New energy station broadband oscillation risk early warning method based on voltage dynamic indication

Abstract

The invention relates to the technical field of power system safety and stability control, and discloses a new energy field station broadband oscillation risk early warning method based on voltage dynamic indications, which comprises the following steps: acquiring instantaneous value data in a new energy electric field station to obtain electric quantity time sequence data; processing the electrical quantity time sequence data according to the target sliding window and outputting risk characteristic quantity; performing nonlinear normalization processing on the risk characteristic quantity by using a weighted fusion technology, and generating a single broadband oscillation risk index reflecting the broadband oscillation level of the new energy station according to a normalization processing result; and carrying out difference comparison on the single broadband oscillation risk index and a dynamic risk early warning threshold, and triggering an oscillation risk early warning signal and implementing oscillation risk early warning when a difference result meets a target warning requirement. According to the method, the defense gateway is moved forward by capturing the dynamic symptoms before oscillation occurs, so that the fundamental conversion from post-event to beforehand is realized, and a precious decision window period is won for operation adjustment.

Description

Technical Field

[0001] This invention relates to the field of power system safety and stability control technology, and more specifically, to a method for early warning of broadband oscillation risks in new energy power plants based on dynamic voltage indicators. Background Technology

[0002] With the rapid increase in the penetration rate of new energy sources such as wind power and photovoltaics in the power system, new energy power generation units are mainly connected to the power grid through power electronic converters. Their large-scale application has profoundly changed the dynamic characteristics of the power system, causing it to evolve from a traditional high-inertia, high-damping system dominated by synchronous generators to a complex dynamic system dominated by converters with low inertia, weak damping, and strong control coupling. The traditional power grid dominated by synchronous generators exhibits a clear electromechanical oscillation mode (0.1-2Hz), and its analysis theory and suppression measures are quite mature.

[0003] However, modern power systems, due to the electromagnetic interaction between the converter's complex nonlinear control system (such as phase-locked loop, inner and outer loop control, etc.) and the power grid over a wide frequency range, have given rise to broadband oscillation phenomena with a wider frequency range (from subsynchronous oscillations of a few hertz to medium-frequency oscillations of several hundred hertz, and even high-frequency oscillations of several thousand hertz) and more hidden and complex mechanisms.

[0004] Once broadband oscillations occur, their consequences are often more severe than those of traditional low-frequency oscillations. They can lead to instability in converter control systems due to signal resonance, damage to key power electronic devices (such as IGBTs) due to overvoltage or overcurrent, large-scale grid disconnection of new energy units, and in severe cases, even trigger DC blocking, threatening the safe and stable operation of the entire power grid. Most existing monitoring and defense measures for broadband oscillations are post-event response mechanisms. These methods usually rely on online identification of the frequency and amplitude of the oscillation signal that has occurred. When the oscillation amplitude exceeds the preset protection setting, a crude emergency control measure of disconnecting the station or unit is taken.

[0005] The fundamental flaw of this type of method lies in its lag; it cannot effectively intervene in the early stages of oscillation. When a significant oscillation amplitude is detected, the system may already be on the verge of instability. Control measures taken at this time are often costly and have limited effectiveness. This is especially true in large-scale renewable energy concentrated access areas such as desert areas, where multiple renewable energy power plants are connected to the same hub substation through convergence lines. The electrical coupling and control interaction between the power plants are extremely close, and the impact range of broadband oscillations is wider and the propagation speed is faster. If the traditional amplitude-based power-off method is still used, it is very easy to cause a serious accident in which the oscillation of one power plant leads to the cascading disconnection of multiple surrounding power plants, causing a severe power impact on the power grid.

[0006] Based on research, practice, and simulation analysis, many broadband oscillations (especially those related to control system stability) undergo a gradual decrease in system stability margin before developing into large-amplitude, persistent oscillations. During this period, although the oscillation amplitude is not yet significant, the system's dynamic response characteristics to small disturbances begin to deteriorate. As described in existing research, with the increase in renewable energy power generation, the system's voltage support capacity decreases, and voltage stability issues become prominent. When renewable energy power generation approaches the static voltage stability limit, even small changes in active and reactive power will cause large fluctuations in the AC voltage at the collection point, thereby increasing the risk of instability.

[0007] This clearly indicates a strong correlation between voltage stability level and oscillation risk. However, most existing studies focus on the assessment of static voltage stability margin or independently analyze broadband oscillation mechanisms, and generally lack in-depth research and engineering applications on the key issue of how voltage stability level quantitatively affects broadband oscillation risk.

[0008] Therefore, the power system field urgently needs a technology that can transform passive to proactive measures and achieve early warning of oscillations, either before or near the event. By capturing weak dynamic signs in the evolution of an unstable state from a stable vector before an oscillation occurs, valuable decision-making windows can be provided for operators to take more refined preventive control measures, such as adjusting control parameters, limiting output, and initiating additional damping control. This will suppress oscillations in their infancy and achieve proactive defense for grid security.

[0009] No effective solutions have yet been proposed to address the problems in the relevant technologies. Summary of the Invention

[0010] To address the problems in related technologies, this invention proposes a broadband oscillation risk early warning method for new energy power plants based on dynamic voltage indicators, in order to overcome the aforementioned technical problems existing in existing related technologies.

[0011] Therefore, the specific technical solution adopted by the present invention is as follows:

[0012] In a first aspect, this invention proposes a method for early warning of broadband oscillation risks in new energy power plants based on dynamic voltage indicators, the method comprising:

[0013] Instantaneous data from various monitoring points within the new energy power plant are collected to obtain electrical quantity time series data. The electrical quantity time series data is then processed according to the target sliding window to output risk characteristic quantities.

[0014] The risk characteristics are nonlinearly normalized using weighted fusion technology, and a single broadband oscillation risk index reflecting the broadband oscillation level of new energy power plants is generated based on the normalization results.

[0015] The difference between the single broadband oscillation risk index and the dynamic risk warning threshold is compared, and when the difference reaches the target alarm requirement, the oscillation risk warning signal is triggered and the oscillation risk warning is implemented.

[0016] Preferably, instantaneous value data from each monitoring point within the new energy power station are collected to obtain electrical quantity time series data. This electrical quantity time series data is then processed according to a target sliding window to output risk characteristic quantities, including:

[0017] Instantaneous values ​​of voltage, active power, and reactive power at various monitoring points within the new energy power station are collected, and the collected data are reconstructed and coupled to obtain electrical quantity time series data.

[0018] Bandpass filtering is performed on the time series data of electrical quantities within the target sliding window. Based on the processing results, the small disturbance components in the operation process of new energy power stations are extracted, and a linear regression model is constructed.

[0019] The least squares method is reconstructed using regular weighting techniques, and the regression coefficients of the linear regression model are solved based on the reconstruction results, which are used as estimates of the dynamic sensitivity of active voltage and reactive voltage.

[0020] The dynamic voltage deviation rate and voltage fluctuation spectrum entropy are calculated based on the voltage amplitude and Shannon entropy technique, and combined with the dynamic sensitivity of active voltage and reactive voltage as risk characteristic quantities.

[0021] Preferably, instantaneous values ​​of voltage, active power, and reactive power are collected from various monitoring points within the new energy power station, and the collected data are reconstructed and coupled for characterization to obtain electrical quantity time series data, including:

[0022] After collecting instantaneous data of voltage, active power, and reactive power at the grid connection point, feeder, and busbar of the new energy power station using a measuring instrument at the sampling frequency, the data are standardized.

[0023] Instantaneous state points are constructed based on the standardized processing results, and the instantaneous state points are used for embedding processing to generate state vectors in the voltage-power coupled phase space to characterize the dynamic potential of the new energy power station.

[0024] Randomly select data points in the voltage-power coupling phase space and find several nearest points of the data points. Combine the data points with the nearest points to generate a local neighborhood and perform component analysis on the local neighborhood.

[0025] Based on the primary and secondary directions of local neighborhood emission from the component analysis results, and by calculating the state vector describing the dynamic characteristics of the new energy power station using the orthogonal divergence rate of the state trajectory, time series data of electrical quantities are obtained.

[0026] Preferably, the least squares method is reconstructed using regularized weighting techniques, and the regression coefficients of the linear regression model are solved based on the reconstruction results. These coefficients serve as estimates of the dynamic sensitivity of active power voltage and reactive power voltage.

[0027] Based on the processed electrical quantity time series data, a coherence function is introduced to quantitatively evaluate the degree of linear correlation between active power, reactive power and voltage at each frequency point, and a confidence weight function is constructed based on the degree of linear correlation and frequency confidence weight.

[0028] The linear regression model is optimized based on the confidence weight function. The composite coefficient is used as a strong physical constraint to reveal the inherent integral constraint relationship between the real and imaginary parts of the response function of the linear time-invariant system, and the inherent integral constraint relationship is used as a constraint condition.

[0029] Based on the constraints, a regularization term with composite coefficients as the core is generated. The regularization term is used to fundamentally reconstruct the least squares objective function, and the linear regression model is solved based on the reconstructed least squares objective function.

[0030] The regression coefficients of the linear regression model are output based on the solution results, and the regression coefficients are used as estimates of the dynamic sensitivity of active voltage and reactive voltage.

[0031] Preferably, the calculation of dynamic voltage deviation rate and voltage fluctuation spectral entropy based on voltage amplitude and Shannon entropy technique includes:

[0032] Calculate the average short-time voltage amplitude and the average long-time voltage amplitude within the target time window, and generate a deviation time series by obtaining the instantaneous voltage deviation based on the average short-time voltage amplitude and the average long-time voltage amplitude.

[0033] The first-order difference approximation technique is used to numerically differentiate the deviation time series. The dynamic voltage deviation rate is obtained from the numerical differentiation result, which characterizes the rate at which the voltage operating point deviates from its short-term stable trajectory.

[0034] The voltage signal is subjected to a fast Fourier transform to obtain the amplitude spectrum of the voltage signal over a wide frequency range, and the amplitude spectrum is divided into several continuous sub-bands according to a preset bandwidth.

[0035] The proportion of spectral energy within a sub-band to the total spectral energy within a window is calculated to generate a probability distribution sequence. The Shannon entropy technique is then used to process the probability distribution sequence to output the spectral entropy of voltage fluctuations.

[0036] Preferably, a weighted fusion technique is used to perform nonlinear normalization on the risk characteristics, and a single broadband oscillation risk index reflecting the broadband oscillation level of new energy power plants is generated based on the normalization result, including:

[0037] The original spectral features are expanded, and the estimated values ​​of modal damping ratio and modal energy are introduced to generate new feature vectors. The new feature vectors are then fed into the online Dirichlet process hybrid model.

[0038] The posterior probability of a new feature vector belonging to all known pattern clusters is calculated using an online Dirichlet process mixture model. Pattern clusters are created based on the posterior probability, and the new feature vector is used as a new pattern sample.

[0039] Based on the pattern cluster and posterior probability, weight values ​​based on pattern evolution are generated, and the risk feature quantities are normalized. The normalized risk feature values ​​and weight values ​​are weighted and summed to obtain a comprehensive risk index, thus generating a single broadband oscillation risk index that reflects the broadband oscillation level of new energy power plants.

[0040] Preferably, the dynamic risk warning threshold is obtained by fitting the probability density distribution of the comprehensive risk index calculated from the historical steady-state operation data of the new energy power station, and selecting the upper limit of the confidence interval below the target probability based on the fitting results.

[0041] Secondly, this invention also proposes a broadband oscillation risk early warning system for new energy power plants based on dynamic voltage indicators. This system includes:

[0042] The data acquisition module is used to collect instantaneous data from various monitoring points within the new energy power plant to obtain time series data of electrical quantities.

[0043] The indicator calculation module is used to process electrical quantity time series data according to the target sliding window and output risk characteristic quantities;

[0044] The risk fusion module is used to perform nonlinear normalization processing on risk characteristics using weighted fusion technology, and generate a single broadband oscillation risk index that reflects the broadband oscillation level of new energy power plants based on the normalization results.

[0045] The early warning decision module is used to compare the difference between a single broadband oscillation risk index and a dynamic risk early warning threshold, and to trigger an oscillation risk early warning signal when the difference result reaches the target alarm requirement, thereby implementing oscillation risk early warning.

[0046] Thirdly, the present invention also proposes an electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program, when executed by the processor, implements the above-described method.

[0047] Fourthly, the present invention also provides a computer-readable storage medium on which a computer program is stored, the computer program implementing the above-described method when executed by a processor.

[0048] The beneficial effects of this invention are as follows:

[0049] 1. This invention captures dynamic signs before oscillations occur, shifting the defense line forward and achieving a fundamental shift from reactive to proactive measures. This provides a valuable decision-making window for preventative control and operational adjustments. Furthermore, by constructing a multi-dimensional and complementary voltage dynamic indicator system and employing a weighted fusion algorithm, it effectively overcomes the problem of misjudgment or omission of single indicators, comprehensively improving the overall performance of early warning.

[0050] 2. This invention supports refined control, avoiding large-scale power outages. Early risk warnings enable the system to adopt more refined control methods such as adjusting converter control parameters and calling energy storage to provide virtual damping, avoiding the one-size-fits-all large-scale power outages caused by oscillation instability. This is of great significance for ensuring the maximum absorption of new energy and grid security. At the same time, the risk indicator library proposed in this invention is configurable and can be tailored or expanded according to the on-site computing capabilities and specific needs. The weights can also be optimized based on offline analysis, which has strong engineering adaptability. Attached Figure Description

[0051] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0052] Figure 1 This is a flowchart of a method for early warning of broadband oscillation risk in new energy power plants based on voltage dynamic indicators according to an embodiment of the present invention;

[0053] Figure 2 This is a schematic diagram of a broadband oscillation risk early warning system for new energy power plants based on voltage dynamic indicators according to an embodiment of the present invention.

[0054] Figure 3 This is a schematic diagram of the hardware operating environment involved in the embodiments of the present invention;

[0055] Figure 4 This is an internal block diagram of the risk indicator calculation process in a broadband oscillation risk early warning method for new energy power plants based on voltage dynamic indicators according to an embodiment of the present invention;

[0056] Figure 5 This is a schematic diagram illustrating the theoretical basis of dynamic-voltage dynamic sensitivity in a broadband oscillation risk early warning method for new energy power plants based on voltage dynamic indicators according to an embodiment of the present invention.

[0057] Figure 6This is a schematic diagram of the voltage fluctuation spectrum entropy calculation process in a broadband oscillation risk early warning method for new energy power plants based on voltage dynamic indicators according to an embodiment of the present invention.

[0058] Figure 7 This is a schematic diagram illustrating the generation of a comprehensive risk index through multi-indicator weighted fusion in a method for early warning of broadband oscillation risk in new energy power plants based on voltage dynamic indicators, according to an embodiment of the present invention.

[0059] Figure 8 This is a schematic diagram of the comparison and decision-making between the dynamic risk warning threshold and the real-time risk index in a broadband oscillation risk warning method for new energy power plants based on voltage dynamic indicators according to an embodiment of the present invention.

[0060] Figure 9 This is a comparison chart of the comprehensive risk factor early warning capabilities in a broadband oscillation risk early warning method for new energy power plants based on voltage dynamic indicators according to an embodiment of the present invention. Detailed Implementation

[0061] To further illustrate the various embodiments, the present invention provides accompanying drawings, which are part of the disclosure of the present invention. These drawings are mainly used to illustrate the embodiments and can be used in conjunction with the relevant descriptions in the specification to explain the operating principles of the embodiments. With reference to these drawings, those skilled in the art should be able to understand other possible implementation methods and the advantages of the present invention.

[0062] According to an embodiment of the present invention, a method for early warning of broadband oscillation risk in new energy power plants based on dynamic voltage indicators is provided.

[0063] The present invention will now be further described in conjunction with the accompanying drawings and specific embodiments, such as... Figure 1 As shown, according to an embodiment of the present invention, a broadband oscillation risk early warning method for new energy power plants based on dynamic voltage indicators includes:

[0064] Step S1: Collect instantaneous data from each monitoring point within the new energy power station to obtain electrical quantity time series data, and process the electrical quantity time series data according to the target sliding window to output risk characteristic quantities;

[0065] Step S2: The risk characteristic quantity is nonlinearly normalized using weighted fusion technology, and a single broadband oscillation risk index reflecting the broadband oscillation level of new energy power stations is generated based on the normalization result.

[0066] Step S3: Compare the difference between the single broadband oscillation risk index and the dynamic risk warning threshold, and trigger the oscillation risk warning signal when the difference result reaches the target alarm requirement, and implement the oscillation risk warning.

[0067] Specifically, using a wideband measurement device (WAMS) or a synchronous phasor measurement unit (PMU) with high-frequency data stream capability, instantaneous values ​​of three-phase voltage, three-phase current, active power, and reactive power at the grid connection point, feeder, or key busbar of the new energy power station are collected synchronously at a sampling frequency of not less than 10kHz, forming a multi-channel, high-resolution electrical quantity time series dataset.

[0068] Within a preset sliding time window with millisecond-level steps, the collected data is processed, and at least two types of risk indicators with voltage dynamic characteristics as the core are calculated in parallel. The set of risk indicators preferentially includes: active power-voltage dynamic sensitivity and reactive power-voltage dynamic sensitivity, and may selectively further include: dynamic voltage deviation rate and voltage fluctuation spectrum entropy.

[0069] A multi-index adaptive weighted fusion algorithm is adopted to perform nonlinear normalization on multiple voltage dynamic risk indicators, and perform weighted summation or more complex nonlinear mapping based on preset or online adaptive weight coefficients to generate a comprehensive single broadband oscillation risk index that can fully reflect the system oscillation risk level.

[0070] Simultaneously, the generated comprehensive broadband oscillation risk index is compared in real time with a dynamic risk warning threshold that can learn itself based on historical power grid operation data and adaptively adjust according to the current operating conditions. When the risk index continuously exceeds the warning threshold for a preset alarm duration, a high-confidence broadband oscillation risk warning signal is triggered.

[0071] The calculation methods for active-voltage dynamic sensitivity and reactive-voltage dynamic sensitivity include: bandpass filtering of voltage, active power and reactive power data sequences within a sliding time window to extract small disturbance components, and constructing a multiple linear regression model with voltage disturbance as the dependent variable and active and reactive power disturbances as independent variables; solving the regression coefficients of the model using the least squares method, and the obtained coefficients are used as real-time estimates of active-voltage dynamic sensitivity (dv / dp) and reactive-voltage dynamic sensitivity (dv / dq), respectively.

[0072] Meanwhile, the multiple linear regression model is based on the total differential theory of the voltage-power relationship in power systems. It assumes that under small disturbances, the change in voltage can be linearly expressed as a linear combination of the changes in active power and reactive power.

[0073] The method for calculating the dynamic voltage deviation rate includes: setting a short-time scale window and a long-time scale window in parallel; calculating the arithmetic mean of the voltage amplitude within the two windows respectively, taking the long-time mean as the dynamic reference, and defining the difference between the short-time mean and the long-time mean as the instantaneous voltage deviation; obtaining the dynamic voltage deviation rate by numerically differentiating the instantaneous voltage deviation time series, which is used to characterize the rate at which the voltage operating point deviates from its short-term stable trajectory.

[0074] The method for calculating the spectral entropy of voltage fluctuations includes: performing a Fast Fourier Transform (FFT) on the instantaneous voltage value sequence within a sliding time window to obtain its amplitude spectrum over a wide frequency range; dividing the spectrum into multiple continuous sub-bands according to a preset bandwidth (e.g., 5Hz or 10Hz); calculating the proportion of spectral energy in each sub-band to the total spectral energy within the window to form a probability distribution sequence; and calculating the spectral entropy of voltage fluctuations based on the Shannon information entropy formula, which is used to quantify the degree of disorder or randomness of voltage harmonics and interharmonic components.

[0075] The weight coefficients of the multi-index adaptive weighted fusion algorithm can be dynamically adjusted based on the preliminary classification results of the identified oscillation modes; for voltage instability oscillation modes, the weights of active power-voltage and reactive power-voltage dynamic sensitivity are increased; for harmonic oscillation or control-interactive oscillation modes, the weights of voltage fluctuation spectral entropy are increased.

[0076] The dynamic risk warning threshold is obtained by fitting the probability density distribution of the comprehensive risk index calculated from massive historical steady-state operation data, and selecting an upper limit of a confidence interval with extremely low probability as the benchmark threshold. This benchmark threshold is also corrected according to current key operating parameters (such as total output of new energy sources, system short-circuit capacity ratio, etc.) to improve the adaptability of the warning. The risk indicator set is a configurable and modular indicator library, which allows operators to selectively enable or disable some risk indicators through the software interface according to the electrical characteristics of a specific site, the resource limitations of the on-site computing device, and the required warning accuracy, and adjust their weight coefficients in the risk index fusion calculation online.

[0077] In one embodiment, instantaneous data of voltage, active power, and reactive power are collected from each monitoring point within the new energy power station to obtain electrical quantity time series data. The electrical quantity time series data is then processed according to a target sliding window to output risk characteristic quantities. This process includes: collecting instantaneous data of voltage, active power, and reactive power from each monitoring point within the new energy power station; reconstructing and coupling the collected data to obtain electrical quantity time series data; performing bandpass filtering on the electrical quantity time series data within the target sliding window; extracting small disturbance components during the operation of the new energy power station based on the processing results; constructing a linear regression model; reconstructing the least squares method using regularized weighting techniques; and solving for the regression coefficients of the linear regression model based on the reconstruction results, using these as estimates of the dynamic sensitivity of active and reactive voltage; calculating the dynamic voltage deviation rate and voltage fluctuation spectral entropy based on the voltage amplitude and Shannon entropy techniques, and combining these with the dynamic sensitivity of active and reactive voltage as risk characteristic quantities.

[0078] In one embodiment, instantaneous values ​​of voltage, active power, and reactive power at various monitoring points within the renewable energy power station are collected. The collected data undergoes reconstruction and coupling characterization processing to obtain electrical quantity time series data. This includes: collecting instantaneous values ​​of voltage, active power, and reactive power at the grid connection point, feeders, and busbars of the renewable energy power station using a measuring instrument at a sampling frequency; standardizing each data point; constructing instantaneous state points based on the standardization results; and using these instantaneous state points for embedding processing to generate state vectors in the voltage-power coupling phase space to characterize the dynamic potential of the renewable energy power station; randomly selecting data points within the voltage-power coupling phase space and finding several nearest neighbor points for each data point; combining the data points with their nearest neighbor points to generate a local neighborhood; and performing component analysis on the local neighborhood; and calculating the state vectors describing the dynamic characteristics of the renewable energy power station based on the primary and secondary directions of emission from the local neighborhood according to the component analysis results, thus obtaining the electrical quantity time series data.

[0079] To address the limitations of single-variable reconstruction in this embodiment, a multivariable joint phase space reconstruction method based on strong physical correlation is proposed. Instead of treating voltage as an isolated time series, this method considers it, along with active and reactive power, as an observation of a three-dimensional state vector describing the dynamic characteristics of the power station port. Specifically, this is achieved by constructing an electric-power coupled phase space and calculating the orthogonal divergence rate of the state trajectory. The implementation process is as follows:

[0080] Constructing a multivariable coupled state vector: within a sliding time window, for synchronously sampled voltages The active power p(t) and reactive power q(t) time series are first standardized in this embodiment to eliminate the influence of different physical dimensions and construct a completely new instantaneous state point:

[0081] ;

[0082] in, , and For the standardized signal, and by embedding this multivariate time series, a state vector in the voltage-power coupled phase space is constructed:

[0083] ;

[0084] Here is the state vector X i It is a 3×m dimensional vector that simultaneously contains the evolution information of voltage, active power, and reactive power within a continuous time window, completely depicting the dynamic potential state of the system at that point.

[0085] Calculating the Orthogonal Divergence Rate (ODR) of state trajectories by only calculating the divergence rate of nearest neighbors (as in the original RLDE) is susceptible to the curse of dimensionality and sensitive to noise in high-dimensional spaces. Therefore, this embodiment introduces a more robust and information-rich divergence rate measurement method, which considers each point X in the phase space... i In this embodiment, the k nearest neighbors are found, and principal component analysis is performed on the local neighborhood formed by these points to find the main and secondary directions of the divergence of the trajectory of the neighborhood.

[0086] When the system is stable, the trajectory converges, and the neighborhood stretches along the tangent of the attractor and flattens along the normal. When the system tends to become unstable, the trajectory diverges, and the neighborhood expands significantly in multiple directions (especially in directions perpendicular to the original stable trajectory).

[0087] Therefore, this embodiment proposes a theoretical formula for the orthogonal divergence rate of state trajectories:

[0088] ;

[0089] In the formula, ODR represents the orthogonal divergence rate of the state trajectory. It is a scalar value characterizing the average divergence rate of the phase space trajectory in the non-principal (orthogonal) directions, which can more sensitively capture early behavior escaping from a stable attractor. d represents the dimension of the local neighborhood, typically taken as d ≤ 3 × m, and X... i,neigh Indicates time i with respect to X i The local neighborhood point set centered on σ j (·) represents the j-th singular value obtained after performing singular value decomposition (SVD) on the covariance matrix of a point set. The first singular value σ1 usually corresponds to the main evolution direction of the trajectory, while σ2,…,σ… d Then it corresponds to its orthogonal direction.

[0090] This represents the logarithmic ratio of the expansion or contraction of the neighborhood in the j-th orthogonal direction after evolution time p·Δt.

[0091] By constructing a multivariable coupled phase space, the monitoring object in this embodiment changes from a one-dimensional projection (voltage) of the system state in a high-dimensional space to the system state itself. This allows the monitoring indicators to reflect the coupling dynamics between multiple physical quantities, and the information completeness far exceeds that of the original scheme. By introducing the ODR index, this embodiment no longer only focuses on the distance change between two points, but analyzes the deformation of a local neighborhood geometry, providing richer dynamic information, especially the capture of orthogonal divergence. This makes the method extremely sensitive to the early warning of oscillatory instability and has better noise resistance. At the same time, this embodiment is based on a deep understanding of the oscillation mechanism of power systems, that is, oscillation is VPQ coupled instability. It carefully designs state vectors and divergence rate measures, thus demonstrating the deep and organic integration of domain knowledge and nonlinear analysis tools.

[0092] In one embodiment, the process of reconstructing the least squares method using regularized weighting techniques and solving for the regression coefficients of the linear regression model based on the reconstruction results, serving as estimates of the dynamic sensitivity of active and reactive voltage, includes: introducing a coherence function based on the processed electrical quantity time series data to quantitatively assess the degree of linear correlation between active power, reactive power, and voltage at each frequency point, and constructing a confidence weight function based on the degree of linear correlation and frequency confidence weights; optimizing the linear regression model based on the confidence weight function, using composite coefficients as strong physical constraints to reveal the inherent integral constraint relationship between the real and imaginary parts of the response function of the linear time-invariant system, and using this inherent integral constraint relationship as a constraint condition; generating a regularization term with composite coefficients as the core based on the constraint condition, fundamentally reconstructing the least squares objective function using the regularization term, and solving the linear regression model based on the reconstructed least squares objective function; outputting the regression coefficients of the linear regression model based on the solution results, and using the regression coefficients as estimates of the dynamic sensitivity of active and reactive voltage.

[0093] In the process of performing Fourier transform on the disturbance signal, completing the frequency domain transformation, calculating the cross-spectral density, and solving the frequency domain problem using weighted least squares, the Kramers-Kronig relation (i.e., the composite coefficient) in power system network theory is introduced as a strong physical constraint into the identification process of frequency-resolved dynamic sensitivity. The Kramers-Kronig relation reveals the inherent integral constraint relationship between the real and imaginary parts of the response function of a linear time-invariant system (such as impedance Z(jw)). This means that the sensitivity (which is essentially a manifestation of the system admittance) cannot be identified in isolation at each frequency point. Instead, it is necessary to ensure that the overall shape of the entire family of identified sensitivity spectrum curves satisfies the fundamental laws of physics.

[0094] To achieve the above-mentioned goals, this embodiment fundamentally reconstructs the original weighted least squares objective function based on Kramers-Kronig constraints (KKR-WLS), proposing a completely new objective function. The original objective function is as follows:

[0095] ;

[0096] This formula only aims to minimize the data fitting error. Therefore, this embodiment introduces a regularization term to avoid solutions that do not conform to the Kramers-Kronig relationship. The resulting solution is as follows:

[0097] ;

[0098] In the formula, J KKR Let W represent the Kramers-Kronig regularization objective function, which is the final objective to be solved. comp (f k ) represents the composite confidence weight function, which is a modification of the original weight function W(f) k A significant improvement is made to this, as it not only considers coherence but also incorporates signal-to-noise ratio information, as defined below:

[0099] ;

[0100] Among them, SNR i (f k ) is at the frequency point f k At point i, the signal-to-noise ratio (SNR) of the power signal i is estimated online by comparing the signal energy with the noise floor energy of adjacent frequency bands. This improvement enables the weighting function to simultaneously suppress irrelevant and heavily noise-contaminated frequency data, resulting in stronger robustness. β′ is a new comprehensive threshold parameter, and V(f k ), P(f k ) and S fr These are the complex spectral vectors of voltage and power, and the frequency-resolved complex sensitivity matrix to be solved, respectively. It should be noted that this embodiment directly identifies the complex sensitivity because its imaginary part contains important phase information, which is crucial for oscillation mechanism analysis. λ KK It is the Kramers-Kronig regularization coefficient. A positive scalar hyperparameter used to balance the relationship between data fit fidelity and conformity to physical laws. Its value can be optimized offline through methods such as cross-validation.

[0101] It is the Hilbert Transform operator, which is the mathematical expression of the Kramers-Kronig relation. This means that performing a Hilbert transform on the entire real part spectrum of the sensitivity should theoretically result in a value equal to the imaginary part spectrum of the sensitivity.

[0102] This represents the regularization penalty term. This term measures the extent to which the currently identified sensitivity spectrum deviates from the Kramers-Kronig relationship; it is zero when the identification result conforms to physical laws.

[0103] Furthermore, by leaping from point to surface, this approach transforms sensitivity identification from a parameter estimation problem of a series of isolated frequency points into a functional optimization problem of solving a smooth function (spectral curve) that satisfies specific physical constraints. Through regularization terms, this embodiment encodes prior knowledge about the physical characteristics of the system (Kramers-Kronig relationship) into the solution algorithm, enabling the recovery of reasonable sensitivity values ​​even when the data quality at certain frequency points is very poor, by leveraging information from other frequency points and physical constraints. This achieves extremely strong noise suppression and data restoration capabilities. At the same time, it concretizes the abstract Kramers-Kronig relationship into a computable regularization term and integrates it with the weighted least squares framework to solve the broadband oscillation early warning problem in engineering. This approach is far beyond what those skilled in the art would easily conceive of; it requires a deep cross-understanding of multiple fields such as power system physics, signal processing, and numerical optimization.

[0104] In one embodiment, calculating the dynamic voltage deviation rate and voltage fluctuation spectral entropy based on the voltage amplitude and Shannon entropy technique includes: calculating the short-time average voltage amplitude and the long-time average voltage amplitude within a target time window; generating a deviation time series by obtaining the instantaneous voltage deviation based on the short-time average voltage amplitude and the long-time average voltage amplitude; numerically differentiating the deviation time series using a first-order difference approximation technique; obtaining the dynamic voltage deviation rate based on the numerical differentiation result, which characterizes the rate at which the voltage operating point deviates from its short-term stable trajectory; performing a fast Fourier transform on the voltage signal to obtain the amplitude spectrum of the voltage signal over a wide frequency range, and dividing the amplitude spectrum into several continuous sub-bands according to a preset bandwidth; calculating the proportion of spectral energy within the sub-bands to the total spectral energy within the window, generating a probability distribution sequence, and processing the probability distribution sequence using Shannon entropy technique to output the spectral entropy of the voltage fluctuation.

[0105] This embodiment examines the incubation process of oscillation from an information theory perspective. A healthy power system, when subjected to continuous background noise and minor disturbances, exhibits a broadband, incoherent dynamic response in the frequency domain, with a diffuse energy distribution. From an information theory perspective, this is a high-entropy (high uncertainty, high complexity) state. When the damping of a certain mode of the system weakens and tends towards instability, the system energy gradually concentrates at the frequency (or band) of this specific, unstable oscillation mode. The originally diffuse energy distribution begins to become ordered, concentrated, and deterministic. This process corresponds to a significant reduction in the system's state information entropy in information theory. Therefore, monitoring the information entropy changes of the system's output signal can provide this embodiment with a novel, non-intrusive risk assessment dimension. This embodiment selects Multiscale Permutation Entropy (MPE) as the specific technical tool to implement the above idea. Compared to other entropies (such as sample entropy and approximate entropy), permutation entropy has a fast calculation speed, good robustness to noise, and is particularly suitable for processing long-term series. Multiscale analysis allows this invention to examine changes in system complexity at different time scales. The specific implementation steps are as follows:

[0106] Time series coarsening: coarsening the original voltage perturbation time series We constructed coarse-grained sequences under different scale factors τ.

[0107] Permutation pattern extraction: At each scale, the coarse-grained sequence is divided into multiple short sub-segments of length m (embedding dimension). For each sub-segment, the corresponding "permutation pattern" is determined according to the numerical relationship of its internal elements (for example, the permutation pattern of the m=3 sequence {5,9,2} is {2,3,1}).

[0108] Calculate permutation entropy: Statistically count the frequency of all possible permutation patterns and calculate the permutation entropy value at this scale according to the definition of Shannon entropy, as a novel risk indicator.

[0109] ;

[0110] In the formula, IRCI(t) represents the Information Rate of Convergence Index. The physical meaning of this index is the rate at which the uncertainty of the system state decreases. A large positive value indicates that the system is rapidly evolving from a complex, disordered state to a simple, ordered state (which may imply a dangerous single oscillation), serving as a strong warning signal. HPE(t,τ) represents the permutation entropy calculated at time t with a scale of τ, where τ... max ω represents the maximum time scale of the analysis. τThe weighting coefficients representing different scales can be designed according to the oscillation frequency bands of interest. For example, to focus on subsynchronous oscillations, a larger weight can be given to the corresponding time scale.

[0111] This represents the negative derivative of the weighted sum of entropy with respect to time. A decrease in entropy is a risk signal, hence its derivative is negative. After taking the negative sign, it becomes a positive indicator that increases with increasing risk.

[0112] Furthermore, it completely breaks away from the traditional framework based on electrical characteristics (voltage, power, impedance), providing a completely new theory and tool for oscillation early warning from the fundamental scientific level of information theory. It possesses strong theoretical originality; entropy theory does not depend on specific system models or oscillation mechanisms. Whether it's subsynchronous oscillation or mid-to-high frequency oscillation, as long as its development process involves a transformation of the system's dynamic behavior from complexity to order, this indicator can theoretically capture it effectively. This gives it strong universality. At the same time, the calculation of permutation entropy only involves comparison and counting, not complex distance calculations, thus its computational efficiency is far higher than other complexity measures such as sample entropy.

[0113] In one embodiment, the process of using weighted fusion technology to perform nonlinear normalization on risk characteristics and generating a single broadband oscillation risk index reflecting the broadband oscillation level of new energy power plants includes: expanding the original spectral characteristics and introducing estimated values ​​of modal damping ratio and modal energy to generate new feature vectors; feeding the new feature vectors into an online Dirichlet process mixture model; calculating the posterior probability that the new feature vectors belong to all known mode clusters using the online Dirichlet process mixture model; creating mode clusters based on the posterior probabilities and using the new feature vectors as new mode samples; generating weight values ​​based on mode evolution according to the mode clusters and posterior probabilities; normalizing the risk characteristics; and weighting and summing the normalized risk characteristics and weight values ​​to obtain a comprehensive risk index, thereby generating a single broadband oscillation risk index reflecting the broadband oscillation level of new energy power plants.

[0114] This embodiment integrates multiple indicators to upgrade a static rule-based expert system into an intelligent diagnostic system with online learning and adaptive capabilities. It must not only be able to identify known patterns, but also detect, cluster, and ultimately adapt to unknown new patterns.

[0115] This embodiment uses an Online Dirichlet Process Mixture Model (DPMM) to replace the static pattern library and GMM / NN classifier in the original scheme. DPMM is a powerful Bayesian nonparametric clustering method, whose greatest advantage is that it does not require pre-specifying the number of clusters (i.e., the number of oscillating patterns K). Its implementation steps are as follows:

[0116] The original spectral eigenvector F is expanded by adding modal damping ratio ζ and modal energy E, in addition to center frequency and bandwidth, taking advantage of the ESPRIT algorithm's ability to accurately estimate modal parameters. mode The estimated value. The new eigenvector is F′=[f] c B w E c N p ,ζ,E mode [The information is richer.]

[0117] Online Pattern Discovery and Membership Calculation: When a new feature vector F′(t) is extracted, this embodiment feeds it into the online DPMM model. The model performs the following operations: calculates the posterior probability (i.e., membership degree μ) that F′(t) belongs to all known pattern clusters k∈{1,…,K(t)}. k (t)); Simultaneously calculate the probability that F′(t) does not belong to any known cluster and a new cluster should be created; if the latter probability exceeds a certain threshold, the system will automatically create a new pattern cluster K(t+1)=K(t)+1, and use F′(t) as the first sample of the new pattern. Then, when a new oscillation precursor that has never been seen before appears in the power grid, the system can automatically detect it and establish a new pattern file for it; and based on the dynamic optimization of the weights according to the pattern evolution, the weight adjustment formula also evolves accordingly, as follows:

[0118] ;

[0119] In the formula, K(t) represents the total number of dynamically changing patterns, and C k (t) represents the pattern confidence factor, a variable in the interval [0,1], which represents the system's confidence in pattern cluster k. For a newly created pattern, its confidence is low (because there are few samples). As more similar feature vectors are assigned to the cluster, its confidence will gradually increase, which can prevent the system from being misled by instantaneous and accidental noise features and increase the robustness of decision-making. The confidence can be designed as a function of the number of samples and variance within the cluster.

[0120] The shift from static matching to dynamic learning transforms early warning systems from static criteria that can only identify known oscillation problems into experts capable of continuously learning and summarizing new problems in real-world scenarios. This enables long-term autonomous operation and adaptation to the ever-changing operating modes of the power grid. Furthermore, the static pattern library is a major limitation of all pattern recognition-based early warning systems. Solving this problem gives the system unprecedented robustness and adaptability in the face of increasingly complex power grid environments. Moreover, by combining cutting-edge machine learning theory (online DPMM) with the domain problem of broadband oscillations in power systems, a closed-loop intelligent system that can automatically discover new patterns and dynamically adjust fusion strategies is constructed. This requires a high level of interdisciplinary integration and innovation capabilities and is by no means an extension of unconventional technical means.

[0121] Furthermore, at the sensitivity calculation level, this embodiment elevates conventional least squares into a physically consistent and highly robust parameter identification method by introducing regularization of Kramers-Kronig physical constraints. At the state monitoring level, this embodiment elevates the monitoring dimension from univariate projection to system-level multivariate coupled dynamics by constructing an electric-power coupled phase space and calculating the orthogonal divergence rate of state trajectories. At the multi-index fusion level, this embodiment upgrades the system from static expert rules to an intelligent diagnostic system with online learning and new pattern discovery capabilities by introducing online Bayesian nonparametric clustering. At the same time, it introduces a novel information theory-based early warning dimension: the information entropy rate convergence index, breaking through the existing conventional technical framework.

[0122] Traditional time-domain multiple linear regression model It inherently contains two assumptions that are difficult to avoid in engineering practice. The model assumes that within the entire selected frequency band, S pv and S q It is a constant real number; however, for a system dominated by power electronic converters, its equivalent impedance is... It is frequency The strongly correlated complex function means that the voltage-power response characteristics of the system in the subsynchronous frequency band are completely different from those in the mid-to-high frequency band. A constant sensitivity coefficient cannot capture this frequency dependence and will inevitably be distorted due to the aliasing of information from multiple frequency bands.

[0123] Conventional least squares treats all data points and all frequency components within the sliding window equally. During the oscillation incubation period, the true dynamic response signal related to unstable modes is often submerged in a large amount of background noise and other irrelevant disturbances. Conventional methods suffer from frequent jumps in calculation results and low signal-to-noise ratio due to the severe contamination of these noises, making it impossible to form a stable and reliable risk criterion.

[0124] To overcome the aforementioned shortcomings, this embodiment proposes a novel weighted identification method in the frequency domain. The core idea is that the voltage-power response relationship of the system differs at different frequency points, and not all frequency components have equal confidence in identifying this relationship. Therefore, it is necessary to adaptively identify and focus on frequency points where voltage and power exhibit a strong linear correlation, and perform parameter estimation based on these "high-confidence" frequency point data. The specific implementation steps are as follows:

[0125] Step 1: Frequency domain transformation and cross-spectral density calculation of the signal. Within the sliding time window, the preprocessed perturbation signal is processed. , , Short-Time Fourier Transform (STFT) is performed. In practice, to obtain good frequency resolution and temporal locality, a Hanning window can be used. The window length is selected based on the sampling rate and target frequency resolution (for example, for a 10kHz sampling rate, a 2048-point window can be used to obtain a resolution of approximately 4.88Hz), and a 50% overlap is set to ensure temporal smoothness. This yields their values ​​at each discrete frequency point. complex spectrum , , Then, the overlapping segments of the spectrum are averaged to obtain a stable cross-power spectral density. , and self-power spectral density , wait.

[0126] Step 2: Calculate the frequency-resolved coherence function and confidence weights, and introduce the coherence function. This is used to quantitatively assess the degree of linear correlation between input (power) and output (voltage) at each frequency point. For example, voltage and active power at different frequencies... The coherence function is defined as:

[0127] ;

[0128] This value is between 0 and 1. The closer it is to 1, the stronger the linear relationship between voltage fluctuations and active power fluctuations at that frequency point. The more reliable the data at that frequency point is for estimating the linear sensitivity model, the better. Conversely, if it is close to 0, it means that the two are not correlated at that frequency, and it may be dominated by noise. The data at that frequency point should be suppressed.

[0129] Based on this theoretical analysis and derivation, a confidence weight function was constructed:

[0130] ;

[0131] In the formula, Frequency confidence weight, one in A dimensionless scalar within an interval, representing the frequency point. At this point, the reliability of the voltage response relationship derived from the power disturbance data is determined, and this weight will be directly used in subsequent weighted least squares calculations. Discrete frequency points: The frequency value of the k-th frequency point obtained after STFT analysis, in Hertz (Hz). Hyperbolic tangent function. A sigmoid activation function with a range of... This is used here to smoothly map linearly varying inputs onto nonlinear weights, avoiding computational instability that may result from hard threshold switching. The slope control parameter of the weighting curve. A dimensionless positive real number used to control the weighting function at the threshold. The width of the nearby transition zone. The larger the value, the steeper the weighting curve, and the stronger the selectivity for high-coherence frequencies. In engineering applications, its value can be determined through offline simulation optimization, and it typically ranges from 5 to 15. Coherence threshold parameter, one in The dimensionless scalar within the interval represents the minimum coherence sum subjectively set by the user to judge the reliability of the power-voltage relationship. Only when the combined coherence of voltage with active and reactive power exceeds this threshold will that frequency point be assigned a significant weight. Its typical value is usually set between 0.5 and 0.8. Coherence function. Among them for (active power) or (Reactive power), this is in A dimensionless scalar within an interval, quantitatively describing the frequency range. At that point, the power spectrum of the voltage signal can be represented by the power signal. The proportion of linear interpretations.

[0132] Step 3: Construct and solve the frequency-domain weighted least squares problem. The linear model in the frequency domain is represented as follows: The goal is to solve for the frequency-resolved dynamic sensitivity. and This makes the weighted sum of squared errors The solution to this weighted least squares problem can be achieved on embedded computing platforms or servers using efficient numerical computing libraries such as LAPACK or Eigen.

[0133] Furthermore, the frequency-resolved dynamic sensitivity calculation method proposed in this embodiment solves the inherent defects of conventional time-domain linear regression models in wideband, low signal-to-noise ratio environments by introducing confidence weights of cross-correlation spectra.

[0134] Dynamic voltage deviation rate (VTRR) is effective for detecting early signs of voltage instability such as "slow drift" caused by integral saturation of the control system. However, it is ineffective for oscillating instability, especially during the oscillation incubation period, when the system may experience small-amplitude, gradually decreasing damping oscillations around a stable equilibrium point. During this process, the long-term average voltage may remain unchanged, rendering the VTRR completely ineffective.

[0135] To overcome the aforementioned deficiencies, this embodiment introduces the phase space reconstruction theory from nonlinear dynamics, aiming to reconstruct the phase space from a single voltage-time series. In this process, the higher-dimensional topological structure of the original dynamic system is restored.

[0136] Step 1: Phase space reconstruction, within the sliding time window, for the voltage time series , construct a series Phase space point: To ensure that the reconstructed phase space can expand the topology of the original attractor to the maximum extent, the embedding dimension is... and delay time Careful selection is required. It can be determined by the false nearest neighbor method, and the value is usually between 3 and 7; The average mutual information method can be used to select the sample when the mutual information function reaches its first minimum value, which usually corresponds to 5 to 15 sampling points.

[0137] Step 2: Calculate the local divergence rate. In a stable system, its phase space trajectory converges to an attractor, and neighboring points remain neighborly after evolution. When the system becomes unstable, the trajectory begins to diverge. To improve computational efficiency, when searching for each point... nearest neighbor In such cases, fast search algorithms such as kd-trees can be used, as detailed below:

[0138] ;

[0139] In the formula, This represents the local divergence rate of the voltage trajectory, a scalar value that characterizes the average exponential divergence rate of adjacent trajectories in the reconstructed phase space within the current analysis window. Its unit is... A larger value indicates that the system's dynamic behavior is more unstable. Represents the evolution time, a time scalar representing the duration of trajectory evolution considered when calculating the divergence rate, in seconds (s). Its value is... ,in It is the data sampling time interval. Indicates the number of valid point pairs, an integer representing the total number of points within the current analysis window that successfully found nearest neighbors and completed the divergence rate calculation, used to calculate the average. Represents the natural logarithm function. Represents the Euclidean norm, used to calculate the Euclidean norm of two state vectors. Straight-line distance in 3D phase space Represents the i-th state vector, one A dimensional column vector, formed by embedding voltage time series data at time delays. Build, The nearest neighbor state vector. In phase space, with The other state vector with the smallest Euclidean distance. The evolution step size, a positive integer, represents the number of discrete-time steps for neighboring point pairs to evolve forward. Its selection is a trade-off: if it is too small, the differences are not obvious and it is easily affected by noise; if it is too large, the nonlinear effect will be too strong due to trajectory folding. It is usually selected as the number of sampling points corresponding to 1 / 4 to 1 / 2 of the period of the main oscillation mode of the system.

[0140] By introducing the local divergence rate of the voltage trajectory, the early warning method is elevated from the traditional "signal processing" level to the nonlinear dynamics level. By mining the hidden system state evolution law behind the time series, the risk is assessed. Its monitoring capability is completely incomparable to conventional voltage deviation indicators.

[0141] like Figure 2 As shown, according to another embodiment of the present invention, a broadband oscillation risk early warning system for new energy power plants based on voltage dynamic indicators is also proposed. This system includes:

[0142] The data acquisition module is used to collect instantaneous data from various monitoring points within the new energy power plant to obtain time series data of electrical quantities.

[0143] The indicator calculation module is used to process electrical quantity time series data according to the target sliding window and output risk characteristic quantities;

[0144] The risk fusion module is used to perform nonlinear normalization processing on risk characteristics using weighted fusion technology, and generate a single broadband oscillation risk index that reflects the broadband oscillation level of new energy power plants based on the normalization results.

[0145] The early warning decision module is used to compare the difference between a single broadband oscillation risk index and a dynamic risk early warning threshold, and to trigger an oscillation risk early warning signal when the difference result reaches the target alarm requirement, thereby implementing oscillation risk early warning.

[0146] To facilitate understanding of the above technical solutions of the present invention, the working principle or operation method of the present invention in actual process will be described in detail below.

[0147] The purpose of this embodiment is to overcome the shortcomings of existing broadband oscillation defense technologies that lack early warning capabilities. It aims to achieve early, accurate, and reliable online identification of oscillation risks, mainly addressing the core issue in broadband oscillation prevention and control: whether the broadband stability margin can be assessed by analyzing the system's response characteristics to endogenous noise and small external disturbances before the oscillation amplitude becomes significant.

[0148] To achieve the above objectives, the technical solution proposed in this embodiment includes four core steps: data acquisition, risk indicator calculation, risk index fusion, and early warning decision-making. Key electrical quantity data of the grid-connected points of new energy power plants are acquired through a high-frequency synchronous acquisition device. Within a sliding time window, an innovative set of dynamic risk indicator indicators, centered on voltage, is calculated in parallel. This indicator set does not rely on significant oscillation amplitudes but assesses system stability by analyzing subtle distortions in the voltage waveform, abnormal distribution of spectral energy, and the deteriorating trend of the dynamic relationship between voltage and power. Specific indicators include, but are not limited to:

[0149] Active-voltage dynamic sensitivity (dv / dp) and reactive-voltage dynamic sensitivity (dv / dq): These parameters draw upon and develop the static voltage stability theory of power systems, and successfully apply it to risk assessment of wideband dynamic processes. By identifying the voltage response sensitivity to power disturbances in real time, they directly quantify the voltage stability margin of the system near the operating point. The abnormal surge in sensitivity has been proven by theory and practice to be a strong precursor to near-voltage instability oscillations in the system.

[0150] Dynamic voltage deviation rate: reflects the dynamic trend of voltage deviating from its normal operating trajectory, and is good at capturing the precursors of oscillations caused by control system misalignment or slow process instability.

[0151] Voltage fluctuation spectrum entropy: Using information entropy theory to quantify the complexity and disorder of harmonic and interharmonic components in voltage waveforms, a sharp increase in spectrum entropy is often a sign of enhanced nonlinearity of the system, oscillation of nearby harmonics, or control interaction instability.

[0152] Considering the potential limitations and biases of a single indicator, this embodiment proposes a multi-indicator adaptive weighted fusion model. This model first normalizes each indicator and assigns weights based on its sensitivity and importance to different oscillation modes, ultimately fusing them into a single, intuitive, comprehensive broadband oscillation risk index R. index This fusion mechanism ensures the comprehensiveness and robustness of early warning decisions, ultimately enabling real-time calculation of R... indexCompared to a dynamically adjustable risk baseline, which is formed by learning the normal behavior patterns of the power grid under different operating modes, this baseline abandons the rigidity of traditional fixed thresholds. index A warning signal is triggered when there is a significant and sustained deviation from the dynamic baseline.

[0153] I. Overall Architecture and Workflow;

[0154] like Figure 2 and Figure 4 As shown, the technical solution proposed in this embodiment relies on a hardware and software integrated system. Functionally, the system can be divided into a data acquisition module, an indicator calculation module, a risk fusion module, and an early warning decision-making module. Figure 4 The working principle of the indicator calculation module is demonstrated. The input real-time high-frequency data stream (i.e. electrical quantity time series data) is distributed to multiple parallel computing units. Each computing unit is responsible for a specific risk indicator algorithm, such as calculating active power-voltage dynamic sensitivity, reactive power-voltage dynamic sensitivity, dynamic voltage deviation rate, and voltage fluctuation spectrum entropy. This parallel processing architecture ensures that all key indicators can be calculated efficiently and synchronously, providing real-time input for subsequent risk fusion.

[0155] Data acquisition module: Deployed at new energy power plants or key collection stations, the core equipment is a broadband measurement device with high-precision time synchronization. This module is responsible for synchronously acquiring the instantaneous values ​​of three-phase voltage and current at the grid connection point at a frequency of not less than 10kHz, and transmitting the data stream to the computing module in real time through a high-speed communication network (such as optical fiber).

[0156] Indicator Calculation Module: This module can be deployed on the edge computing unit at the station or on the high-performance server of the remote master station. It is the core computing unit in this embodiment. It processes the real-time data stream in a sliding time window (e.g., the window length is 100ms and the sliding step is 10ms) and calculates multiple voltage dynamic risk indicators in parallel.

[0157] Risk fusion module: Receives multiple parallel outputs from the indicator calculation module, executes a multi-indicator weighted fusion algorithm, and calculates a single comprehensive risk index R. index .

[0158] Early warning decision module: will provide real-time R... index With dynamic risk warning threshold R th To prevent false alarms caused by momentary disturbances, this module incorporates alarm persistence judgment logic; it only checks for alarm persistence when R... index >R th An alarm is only formally issued to the monitoring system or operators when the status lasts for more than a preset value (e.g., 50ms).

[0159] II. Theoretical basis and detailed calculation of voltage dynamic risk indicators;

[0160] The core innovation of this embodiment lies in proposing and implementing a set of risk assessment indicators based on voltage dynamic indicators. The theoretical basis, calculation method and physical meaning of each indicator will be explained in detail below.

[0161] (1) Active power-voltage dynamic sensitivity S pv With reactive power-voltage dynamic sensitivity S qv :

[0162] Theoretical Basis and Innovative Demonstration: In classical power system static stability analysis, the PV curve and QV curve are the cornerstones for evaluating voltage stability, such as... Figure 5 The figure shows a classic PV curve for a power system, shaped like a nose. The horizontal axis represents the injected active power P, and the vertical axis represents the node voltage V. Point A represents the stable operating point of the system, located in the upper half of the curve. Here, the curve is relatively flat, and the slope (representing sensitivity |dv / dp|) is small, indicating that the system is not sensitive to power disturbances and the voltage is stable. Point B represents the critical operating point of the system, close to the nose of the PV curve. At this point, the curve becomes very steep, and its slope (sensitivity |dv / dp|) increases sharply, indicating that even a small increase in power will cause a large drop in voltage, and the system is on the verge of instability. This dynamic sensitivity can be calculated online in real time. (It can identify the accumulation of risks in advance during the process of the system moving from point A to point B). The PV curve describes the law of change of the voltage amplitude of a certain node (load node) with the active power injected into that node. The nose of the curve represents the maximum power transmission limit that the system can maintain voltage stability. Mathematically, the slope of the tangent at the nose is vertical, that is, the voltage sensitivity (derivative) dV / dP to active power tends to infinity at this point. This means that when the system operating point is close to the nose, even a very small active power disturbance will cause a sharp drop in voltage, and the system is on the verge of collapse. Similarly, the sharp increase in dV / dQ also indicates voltage instability caused by insufficient reactive power.

[0163] The first innovation of this embodiment is to dynamically apply this widely accepted static stability criterion to the early warning of broadband oscillations through an engineering method. Traditional views hold that PV / QV analysis is not applicable to high-frequency dynamic processes, but the underlying physical nature of the system's response to disturbances deteriorates as the stability margin decreases is universal. It can be realized through the development and application of existing high-speed, synchronous acquisition devices for high-frequency dynamic analysis and early warning of degradation processes.

[0164] Wideband oscillations, especially voltage source oscillations caused by the interaction between the converter control system and a weak power grid, can essentially be viewed as the equivalent impedance of the system exhibiting negative damping characteristics at a certain oscillation frequency. This leads to the continuous amplification of disturbances at that frequency. This negative damping characteristic manifests externally as an abnormal voltage-power dynamic relationship. In a stable system exhibiting positive impedance, when an injected active power disturbance Δp > 0, its terminal voltage will decrease slightly, i.e., dV / dP < 0 (under generator conventions). However, when the converter control system (such as a phase-locked loop (PLL), current inner loop, etc.) experiences poor interaction with the power grid, causing it to exhibit negative damping, it may exhibit a positive feedback characteristic: a small active power injection disturbance Δp may cause a large change in voltage in the same direction, or the system itself may generate a power response in the same direction as the disturbance, thus affecting the calculated dynamic sensitivity |S pv |or|S qv The value has increased abnormally.

[0165] From the perspective of voltage stability, as the active power of new energy sources continues to increase, the voltage stability level decreases, and the voltage active power sensitivity coefficient (dv / dp) will increase accordingly. When dv / dp increases, even small fluctuations in the active power of new energy units will cause significant voltage changes. This voltage disturbance affects the output of the current loop control through the phase-locked loop (PLL), ultimately impacting the inverter inductor current. This closed-loop feedback path, composed of grid impedance and the PLL, is more prone to causing oscillations in the converter grid-connected system under weak grid conditions (i.e., high impedance and low short-circuit ratio). In other words, the voltage active power sensitivity S... pv It is a direct and measurable external characterization of the gain of the critical instability loop. The higher the sensitivity, the greater the voltage disturbance generated for the same internal power disturbance, the stronger the disturbance signal fed back to the control loop, the smaller the stability margin of the entire system (such as the margin obtained by impedance analysis), and the higher the risk of oscillation.

[0166] Therefore, based on existing research and extensive simulations in this embodiment, by continuously monitoring S pv and S qv The dynamic evolution trend is equivalent to the local dynamic PV / QV characteristics of the online scanning system. The continuous and rapid increase of this sensitivity index is a direct and quantitative manifestation of the weakening of damping and the decrease of stability margin in this frequency band. It is an inevitable stage for the oscillation to evolve from a small disturbance to a large-scale instability. This allows this embodiment to detect the risk in advance when the oscillation amplitude is still very small and has not yet triggered traditional protection, thus achieving true early warning.

[0167] Calculation method: Within the sliding time window, there are small disturbances Δv, Δp, and Δq in voltage v(t), active power p(t), and reactive power q(t). The partial derivatives are estimated by numerical methods.

[0168] Data preprocessing: Apply bandpass filters to the v(t), p(t), and q(t) sequences to extract small perturbation components in a specific frequency band (e.g., 1-100Hz), denoted as Δv(t), Δp(t), and Δq(t).

[0169] Sensitivity estimation: Multiple linear regression is used to fit the data points within a window. According to the principle of total differential, a small change in voltage can be approximated as:

[0170] ;

[0171] It is a multiple linear regression problem, with a length of Within the window, you can construct an overdetermined system of equations:

[0172] ;

[0173] Solving using the least squares method, such that the sum of squared errors is obtained. Minimum, to obtain sensitivity:

[0174] ;

[0175] In the formula, This indicates the active-voltage dynamic sensitivity, expressed in volts per watt (V / W) or per unit value. This indicates the reactive-voltage dynamic sensitivity, expressed in volts per var (V / var) or per unit value, where:

[0176] ;

[0177] ;

[0178] A represents the design matrix consisting of active and reactive power disturbances from N sampling points, B represents the corresponding voltage disturbance vector, and ε i Let A represent the regression error for the i-th data point. T A) -1 A T This represents the pseudoinverse of A.

[0179] Physical meaning: |S pv |or|S qv A sustained and rapid increase in | is a strong signal of weakened system voltage support and impending voltage instability oscillations. This indicator is highly sensitive to subsynchronous, mid-frequency voltage oscillations caused by improper control system parameters or interaction with a weak power grid.

[0180] (2) Dynamic voltage deviation rate R DVDR :

[0181] In the theoretical basis and validity demonstration, the power system is a high-dimensional nonlinear system, and its instability process does not always appear in the form of violent oscillations. Some types of oscillations, especially slow-process instability caused by the saturation of the integral link of the control system or the competition between multiple slow controllers, may not initially manifest as energy diffusion in the spectrum, but rather as a slow but irreversible drift of the operating point of one or more key state variables (such as voltage). The theoretical basis of this index is that a stable control system should have a strong negative feedback capability, which can firmly anchor the state variables near their setpoints, and any deviation will be quickly corrected. When the stability margin of the system decreases, the anchoring capability will weaken. Even in the absence of large disturbances, small noises within the system can cause the operating point to begin to deviate, resulting in the dynamic voltage deviation rate R. DVDR It is designed precisely to quantify the acceleration of this deviation. A stationary deviation (R0) DVDR ≈0) may simply represent a new steady state, but a continuously increasing deviation rate (R) DVDR If the value is consistently positive or negative and its absolute value increases, it clearly indicates that the system is losing control and the state variable is accelerating toward the instability boundary. Therefore, this indicator is an effective tool for capturing early signs of instability in slow processes.

[0182] Calculation method: Calculate the short-time window (length T) s For example, the average voltage amplitude V within 100ms av,q,s (t); Calculate the long-term window (length T) l For example, the average voltage amplitude V within 1 second. av,q,l (t), serving as a dynamically changing voltage reference, calculate the instantaneous voltage deviation ΔV(t) = V av,q,s (t)-V av,q,l (t).

[0183] The rate of change of instantaneous voltage deviation is calculated to obtain the dynamic voltage deviation rate. In the discrete implementation, a first-order difference approximation can be used:

[0184] ;

[0185] In the formula, R DVDR (k) represents the dynamic voltage deviation rate at the k-th sampling time, in units of V / s or pu / s, ΔV(k) and ΔV(k-1) represent the instantaneous voltage deviation between the current and previous sampling times, and Δt represents the sampling time interval.

[0186] Physical meaning: A continuously positive and increasing R DVDR This indicates that the system voltage is rapidly deviating from its stable operating trajectory, which is an important signal of accumulating oscillation risk.

[0187] (3) Voltage fluctuation spectrum entropy EVSF :

[0188] Theoretical Basis and Validity Demonstration: Based on nonlinear dynamics and bifurcation theory, the process of a power system transitioning from stable periodic motion (steady state) to complex aperiodic motion (oscillation, chaos) is often accompanied by the redistribution and diffusion of system energy between different modes. A healthy power system should have a highly ordered voltage waveform, with almost all energy concentrated on the fundamental frequency component, exhibiting a very simple spectrum of just a few lines and low complexity. This index innovatively introduces the mathematical tool of information entropy to quantify the complexity of this spectrum or the disorder of energy distribution. When the system approaches the instability boundary, various potential nonlinear interactions (such as the switching nonlinearity of converter control and the nonlinear coupling between different controllers) are excited by small disturbances, thus generating a large number of new frequency components in the system, including harmonics, interharmonics, and broadband noise. This makes the originally clean spectrum cluttered, with energy diffusing from a single fundamental frequency to the entire broadband domain, such as... Figure 6 As shown in Figure A, the voltage spectrum of the power grid in a stable state and a state of impending instability is compared. The blue spectral line represents the stable state, where energy is highly concentrated near the power frequency and the spectral entropy is very low. The red spectral line represents the impending instability, where multiple speech / interharmonic components appear and energy is dispersed in the frequency domain, resulting in a significant increase in spectral entropy. Figure B shows the trend of the spectral entropy index over time. In the stable operating region, the index value fluctuates steadily at a low level. When the system stability begins to deteriorate, the index enters the risk accumulation region and the value continues to rise rapidly. When it exceeds the warning threshold, an alarm can be triggered.

[0189] Shannon information entropy is the best measure of the uncertainty or information content of a system. Treating the energy distribution of the voltage spectrum as a probability distribution, the spectral entropy E... VSF The value of E directly reflects the degree of uniformity or disorder in this distribution; a continuously increasing E VSF The value, in its physical sense, is very clear: the dynamic modes within the system are becoming increasingly rich and unpredictable, and energy is shifting from the stable dominant mode to a large number of potentially unstable modes. This is a typical path for the system to become unstable. Therefore, E VSF It is an extremely sensitive precursor indicator of oscillations that profoundly reflects the trend of increasing complexity in the dynamic behavior of a system.

[0190] Calculation method: Perform Fast Fourier Transform (FFT) on the voltage signal v(t) within the current sliding window to obtain its spectrum V(f).

[0191] Calculate the total energy P total :

[0192] ;

[0193] Divide the entire wideband frequency range (e.g., 0-5000Hz) into M non-overlapping sub-bands. Calculate the... The energy P of each sub-band j .

[0194] Calculate the energy percentage p of the j-th sub-band. j :

[0195] ;

[0196] Calculate the spectral entropy of voltage fluctuations using the Shannon entropy formula:

[0197] ;

[0198] In the formula: E VSF N represents the voltage fluctuation spectral entropy, measured in bits. f V(f) represents the total number of frequency points after the FFT transformation. i ) represents frequency f i The corresponding spectral amplitude, M represents the total number of sub-bands in the spectrum division, P j p represents the energy of the j-th subband. j This represents the proportion of the energy of the j-th sub-band to the total energy, satisfying Σp j =1.

[0199] E in a physical sense VSF The larger the value, the more dispersed and disordered the voltage energy distribution in the frequency domain, the stronger the nonlinear characteristics of the system, and the higher the risk of oscillation. This index is particularly sensitive to oscillations caused by harmonic resonance, control interaction, etc.

[0200] III. Complementarity, Customization, and Feasibility of Indicators;

[0201] Complementary Advantages: The indicator set proposed in this embodiment has significant complementarity, constituting a multi-perspective risk observation system. pv and S qv It is evaluated from the perspective of input-output relationship and stability margin, and is most sensitive to voltage-instability oscillations; R DVDR Focuses on the macroscopic drift trend of the average voltage and excels at capturing slow-process instability; E VSF It evaluates from the perspective of signal morphology and spectral complexity, and has unique insights into harmonic resonance and complex control interaction oscillations. Different types of broadband oscillations have different precursor characteristics in these indicators. Through fusion analysis, the coverage and accuracy of early warning can be greatly improved.

[0202] Customized selection: In engineering practice, this index set serves as a toolbox. For monitoring devices at the site with limited computing resources, R, which requires less computation, can be selected.DVDR and E VSF Initial warnings are issued. For regional master stations with strong computing capabilities, a full set of indicators, including dynamic sensitivity, can be used for more refined risk assessment. Furthermore, the weights w of each indicator in the final fusion model are... i It is also possible to make targeted adjustments and optimizations based on the historical oscillation experience of specific stations, so as to achieve a "one station, one policy" approach.

[0203] Practical feasibility: The data source (high-frequency synchronous electrical quantity) required in this embodiment is already a standard configuration of modern new energy power plants and smart substations. The calculation methods described, whether FFT, numerical differentiation or least squares, are all mature algorithms in computer science and can be efficiently implemented on existing edge computing platforms or servers, fully meeting the needs of online real-time early warning.

[0204] IV. Risk Integration and Early Warning Decision-Making;

[0205] The normalized indicator I j ′(j=1,2,…,) and weight w j The weighted sum is used to obtain the comprehensive risk index R. index The core of early warning decision-making lies in the dynamic threshold R. th The settings.

[0206] ;

[0207] R th It is not a fixed value, but rather a value that varies over a long period of time depending on the power grid's operating conditions (such as different wind / sunlight periods and different main grid structures). index The data is obtained through statistical analysis of historical data. It can be a function that changes over time, or a lookup table based on current operating conditions. For example, it can learn that during periods of strong winds at night, R... index The normal baseline value will be slightly higher than during the calm midday period. The warning is triggered when the current R... index Significantly higher than the normal baseline under its operating conditions, such as Figure 7 As shown in the figure (this figure illustrates the workflow of the risk fusion module, with multiple parallel calculated risk indicators S input on the left), pv S qv After normalization to eliminate the influence of dimensions, each normalized indicator is multiplied by its corresponding weight coefficient. All weighted indicators are then combined in a weighted fusion unit into a single, quantitative comprehensive risk index R. index (This serves as the final basis for early warning decisions).

[0208] ;

[0209] In the formula, Rindex (t) represents the real-time comprehensive risk index calculated at time t, I j ′ represents the normalized value of the j-th risk indicator, w j The weight of the j-th risk indicator, t duration R indicates the duration of sustained overshooting of the risk index. th (Mode(t)) represents the dynamic risk threshold corresponding to the current power grid operation mode Mode(t), T persist This indicates the alarm duration threshold set to prevent false alarms; such as... Figure 8 As shown in the figure (this figure simulates the early warning decision-making process, with the horizontal axis representing time and the vertical axis representing the comprehensive risk index R), index The blue dashed line represents the dynamic risk threshold, which is not a fixed value but adaptively adjusts according to changes in the power grid's operating mode. The red solid line represents the real-time calculated risk index. In the left half of the graph, this index is below the threshold, indicating a safe system. As time progresses, the risk index continuously rises and eventually exceeds the dynamic threshold. When the index remains above the threshold for a preset alarm duration, the system will officially trigger a warning signal.

[0210] By utilizing the above-mentioned technical solutions of this invention, this invention constructs a voltage dynamic risk indicator system that is theoretically sound, computationally feasible, and complementary in its advantages. Combined with multi-index fusion and dynamic threshold technology, it can effectively solve the major technical problem of the lack of early warning means in current broadband oscillation defense. It has extremely high theoretical innovation value and broad engineering application prospects.

[0211] like Figure 9 As shown, a weak AC system model (short-circuit ratio SCR=2.5) with a 100MW doubly-fed wind turbine (using the standard Type4DFIG model in the PSCAD library) was built in PSCAD / EMTDC. The simulation was set at t=6.0s, applying a very short-duration (0.02s) and small-amplitude (5% of the rated value) pulse disturbance to the dq-axis current command. Subsequently, the system experienced a 22Hz subsynchronous oscillation. The performance of the indicator warning method proposed in this embodiment was compared. Based on the performance results, it is shown that this embodiment, by introducing a frequency domain weighting mechanism based on cross-correlation spectrum, solves the fundamental defects of conventional time-domain regression methods in wide-bandwidth, low signal-to-noise ratio environments, achieving a qualitative improvement in indicator performance. Furthermore, by introducing the voltage trajectory local divergence rate based on phase space reconstruction, it reveals instability precursors that conventional single-dimensional indicators cannot observe from a higher dimension of nonlinear dynamics. Simultaneously, by constructing a weighted dynamic optimization algorithm based on precursor spectrum feature identification, fuzzy expert experience is transformed into a rigorous online adaptive mechanism, realizing the intelligence of the warning system.

[0212] Furthermore, the present invention also provides an electronic device. For example... Figure 3 The diagram illustrates the hardware operating environment of an electronic device, which may include: a processor (e.g., CPU), memory, a user interface, a network interface, and a communication bus. The communication bus is used to enable communication between components. The user interface may include a display screen and an input unit such as a keyboard; optionally, the user interface may also include a standard wired interface or a wireless interface. The network interface may optionally include a standard wired interface or a wireless interface. The memory may be high-speed RAM or stable non-volatile memory, such as disk storage. Alternatively, the memory may be a storage device independent of the aforementioned processor.

[0213] Those skilled in the art will understand that Figure 3 The electronic devices shown do not constitute a limitation on electronic devices and may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0214] like Figure 3 As shown, a memory, as a type of computer storage medium, may include an operating system, a network communication module, a user interface module, and device management programs. The operating system is a program that manages and controls the hardware and software resources of electronic devices, supporting the operation of electronic devices and other software or programs. Figure 3 In the electronic device shown, the user interface is mainly used to connect to the terminal and communicate with the terminal, such as receiving user signaling data sent by the terminal; the network interface is mainly used to communicate with the backend server; the processor can be used to call the program stored in the memory and execute the steps of the method or system described above.

[0215] Furthermore, the present invention also proposes a computer-readable storage medium storing a device management program, which, when executed by a processor, implements the steps of the method or system described above.

[0216] The specific embodiments of the computer-readable storage medium of the present invention are basically the same as those of the above-described methods or systems, and will not be repeated here. Furthermore, to achieve the above objectives, the present invention also provides a computer program product, comprising: a computer program, which, when executed by a processor, implements the steps of the methods or systems described above.

[0217] Those skilled in the art will recognize that the units and algorithm steps described in conjunction with the embodiments herein can be implemented in electronic hardware or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0218] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for early warning of broadband oscillation risk in new energy power plants based on dynamic voltage indicators, characterized in that, The method includes: Instantaneous data from various monitoring points within the new energy power plant are collected to obtain electrical quantity time series data. The electrical quantity time series data is then processed according to the target sliding window to output risk characteristic quantities. The risk characteristics are nonlinearly normalized using weighted fusion technology, and a single broadband oscillation risk index reflecting the broadband oscillation level of new energy power plants is generated based on the normalization results. The difference between the single broadband oscillation risk index and the dynamic risk warning threshold is compared, and when the difference reaches the target alarm requirement, the oscillation risk warning signal is triggered and the oscillation risk warning is implemented.

2. The method for early warning of broadband oscillation risk in new energy power plants based on dynamic voltage indicators according to claim 1, characterized in that, The instantaneous value data of each monitoring point within the new energy power station is collected to obtain electrical quantity time series data. This electrical quantity time series data is then processed according to a target sliding window to output risk characteristic quantities, including: Instantaneous values ​​of voltage, active power, and reactive power at various monitoring points within the new energy power station are collected, and the collected data are reconstructed and coupled to obtain electrical quantity time series data. Bandpass filtering is performed on the time series data of electrical quantities within the target sliding window. Based on the processing results, the small disturbance components in the operation process of new energy power stations are extracted, and a linear regression model is constructed. The least squares method is reconstructed using regular weighting techniques, and the regression coefficients of the linear regression model are solved based on the reconstruction results, which are used as estimates of the dynamic sensitivity of active voltage and reactive voltage. The dynamic voltage deviation rate and voltage fluctuation spectrum entropy are calculated based on the voltage amplitude and Shannon entropy technique, and combined with the dynamic sensitivity of active voltage and reactive voltage as risk characteristic quantities.

3. The method for early warning of broadband oscillation risk in new energy power plants based on dynamic voltage indicators according to claim 2, characterized in that, The instantaneous values ​​of voltage, active power, and reactive power at each monitoring point within the new energy power station are collected, and the collected data are reconstructed and coupled for characterization to obtain electrical quantity time series data, including: After collecting instantaneous data of voltage, active power, and reactive power at the grid connection point, feeder, and busbar of the new energy power station using a measuring instrument at the sampling frequency, the data are standardized. Instantaneous state points are constructed based on the standardized processing results, and the instantaneous state points are used for embedding processing to generate state vectors in the voltage-power coupled phase space to characterize the dynamic potential of the new energy power station. Randomly select data points in the voltage-power coupling phase space and find several nearest points of the data points. Combine the data points with the nearest points to generate a local neighborhood and perform component analysis on the local neighborhood. Based on the primary and secondary directions of local neighborhood emission from the component analysis results, and by calculating the state vector describing the dynamic characteristics of the new energy power station using the orthogonal divergence rate of the state trajectory, time series data of electrical quantities are obtained.

4. The method for early warning of broadband oscillation risk in new energy power plants based on dynamic voltage indicators according to claim 3, characterized in that, The process of reconstructing the least squares method using regularized weighting techniques and solving for the regression coefficients of the linear regression model based on the reconstruction results, as estimates of the dynamic sensitivity of active power voltage and reactive power voltage, includes: Based on the processed electrical quantity time series data, a coherence function is introduced to quantitatively evaluate the degree of linear correlation between active power, reactive power and voltage at each frequency point, and a confidence weight function is constructed based on the degree of linear correlation and frequency confidence weight. The linear regression model is optimized based on the confidence weight function. The composite coefficient is used as a strong physical constraint to reveal the inherent integral constraint relationship between the real and imaginary parts of the response function of the linear time-invariant system, and the inherent integral constraint relationship is used as a constraint condition. Based on the constraints, a regularization term with composite coefficients as the core is generated. The regularization term is used to fundamentally reconstruct the least squares objective function, and the linear regression model is solved based on the reconstructed least squares objective function. The regression coefficients of the linear regression model are output based on the solution results, and the regression coefficients are used as estimates of the dynamic sensitivity of active voltage and reactive voltage.

5. The method for early warning of broadband oscillation risk in new energy power plants based on dynamic voltage indicators according to claim 4, characterized in that, The calculation of dynamic voltage deviation rate and voltage fluctuation spectrum entropy based on voltage amplitude and Shannon entropy technique includes: Calculate the average short-time voltage amplitude and the average long-time voltage amplitude within the target time window, and generate a deviation time series by obtaining the instantaneous voltage deviation based on the average short-time voltage amplitude and the average long-time voltage amplitude. The first-order difference approximation technique is used to numerically differentiate the deviation time series. The dynamic voltage deviation rate is obtained from the numerical differentiation result, which characterizes the rate at which the voltage operating point deviates from its short-term stable trajectory. The voltage signal is subjected to a fast Fourier transform to obtain the amplitude spectrum of the voltage signal over a wide frequency range, and the amplitude spectrum is divided into several continuous sub-bands according to a preset bandwidth. The proportion of spectral energy within a sub-band to the total spectral energy within a window is calculated to generate a probability distribution sequence. The Shannon entropy technique is then used to process the probability distribution sequence to output the spectral entropy of voltage fluctuations.

6. The method for early warning of broadband oscillation risk in new energy power plants based on dynamic voltage indicators according to claim 1, characterized in that, The process of using weighted fusion technology to perform nonlinear normalization on risk characteristics and generating a single broadband oscillation risk index reflecting the broadband oscillation level of new energy power plants based on the normalization results includes: The original spectral features are expanded, and the estimated values ​​of modal damping ratio and modal energy are introduced to generate new feature vectors. The new feature vectors are then fed into the online Dirichlet process hybrid model. The posterior probability of a new feature vector belonging to all known pattern clusters is calculated using an online Dirichlet process mixture model. Pattern clusters are created based on the posterior probability, and the new feature vector is used as a new pattern sample. Based on the pattern cluster and posterior probability, weight values ​​based on pattern evolution are generated, and the risk feature quantities are normalized. The normalized risk feature values ​​and weight values ​​are weighted and summed to obtain a comprehensive risk index, thus generating a single broadband oscillation risk index that reflects the broadband oscillation level of new energy power plants.

7. The method for early warning of broadband oscillation risk in new energy power plants based on dynamic voltage indicators according to claim 1, characterized in that, The dynamic risk warning threshold is obtained by fitting the probability density distribution of the comprehensive risk index calculated from the historical steady-state operation data of new energy power stations, and selecting the upper limit of the confidence interval below the target probability based on the fitting results.

8. A broadband oscillation risk early warning system for new energy power plants based on dynamic voltage indicators, used to implement the broadband oscillation risk early warning method for new energy power plants based on dynamic voltage indicators as described in any one of claims 1-7, characterized in that, The system includes: The data acquisition module is used to collect instantaneous data from various monitoring points within the new energy power plant to obtain time series data of electrical quantities. The indicator calculation module is used to process electrical quantity time series data according to the target sliding window and output risk characteristic quantities; The risk fusion module is used to perform nonlinear normalization processing on risk characteristics using weighted fusion technology, and generate a single broadband oscillation risk index that reflects the broadband oscillation level of new energy power plants based on the normalization results. The early warning decision module is used to compare the difference between a single broadband oscillation risk index and a dynamic risk early warning threshold, and to trigger an oscillation risk early warning signal when the difference result reaches the target alarm requirement, thereby implementing oscillation risk early warning.

9. An electronic device, characterized in that, The electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When executed by the processor, the computer program implements the steps of the broadband oscillation risk warning method for new energy power plants based on voltage dynamic indicators as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the broadband oscillation risk warning method for new energy power plants based on voltage dynamic indicators as described in any one of claims 1 to 7.

Images