Comprehensive evaluation and analysis method for state estimation precision of power system

Abstract

The invention belongs to the technical field of power system operation monitoring and control, and relates to a power system state estimation precision comprehensive evaluation and analysis method. A traditional method depends on a single data source and is poor in model adaptability and incomplete in precision evaluation. According to the method, firstly, data are collected and fused through an SCADA system and a PMU, weights are calculated for all features according to measurement accuracy, and key features such as a power factor, a voltage-current ratio and a power change rate are extracted; then, an SVM and an RBF kernel function are selected to construct a model, and after parameters are reasonably initialized, grid search and cross validation optimization are carried out; and finally, evaluating the performance of the model from multiple dimensions by using multivariate indexes such as a mean square error and the like. The method effectively fuses multi-source data, optimizes the model and comprehensively evaluates, significantly improves the precision, reliability and adaptability of state estimation of the power system, ensures safe and stable operation of the system, and provides a better scheme for state estimation of the power system.

Description

Technical Field

[0001] The present invention relates to the technical field of power system operation monitoring and control, and in particular to a comprehensive evaluation and analysis method for power system state estimation accuracy. Background Art

[0002] In the operation of power systems, accurate state estimation is crucial for system monitoring, scheduling and optimization. Traditional power system state estimation methods mainly rely on data from a single data source (such as SCADA system) for analysis, but with the continuous expansion of power system scale and the increasing complexity of the operating environment, this method has exposed many defects.

[0003] First, although the SCADA system can provide basic measurement data of the power system, including voltage amplitude, voltage phase angle, current amplitude, current phase angle, active power, reactive power, etc., due to its relatively low data acquisition frequency and certain measurement errors, it is difficult to provide sufficiently accurate and timely information when facing the dynamic changes and complex working conditions of the power system, resulting in limited accuracy of state estimation. For example, when a rapid load fluctuation or failure occurs in the power system, the SCADA system may not be able to capture these changes in time, resulting in deviations in the state estimation results, affecting the normal operation decision of the power system.

[0004] Secondly, traditional state estimation models often lack sufficient adaptability and accuracy when processing complex power system data. For example, some methods based on simple mathematical models cannot effectively estimate the system state when faced with nonlinear factors and uncertainties in the power system, and are prone to large errors.

[0005] In addition, the existing power system state estimation methods are not comprehensive and in-depth enough in terms of accuracy evaluation. They usually only use single or limited indicators for evaluation, and are unable to accurately measure the performance of the model from multiple dimensions. It is also difficult to make effective improvements to the shortcomings of the model, making it difficult to substantially improve the quality of power system state estimation. Summary of the invention

[0006] In view of the above deficiencies in the prior art, the purpose of the present invention is to propose a comprehensive evaluation and analysis method for the accuracy of power system state estimation, aiming to improve the accuracy and reliability of power system state estimation and ensure the safe and stable operation of the power system by fusing multi-source data, building a suitable model and precise evaluation and improvement mechanism.

[0007] In order to achieve the above objectives, the present invention adopts the following technical solutions: A method for comprehensive evaluation and analysis of power system state estimation accuracy includes the following steps: S1. Data preprocessing and feature extraction: obtain the measurement data of the power system in real time through the SCADA system, obtain the synchronized phasor data using the synchronized phasor measurement unit (PMU), fuse the SCADA data and the synchronized phasor data obtained using the synchronized phasor measurement unit (PMU), and extract features that have an important impact on the state estimation accuracy from the fused data; S2. Construction and training of state estimation model: select support vector machine (SVM) as the model for power system state estimation, select radial basis function (RBF) as the kernel function of the model, and then set the penalty parameters and kernel function parameters; use the training set to train the SVM model, and optimize the model through grid search method combined with cross validation; S3. Evaluation and analysis of state estimation accuracy, based on the constructed and optimized SVM state estimation model, using multivariate accuracy evaluation indicators, comprehensively measure the difference between the model prediction value and the true value from different dimensions, and quantitatively evaluate the prediction performance of the model.

[0008] Furthermore, the measurement data in S1 includes voltage amplitude, voltage phase angle, current amplitude, current phase angle, active power, reactive power, etc.

[0009] Further, in said S1, further, in said S1, a method for acquiring measurement data of the power system in real time through a SCADA system and acquiring synchronized phasor data using a synchronized phasor measurement unit (PMU) is as follows: voltage amplitude, voltage phase angle, current amplitude, current phase angle, active power and reactive power data are collected from the SCADA system at fixed time intervals; at the same time, the synchronized phasor measurement unit (PMU) collects high-precision voltage phasors and current phasors at a frequency higher than that of the SCADA system.

[0010] Furthermore, in S1, the method for fusing SCADA data and synchronized phasor data obtained by synchronized phasor measurement unit (PMU) is as follows: for each feature, the accuracy of the SCADA system and synchronized phasor measurement unit (PMU) for measuring the feature is first evaluated, and the measurement accuracy is obtained by long-term data statistical analysis to obtain the respective standard deviations, which are recorded as and ; Then calculate the weight and , the fused eigenvalues ​​are expressed as: ; ; ; in, is the characteristic value measured by the SCADA system, It is the characteristic value measured by the synchronized phasor measurement unit (PMU).

[0011] In S1, the characteristics that have an important impact on the accuracy of state estimation include power factor, voltage-current ratio, and the rate of change of active power and reactive power; Among them, the power factor The calculation formula is: ; and is the fused active power and reactive power; the power factor reflects the balance between reactive power and active power in the power system, and has an important influence on the accuracy of state estimation; The voltage-to-current ratio is calculated as: ; is the voltage-to-current ratio; is the voltage amplitude after fusion; is the current amplitude after fusion; The calculation formula for the rate of change of active power and reactive power is: ; ; and are two adjacent sampling moments; and Respectively represent the two adjacent sampling moments and The fused active power value; and Respectively represent the two adjacent sampling moments and The fused reactive power value.

[0012] Further, the S2 includes: S21. Model selection and parameter initialization, determine the support vector machine (SVM) as the state estimation model, and select the radial basis function (RBF) as the kernel function. The kernel function expression is: ; in, and is the input sample, that is, the input feature vector, is the kernel function parameter, Representation vector and The norm of two input sample vectors and The distance between Then, the penalty parameters and kernel function parameters are initialized based on experience or preliminary data analysis, and the penalty parameters are set at In the range, the kernel function parameters Set in Within the range, adjustments will be made later through optimization methods; S22. Training set and test set division: the feature data extracted after fusion is divided into training set and test set according to a specific ratio, specifically: Sample data, the training set samples are: ; in, For The number of training sets extracted from sample data; is the true state value corresponding to the input feature vector; The test set samples are: ; S23. Model training, use the training set to train the SVM model, the goal is to minimize the objective function: ; It is a regularization term used to control the complexity of the model and prevent overfitting; is the weight vector, which determines the direction of the hyperplane, The smaller the value of is, the simpler the model is. is the weight vector The transpose of is the bias term, is the empirical risk term, is a slack variable vector, which is used to allow some samples to violate the constraints to a certain extent. is the first slack variables, is the penalty parameter; The constraints are: ; is the category label, is through the real state value Transformed, when the model learns the input feature vector with the "abnormal" category label When the model learns the input feature vector with the "Normal" category label When it is associated, it is predicted to be in a normal state; Is to input Functions that map to high-dimensional spaces; Constraints Represents the distance from the sample to the hyperplane plus the slack variable in high-dimensional space Must be greater than or equal to 1, which means that the sample is required to be correctly classified or close to correctly classified as much as possible; Constraints The non-negativity of the slack variable is guaranteed, that is, unreasonable negative values ​​cannot appear; During the model training process, the Sequential Minimum Optimization (SMO) algorithm is used to solve the above objective function and constraints to obtain the optimal weight vector , bias term and slack variables , thereby determining the final SVM model; During the tuning process, by adjusting the penalty parameters and kernel function parameters , observe the performance of the model on the training set, and select the optimal parameter combination to improve the performance and generalization ability of the model.

[0013] The method for selecting the optimal parameter combination includes: S231. Using the penalty parameters set in S21 And the kernel function parameters The range determines the parameter search range and step size; S232. Perform cross validation using fold cross validation, dividing the training set equally into For each group Parameter combination, select one of them in turn The remaining one is used as the validation subset; the training set is used to train the SVM model on the training subset, and the objective function is used. and constraints Conduct training and solve the above objective function and constraints through the sequential minimum optimization algorithm (SMO) to obtain the optimal weight vector, bias term and slack variable, thereby determining the final SVM model; The mean square error (MSE) is calculated on the validation subset as the evaluation indicator, and each set of parameter combinations is recorded in The average MSE value in the fold cross validation is compared, the average MSE value of all parameter combinations is compared, and the group with the smallest MSE value is selected as the final model parameter.

[0014] Furthermore, in S22, in the process of dividing the training set and the test set, a stratified sampling method is adopted to ensure that the data distribution of the training set and the test set are similar. If the data is classified according to different operating conditions of the power system, it is ensured that each operating condition has a reasonable proportion in the training set and the test set during the division.

[0015] Furthermore, the multivariate accuracy evaluation indicators in S3 include mean square error (MSE), mean absolute error (MAE), maximum absolute error (MaxAE) and root mean square error (RMSE).

[0016] Furthermore, in S3, the multivariate accuracy evaluation index is used to comprehensively measure the difference between the model prediction value and the true value from different dimensions. The method for quantitatively evaluating the prediction performance of the model includes: S31. Obtain the predicted value and the true value, input the test set data into the trained SVM model, obtain the predicted value of each sample, and at the same time, obtain the true state value corresponding to the input feature vector ; S32. Calculate the multivariate accuracy evaluation index, Mean Squared Error (MSE) calculation: ; Mean Absolute Error (MAE) calculation: ; Maximum absolute error (MaxAE) calculation: ; Root mean square error (RMSE) calculation: ; To obtain the predicted value of each sample according to the SVM model; S33. Analyze the multivariate accuracy evaluation indicators and make a comprehensive comparison of the calculated indicators such as MSE, MAE, MaxAE and RMSE. The smaller the value of the indicator, the better the prediction performance of the model.

[0017] In summary, due to the adoption of the above technical solution, the beneficial technical effects of the invention are: Data fusion improves accuracy: By integrating the data of the SCADA system and the synchronized phasor measurement unit (PMU), the advantages of both are fully utilized. The SCADA system provides basic measurement data, and the synchronized phasor measurement unit (PMU) provides high-precision synchronized phasor data with a higher acquisition frequency. During the fusion process, the weight of each feature is calculated according to the measurement accuracy of both. This method makes the fused data more accurately reflect the actual state of the power system. Feature extraction enhances adaptability: Features that have an important impact on the accuracy of state estimation, such as power factor, voltage-current ratio, and the rate of change of active power and reactive power, are extracted from the fused data. These features can fully reflect the operating state of the power system. For example, the power factor reflects the balance between reactive power and active power, the voltage-current ratio has a specific range under different operating conditions, and the power change rate reflects the dynamic characteristics of the system. Compared with traditional methods that rely only on raw measurement data, these features provide richer information for the model, enabling the model to better adapt to the complex operating environment and operating conditions of the power system, and improve the accuracy and reliability of state estimation.

[0018] Model construction and optimization to improve performance: Support vector machine (SVM) is selected as the state estimation model, and radial basis function (RBF) is used as the kernel function, which can effectively handle nonlinear factors in power system data. By reasonably setting the penalty parameters and kernel function parameters, and using the grid search method combined with cross-validation to tune the model, the complexity of the model and the training error are balanced during the training process to find the optimal parameter combination. This makes the model have better generalization ability and can accurately estimate the state of the power system, overcoming the shortcomings of traditional simple mathematical models in dealing with nonlinearity and uncertainty.

[0019] Multivariate accuracy evaluation achieves comprehensive evaluation: using multivariate accuracy evaluation indicators such as mean square error (MSE), mean absolute error (MAE), maximum absolute error (MaxAE) and root mean square error (RMSE) to comprehensively measure the difference between the model prediction value and the true value from different dimensions. This multi-indicator evaluation method can more accurately and comprehensively understand the performance of the model. Compared with the traditional method of using only a single or limited indicator evaluation, it can more deeply analyze the advantages and disadvantages of the model, thereby providing a strong basis for further improvement of the model and helping to continuously improve the quality of power system state estimation.

[0020] In summary, the technical solution of the present invention effectively solves the problems existing in the prior art, significantly improves the accuracy, reliability and adaptability of power system state estimation, and ensures the safe and stable operation of the power system. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1It is a logic block diagram of a comprehensive evaluation and analysis method for power system state estimation accuracy; Figure 2 It is a logical block diagram of the construction and training of the state estimation model in a comprehensive evaluation and analysis method of power system state estimation accuracy; Figure 3 The logical block diagram is a method for quantitatively evaluating the prediction performance of a model in a comprehensive evaluation and analysis method for power system state estimation accuracy. DETAILED DESCRIPTION

[0022] In order to make the purpose, technical solution and advantages of the invention more clear, the invention is further described in detail below in combination with the embodiments. It should be understood that the specific embodiments described here are only used to explain the invention and are not used to limit the invention.

[0023] like Figure 1 A method for comprehensive evaluation and analysis of power system state estimation accuracy is shown, comprising the following steps: S1. Data preprocessing and feature extraction: obtain the measurement data of the power system in real time through the SCADA system, obtain the synchronized phasor data using the synchronized phasor measurement unit (PMU), fuse the SCADA data and the synchronized phasor data obtained using the synchronized phasor measurement unit (PMU), and extract features that have an important impact on the state estimation accuracy from the fused data; S2. Construction and training of state estimation model: select support vector machine (SVM) as the model for power system state estimation, select radial basis function (RBF) as the kernel function of the model, and then set the penalty parameter C and kernel function parameters; use the training set to train the SVM model, and tune the model through grid search method combined with cross validation; S3. Evaluation and analysis of state estimation accuracy, based on the constructed and optimized SVM state estimation model, using multivariate accuracy evaluation indicators, comprehensively measure the difference between the model prediction value and the true value from different dimensions, and quantitatively evaluate the prediction performance of the model.

[0024] The measured data in S1 include voltage amplitude, voltage phase angle, current amplitude, current phase angle, active power, reactive power, etc.

[0025] In S1, the method of obtaining the measurement data of the power system in real time through the SCADA system and obtaining the synchronized phasor data using the synchronized phasor measurement unit (PMU) is as follows: the voltage amplitude, voltage phase angle, current amplitude, current phase angle, active power and reactive power data are collected from the SCADA system at fixed time intervals (such as 1 minute); at the same time, the synchronized phasor measurement unit (PMU) collects high-precision voltage phasors and current phasors at a higher frequency (such as 30 frames / second).

[0026] In S1, the method of fusing SCADA data and synchronized phasor data obtained by synchronized phasor measurement unit (PMU) is as follows: for each feature (such as current amplitude, active power, reactive power, etc.), firstly, the accuracy of the SCADA system and synchronized phasor measurement unit (PMU) for the feature measurement is evaluated, and the measurement accuracy is obtained by long-term data statistical analysis to obtain the respective standard deviations, which are recorded as and ; Then calculate the weight and , the fused eigenvalues ​​are expressed as: ; ; ; in, is the characteristic value measured by the SCADA system, It is the characteristic value measured by the synchronized phasor measurement unit (PMU).

[0027] In S1, the characteristics that have an important impact on the accuracy of state estimation include power factor, voltage-current ratio, and the rate of change of active power and reactive power; Among them, the power factor The calculation formula is: ; and is the fused active power and reactive power; the power factor reflects the balance between reactive power and active power in the power system, and has an important influence on the accuracy of state estimation; The voltage-to-current ratio is calculated as: ; is the voltage-to-current ratio; is the voltage amplitude after fusion; is the current amplitude after fusion; The calculation formula for the rate of change of active power and reactive power is: ; ; and are two adjacent sampling moments; and Respectively represent the two adjacent sampling moments and The fused active power value; and Respectively represent the two adjacent sampling moments and The fused reactive power value.

[0028] The S2 includes: S21. Model selection and parameter initialization, determine the support vector machine (SVM) as the state estimation model, and select the radial basis function (RBF) as the kernel function. The kernel function expression is: ; in, and is the input sample, that is, the input feature vector, is the kernel function parameter, Representation vector and The norm of two input sample vectors and The distance between Then, the penalty parameters and kernel function parameters are initialized based on experience or preliminary data analysis. The penalty parameters can be set at In the range, the kernel function parameters Set in Within the range, adjustments will be made later through optimization methods; S22. Training set and test set division: the feature data extracted after fusion is divided into training set and test set according to a specific ratio (70:30), specifically: there are a total of Sample data, the training set samples are: ; in, For The number of training sets extracted from sample data; is the true state value corresponding to the input feature vector; The test set samples are: ; In the process of division, the stratified sampling method is used to ensure that the data distribution of the training set and the test set is similar; if the data is classified according to different operating conditions of the power system (such as peak, valley, normal load, etc.), it is ensured that each operating condition has a reasonable proportion in the training set and the test set during division; S23. Model training and tuning, use the training set to train the SVM model, the goal is to minimize the objective function: ; It is a regularization term used to control the complexity of the model and prevent overfitting; is the weight vector, which determines the direction of the hyperplane, The smaller the value of is, the simpler the model is. is the weight vector The transpose of is the bias term; is the empirical risk term, is a slack variable vector, used to allow some samples to violate the constraints to a certain extent (that is, to allow some samples to be misclassified or fall into the interval), is the first slack variables, is a penalty parameter that controls the degree of penalty for these samples that violate the constraint; the objective function aims to find a balance between minimizing the model complexity and minimizing the training error, so that the model can both fit the training data well and have good generalization ability; The constraints are: ; is the category label, is through the real state value The conversion process needs to reasonably determine the standards and boundaries of classification based on the actual operation requirements and experience of the power system, and the category labels Usually + 1 or - 1; when the model learns the input feature vector with the "abnormal" category label When the model learns the input feature vector with the "Normal" category label When it is associated, it is predicted to be in a normal state; Is to input Functions mapped to high-dimensional space, Constraints Represents the distance from the sample to the hyperplane (after scaling) plus the slack variable in high-dimensional space Must be greater than or equal to 1, which means that the sample is required to be correctly classified or close to correctly classified as much as possible; Constraints The non-negativity of the slack variable is guaranteed, that is, unreasonable negative values ​​cannot appear; During the model training process, the Sequential Minimum Optimization (SMO) algorithm is used to solve the above objective function and constraints to obtain the optimal weight vector , bias term and slack variables , thereby determining the final SVM model; During the tuning process, by adjusting the penalty parameters and kernel function parameters , observe the performance of the model on the training set (such as accuracy, mean square error, etc.), and select the optimal parameter combination to improve the performance and generalization ability of the model.

[0029] The method for selecting the optimal parameter combination includes: S231. Using the penalty parameters set in S21 And the kernel function parameters The range determines the parameter search range and step size; S232. Perform cross validation using fold cross validation, dividing the training set equally into For each group Parameter combination, select one of them in turn The remaining one is used as the validation subset; the training set is used to train the SVM model on the training subset, and the objective function is used. and constraints Conduct training and solve the above objective function and constraints through the sequential minimum optimization algorithm (SMO) to obtain the optimal weight vector, bias term and slack variable, thereby determining the final SVM model; The mean square error (MSE) is calculated on the validation subset as the evaluation indicator, and each set of parameter combinations is recorded in The average MSE value in the fold cross validation is compared, the average MSE value of all parameter combinations is compared, and the group with the smallest MSE value is selected as the final model parameter.

[0030] The multivariate accuracy evaluation indicators in S3 include mean square error (MSE), mean absolute error (MAE), maximum absolute error (MaxAE) and root mean square error (RMSE).

[0031] In S3, multivariate accuracy evaluation indicators are used to comprehensively measure the difference between the model prediction value and the true value from different dimensions. The method for quantitatively evaluating the prediction performance of the model includes: S31. Obtain the predicted value and the true value, input the test set data into the trained SVM model, obtain the predicted value of each sample, and at the same time, obtain the true state value corresponding to the input feature vector ; S32. Calculate the multivariate accuracy evaluation index, Mean Squared Error (MSE) calculation: ; Mean Absolute Error (MAE) calculation: ; Maximum absolute error (MaxAE) calculation: ; Root mean square error (RMSE) calculation: ; To obtain the predicted value of each sample according to the (SVM) model; S33. Analyze the multivariate accuracy evaluation indicators and make a comprehensive comparison of the calculated indicators such as MSE, MAE, MaxAE and RMSE. The smaller the value of the indicator, the better the prediction performance of the model.

[0032] The above description is a preferred embodiment of the invention and is not intended to limit the invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the invention should be included in the protection scope of the invention.

Claims

1. A comprehensive evaluation and analysis method for power system state estimation accuracy, characterized in that: The following steps are involved: S1. Data preprocessing and feature extraction: obtaining the measurement data of the power system in real time through the SCADA system, obtaining the synchronized phasor data using the synchronized phasor measurement unit PMU, fusing the SCADA data and the synchronized phasor data obtained using the synchronized phasor measurement unit PMU, and extracting features that have an important impact on the state estimation accuracy from the fused data; S2. Construction and training of state estimation model: select support vector machine (SVM) as the model for power system state estimation, select radial basis function (RBF) as the kernel function of the model, and then set penalty parameters and kernel function parameters; use the training set to train the SVM model, and optimize the model through grid search method combined with cross validation; S3. Evaluation and analysis of state estimation accuracy, based on the constructed and optimized SVM state estimation model, using multivariate accuracy evaluation indicators, comprehensively measure the difference between the model prediction value and the true value from different dimensions, and quantitatively evaluate the prediction performance of the model.

2. A method for comprehensive evaluation and analysis of power system state estimation accuracy according to claim 1, characterized in that: The measured data in S1 include voltage amplitude, voltage phase angle, current amplitude, current phase angle, active power, and reactive power.

3. A method for comprehensive evaluation and analysis of power system state estimation accuracy according to claim 1, characterized in that: In S1, the method of obtaining the measurement data of the power system in real time through the SCADA system and obtaining the synchronized phasor data using the synchronized phasor measurement unit PMU is as follows: the voltage amplitude, voltage phase angle, current amplitude, current phase angle, active power and reactive power data are collected from the SCADA system at fixed time intervals; at the same time, the synchronized phasor measurement unit PMU collects high-precision voltage phasors and current phasors at a frequency higher than that of the SCADA system.

4. A method for comprehensive evaluation and analysis of power system state estimation accuracy according to claim 1, characterized in that: In S1, the method for fusing SCADA data and synchronized phasor data obtained by synchronized phasor measurement unit PMU is as follows: for each feature, the accuracy of the SCADA system and synchronized phasor measurement unit PMU for measuring the feature is first evaluated, and the measurement accuracy is obtained by long-term data statistical analysis to obtain the respective standard deviations, which are respectively recorded as and ; Then calculate the weight and , the fused eigenvalues ​​are expressed as: ； ； ； in, is the characteristic value measured by the SCADA system, It is the characteristic value measured by the synchronized phasor measurement unit PMU.

5. A method for comprehensive evaluation and analysis of power system state estimation accuracy according to claim 1, characterized in that: In S1, the characteristics that have an important impact on the accuracy of state estimation include power factor, voltage-current ratio, and the rate of change of active power and reactive power; Among them, the power factor The calculation formula is: ； and is the fused active power and reactive power; the power factor reflects the balance between reactive power and active power in the power system, and has an important influence on the accuracy of state estimation; The voltage-to-current ratio is calculated as: ； is the voltage-to-current ratio; is the voltage amplitude after fusion; is the current amplitude after fusion; The calculation formula for the rate of change of active power and reactive power is: ； ； and are two adjacent sampling moments; and Respectively represent the two adjacent sampling moments and The fused active power value; and Respectively represent the two adjacent sampling moments and The fused reactive power value.

6. A method for comprehensive evaluation and analysis of power system state estimation accuracy according to claim 1, characterized in that: The S2 includes: S21. Model selection and parameter initialization, determine the support vector machine SVM as the state estimation model, and select the radial basis function RBF as the kernel function. The kernel function expression is: ； in, and is the input sample, that is, the input feature vector, is the kernel function parameter, Representation vector and The norm of two input sample vectors and The distance between Then, the penalty parameters and kernel function parameters are initialized based on experience or preliminary data analysis, and the penalty parameters are set at In the range, the kernel function parameters Set in Within the range, adjustments will be made later through optimization methods; S22. Training set and test set division: the feature data extracted after fusion is divided into training set and test set according to a specific ratio, specifically: Sample data, the training set samples are: ； in, For The number of training sets extracted from the sample data; is the true state value corresponding to the input feature vector; The test set samples are: ； S23. Model training and tuning, use the training set to train the SVM model, the goal is to minimize the objective function: ； It is a regularization term used to control the complexity of the model and prevent overfitting; is the weight vector, which determines the direction of the hyperplane, The smaller the value of is, the simpler the model is. is the weight vector The transpose of is the bias term; is the empirical risk term, is a slack variable vector, which is used to allow some samples to violate the constraints to a certain extent. is the first slack variables, is the penalty parameter, which controls the degree of penalty for these constraint violation samples; The constraints are: ； is the category label, is through the real state value The conversion process needs to be based on the actual operation requirements and experience of the power system. When the model learns the input feature vector with the "abnormal" category label When the model learns the input feature vector with the "Normal" category label When it is associated, it is predicted to be in a normal state; Is to input Functions that map to high-dimensional spaces; Constraints Represents the distance from the sample to the hyperplane plus the slack variable in high-dimensional space Must be greater than or equal to 1, which means that the sample is required to be correctly classified or close to correctly classified as much as possible; Constraints The non-negativity of the slack variable is guaranteed, that is, unreasonable negative values ​​cannot appear; During the model training process, the sequential minimum optimization algorithm SMO is used to solve the above objective function and constraints to obtain the optimal weight vector , bias term and slack variables , thereby determining the final SVM model; During the tuning process, by adjusting the penalty parameters and kernel function parameters , observe the performance of the model on the training set, and select the optimal parameter combination to improve the performance and generalization ability of the model.

7. A method for comprehensive evaluation and analysis of power system state estimation accuracy according to claim 6, characterized in that: The method for selecting the optimal parameter combination includes: S241. Using the penalty parameters set in S21 And the kernel function parameters The range determines the parameter search range and step size; S242. Perform cross validation using Fold cross validation, divide the training set equally into For each group Parameter combination, select one of them in turn The remaining one is used as the validation subset; the training set is used to train the SVM model on the training subset, and the objective function is used. and constraints Conduct training, solve the above objective function and constraints through the sequential minimum optimization algorithm SMO, obtain the optimal weight vector, bias term and slack variable, and thus determine the final SVM model; Calculate the mean square error MSE as the evaluation indicator on the validation subset and record each set of parameter combinations in The average MSE value in the fold cross validation is compared, the average MSE value of all parameter combinations is compared, and the group with the smallest MSE value is selected as the final model parameter.

8. A method for comprehensive evaluation and analysis of power system state estimation accuracy according to claim 6, characterized in that: In S22, in the process of dividing the training set and the test set, a stratified sampling method is adopted to ensure that the data distribution of the training set and the test set are similar.

9. A method for comprehensive evaluation and analysis of power system state estimation accuracy according to claim 1, characterized in that: The multivariate accuracy evaluation indicators in S3 include mean square error (MSE), mean absolute error (MAE), maximum absolute error (MaxAE) and root mean square error (RMSE).

10. A method for comprehensive evaluation and analysis of power system state estimation accuracy according to claim 1, characterized in that: In S3, multivariate accuracy evaluation indicators are used to comprehensively measure the difference between the model prediction value and the true value from different dimensions. The method for quantitatively evaluating the prediction performance of the model includes: S31. Obtain the predicted value and the true value, input the test set data into the trained SVM model, obtain the predicted value of each sample, and at the same time, obtain the true state value corresponding to the input feature vector ; S32. Calculate the multivariate accuracy evaluation index, Mean square error MSE calculation: ； Mean absolute error MAE calculation: ； Maximum absolute error MaxAE calculation: ； Root mean square error RMSE calculation: ； To obtain the predicted value of each sample according to the SVM model; S33. Analyze the multivariate accuracy evaluation indicators and make a comprehensive comparison of the calculated MSE, MAE, MaxAE and RMSE indicators. The smaller the value of the indicator, the better the prediction performance of the model.

Images