in that: Step S13 is further as follows: S131. Perform fast Fourier transform based on the discrete-time voltage sequence and the discrete-time current sequence to obtain frequency spectrum information of the voltage and current signals; S132. Calculate the signal-to-noise ratio and total harmonic distortion of each phase based on the frequency spectrum information of the voltage and current signals; and evaluate the phase balance among the three phases based on the signal-to-noise ratio and total harmonic distortion of each phase. S133. Based on the phase balance degree, a multi-objective optimization problem is constructed and solved using a non-dominated sorting genetic algorithm to obtain a Pareto optimal solution set. S134. Based on the Pareto optimal solution set, determine the optimal single-phase voltage and current sequence.

8. The voltage characteristic analysis method considering line impedance in a new energy high penetration power grid according to claim 5 is characterized in that: Step S15 is further as follows: S151. Perform wavelet packet decomposition on the voltage and current sequence with the fundamental component removed to obtain coefficients of different frequency bands. S152. Performing empirical mode decomposition on the coefficients of each frequency band to obtain multiple eigenmode functions; calculating the instantaneous frequency of each eigenmode function using Hilbert transform; identifying an eigenmode function including predetermined subharmonics based on the instantaneous frequency information to obtain a selected eigenmode function; S153, denoising the selected intrinsic mode function using an adaptive soft threshold method to obtain a denoised intrinsic mode function; S154. Reconstruct the harmonic signal based on the denoised intrinsic mode function to obtain multi-dimensional harmonic components.

9. The voltage characteristic analysis method considering line impedance in a new energy high penetration power grid according to claim 5 is characterized in that: Step S14 is further as follows: S141. Based on the optimal single-phase voltage and current sequence, a recursive least squares algorithm is used to dynamically estimate the actual grid frequency. S142, initializing the notch filter, and adjusting the center frequency of the notch filter based on the estimated actual grid frequency to obtain an adjusted notch filter; S143, calling a preconfigured support vector regression model to optimize the adjusted notch filter to obtain an optimized notch filter, repeating step S143 until a preset performance indicator is achieved to obtain a final optimized notch filter; S144. Use the finally optimized notch filter to filter out the fundamental wave of the optimal single-phase voltage and current sequence to obtain a voltage and current sequence with the fundamental wave component removed.

Images