Generator state estimation method and system considering noise and parameter uncertainty constraint

Abstract

The invention discloses a generator state estimation method and system considering noise and parameter uncertainty constraints. The method comprises the following steps: acquiring model parameters and dynamic state vectors of a generator, and establishing augmented state vectors; performing unscented transformation on the augmented state vector to obtain a particle set; improving to obtain robust mixed Kalman particle filtering, and in the process of performing unscented Kalman filtering on an augmented state vector, taking correlation entropy maximization of a measurement information sequence as a target function, and solving by adopting a fixed point iteration method to obtain a filtering gain; determining a filtering gain according to the updated state of the measurement information; robust mixed Kalman particle filtering is executed, physical constraints of model parameters serve as a feasible region, after resampling, projections of the model parameters in new-generation particles exceed the feasible region, the model parameters are set to be closest boundary points, and then resampling is conducted again; a weighted average value of the particle set is an optimal joint estimation value, a dynamic state estimation value and a model parameter identification result are separated, and reliable uncertainty quantization is provided for state and parameter estimation.

Description

Technical Field

[0001] This invention belongs to the field of power system monitoring and relay protection technology, and specifically relates to a generator state estimation method and system that takes into account noise and parameter uncertainty constraints. Background Technology

[0002] Synchronous generators are core equipment in power systems, and accurate estimation of their dynamic states, such as power angle, rotor speed, and transient electromotive force, is crucial for stability assessment, adaptive protection, and control. However, in practical applications, generator dynamic state estimation faces three major challenges: First, noise non-Gaussianity. Process noise from the prime mover and excitation system disturbances, as well as measurement noise from sensors such as PMUs, often do not conform to an ideal Gaussian distribution, exhibiting characteristics such as heavy tails and skewness. Traditional Kalman filter algorithms, based on the Gaussian assumption, experience a sharp performance decline in such scenarios. Second, model parameter uncertainty. Parameters in the generator model, such as transient reactance... , The parameters, such as operating temperature, saturation level, and aging condition, will vary and are not constant. Estimation algorithms based on fixed and accurate models will produce significant errors due to model mismatch. Poor data in measurements, sensor failures, or communication errors will introduce gross errors (poor data) and disrupt the estimation process.

[0003] Most existing state estimation methods address these problems in isolation. Existing techniques use the minimum mean square error (MMSE) criterion when filtering particle sets, relying on manual adjustment of the noise covariance matrix. In impulse noise environments, an additional threshold judgment mechanism is needed. Furthermore, they employ a noise covariance update strategy with a fixed forgetting factor, which needs to be determined offline. This approach is poorly adapted to time-varying noise, unable to distinguish between normal noise and outliers, and impulse noise can cause abnormal increases in the Q and R matrices. Robust filtering can handle poor data but has limited effectiveness against non-Gaussianity and parameter uncertainty. Particle filtering can handle non-Gaussian problems but is computationally intensive and susceptible to model errors. Parameter identification algorithms typically assume known states. Currently, there is a lack of a comprehensive solution that can uniformly address the non-Gaussianity of noise, model parameter uncertainty, and constraints imposed by poor data. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a generator state estimation method and system that takes into account noise and parameter uncertainty constraints. Under multiple constraints, including non-Gaussian noise, poor measurement data, and generator model parameter uncertainties, it achieves high-precision, robust, and accurate estimation of the generator dynamic state, while simultaneously identifying key model parameters online and providing reliable uncertainty quantification for state and parameter estimation.

[0005] The present invention adopts the following technical solution.

[0006] This invention proposes a generator state estimation method that takes into account noise and parameter uncertainty constraints, comprising: Obtain the generator's model parameters and dynamic state vector, and establish an augmented state vector; apply an unscented transformation to the augmented state vector to obtain the particle set; The robust hybrid Kalman particle filter is obtained by improving the unscented Kalman filter. The process includes: during the unscented Kalman filtering of the augmented state vector, the objective function is to maximize the correlation entropy of the measurement innovation sequence, and the filter gain is obtained by solving the objective function using the fixed-point iteration method; when performing state update using the filter gain, the corresponding filter gain is determined according to the state of the updated measurement innovation. Based on the particle set, a robust hybrid Kalman particle filter is performed. The feasible region is established by the physical constraints of the model parameters. After the particle set is resampled, if the projection of the model parameters in the new generation of particles exceeds the feasible region, the model parameters are set to the nearest boundary point and resampled again. The weighted average of the particle set is used as the optimal joint estimate, and the dynamic state estimate and model parameter identification results are separated from the optimal joint estimate.

[0007] Model parameters of the generator With dynamic state Merge to form an augmented state vector Among them, model parameters Including: transient reactance Dynamic state Includes: functional angles Rotor speed Transient potential .

[0008] The Gaussian kernel function is introduced as the cost function for measuring innovation, as shown in the following equation:

[0009] In the formula, To measure new information The cost function; For kernel width; Establish an adaptively adjustable kernel width, as shown in the following formula:

[0010] In the formula, For the first The kernel width of the next iteration. For smoothing coefficients, The trace function of a matrix. To estimate the error covariance matrix, This represents the maximum kernel width.

[0011] The objective function is to maximize the correlation entropy of the measured information sequence, as shown in the following equation:

[0012] In the formula, State variables The corresponding objective function, For the first Individual Measurement News The cost function, To measure the amount of new information; Using unscented Kalman filtering to... The measurement equations are statistically linearized to approximate the following: Then the derivative of the objective function is shown in the following equation:

[0013] In the formula, For the first The estimated value of each measurement equation, For state variable residuals - The Each measurement function value.

[0014] The filter gain is obtained by solving the objective function using the fixed-point iteration method, including: Set initial values ​​for state estimation Estimating the initial matrix of error covariance initial value of kernel width Initial value of filter gain , The gain of the unscented Kalman filter; Define filter gain Mapping function As shown in the following formula:

[0015] In the formula, For filter gain, Let be the autocovariance matrix of the state variables. For the measurement matrix, For the observation noise matrix, This is a weight matrix based on filter gain; The fixed point is solved iteratively using the following formula:

[0016] In the formula, For the first The filter gain of the next iteration; When the sufficient condition for iterative convergence is met The MCC filter gain is obtained as shown in the following formula:

[0017] In the formula, For filter gain, Let be the autocovariance matrix of the state variables. For the measurement matrix, For the observation noise matrix, This is the weight matrix. , No. Individual weighted coefficients .

[0018] The state is updated using the filter gain to update the measurement information. If the updated measurement information is in a normal state, the kernel weighting coefficient is the square of the kernel width to determine the filter gain. If the updated measurement information is in an abnormal state, the kernel weighting coefficient is 0 to determine the filter gain.

[0019] The adaptively adjusted kernel width is updated using the updated measurement information, as shown in the following equation:

[0020]

[0021] In the formula, For the first The kernel width of the next iteration. For the first Measurement updates in the next iteration For smoothing coefficients, , These represent the lower and upper boundaries of the kernel width, respectively.

[0022] This invention also proposes a generator state estimation system that takes into account noise and parameter uncertainty constraints, comprising: The data processing module is used to acquire the generator's model parameters and dynamic state vector, establish the augmented state vector, and apply an unscented transformation to the augmented state vector to obtain the particle set. The filtering improvement module is used to improve the unscented Kalman filter to obtain a robust hybrid Kalman particle filter. It includes: during the unscented Kalman filtering of the augmented state vector, taking the maximization of the correlation entropy of the measurement innovation sequence as the objective function, and using the fixed-point iteration method to solve the objective function to obtain the filtering gain; when performing state updates using the filtering gain, determining the corresponding filtering gain based on the updated state of the measurement innovation. The filtering execution module is used to perform robust hybrid Kalman particle filtering based on the particle set. The feasible region is established by the physical constraints of the model parameters. After resampling the particle set, if the projection of the model parameters in the new generation of particles exceeds the feasible region, the model parameters are set to the nearest boundary point and resampling is performed again. The weighted average value of the particle set is used as the optimal joint estimate, and the dynamic state estimate and model parameter identification result are separated from the optimal joint estimate.

[0023] The present invention is also a terminal, including a processor and a storage medium; the storage medium is used to store instructions; the processor is used to perform operations according to the instructions to execute the steps of the method.

[0024] The present invention is also a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method.

[0025] The beneficial effects of this invention are as follows, compared with the prior art, at least including: This invention solves generator state estimation and parameter identification in a unified manner within the augmented space, avoiding the error propagation problem of the traditional two-step method and realizing integrated joint estimation; the robust UKF kernel based on the MCC criterion can effectively suppress the negative impact of non-Gaussian noise and poor measurement data simultaneously, ensuring the stability of the estimation process and achieving strong robustness of the estimation; through the constraint processing mechanism, it ensures that the model parameters identified online are always within a physically reasonable range, improving the reliability, practicality, and physical consistency of the results; the hybrid particle filter framework provides the ability to approximate the optimal Bayesian estimation, can handle complex non-Gaussian distributions, and gives the uncertainty range of the estimated value in the form of particle distribution, providing information support for advanced applications (such as risk warning), with the advantages of high accuracy and quantifiable uncertainty; the system can adaptively manage parameter uncertainty and adjust the learning rate after parameter convergence, combining the advantages of fast tracking and stable convergence. Attached Figure Description

[0026] Figure 1 This is a flowchart of the generator state estimation method that takes into account noise and parameter uncertainty constraints proposed in this invention. Detailed Implementation

[0027] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of this invention. The embodiments described in this application are merely some embodiments of this invention, and not all embodiments. Based on the spirit of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the protection scope of this invention.

[0028] This invention proposes a generator dynamic state estimation method that takes into account noise and model parameter uncertainty constraints, such as... Figure 1 As shown, the method includes: Step S1: Obtain the model parameters and dynamic state vector of the generator, and establish the augmented state vector; apply an unscented transformation to the augmented state vector to obtain the particle set.

[0029] Specifically, augmented state-space modeling is performed to determine the unknown key model parameters of the generator. With dynamic state Merge to form an augmented state vector Among them, model parameters Including but not limited to transient reactance Dynamic state Including but not limited to: angles Rotor speed Transient potential .

[0030] This invention constructs a joint dynamic state and model parameter estimation framework by extending the unknown key model parameters and dynamic state vector of the generator to the state vector. This integrates the uncertainty constraints of the model parameters into the dynamic state estimation process and enables the collaborative estimation of dynamic state and model parameters.

[0031] Specifically, when performing an unscented transformation on the augmented state vector, the first... Sigma points are , This is the state estimate. As a dimension, For proportional parameters, To estimate the error covariance matrix, the measurement innovation is: , For measurement value, For the first The weight of each Sigma point , For the first Measurement equations at Sigma points The value of .

[0032] Step S2 involves improving the unscented Kalman filter to obtain a robust hybrid Kalman particle filter, including: during the unscented Kalman filtering of the augmented state vector, taking the maximization of the correlation entropy of the measurement innovation sequence as the objective function, and using the fixed-point iteration method to solve the objective function to obtain the filter gain; when performing state updates using the filter gain, determining the corresponding filter gain based on the updated state of the measurement innovation.

[0033] This invention constructs a robust hybrid Kalman-Particle Filter (RHKPF) algorithm through step S2.

[0034] Specifically, step S2 includes: Step S2.1: Introduce a Gaussian kernel function as the cost function for measuring innovation, and establish an adaptively adjusted kernel width; Specifically, the cost function for measuring new information is shown in the following equation:

[0035] In the formula, To measure new information The cost function is the measurement innovation, which is the measurement value minus the predicted measurement value. For kernel width; This invention uses the maximum correlation entropy criterion (MCC) as the objective function for filtering optimization. By introducing a Gaussian kernel function, a non-quadratic cost function is given to the measurement information. MCC maximizes the cross-entropy of the error probability density function and takes advantage of the inherent ability of the kernel function to suppress non-Gaussian noise.

[0036] The adaptively adjusted kernel width is shown in the following formula:

[0037] In the formula, For the first The kernel width of the next iteration. For smoothing coefficients, The trace function of a matrix. To estimate the error covariance matrix, This represents the maximum kernel width. This invention enables the kernel width to automatically vary with the estimation error covariance matrix. Dynamic adjustment applies an exponential penalty to large residuals using a Gaussian kernel function; it achieves robust estimation driven by data without requiring prior knowledge of noise statistics. Step S2.2: Taking the maximization of the correlation entropy of the measured innovation sequence as the objective function, the filter gain is obtained by solving the objective function using the fixed-point iteration method. The objective function of the MCC optimization problem is to maximize the correlation entropy of the measurement news sequence, as shown in the following equation:

[0038] In the formula, State variables The corresponding objective function, For the first Individual Measurement News The cost function, To measure the amount of new information; right Regarding the differentiation of the state and setting the derivative to zero:

[0039] In the formula, Let be the derivative of the objective function. For the first A measurement equation with respect to state variables The derivative; Using unscented Kalman filtering to... The measurement equations are statistically linearized to approximate the following: Then the derivative of the objective function is shown in the following equation:

[0040] In the formula, For the first The estimated value of each measurement equation, For state variable residuals - The Each measured function value; The filter gain is obtained by solving the problem of maximizing the correlation entropy of the measurement innovation sequence using the fixed-point iteration method. 1) Initialization: Set the initial values ​​for state estimation. Estimating the initial matrix of error covariance initial value of kernel width Initial value of filter gain , The gain of the unscented Kalman filter; 2) Define the filter gain Mapping function As shown in the following formula:

[0041] In the formula, For filter gain, Let be the autocovariance matrix of the state variables. For the measurement matrix, For the observation noise matrix, This is a weight matrix based on filter gain; The fixed point is solved iteratively using the following formula:

[0042] In the formula, For the first The filter gain of the next iteration; When the sufficient condition for iterative convergence is met The MCC filter gain is obtained as shown in the following formula:

[0043] In the formula, For filter gain, Let be the autocovariance matrix of the state variables. For the measurement matrix, For the observation noise matrix, This is the weight matrix. , No. Individual weighted coefficients ; Step S2.3: Perform state update using the filter gain to update the measurement information; if the updated measurement information is in a normal state, the square of the kernel width is used as the kernel weighting coefficient to determine the filter gain; if the updated measurement information is in an abnormal state, the kernel weighting coefficient is 0 to determine the filter gain. Under ideal filtering conditions, such as accurate models and known and stable noise statistics, the innovation in a normal state satisfies the characteristics of zero mean, white noise, and a Gaussian distribution. Therefore, the innovation in an abnormal state deviates from these statistical characteristics, exhibiting significantly non-zero mean, mismatched covariance, and significant sequence correlation. Existing technologies primarily employ strategies for handling innovations in abnormal states, including online adjustment of noise covariance for adaptive filtering, robust filtering using robust estimation methods (such as M-estimation and Huber functions) to reduce the weight of abnormal innovations, combining multi-sensor data to locate fault sources, eliminating abnormal observations, and diagnosing and isolating faults, and adjusting system model parameters based on innovation feedback. This invention, however, uses different filtering gains to process measurement innovations in different states, as detailed below: When the state of the measured information is normal, Then the kernel weighting coefficient is The gain equation simplifies to:

[0044] when hour, It exhibits standard UKF characteristics; When the state of the measured information is abnormal, Then the kernel weighting coefficient is The filter gain is suppressed to: The gain degenerates into the pseudo-inverse of the measurement matrix, thus achieving constant value suppression.

[0045] In the method proposed in this invention, the filtering gain directly determines the weight of the innovation. If the filtering gain is too large, it may amplify the interference of observation noise, causing the estimated value to oscillate. If the filtering gain is too small, the system will be slow to respond to new observation information and the estimation will converge slowly. Therefore, this invention proposes to determine different kernel weighting coefficients and set different filtering gains according to the state of the innovation, thereby affecting the posterior covariance and improving the prediction accuracy of the next time step. In addition, the filtering gain also affects the identification of model parameters. Since there is a coupling between model parameters and state estimation, when the model parameters are augmented to the state vector, the update equation of the model parameters is completely consistent with the state variables. Therefore, the filtering gain also has a direct regulatory effect on the correction magnitude of parameter identification. However, the regulation of state variables and model parameters by the filtering gain has different dynamic characteristics. Since state variables usually have clear physical dynamics (such as displacement and velocity), the filtering gain needs to match their time correlation. Model parameters are usually slow time-varying or constant (such as friction coefficient and stiffness coefficient), and are less sensitive to the filtering gain. They need to be designed specifically through the initialization of parameter blocks in the covariance matrix. In the embodiments, the filter gain is obtained by solving the MCC optimization problem using the fixed-point iteration method. The kernel weighting coefficient in the filter gain is related to the kernel width and the measurement innovation, which enables the filter gain to be adaptively adjusted according to the magnitude of the measurement innovation. For small measurement innovations in normal states, the robust hybrid Kalman particle filter proposed in this invention is similar to the standard UKF. For large amounts of measurement innovations in abnormal states caused by bad data or model mismatch, the kernel function suppresses the influence of the filter gain to a constant value, thus naturally endowing the robust hybrid Kalman particle filter algorithm proposed in this invention with robustness.

[0046] Step S2.4: Update the adaptively adjusted kernel width using the updated measurement information, as shown in the following equation:

[0047]

[0048] In the formula, For the first Measurement updates in the next iteration As a smoothing coefficient, in the example The value is 0.95; , These are the lower and upper boundaries of the kernel width, respectively. In this embodiment, the lower boundary is... To suppress excessive decay, upper boundary To ensure normal information sensitivity; In this embodiment, the kernel width is updated three times using the updated measurement information to balance accuracy and real-time performance.

[0049] Step S3: Based on the particle set, perform robust hybrid Kalman particle filtering. The feasible region is established by the physical constraints of the model parameters. After resampling the particle set, if the projection of the model parameters in the new generation of particles exceeds the feasible region, the model parameters are set to the nearest boundary point and resampling is performed again.

[0050] Specifically, step S3 includes: Step S3.1: Resample the particle set using a robust hybrid Kalman particle filter; A robust hybrid Kalman particle filter is used to generate a proposal distribution for each particle, as shown in the following equation:

[0051] In the formula, Let be the probability density function of the proposed distribution. For iteration step The One particle, For iteration step Measurement information, This represents the mean value at the previous iteration step. This represents the covariance at the previous time step in the iteration.

[0052] The proposed distribution incorporates the latest measurement information determined based on the MCC criterion. It is not sensitive to bad data, so the generated proposal distribution is more reliable and accurate; Sampling from the proposed distribution yields a new generation of particles. ; Calculate and update particle weights The weight calculation includes the ratio of the actual state transition probability to the proposed distribution probability; Step S3.2: Establish the feasible region based on the physical constraints of the model parameters; after resampling the particle set, if the projection of the model parameters in the new generation of particles exceeds the feasible region, then set the model parameters to the nearest boundary point and resample again. Specifically, during the filtering process, physical constraints (such as...) are applied to the augmented state components representing the model parameters. The value of the measurement matrix , , These are the lower and upper bounds of the elements of the measurement matrix, respectively. After resampling, the parameter components of each particle are projected, and if they exceed the feasible region, they are pulled back to the nearest boundary point, which ensures the physical meaning of the parameter estimation results.

[0053] Step S3.3: Perform the MCMC move step on the resampled next-generation particles to mitigate the particle degradation problem; Step S3.4: Monitor the posterior covariance matrix of the estimation results of the new generation of particle sets in real time. If the uncertainty (variance) of the state estimation equation parameter continues to decrease and is below the threshold, it is considered that the parameter has converged. The corresponding process noise can be reduced as appropriate to lock in the optimal estimate.

[0054] Step S4: Use the weighted average of the particle set as the optimal joint estimate, and separate the dynamic state estimate and model parameter identification results from the optimal joint estimate. Specifically, the weighted average of the particle set is used as the optimal joint estimate at the prediction time, from which the dynamic state estimation and model parameter identification results can be separated simultaneously. The discrete distribution of the particle set provides the uncertainty quantification of the state and parameter estimates.

[0055] This invention also proposes a generator state estimation system that takes into account noise and parameter uncertainty constraints, comprising: The data processing module is used to acquire the generator's model parameters and dynamic state vector, establish the augmented state vector, and apply an unscented transformation to the augmented state vector to obtain the particle set. The filtering improvement module is used to improve the unscented Kalman filter to obtain a robust hybrid Kalman particle filter. It includes: during the unscented Kalman filtering of the augmented state vector, taking the maximization of the correlation entropy of the measurement innovation sequence as the objective function, and using the fixed-point iteration method to solve the objective function to obtain the filtering gain; when performing state updates using the filtering gain, determining the corresponding filtering gain based on the updated state of the measurement innovation. The filtering execution module is used to perform robust hybrid Kalman particle filtering based on the particle set. The feasible region is established by the physical constraints of the model parameters. After resampling the particle set, if the projection of the model parameters in the new generation of particles exceeds the feasible region, the model parameters are set to the nearest boundary point and resampling is performed again. The weighted average value of the particle set is used as the optimal joint estimate, and the dynamic state estimate and model parameter identification result are separated from the optimal joint estimate.

[0056] This disclosure can be a system, method, and / or computer program product. A computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for causing a processor to implement various aspects of this disclosure.

[0057] Computer-readable storage media can be tangible devices capable of holding and storing instructions for use by an instruction execution device. Computer-readable storage media can be, for example—but not limited to—electrical storage devices, magnetic storage devices, optical storage devices, electromagnetic storage devices, semiconductor storage devices, or any suitable combination of the foregoing. More specific examples (a non-exhaustive list) of computer-readable storage media include: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), static random access memory (SRAM), portable compact disc read-only memory (CD-ROM), digital multifunction disc (DVD), memory sticks, floppy disks, mechanical encoding devices, such as punch cards or recessed protrusions storing instructions thereon, and any suitable combination of the foregoing. The computer-readable storage media used herein are not to be construed as transient signals themselves, such as radio waves or other freely propagating electromagnetic waves, electromagnetic waves propagating through waveguides or other transmission media (e.g., light pulses through fiber optic cables), or electrical signals transmitted through wires.

[0058] The computer-readable program instructions described herein can be downloaded from computer-readable storage media to various computing / processing devices, or downloaded via a network, such as the Internet, local area network, wide area network, and / or wireless network, to an external computer or external storage device. The network may include copper transmission cables, fiber optic transmission, wireless transmission, routers, firewalls, switches, gateway computers, and / or edge servers. A network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards them to the computer-readable storage media in the respective computing / processing device.

[0059] Computer program instructions used to perform the operations of this disclosure may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, status setting data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Smalltalk, C++, etc., and conventional procedural programming languages ​​such as the "C" language or similar programming languages. The computer-readable program instructions may execute entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving a remote computer, the remote computer may be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or may be connected to an external computer (e.g., via the Internet using an Internet service provider). In some embodiments, electronic circuitry, such as programmable logic circuitry, field-programmable gate arrays (FPGAs), or programmable logic arrays (PLAs), is personalized by utilizing the status information of the computer-readable program instructions to implement various aspects of this disclosure.

[0060] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the protection scope of the claims of the present invention.

Claims

1. A generator state estimation method considering noise and parameter uncertainty constraints, characterized in that, include: Obtain the generator's model parameters and dynamic state vector, and establish an augmented state vector; The particle set is obtained by applying an unscented transformation to the augmented state vector; The robust hybrid Kalman particle filter is obtained by improving the unscented Kalman filter. The process includes: during the unscented Kalman filtering of the augmented state vector, the objective function is to maximize the correlation entropy of the measurement innovation sequence, and the filter gain is obtained by solving the objective function using the fixed-point iteration method; when performing state update using the filter gain, the corresponding filter gain is determined according to the state of the updated measurement innovation. Based on the particle set, a robust hybrid Kalman particle filter is performed. The feasible region is established by the physical constraints of the model parameters. After the particle set is resampled, if the projection of the model parameters in the new generation of particles exceeds the feasible region, the model parameters are set to the nearest boundary point and resampled again. The weighted average of the particle set is used as the optimal joint estimate, and the dynamic state estimate and model parameter identification results are separated from the optimal joint estimate.

2. The generator state estimation method considering noise and parameter uncertainty constraints according to claim 1, characterized in that, Model parameters of the generator With dynamic state Merge to form an augmented state vector Among them, model parameters Including: transient reactance Dynamic state Includes: functional angles Rotor speed Transient potential .

3. The generator state estimation method considering noise and parameter uncertainty constraints according to claim 1, characterized in that, The Gaussian kernel function is introduced as the cost function for measuring innovation, as shown in the following equation: In the formula, To measure new information The cost function; For kernel width; Establish an adaptively adjustable kernel width, as shown in the following formula: In the formula, For the first The kernel width of the next iteration. For smoothing coefficients, The trace function of a matrix. To estimate the error covariance matrix, This represents the maximum kernel width.

4. The generator state estimation method considering noise and parameter uncertainty constraints according to claim 3, characterized in that, The objective function is to maximize the correlation entropy of the measured information sequence, as shown in the following equation: In the formula, State variables The corresponding objective function, For the first Individual Measurement News The cost function, To measure the amount of new information; Using unscented Kalman filtering to... The measurement equations are statistically linearized to approximate the following: Then the derivative of the objective function is shown in the following equation: In the formula, For the first The estimated value of each measurement equation, For state variable residuals - The Each measurement function value.

5. The generator state estimation method considering noise and parameter uncertainty constraints according to claim 4, characterized in that, The filter gain is obtained by solving the objective function using the fixed-point iteration method, including: Set initial values ​​for state estimation Estimating the initial matrix of error covariance initial value of kernel width Initial value of filter gain , The gain of the unscented Kalman filter; Define filter gain Mapping function As shown in the following formula: In the formula, For filter gain, Let be the autocovariance matrix of the state variables. For the measurement matrix, For the observation noise matrix, This is a weight matrix based on filter gain; The fixed point is solved iteratively using the following formula: In the formula, For the first The filter gain of the next iteration; When the sufficient condition for iterative convergence is met The MCC filter gain is obtained as shown in the following formula: In the formula, For filter gain, Let be the autocovariance matrix of the state variables. For the measurement matrix, For the observation noise matrix, This is the weight matrix. , No. Individual weighted coefficients .

6. The generator state estimation method considering noise and parameter uncertainty constraints according to claim 5, characterized in that, The state is updated using the filter gain to update measurement information; If the updated measurement information is normal, the kernel weighting coefficient is the square of the kernel width to determine the filter gain; if the updated measurement information is abnormal, the kernel weighting coefficient is 0 to determine the filter gain.

7. The generator state estimation method considering noise and parameter uncertainty constraints according to claim 6, characterized in that, The adaptively adjusted kernel width is updated using the updated measurement information, as shown in the following equation: In the formula, For the first The kernel width of the next iteration. For the first Measurement updates in the next iteration For smoothing coefficients, , These represent the lower and upper boundaries of the kernel width, respectively.

8. A generator state estimation system considering noise and parameter uncertainty constraints, used to implement the generator state estimation method considering noise and parameter uncertainty constraints as described in any one of claims 1 to 7, characterized in that, include: The data processing module is used to acquire the generator's model parameters and dynamic state vectors, and to establish augmented state vectors. The particle set is obtained by applying an unscented transformation to the augmented state vector; The filtering improvement module is used to improve the unscented Kalman filter to obtain a robust hybrid Kalman particle filter. It includes: during the unscented Kalman filtering of the augmented state vector, taking the maximization of the correlation entropy of the measurement innovation sequence as the objective function, and using the fixed-point iteration method to solve the objective function to obtain the filtering gain; when performing state updates using the filtering gain, determining the corresponding filtering gain based on the updated state of the measurement innovation. The filtering execution module is used to perform robust hybrid Kalman particle filtering based on the particle set. The feasible region is established by the physical constraints of the model parameters. After resampling the particle set, if the projection of the model parameters in the new generation of particles exceeds the feasible region, the model parameters are set to the nearest boundary point and resampling is performed again. The weighted average value of the particle set is used as the optimal joint estimate, and the dynamic state estimate and model parameter identification result are separated from the optimal joint estimate.

9. A terminal, comprising a processor and a storage medium; characterized in that: The storage medium is used to store instructions; The processor is configured to operate according to the instructions to perform the steps of the method according to any one of claims 1-7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the program implements the steps of the method according to any one of claims 1-7.

Images