A method and system for generator parameter identification processing

Abstract

The application discloses a kind of generator parameter identification processing method and system, and the application relates to signal processing technical field, comprising the following steps: collecting the phasor time series data of target generator terminal, determine active power signal and excitation voltage signal;Active power signal is adaptively variable modal decomposition, gradually increase modal decomposition number, calculate residual energy proportion and change amplitude, until meet termination condition, determine optimal modal decomposition number and output corresponding modal component;Based on the average and standard deviation of transfer entropy, effective excitation modal component is screened, and pure active power signal is reconstructed;The generator state space model considering magnetic circuit saturation is established, the augmented state vector is constructed, the recursive estimation method is used, and the measurement noise covariance matrix is dynamically adjusted according to the pure active power signal, and the generator state variable and parameter set are estimated synchronously, so as to realize high-precision parameter identification.

Description

Technical Field

[0001] This invention relates to the field of signal processing technology, specifically to a generator parameter identification and processing method and system. Background Technology

[0002] In modern power systems, generators are core equipment, and their performance and stability directly affect the reliable supply and quality of electricity. Therefore, accurate identification of generator parameters is crucial for ensuring the safe operation of the power system. Currently, generator operation status monitoring and parameter identification typically rely on real-time acquisition of time-series data such as voltage, current, excitation voltage, and rotor angular velocity. This information provides a foundation for generator performance evaluation and fault diagnosis. However, accurately and effectively extracting the information contained in this data in complex operating environments remains a challenge. Existing technologies mostly employ model-based methods, using numerical simulation and theoretical analysis to identify generator parameters. These methods generally rely on linearized models, simplifying the generator's dynamic characteristics into linear equations for parameter estimation. However, the assumption of linearization is not applicable to all operating conditions, especially when the generator exhibits nonlinearity, saturation effects, or large load fluctuations. In practical applications, these methods struggle to cope with the complexity of various operating conditions and lack robustness to measurement noise and system uncertainties, leading to reduced accuracy and reliability of the identification results. The shortcoming of existing technologies is that they often fail to effectively capture the real behavior of generators under nonlinear and dynamic conditions, especially in environments with strong disturbances and frequent changes. Therefore, there is an urgent need for a technology that can accurately identify generator parameters in order to improve the operating efficiency and reliability of generators.

[0003] The information disclosed in the background section is only intended to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0004] The purpose of this invention is to provide a generator parameter identification and processing method and system to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: A generator parameter identification and processing method, the specific steps of which include: The raw phasor timing data of the target generator terminal is collected, and the collected raw phasor timing data is preprocessed to obtain the active power signal and the excitation voltage signal. Adaptive variational mode decomposition is performed on the active power signal. The number of mode decompositions is increased stepwise, and the residual energy ratio between the active power reconstructed signal and the active power signal is calculated in each decomposition round. When the residual energy ratio and the change amplitude of the ratio meet the preset termination condition, the optimal number of mode decompositions is determined, and all mode components under the optimal number of mode decompositions are output. For any modal component under the optimal number of modal decompositions, calculate its transfer entropy with the excitation voltage signal, and analyze the mean and standard deviation of the transfer entropy of all modal components. Based on the mean and standard deviation of the transfer entropy, select the effective excitation modal components, and superimpose the effective excitation modal components to reconstruct a pure active power signal. A generator state-space model with magnetic circuit saturation effect is established, and the parameter set of the target generator is determined. The parameter set of the target generator and the generator state variable set are merged to form an augmented state vector. The augmented state vector is recursively estimated. In each recursive estimation, the innovation amplitude is calculated based on the recursive estimate value at each time and the pure active power signal. The measurement noise covariance matrix is ​​dynamically adjusted so that the generator state variable set and parameter set of the target generator can be estimated simultaneously during the recursive process.

[0006] Furthermore, the original phasor timing data specifically includes timing data of three-phase voltage, three-phase current, excitation voltage, and rotor angular velocity. The original phasor timing data of the target generator is collected from the phasor measurement unit at the generator terminal at a fixed sampling frequency within a sampling time window. The sampling time window length covers at least one low-frequency oscillation cycle. The preprocessing of the acquired raw phasor time series data specifically includes feature analysis preprocessing and noise filtering preprocessing. The feature analysis preprocessing specifically involves: analyzing the time series data of three-phase voltage, three-phase current and rotor angular velocity in the raw phasor time series data using coordinate transformation method, calculating the active power value of the target generator at each sampling point, and using cubic spline interpolation to analyze the active power value and excitation voltage value at each sampling point to obtain the active power signal and excitation voltage signal of the target generator within the sampling time window. The noise filtering preprocessing specifically uses a low-pass filter with a preset cutoff frequency to perform low-pass filtering on the active power signal and excitation voltage signal to eliminate high-frequency noise.

[0007] Furthermore, the specific steps for adaptive variational mode decomposition of the active power signal include: setting characteristic parameters for variational mode decomposition in each decomposition round, wherein the characteristic parameters specifically include a penalty factor, a fidelity coefficient, and the number of mode decompositions, and the number of mode decompositions increases progressively with the number of decomposition rounds; in each decomposition round, variational mode decomposition is performed on the active power signal to obtain several mode components with the same number of mode decompositions in that decomposition round; and all mode components in that decomposition round are reconstructed to obtain the reconstructed active power signal, wherein the mode components are specifically represented as a function with time within the sampling time window as the independent variable; For any decomposition round, the logic for reconstructing all modal components under it is as follows: accumulate the values ​​of all modal components at the same time to obtain the active power reconstruction value at that time, and summarize the active power reconstruction values ​​at each time within the sampling time window to obtain the active power reconstruction signal.

[0008] Furthermore, for any decomposition round, the specific logic for calculating the residual energy ratio between the active power reconstructed signal and the active power signal is as follows: calculate the squared difference between the active power value and the active power reconstructed value at the same moment, and integrate the squared difference at each moment within the sampling time window to obtain the residual energy between the active power reconstructed signal and the active power signal in that decomposition round. Then, integrate the squared active power value at each moment within the sampling time window to obtain the active energy quantity. Calculate the ratio of the residual energy to the active energy quantity, which is used as the residual energy ratio between the active power reconstructed signal and the active power signal in that decomposition round.

[0009] Furthermore, the logic for determining the optimal number of mode decompositions is as follows: the number of mode decompositions that first meet the preset termination condition during the adaptive variational mode decomposition process is taken as the optimal number of mode decompositions. The preset termination condition is as follows: the residual energy ratio is not greater than a preset threshold and in at least two subsequent decomposition rounds, the decrease in the residual energy ratio is less than a preset decrease threshold.

[0010] Furthermore, the logic underlying the calculation of the transfer entropy between any modal component and the excitation voltage signal under the optimal number of modal decompositions is as follows: for any modal component, analyze its probability density distribution and that of the excitation voltage within the same sampling time window, in order to calculate the transfer entropy between this modal component and the excitation voltage signal within the sampling time window, thereby obtaining the transfer entropy between each modal component and the excitation voltage signal under the sampling time window; Calculate the average and standard deviation of the transfer entropy corresponding to all modal components. Based on the mean and standard deviation of the transfer entropy, select effective excitation modal components. The specific logic is as follows: use the sum of the mean and standard deviation of the transfer entropy as the effective selection coefficient, calculate the difference between the transfer entropy of each modal component and the effective selection coefficient, and mark the modal components corresponding to the difference that is not less than 0 as effective excitation modal components. The specific method for obtaining the clean active power signal is as follows: superimpose the signal values ​​of the effective excitation mode components at the same time to obtain the clean active power signal value at that time, thereby determining the clean active power signal within the sampling time window.

[0011] Furthermore, the parameter set of the target generator includes inertial time constant, damping coefficient, saturation coefficient, excitation winding transient time constant, d-axis synchronous reactance, q-axis synchronous reactance and d-axis transient reactance, and the generator state variable set includes power angle, angular velocity and q-axis transient electromotive force; The generator state-space model includes state equations and measurement equations. The state equations include: a power angle state equation determined based on the deviation between angular velocity and synchronous speed; an angular velocity state equation determined based on mechanical power, active power, inertial time constant, and damping coefficient; and a q-axis transient electromotive force state equation determined based on excitation voltage, d-axis current, d-axis synchronous reactance, d-axis transient reactance, and excitation winding transient time constant. The measurement equation is used to characterize the active power output and includes an active power calculation term based on the power angle, q-axis transient electromotive force, bus voltage and system equivalent transient reactance, as well as a saturation correction term based on the relationship between the q-axis transient electromotive force and a preset saturation threshold.

[0012] Furthermore, the parameter set of the target generator is merged with the generator state variable set to form an augmented state vector; An extended Kalman filter framework is used to recursively estimate the augmented state vector. In each recursive step, the innovation amplitude at the same time is determined based on the measured predicted value and the pure active power signal value at the same time, and then the recursive estimation is performed. The specific steps are as follows: set the augmented state estimate at the initial time and set the initial error covariance matrix. For each time within the sampling time window, the augmented state estimate at that time is calculated based on the augmented state estimate and state equation at the previous time. At the same time, the error covariance matrix is ​​updated based on the Jacobian matrix and the process noise covariance matrix. The specific logic for calculating the innovation amplitude is as follows: the measurement equation is used to calculate the measurement prediction value at different times, and the difference between the pure active power signal value and the measurement prediction value at the same moment is calculated as the innovation amplitude at that moment; Based on the innovation amplitude and Kalman gain, the augmented state estimate and its covariance matrix are updated. The recursive estimation steps are repeated for each sampling time until the extended Kalman filter converges. The convergence criterion is specifically set as follows: the changes in the estimated values ​​of all target generator parameters are less than the preset change threshold for several consecutive time steps. Once the convergence criterion is met, the extended Kalman filter is considered to have reached stability. At this time, the parameter part in the augmented state estimate vector is the final identification result.

[0013] The present invention also provides a generator parameter identification and processing system, which is used to execute the above-described generator parameter identification and processing method, including: The data acquisition module is used to acquire the raw phasor timing data of the target generator terminal and preprocess the acquired raw phasor timing data to obtain the active power signal and excitation voltage signal. The signal decomposition optimization module is used to perform adaptive variational mode decomposition on the active power signal. It increases the number of mode decompositions in a stepwise manner and calculates the residual energy ratio between the active power reconstructed signal and the active power signal in each decomposition round. When the residual energy ratio and the change amplitude of the ratio meet the preset termination conditions, the optimal number of mode decompositions is determined and all mode components under the optimal number of mode decompositions are output. The signal filtering and reconstruction module is used to calculate the transfer entropy between any mode component and the excitation voltage signal for any mode component under the optimal number of mode decompositions, and analyze the mean and standard deviation of the transfer entropy of all mode components. Based on the mean and standard deviation of the transfer entropy, the effective excitation mode components are selected, and the effective excitation mode components are superimposed to reconstruct a clean active power signal. The parameter recursive identification module is used to establish a generator state-space model with magnetic circuit saturation effect and determine the parameter set of the target generator. The parameter set of the target generator is merged with the generator state variable set to form an augmented state vector. The augmented state vector is recursively estimated. In each recursive estimation, the innovation amplitude is calculated based on the recursive estimate value at each time and the pure active power signal. The measurement noise covariance matrix is ​​dynamically adjusted so as to simultaneously estimate the generator state variable set and parameter set of the target generator during the recursive process.

[0014] Compared with the prior art, the beneficial effects of the present invention are: This invention introduces adaptive variational mode decomposition technology, enabling the dynamic adjustment of the number of mode decompositions based on signal characteristics when processing active power signals. This ensures that the residual energy ratio is minimized. By calculating the residual energy ratio between the reconstructed signal and the original signal after each decomposition, the optimal number of mode decompositions can be effectively identified, and multiple mode components can be extracted. This mode information not only intuitively reflects the dynamic characteristics of the system but also quantifies the causal influence of excitation voltage on each mode component through transfer entropy analysis. This allows for the selection of effective excitation modes, enabling the extraction of pure active power signals from complex signals and providing reliable data support for subsequent parameter identification. Secondly, considering the establishment of a nonlinear state-space model for the magnetic circuit saturation effect, the constructed model is closer to the actual operating state of the generator. By combining it with an extended Kalman filter, the augmented state vector is recursively estimated. This not only allows for the simultaneous estimation of the generator's state variables and parameter set, but also improves the robustness and accuracy of parameter identification by dynamically adjusting the measurement noise covariance matrix. This enables the system to adapt and converge to the true state more quickly when facing different operating conditions and external disturbances, thereby improving the safety and stability of generator operation. Attached Figure Description

[0015] Figure 1 This is a schematic diagram of the overall method flow of the present invention; Figure 2 A diagram showing the proportion of residual energy across different rounds; Figure 3 This is a schematic diagram of the overall system structure of the present invention. Detailed Implementation

[0016] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.

[0017] It should be noted that, unless otherwise defined, the technical or scientific terms used in this invention should have the ordinary meaning understood by one of ordinary skill in the art to which this invention pertains. The terms "first," "second," and similar terms used in this invention do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "comprising" or "including" mean that the element or object preceding the word encompasses the elements or objects listed following the word and their equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as "upper," "lower," "left," and "right" are used only to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0018] Example: Please see Figures 1-2 The present invention provides a technical solution: A generator parameter identification and processing method, the specific steps of which include: Step 1: Collect the raw phasor timing data of the target generator terminal, and preprocess the collected raw phasor timing data to obtain the active power signal and excitation voltage signal.

[0019] The original phasor timing data specifically includes timing data of three-phase voltage, three-phase current, excitation voltage, and rotor angular velocity. The original phasor timing data of the target generator is collected from the phasor measurement unit at the generator terminal at a fixed sampling frequency within a sampling time window. The sampling time window length covers at least one low-frequency oscillation cycle. The specific method for acquiring raw phasor time series data is as follows: Select a suitable phasor measurement device, such as a phasor measurement unit (PMU), which can monitor and record voltage, current and other relevant parameters in the power system in real time; set an appropriate sampling frequency, such as 1 kHz or higher, according to the characteristics of the target generator and the dynamic characteristics of the system, to ensure that rapidly changing signals are captured. Connect the PMU to the electrical port of the target generator to ensure that the timing data of three-phase voltage, three-phase current, excitation voltage and rotor angular velocity can be acquired in real time. Under the normal operation of the target generator, start the PMU to acquire data. The device will record relevant parameters at a set fixed frequency and generate raw phasor timing data. During the acquisition process, monitor the stability of the data in real time and record any abnormalities so that possible interference factors can be considered in subsequent data processing.

[0020] The preprocessing of the acquired raw phasor time-series data specifically includes feature analysis preprocessing and noise filtering preprocessing. The feature analysis preprocessing specifically involves: analyzing the time-series data of three-phase voltage, three-phase current, and rotor angular velocity in the raw phasor time-series data using coordinate transformation to calculate the active power value of the target generator at each sampling point. Specifically, this includes converting the three-phase voltage and current signals into phasor form and calculating the complex phasor representation of each sampling point; combining the voltage and current amplitudes with the phase angles to obtain the complex phasor of each sampling point; and calculating the active power value based on the complex phasor of each sampling point using the power calculation formula. The power calculation formula is a conventional technique and will not be elaborated upon here. Cubic spline interpolation is used to analyze the active power and excitation voltage values ​​at each sampling point to obtain the active power and excitation voltage signals of the target generator within the sampling time window. Specifically, a continuous smooth curve is constructed between the calculated active power values ​​to make the power values ​​at each sampling point smoother, thereby improving the availability of the data. The interpolation method for the excitation voltage values ​​is similar.

[0021] The noise filtering preprocessing specifically involves using a low-pass filter with a preset cutoff frequency to perform low-pass filtering on the active power signal and excitation voltage signal to eliminate high-frequency noise. The appropriate cutoff frequency is selected based on the signal characteristics. The cutoff frequency should be set below the upper limit of the main frequency components in the signal to ensure effective preservation of useful signals. Common types of low-pass filters include RC low-pass filters, Butterworth filters, and Chebyshev filters.

[0022] Step 2: Perform adaptive variational mode decomposition on the active power signal. Increase the number of mode decompositions in a stepwise manner and calculate the residual energy ratio between the active power reconstructed signal and the active power signal in each decomposition round. Continue until the residual energy ratio and the change amplitude of the ratio meet the preset termination condition. Determine the optimal number of mode decompositions and output all mode components under the optimal number of mode decompositions.

[0023] The specific steps for adaptive variational mode decomposition (VMD) of active power signals include: setting characteristic parameters for VMD in each decomposition round, specifically including a penalty factor, a fidelity coefficient, and the number of mode decompositions. Specifically, the penalty factor is set based on the bandwidth of the control mode, the fidelity coefficient is initially set to 0, and the number of mode decompositions is initially set to 1 and increases incrementally with each decomposition round. In each decomposition round, VMD is performed on the active power signal to obtain several mode components, the same number as the number of mode decompositions in that round. All mode components in that round are then reconstructed to obtain the reconstructed active power signal. The mode components are specifically represented as functions with time within the sampling time window as the independent variable; the specific formula used is: In the formula, This is the set of modal components at time t after modal decomposition. Let K be the signal value of the Kth mode component at time t, where K is the total number of mode decompositions and t is the time variable within the sampling time window; For any decomposition round, the logic for reconstructing all modal components under it is as follows: Accumulate the values ​​of all modal components at the same time moment to obtain the reconstructed active power value at that time moment; summarize the reconstructed active power values ​​at each time moment within the sampling time window to obtain the reconstructed active power signal; where the reconstructed active power value at any given time moment is specifically represented as follows: In the formula, Let be the reconstructed active power value at time t. Let be the signal value of the i-th mode component at time t, where i is the index of the number of mode decompositions.

[0024] For any decomposition round, the logic for calculating the residual energy ratio between the reconstructed active power signal and the active power signal is as follows: Calculate the squared difference between the active power value and the reconstructed active power value at the same moment, and integrate the squared difference over each moment within the sampling time window to obtain the residual energy between the reconstructed active power signal and the active power signal in that decomposition round. Then, integrate the squared active power value over each moment within the sampling time window to obtain the active power quantity. Calculate the ratio of the residual energy to the active power quantity, which is used as the residual energy ratio between the reconstructed active power signal and the active power signal in that decomposition round. The specific formula for the residual energy ratio is as follows: In the formula, The percentage of residual energy. The sampling time window length, Let be the residual signal at time t. Let be the active power value at time t. This represents the residual energy. It is a functional quantity.

[0025] The logic for determining the optimal number of mode decompositions is as follows: the number of mode decompositions that first meet the preset termination condition during the adaptive variational mode decomposition process is taken as the optimal number of mode decompositions. The preset termination condition is as follows: the residual energy ratio is not greater than a preset threshold and in at least two subsequent decomposition rounds, the decrease in the residual energy ratio is less than a preset decrease threshold.

[0026] By using the number of mode decompositions that first meet a preset termination condition as the optimal number of mode decompositions, the problems of over-decomposition and under-decomposition are effectively avoided. This ensures that the number of modes accurately reflects the signal characteristics without introducing redundant modes, thus maintaining the effectiveness and simplicity of the decomposition. Setting the residual energy proportion to be no greater than a preset threshold effectively controls the complexity of mode decomposition. By limiting the magnitude of the residual energy, unnecessary mode decomposition is avoided, thereby reducing the amount of computation and improving the processing speed. When the decrease in the proportion of residual energy is less than the preset decrease threshold, it indicates that increasing the number of mode decompositions has a limited effect on improving signal reconstruction. Increasing the number of modes will not significantly improve the signal representation, thus avoiding the degradation of reconstructed signal quality due to excessive decomposition.

[0027] The preset threshold for the residual energy ratio is usually set between 0.01 and 0.05. The specific value can be adjusted according to expert experience. A value below 0.01 usually means that the signal is over-decomposed and important information may be lost; a value above 0.05 may result in insufficient decomposition of important features of the signal; output all modal components under the optimal number of modal decompositions. The preset drop threshold can be set according to the complexity of the signal. For power fluctuation signals, the drop amplitude may be small, so the drop threshold is set between 0.1% and 0.5% to ensure the fineness of the decomposition.

[0028] Step 3: For any mode component under the optimal number of mode decompositions, calculate its transfer entropy with the excitation voltage signal, and analyze the mean and standard deviation of the transfer entropy of all mode components. Based on the mean and standard deviation of the transfer entropy, select the effective excitation mode components, and superimpose the effective excitation mode components to reconstruct a pure active power signal.

[0029] The logic underlying the calculation of the transfer entropy between any modal component and the excitation voltage signal under the optimal number of mode decompositions is as follows: For any modal component, analyze its probability density distribution and that of the excitation voltage within the same sampling time window to calculate the transfer entropy between this modal component and the excitation voltage signal within that sampling time window. This yields the transfer entropy between each modal component and the excitation voltage signal within that sampling time window. The specific formula for calculating the transfer entropy is: In the formula, Let the transfer entropy be the distance between the j-th modal component and the excitation voltage signal. For the j-th modal component in The signal value at time [time] Indicates that the j-th modal component is in The signal value at time , This represents the value of the preprocessed excitation voltage signal at time t. denoted as probability density function; j is the index of the modal component under the optimal number of mode decompositions.

[0030] It should be noted that, from the decomposed modes, the dynamic response modes that are truly excited by the excitation system disturbance are identified by the transfer entropy, while the continuous environmental noise or forced oscillation modes that are unrelated to the excitation are eliminated.

[0031] Calculate the mean and standard deviation of the transfer entropy for all modal components. Based on the mean and standard deviation of the transfer entropy, select effective excitation modal components. Specifically, the sum of the mean and standard deviation of the transfer entropy is used as the effective selection coefficient. Calculate the difference between the transfer entropy of each modal component and the effective selection coefficient. Mark the modal components whose difference is not less than 0 as effective excitation modal components. Specifically, if... If the modal component is valid, then it is recorded as an effective excitation modal component; otherwise, the modal component is discarded. It should be noted that by using effective screening coefficients, the screening boundary can be dynamically determined based on the statistical characteristics of the data itself, avoiding the subjectivity and inapplicability of manually setting fixed thresholds, and effectively distinguishing the forced response component caused by active excitation control from other free oscillations or background noise components in the system.

[0032] The specific method for obtaining the clean active power signal is as follows: superimpose the signal values ​​of the effective excitation mode components at the same time to obtain the clean active power signal value at that time, thereby determining the clean active power signal within the sampling time window.

[0033] Step 4: Establish a generator state-space model for magnetic circuit saturation effect and determine the parameter set of the target generator. Merge the parameter set of the target generator with the generator state variable set to form an augmented state vector. Perform recursive estimation on the augmented state vector. In each recursive estimation, calculate the innovation amplitude based on the recursive estimate value at each time and the pure active power signal, and dynamically adjust the measurement noise covariance matrix to simultaneously estimate the generator state variable set and parameter set of the target generator during the recursive process.

[0034] The parameter set of the target generator includes inertia time constant, damping coefficient, saturation coefficient, excitation winding transient time constant, d-axis synchronous reactance, q-axis synchronous reactance and d-axis transient reactance; the generator state variable set includes power angle, angular velocity and q-axis transient electromotive force. The generator state-space model includes state equations and measurement equations. The state equations include: a power angle state equation determined by the deviation between angular velocity and synchronous speed; an angular velocity state equation determined by mechanical power, active power, inertial time constant, and damping coefficient; and a q-axis transient electromotive force state equation determined by excitation voltage, d-axis current, d-axis synchronous reactance, d-axis transient reactance, and excitation winding transient time constant. The generator state-space model consists of state equations and measurement equations. The state equations consist of a set of generator state variables and a set of parameters of the target generator. The measurement equations are specifically expressed as equations that take the parameters of the target generator as input and active power as output. The state equation is specifically represented as follows: the state equation consists of three differential equations, which describe the relationship between the power angle, angular velocity and q-axis transient electromotive force with time, and are respectively denoted as the first, second and third differential equations. The first differential equation is specifically represented as the difference between the angular velocity and the rotational speed of the target generator at the same time. The second differential equation is expressed as follows: calculate the quotient of the rotational speed divided by twice the inertial time constant, calculate the difference between the generator mechanical power and the active power, calculate the product of the first differential equation and the damping coefficient, subtract the product from the difference between the generator mechanical power and the active power to obtain the second difference, and use the product of the quotient and the second difference as the expression of the second differential equation. The third differential equation is expressed as follows: Calculate the reciprocal of the transient time constant of the excitation winding, calculate the difference between the d-axis synchronous reactance and the d-axis transient reactance, multiply this difference by the d-axis current to obtain the first correlation product, calculate the difference between the q-axis transient electromotive force and the first correlation product, subtract this difference from the excitation voltage to calculate the third difference, multiply the third difference by the reciprocal of the transient time constant of the excitation winding, and use this product as the expression of the third differential equation. The specific formula used is as follows: In the formula, The rate of change of the work angle, The target generator angular velocity, For rotational speed, This refers to angular acceleration, i.e., the relationship between angular velocity and time. The inertial time constant, The damping coefficient is... For generator mechanical power, Active power Let be the rate of change of transient electromotive force along the q-axis. The transient time constant of the excitation winding is... This is the excitation voltage. Let be the q-axis transient electromotive force. For d-axis synchronous reactance, For d-axis transient reactance, This refers to the d-axis current. This is the first differential equation. This is the second differential equation. This is the third differential equation; The measurement equation is used to characterize the active power output and includes an active power calculation term based on the power angle, q-axis transient electromotive force, bus voltage and system equivalent transient reactance, as well as a saturation correction term based on the relationship between the q-axis transient electromotive force and a preset saturation threshold. The specific method for expressing the measurement equation is as follows: calculate the product of the q-axis transient electromotive force and the bus voltage, divide the product by the d-axis system equivalent transient reactance, and record the product of the resulting quotient and the sine of the power angle as the first measurement term; Calculate the difference between the reciprocal of the equivalent transient reactance of the q-axis system and the reciprocal of the equivalent transient reactance of the d-axis system. Calculate the quotient of the square of the bus voltage divided by a constant 2. Calculate the product of the quotient and the difference. Multiply the product of the obtained product and twice the sine of the power angle as the second measurement item. The third measurement term is defined based on the calculation of the saturation correction function. When the q-axis transient electromotive force is greater than the saturation threshold, the saturation correction function takes the value of the product of the saturation coefficient and the first correction exponent; otherwise, it takes the value of zero. The first correction exponent is specifically expressed as the square of the difference between the q-axis transient electromotive force and the saturation threshold. The first, second, and third measurement terms are added together to obtain the measurement equation, which is expressed as follows: In the formula, Bus voltage The equivalent transient reactance of the d-axis system, For the angle of attack, The equivalent transient reactance of the q-axis system, For saturation correction function, That is, the first measurement item. The second measurement item, This is the third measurement item.

[0035] The specific formula upon which the saturation correction function is based is: In the formula, The saturation coefficient is... This is the saturation threshold; It should be noted that when the generator enters the saturation region, the actual output active power will be slightly larger than that calculated by the linear model due to the change in equivalent reactance caused by saturation. Therefore, a saturation correction function is introduced to compensate for the difference. The saturation effect is not linear; once the threshold is exceeded, the rate of increase in magnetoresistance accelerates. Therefore, [the following is a possible interpretation:] ... The squared term can simulate this acceleration enhancement characteristic, when Slightly larger At that time, the correction amount is very small, when When the value is large, the correction amount increases rapidly.

[0036] The parameter set of the target generator is merged with the generator state variable set to form an augmented state vector. The augmented state vector specifically includes the inertial time constant, damping coefficient, saturation coefficient, excitation winding transient time constant, d-axis and q-axis synchronous reactance, d-axis transient reactance, generator power angle, angular velocity and transient electromotive force of the target generator, with a dimension of 10. An extended Kalman filter framework is used to recursively estimate the augmented state vector. In each recursive step, the innovation amplitude at the same time is determined based on the measured predicted value and the pure active power signal value at the same time, and then the recursive estimation is performed. The specific steps are as follows: set the augmented state estimate at the initial time and set the initial error covariance matrix. For each time within the sampling time window, the augmented state estimate at that time is calculated based on the augmented state estimate and state equation at the previous time. At the same time, the error covariance matrix is ​​updated based on the Jacobian matrix and the process noise covariance matrix. The specific formula used to update the error covariance matrix is ​​as follows: In the formula, The updated error covariance matrix, In order to be in At time t, the Jacobian matrix associated with the state equation and the augmented state vector, This is the process noise covariance matrix, used to reflect model uncertainty, and is usually set as a constant diagonal matrix.

[0037] It should be noted that the Jacobian matrix related to the state equation and the augmented state vector is obtained by taking the partial derivative of the expression of the state equation with respect to each state variable, thereby obtaining the elements of the Jacobian matrix. Modern computing platforms support automatic differentiation tools, such as TensorFlow and PyTorch, which can automatically calculate the Jacobian matrix of complex functions. Process noise covariance matrix This is used to describe the statistical characteristics of model errors and unmodeled dynamics during state transitions, reflecting the system's dynamic uncertainty and disturbance level. Its dimension is the square of the state vector dimension, typically a symmetric positive definite matrix. Specifically, it can be set as a diagonal matrix based on prior knowledge to estimate process noise intensity and correlation, with the main diagonal elements representing the process noise variance of each state variable. Alternatively, it can be obtained by fitting and analyzing system dynamics using experimental data to obtain the statistical characteristics of process noise. Matrix parameters are determined using statistical methods such as maximum likelihood estimation (MLE) and Bayesian estimation.

[0038] The logic underlying the calculation of the new information amplitude is as follows: the measurement equation is used to calculate the measurement prediction value at different times, specifically, the measurement equation is used to determine the active power value of the augmented state vector under different augmented state estimates. The difference between the current clean active power signal value and the measured predicted value is calculated and used as the innovation amplitude at that moment; the specific formula used is as follows: In the formula, Let be the innovation amplitude at time t. Let be the pure active power signal value at time t. The predicted value is the value measured at time t.

[0039] The augmented state estimate and its covariance matrix are updated based on the innovation magnitude and Kalman gain. The specific formula used is as follows: In the formula, The augmented state estimate is updated at time t. Let be the augmented state estimate at time t. Let be the Kalman gain matrix at time t, specifically expressed as: In the formula, Let be the covariance matrix updated at time t. It is the identity matrix. Let be the Jacobian matrix relating the measurement equation and the augmented state vector at time t.

[0040] The specific method for obtaining the Jacobian matrix related to the measurement equation and the augmented state vector is as follows: Based on the expression of the measurement equation, the Jacobian matrix related to the measurement equation and the augmented state vector is obtained by taking its derivative with respect to the augmented state vector. Partial derivatives need to be calculated for each component of the state variable. Alternatively, automatic differentiation tools such as TensorFlow and PyTorch can be provided to automatically calculate the Jacobian matrix based on the implementation of the measurement function.

[0041] The recursive estimation steps are repeated for each sampling time until the extended Kalman filter converges. The convergence criterion is specifically set as follows: the estimated values ​​of all target generator parameters change less than a preset change threshold at several consecutive times. Once the convergence criterion is met, the extended Kalman filter is considered to have reached stability, and the parameter part in the augmented state estimation vector is the final identification result.

[0042] The preset change threshold is set according to the physical meaning and value range of the parameter to be identified. For example, the impedance value in the generator system or the rotor angular velocity of the motor usually has a small change range, so a low threshold can be set, generally between 0.001 and 0.01.

[0043] Please see Figure 3 The present invention also provides a generator parameter identification and processing system, which is used to execute the above-described generator parameter identification and processing method, including: The data acquisition module is used to acquire the raw phasor timing data of the target generator terminal and preprocess the acquired raw phasor timing data to obtain the active power signal and excitation voltage signal. The signal decomposition optimization module is used to perform adaptive variational mode decomposition on the active power signal. It increases the number of mode decompositions in a stepwise manner and calculates the residual energy ratio between the active power reconstructed signal and the active power signal in each decomposition round. When the residual energy ratio and the change amplitude of the ratio meet the preset termination conditions, the optimal number of mode decompositions is determined and all mode components under the optimal number of mode decompositions are output. The signal filtering and reconstruction module is used to calculate the transfer entropy between any mode component and the excitation voltage signal for any mode component under the optimal number of mode decompositions, and analyze the mean and standard deviation of the transfer entropy of all mode components. Based on the mean and standard deviation of the transfer entropy, the effective excitation mode components are selected, and the effective excitation mode components are superimposed to reconstruct a clean active power signal. The parameter recursive identification module is used to establish a generator state-space model with magnetic circuit saturation effect and determine the parameter set of the target generator. The parameter set of the target generator is merged with the generator state variable set to form an augmented state vector. The augmented state vector is recursively estimated. In each recursive estimation, the innovation amplitude is calculated based on the recursive estimate value at each time and the pure active power signal. The measurement noise covariance matrix is ​​dynamically adjusted so as to simultaneously estimate the generator state variable set and parameter set of the target generator during the recursive process.

[0044] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0045] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented in software, the above embodiments can be implemented, in whole or in part, as a computer program product. Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented by 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.

[0046] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment, depending on actual needs.

[0047] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. A generator parameter identification and processing method, characterized in that, The specific steps include: The raw phasor timing data of the target generator terminal is collected, and the collected raw phasor timing data is preprocessed to obtain the active power signal and the excitation voltage signal. Adaptive variational mode decomposition is performed on the active power signal. The number of mode decompositions is increased stepwise, and the residual energy ratio between the active power reconstructed signal and the active power signal is calculated in each decomposition round. When the residual energy ratio and the change amplitude of the ratio meet the preset termination condition, the optimal number of mode decompositions is determined, and all mode components under the optimal number of mode decompositions are output. For any modal component under the optimal number of modal decompositions, calculate its transfer entropy with the excitation voltage signal, and analyze the mean and standard deviation of the transfer entropy of all modal components. Based on the mean and standard deviation of the transfer entropy, select the effective excitation modal components, and superimpose the effective excitation modal components to reconstruct a pure active power signal. A generator state-space model with magnetic circuit saturation effect is established, and the parameter set of the target generator is determined. The parameter set of the target generator and the generator state variable set are merged to form an augmented state vector. The augmented state vector is recursively estimated. In each recursive estimation, the innovation amplitude is calculated based on the recursive estimate value at each time and the pure active power signal. The measurement noise covariance matrix is ​​dynamically adjusted so that the generator state variable set and parameter set of the target generator can be estimated simultaneously during the recursive process. The parameter set of the target generator is merged with the generator state variable set to form an augmented state vector; An extended Kalman filter framework is used to recursively estimate the augmented state vector. In each recursive step, the innovation amplitude at the same time is determined based on the measured predicted value and the pure active power signal value at the same time, and then the recursive estimation is performed. The specific steps are as follows: set the augmented state estimate at the initial time and set the initial error covariance matrix. For each time within the sampling time window, the augmented state estimate at that time is calculated based on the augmented state estimate and state equation at the previous time. At the same time, the error covariance matrix is ​​updated based on the Jacobian matrix and the process noise covariance matrix. The specific logic for calculating the innovation amplitude is as follows: the measurement equation is used to calculate the measurement prediction value at different times, and the difference between the pure active power signal value and the measurement prediction value at the same moment is calculated as the innovation amplitude at that moment. Based on the innovation amplitude and Kalman gain, the augmented state estimate and its covariance matrix are updated. The recursive estimation steps are repeated for each sampling time until the extended Kalman filter converges. The convergence criterion is specifically set as follows: the changes in the estimated values ​​of all target generator parameters are less than the preset change threshold for several consecutive time steps. Once the convergence criterion is met, the extended Kalman filter is considered to have reached stability. At this time, the parameter part in the augmented state estimate vector is the final identification result.

2. The generator parameter identification and processing method according to claim 1, characterized in that: The original phasor timing data specifically includes timing data of three-phase voltage, three-phase current, excitation voltage, and rotor angular velocity. The original phasor timing data of the target generator is collected from the phasor measurement unit at the generator terminal at a fixed sampling frequency within a sampling time window. The sampling time window length covers at least one low-frequency oscillation cycle. The preprocessing of the acquired raw phasor time series data specifically includes feature analysis preprocessing and noise filtering preprocessing. The feature analysis preprocessing specifically involves: analyzing the time series data of three-phase voltage, three-phase current and rotor angular velocity in the raw phasor time series data using coordinate transformation method, calculating the active power value of the target generator at each sampling point, and using cubic spline interpolation to analyze the active power value and excitation voltage value at each sampling point to obtain the active power signal and excitation voltage signal of the target generator within the sampling time window. The noise filtering preprocessing specifically uses a low-pass filter with a preset cutoff frequency to perform low-pass filtering on the active power signal and excitation voltage signal to eliminate high-frequency noise.

3. The generator parameter identification and processing method according to claim 2, characterized in that: The specific steps for adaptive variational mode decomposition of active power signals include: setting characteristic parameters for variational mode decomposition in each decomposition round, wherein the characteristic parameters specifically include a penalty factor, a fidelity coefficient, and the number of mode decompositions, and the number of mode decompositions increases incrementally with the number of decomposition rounds; in each decomposition round, variational mode decomposition is performed on the active power signal to obtain several mode components with the same number of mode decompositions in that decomposition round; and all mode components in that decomposition round are reconstructed to obtain the reconstructed active power signal, wherein the mode components are specifically represented as a function with time within the sampling time window as the independent variable; For any decomposition round, the logic for reconstructing all modal components under it is as follows: accumulate the values ​​of all modal components at the same time to obtain the active power reconstruction value at that time, and summarize the active power reconstruction values ​​at each time within the sampling time window to obtain the active power reconstruction signal.

4. The generator parameter identification and processing method according to claim 3, characterized in that: For any decomposition round, the logic for calculating the residual energy ratio between the active power reconstructed signal and the active power signal is as follows: calculate the squared difference between the active power value and the active power reconstructed value at the same moment, and integrate the squared difference at each moment within the sampling time window to obtain the residual energy between the active power reconstructed signal and the active power signal in that decomposition round. Then, integrate the squared active power value at each moment within the sampling time window to obtain the active energy quantity. Calculate the ratio of the residual energy to the active energy quantity, which is used as the residual energy ratio between the active power reconstructed signal and the active power signal in that decomposition round.

5. The generator parameter identification and processing method according to claim 4, characterized in that: The logic for determining the optimal number of mode decompositions is as follows: the number of mode decompositions that first meet the preset termination condition during the adaptive variational mode decomposition process is taken as the optimal number of mode decompositions. The preset termination condition is as follows: the residual energy ratio is not greater than a preset threshold and in at least two subsequent decomposition rounds, the decrease in the residual energy ratio is less than a preset decrease threshold.

6. The generator parameter identification and processing method according to claim 3, characterized in that, The logic behind calculating the transfer entropy between any modal component and the excitation voltage signal under the optimal number of modal decompositions is as follows: For any modal component, analyze its probability density distribution and that of the excitation voltage within the same sampling time window to calculate the transfer entropy between this modal component and the excitation voltage signal within that sampling time window, thereby obtaining the transfer entropy between each modal component and the excitation voltage signal under that sampling time window. Calculate the average and standard deviation of the transfer entropy corresponding to all modal components. Based on the mean and standard deviation of the transfer entropy, select effective excitation modal components. The specific logic is as follows: use the sum of the mean and standard deviation of the transfer entropy as the effective selection coefficient, calculate the difference between the transfer entropy of each modal component and the effective selection coefficient, and mark the modal components corresponding to the difference that is not less than 0 as effective excitation modal components. The specific method for obtaining the clean active power signal is as follows: superimpose the signal values ​​of the effective excitation mode components at the same time to obtain the clean active power signal value at that time, thereby determining the clean active power signal within the sampling time window.

7. The generator parameter identification and processing method according to claim 6, characterized in that: The parameter set of the target generator includes inertia time constant, damping coefficient, saturation coefficient, excitation winding transient time constant, d-axis synchronous reactance, q-axis synchronous reactance and d-axis transient reactance; the generator state variable set includes power angle, angular velocity and q-axis transient electromotive force. The generator state-space model includes state equations and measurement equations. The state equations include: a power angle state equation determined based on the deviation between angular velocity and synchronous speed; an angular velocity state equation determined based on mechanical power, active power, inertial time constant, and damping coefficient; and a q-axis transient electromotive force state equation determined based on excitation voltage, d-axis current, d-axis synchronous reactance, d-axis transient reactance, and excitation winding transient time constant. The measurement equation is used to characterize the active power output and includes an active power calculation term based on the power angle, q-axis transient electromotive force, bus voltage and system equivalent transient reactance, as well as a saturation correction term based on the relationship between the q-axis transient electromotive force and a preset saturation threshold.

8. A generator parameter identification and processing system, characterized in that: The generator parameter identification and processing system is used to execute the generator parameter identification and processing method according to any one of claims 1-7, including: The data acquisition module is used to acquire the raw phasor timing data of the target generator terminal and preprocess the acquired raw phasor timing data to obtain the active power signal and excitation voltage signal. The signal decomposition optimization module is used to perform adaptive variational mode decomposition on the active power signal. It increases the number of mode decompositions in a stepwise manner and calculates the residual energy ratio between the active power reconstructed signal and the active power signal in each decomposition round. When the residual energy ratio and the change amplitude of the ratio meet the preset termination conditions, the optimal number of mode decompositions is determined and all mode components under the optimal number of mode decompositions are output. The signal filtering and reconstruction module is used to calculate the transfer entropy between any mode component and the excitation voltage signal for any mode component under the optimal number of mode decompositions, and analyze the mean and standard deviation of the transfer entropy of all mode components. Based on the mean and standard deviation of the transfer entropy, the effective excitation mode components are selected, and the effective excitation mode components are superimposed to reconstruct a clean active power signal. The parameter recursive identification module is used to establish a generator state-space model with magnetic circuit saturation effect and determine the parameter set of the target generator. The parameter set of the target generator is merged with the generator state variable set to form an augmented state vector. The augmented state vector is recursively estimated. In each recursive estimation, the innovation amplitude is calculated based on the recursive estimate value at each time and the pure active power signal. The measurement noise covariance matrix is ​​dynamically adjusted so as to simultaneously estimate the generator state variable set and parameter set of the target generator during the recursive process.

Images