Online prediction method for transient frequency track of power grid under coexistence of wind power low voltage ride through and off-grid

Abstract

The invention discloses an online prediction method for a transient frequency track of a power grid under coexistence of wind power low voltage ride through and off-grid, and belongs to the technical field of operation and control of a power system. A multi-source heterogeneous data fusion monitoring system is constructed, power grid and wind turbine generator data are collected, faults are recognized through an improved algorithm, and feature vectors are output; building an energy flow model based on a fault result, calculating a trajectory divergence index by using technologies such as phase-space reconstruction, estimating power vacancy, and obtaining a power unbalance sequence; designing a prediction algorithm by using the sequence, predicting a frequency trajectory in combination with an improved K-nearest neighbor algorithm and a trajectory feature library, and introducing a confidence coefficient to evaluate a correction error; and finally, establishing a three-level control response system, and implementing multi-time scale cooperative control according to a prediction result. The method can accurately predict the frequency trajectory, effectively deal with the wind power fault, improve the stability of the power grid and the wind power consumption capability, and provide powerful guarantee for the safe and stable operation of the power grid.

Description

Technical Field

[0001] This invention relates to the field of power system operation and control technology, specifically to an online prediction method for the transient frequency trajectory of the power grid under the coexistence of wind power low voltage ride-through and grid disconnection. Background Technology

[0002] In the current power system, with the continuous increase in wind power installed capacity, wind power integration has brought many challenges to grid operation.

[0003] Frequency stability issues under fault conditions: When wind power experiences low-voltage ride-through and grid disconnection faults, the power balance of the power grid is severely disrupted, leading to frequency fluctuations. Existing technologies struggle to quickly and accurately predict the transient frequency trajectory of the power grid under such complex conditions, failing to provide a reliable basis for subsequent control strategy formulation. For example, traditional frequency prediction methods are often based on simple models, failing to fully consider the diversity of wind power faults and the dynamic characteristics of the power grid, resulting in significant deviations between prediction results and actual conditions, and failing to meet the needs of real-time power grid control. Data monitoring and fault identification challenges: Power grid operation data is characterized by multi-source heterogeneity, with different types of sensors collecting data in varying formats and communication protocols, making data fusion difficult. Furthermore, in terms of fault identification, existing methods struggle to accurately distinguish between low-voltage ride-through and grid disconnection fault types, easily leading to misjudgments or omissions. For instance, some fault identification algorithms do not comprehensively extract features of voltage fluctuations and frequency changes, failing to accurately identify faults in complex interference environments, affecting subsequent fault handling and power grid recovery. Low accuracy in power imbalance calculation and processing: Accurate calculation of power imbalance is crucial for stabilizing the power grid frequency, but existing technologies are insufficient in this regard. On the one hand, the established energy flow model is not perfect and fails to fully consider the impact of factors such as inverter reactive power support and spinning reserve response delay on power imbalance. On the other hand, the method for handling measurement noise is relatively simple and cannot effectively eliminate noise interference, leading to inaccurate calculation of power imbalance and affecting the accuracy of subsequent frequency control. The control strategy lacks synergy: In terms of grid frequency control, most existing control strategies are aimed at a single time scale and lack multi-time-scale coordinated control. For example, at the millisecond, second, and minute levels, there is a lack of effective coordination between control links, making it impossible to make reasonable adjustments based on the urgency and trend of grid frequency changes. This makes it difficult to achieve stable control of the grid's transient frequency, reducing the reliability and stability of grid operation.

[0004] Based on the shortcomings of the existing technologies, we propose an online prediction method for the transient frequency trajectory of the power grid under the condition of simultaneous low voltage ride-through and disconnection of wind power. Summary of the Invention

[0005] In view of the shortcomings of the prior art, the purpose of this invention is to propose an online prediction method for the transient frequency trajectory of the power grid under the coexistence of low voltage ride-through and disconnection of wind power, which aims to ensure the safe and stable operation of the power grid under complex wind power conditions and improve the adaptability and absorption capacity of the power grid to wind power.

[0006] To achieve the above objectives, the present invention adopts the following technical solution:

[0007] A method for online prediction of grid transient frequency trajectory under the coexistence of low voltage ride-through and grid disconnection of wind power includes the following steps:

[0008] S1. Power grid condition monitoring and fault identification: Construct a monitoring system that integrates multi-source heterogeneous data to collect voltage, current, frequency and wind turbine status parameters in real time; Track voltage amplitude through an improved sliding window dynamic threshold algorithm and set dual criteria to identify fault types; Utilize a fault feature library and dynamic time bending matching algorithm to perform pattern recognition on voltage waveform distortion features and output fault feature vectors.

[0009] S2. Power imbalance calculation: Based on the fault identification results, a real-time energy flow model is built. The phase space reconstruction technology is used to embed the voltage and frequency time series. The trajectory divergence index is calculated to characterize the instability of the system. The power deficit is estimated based on the Lyapunov exponent, and the measurement noise is eliminated to obtain the power imbalance sequence.

[0010] S3. Transient frequency trajectory prediction: Using the power imbalance sequence, a prediction algorithm based on phase space trajectory similarity is designed, a trajectory feature library is constructed, and the current trajectory is matched with historical trajectories by improving the K-nearest neighbor algorithm. The reference trajectory is weighted and fused to generate the future frequency prediction trajectory. A confidence evaluation mechanism is introduced to correct the prediction error and output the frequency prediction envelope to support the generation of control strategies.

[0011] S4. A multi-timescale collaborative control strategy is generated. Based on the frequency trajectory prediction results, a three-level control response system is established. The millisecond level is responded by supercapacitor energy storage, the second level is switched by unloading circuit, and the minute level coordinates the AGC regulation of conventional units. A control strategy based on trajectory trend similarity is designed to ensure operational safety and achieve stable control of the power grid transient frequency.

[0012] Furthermore, in S1, the method for constructing a multi-source heterogeneous data fusion monitoring system is as follows:

[0013] S111. Hardware deployment: Install voltage sensors, current sensors, frequency sensors, and various sensors for monitoring the status parameters of wind turbines at key nodes of the power grid, such as substations and wind farm access points. These sensors are responsible for collecting relevant physical quantity data in real time. For example, a capacitive voltage transformer can be used as a voltage sensor, which can accurately measure the grid voltage, and a Rogowski coil current sensor can be used as a current sensor.

[0014] S112. Data transmission and aggregation: Different types of sensors collect data with different formats and communication protocols. To achieve data fusion, edge computing devices are used to perform preliminary processing and conversion of the field data. The edge computing devices parse and encapsulate the collected data and transmit it to the data center through optical fiber, wireless communication and other means. In the data center, data fusion technology is used to integrate data such as voltage, current, frequency and wind turbine status parameters from different sensors to form a unified dataset.

[0015] Furthermore, in step S1, the voltage amplitude is tracked using an improved sliding window dynamic threshold algorithm:

[0016] S121. Sliding window setting: The sliding window size is determined based on the frequency characteristics of the grid voltage fluctuation and the sampling frequency. The window slides on the time series with a fixed step size.

[0017] S122. Calculate window statistics. Within each sliding window, perform statistical analysis on the voltage amplitude sequence. First, calculate the mean of the voltage amplitude within the window, and then calculate the standard deviation.

[0018] S123. An adaptive adjustment factor is introduced. This factor dynamically adjusts the threshold based on historical power grid operating data and real-time fluctuations. The adjustment formula is as follows:

[0019]

[0020] Among them, T d Indicates a dynamic threshold; This represents the mean voltage amplitude within the window; k is an empirical coefficient, usually taken as 2; α is an adaptive adjustment factor; σ is the standard deviation of the voltage amplitude within the window.

[0021] S124. The adjustment factor is correlated with the power grid status. Real-time and historical data of the power grid are collected to determine the operating status of the power grid. If the current sensor data shows that the power grid load changes drastically, or the frequency sensor shows that the frequency fluctuates greatly, it is inferred that the voltage fluctuation is aggravated. At this time, the α value is increased to make the dynamic threshold more sensitive and to capture abnormal voltage changes in time. Conversely, if the data shows that the power grid is operating relatively stably and the voltage fluctuation is small, the α value is decreased to avoid misjudgment due to the threshold being too sensitive.

[0022] Furthermore, in S1, the dual criteria include voltage amplitude criteria and frequency change rate criteria;

[0023] The voltage amplitude criterion is: when the voltage amplitude within the sliding window is less than the dynamic threshold T. d When the proportion of the number of sampling points to the total number of points in the window exceeds a set proportion, a preliminary judgment is made that a fault has occurred, and the time at this time is recorded. This criterion is directly based on the dynamic threshold obtained by the improved sliding window dynamic threshold algorithm, which ensures that abnormal voltage drops can be detected in a timely manner.

[0024] The frequency change rate criterion is as follows: the power grid frequency change rate is calculated in real time by differential calculation of the frequency sampling sequence; a frequency change rate threshold is set, and if the voltage amplitude is continuously lower than the threshold while the frequency change rate is greater than the set frequency change rate threshold, then a fault is judged to have occurred in the power grid. Based on the specific numerical range of the voltage amplitude and the frequency change rate, the severity of the fault is initially judged.

[0025] The formula for calculating the difference is:

[0026]

[0027] in, The real-time frequency change rate is represented by Δt, which is the sampling time interval; f n It is the frequency sample value at the current moment, f n-1 It is the frequency sample value from the previous moment;

[0028] Furthermore, in S1, a pattern recognition method is performed using a fault feature library and a dynamic time bending matching algorithm:

[0029] S131. Construct a fault feature library, collect a large amount of historical fault data, including voltage waveform data of different types of low voltage ride-through faults and grid disconnection faults, preprocess these voltage waveform data, such as filtering and noise reduction, extract their key features, form standard fault feature vectors, and store them in the fault feature library; each fault feature vector corresponds to a fault type label, severity score and timestamp, which facilitates subsequent matching and querying.

[0030] S132. The dynamic time bending matching algorithm dynamically bends the fault voltage waveform sequence acquired in real time with the standard voltage waveform sequence corresponding to each feature vector in the fault feature library; it calculates the dynamic time bending distance between the two; the formula for calculating the DTW distance between the two is as follows:

[0031]

[0032] Where A is the real-time acquired fault voltage waveform sequence; B is the standard voltage waveform sequence corresponding to each feature vector in the fault feature library; D(A,B) is the dynamic time bending distance between A and B; m is the length of the real-time acquired voltage waveform sequence A, i.e., the number of sampling points in the sequence; n is the length of the voltage waveform sequence B corresponding to the standard fault feature vector in the fault feature library; ω(i,j) is the path weight, satisfying...

[0033] The cumulative path weight is typically used for calculation; a i b represents the voltage amplitude at the i-th sampling time in sequence A; i It is the voltage amplitude at the j-th sampling time in the B sequence; d(ai,b) j (ai - bi)2 is the distance metric between two points;

[0034] Calculate all D(A,B) in the A and B libraries, compare the minimum distance with a set threshold; if the minimum distance is greater than the set threshold, find the matching fault type, and output a standardized fault feature vector containing fault type label, severity score and current timestamp.

[0035] Furthermore, S2 includes:

[0036] S21. Establish a real-time energy flow model. Based on the fault identification results in S1, build a real-time energy flow model:

[0037] For low-voltage ride-through conditions, considering the power imbalance caused by inverter reactive power support, in the equivalent circuit model, the reactive power is converted into equivalent active power through inverter control strategies and related circuit parameters. The active power imbalance at this point is:

[0038] ΔP1=Pw-Pl+ΔP Q ;

[0039] ΔP1 represents the imbalance of active power under low voltage ride-through conditions; P w For the fan output; Pl is the load demand; ΔP Q Inverter reactive power support;

[0040] For off-grid operation, an islanded system model is constructed, considering the spinning reserve response delay; then the active power imbalance is:

[0041] ΔP2=P w -P l -P r (tt d );

[0042] ΔP2 represents the active power imbalance under grid disconnection conditions; P r To set the rotational reserve power; td t represents the response delay time; t represents the current time.

[0043] S22. Phase space reconstruction technology: Acquiring voltage and frequency time series, selecting embedding parameters: To transform these one-dimensional time series into multi-dimensional space vectors that reflect the dynamic characteristics of the system, phase space reconstruction is required. First, the embedding dimension is selected using the Cao method. The Cao method calculates the changes of a series of statistics with the embedding dimension; when the statistics no longer change significantly with the embedding dimension, the corresponding embedding dimension value is the appropriate embedding dimension. The time delay is determined using the mutual information method. The mutual information function of the voltage or frequency time series is calculated, and the value is taken when the mutual information function first decreases to its maximum value. This is a time delay;

[0044] Reconstructing phase space vectors: Based on the selected embedding dimension and time delay, the voltage sequence and frequency sequence are reconstructed in phase space to obtain the corresponding voltage change trajectory sequence and frequency change trajectory sequence. These vectors describe the state of the system at different times in phase space, and the changes in the phase space trajectory can reflect the dynamic characteristics of the system.

[0045] S23. Calculate the trajectory divergence index; find the nearest neighbor point and calculate the trajectory divergence index using the local Lyapunov exponent method; in the reconstructed phase space vector, for any point... Find the x nearest neighbors of this point. Calculate the initial distance After δ-step evolution, the distance becomes Local Lyapunov index The trajectory divergence index is the average of all local Lyapunov indices: This index is used to characterize the degree of system instability. It reflects the local divergence or convergence of the trajectory near a point in phase space. The larger the value of λ, the more unstable the system. T represents the total length of the voltage or frequency time series.

[0046] S24. Using the Lyapunov exponent to estimate the power deficit and eliminate measurement noise, according to the Lyapunov stability theory, the power deficit is related to the average value of the local Lyapunov exponent:

[0047] ΔP=Cλ;

[0048] ΔP is the power deficit; C is the proportionality coefficient;

[0049] In actual measurements, power deficit data is affected by noise. To obtain a more accurate power imbalance sequence, a moving average algorithm with an adaptive forgetting factor is introduced to eliminate measurement noise. The measured power deficit sequence is {ΔP}. m1,ΔP m2 ,…,ΔP mT}, ΔP mi (i = 1, 2, ..., T) represents the power deficit measured at different times. The filtered power imbalance sequence is recursively calculated as follows:

[0050] ΔP f (t)=αΔP f (t-1)+(1-α)ΔP m (t);

[0051] ΔP f (t) represents the power imbalance sequence at time t after filtering; ΔP f (t-1) represents the power imbalance sequence at time t-1 after filtering; α is the forgetting factor.

[0052] By continuously adjusting the value of α and optimizing the filtering effect based on the data's changing trends and noise characteristics, a time-scale-aligned power imbalance sequence {ΔP} is finally obtained. f1 ,ΔP f2 ,…,ΔP fT}, ΔP fi (i = 1, 2, ..., T) represents the power imbalance at different times after filtering; this provides accurate data support for subsequent transient frequency trajectory prediction.

[0053] Furthermore, S3 includes:

[0054] S31. Construct a trajectory feature library, comprehensively collect various historical fault cases, covering different fault scenarios, including but not limited to low voltage ride-through faults and grid disconnection faults of different degrees, as well as situations under various power grid operating conditions; each case records the power imbalance sequence and the corresponding frequency change trajectory sequence in detail. The power imbalance sequence reflects the dynamic changes in power during the fault, while the frequency change trajectory sequence reflects the frequency fluctuations over time. Normalize these data, map the power imbalance and frequency data to the [0,1] interval, classify and store the data according to fault type, fault severity, etc., and construct the trajectory feature library;

[0055] S32. Improved k-nearest neighbor algorithm for trajectory matching, for obtaining time-stamped aligned power imbalance sequence {ΔP} f1 ,ΔP f2 ,…,ΔP fT An improved k-nearest neighbor algorithm is used to perform real-time pattern matching in phase space, calculating the Hausdorff distance between the current sequence and various typical trajectories in the feature library. The formula is as follows:

[0056]

[0057] Where, F={ΔP f1 ,ΔP f2 ,…,ΔP fT};M={ΔP m1 ,ΔP m2 ,…,ΔP mT} is the power imbalance sequence corresponding to a typical trajectory in the trajectory feature library, ΔP mj (j=1,2,…,T) is the power imbalance value at the j-th time in the sequence; First, for each i, find the value within the range of j that makes... The minimum value is found in sequence F, which is the point with the smallest distance to the i-th point in sequence M; then the maximum value is taken from these minimum values. Similarly;

[0058] The top K samples with the smallest Hausdorff distance are selected as reference trajectories and their corresponding frequency trajectories. The value of K is determined by cross-validation. Prediction experiments are conducted on historical data under different K values, and the K value with the smallest prediction error is selected.

[0059] S33. Weighted fusion of reference trajectories generates predicted trajectories, calculating the weight ω of each reference trajectory based on the Hausdorff distance. k The formula is:

[0060]

[0061] The power imbalance sequence of the d-th reference trajectory is a historical data sequence selected from the trajectory feature library that is similar to the current system power imbalance sequence.

[0062] The subsequent evolution paths of the reference trajectory are weighted and fused to generate the future frequency prediction trajectory. The calculation formula is as follows:

[0063]

[0064] f p (t) represents the value of the future frequency prediction trajectory at time t; frd represents the frequency trajectory corresponding to the d-th reference trajectory, which is a sequence of power imbalances relative to the d-th reference trajectory. The corresponding frequency change sequence is used to provide frequency information during the weighted fusion process;

[0065] S34. Introduce a confidence assessment mechanism to correct prediction errors, and calculate the root mean square error (RMSE) between the predicted trajectory and real-time measurements:

[0066]

[0067] f m(t) represents the real-time measured frequency. A root mean square error threshold is set. If the RMSE is greater than the threshold, a prediction correction mechanism is triggered. The correction method can be to reselect the reference trajectory or adjust the weight of the reference trajectory. Then, the prediction trajectory is recalculated, and finally, the frequency prediction envelope with probability density interval is output.

[0068] Furthermore, in S4, the millisecond-level response is handled by the supercapacitor energy storage. Based on the frequency prediction envelope obtained in S3, combined with the power balance relationship, the power deficit curve is predicted. Considering constraints such as the supercapacitor's charging and discharging efficiency and lifetime, an optimization model is established, with the objective function being:

[0069] min(|ΔP p (t)-P sc (t)|);

[0070] P sc (t) represents the charging and discharging power of the supercapacitor at time t. The constraints include voltage and current limits as well as depth of charge and discharge limits for the supercapacitor.

[0071] The optimal charging and discharging power command is obtained by solving the optimization model. Then, the charging and discharging power Psc of the supercapacitor is calculated according to the power balance equation: Psc = ΔP p (t) Combined with the voltage V of the supercapacitor energy storage system sc and current I sc Relationship, P sc =V sc I sc ; and the voltage conversion relationship of the Buck / Boost converter. V o V is the output voltage of the converter. i The input voltage is given by D, and the duty cycle is given by the formula:

[0072]

[0073] The duty cycle of the Buck / Boost converter is dynamically adjusted using real-time feedback control technology. The duty cycle is fine-tuned based on the real-time voltage and current of the supercapacitor to ensure that the supercapacitor can respond quickly and accurately to changes in power deficit.

[0074] Furthermore, in S4, the second-level response is achieved through the graded switching of the unloading circuit. A fuzzy control algorithm is used to optimize the resistor switching combination. The frequency deviation predicted in S3 and the power imbalance calculated in S2 are used as the inputs to the fuzzy controller, and the output is the switching state of the unloading resistor. Fuzzy sets are defined, such as the fuzzy set of power imbalance as {negative large, negative small, zero, positive small, positive large}, the fuzzy set of frequency deviation as {negative large, negative small, zero, positive small, positive large}, and the fuzzy set of unloading resistor switching state as {all unloaded, partially unloaded, no operation, partially loaded, all loaded}. A fuzzy rule base is established, such as "if the power imbalance is positive large and the frequency deviation is positive large, then all unloading resistors are loaded". To improve the adaptability of fuzzy control, an adaptive fuzzy control algorithm is used to adjust the fuzzy rules online according to the real-time operation data of the power grid. The specific resistor switching combination is obtained through fuzzy inference and defuzzification.

[0075] Furthermore, in S4, the minute-level response coordinates the AGC (Automatic Generation Control) regulation of the conventional unit. Based on the frequency trajectory trend predicted in S3, the required power adjustment for the conventional unit is calculated. Considering constraints such as the ramp rate and regulation range of the conventional unit, the required power adjustment is:

[0076] ΔP A =r×t A ;

[0077] r is the power regulation rate of a conventional unit, t A To adjust the time, and to avoid over- or under-adjustment, the adjustment time t is dynamically adjusted based on the deviation between the predicted frequency and the rated frequency. A ;

[0078] Unit coordination control involves sending regulation commands to conventional units through the control system to achieve power regulation; establishing a coordination control mechanism among conventional units, and rationally allocating regulation tasks based on factors such as the capacity and regulation performance of each unit to avoid regulation conflicts between different units and ensure the stable operation of the entire power grid; for example, prioritizing the regulation of units with good regulation performance and fast response speed, while also considering the economic operation of the units, so that the regulation process can minimize power generation costs while ensuring the stability of the power grid frequency.

[0079] Furthermore, in S4, a method for selecting a control strategy based on trajectory trend similarity is designed:

[0080] Similarity calculation: The similarity between the predicted trajectory in S3 and the trajectory trend of historical successful control cases is calculated. The reciprocal of the dynamic time curvature distance between the predicted trajectory and the trajectory of historical successful control cases is used as the similarity index. The larger the value, the higher the similarity.

[0081] In terms of strategy selection, when the similarity index is greater than the preset value, the preset successful control strategy is automatically invoked; if the similarity index is less than the preset value, the control strategy is re-formulated based on the current power grid operating status and predicted trajectory through an expert system combined with reinforcement learning algorithms to ensure operational safety and achieve stable control of the power grid transient frequency.

[0082] In summary, due to the adoption of the above technical solution, the beneficial technical effects of the invention are as follows:

[0083] Synergistic Effect of the Overall Technical Solution: The online prediction method for grid transient frequency trajectory under the coexistence of low-voltage ride-through and grid disconnection of wind power proposed in this invention integrates closely related and collaborative technical steps. From grid condition monitoring and fault identification to power imbalance calculation, transient frequency trajectory prediction, and multi-timescale collaborative control strategy generation, a complete and efficient system is formed. This system effectively solves the frequency instability problem caused by low-voltage ride-through and grid disconnection faults after wind power is connected to the grid, significantly improving the stability and reliability of the grid.

[0084] Enhancing the power grid's ability to cope with wind power faults: In the context of frequent wind power faults, accurate and timely fault identification is the primary link in ensuring power grid stability. This invention constructs a monitoring system that integrates multi-source heterogeneous data, enabling comprehensive collection of various data. Combined with an improved sliding window dynamic threshold algorithm and dual criteria, as well as a fault feature database and dynamic time-warping matching algorithm, it accurately identifies low-voltage ride-through and grid disconnection fault types. This allows the power grid to respond quickly to faults, providing a correct basis for subsequent power imbalance calculations and frequency control, and significantly reducing the impact of faults on the power grid.

[0085] Enhancing the accuracy and reliability of frequency trajectory prediction: Accurate calculation of power imbalance is crucial for accurate frequency trajectory prediction. This invention builds a real-time energy flow model based on fault identification results, fully considering factors such as inverter reactive power support and spinning reserve response delay. It combines phase space reconstruction technology and Lyapunov exponents to estimate power deficit and effectively eliminates measurement noise, resulting in a high-precision power imbalance sequence. Based on this, a prediction algorithm based on phase space trajectory similarity and an improved K-nearest neighbor algorithm are used, combined with a trajectory feature library, for prediction. A confidence assessment mechanism is introduced to correct errors, ultimately outputting a frequency prediction envelope with a probability density interval. This method significantly improves the accuracy and reliability of frequency trajectory prediction, providing grid operators with more valuable information and helping to formulate reasonable control strategies in advance to prevent frequency instability accidents.

[0086] Achieving precise frequency control across multiple time scales: The multi-time scale coordinated control strategy is one of the core advantages of this invention in ensuring grid frequency stability. In millisecond-level response, the supercapacitor energy storage, based on the predicted power deficit curve, dynamically adjusts the duty cycle of the Buck / Boost converter to quickly replenish or absorb power, suppressing rapid frequency fluctuations. In second-level response, the unloading circuit uses a fuzzy control algorithm to optimize the resistor input combination, performing precise control based on frequency deviation and power imbalance to further stabilize the frequency. Minute-level response coordinates the AGC regulation of conventional units, considering factors such as the unit's ramp rate and adjustment range to achieve long-term stable frequency regulation. Simultaneously, the control strategy selection mechanism based on trajectory trend similarity can intelligently select or formulate control strategies based on predicted trajectories and historical success cases, ensuring precise control of the grid frequency under different operating conditions, effectively avoiding large frequency fluctuations and guaranteeing the safe and stable operation of the grid.

[0087] Improving the grid's adaptability and absorption capacity for wind power: As the proportion of wind power in the power grid continues to increase, the grid's adaptability and absorption capacity for wind power are crucial. This invention, through precise fault identification, accurate frequency trajectory prediction, and effective control strategies, can quickly handle wind power low-voltage ride-through and grid disconnection faults, stabilize grid frequency, and reduce grid fluctuations and power outages caused by wind power faults. This not only improves the grid's capacity to accommodate wind power but also reduces the negative impact of wind power on the grid, promotes the large-scale and efficient utilization of wind power, and accelerates the power system's transition to clean energy. Attached Figure Description

[0088] Figure 1 This is a flowchart of an online prediction method for the transient frequency trajectory of the power grid under the condition of simultaneous low voltage ride-through and disconnection of wind power. Detailed Implementation

[0089] To make the objectives, technical solutions, and advantages of the invention clearer, the invention will be further described in detail below with reference to embodiments. It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the scope of the invention.

[0090] like Figure 1 As shown, an online prediction method for the transient frequency trajectory of the power grid under the coexistence of low-voltage ride-through and grid disconnection of wind power includes the following steps:

[0091] A method for online prediction of grid transient frequency trajectory under the coexistence of low voltage ride-through and grid disconnection of wind power includes the following steps:

[0092] S1. Power Grid Condition Monitoring and Fault Identification: A multi-source heterogeneous data fusion condition monitoring system is constructed to collect three-phase voltage, current waveforms, frequency change rate, and wind turbine operating status parameters in real time. Through an improved sliding window dynamic threshold algorithm, the voltage amplitude is tracked and analyzed in real time, and dual criteria are set. At the same time, using the established fault feature library and dynamic time curvature matching algorithm, the voltage waveform distortion features are pattern recognized to accurately distinguish between low voltage ride-through and grid disconnection fault types, and output a standardized fault feature vector containing fault type label, severity score, and timestamp.

[0093] S2. Power imbalance calculation: Based on fault identification results, a real-time energy flow model is established, including turbine output, load demand, and energy storage status. For low voltage ride-through conditions, an equivalent circuit model is constructed, considering the power imbalance caused by inverter reactive power support. For grid disconnection conditions, an islanded system model is constructed, taking into account spinning reserve response delay. Based on this, phase space reconstruction technology is used to embed the collected voltage and frequency time series into a three-dimensional phase space, and the instability of the system is quantitatively characterized by calculating the trajectory divergence index. The Lyapunov exponent is used to estimate the power deficit in real time, and a moving average algorithm with an adaptive forgetting factor is introduced to eliminate the influence of measurement noise, ultimately obtaining a time-scale-aligned power imbalance sequence.

[0094] S3. Transient Frequency Trajectory Prediction: With the key data of power imbalance, a prediction algorithm based on phase space trajectory similarity was designed, abandoning the use of traditional neural networks. A trajectory feature library containing historical fault cases was constructed, and an improved K-nearest neighbor algorithm was used for real-time pattern matching in phase space. For the trajectory of the current system, the Hausdorff distance between it and various typical trajectories in the feature library was calculated, and the top K samples with the highest similarity were selected as reference trajectories. By weighted fusion of the subsequent evolution paths of these reference trajectories, the future frequency prediction trajectory was generated. At the same time, a trajectory confidence evaluation mechanism was introduced. When the root mean square error between the predicted trajectory and the real-time measurement value exceeded a set threshold, prediction correction was triggered, and the final output was a frequency prediction envelope with a probability density interval.

[0095] S4. Multi-timescale collaborative control strategy generation: Based on frequency trajectory prediction, a three-level control response system (millisecond-level response, second-level response, and minute-level response) is established to achieve multi-timescale collaborative control. The millisecond-level response is handled by supercapacitor energy storage, dynamically adjusting the duty cycle of the Buck / Boost converter according to the predicted power deficit curve; the second-level response is achieved through graded switching of the unloading circuit, using a fuzzy control algorithm to optimize the resistor input combination; the minute-level response coordinates the AGC regulation of conventional units. Furthermore, a control strategy selection mechanism based on trajectory trend similarity is designed; when the matching degree between the predicted trajectory and historical successful control cases exceeds a preset value, a preset strategy is automatically invoked.

[0096] In S1, the method for constructing a multi-source heterogeneous data fusion monitoring system is as follows:

[0097] S111. Hardware deployment: Install voltage sensors, current sensors, frequency sensors, and various sensors for monitoring the status parameters of wind turbines at key nodes of the power grid, such as substations and wind farm access points. These sensors are responsible for collecting relevant physical quantity data in real time. For example, a capacitive voltage transformer can be used as a voltage sensor, which can accurately measure the grid voltage, and a Rogowski coil current sensor can be used as a current sensor.

[0098] S112. Data transmission and aggregation: Different types of sensors collect data with different formats and communication protocols. To achieve data fusion, edge computing devices are used to perform preliminary processing and conversion of the field data. The edge computing devices parse and encapsulate the collected data and transmit it to the data center through optical fiber, wireless communication (such as 4G / 5G), etc. In the data center, data fusion technology is used to integrate data such as voltage, current, frequency and wind turbine status parameters from different sensors to form a unified dataset.

[0099] In step S1, the voltage amplitude is tracked using an improved sliding window dynamic threshold algorithm:

[0100] S121. Sliding window setting: The sliding window size is determined based on the frequency characteristics of the grid voltage fluctuation and the sampling frequency. The window slides on the time series with a fixed step size (usually 1 sampling point). For example, if the sampling frequency is 10kHz and the grid voltage fluctuation is mainly concentrated in the low frequency band, n=100 can be set, that is, the window contains 100 sampling points and the corresponding duration is 10ms.

[0101] S122. Calculate window statistics. Within each sliding window, perform statistical analysis on the voltage amplitude sequence. First, calculate the mean of the voltage amplitude within the window, and then calculate the standard deviation.

[0102] S123. An adaptive adjustment factor is introduced. This factor dynamically adjusts the threshold based on historical power grid operating data and real-time fluctuations. The adjustment formula is as follows:

[0103]

[0104] Among them, T d Indicates a dynamic threshold; This represents the mean voltage amplitude within the window; k is an empirical coefficient, usually taken as 2; α is an adaptive adjustment factor; σ is the standard deviation of the voltage amplitude within the window.

[0105] S124. The adjustment factor is correlated with the power grid status. Real-time and historical data of the power grid are collected to determine the operating status of the power grid. If the current sensor data shows that the power grid load changes drastically, or the frequency sensor shows that the frequency fluctuates greatly, it is inferred that the voltage fluctuation is aggravated. At this time, the α value is increased to make the dynamic threshold more sensitive and to capture abnormal voltage changes in time. Conversely, if the data shows that the power grid is operating relatively stably and the voltage fluctuation is small, the α value is decreased to avoid misjudgment due to the threshold being too sensitive.

[0106] In S1, the dual criteria include voltage amplitude criteria and frequency change rate criteria;

[0107] The voltage amplitude criterion is: when the voltage amplitude within the sliding window is less than the dynamic threshold T. d When the proportion of the number of sampling points to the total number of points in the window exceeds a set proportion (e.g., p = 0.3), a preliminary judgment is made that a fault has occurred, and the time at this time is recorded. This criterion is directly based on the dynamic threshold obtained by the improved sliding window dynamic threshold algorithm, which ensures that abnormal voltage drops can be detected in a timely manner.

[0108] The frequency change rate criterion is as follows: the power grid frequency change rate is calculated in real time by differential calculation of the frequency sampling sequence; a frequency change rate threshold is set (e.g., 0.5 Hz / s). If the voltage amplitude is continuously lower than the threshold while the frequency change rate is greater than the set frequency change rate threshold, it is judged that a fault has occurred in the power grid. The severity of the fault is preliminarily judged based on the specific numerical range of the voltage amplitude and the frequency change rate.

[0109] The formula for calculating the difference is:

[0110]

[0111] in, The real-time frequency change rate is represented by Δt, which is the sampling time interval; f n It is the frequency sample value at the current moment, f n-1 It is the frequency sample value from the previous moment;

[0112] The method for pattern recognition using a fault feature database and a dynamic time warp matching algorithm:

[0113] S131. Construct a fault feature library by collecting a large amount of historical fault data, including voltage waveform data of different types of low-voltage ride-through faults and grid disconnection faults. Preprocess these voltage waveform data, such as filtering and denoising, to extract their key features (voltage amplitude change trend, waveform distortion degree, duration, etc.) to form standard fault feature vectors, which are then stored in the fault feature library. Each fault feature vector corresponds to a fault type label, severity score, and timestamp for easy matching and querying later.

[0114] S132. The dynamic time bending matching algorithm dynamically bends the fault voltage waveform sequence acquired in real time with the standard voltage waveform sequence corresponding to each feature vector in the fault feature library; it calculates the dynamic time bending distance between the two; the formula for calculating the DTW distance between the two is as follows:

[0115]

[0116] Where A is the real-time acquired fault voltage waveform sequence; B is the standard voltage waveform sequence corresponding to each feature vector in the fault feature library; D(A,B) is the dynamic time bending distance between A and B; m is the length of the real-time acquired voltage waveform sequence A, i.e., the number of sampling points in the sequence; n is the length of the voltage waveform sequence B corresponding to the standard fault feature vector in the fault feature library; ω(i,j) is the path weight, satisfying...

[0117] The cumulative path weight method is typically used for calculation.

[0118] a i b represents the voltage amplitude at the i-th sampling time in sequence A; i It is the voltage amplitude at the j-th sampling time in the B sequence; d(ai,b) j (ai - bi)2 is the distance metric between two points;

[0119] Calculate all D(A,B) in the A and B libraries, compare the minimum distance with a set threshold; if the minimum distance is greater than the set threshold, find the matching fault type, and output a standardized fault feature vector containing fault type label, severity score and current timestamp.

[0120] S2 includes:

[0121] S21. Establish a real-time energy flow model. Based on the fault identification results in S1, build a real-time energy flow model:

[0122] For low-voltage ride-through conditions, considering the power imbalance caused by inverter reactive power support, in the equivalent circuit model, the reactive power is converted into equivalent active power through inverter control strategies and related circuit parameters. The active power imbalance at this point is:

[0123] ΔP1=Pw-Pl+ΔP Q ;

[0124] ΔP1 represents the imbalance of active power under low voltage ride-through conditions; P w For the fan output; Pl is the load demand; ΔP Q Inverter reactive power support;

[0125] For off-grid operation, an islanded system model is constructed, considering the spinning reserve response delay; then the active power imbalance is:

[0126] ΔP2=P w -P l -P r (tt d );

[0127] ΔP2 represents the active power imbalance under grid disconnection conditions; P r To set the rotational reserve power; t d t represents the response delay time; t represents the current time.

[0128] S22. Phase space reconstruction technology: Acquiring voltage and frequency time series, selecting embedding parameters: To transform these one-dimensional time series into multi-dimensional space vectors that reflect the dynamic characteristics of the system, phase space reconstruction is required. First, the embedding dimension is selected using the Cao method. The Cao method calculates the changes of a series of statistics with the embedding dimension; when the statistics no longer change significantly with the embedding dimension, the corresponding embedding dimension value is the appropriate embedding dimension. The time delay is determined using the mutual information method. The mutual information function of the voltage or frequency time series is calculated, and the value is taken when the mutual information function first decreases to its maximum value. This is a time delay;

[0129] Reconstructing phase space vectors: Based on the selected embedding dimension and time delay, the voltage sequence and frequency sequence are reconstructed in phase space to obtain the corresponding voltage change trajectory sequence and frequency change trajectory sequence. These vectors describe the state of the system at different times in phase space, and the changes in the phase space trajectory can reflect the dynamic characteristics of the system.

[0130] S23. Calculate the trajectory divergence index; find the nearest neighbor point and calculate the trajectory divergence index using the local Lyapunov exponent method; in the reconstructed phase space vector, for any point... Find the x nearest neighbors of this point. Calculate the initial distance After δ-step evolution, the distance becomes Local Lyapunov index The trajectory divergence index is the average of all local Lyapunov indices: This index is used to characterize the degree of system instability. It reflects the local divergence or convergence of the trajectory near a point in phase space. The larger the value of λ, the more unstable the system. T represents the total length of the voltage or frequency time series.

[0131] S24. Using the Lyapunov exponent to estimate the power deficit and eliminate measurement noise, according to the Lyapunov stability theory, the power deficit is related to the average value of the local Lyapunov exponent:

[0132] ΔP=Cλ;

[0133] ΔP is the power deficit; C is the proportionality coefficient;

[0134] In actual measurements, power deficit data is affected by noise. To obtain a more accurate power imbalance sequence, a moving average algorithm with an adaptive forgetting factor is introduced to eliminate measurement noise. The measured power deficit sequence is {ΔP}. m1 ,ΔP m2 ,…,ΔP mT}, ΔP mi (i = 1, 2, ..., T) represents the power deficit measured at different times. The filtered power imbalance sequence is recursively calculated as follows:

[0135] ΔP f (t)=αΔP f (t-1)+(1-α)ΔP m (t);

[0136] ΔP f (t) represents the power imbalance sequence at time t after filtering; ΔP f (t-1) represents the power imbalance sequence at time t-1 after filtering; α is the forgetting factor.

[0137] By continuously adjusting the value of α and optimizing the filtering effect based on the data's changing trends and noise characteristics, a time-scale-aligned power imbalance sequence {ΔP} is finally obtained. f1 ,ΔP f2 ,…,ΔP fT}, ΔP fi (i = 1, 2, ..., T) represents the power imbalance at different times after filtering; this provides accurate data support for subsequent transient frequency trajectory prediction.

[0138] The S3 mentioned above includes:

[0139] S31. Construct a trajectory feature library, comprehensively collect various historical fault cases, covering different fault scenarios, including but not limited to low voltage ride-through faults and grid disconnection faults of different degrees, as well as situations under various power grid operating conditions; each case records the power imbalance sequence and the corresponding frequency change trajectory sequence in detail. The power imbalance sequence reflects the dynamic changes in power during the fault, while the frequency change trajectory sequence reflects the frequency fluctuations over time. Normalize these data, map the power imbalance and frequency data to the [0,1] interval, classify and store the data according to fault type, fault severity, etc., and construct the trajectory feature library;

[0140] S32. Improved k-nearest neighbor algorithm for trajectory matching, for obtaining time-stamped aligned power imbalance sequence {ΔP} f1 ,ΔP f2 ,…,ΔP fT An improved k-nearest neighbor algorithm is used to perform real-time pattern matching in phase space, calculating the Hausdorff distance between the current sequence and various typical trajectories in the feature library. The formula is as follows:

[0141]

[0142] Where, F={ΔP f1 ,ΔP f2 ,…,ΔP fT};M={ΔP m1 ,ΔP m2 ,…,ΔP mT} is the power imbalance sequence corresponding to a typical trajectory in the trajectory feature library, ΔP mj (j=1,2,…,T) is the power imbalance value at the j-th time in the sequence; First, for each i, find the value within the range of j that makes... The minimum value is found in sequence F, which is the point with the smallest distance to the i-th point in sequence M; then the maximum value is taken from these minimum values. Similarly;

[0143] The top K samples with the smallest Hausdorff distance are selected as reference trajectories and their corresponding frequency trajectories. The value of K is determined by cross-validation. Prediction experiments are conducted on historical data under different K values, and the K value with the smallest prediction error is selected.

[0144] S33. Weighted fusion of reference trajectories generates predicted trajectories, calculating the weight ω of each reference trajectory based on the Hausdorff distance. k The formula is:

[0145]

[0146] The power imbalance sequence of the d-th reference trajectory is a historical data sequence selected from the trajectory feature library that is similar to the current system power imbalance sequence.

[0147] The subsequent evolution paths of the reference trajectory are weighted and fused to generate the future frequency prediction trajectory. The calculation formula is as follows:

[0148]

[0149] f p (t) represents the value of the future frequency prediction trajectory at time t; f rd The frequency trajectory corresponding to the d-th reference trajectory is a sequence of power imbalances relative to the d-th reference trajectory. The corresponding frequency change sequence is used to provide frequency information during the weighted fusion process;

[0150] S34. Introduce a confidence assessment mechanism to correct prediction errors, and calculate the root mean square error (RMSE) between the predicted trajectory and real-time measurements:

[0151]

[0152] f m (t) represents the real-time measured frequency. A root mean square error threshold is set. If the RMSE is greater than the threshold, a prediction correction mechanism is triggered. The correction method can be to reselect the reference trajectory or adjust the weight of the reference trajectory. Then, the prediction trajectory is recalculated, and finally, the frequency prediction envelope with probability density interval is output.

[0153] In step S4, the millisecond-level response is handled by the supercapacitor energy storage. Based on the frequency prediction envelope obtained in step S3 and combined with the power balance relationship, the power deficit curve is predicted. Considering constraints such as the supercapacitor's charging and discharging efficiency and lifetime, an optimization model is established with the objective function as follows:

[0154] min(|ΔP p (t)-P sc (t)|);

[0155] P sc (t) represents the charging and discharging power of the supercapacitor at time t. The constraints include voltage and current limits as well as depth of charge and discharge limits for the supercapacitor.

[0156] The optimal charging and discharging power command is obtained by solving the optimization model. Then, the charging and discharging power Psc of the supercapacitor is calculated according to the power balance equation: P sc =ΔP p (t) Combined with the voltage V of the supercapacitor energy storage system sc and current I sc Relationship, Psc =V sc I sc ; and the voltage conversion relationship of the Buck / Boost converter. V o V is the output voltage of the converter. i The input voltage is given by D, and the duty cycle is given by the formula:

[0157]

[0158] The duty cycle of the Buck / Boost converter is dynamically adjusted using real-time feedback control technology. The duty cycle is fine-tuned based on the real-time voltage and current of the supercapacitor to ensure that the supercapacitor can respond quickly and accurately to changes in power deficit.

[0159] In S4, the second-level response is achieved through the graded switching of the unloading circuit. A fuzzy control algorithm is used to optimize the resistor switching combination. The frequency deviation predicted in S3 and the power imbalance calculated in S2 are used as the inputs of the fuzzy controller, and the output is the switching state of the unloading resistor. Fuzzy sets are defined, such as the fuzzy set of power imbalance as {negative large, negative small, zero, positive small, positive large}, the fuzzy set of frequency deviation as {negative large, negative small, zero, positive small, positive large}, and the fuzzy set of unloading resistor switching state as {all unloaded, partially unloaded, no operation, partially loaded, all loaded}. A fuzzy rule base is established, such as "if the power imbalance is positive large and the frequency deviation is positive large, then all unloading resistors are loaded". To improve the adaptability of fuzzy control, an adaptive fuzzy control algorithm is used to adjust the fuzzy rules online according to the real-time operation data of the power grid. The specific resistor switching combination is obtained through fuzzy inference and defuzzification.

[0160] In S4, the minute-level response coordinates the AGC (Automatic Generation Control) regulation of the conventional unit. Based on the frequency trajectory trend predicted in S3, the required power adjustment for the conventional unit is calculated. Considering constraints such as the ramp rate and regulation range of the conventional unit, the required power adjustment is:

[0161] ΔP A =r×t A ;

[0162] r is the power regulation rate of a conventional unit, t A To adjust the time, and to avoid over- or under-adjustment, the adjustment time t is dynamically adjusted based on the deviation between the predicted frequency and the rated frequency. A ;

[0163] Unit coordination control involves sending regulation commands to conventional units through the control system to achieve power regulation; establishing a coordination control mechanism among conventional units, and rationally allocating regulation tasks based on factors such as the capacity and regulation performance of each unit to avoid regulation conflicts between different units and ensure the stable operation of the entire power grid; for example, prioritizing the regulation of units with good regulation performance and fast response speed, while also considering the economic operation of the units, so that the regulation process can minimize power generation costs while ensuring the stability of the power grid frequency.

[0164] In S4, a method for selecting a control strategy based on trajectory trend similarity is designed:

[0165] Similarity calculation: The similarity between the predicted trajectory in S3 and the trajectory trend of historical successful control cases is calculated. The reciprocal of the dynamic time curvature distance between the predicted trajectory and the trajectory of historical successful control cases is used as the similarity index. The larger the value, the higher the similarity.

[0166] In terms of strategy selection, when the similarity index is greater than the preset value, the preset successful control strategy is automatically invoked; if the similarity index is less than the preset value, the control strategy is re-formulated based on the current power grid operating status and predicted trajectory through an expert system combined with reinforcement learning algorithms to ensure operational safety and achieve stable control of the power grid transient frequency.

[0167] The above description is a preferred embodiment of the invention and is not intended to limit the scope of the invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the invention should be included within the scope of protection of the invention.

Claims

1. A method for online prediction of grid transient frequency trajectory under the coexistence of wind power low-voltage ride-through and grid disconnection, characterized in that, Includes the following steps: S1. Power grid condition monitoring and fault identification: Construct a monitoring system that integrates multi-source heterogeneous data to collect voltage, current, frequency and wind turbine status parameters in real time; Track voltage amplitude through an improved sliding window dynamic threshold algorithm and set dual criteria to identify fault types; Utilize a fault feature library and dynamic time bending matching algorithm to perform pattern recognition on voltage waveform distortion features and output fault feature vectors. S2. Power imbalance calculation: Based on the fault identification results, a real-time energy flow model is built. The phase space reconstruction technology is used to embed voltage and frequency time series, and the trajectory divergence index is calculated to characterize the instability of the system. Power deficit is estimated based on Lyapunov exponents, measurement noise is eliminated, and a power imbalance sequence is obtained. S3. Transient frequency trajectory prediction: Using the power imbalance sequence, a prediction algorithm based on phase space trajectory similarity is designed, a trajectory feature library is constructed, and the current trajectory is matched with historical trajectories by improving the K-nearest neighbor algorithm. The reference trajectory is weighted and fused to generate the future frequency prediction trajectory. A confidence evaluation mechanism is introduced to correct the prediction error and output the frequency prediction envelope to support the generation of control strategies. S4. A multi-timescale collaborative control strategy is generated. Based on the frequency trajectory prediction results, a three-level control response system is established. The millisecond level is responded by supercapacitor energy storage, the second level is switched by unloading circuit, and the minute level coordinates the AGC regulation of conventional units. A control strategy based on trajectory trend similarity is designed to ensure operational safety and achieve stable control of the power grid transient frequency.

2. The method for online prediction of grid transient frequency trajectory under the coexistence of wind power low voltage ride-through and grid disconnection as described in claim 1, characterized in that, In S1, the method for constructing a multi-source heterogeneous data fusion monitoring system is as follows: S111. Hardware deployment: At various key nodes of the power grid, such as substations and wind farm access points, install voltage sensors, current sensors, frequency sensors, and various sensors used to monitor the status parameters of wind turbine units. These sensors are responsible for collecting the corresponding physical quantity data in real time. S112. Data transmission and aggregation: Different types of sensors collect data with different formats and communication protocols. To achieve data fusion, edge computing devices are used to perform preliminary processing and conversion of the field data. The edge computing devices parse and encapsulate the collected data and transmit it to the data center through optical fiber and wireless communication. In the data center, data fusion technology is used to integrate the voltage, current, frequency and wind turbine status parameter data from different sensors to form a unified dataset.

3. The method for online prediction of grid transient frequency trajectory under the coexistence of wind power low voltage ride-through and grid disconnection as described in claim 1, characterized in that, In step S1, the voltage amplitude is tracked using an improved sliding window dynamic threshold algorithm: S121. Sliding window setting: The sliding window size is determined based on the frequency characteristics of the grid voltage fluctuation and the sampling frequency. The window slides on the time series with a fixed step size. S122. Calculate window statistics. Within each sliding window, perform statistical analysis on the voltage amplitude sequence. First, calculate the mean of the voltage amplitude within the window, and then calculate the standard deviation. S123. An adaptive adjustment factor is introduced. This factor dynamically adjusts the threshold based on historical power grid operating data and real-time fluctuations. The adjustment formula is as follows: Among them, T d Indicates a dynamic threshold; The mean of the voltage amplitude within the window is represented by k; k is an empirical coefficient; α is an adaptive adjustment factor; and σ is the standard deviation of the voltage amplitude within the window. S124. The adjustment factor is correlated with the power grid status. Real-time and historical data of the power grid are collected to determine the operating status of the power grid. If the current sensor data shows that the power grid load changes drastically, or the frequency sensor shows that the frequency fluctuation is large, it is inferred that the voltage fluctuation is aggravated. At this time, the α value is increased to make the dynamic threshold more sensitive and to capture abnormal voltage changes in time. Conversely, if the data shows that the power grid is operating relatively stably and the voltage fluctuation is small, the α value is decreased to avoid misjudgment due to the threshold being too sensitive.

4. The method for online prediction of grid transient frequency trajectory under the coexistence of wind power low voltage ride-through and grid disconnection as described in claim 1, characterized in that, In S1, the dual criteria include voltage amplitude criteria and frequency change rate criteria; The voltage amplitude criterion is: when the voltage amplitude within the sliding window is less than the dynamic threshold T. d If the proportion of the number of sampling points to the total number of points in the window exceeds the set proportion, a preliminary judgment is made that a fault has occurred, and the time at this time is recorded. The frequency change rate criterion is as follows: the frequency change rate of the power grid is calculated in real time by differential calculation of the frequency sampling sequence; a frequency change rate threshold is set, and if the voltage amplitude is continuously lower than the threshold while the frequency change rate is greater than the set frequency change rate threshold, it is judged that a fault has occurred in the power grid. The severity of the fault is preliminarily judged based on the specific numerical range of the voltage amplitude and the frequency change rate. The formula for calculating the difference is: in, The real-time frequency change rate is represented by Δt, which is the sampling time interval; f n It is the frequency sample value at the current moment, f n-1 It is the frequency sample value from the previous moment.

5. The method for online prediction of grid transient frequency trajectory under the coexistence of wind power low voltage ride-through and grid disconnection as described in claim 1, characterized in that, The method for pattern recognition using a fault feature database and a dynamic time warp matching algorithm: S131. Construct a fault feature library, collect a large amount of historical fault data, including voltage waveform data of different types of low-voltage ride-through faults and grid disconnection faults, preprocess these voltage waveform data, and extract their key features. Standardized fault feature vectors are generated and stored in a fault feature library; Each fault feature vector corresponds to a fault type label, severity score, and timestamp, which facilitates subsequent matching and querying. S132. The dynamic time bending matching algorithm dynamically bends the fault voltage waveform sequence acquired in real time with the standard voltage waveform sequence corresponding to each feature vector in the fault feature library; the dynamic time bending distance between the two is calculated based on the following formula: Where A is the real-time acquired fault voltage waveform sequence; B is the standard voltage waveform sequence corresponding to each feature vector in the fault feature library; D(A,B) is the dynamic time bending distance between A and B; m is the length of the real-time acquired voltage waveform sequence A, i.e., the number of sampling points in the sequence; n is the length of the voltage waveform sequence corresponding to the standard fault feature vector in the fault feature library according to B; ω(i,j) is the path weight, satisfying... The cumulative path weight is typically used for calculation; a i b represents the voltage amplitude at the i-th sampling time in sequence A; i It is the voltage amplitude at the j-th sampling time in the B sequence; d(a i ,b j )=(a i -b i ) 2 It is a distance metric between two points; Calculate the distance between the base A and all bases in the base B, and compare the minimum distance with a set threshold. If the minimum distance is greater than the set threshold, a matching fault type is found, and a standardized fault feature vector containing fault type label, severity score and current timestamp is output.

6. The method for online prediction of grid transient frequency trajectory under the coexistence of wind power low voltage ride-through and grid disconnection as described in claim 1, characterized in that, S2 includes: S21. Establish a real-time energy flow model. Based on the fault identification results in S1, build a real-time energy flow model: For low-voltage ride-through conditions, considering the power imbalance caused by inverter reactive power support, in the effective circuit model, reactive power is converted into effective active power through inverter control strategies and related circuit parameters. Based on this, the active power imbalance is as follows: ΔP1=P w -P l +ΔP Q ; ΔP1 represents the imbalance of active power under low voltage ride-through conditions; P w For the fan output; Pl is the load demand; ΔP Q Inverter reactive power support; For off-grid operation, an islanded system model is constructed, considering the spinning reserve response delay; then the active power imbalance is based on: ΔP2=P w -P l -P r (tt d )according to; ΔP2 represents the active power imbalance under grid disconnection conditions; P r To set the rotational reserve power; t d t represents the response delay time; t represents the current time. S22. Phase Space Reconstruction Technology: Acquiring voltage and frequency time series, selecting embedding parameters: To transform these one-dimensional time series into multi-dimensional space vectors that reflect the dynamic characteristics of the system, phase space reconstruction is required. First, the embedding dimension is selected based on the Caau method. The Caau method calculates the changes of a series of statistics with the embedding dimension; when the statistics no longer change significantly with the embedding dimension, the corresponding embedding dimension is the appropriate embedding dimension. The time delay is determined using the mutual information method. The mutual information function of the voltage or frequency time series is calculated, and the value is taken when the mutual information function first decreases to its maximum value. This is a time delay; Reconstructing phase space vectors: Based on the selected embedding dimension and time delay, the voltage sequence and frequency sequence are reconstructed in phase space to obtain the corresponding voltage change trajectory sequence and frequency change trajectory sequence. These vectors describe the state of the system at different times in phase space, and the changes in the phase space trajectory can reflect the dynamic characteristics of the system. S23. Calculate the trajectory divergence index; find the nearest neighbor point and calculate the trajectory divergence index using the local Lyapunov exponent method; in the reconstructed phase space vector, for any point... Based on this, find the x nearest neighbors of the point. Calculate the initial distance According to the equation, after δ-step evolution, the distance becomes... Local Lyapunov index based on The trajectory divergence index is the average of all local Lyapunov indices: According to this index, it is used to characterize the degree of system instability. It reflects the local divergence or convergence of the trajectory near a point in phase space. The larger the value of λ, the more unstable the system. T represents the total length of the voltage or frequency time series. S24. Using the Lyapunov exponent to estimate the power deficit and eliminate measurement noise, according to Lyapunov stability theory, the power deficit is related to the average value of the local Lyapunov exponent: [following...] ΔP=Cλ; ΔP is the power deficit; C is the proportionality coefficient; In actual measurements, power deficit data is affected by noise. To obtain a more accurate power imbalance sequence, a moving average algorithm with an adaptive forgetting factor is introduced to eliminate measurement noise. The measured power deficit sequence is {ΔP}. m1 ,ΔP m2 ,…,ΔP mT }, ΔP mi (i = 1, 2, ..., T) represents the power deficit measured at different times. The filtered power imbalance sequence is recursively calculated as follows: ΔP f (t)=αΔP f (t-1)+(1-α)ΔP m (t); ΔP f (t) represents the power imbalance sequence at time t after filtering; ΔP f (t-1) represents the power imbalance sequence at time t-1 after filtering; α is the forgetting factor. By continuously adjusting the value of α and optimizing the filtering effect based on data trends and noise characteristics, a time-scale-aligned power imbalance sequence {ΔP} is finally obtained. f1 ,ΔP f2 ,…,ΔP fT }, ΔP fi (i = 1, 2, ..., T) represents the power imbalance at different times after filtering; this provides accurate data support for subsequent transient frequency trajectory prediction.

7. The method for online prediction of grid transient frequency trajectory under the coexistence of wind power low voltage ride-through and grid disconnection as described in claim 1, characterized in that, The S3 mentioned above includes: S31. Construct a trajectory feature library, comprehensively collect various historical fault cases, covering different fault scenarios, including but not limited to low voltage ride-through faults and grid disconnection faults of different degrees, as well as situations under various power grid operating conditions; each case records the power imbalance sequence and the corresponding frequency change trajectory sequence in detail. The power imbalance sequence reflects the dynamic changes in power during the fault, while the frequency change trajectory sequence reflects the frequency fluctuations over time. Normalize these data, map the power imbalance and frequency data to the [0,1] interval, classify and store the data according to fault type and fault severity, and construct the trajectory feature library; S32. Improved k-nearest neighbor algorithm for trajectory matching, for obtaining time-stamped aligned power imbalance sequence {ΔP} f1 ,ΔP f2 ,…,ΔP fT An improved k-nearest neighbor algorithm is used for real-time pattern matching in phase space. The Hausdorff distance between the current sequence and various typical trajectories in the feature library is calculated based on the following formula: Where, F={ΔP f1 ,ΔP f2 ,…,ΔP fT };M={ΔP m1 ,ΔP m2 ,…,ΔP mT } is the power imbalance sequence corresponding to a typical trajectory in the trajectory feature library, ΔP mj (j=1,2,…,T) is the power imbalance value at the j-th time in the sequence; First, for each i, find the value within the range of j that makes... The minimum value is the point in sequence F that has the smallest distance to the i-th point in sequence M; then the maximum value is taken from these minimum values. Similarly; The top K samples with the smallest Hausdorff distance are selected as reference trajectories and their corresponding frequency trajectories. The value of K is determined by cross-validation. Prediction experiments are conducted on historical data under different K values, and the K value with the smallest prediction error is selected. S33. Weighted fusion of reference trajectories generates predicted trajectories, calculating the weight ω of each reference trajectory based on the Hausdorff distance. k The formula is: The power imbalance sequence based on the d-th reference trajectory is a historical data sequence selected from the trajectory feature library that is similar to the current system power imbalance sequence. The subsequent evolution paths of the reference trajectory are weighted and fused to generate the future frequency prediction trajectory. The calculation formula is as follows: f p (t) represents the value of the future frequency prediction trajectory at time t; frd represents the frequency trajectory corresponding to the d-th reference trajectory, which is the frequency change sequence corresponding to the power imbalance sequence of the d-th reference trajectory, used to provide frequency information during the weighted fusion process; S34. Introduce a confidence assessment mechanism to correct prediction errors, and calculate the root mean square error (RMSE) between the predicted trajectory and real-time measurements: according to; f m (t) represents the real-time measured frequency. A root mean square error threshold is set based on the data. If the RMSE is greater than the threshold, the prediction correction mechanism is triggered, and the prediction trajectory is recalculated. Finally, the frequency prediction envelope with probability density interval is output.

8. The method for online prediction of grid transient frequency trajectory under the coexistence of wind power low voltage ride-through and grid disconnection as described in claim 1, characterized in that, The three-level control response system in S4 includes millisecond-level response, second-level response, and minute-level response; among which... The millisecond-level response is handled by supercapacitor energy storage. Based on the frequency prediction envelope obtained from S3 and combined with the power balance relationship, the power deficit curve is predicted. Considering the charging and discharging efficiency and lifetime constraints of the supercapacitor, an optimization model is established with the objective function as follows: min(|ΔP p (t)-P sc (t)|); P sc (t) represents the supercapacitor's charging and discharging power at time t, with constraints including voltage and current limits as well as depth of charge and discharge limits. The optimal charging and discharging power command is obtained by solving the optimization model, and then the charging and discharging power P of the supercapacitor is calculated according to the power balance equation. sc :P sc =ΔP p (t) Combined with the voltage V of the supercapacitor energy storage system sc and current I sc Relationship, P sc =V sc I sc Based on; and the voltage conversion relationship of the Buck / Boost converter According to V o V is the output voltage of the converter. i The input voltage is given by D, and the duty cycle is given by the formula: The duty cycle of the Buck / Boost converter is dynamically adjusted using real-time feedback control technology. The duty cycle is finely adjusted based on the real-time voltage and current of the supercapacitor to ensure that the supercapacitor can respond quickly and accurately to changes in power deficit. The second-level response is achieved through the graded switching of the unloading circuit. A fuzzy control algorithm is used to optimize the resistor switching combination. The frequency deviation predicted by S3 and the power imbalance calculated by S2 are used as the inputs of the fuzzy controller, and the output is the switching state of the unloading resistor. Fuzzy sets are defined, such as the fuzzy set of power imbalance as {negative large, negative small, zero, positive small, positive large}, the fuzzy set of frequency deviation as {negative large, negative small, zero, positive small, positive large}, and the fuzzy set of unloading resistor switching state as {all off, partially off, no operation, partially on, all on}. A fuzzy rule base is formulated, such as "if the power imbalance is positive large and the frequency deviation is positive large, then all unloading resistors are on". To improve the adaptability of fuzzy control, an adaptive fuzzy control algorithm is used to adjust the fuzzy rules online according to the real-time operation data of the power grid. The specific resistor switching combination is obtained through fuzzy inference and defuzzification. The AGC regulation of the conventional unit is coordinated with a minute-level response. Based on the frequency trajectory trend predicted by S3, the power adjustment required for the conventional unit is calculated. Considering the ramp rate and regulation range constraints of the conventional unit, the required power adjustment is: ΔP A r×t A ; r is the power regulation rate of a conventional unit, t A To adjust the time, and to avoid over- or under-adjustment, the adjustment time t is dynamically adjusted based on the deviation between the predicted frequency and the rated frequency. A ; Unit coordination control involves sending regulation commands to conventional units through the control system to achieve power regulation; establishing a coordination control mechanism among conventional units, rationally allocating regulation tasks based on the capacity and regulation performance of each unit, avoiding regulation conflicts between different units, and ensuring the stable operation of the entire power grid; for example, prioritizing the regulation of units with good regulation performance and fast response speed, while also considering the economic operation of the units, so that the regulation process can minimize power generation costs while ensuring the stability of the power grid frequency.

9. The method for online prediction of grid transient frequency trajectory under the coexistence of wind power low voltage ride-through and grid disconnection as described in claim 1, characterized in that, In S4, a method for selecting a control strategy based on trajectory trend similarity is designed: Similarity calculation: The similarity between the predicted trajectory in S3 and the trajectory trend of historical successful control cases is calculated. The reciprocal of the dynamic time curvature distance between the predicted trajectory and the trajectory of historical successful control cases is used as the similarity index. The larger the value, the higher the similarity. In terms of strategy selection, when the similarity index is greater than the preset value, the preset successful control strategy is automatically invoked; if the similarity index is less than the preset value, the control strategy is re-formulated based on the current power grid operating status and predicted trajectory through an expert system combined with reinforcement learning algorithms to ensure operational safety and achieve stable control of the power grid transient frequency.

Images