Power grid prediction state estimation method and system based on square root unscented particle filtering

By using a square root unscented particle filter-based method, the problem of insufficient estimation accuracy and robustness in power grid dynamic state estimation is solved. It achieves high accuracy and stability improvement under non-Gaussian noise conditions and is applicable to power systems with new energy access.

CN121434552APending Publication Date: 2026-01-30STATE GRID HENAN ELECTRIC POWER ELECTRIC POWER SCI RES INST +2
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202511486753.2
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2025-10-17
Publication Date
2026-01-30

AI Technical Summary

Technical Problem

Among existing power grid dynamic state estimation methods, traditional extended Kalman filters and unscented Kalman filters have low estimation accuracy when facing nonlinear and non-Gaussian noise, and are prone to non-positive definiteness of the covariance matrix and rounding errors. Particle filters are greatly affected by the target density function, resulting in low filtering accuracy.

Method used

A method based on square root unscented particle filtering is adopted. By establishing a predictive state estimation model of the power system, the square root unscented Kalman filter is used to generate a proposal distribution for the resampled particles. The weights are updated through Cholisky decomposition and MCMC moving step to improve the estimation accuracy and robustness.

Benefits of technology

Simulations on the IEEE-30 bus system show that the proposed method reduces the accumulation of truncation error, maintains the positive definiteness of the state prediction error covariance matrix, and improves estimation accuracy and robustness, especially exhibiting better accuracy and stability under non-Gaussian noise conditions.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN121434552A_ABST
    Figure CN121434552A_ABST
Patent Text Reader

Abstract

The invention discloses a power grid prediction state estimation method and system based on square root unscented particle filtering, and the method comprises the steps: obtaining the historical operation data of a power grid, generating an initial particle set, and obtaining a state variable estimation value of the power grid and the weight of the state variable estimation value through employing a square root unscented particle filtering method based on a prediction state estimation model; in iteration, calculating a weighted covariance matrix by using a state variable estimation value of the power grid and a corresponding normalized weight, performing Cholesky decomposition on the weighted covariance matrix to obtain a square root of the weighted covariance matrix, and taking the square root as a square root estimation value of each particle after resampling; the number of effective particles is used as a normalized weight initial value of each particle after resampling; the state variable estimation value of the power grid and the weighted mean value of the weight of the state variable estimation value serve as the prediction state estimation result of the power system, and the accuracy and robustness of state estimation of the power system with new energy access are improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention belongs to the field of power system state estimation technology, specifically, it relates to a robust prediction-aided state estimation method and system for power systems based on square root unscented particle filtering. Background Technology

[0002] Power system state estimation is mainly divided into static estimation (SSE) and dynamic estimation (DSE). Static estimation is mainly based on the measurement data of a certain cross-section at a certain moment to estimate the real-time state of the power system. Dynamic state estimation not only uses the measurement information at the current moment, but also the state prediction value at the previous moment, and obtains the estimated and predicted values ​​of the grid operating state, providing reliable information support for the grid energy management system.

[0003] In existing methods for establishing and estimating power grid dynamic state estimation models, the Extended Kalman Filter (EKF) and its improved versions are widely used. However, traditional EKF-based power system dynamic estimation methods neglect higher-order terms when linearizing nonlinear functions using Taylor series, resulting in low estimation accuracy. Furthermore, the Unscented Kalman Filter (UKF) algorithm does not require calculating the Jacobian matrix but uses deterministic sampling to approximate the probability distribution of the nonlinear function; however, with further filtering calculations, rounding errors caused by the calculations lead to the loss of the nonnegativity and symmetry of the covariance matrix. Additionally, a square root unscented Kalman filter (SR-UKF) based on UKF is established directly by updating the square root of the state covariance matrix through matrix QR decomposition and Cholesky decomposition. However, the Kalman filter used can only be applied to Gaussian distribution models, while the Particle Filter (PF) is unaffected by Gaussian distribution. The PF filter is an optimal recursive Bayesian filtering method based on Monte Carlo simulation, and its filtering accuracy is largely influenced by the selected target density function. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a power grid prediction state estimation method and system based on square root unscented particle filtering, which improves the accuracy and robustness of state estimation for power systems with renewable energy access.

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

[0006] This invention proposes a power grid predictive state estimation method based on square root unscented particle filtering, comprising: Step 1: Establish a predictive state estimation model for the power system; Step 2: Obtain historical power grid operation data to generate an initial particle set. Based on the predictive state estimation model, use the square root unscented particle filter method to obtain the estimated values ​​of the power grid state variables and their weights. During the iteration, the weighted covariance matrix is ​​calculated using the estimated state variables of the power grid and their corresponding normalized weights. The square root of the weighted covariance matrix is ​​obtained by performing Choliski decomposition on the weighted covariance matrix, which is used as the square root estimate of each particle after resampling. The number of effective particles is used as the initial value of the normalized weight of each particle after resampling. The square root unscented Kalman filter method is used to generate a proposal distribution for each particle after resampling, and a new particle set is obtained by sampling from the proposal distribution. After performing the MCMC moving step on the new particle set, the normalized weight of each particle after resampling is updated. Step 3: Use the estimated values ​​of the power grid's state variables and their weighted average as the predicted state estimation result of the power system.

[0007] Predictive state estimation model; described as follows:

[0008] In the formula, For a moment State variables; The system equations; For a moment System noise; For a moment The observed variables; For measurement equations; For a moment Observation noise.

[0009] Step 2 includes: Step 2.1: Obtain historical power grid operation data and generate an initial particle set and initial weights; Step 2.2, use Sigma point scaling correction sampling, around time... The Each particle constructs a Sigma point and weights; Step 2.3: Based on the system equations, perform a one-step prediction to obtain the predicted point set and its weighted mean. Step 2.4: Construct the elements of the first matrix using the predicted point set and its weighted mean. Satisfying the state quantity prediction error covariance matrix ; Step 2.5, for The square root of the state variable prediction error covariance matrix is ​​obtained by performing QR decomposition. The obtained square root is updated using the Jolesi rank-1 decomposition algorithm. Update.

[0010] Step 2.6: Based on the measurement equation, use the predicted point set to update the observed variables and calculate the measurement mean of the Sigma points; Step 2.7: Construct the elements of the second matrix using the observed variables and the measurement mean. ; Step 2.8, for The square root of the autocovariance matrix of the measurement values ​​is obtained by performing QR decomposition. The obtained square root is updated using the Jolesi rank-1 decomposition algorithm. Update; Step 2.9, Calculate the Kalman gain ; Step 2.10: Obtain the estimated state variables of the system using Kalman gain. ; Step 2.11, using Kalman gain The square root of the autocovariance matrix of the measured values The square root of the covariance matrix of the state quantity prediction error The state variable prediction error covariance matrix is ​​obtained by using the Jolesi rank-1 decomposition update algorithm. Square root estimate ; Step 2.12: Update and normalize the weights of the state variable estimates; Step 2.13: Calculate the effective number of particles using normalized weights. ,like , If the number of particles is [value], then resampling is performed on the particle set; if [value] is [value], then [value] is performed on the particle set. Then output the estimated value of the state variable. ;

[0011] In the formula, These are the normalized weights for the estimated values ​​of the state variables.

[0012] Step 2.14: Calculate the weighted covariance matrix using the estimated state variables of the power grid and their corresponding normalized weights, as shown in the following equation: ,

[0013] In the formula, It is a weighted covariance matrix; Step 2.15: Perform Choliski decomposition on the weighted covariance matrix to obtain the square root of the weighted covariance matrix. ;by As the square root estimate of each particle after resampling, and with These are the initial normalized weights for each particle after resampling; Step 2.16: Generate a proposal distribution for each resampled particle using the square root unscented Kalman filter method; sample a new particle set from the proposal distribution; update the normalized weights of each resampled particle after performing the MCMC shift step on the new particle set; and estimate the resampled state variables. , The updated normalized weights are then substituted into step 2.2 for the next iteration.

[0014] Update the normalized weights of each particle after resampling using the following relationship: :

[0015] In the formula, For new particles Obtain the observed variables The prior probability; For state variable estimates New particles were obtained The prior probability; For state variable estimates and observed variables New particles obtained The proposed distribution density function.

[0016] A Gaussian distribution is obtained by randomly selecting a particle as the center from the resampled particle set, and candidate particles are obtained by sampling from the Gaussian distribution. ; The probability of acceptance is calculated using the following formula. :

[0017] In the formula, Based on observed variables Obtain candidate particles The prior probability; Based on observed variables New particles were obtained The prior probability; Based on acceptance probability Decide whether to use candidate particles Replacement particles When accepting probability When the value is not less than 0.7, use candidate particles. Replacement particles Otherwise, retain the particles. .

[0018] This invention also proposes a power grid predictive state estimation system based on square root unscented particle filtering, comprising: The model building module is used to build predictive state estimation models for power systems. The square root unscented particle filter module is used to acquire historical power grid operation data and historical power flow data to generate an initial particle set. Based on the predictive state estimation model, the square root unscented particle filter method is used to obtain the estimated values ​​of the power grid's state variables and their weights. In the iteration, the weighted covariance matrix is ​​calculated using the estimated values ​​of the power grid's state variables and their corresponding normalized weights. The square root of the weighted covariance matrix is ​​obtained by performing Choliski decomposition on the weighted covariance matrix, which is used as the square root estimate of each particle after resampling. The number of effective particles is used as the initial value of the normalized weight of each particle after resampling. The square root unscented Kalman filter method is used to generate a proposal distribution for each particle after resampling, and a new particle set is obtained by sampling from the proposal distribution. After implementing the MCMC moving step on the new particle set, the normalized weight of each particle after resampling is updated. The state estimation module is used to take the weighted average of the estimated values ​​of the power grid's state variables and their weights as the predicted state estimation result of the power system.

[0019] 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.

[0020] 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.

[0021] The beneficial effects of this invention are as follows: Compared with the prior art, it addresses the problem of reduced dynamic estimation accuracy or even filter divergence caused by the non-positive definiteness of the state prediction error covariance matrix. This invention introduces square root technology within the framework of the unscented particle filter algorithm, establishing a power system dynamic estimation method based on the square root-based unscented particle filter algorithm. Simulation experiments were conducted on the IEEE-30 bus system. During the iterative calculation of dynamic estimation, the method proposed in this invention can reduce the accumulation of truncation errors, maintain the positive definiteness of the state prediction error covariance matrix, and improve estimation accuracy. Using the square root of the state prediction error covariance matrix instead of the state prediction error variance matrix for calculation avoids calculating the square root of the state prediction error covariance matrix in each iteration, effectively overcoming the non-positive definiteness of the state prediction error covariance matrix and improving the robustness of dynamic estimation. Under non-Gaussian noise conditions, the square root-based unscented particle filter algorithm has better accuracy and robustness than both the unscented particle filter algorithm and the unscented Kalman filter algorithm. This invention ensures the convergence speed and stability of the filter and can be well applied to power systems with renewable energy integration. Attached Figure Description

[0022] Figure 1This is a flowchart of a power grid prediction state estimation method based on square root unscented particle filtering proposed in this invention. Detailed Implementation

[0023] 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.

[0024] Traditional state estimation methods (such as weighted least squares WLS) suffer from a sharp decline in performance when faced with strong nonlinearity (such as power fluctuations in wind and solar power) and non-Gaussian noise (measurement and prediction errors) brought by new energy sources. To improve the accuracy and robustness of state estimation for power systems with new energy access, this invention proposes a grid prediction state estimation method based on square root unscented particle filtering. In the particle filtering process, the Markov chain Monte Carlo method is used to maintain the diversity of particles. like Figure 1 As shown, the method includes: Step 1: Establish a predictive state estimation model for the power system; described as follows:

[0025] In the formula, For a moment State variables; For the system equations, the two-parameter exponential smoothing method is used to establish the system equations in the embodiment. The exponential smoothing method is a simple short-term load forecasting method with fewer stored variables, and has the advantages of fewer stored variables and fast calculation speed. For a moment System noise; For a moment The observed variables include: the phase angle and amplitude of the power system node voltage, the measured values ​​of branch active power and reactive power, and the bus active power and reactive power; The measurement equation is a nonlinear function; For a moment Observation noise.

[0026] In the embodiment, power flow calculation is used to obtain the power flow value of each part, namely the step voltage phase angle and voltage amplitude, by combining the given load factor and generator factor. The value of each direct measurement signal is obtained by superimposing zero-average Gaussian white noise on the true value of the state variables. In the embodiments, using the two-parameter exponential smoothing method to construct the system equations is a non-limiting but preferred choice.

[0027] Step 2: Obtain historical power grid operation data to generate an initial particle set. Based on the predictive state estimation model, use the square root unscented particle filter method to obtain the estimated values ​​of the power grid state variables and their normalized weights. During the iteration, the weighted covariance matrix is ​​calculated using the estimated state variables of the power grid and their normalized weights. The square root of the weighted covariance matrix is ​​obtained by performing Choliski decomposition on the weighted covariance matrix, which is used as the square root estimate of each particle after resampling. The number of effective particles is used as the initial value of the normalized weight of each particle after resampling. The square root unscented Kalman filter method is used to generate a proposal distribution for each particle after resampling, and a new particle set is obtained by sampling from the proposal distribution. After performing the MCMC moving step on the new particle set, the normalized weight of each particle after resampling is updated.

[0028] Specifically, step 2 includes: Step 2.1: Obtain historical power grid operation data and generate an initial particle set and initial weights; When =0, historical running data is used as the initial state variable. Generate an initial particle set near the initial state variables. , , For the number of particles, ~Prior Distribution , No. Initial weights of each particle ; The output of new energy sources, load fluctuations, and related measurement errors often do not follow a Gaussian distribution. Particle filtering does not rely on any distribution assumptions and directly uses a set of particles to approximate arbitrarily complex posterior probability distributions, thus it can more accurately describe the true uncertainty of the state.

[0029] Step 2.2, use Sigma point scaling correction sampling, around time... The Particles Construct Sigma points and weight , ; ,

[0030]

[0031] In the formula, As a scaling correction factor, it determines the Sigma point's position around the particle. The range of fluctuations; It is prior knowledge; , This is the scaling parameter. The scaling parameter is set to 0 in state estimation. The dimension of the state variables; For a moment particle The square root of the corresponding state quantity error covariance matrix.

[0032] Step 2.3: Based on the system equations, perform a one-step prediction to obtain the prediction point set. and its weighted mean ; ,

[0033] In the formula, For at any time The time obtained by performing a one-step prediction +1th The first state variable One prediction point.

[0034] Step 2.4, using the predicted point set and its weighted mean Construct the first matrix; the elements of the first matrix satisfy ,in, The state variable prediction error covariance matrix is... .

[0035] Step 2.5, for QR decomposition yields the state variable prediction error covariance matrix. square root The obtained square root is updated using the Jolesi rank-1 decomposition algorithm. Update;

[0036]

[0037] In the formula, qr() is the QR decomposition function, and choleupdate() is the Cholesi rank-1 decomposition update algorithm function.

[0038] Step 2.6, based on the measurement equation, utilize the predicted point set. Update observed variables And calculate the mean of measurements at Sigma points ; ,

[0039] Step 2.7, using observed variables and measurement mean Construct a second matrix, the elements of the second matrix. As shown in the following formula:

[0040] Step 2.8, for The square root of the autocovariance matrix of the measurement values ​​is obtained by performing QR decomposition. The obtained square root is updated using the Jolesi rank-1 decomposition algorithm. Update;

[0041]

[0042] In the formula, qr() is the QR decomposition function, and choleupdate() is the Cholesi rank-1 decomposition update algorithm function.

[0043] Step 2.9, Calculate the Kalman gain As shown in the following formula:

[0044] In the formula, The square root matrix of the cross-covariance between predicted and measured values; Step 2.10: Obtain the estimated state variables of the system using Kalman gain. As shown in the following formula:

[0045] Step 2.11, using Kalman gain The square root of the autocovariance matrix of the measured values The square root of the covariance matrix of the state quantity prediction error The state variable prediction error covariance matrix is ​​obtained by using the Jolesi rank-1 decomposition update algorithm. Square root estimate As shown in the following formula:

[0046] Step 2.12, update and normalize the weights of the state variable estimates, as shown in the following equation:

[0047]

[0048] In the formula, The distribution function, Let be the probability density function. These are the normalized weights for the estimated values ​​of the state variables.

[0049] Step 2.13: Calculate the effective number of particles using normalized weights. ,like If so, then resampling is performed on the particle set in step 2.14; if Then output the estimated value of the state variable. ; The effective number of particles is shown in the following formula:

[0050] Step 2.14: Calculate the weighted covariance matrix using the estimated state variables of the power grid and their corresponding normalized weights, as shown in the following equation: ,

[0051] In the formula, It is a weighted covariance matrix; Step 2.15: Perform Choliski decomposition on the weighted covariance matrix to obtain the square root of the weighted covariance matrix. ;by As the square root estimate of each particle after resampling, and with These are the initial normalized weights for each particle after resampling; Particle filtering is very sensitive to bad data. A single bad data point can contaminate the entire estimation result. The weights of some particles affected by bad data will be significantly reduced, and they will be eliminated along with the particles corresponding to the bad data, making the state estimation inaccurate. This invention redefines the weights before performing resampling. Through resampling and multiple iterations, particles corresponding to bad data are accurately eliminated, while particles affected by bad data are still retained, thus giving the system strong robustness.

[0052] This invention avoids the underestimation of filtering error caused by simple multiple copies during resampling by introducing the square root of the weighted covariance matrix, representing the overall uncertainty, into the resampling process, thus maintaining the consistency of uncertainty estimation and the stability of the algorithm. As the square root estimate of each particle after resampling, it avoids the covariance matrix from losing its positive definiteness during the iteration process, and can also save computation time for the next iteration. After step 2.16, it is directly substituted into step 2.5 for the next iteration, which ensures the numerical stability and reliability of the entire filtering process and improves the filtering efficiency.

[0053] Step 2.16: Generate a proposal distribution for each resampled particle using the square root unscented Kalman filter method; sample a new particle set from the proposal distribution; update the normalized weights of each resampled particle after performing the MCMC shift step on the new particle set; and estimate the resampled state variables. , The updated normalized weights are then substituted into step 2.5 for the next iteration; This invention uses the square root unscented Kalman filter method to generate a proposal distribution for each resampled particle, which can more accurately capture the nonlinear characteristics of the state equation and measurement equation, providing a higher quality proposal distribution for particle filtering, making particles more likely to appear near the true state. Specifically, in At time 1, the square root unscented Kalman filter (SR-UKF) method is used to estimate the state variables. Generate a distribution that follows the suggested distribution new particles Since the proposed distribution provided by SR-UKF is closer to the true posterior distribution, this invention uses SR-UKF and performs the MCMC moving step to generate a superior new particle that follows the proposed distribution for each particle, thereby improving particle quality. Update the normalized weights of each particle after resampling using the following relationship: :

[0054] In the formula, For new particles Obtain the observed variables The prior probability; For state variable estimates New particles were obtained The prior probability; For state variable estimates and observed variables New particles obtained The proposed distribution density function; The MCMC moving steps include, but are not limited to, the Metropolis-Hastings algorithm. The MCMC step is crucial for accuracy. Without diversity, all particles cluster around a potentially "wrong" point (especially when the system undergoes a sudden change), leading to estimates that deviate significantly from the true value. MCMC, by moving particles, allows the particle set to cover multiple possible high-probability regions (i.e., addressing multimodal distribution problems), thereby suppressing divergence, maintaining diversity, and finding estimates that are closer to the true state.

[0055] A Gaussian distribution is obtained by randomly selecting a particle as the center from the resampled particle set, and candidate particles are obtained by sampling from the Gaussian distribution. ; The probability of acceptance is calculated using the following formula. :

[0056] In the formula, Based on observed variables Obtain candidate particles The prior probability; Based on observed variables New particles were obtained The prior probability; Based on acceptance probability Decide whether to use candidate particles Replacement particles When accepting probability When the value is not less than 0.7, use candidate particles. Replacement particles Otherwise, retain the particles. The acceptance probability can be obtained by training the model's target value. The MCMC moving step allows particles to explore and move in high-probability regions. Even identical particles that are copied will become different after the MCMC moving step, thus greatly increasing particle diversity and avoiding sample impoverishment.

[0057] Step 3: Use the estimated values ​​of the power grid's state variables and their weighted average as the predicted state estimation result of the power system.

[0058] This invention utilizes simulation analysis of the IEEE 30-node system. In the simulation, the system power flow calculation results are real values, while the measured data are formed by superimposing corresponding random measurement errors onto the power flow results of the test system. MATLAB was used to test the program under normal Gaussian noise conditions, and simulation analysis was performed on two test systems of the IEEE 30 system with a particle count of 100. Table 1 shows the specific values ​​of the root mean square error (RMSE) for different algorithms. The RMSE analysis results in Table 1 show that the square root-based unscented particle filter algorithm proposed in this invention has higher filtering accuracy and is significantly superior to the unscented Kalman filter and unscented particle filter algorithms in estimating the RMSE performance index. The square root-based unscented particle filter algorithm uses an improved square root decomposition. During the covariance matrix update calculation, the square root of the covariance matrix is ​​iteratively updated, thus avoiding the damage to the positive definite covariance caused by error accumulation in numerical calculations. This enhances robustness and tracking ability, while improving the stability of the filter during numerical calculations.

[0059] Table 1. Root Mean Square Error of Gaussian Noise

[0060] Different algorithms were used to predict the auxiliary state estimation, and the results are shown in the figure. Table 2 shows the average and maximum values ​​of the step voltage phase angle and voltage amplitude.

[0061] Table 2. Root Mean Square Error of Non-Gaussian Noise

[0062] Based on the voltage amplitude and phase angle estimation results of the IEEE 30 test system, the estimation performance of the square root-based unscented particle filter algorithm, which incorporates non-Gaussian noise, is significantly better than that of the unscented Kalman filter and the unscented particle filter algorithms. Under normal operating conditions and in non-Gaussian noise environments, the root mean square error (RMSE) of both the unscented Kalman filter and the unscented particle filter algorithms is on the order of magnitude. Under normal operating conditions, the RMSE of the square root-based unscented particle filter algorithm is approximately one-tenth that of the unscented particle filter algorithm. Under non-Gaussian noise conditions, the RMSE of the square root-based unscented particle filter algorithm is approximately one-thousandth that of the unscented particle filter algorithm. The square root-based unscented particle filter algorithm used in this invention exhibits excellent accuracy and robustness under non-Gaussian noise conditions.

[0063] This invention also proposes a power grid predictive state estimation system based on square root unscented particle filtering, comprising: The model building module is used to build predictive state estimation models for power systems. The square root unscented particle filter module is used to acquire historical power grid operation data and historical power flow data to generate an initial particle set. Based on the predictive state estimation model, the square root unscented particle filter method is used to obtain the estimated values ​​of the power grid's state variables and their weights. In the iteration, the weighted covariance matrix is ​​calculated using the estimated values ​​of the power grid's state variables and their corresponding normalized weights. The square root of the weighted covariance matrix is ​​obtained by performing Choliski decomposition on the weighted covariance matrix, which is used as the square root estimate of each particle after resampling. The number of effective particles is used as the initial value of the normalized weight of each particle after resampling. The square root unscented Kalman filter method is used to generate a proposal distribution for each particle after resampling, and a new particle set is obtained by sampling from the proposal distribution. After implementing the MCMC moving step on the new particle set, the normalized weight of each particle after resampling is updated. The state estimation module is used to take the weighted average of the estimated values ​​of the power grid's state variables and their weights as the predicted state estimation result of the power system.

[0064] 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.

[0065] 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.

[0066] 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.

[0067] 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.

[0068] 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 power grid predictive state estimation method based on square root unscented particle filter, characterized in that, Comprising: Step 1, establishing a predictive state estimation model of the power system; Step 2, obtaining historical operation data of the power grid to generate an initial particle set, and based on the predictive state estimation model, using a square root unscented particle filtering method to obtain state variable estimation values and their weights of the power grid; In iteration, the state variable estimation values and their corresponding normalized weights of the power grid are used to calculate a weighted covariance matrix, and Cholesky decomposition is performed on the weighted covariance matrix to obtain the square root of the weighted covariance matrix as the square root estimation value of each particle after resampling, and the number of effective particles is taken as the initial value of the normalized weight of each particle after resampling; a square root unscented Kalman filtering method is used to generate a proposal distribution for each particle after resampling, and new particles are sampled from the proposal distribution; after the MCMC moving step is performed on the new particle set, the normalized weight of each particle after resampling is updated; Step 3, taking the weighted mean of the state variable estimation values and their weights of the power grid as the predictive state estimation result of the power system.

2. The power grid predictive state estimation method based on square root unscented particle filtering according to claim 1, wherein the predictive state estimation model is described as follows:

3. The power grid predictive state estimation method based on square root unscented particle filtering according to claim 2, wherein step 2 comprises: wherein is the state variable at time ; is the system equation; is the system noise at time ; is the observation variable at time ; is the measurement equation; is the observation noise at time . Step 2.1, obtaining historical operation data of the power grid to generate an initial particle set and initial weights; Step 2.3, performing one-step prediction based on the system equation to obtain a prediction point set and their weighted mean; 4. The power grid predictive state estimation method based on square root unscented particle filtering according to claim 3, wherein step 2 further comprises: Step 2.2, Sigma point proportional modification sampling, construct Sigma points and weights around the time instant of the first particle; Step 2.6, based on the measurement equation, using the prediction point set to update the measurement mean of the observation variable and calculate the Sigma point; Step 2.

4. Constructing the elements in the first matrix using the set of prediction points and their weighted mean satisfying the state quantity prediction error covariance matrix ; Step 2.5, QR decomposition of the state prediction error covariance matrix to obtain its square root and update the obtained square root using the Cholesky rank-1 update algorithm. Step 2.12, updating the weight of the state variable estimation value and normalizing; 5. The power grid predictive state estimation method based on square root unscented particle filtering according to claim 4, wherein step 2 further comprises: Step 2.14, using the state variable estimation values and their corresponding normalized weights of the power grid to calculate a weighted covariance matrix, as shown in the following formula: Step 2.

7. Constructing elements in the second matrix using the observed variables and the measured mean values ; Step 2.8, for The square root of the autocovariance matrix of the measurement values ​​is obtained by performing QR decomposition. The obtained square root is updated using the Jolesi rank-1 decomposition algorithm. Update; Step 2.9, compute Kalman gain ; Step 2.

10. Obtain the state variable estimates of the system using the Kalman gain ; Step 2.11, using Kalman gain , square root of measurement self-covariance matrix , square root of state quantity prediction error covariance matrix , square root estimation value of state quantity prediction error covariance matrix obtained by using Cholesky rank 1 decomposition update algorithm ;​ 6. The power grid predictive state estimation method based on square root unscented particle filtering according to claim 5, wherein the step of performing Cholesky decomposition on the weighted covariance matrix to obtain the square root of the weighted covariance matrix comprises: Step 2.

13. Calculate the effective number of particles using the normalization weight , if , is the number of particles, then perform resampling on the particle set; if , output the state variable estimate ; In the formula, is the normalized weight of the state variable estimate.

7. The power grid predictive state estimation method based on square root unscented particle filtering according to claim 5, wherein the step of using a square root unscented Kalman filtering method to generate a proposal distribution for each particle after resampling comprises: Comprising: A model establishing module for establishing a predictive state estimation model of the power system; , In the formula, is a weighted covariance matrix; Step 2.

15. Cholesky decomposition of the weighted covariance matrix to obtain the square root of the weighted covariance matrix ; and as the initial values of the normalized weights of the particles after resampling . Step 2.

16. Generate a proposal distribution for each resampled particle using the square root unscented Kalman filter method, sample a new set of particles from the proposal distribution, update the normalized weight of each resampled particle after implementing the MCMC move step on the new set of particles, and update the state variable estimate 、 and the updated normalized weight into Step 2.2 for the next iteration. A square root unscented particle filtering module for obtaining historical operation data and historical power flow data of the power grid to generate an initial particle set, and based on the predictive state estimation model, using a square root unscented particle filtering method to obtain state variable estimation values and their weights of the power grid; The normalized weight of each particle after resampling is updated with the following relationship : where is the new particle based on the state variable estimate and the observation variable is the prior probability of the new particle based on the state variable estimate is the prior probability of the new particle based on the state variable estimate and the observation variable is the density function of the proposal distribution of the new particle based on the state variable estimate ​ selecting a particle randomly from the resampled particle set as a center to obtain a Gaussian distribution, sampling a candidate particle from the Gaussian distribution ; The acceptance probability is calculated in the following relation : wherein is based on the observation variable obtains the prior probability of the candidate particle ; is based on the observation variable obtains the prior probability of the new particle ; according to acceptance probability decide whether to replace the particle with the candidate particle replace the particle when the acceptance probability is not less than 0.7, replace the particle with the candidate particle replace the particle , otherwise, keep the particle .

8. A power grid predictive state estimation system based on square root unscented particle filter, for implementing the power grid predictive state estimation method based on square root unscented particle filter according to any one of claims 1 to 7, characterized in that, ​ ​ ​ In the iteration, the weighted covariance matrix is calculated by using the state variable estimation value of the power grid and its corresponding normalized weight, the weighted covariance matrix is subjected to Cholesky decomposition to obtain the square root of the weighted covariance matrix as the square root estimation value of each particle after resampling, and the effective particle number is taken as the initial value of the normalized weight of each particle after resampling; the square root unscented Kalman filtering method is used to generate a proposal distribution for each particle after resampling, and a new particle set is obtained by sampling from the proposal distribution; the normalized weight of each particle after resampling is updated after the MCMC moving step is implemented on the new particle set; The state estimation module is configured to take the weighted mean of the state variable estimation value of the power grid and its weight as the predicted state estimation result of the power system. 9.A terminal, comprising a processor and a storage medium; characterized in that: The storage medium is configured to store instructions; The processor is configured to operate according to the instructions to perform the steps of the method of any one of claims 1-7.

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