A multi-working-condition adaptive current transformer online monitoring method and system

Abstract

The application provides a multi-working-condition adaptive current transformer online monitoring method and system, which comprises the following steps: obtaining measurement data of a three-phase current transformer, and respectively constructing corresponding modeling data sets and monitoring data sets; performing sequence current decomposition on three-phase fundamental wave currents in the modeling data sets, calculating the relative mean deviation of the positive sequence current percentage and the zero sequence percentage of samples, and constructing a sample input variable matrix; constructing an infinite Gaussian mixture model, and optimizing the model through a multi-objective optimization algorithm; dividing the samples into corresponding numbers of Gaussian components through the infinite Gaussian mixture model; calculating the local probability and posterior probability of each sample in the monitoring data set belonging to each Gaussian component, and calculating a global BIP index; comparing the global BIP index of the sample with an abnormal threshold under a significance level, and judging whether the monitoring data set has a measurement error abnormality through a weighted majority voting mechanism. Through the scheme, current transformer error online monitoring can be realized, and the adaptability and monitoring precision are high.

Description

Technical Field

[0001] This invention belongs to the field of power data monitoring technology, and particularly relates to a multi-condition adaptive online monitoring method and system for current transformers. Background Technology

[0002] Current transformers (CTs) are measuring devices that transform large currents in high-voltage power grids into smaller current signals on the secondary side at a fixed ratio. They play a crucial role in power systems and require timely calibration to address any potential errors or anomalies. Traditional "outage calibration" and "live-line calibration" methods require periodic testing and lack real-time capability, posing a risk of transformers becoming inaccurate during operation without timely identification.

[0003] Currently, with the widespread attention given to online monitoring technology for current transformer measurement errors, some solutions propose to extract abnormal change features of measurement errors from the secondary output signal of a single CT using signal processing, artificial intelligence, and other methods. However, these methods face difficulties in feature extraction in practical engineering applications and suffer from poor applicability.

[0004] Another approach is to construct multiple instrument transformers with electrical physical connections into a single monitoring group, using the physical connections within the group as constraints to achieve accurate assessment of measurement error states. For larger instrument transformer groups, voltage and current constraint equations can be constructed based on Kirchhoff's voltage and current laws. However, considering the dependence on line parameters and the requirements for equipment synchronization, the practical application requirements are high, and this type of method has poor applicability. For smaller instrument transformer groups, the stability of short-time three-phase voltage imbalance has been studied, and constraint equations involving a set of three-phase voltage transformers have been established to achieve abnormal monitoring of measurement errors. However, since the amplitude and fluctuation of current imbalance in power systems are much higher than those of voltage imbalance, the accuracy of three-phase current transformer error monitoring is poor.

[0005] Meanwhile, with the periodic changes in power load, the amplitude of the primary current in the power grid also fluctuates drastically. According to the measurement principle of current transformers, changes in the amplitude of the primary current will have a certain impact on the measurement error. When the current is small, the measurement error increases significantly, which will further affect the error monitoring accuracy of the current transformer. Summary of the Invention

[0006] In view of this, embodiments of the present invention provide a multi-condition adaptive online monitoring method and system for current transformers, which is used to solve the problems of poor applicability and poor monitoring accuracy of current current transformer error monitoring schemes.

[0007] In a first aspect of the present invention, a multi-condition adaptive online monitoring method for current transformers is provided, comprising:

[0008] Acquire the measurement data of the fundamental current amplitude and phase in the three-phase current transformer, and construct the corresponding modeling dataset and monitoring dataset based on the normal error data and real-time monitoring data in the measurement data, respectively.

[0009] Sequence current decomposition is performed on the three-phase fundamental current in the modeling dataset. The relative mean deviation of the positive sequence current percentage and the zero sequence current percentage for each sample in the modeling dataset is calculated, and a sample input variable matrix is ​​constructed.

[0010] An infinite Gaussian mixture model is constructed based on the sample input variable matrix, and the infinite Gaussian mixture model is optimized by a multi-objective optimization algorithm. Based on the optimized infinite Gaussian mixture model, the samples in the modeling dataset are adaptively divided into a corresponding number of Gaussian components.

[0011] Based on the optimized infinite Gaussian mixture model, the local probability and posterior probability of each sample in the monitoring dataset belonging to each Gaussian component are calculated, and the global BIP index is calculated based on the local probability and posterior probability of each sample.

[0012] The global BIP index for each sample is compared with the anomaly threshold at the significance level. Based on the comparison results, a weighted majority voting mechanism is used to determine whether there are measurement error anomalies in the monitoring dataset.

[0013] In a second aspect of the present invention, a multi-condition adaptive online monitoring system for current transformers is provided, comprising:

[0014] The data acquisition module is used to acquire measurement data of the fundamental current amplitude and phase in the three-phase current transformer, and to construct corresponding modeling datasets and monitoring datasets based on the normal error data and real-time monitoring data in the measurement data.

[0015] The feature extraction module is used to perform sequence current decomposition on the three-phase fundamental current in the modeling dataset, calculate the relative mean deviation of the positive sequence current percentage and the zero sequence current percentage for each sample in the modeling dataset, and construct the sample input variable matrix.

[0016] The model building module is used to construct an infinite Gaussian mixture model based on the sample input variable matrix, and optimize the infinite Gaussian mixture model through a multi-objective optimization algorithm. Based on the optimized infinite Gaussian mixture model, the samples in the modeling dataset are adaptively divided into a corresponding number of Gaussian components.

[0017] The index calculation module is used to calculate the local probability and posterior probability of each sample in the monitoring dataset belonging to each Gaussian component based on the optimized infinite Gaussian mixture model, and to calculate the global BIP index based on the local probability and posterior probability of each sample.

[0018] The anomaly detection module compares the global BIP index of each sample with the anomaly threshold at the significance level, and determines whether there are measurement error anomalies in the monitoring dataset based on the comparison results through a weighted majority voting mechanism.

[0019] In a third aspect of the present invention, an electronic device is provided, including a memory, a processor, and a computer program stored in the memory and executable by the processor, wherein the processor executes the computer program to implement the steps of the method as described in the first aspect of the present invention.

[0020] In a fourth aspect of the present invention, a computer-readable storage medium is provided, the computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the method provided in the first aspect of the present invention.

[0021] In this embodiment of the invention, by utilizing the stability of the zero-sequence current in three-phase current, a characteristic quantity related to measurement error is established as the model input. An infinite Gaussian mixture model adaptively divides the measurement data into different modes, and a unified BIP index is provided to quantify the degree of anomaly in the monitored samples. A weighted majority voting mechanism is then used to determine the measurement error state of the current transformer. This method, through the adaptive division of the current transformer's operating modes, enables online error monitoring of the current transformer in various complex scenarios, with high monitoring accuracy and stable, reliable results. Furthermore, it eliminates the dependence on physical standards, does not require power outages, and has good applicability and versatility. Attached Figure Description

[0022] To more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0023] Figure 1 A flowchart illustrating a multi-condition adaptive online monitoring method for current transformers provided in one embodiment of the present invention;

[0024] Figure 2 This is a schematic diagram of the temporal distribution of each element in the input variable matrix of the modeling dataset provided in one embodiment of the present invention;

[0025] Figure 3A schematic diagram of a multi-condition adaptive online monitoring system for current transformers provided in one embodiment of the present invention;

[0026] Figure 4 This is a schematic diagram of the structure of an electronic device provided in one embodiment of the present invention. Detailed Implementation

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

[0028] It should be understood that the terms "comprising" and other similar expressions in the specification, claims, and accompanying drawings of this invention are intended to cover a non-exclusive inclusion, such as a process, method, system, or apparatus that includes a series of steps or units and is not limited to the listed steps or units. Furthermore, "first" and "second" are used to distinguish different objects and are not intended to describe a specific order.

[0029] Please see Figure 1 The present invention provides a flowchart illustrating a multi-condition adaptive online monitoring method for current transformers, comprising:

[0030] S101. Obtain the measurement data of the fundamental current amplitude and phase in the three-phase current transformer, and construct the corresponding modeling dataset and monitoring dataset based on the normal error data and real-time monitoring data in the measurement data.

[0031] Three-phase current transformers contain fundamental current. By acquiring the amplitude and phase data of the three-phase fundamental current, it is convenient to perform sequence current decomposition on the three-phase current. The fundamental current is the sinusoidal component of the three-phase current that is synchronized with the power frequency. Its three phase amplitudes are equal and its phases differ by 120°, and it can be measured by the current transformer.

[0032] Among them, after the power outage verification, the fundamental current measurement data of the three-phase current transformers within the error limit range are collected as the modeling dataset; and the real-time monitoring data is used as the monitoring dataset.

[0033] The modeling dataset can be used to extract sequence current features and then used as input to construct an infinite Gaussian mixture model. Real-time monitoring data is used as real-time measurement data to determine whether there are any abnormalities in its error.

[0034] S102. Perform sequence current decomposition on the three-phase fundamental current in the modeling dataset, calculate the relative mean deviation of the positive sequence current percentage and the zero sequence current percentage for each sample in the modeling dataset, and construct the sample input variable matrix.

[0035] Sequence current decomposition involves decomposing the three-phase fundamental current into positive-sequence, negative-sequence, and zero-sequence currents, yielding the corresponding positive-sequence, negative-sequence, and zero-sequence currents. The positive-sequence current percentage refers to the percentage of positive-sequence current relative to the rated current, and the deviation of the zero-sequence current percentage from the mean refers to the deviation of the zero-sequence current percentage from its mean. Based on the positive-sequence current percentage and the deviation of the zero-sequence current percentage from the mean for each sample, a sample input variable matrix for the modeling dataset can be constructed.

[0036] Optionally, based on the measurement data of the amplitude and phase of the three-phase fundamental current, the three-phase fundamental current can be decomposed into positive-sequence, negative-sequence, and zero-sequence components using the symmetrical component method:

[0037] (1)

[0038] In the formula, This represents the three-phase fundamental current vector. For complex operators, this indicates rotating the current phasor in the positive direction by 120°, and , Represents variables, Represents the positive-order components. Indicates the negative order component. It is a zero-order component;

[0039] Calculate the percentage of positive sequence current relative to rated current using formula (2). :

[0040] (2)

[0041] In the formula, This represents the positive order component of the t-th sample. Indicates the rated current;

[0042] Calculate the sample according to formula (3) The proportion of zero-sequence current to positive-sequence current :

[0043] (3)

[0044] In the formula, Let represent the zero-order component of the t-th sample;

[0045] Calculate the sample according to formula (4) Deviation of the proportion of zero-sequence current relative to the average value :

[0046] (4)

[0047] In the formula, To model the proportion of zero-sequence current in all samples of the dataset The mean, This represents the ratio of the zero-sequence current to the positive-sequence current in the t-th sample.

[0048] Based on the deviation of the positive-sequence current percentage and zero-sequence current percentage from the mean for each sample, a sample input variable matrix X is constructed:

[0049] (5)

[0050] In the formula, This represents the last sample in the modeling dataset. This represents the percentage of the positive-sequence current relative to the rated current of the T-th sample in the modeling dataset. This represents the magnitude of the deviation of the zero-order proportion of the T-th sample from the mean.

[0051] S103. Construct an infinite Gaussian mixture model based on the sample input variable matrix, and optimize the infinite Gaussian mixture model through a multi-objective optimization algorithm. Based on the optimized infinite Gaussian mixture model, adaptively divide the samples in the modeling dataset into a corresponding number of Gaussian components.

[0052] The infinite Gaussian mixture model is a linear combination of an unlimited number of independent Gaussian distributions, capable of representing the distribution of any data. The multi-objective optimization algorithm is an algorithm that simultaneously solves multiple objective optimization problems, which may include genetic algorithms, particle swarm optimization, simulated annealing, etc., and is used to optimize the parameters of the infinite Gaussian mixture model.

[0053] A Gaussian component represents a subset of Gaussian sub-distributions within the overall distribution of an infinite Gaussian mixture model. An infinite Gaussian mixture model can represent the data distribution of a variable matrix X as a weighted sum of K Gaussian components:

[0054]

[0055] In the formula, These represent the k-th Gaussian components. The mixing coefficients, mean, and covariance matrix of , where K is the number of Gaussian components.

[0056] In some embodiments, constraints are set for the infinite Gaussian mixture model, and these constraints are expressed as follows:

[0057]

[0058] In the formula, m represents a preset positive value. , They represent the variables respectively. Direction first , The mean parameter of a Gaussian component, , All are count variables;

[0059] By using the logarithmic barrier method to treat the constraints as penalty terms, the objective function of the infinite Gaussian mixture model is constructed as follows:

[0060] ;

[0061] In the formula, The KL divergence between the model distribution and the real data distribution. The weight parameters for the penalty term, express The minimum value of K is the number of Gaussian components in the model.

[0062] Optionally, an infinite Gaussian mixture model can be constructed based on the Dirichlet process;

[0063] The sample input variable matrix is ​​used as the input to the infinite Gaussian mixture model, and the modeling process is transformed into an optimization problem. The model is then optimized using the multi-objective genetic algorithm NSGA-II.

[0064] Based on the optimized infinite Gaussian mixture model, the samples in the modeling dataset are adaptively divided into multiple Gaussian components to characterize the primary current amplitude and measurement error conditions under different modes.

[0065] For a total of T samples in the modeling dataset, the T-dimensional decision variable is the component label corresponding to each sample. .

[0066] The optimization objectives are to minimize the KL divergence and the number of Gaussian components. The objective function is:

[0067] ;

[0068] In the formula, Let KL divergence be a metric, and , The model distribution to be estimated, This is the maximum likelihood estimate of the true data distribution of the input matrix X.

[0069] By setting constraints on the objective function, the mean parameters corresponding to any two Gaussian components in the model are expressed in terms of variables. The difference in direction needs to be greater than a small positive number m to ensure the effectiveness of classification; then, the constraints are constructed as penalty terms using the logarithmic barrier method, and the objective function is modified to:

[0070] .

[0071] The input variable matrix X is used as the model input for solving the optimization problem. Considering the requirements for computational speed and convergence, the multi-objective optimization algorithm NSGA-II is selected. X is updated using the results of each iteration. The component labels corresponding to the samples in a given iteration are... ,sample In The calculation is performed according to the following formula, where Indicates the centrality of the dataset and Mean percentage of zero-sequence current for samples with the same label:

[0072]

[0073] After optimization, samples with the same component label are considered to operate in the same modality, corresponding to one Gaussian component, for a total of K. For the k-th Gaussian component... Assuming a total of For each sample, calculate its mixing coefficient: ;

[0074] Gaussian components mean Covariance Matrix The parameters are calculated according to the following formulas:

[0075]

[0076]

[0077] In the formula, , They are respectively The corresponding variables of the samples contained in and variance , Let be the covariance between the two variables.

[0078] Based on the above parameters, an infinite Gaussian mixture model is established. All samples in the modeling dataset are adaptively divided into K modes, corresponding to different primary current amplitudes and measurement error conditions.

[0079] S104. Based on the optimized infinite Gaussian mixture model, calculate the local probability and posterior probability of each sample in the monitoring dataset belonging to each Gaussian component, and calculate the global BIP index based on the local probability and posterior probability of each sample.

[0080] For the monitoring dataset, the probability of each sample belonging to each Gaussian component is determined using an infinite Gaussian mixture model, including calculating its local probability and posterior probability. The local probability can be determined based on the Mahalanobis distance from the sample to the Gaussian component, reflecting the degree of deviation of the sample relative to the Gaussian component. The posterior probability can be calculated based on the mixing coefficients of the Gaussian component, representing the probability correction of the hypothesis after knowing relevant evidence.

[0081] Optionally, the weight matrix can be set based on the pre-whitening method;

[0082] The covariance matrix is ​​preprocessed based on the weight matrix. The Mahalanobis distance of the samples in the monitoring dataset to the Gaussian components is calculated based on the formula. Then, the chi-square distribution critical value table is consulted based on the Mahalanobis distance to obtain the local probability of a sample belonging to each Gaussian component.

[0083]

[0084] In the formula, Represents Mahalanobis distance, Represents the weight matrix;

[0085] Calculate the posterior probability of a sample belonging to each Gaussian component using the formula:

[0086]

[0087] In the formula, Indicates sample Belongs to components The posterior probability, Represents the k-th Gaussian component The mixing coefficient, Represents the multivariate Gaussian distribution function. This represents the t-th sample in the monitoring dataset. Indicates Gaussian component The mean, Indicates Gaussian component The covariance matrix, where K represents the total number of Gaussian components. This represents the mixing coefficient of the i-th Gaussian component. Let represent the mean of the i-th Gaussian component. Let represent the covariance matrix of the i-th Gaussian component.

[0088] For the k-th Gaussian component Calculate the samples respectively Corresponding variables That is, the percentage of positive sequence current relative to rated current and the deviation of the percentage of zero sequence current relative to the average value. Represents component labels in the modeling dataset Zero-sequence current percentage of all samples The mean.

[0089] Based on expert experience, the monitoring data contains samples. In variables Changes in direction are more likely to reflect anomalies in measurement error; therefore, the weight matrix is ​​set based on the pre-whitening method. :

[0090]

[0091] In the formula, Choose according to the actual application.

[0092] Calculate the sample according to the following formula For Gaussian components Mahalanobis distance:

[0093]

[0094] Sample calculation based on Mahalanobis distance For the local probability of each Gaussian component, this metric reflects the sample's relative probability to the Gaussian component. The degree of deviation. Due to the Mahalanobis distance It follows a chi-square distribution with 2 degrees of freedom, and its local probability is... It can be obtained by looking up the table of critical values ​​for the chi-square distribution.

[0095] Then, calculate the sample according to the formula. posterior probability .

[0096] Furthermore, based on the local and posterior probabilities of each sample in the monitoring dataset, the global BIP index corresponding to each sample is calculated according to the formula:

[0097] ;

[0098] In the formula, This represents the BIP metric of the t-th sample in the monitoring dataset. Indicates sample The local probability, Indicates sample The posterior probability, For counting variables, This represents the total number of components.

[0099] Integrated Samples Local probabilities involving all Gaussian components With posterior probability The corresponding global BIP index is calculated according to the formula, and the above calculation is repeated for all samples in the monitoring dataset.

[0100] S105. Compare the global BIP index of each sample with the anomaly threshold at the significance level, and determine whether there is measurement error anomaly in the monitoring dataset based on the comparison results through a weighted majority voting mechanism.

[0101] The significance level is the probability of making an error when estimating a population parameter to fall within a certain interval. It is used to determine whether the sample data is sufficient to reject the null hypothesis. It is typically expressed as... α Indicates significance level. α Indicates when the null hypothesis H When 0 is true, an error is rejected. H The maximum allowed probability of 0, for example, in At a significance level of , if H If 0 is true, there is a 5% risk of incorrectly rejecting it, meaning the threshold is set to 0.95.

[0102] The weighted majority voting mechanism is a voting mechanism that sets the voting weight of each voter based on their importance, influence, etc., which can ensure the fairness and representativeness of the decision-making process.

[0103] Among them, the anomaly threshold corresponding to the significance level is determined based on the Monte Carlo test.

[0104] This embodiment determines the anomaly monitoring threshold based on the Monte Carlo test. For example, at a significance level of α=0.05, the threshold is selected as 0.95.

[0105] Different weights are assigned to the comparison results between the global BIP index and the anomaly threshold. An anomaly score of the monitoring dataset is calculated through a weighted majority voting mechanism to determine whether there are measurement error anomalies in the monitoring dataset.

[0106] Optionally, compare the global BIP index for each sample with the outlier threshold at the significance level, and express the comparison results as follows:

[0107] ;

[0108] The comparison results of all samples are expressed as follows: ;

[0109] In the formula, This represents the comparison result of the t-th sample in the monitoring dataset. This represents the BIP index of the t-th sample. Indicates the significance level, and D represents the comparison result dataset. This indicates the comparison result of the Nth sample;

[0110] Based on the magnitude of the primary current, different weights are assigned to the comparison results according to the formula:

[0111] ;

[0112] The final anomaly score for the monitoring dataset is calculated based on a weighted majority voting mechanism.

[0113]

[0114] In the formula, This represents the weight of the t-th sample in the monitoring dataset. The current represents the percentage of the positive-sequence current of the t-th sample relative to the rated current, and N represents the number of samples in the monitoring dataset. This indicates the anomaly score in the monitoring dataset. Indicates the results of sample comparison;

[0115] When anomaly scores exceed a preset threshold, it is determined that there is an anomaly in the monitoring dataset. For example, when At this time, it is determined that there is an anomaly in the measurement error of the monitoring dataset.

[0116] In this embodiment, the stability of the zero-sequence current in the three-phase current is used to accurately identify the error of the current transformer. An infinite Gaussian mixture model is used to adaptively divide the data into different modes, enabling it to adapt to complex operating conditions with drastic changes in primary current. A unified BIP index is calculated to quantify the degree of anomaly in the monitored samples, and a weighted majority voting mechanism is used to determine the measurement error status. Therefore, not only is online monitoring of the current transformer error achieved, but the accuracy of error monitoring is also guaranteed, improving adaptability and versatility in different environments. Furthermore, it eliminates the dependence on physical standards. Simulation experiments have verified that it can reliably identify measurement error changes in a 0.2-level CT.

[0117] In some embodiments, a set of three-phase current transformers on the outgoing line side of a 220kV power plant is selected as an application example. Their accuracy class is 0.2S, and their transformation ratio is 1600A / 1A. The secondary side output signal of the three-phase current transformers is acquired using a 0.05-level high-precision multi-channel synchronous signal acquisition system. The online error monitoring method for the current transformers based on zero-sequence current constraints is as follows:

[0118] The fundamental amplitude and phase measurement data of the three-phase current transformers were collected at 1-minute intervals. Before data acquisition, a power outage calibration was performed, and the measurement errors of the three-phase current transformers were all within the error limits. 5000 sets of current transformer measurement data were collected as the modeling dataset. After a period of operation, the measurement error of phase A of the three-phase current transformers was manually altered to exceed the error limit, and another 5000 sets of real-time measurement data were collected as the monitoring dataset.

[0119] Based on the symmetrical component method, the three-phase current data in the modeling dataset are decomposed into sequence currents. The percentage of positive sequence current P and the deviation of the zero sequence current from the mean for each sample in the modeling dataset are calculated. δ Establish the input variable matrix X, and the two-dimensional time series distribution of the variables is as follows: Figure 2 As shown.

[0120] An infinite Gaussian mixture model was established using the input variable matrix X, transforming the modeling process into an optimization problem, which was then solved using the multi-objective optimization algorithm NSGA-II. The established infinite Gaussian mixture model contains six Gaussian components, and their specific parameters are shown in Table 1.

[0121]

[0122] Table 1 Parameters of the Infinite Gaussian Mixture Model

[0123] Based on the infinite Gaussian mixture model, the local probability and posterior probability of each sample in the monitoring dataset belonging to each Gaussian component are calculated, and the global BIP index is calculated.

[0124] Taking the first sample in the monitoring dataset as an example, the calculation results of its posterior probability, local probability, and global BIP index are shown in Table 2:

[0125]

[0126] Table 2 Monitoring results of the first sample in the monitoring dataset

[0127] exist At a significance level, a threshold of 0.95 was selected. This threshold was compared with the BIP index corresponding to each sample in the monitoring dataset. The number of samples exceeding the threshold in the monitoring dataset was 4321, resulting in anomaly scores. The value is greater than 0.5. Therefore, it is determined that there is an anomaly in the monitoring data at this time. The monitoring result is consistent with the difference result, and the method has high reliability.

[0128] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0129] Figure 3 This is a schematic diagram of a multi-condition adaptive online monitoring system for current transformers provided in an embodiment of the present invention. The system includes:

[0130] The data acquisition module 310 is used to acquire the measurement data of the fundamental current amplitude and phase in the three-phase current transformer, and to construct the corresponding modeling dataset and monitoring dataset based on the normal error data and real-time monitoring data in the measurement data.

[0131] The data acquisition module 310 includes:

[0132] The dataset construction unit is used to collect fundamental current measurement data within a preset error limit range from three-phase current transformers as a modeling dataset after power outage verification; and to collect real-time monitoring data as a monitoring dataset.

[0133] The feature extraction module 320 is used to perform sequence current decomposition on the three-phase fundamental current in the modeling dataset, calculate the relative mean deviation of the positive sequence current percentage and the zero sequence current percentage for each sample in the modeling dataset, and construct the sample input variable matrix.

[0134] The feature extraction module 320 includes:

[0135] The current decomposition unit is used to decompose the three-phase fundamental current into positive-sequence, negative-sequence, and zero-sequence components based on the measurement data of the amplitude and phase of the three-phase fundamental current using the symmetrical component method.

[0136] (1)

[0137] In the formula, This represents the three-phase fundamental current vector. For complex operators, this indicates rotating the current phasor in the positive direction by 120°, and , Represents the positive-order components. Indicates the negative order component. It is a zero-order component;

[0138] The first calculation unit is used to calculate the percentage of the positive sequence current relative to the rated current according to formula (2). :

[0139] (2)

[0140] In the formula, This represents the positive order component of the t-th sample. Indicates the rated current;

[0141] The second calculation unit is used to calculate the sample according to formula (3). The proportion of zero-sequence current to positive-sequence current :

[0142] (3)

[0143] In the formula, Let represent the zero-order component of the t-th sample;

[0144] And, calculate the sample according to formula (4). Deviation of the proportion of zero-sequence current relative to the average value :

[0145] (4)

[0146] In the formula, To model the proportion of zero-sequence current in all samples of the dataset The mean, This represents the ratio of the zero-sequence current to the positive-sequence current in the t-th sample.

[0147] The input matrix construction unit is used to construct the sample input variable matrix X based on the relative mean deviation of the positive-sequence current percentage and the zero-sequence percentage for each sample:

[0148] (5)

[0149] In the formula, This represents the last sample in the modeling dataset. This represents the percentage of the positive-sequence current relative to the rated current of the T-th sample in the modeling dataset. This represents the deviation of the zero-order proportion of the T-th sample from the mean.

[0150] The model building module 330 is used to build an infinite Gaussian mixture model based on the sample input variable matrix, and optimize the infinite Gaussian mixture model through a multi-objective optimization algorithm. Based on the optimized infinite Gaussian mixture model, the samples in the modeling dataset are adaptively divided into a corresponding number of Gaussian components.

[0151] Optionally, set constraints for the infinite Gaussian mixture model, and express the constraints as follows:

[0152] ,and ;

[0153] In the formula, m represents a preset positive value. , They represent the variables respectively. Direction first , The mean parameter of a Gaussian component, , All are count variables;

[0154] By using the logarithmic barrier method to treat the constraints as penalty terms, the objective function of the infinite Gaussian mixture model is constructed as follows:

[0155] ;

[0156] In the formula, The KL divergence between the model distribution and the real data distribution. The weight parameters for the penalty term, express The minimum value of K is the number of Gaussian components in the model.

[0157] Optionally, the model building module 330 includes:

[0158] Model building unit, used to build infinite Gaussian mixture models based on Dirichlet processes;

[0159] The model optimization unit is used to take the sample input variable matrix as input to the infinite Gaussian mixture model and optimize the model using the multi-objective genetic algorithm NSGA-II.

[0160] The component partitioning unit is used to adaptively divide the samples in the modeling dataset into multiple Gaussian components based on the optimized infinite Gaussian mixture model, so as to characterize the primary current amplitude and measurement error conditions under different modes.

[0161] The indicator calculation module 340 is used to calculate the local probability and posterior probability of each sample in the monitoring dataset belonging to each Gaussian component based on the optimized infinite Gaussian mixture model, and to calculate the global BIP indicator based on the local probability and posterior probability of each sample.

[0162] Optionally, the indicator calculation module 340 includes:

[0163] The weight setting unit is used to set the weight matrix based on the pre-whitening method;

[0164] The local probability calculation unit is used to preprocess the covariance matrix based on the weight matrix, calculate the Mahalanobis distance of the samples in the monitoring dataset to the Gaussian components based on the formula, and look up the chi-square distribution critical value table based on the Mahalanobis distance to obtain the local probability of a sample belonging to each Gaussian component.

[0165]

[0166] In the formula, Represents Mahalanobis distance, Represents the weight matrix;

[0167] The posterior probability calculation unit is used to calculate the posterior probability of a sample belonging to each Gaussian component according to the formula:

[0168]

[0169] In the formula, Indicates sample Belongs to components The posterior probability, Represents the k-th Gaussian component The mixing coefficient, Represents the multivariate Gaussian distribution function. This represents the t-th sample in the monitoring dataset. Indicates Gaussian component The mean, Indicates Gaussian component The covariance matrix, where K represents the total number of Gaussian components. This represents the mixing coefficient of the i-th Gaussian component. Let represent the mean of the i-th Gaussian component. Let represent the covariance matrix of the i-th Gaussian component.

[0170] Optionally, the indicator calculation module 340 further includes:

[0171] The BIP metric calculation unit is used to calculate the global BIP metric for each sample based on the local and posterior probabilities of each sample in the monitoring dataset, according to the following formula:

[0172] ;

[0173] In the formula, This represents the BIP metric of the t-th sample in the monitoring dataset. Indicates sample The local probability, Indicates sample The posterior probability.

[0174] The anomaly detection module 350 is used to compare the global BIP index of each sample with the anomaly threshold at the significance level, and to determine whether there is a measurement error anomaly in the monitoring dataset based on the comparison results through a weighted majority voting mechanism.

[0175] The step of comparing the global BIP index of each sample with the outlier threshold at the significance level includes:

[0176] The significance level is determined based on the Monte Carlo test to identify the anomaly threshold.

[0177] Optionally, the anomaly detection module 350 includes:

[0178] The comparison unit compares the global BIP index of each sample with the outlier threshold at the significance level, and the comparison results are expressed as follows:

[0179] ;

[0180] The comparison results of all samples are represented as: D ;

[0181] In the formula, This represents the comparison result of the t-th sample in the monitoring dataset. This represents the BIP index of the t-th sample. Indicates the significance level, and D represents the comparison result dataset. This indicates the comparison result of the Nth sample;

[0182] The weighting unit is used to assign different weights to the comparison results according to the magnitude of the primary current and a formula:

[0183]

[0184] The anomaly calculation unit is used to calculate the final anomaly score of the monitoring dataset based on a weighted majority voting mechanism.

[0185]

[0186] In the formula, This represents the weight of the t-th sample in the monitoring dataset. The current represents the percentage of the positive-sequence current of the t-th sample relative to the rated current, and N represents the number of samples in the monitoring dataset. This indicates the anomaly score in the monitoring dataset. Indicates the results of sample comparison;

[0187] The anomaly detection unit is used to determine that there is a measurement error anomaly in the monitoring dataset when the anomaly score exceeds a preset threshold.

[0188] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working process of the system and modules described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0189] Figure 4 This is a schematic diagram of an electronic device according to an embodiment of the present invention. The electronic device is used for error monitoring of a current transformer. Figure 4 As shown, the electronic device 4 in this embodiment includes a memory 410, a processor 420, and a system bus 430. The memory 410 includes an executable program 4101 stored thereon. As those skilled in the art will understand, Figure 4 The electronic device structure shown does not constitute a limitation on the electronic device and may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0190] The following is combined with Figure 4 A detailed introduction to each component of the electronic device:

[0191] The memory 410 can be used to store software programs and modules. The processor 420 executes various functional applications and data processing of the electronic device by running the software programs and modules stored in the memory 410. The memory 410 may mainly include a program storage area and a data storage area. The program storage area may store the operating system, application programs required for at least one function (such as sound playback function, image playback function, etc.), etc.; the data storage area may store data created according to the use of the electronic device (such as cached data), etc. In addition, the memory 410 may include high-speed random access memory, and may also include non-volatile memory, such as at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0192] The memory 410 contains an executable program 4101 with a network request method. This executable program 4101 can be divided into one or more modules / units, which are stored in the memory 410 and executed by the processor 420 to achieve online monitoring of current transformer errors, etc. The one or more modules / units can be a series of computer program instruction segments capable of performing specific functions, describing the execution process of the computer program 4101 in the electronic device 4. For example, the computer program 4101 can be divided into functional modules such as a data acquisition module, a feature extraction module, a model building module, an indicator calculation module, and an anomaly detection module.

[0193] The processor 420 is the control center of the electronic device. It connects various parts of the electronic device via various interfaces and lines. By running or executing software programs and / or modules stored in the memory 410, and by calling data stored in the memory 410, it performs various functions and processes data, thereby monitoring the overall status of the electronic device. Optionally, the processor 420 may include one or more processing units; preferably, the processor 420 may integrate an application processor and a modem processor, wherein the application processor mainly handles the operating system, application programs, etc., and the modem processor mainly handles wireless communication. It is understood that the modem processor may also not be integrated into the processor 420.

[0194] The system bus 430 is used to connect various functional components inside the computer, transmitting data, address, and control information. Its type can be, for example, a PCI bus, an ISA bus, or a CAN bus. Instructions from the processor 420 are transmitted to the memory 410 via the bus, and the memory 410 sends data back to the processor 420. The system bus 430 is responsible for data and instruction exchange between the processor 420 and the memory 410. Of course, the system bus 430 can also connect to other devices, such as network interfaces and display devices.

[0195] In this embodiment of the invention, the executable program executed by the processor 420 included in the electronic device includes:

[0196] Acquire the measurement data of the fundamental current amplitude and phase in the three-phase current transformer, and construct the corresponding modeling dataset and monitoring dataset based on the normal error data and real-time monitoring data in the measurement data, respectively.

[0197] Sequence current decomposition is performed on the three-phase fundamental current in the modeling dataset. The relative mean deviation of the positive sequence current percentage and the zero sequence current percentage for each sample in the modeling dataset is calculated, and a sample input variable matrix is ​​constructed.

[0198] An infinite Gaussian mixture model is constructed based on the sample input variable matrix, and the infinite Gaussian mixture model is optimized by a multi-objective optimization algorithm. Based on the optimized infinite Gaussian mixture model, the samples in the modeling dataset are adaptively divided into a corresponding number of Gaussian components.

[0199] Based on the optimized infinite Gaussian mixture model, the local probability and posterior probability of each sample in the monitoring dataset belonging to each Gaussian component are calculated, and the global BIP index is calculated based on the local probability and posterior probability of each sample.

[0200] The global BIP index for each sample is compared with the anomaly threshold at the significance level. Based on the comparison results, a weighted majority voting mechanism is used to determine whether there are measurement error anomalies in the monitoring dataset.

[0201] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and modules described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0202] In the above embodiments, the descriptions of each embodiment have different focuses. For parts that are not described in detail or recorded in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0203] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A multi-condition adaptive online monitoring method for current transformers, characterized in that, include: Acquire the measurement data of the fundamental current amplitude and phase in the three-phase current transformer, and construct the corresponding modeling dataset and monitoring dataset based on the normal error data and real-time monitoring data in the measurement data, respectively. Sequence current decomposition is performed on the three-phase fundamental current in the modeling dataset. The relative mean deviation of the positive sequence current percentage and the zero sequence current percentage for each sample in the modeling dataset is calculated, and a sample input variable matrix is ​​constructed. An infinite Gaussian mixture model is constructed based on the sample input variable matrix, and the infinite Gaussian mixture model is optimized by a multi-objective optimization algorithm. Based on the optimized infinite Gaussian mixture model, the samples in the modeling dataset are adaptively divided into a corresponding number of Gaussian components. Based on the optimized infinite Gaussian mixture model, the local probability and posterior probability of each sample in the monitoring dataset belonging to each Gaussian component are calculated, and the global BIP index is calculated based on the local probability and posterior probability of each sample. The global BIP index for each sample is compared with the anomaly threshold at the significance level. Based on the comparison results, a weighted majority voting mechanism is used to determine whether there are measurement error anomalies in the monitoring dataset.

2. The method according to claim 1, characterized in that, The process of performing sequence current decomposition on the three-phase fundamental currents in the modeling dataset, calculating the relative mean deviation of the positive-sequence current percentage and the zero-sequence current percentage for each sample in the modeling dataset, and constructing the sample input variable matrix includes: Based on the measurement data of the amplitude and phase of the three-phase fundamental current, the three-phase fundamental current is decomposed into positive-sequence, negative-sequence, and zero-sequence components using the symmetrical component method: ；（1） In the formula, This represents the three-phase fundamental current vector. For complex operators, this indicates rotating the current phasor in the positive direction by 120°, and , Represents the positive-order components. Indicates the negative-order component. It is a zero-order component; Calculate the percentage of positive sequence current relative to rated current using formula (2). : ;（2） In the formula, This represents the positive order component of the t-th sample. Indicates the rated current of the current transformer; Calculate the sample according to formula (3) The proportion of zero-sequence current to positive-sequence current : ；（3） In the formula, Let represent the zero-order component of the t-th sample; Calculate the sample according to formula (4) Deviation of the proportion of zero-sequence current relative to the average value : ；（4） In the formula, To model the proportion of zero-sequence current in all samples of the dataset The mean, This represents the ratio of the zero-sequence current to the positive-sequence current in the t-th sample. Based on the deviation of the positive-sequence current percentage and zero-sequence current percentage from the mean for each sample, a sample input variable matrix X is constructed: ；（5） In the formula, This represents the last sample in the modeling dataset. This represents the percentage of the positive-sequence current relative to the rated current of the T-th sample in the modeling dataset. This represents the magnitude of the deviation of the zero-order proportion of the T-th sample from the mean.

3. The method according to claim 1, characterized in that, The process of constructing an infinite Gaussian mixture model based on the sample input variable matrix and optimizing the infinite Gaussian mixture model using a multi-objective optimization algorithm includes: Set the constraints for the infinite Gaussian mixture model, and express the constraints as follows: ,and ; In the formula, m represents a preset positive value. , They represent the variables respectively. Direction first , The mean parameter of a Gaussian component, , All are count variables; By using the logarithmic barrier method to treat the constraints as penalty terms, the objective function of the infinite Gaussian mixture model is constructed as follows: ； In the formula, The KL divergence between the model distribution and the real data distribution. The weight parameters for the penalty term, express The minimum value of K is the number of Gaussian components in the model.

4. The method according to claim 1, characterized in that, The process involves constructing an infinite Gaussian mixture model based on the sample input variable matrix, optimizing the infinite Gaussian mixture model using a multi-objective optimization algorithm, and adaptively dividing the samples in the modeling dataset into a corresponding number of Gaussian components based on the optimized infinite Gaussian mixture model, including: Constructing an infinite Gaussian mixture model based on the Dirichlet process; The sample input variable matrix is ​​used as the input to the infinite Gaussian mixture model, and the model is optimized by the multi-objective genetic algorithm NSGA-II. Based on the optimized infinite Gaussian mixture model, the samples in the modeling dataset are adaptively divided into multiple Gaussian components to characterize the primary current amplitude and measurement error conditions under different modes.

5. The method according to claim 1, characterized in that, The local probability and posterior probability of each sample in the monitoring dataset belonging to each Gaussian component, based on the optimized infinite Gaussian mixture model, include: The weight matrix is ​​set based on the pre-whitening method; The covariance matrix is ​​preprocessed based on the weight matrix. The Mahalanobis distance of the samples in the monitoring dataset to the Gaussian components is calculated based on the formula. Then, the chi-square distribution critical value table is consulted based on the Mahalanobis distance to obtain the local probability of a sample belonging to each Gaussian component. In the formula, Represents Mahalanobis distance, Represents the weight matrix; Calculate the posterior probability of a sample belonging to each Gaussian component using the formula: In the formula, Indicates sample Belongs to components The posterior probability, Represents the k-th Gaussian component The mixing coefficient, Represents the multivariate Gaussian distribution function. This represents the t-th sample in the monitoring dataset. Indicates Gaussian component The mean, Indicates Gaussian component The covariance matrix, where K represents the total number of Gaussian components. This represents the mixing coefficient of the i-th Gaussian component. Let represent the mean of the i-th Gaussian component. Let represent the covariance matrix of the i-th Gaussian component.

6. The method according to claim 1, characterized in that, The calculation of the global BIP index based on the local and posterior probabilities of each sample includes: Based on the local and posterior probabilities of each sample in the monitoring dataset, the global BIP metric for each sample is calculated according to the formula: ； In the formula, This represents the BIP metric of the t-th sample in the monitoring dataset. Indicates sample The local probability, Indicates sample The posterior probability.

7. The method according to claim 1, characterized in that, The comparison of the global BIP index of each sample with the anomaly threshold at the significance level, and the determination of whether there are measurement error anomalies in the monitoring dataset based on the comparison results through a weighted majority voting mechanism, includes: The global BIP index for each sample is compared with the outlier threshold at the significance level, and the comparison results are expressed as follows: ； The comparison results of all samples are expressed as follows: ; In the formula, This represents the comparison result of the t-th sample in the monitoring dataset. This represents the BIP index of the t-th sample. Indicates the significance level, and D represents the comparison result dataset. This indicates the comparison result of the Nth sample; Based on the magnitude of the primary current, different weights are assigned to the comparison results according to the formula: ； The final anomaly score for the monitoring dataset is calculated based on a weighted majority voting mechanism. ； In the formula, This represents the weight of the t-th sample in the monitoring dataset. The current represents the percentage of the positive-sequence current of the t-th sample relative to the rated current, and N represents the number of samples in the monitoring dataset. This indicates the anomaly score in the monitoring dataset. Indicates the results of sample comparison; When the abnormal score exceeds the preset threshold, it is determined that there is an abnormal measurement error in the monitoring dataset.

8. A multi-condition adaptive online monitoring system for current transformers, characterized in that, include: The data acquisition module is used to acquire measurement data of the fundamental current amplitude and phase in the three-phase current transformer, and to construct corresponding modeling datasets and monitoring datasets based on the normal error data and real-time monitoring data in the measurement data. The feature extraction module is used to perform sequence current decomposition on the three-phase fundamental current in the modeling dataset, calculate the relative mean deviation of the positive sequence current percentage and the zero sequence current percentage for each sample in the modeling dataset, and construct the sample input variable matrix. The model building module is used to construct an infinite Gaussian mixture model based on the sample input variable matrix, and optimize the infinite Gaussian mixture model through a multi-objective optimization algorithm. Based on the optimized infinite Gaussian mixture model, the samples in the modeling dataset are adaptively divided into a corresponding number of Gaussian components. The index calculation module is used to calculate the local probability and posterior probability of each sample in the monitoring dataset belonging to each Gaussian component based on the optimized infinite Gaussian mixture model, and to calculate the global BIP index based on the local probability and posterior probability of each sample. The anomaly detection module compares the global BIP index of each sample with the anomaly threshold at the significance level, and determines whether there are measurement error anomalies in the monitoring dataset based on the comparison results through a weighted majority voting mechanism.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the multi-condition adaptive online monitoring method for current transformers as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed, it implements the steps of the multi-condition adaptive online monitoring method for current transformers as described in any one of claims 1 to 7.

Images