Circuit breaker data error identification and correction method and system

Abstract

The invention discloses a circuit breaker data error identification and correction method and system, and belongs to the technical field of circuit breaker abnormal data identification, and the method comprises the steps: collecting the electrical data and equipment state data of a circuit breaker in the operation process, and obtaining an original data sequence; performing frequency multi-scale decomposition on the original data sequence through wavelet transform, and extracting a feature vector of the original data sequence; based on the feature vector, performing error identification on the original data sequence through a preset error identification model, and outputting an error identification result and a corresponding error type; and a correction algorithm corresponding to the error type is selected to correct the error identification result, and a corrected data sequence is obtained. Therefore, by implementing the application, the problems that in the prior art, the collected electrical data has a relatively large error and the error cannot be accurately corrected can be solved.

Description

Technical Field

[0001] This application belongs to the technical field of circuit breaker abnormal data identification, and specifically relates to a method and system for error identification and correction of circuit breaker data. Background Art

[0002] As a core protection device in modern power systems, intelligent circuit breakers are widely used in distribution networks, power equipment, and various industrial automation systems, mainly for real-time monitoring and protection of power lines. Through embedded sensors, communication modules, and data processing units, intelligent circuit breakers can collect various electrical parameters such as current, voltage, and power in real time, and achieve fault detection, protection actions, and operation optimization through intelligent algorithms. However, in actual operation, the environment of the power system is complex and changeable, and intelligent circuit breakers face a series of challenges such as data acquisition accuracy, timeliness of fault diagnosis, and reliability of device response.

[0003] Currently, intelligent circuit breakers mainly rely on statically set thresholds and simple judgment logics for fault detection and protection actions. By collecting data through analog sensors, the acquisition frequency is low and the accuracy is limited, unable to capture rapid changes or minute fluctuations in electrical parameters, resulting in large errors in the collected electrical data. Moreover, in the data processing process, traditional methods often rely on fixed rules and simple filtering algorithms to remove noise and abnormal data, and cannot effectively adapt to the dynamic changes of data in complex power environments. Summary of the Invention

[0004] This application proposes a method and system for error identification and correction of circuit breaker data, which can solve the problems that the collected electrical data in the prior art has large errors and cannot accurately correct the errors.

[0005] The first aspect of this application provides a method for error identification and correction of circuit breaker data, and the method includes:

[0006] Collect electrical data and equipment status data during the operation of the circuit breaker to obtain an original data sequence;

[0007] Perform multi-scale decomposition in frequency on the original data sequence through wavelet transform to extract the feature vector of the original data sequence;

[0008] Based on the feature vector, perform error identification on the original data sequence through a preset error identification model, and output the error identification result and the corresponding error type;

[0009] Select the correction algorithm corresponding to the error type to correct the error identification result to obtain a corrected data sequence.

[0010] The above solution first collects the change data of the electrical parameters of the circuit breaker during operation and the equipment status data generated by operations in units of time to obtain the original data sequence with errors to be corrected. Then, wavelet transform is used to decompose the original data sequence in different frequency bands, and the characteristic information of the circuit breaker data in different frequency bands is extracted, so that the data points in the original data sequence provide information in different frequencies and statistical perspectives, which can improve the accuracy of error identification in the subsequent error identification process and help the model capture more potential error content. After error identification, the error identification result and the corresponding error type are output, and then different error correction methods are adopted based on different error types to ensure the accuracy and reliability of the data, and the corrected high-accuracy circuit breaker data related to the actual operation situation is obtained, providing guarantee for the long-term stable operation of the power system.

[0011] In a possible implementation method of the first aspect, the original data sequence is decomposed in multiple scales in terms of frequency through wavelet transform, and the feature vector of the original data sequence is extracted, specifically:

[0012] Based on the preset decomposition level, the original data sequence is segmented in frequency through wavelet transform, and the characteristic information of the original data sequence in each frequency band is extracted; wherein, the decomposition level is determined by the complexity and fluctuation of the original data sequence;

[0013] According to the characteristic information, the statistical features of each frequency band are extracted from the original data sequence; wherein, the statistical features include signal energy, data amplitude, and data deviation degree;

[0014] The feature vector of the original data sequence is constructed through the statistical features and the characteristic information.

[0015] The above solution uses wavelet transform to divide the original data sequence into multiple frequency bands. The more levels of decomposition, the more detailed the capture of data features, and the more the obtained feature vector can reflect the true situation of the data. Then, based on each frequency band, the data features within that frequency band are extracted, such as the fluctuation situation, energy intensity, and overall symmetry of the data in a certain frequency band, to highlight the mutation and abnormal fluctuation of the error points in the original data sequence, facilitating more accurate positioning of the error points in the subsequent process.

[0016] In a possible implementation method of the first aspect, the statistical features are specifically:

[0017] The signal energy, the specific expression is:

[0018]

[0019] In the formula, E

[0019] , , jis the total energy of the signal in the j-th frequency band, n is the total number of the feature information, and c ji is the i-th feature information in the j-th frequency band;

[0020] The data amplitude, and the specific expression is:

[0021]

[0022] In the formula, is the fluctuation degree of the feature information in the j-th frequency band, is the mean value of the feature information in the j-th frequency band;

[0023] The data deviation degree, and the specific expression is:

[0024]

[0025] In the formula, γ j is the symmetry of the signal in the j-th frequency band.

[0026] In a possible implementation method of the first aspect, before performing multi-scale decomposition on the original data sequence in terms of frequency through wavelet transform, it further includes:

[0027] Performing outlier removal, data cleaning, and data standardization on the original data sequence in sequence to obtain a preprocessed original data sequence;

[0028] Among them, the outlier removal is to delete the data in the original data sequence that deviates from the preset normal range; the data cleaning is to delete the data in the original data sequence that exceeds the preset frequency; the data standardization is to map different electrical parameters in the original data sequence to a preset same standard.

[0029] In the above solution, after collecting data, the obtained original data sequence is preprocessed. By removing outliers, data cleaning, and data standardization, the data quality is improved, and the influence of some redundant data on subsequent feature extraction and error identification is avoided.

[0030] In a possible implementation method of the first aspect, based on the feature vector, the original data sequence is subjected to error identification through a preset error identification model, and an error identification result and the corresponding error type are output. Specifically:

[0031] Input the feature vector into the error identification model, and by calculating the matching degree between the feature vector and the preset error label, the error existing in the original data sequence is identified to generate an error identification result; among them, the error label is formed by performing error annotation on historical circuit breaker data;

[0032] Determine the error type by identifying the error pattern of the error identification result.

[0033] The above solution can more accurately and quickly identify the errors existing in the original data sequence through pre-set error tags.

[0034] In a possible implementation method of the first aspect, select the correction algorithm corresponding to the error type to correct the error identification result, and obtain the corrected data sequence, specifically:

[0035] If the error type is the existence of outliers, correct the error identification result by the mean replacement method or the interpolation correction method;

[0036] If the error type is the existence of trend errors, correct the error identification result by the fitting prediction method;

[0037] When the corrected error identification result meets the preset device characteristic constraints, obtain the corrected data sequence through the corrected error identification result; wherein, the device characteristic constraints are determined by the physical characteristics of the circuit breaker and the normal fluctuation range of the electrical data.

[0038] The above solution adopts different error correction methods for different error types. For discrete outliers, directly use the mean replacement and interpolation methods to replace the error data points. For trend errors, predict the future value through the fitting prediction method with normal historical data, and replace the trend error with the predicted data. Both error correction methods take into account that the data after error correction should have smoothness and a certain degree of accuracy. Therefore, different error correction methods are adopted according to the different arrangements of errors to obtain more accurate error correction values.

[0039] In a possible implementation method of the first aspect, correct the error identification result by the fitting prediction method, specifically:

[0040] Select the first data sequence before the error data point in the error identification result, and perform trend fitting on the first data sequence to obtain the predicted data value;

[0041] Use the predicted data value to replace the error data point of the error identification result to obtain the corrected error identification result.

[0042] In a possible implementation method of the first aspect, after obtaining the corrected data sequence, it further includes:

[0043] Compare the corrected data sequence with the obtained historical data sequence, and quantify the effect of this error correction by calculating the correction effect index; wherein, the correction effect index includes the root mean square error, mean absolute error, and correlation coefficient between the corrected data sequence and the historical data sequence.

[0044] After the above solution completes the error correction, compare the corrected data with the normal historical data sequence to evaluate the error correction effect. Evaluate whether this error correction conforms to the electrical law and the actual operation of the circuit breaker by calculating the correction effect index, improve the rationality and accuracy of the error correction, and ensure that the data after the error correction can truly reflect the operating state of the intelligent circuit breaker.

[0045] The second aspect of this application provides a system for error identification and correction of circuit breaker data, and the system includes: a data sequence acquisition module, a feature extraction module, an error identification module, and an error correction module;

[0046] Among them, the data sequence acquisition module is used to collect electrical data and equipment status data during the operation of the circuit breaker to obtain the original data sequence;

[0047] The feature extraction module is used to perform multi-scale decomposition in frequency on the original data sequence through wavelet transform to extract the feature vector of the original data sequence;

[0048] The error identification module is used to perform error identification on the original data sequence based on the feature vector through a preset error identification model, and output the error identification result and the corresponding error type;

[0049] The error correction module is used to select the correction algorithm corresponding to the error type to correct the error identification result to obtain the corrected data sequence. Brief Description of the Drawings

[0050] In order to more clearly illustrate the technical solutions of this application, the drawings required for the implementation will be briefly introduced below. Obviously, the drawings in the following description are only some implementations of this application. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0051] Figure 1 It is a schematic flow chart of a method for error identification and correction of circuit breaker data provided by an embodiment of this application;

[0052] Figure 2 It is a structural diagram of a system for error identification and correction of circuit breaker data provided by an embodiment of this application. Detailed Embodiments

[0053] Next, the technical solutions in the embodiments of the present application will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present application.

[0054] It should be understood that the step numbers used in the text are only for convenience of description and are not intended to limit the execution order of the steps.

[0055] First Embodiment

[0056] There are significant deficiencies in the automation of error identification and the accuracy of error correction in the prior art. It is impossible to identify the errors existing in the collected electrical data in real time and accurately, resulting in the errors not being processed in time, thus affecting the accuracy of the data and the reliability of power grid decision-making. Therefore, how to improve the deficiencies in error identification and correction of the prior art, realize accurate correction of the errors in the circuit breaker data, provide a basis for timely detection of equipment failures on the transmission line, and thus enhance the operation stability of the system.

[0057] As Figure 1 shown, to solve the problem that there are large errors in the collected electrical data and the errors cannot be accurately corrected in the prior art, the first embodiment of the present application provides a specific flow schematic diagram of a method for error identification and correction of circuit breaker data. The method for error identification and correction of circuit breaker data in this embodiment includes steps S1 to S4, which are described in detail as follows:

[0058] Step S1, collect the electrical data and equipment status data of the circuit breaker during operation to obtain the original data sequence.

[0059] In the embodiments of the present application, the electrical data and equipment status data of the circuit breaker during operation are collected in real time at a high sampling frequency by a high-precision sensor with good anti-interference ability. Among them, the electrical data includes the current, voltage, and power data of the branch where the circuit breaker is located, and the equipment status data includes the switch status and action time of the circuit breaker during operation, etc.

[0060] During the data acquisition process, it is crucial to accurately capture rapidly changing electrical data. To ensure the accuracy and reliability of the data, it is essential to use high-precision sensors. In the embodiments of this application, the set sampling frequency is not less than 1 kHz to ensure that rapid fluctuations in parameters such as instantaneous current and voltage can be precisely captured. This is because a lower sampling frequency may cause important transient changes to be missed, thus affecting subsequent error identification and correction. Therefore, by increasing the sampling frequency, the system can more meticulously capture the fluctuations of electrical data over time, especially in the case of sudden changes in current and voltage. Moreover, the accuracy requirement of the sensor reaches more than 0.1%, which means that the error of each collected data will not exceed 0.1%, thereby ensuring the high accuracy of the data and avoiding identification errors caused by data errors.

[0061] In addition to high precision and high sampling frequency, the anti-interference ability of the sensor is also very important. During the operation of electrical equipment, electromagnetic interference may come from nearby high-voltage equipment or other electronic devices. Such interference may affect the output signal of the sensor, resulting in large errors in the collected data. Therefore, choosing a sensor with good anti-interference ability, especially one that can effectively shield external electromagnetic interference, can reduce the generation of errors. To further improve the reliability of the data, hardware filtering techniques and signal correction algorithms are usually used to process the collected data.

[0062] Exemplarily, a designed low-pass filter is used to filter out high-frequency noise in the collected data or the Kalman filtering algorithm is used to denoise the collected data, enhancing the accuracy and stability of the data.

[0063] Optionally, in the embodiments of this application, a Butterworth filter is selected to implement data filtering. The Butterworth filter is fully called the Butterworth filter, and its characteristic is that the frequency response curve in the passband is maximally flat without ripples, while in the stopband it gradually drops to zero.

[0064] During the data acquisition process, all the collected data will be stored according to the time series to form an original data sequence. The original data sequence has the characteristics of high frequency and low error.

[0065] Specifically, assuming that at time point t, the electrical data such as current and voltage collected are I(t) and V(t) respectively, then the corresponding original data sequence is {I(t1), I(t2),..., I(t n )} V(t1), V(t2),..., V(t n ).

[0066] Furthermore, the data collected by the sensors needs to ensure its time synchronization and accuracy. For example, the sampling period must be kept constant, and the output of the sensors should pass through a clock synchronization mechanism to ensure that the data tags at each time point are consistent. In practical applications, the collected time series data should use the following formula to represent the relationship between current and voltage data:

[0067] P(t) = V(t) × I(t);

[0068] In the formula, P(t) is the instantaneous power at time t, V(t) is the voltage at time t, and I(t) is the current at time t.

[0069] By calculating the instantaneous power, the working state of the circuit breaker can be monitored in real time, and further provide data support for the error identification model. This precise data collection can not only ensure high-quality data input, but also lay a foundation for subsequent data preprocessing, feature extraction and error correction steps, ensuring the accuracy of error identification.

[0070] After the data is collected, the obtained original data sequence is preprocessed, and outlier removal, data cleaning and data standardization are carried out in turn. This is a key operation to ensure data quality and provide a reliable basis for subsequent analysis. In the operating data of the circuit breaker, outliers usually refer to those points that significantly deviate from the normal data range. These outliers may be caused by sensor failures, external interferences or other problems. To eliminate these outliers, a method based on statistical principles is adopted. The mean and standard deviation of the data points in the original data sequence are calculated, and then the data is filtered according to the mean and standard deviation using the normal distribution hypothesis to complete outlier removal.

[0071] Specifically, let the mean of each data point in the original data sequence be μ and the standard deviation be σ, then the outliers can be determined by the following formula:

[0072] |x i- μ| > 3σ;

[0073] In the formula, x i is the value of a certain data point in the original data sequence. If x i exceeds the range of ±3 times the standard deviation of the mean, it is considered an outlier and deleted from the original data sequence. In this way, the elimination of outliers helps to improve the quality of the data and avoid their impact on subsequent analysis and model training.

[0074] In the data cleaning stage, the moving average method is used to remove high-frequency noise from the original data sequence. Since the operating data of the circuit breaker may be affected by factors such as short-term fluctuations, sensor noise, or electromagnetic interference, the moving average can effectively smooth the data and eliminate this high-frequency noise. The moving average filtering method performs smoothing by calculating the average value of the data within a fixed window around each data point. Specifically: Assume the original data sequence is {x1, x2,..., x n}, and the window size is w. Then the data sequence after moving average processing is Specifically as follows:

[0075]

[0076] In the formula, is the data point after moving average filtering. The moving window size w is a preset constant, usually taking an odd value. By performing data cleaning through the moving average, the time series data can be effectively smoothed, and the noise caused by instantaneous fluctuations can be reduced.

[0077] Finally, data standardization is performed on the original data sequence. This is to map the data of different electrical parameters to the same dimension, facilitating subsequent feature extraction and model training. Due to the magnitude differences of different parameters, if standardization is not performed, the error recognition model may be affected by certain parameters with larger dimensions, resulting in deviations in the prediction results. Commonly used standardization methods include normalization, which generally maps the data to the interval [0, 1] or [-1, 1]. For a given parameter x i , the normalized value x′ i can be calculated through the following formula:

[0078]

[0079] Among them, x i is a data point in the original data sequence, and min(x) and max(x) are the maximum and minimum values in the original data sequence respectively. The normalized data x′ i will be mapped to the interval [0, 1], thus eliminating the dimension difference and making subsequent feature extraction and error recognition model training more efficient and accurate. In this way, different electrical parameters (such as voltage, current, etc.) can be compared and analyzed on the same scale, providing unified input data for the error recognition model.

[0080] These preprocessing operations complement each other, jointly ensuring the high quality and consistency of the data. After removing outliers, the moving average filtering further smooths the noise in the data, and the final standardization process unifies the data of different features on the same scale, providing clear and standardized data input for the error recognition model, effectively improving the accuracy of subsequent error recognition.

[0081] Step S2: Perform multi-scale decomposition on the original data sequence in terms of frequency through wavelet transform to extract the feature vector of the original data sequence.

[0082] In the embodiment of the present application, wavelet transform is adopted to perform multi-scale decomposition on the original data sequence, aiming to extract the feature information of the breaker data in different frequency bands. Among them, wavelet transform is a transform analysis method. By providing a "time-frequency" window that changes with frequency, it conducts time-frequency analysis and processing of signals. Through dilation and translation operations, the signal is gradually refined at multiple scales. Finally, it achieves fine time division at high frequencies and fine frequency division at low frequencies, and can automatically adapt to the requirements of time-frequency signal analysis, so as to focus on any details of the signal to highlight the characteristics of the signal.

[0083] Optionally, the embodiment of the present application selects the Daubechies wavelet basis function as the transform function. The full name of the Daubechies wavelet basis function is the multi-Bezier wavelet, which is mainly applied to discrete wavelet transforms. Because of its good localization properties, the multi-Bezier wavelet can provide the resolution of both time and frequency, and is suitable for processing signals containing transient changes and high-frequency noise. The multi-Bezier wavelet has good smoothness and finite support, which is particularly effective for dealing with the common noise and mutation points in electrical data.

[0084] The decomposition level of wavelet transform is generally determined according to the complexity and fluctuation of the data, and is generally 3 to 5 levels. The more levels there are, the finer the decomposed frequency bands will be, and more detailed signal characteristics can be captured.

[0085] Exemplarily, if the decomposition level is set to 3 levels and the sampling frequency is assumed to be 1 kHz, the frequency bands can be divided into: low-frequency band 0 - 62.5 Hz, medium-low frequency band 62.5 - 125 Hz, medium-high frequency band 125 - 250 Hz, and high-frequency band 250 - 500 Hz.

[0086] According to the preset decomposition level, the original data sequence is segmented in terms of frequency through the Daubechies wavelet basis function, and the wavelet coefficients of each frequency band are obtained as the feature information of the original data sequence. The wavelet coefficients reflect the multi-scale characteristics of the data in different frequency ranges. The low-frequency part can reflect the overall trend of the data, while the high-frequency part can reveal the mutations, abnormal fluctuations, and short-term changes in the signal. Therefore, for the wavelet coefficients of each frequency band, the signal energy, data amplitude, and data deviation degree of each frequency band can be extracted as the corresponding statistical features to supplement the feature information.

[0087] Signal energy is an important indicator for describing signal strength and can reflect the energy distribution of the signal in that frequency band. The data amplitude is generally the variance of the calculated data. The statistical features can jointly form a complete feature vector with the wavelet coefficients, improving the accuracy of the error recognition model and helping the model capture more potential error patterns.

[0088] Specifically, for the wavelet coefficient sequence {c j1 , c j2 ,..., c jn} in the j-th frequency band, the following formulas can be used to calculate statistical features such as signal energy, data amplitude, and data deviation degree:

[0089] The signal energy, with the specific expression:

[0090]

[0091] In the formula, E j is the total energy of the signal in the j-th frequency band, reflecting the signal strength. n is the total number of the wavelet coefficients, and c ji is the i-th wavelet coefficient in the j-th frequency band;

[0092] The data amplitude, with the specific expression:

[0093]

[0094] In the formula, is the fluctuation degree of the wavelet coefficients in the j-th frequency band. A larger variance means greater signal fluctuation in that frequency band, which may be related to abnormal changes; is the mean value of the wavelet coefficients in the j-th frequency band;

[0095] The data deviation degree, with the specific expression:

[0096]

[0097] In the formula, γ j is the symmetry of the signal in the j-th frequency band. A larger data deviation degree may indicate a greater asymmetry in the signal, further providing a basis for error recognition.

[0098] Construct a feature vector by combining these statistical features with the multi-level feature information composed of wavelet coefficients, enabling each data point in the original data sequence to provide information from different frequencies and statistical perspectives, enhancing the error recognition model's ability to capture complex data patterns. These features can be provided as input data to the error recognition model to improve the model's recognition accuracy for outliers, especially in electrical parameters where instantaneous current, voltage changes, etc. may occur. In this way, the feature vector not only contains the time-frequency information of the wavelet coefficients themselves but also combines the statistical properties of the signal, enabling the error recognition model to analyze the data more comprehensively and ensuring higher accuracy and robustness.

[0099] Step S3, based on the feature vector, perform error recognition on the original data sequence through a preset error recognition model, and output the error recognition result and the corresponding error type.

[0100] Then input the obtained feature vector and the original data sequence into the error recognition model, and perform error recognition on the original data sequence through the preset error recognition model.

[0101] The error recognition model is jointly constructed by a support vector machine and a neural network. The support vector machine is a supervised learning method that classifies data by finding an optimal decision boundary and has strong generalization ability for high-dimensional data. To improve the performance of the support vector machine in circuit breaker data error recognition, a radial basis function is used as the kernel function. The expression of the radial basis function is:

[0102]

[0103] In the formula, x and y are the feature vectors of the input data, and σ is the coefficient of the radial basis function, which determines the similarity between data points. By adjusting σ, the influence range between data points can be controlled. A smaller σ will result in a more local influence, while a larger σ will expand the influence range. Therefore, during the training process of the error recognition model, the cross-validation method is used to optimize σ and the penalty factor C (the penalty factor determines the penalty degree for misclassification). Through the cross-validation method, the generalization ability of the model can be evaluated under different combinations of hyperparameters, so as to select the best parameters, avoid overfitting, and improve the accuracy of the model on the test set.

[0104] For the neural network algorithm, a multi-layer perceptron structure is selected, which is a feedforward neural network with at least one hidden layer. The number of nodes in the input layer of the multi-layer perceptron structure is determined by the dimension of the feature vector. Usually, the input data will have a relatively high dimension after feature extraction. Therefore, the number of nodes in the input layer should be consistent with the dimension of the feature vector. The number of nodes in the middle hidden layer is usually determined by the trial-and-error method. A common approach is to start with a small number of nodes and gradually increase and evaluate the performance of the model. Generally, a hidden layer structure with 1 to 2 layers can effectively extract complex patterns in the data. Between each layer, the neural network non-linearly transforms the data through activation functions (such as ReLU or sigmoid) to enhance the learning ability of the model.

[0105] The neural network algorithm is trained using the backpropagation function. The core idea of the backpropagation function is to calculate the gradient of each weight and update the weights according to the gradient descent method to minimize the loss function. The loss function is usually selected as the mean squared error or cross-entropy, depending on the nature of the task. The weight update formula for the neural network is:

[0106]

[0107] where w ij is the weight from the i-th layer to the j-th layer, η is the learning rate, and E is the loss function. By continuously adjusting the weights, the neural network gradually learns the features of the data.

[0108] To ensure the stability of the neural network training process and prevent overfitting, the early stopping method is adopted. The early stopping method monitors the loss value on the validation set. If the validation set loss does not decrease for multiple training epochs, the training is stopped to avoid overfitting of the model on the training set and improve the generalization ability of the model. During the training process, parameters such as the learning rate and batch size are usually adjusted to accelerate convergence and ensure that the neural network can find the optimal solution in a relatively short time.

[0109] Through the above training strategies, both the support vector machine and the neural network can make full use of the feature information extracted from the wavelet transform for error identification. The support vector machine can handle complex non-linear classification problems through the radial basis kernel function, while the neural network automatically extracts complex features of the data through a multi-layer structure and uses backpropagation to optimize the network structure. Both combine the time-frequency features and statistical characteristics of the data, can effectively identify error patterns in the intelligent circuit breaker data, and provide accurate error type determination.

[0110] The error identification model is based on a large number of preset error labels, calculates the matching degree between the feature vector and the preset error labels, and identifies outliers and the trend change characteristics of the low-frequency coefficients through the abnormal fluctuations of the high-frequency coefficients provided by the feature vector, and identifies the error identification results and the corresponding error types.

[0111] Among them, the error tags are generated from manual annotation or by combining physical models and historical breaker data anomaly detection.

[0112] Step S4, select the correction algorithm corresponding to the error type to correct the error recognition result, and obtain the corrected data sequence.

[0113] In the embodiments of the present application, the error types include the existence of outliers and the existence of trend errors. The judgment of outliers is identified by the abnormal fluctuation of the high-frequency coefficients provided by the eigenvectors, and the trend error is obtained from the trend change characteristics of the low-frequency coefficients provided by the eigenvectors. In the data with trend errors, regular fluctuations will appear, such as long-term drift and periodic deviation. In the data with outliers, the outliers will appear in the form of discrete points.

[0114] In view of the characteristics of the above error data, the embodiments of the present application adopt different error correction methods to correct the error recognition results to ensure the accuracy and reliability of the data.

[0115] For the correction of outliers, the embodiments of the present application adopt the mean replacement method. According to the time series characteristics of the original data sequence, the mean value is calculated by selecting the data within a certain time window before and after the outlier, and this mean value is used as the replacement value to replace the outlier.

[0116] Specifically, assume that the original data sequence is {x1, x2,..., x n}}, if a certain data point x i is determined to be an outlier, then the data at w time steps before and after this point can be selected for mean calculation, as follows:

[0117]

[0118] In the formula, is the corrected value, w is the size of the time window, which determines the range of the front and back data used to calculate the mean. Through the mean replacement method, accidental noise and mutation values can be effectively eliminated, so that the corrected data maintains continuity and consistency.

[0119] In addition, in other embodiments, the interpolation correction method is used to correct the outliers in the original data sequence. The interpolation correction method generates reasonable estimated values by using the information of adjacent data points. Common interpolation methods include cubic spline interpolation and Lagrange interpolation. The cubic spline interpolation method constructs a smooth polynomial function, which is continuous at all nodes and has continuous derivatives, and can provide smoother and more accurate interpolation results. The expression of cubic spline interpolation is:

[0120] S(x) = a(x - x0) 3 + b(x - x0)2 +c(x - x0)+d;

[0121] In the formula, a, b, c, and d are undetermined coefficients of the cubic spline interpolation formula, x0 is the abscissa of the interpolation point, and x is the point to be interpolated. By solving a set of equations, these coefficients can be determined, thereby obtaining a smooth interpolation curve. Through cubic spline interpolation, interpolation is performed on the outliers in the original data sequence to obtain a smooth and corrected data sequence.

[0122] Lagrange interpolation rule is to construct an interpolation polynomial using the values of all known data points in the original data sequence. The specific expression of the interpolation polynomial is:

[0123]

[0124] In the formula, y i is the value of the known data point, and x i is the abscissa of the known data point. Lagrange interpolation is suitable for interpolating data points in a small range, but for large data sets, the computational cost is relatively high. Whether it is cubic spline interpolation or Lagrange interpolation, they can effectively estimate reasonable interpolation results based on the surrounding data points, thereby correcting outliers.

[0125] For the correction of trend errors, the embodiment of the present application adopts a fitting prediction method. Trend errors usually manifest as a certain regular fluctuation in the data. The future data values can be predicted by fitting the trend of historical data. Common fitting methods include polynomial fitting and exponential smoothing method. Polynomial fitting captures the trend of the data by fitting a high-degree polynomial function. Assuming that the trend of the original data sequence can be represented by an m-degree polynomial, the fitting function is:

[0126] f(x) = a m x m + a m-1 x m-1 + … + a1x + a0;

[0127] In the formula, a m , a m-1 , …, a0 are polynomial coefficients solved by methods such as the least squares method. Through this method, the future data values can be predicted, and these predicted values are used to replace the error data points. Therefore, in the embodiment of the present application, by performing fitting prediction on a segment of the data sequence before the error data point in the original data sequence, the predicted data values are obtained, and then the error data points are replaced with the predicted data values to obtain the corrected data sequence.

[0128] In addition, in other embodiments, the exponential smoothing method can also be used to correct the trend error. The exponential smoothing method is a forecasting method based on weighted average. It smooths the data and predicts future values by assigning exponentially decaying weights to historical data. The basic formula of the exponential smoothing method is as follows:

[0129]

[0130] In the formula, is the predicted value, y t-1 is the actual value at the previous moment, is the predicted value at the previous moment, and α is the smoothing coefficient, whose value range is (0, 1). A larger α value gives more weight to recent historical data, while a smaller α value makes the influence of the long-term trend greater. Through this method, the long-term trend in the circuit breaker data can be effectively captured, and future data can be predicted, thereby correcting the anomalies caused by trend errors.

[0131] Therefore, the exponential smoothing method assigns exponentially decaying weights to the data before the error data points in the original data sequence to predict the corresponding future values, and uses the future values to correct the existing trend error data.

[0132] During the error identification process and data correction process, the physical meaning of the data and the actual operating conditions of the circuit breaker also need to be fully considered. For example, it is normal for parameters such as current and voltage to fluctuate within a certain range, but changes beyond a certain range may cause equipment damage. Therefore, the corrected data must conform to electrical laws and cannot deviate from the operating characteristics of the equipment. By introducing physical models and equipment characteristic constraints into the error identification method and correction method, the rationality and accuracy of the results can be further improved, ensuring that the data after error correction can truly reflect the operating state of the intelligent circuit breaker.

[0133] As an improvement to the above solution, a series of evaluations also need to be performed on the corrected data sequence to quantify the quality of the correction effect, so as to ensure that the corrected data has sufficient accuracy and reliability. The correction effect indicators used in the embodiments of this application include root mean square error, mean absolute error, and correlation coefficient, and these indicators can evaluate the quality of the data after correction from different perspectives.

[0134] The embodiments of this application mainly compare the historical data sequence of the circuit breaker with the corrected data sequence, and calculate the corresponding root mean square error, mean absolute error, and correlation coefficient to evaluate the effect of this error correction. The historical data sequence is the true data of the circuit breaker obtained previously and does not contain errors.

[0135] Specifically, the root mean square error (RMSE) is one of the important indicators for evaluating data correction errors, which can reflect the difference between the corrected data and the actual data. The calculation formula for the root mean square error is:

[0136]

[0137] where x i is the actual data value (referring to the data point in the historical data sequence in the embodiments of the present application), is the corrected data value, and n is the number of data points. The smaller the root mean square error, the closer the corrected data is to the actual data, and the better the correction effect. This indicator can intuitively reflect the overall error size of the data, especially suitable for measuring the uniformity of the differences between data points.

[0138] The mean absolute error (MAE) is another indicator for measuring the error size, which evaluates the correction effect by calculating the absolute value of the error for each data point and taking its average. The calculation formula for the mean absolute error is:

[0139]

[0140] where x i is the actual data value (referring to the data point in the historical data sequence in the embodiments of the present application), is the corrected data value, and n is the number of data points. Different from the root mean square error, the mean absolute error pays more attention to the average level of the errors, giving the same weight to large errors and small errors. The smaller the mean absolute error, the more accurate the correction result, especially suitable for scenarios with fewer outliers and a higher tolerance for errors. Through these two indicators, the difference between the corrected data and the actual data can be quantified, so as to determine the effect of the error correction method.

[0141] The correlation coefficient is used to measure the linear correlation degree between the corrected data and the actual data. The calculation formula for the correlation coefficient is:

[0142]

[0143] where a and b are the means of the actual data and the corrected data respectively. The value range of the correlation coefficient is from -1 to 1. When the correlation coefficient approaches 1, it indicates a strong positive correlation between the corrected data and the actual data, and the correction effect is good; when the correlation coefficient approaches -1, it indicates a negative correlation between the corrected data and the actual data, and the correction effect is poor; when the correlation coefficient is close to 0, it indicates that there is no linear relationship between the two.

[0144] To ensure the effectiveness of the error correction method, the above-mentioned correction effect indicators need to meet the preset threshold requirements. For example, if the root mean square error is greater than a certain set value or the correlation coefficient is less than a certain set value, the correction effect is considered unsatisfactory. At this time, the system will automatically return to the error recognition model training step, reselect a suitable model algorithm, adjust the model parameters, or increase the amount of training data to ensure that the model can better identify and correct errors in the data. In this process, it may be necessary to expand the existing data set and add more labeled error data to enhance the generalization ability and accuracy of the model. In addition, it is also crucial to regularly monitor and evaluate the correction effect to adapt to the changes in the operating state of the intelligent circuit breaker and the drift of data characteristics. As the operating state of the device changes continuously, the data characteristics may also drift, so it is necessary to continuously adjust the correction method to ensure that it can always effectively handle newly emerging error patterns and ensure the long-term stable operation of the system. Through this dynamic correction and evaluation mechanism, the accuracy and adaptability of error correction can be continuously improved, ensuring that the corrected data always meets the actual working requirements of the intelligent circuit breaker.

[0145] Among them, the threshold requirements are related to the error correction accuracy requirements, the allowable error range of electrical parameters for the safe operation of the device, and industry standards and specifications.

[0146] Implementing the embodiments of the present application has the following beneficial effects:

[0147] In the embodiments of the present application, first, the change data of the electrical parameters of the circuit breaker during operation and the device state data generated by operations are collected in units of time to obtain the original data sequence of the error to be corrected. Then, wavelet transform is used to decompose the original data sequence in different frequency bands, and the characteristic information of the circuit breaker data in different frequency bands is extracted, so that the data points in the original data sequence provide information in different frequencies and statistical perspectives, which can improve the accuracy of error recognition in the subsequent error recognition process and help the model capture more potential error contents. After error recognition, the error recognition result and the corresponding error type are output, and then different error correction methods are adopted based on different error types to ensure the accuracy and reliability of the data, and obtain the highly accurate circuit breaker data related to the actual operation situation, providing guarantee for the long-term stable operation of the power system.

[0148] Second Embodiment

[0149] Furthermore, in order to execute the error recognition and correction system for circuit breaker data corresponding to the above method embodiments to achieve the corresponding functions and technical effects, Figure 2 A structural diagram of an error recognition and correction system for circuit breaker data is provided. For the sake of convenience of description, only the parts related to this embodiment are shown. The error recognition and correction system for circuit breaker data provided by the embodiments of the present application includes:

[0150] The data sequence acquisition module 201 is configured to collect electrical data and equipment status data during the operation of the circuit breaker to obtain an original data sequence.

[0151] The feature extraction module 202 is configured to perform multi-scale decomposition in terms of frequency on the original data sequence through wavelet transform to extract the feature vector of the original data sequence.

[0152] In the embodiment of the present application, based on a preset decomposition level, the original data sequence is segmented in terms of frequency through wavelet transform, and the feature information of the original data sequence on each frequency segment is extracted; wherein, the decomposition level is determined by the complexity and fluctuation of the original data sequence;

[0153] According to the feature information, the statistical features of each frequency segment are extracted from the original data sequence; wherein, the statistical features include signal energy, data amplitude, and data deviation degree;

[0154] Through the statistical features and the feature information, a feature vector of the original data sequence is constructed.

[0155] The error identification module 203 is configured to perform error identification on the original data sequence through a preset error identification model based on the feature vector, and output an error identification result and a corresponding error type.

[0156] In the embodiment of the present application, the feature vector is input into the error identification model, and by calculating the matching degree between the feature vector and a preset error label, the error existing in the original data sequence is identified to generate an error identification result; wherein, the error label is formed by error annotation of historical circuit breaker data;

[0157] By identifying the error mode of the error identification result, the error type is determined.

[0158] The error correction module 204 is configured to select a correction algorithm corresponding to the error type to correct the error identification result to obtain a corrected data sequence.

[0159] In the embodiment of the present application, if the error type is the existence of outliers, the error identification result is corrected by the mean replacement method or the interpolation correction method;

[0160] If the error type is the existence of trend errors, the error identification result is corrected by the fitting prediction method;

[0161] When the corrected error recognition result meets the preset device characteristic constraints, a corrected data sequence is obtained through the corrected error recognition result; wherein, the device characteristic constraints are determined by the physical characteristics of the circuit breaker and the normal fluctuation range of the electrical data.

[0162] In some embodiments, the data sequence acquisition module 201 is specifically:

[0163] First, high-precision sensors with good anti-interference capabilities are used to collect electrical data and device status data of the circuit breaker in real time at a high sampling frequency. Among them, the electrical data includes current, voltage, and power data of the branch where the circuit breaker is located, and the device status data includes the switch status and action time of the circuit breaker during operation, etc.

[0164] During data acquisition, it is crucial to accurately capture rapidly changing electrical data. To ensure the accuracy and reliability of the data, it is essential to use high-precision sensors. In the embodiments of the present application, the set sampling frequency is not less than 1 kHz to ensure that rapid fluctuations in instantaneous current, voltage, and other parameters can be accurately captured. This is because a lower sampling frequency may cause important transient changes to be missed, thus affecting subsequent error recognition and correction. Therefore, by increasing the sampling frequency, the system can more finely capture the fluctuations of electrical data over time, especially in the case of sudden changes in current and voltage. Moreover, the accuracy requirement of the sensor reaches more than 0.1%, which means that the error of each collected data does not exceed 0.1%, thereby ensuring the high accuracy of the data and avoiding recognition errors caused by data errors.

[0165] In addition to high precision and high sampling frequency, the anti-interference ability of the sensor is also very important. During the operation of electrical equipment, electromagnetic interference may come from nearby high-voltage equipment or other electronic devices. This interference may affect the output signal of the sensor, resulting in large errors in the collected data. Therefore, selecting sensors with good anti-interference capabilities, especially sensors that can effectively shield external electromagnetic interference, can reduce the generation of errors. To further improve the reliability of the data, hardware filtering techniques and signal correction algorithms are usually used to process the collected data.

[0166] Exemplarily, a designed low-pass filter is used to filter out high-frequency noise in the collected data or the Kalman filtering algorithm is used to denoise the collected data, enhancing the accuracy and stability of the data.

[0167] Optionally, in the embodiments of the present application, a Butterworth filter is selected to implement data filtering. The Butterworth filter is also known as the Butterworth filter, and its characteristic is that the frequency response curve in the passband is maximally flat without ripples, while it gradually drops to zero in the stopband.

[0168] During the data acquisition process, all the acquired data will be stored according to the time series, forming the original data sequence. The original data sequence has the characteristics of high frequency and low error.

[0169] Specifically, assuming that at time point t, the electrical data such as current and voltage acquired are I(t) and V(t) respectively, then the corresponding original data sequence is {I(t1), I(t2),..., I(t n )} V(t1), V(t2),..., V(t n ).

[0170] Furthermore, the data acquired by the sensor needs to ensure its time synchronization and accuracy. For example, the sampling period must be kept constant, and the output of the sensor should pass through the clock synchronization mechanism to ensure that the data markers at each time point are consistent. In practical applications, the acquired time series data should use the following formula to represent the relationship between current and voltage data:

[0171] P(t) = V(t) × I(t);

[0172] In the formula, P(t) is the instantaneous power at time t, V(t) is the voltage at time t, and I(t) is the current at time t.

[0173] By calculating the instantaneous power, the working state of the circuit breaker can be monitored in real time, and further provide data support for the error identification model. This precise data acquisition can not only ensure high-quality data input, but also lay a foundation for subsequent data preprocessing, feature extraction and error correction steps, ensuring the accuracy of error identification.

[0174] After completing the data acquisition, preprocess the obtained original data sequence, successively perform outlier removal, data cleaning and data standardization, which are key operations to ensure data quality and provide a reliable basis for subsequent analysis. In the operating data of the circuit breaker, outliers usually refer to those points that significantly deviate from the normal data range, and these outliers may be caused by sensor failures, external interferences or other problems. In order to eliminate these outliers, a method based on statistical principles is adopted. Calculate the mean and standard deviation of the data points in the original data sequence, and then use the normal distribution assumption to filter the data according to the mean and standard deviation to complete outlier removal.

[0175] Specifically, let the mean of each data point in the original data sequence be μ and the standard deviation be σ, then the outliers can be determined by the following formula:

[0176] |x i- μ| > 3σ;

[0177] In the formula, x iis the value of a certain data point in the original data sequence. If x i exceeds the range of ±3 times the standard deviation from the mean, it is considered an outlier and removed from the original data sequence. In this way, the elimination of outliers helps improve the quality of the data and avoid their impact on subsequent analysis and model training.

[0178] In the data cleaning stage, the moving average method is used to remove high-frequency noise from the original data sequence. Since the operating data of the circuit breaker may be affected by factors such as short-term fluctuations, sensor noise, or electromagnetic interference, the moving average can effectively smooth the data and eliminate this high-frequency noise. The moving average filtering method performs smoothing by calculating the average value of the data within a fixed window around each data point. Specifically: Assume the original data sequence is {x1, x2,..., x n}, and the window size is w. Then the data sequence after moving average processing is Specifically as follows:

[0179]

[0180] where is the data point after moving average filtering. The moving window size w is a preset constant, usually taking an odd value. By performing data cleaning through the moving average, the time series data can be effectively smoothed, reducing the noise caused by instantaneous fluctuations.

[0181] Finally, the original data sequence is normalized. This is to map the data of different electrical parameters to the same dimension for subsequent feature extraction and model training. Due to the magnitude differences of different parameters, if not normalized, the error recognition model may be affected by some parameters with larger dimensions, resulting in deviations in the prediction results. Commonly used normalization methods include normalization, which generally maps the data to the interval [0, 1] or [-1, 1]. For a given parameter x i , the normalized value x' i can be calculated by the following formula:

[0182]

[0183] where x i is a certain data point in the original data sequence, and min(x) and max(x) are the maximum and minimum values in the original data sequence respectively. The normalized data x' i will be mapped to the interval [0, 1], thus eliminating the dimension differences and making subsequent feature extraction and error recognition model training more efficient and accurate. In this way, different electrical parameters (such as voltage, current, etc.) can be compared and analyzed on the same scale, providing unified input data for the error recognition model.

[0184] These preprocessing operations complement each other and jointly ensure the high quality and consistency of the data. After removing outliers, the moving average filter further smooths the noise in the data, and the final normalization process unifies the data of different features on the same scale, providing clear and standardized data input for the error identification model and effectively improving the accuracy of subsequent error identification.

[0185] In some embodiments, the feature extraction module 202 is specifically:

[0186] Wavelet transform is used to perform multi-scale decomposition on the original data sequence, aiming to extract the feature information of circuit breaker data in different frequency bands. Among them, wavelet transform is a transform analysis method. By providing a "time-frequency" window that changes with frequency, it conducts time-frequency analysis and processing of signals. Through stretching and translation operations, the signal is gradually refined at multiple scales. Finally, it achieves fine time division at high frequencies and fine frequency division at low frequencies, and can automatically adapt to the requirements of time-frequency signal analysis, so as to focus on any details of the signal to highlight the characteristics of the signal.

[0187] Optionally, the Daubechies wavelet basis function is selected as the transformation function in the embodiments of the present application. The Daubechies wavelet basis function is also known as the Daubechies wavelet and is mainly applied to discrete wavelet transforms. Because of its good localization properties, the Daubechies wavelet can provide both time and frequency resolutions and is suitable for processing signals containing transient changes and high-frequency noise. The Daubechies wavelet has good smoothness and finite support, which is particularly effective for dealing with common noise and mutation points in electrical data.

[0188] The decomposition level of wavelet transform is generally determined according to the complexity and fluctuation of the data, usually 3 to 5 levels. The more levels, the finer the decomposed frequency bands, and the more detailed signal features can be captured.

[0189] Exemplarily, if the decomposition level is set to 3 levels and the sampling frequency is assumed to be 1 kHz, the frequency bands can be divided into: low-frequency band 0 - 62.5 Hz, medium-low frequency band 62.5 - 125 Hz, medium-high frequency band 125 - 250 Hz, and high-frequency band 250 - 500 Hz.

[0190] According to the preset decomposition level, the original data sequence is segmented by frequency through the Daubechies wavelet basis function, and the wavelet coefficients of each frequency band are obtained as the characteristic information of the original data sequence. The wavelet coefficients reflect the multi-scale characteristics of the data in different frequency ranges. The low-frequency part can reflect the overall trend of the data, while the high-frequency part can reveal mutations, abnormal fluctuations, and short-term changes in the signal. Therefore, for the wavelet coefficients of each frequency band, the signal energy, data amplitude, and data deviation degree of each frequency band can be extracted as corresponding statistical features to supplement the characteristic information.

[0191] Signal energy is an important index to describe the signal intensity and can reflect the energy distribution of the signal in this frequency band. The data amplitude is generally calculated as the variance of the data. The statistical features can jointly form a complete feature vector with the wavelet coefficients, improve the accuracy of the error recognition model, and help the model capture more potential error patterns.

[0192] Specifically, for the wavelet coefficient sequence {c j1 ,c j2 ,...,c jn} of the j-th frequency band, the statistical features such as signal energy, data amplitude, and data deviation degree can be calculated through the following formulas:

[0193] The signal energy, the specific expression is:

[0194]

[0195] In the formula, E j is the total energy of the signal in the j-th frequency band, reflecting the signal intensity, n is the total number of the wavelet coefficients, and c ji is the i-th wavelet coefficient in the j-th frequency band;

[0196] The data amplitude, the specific expression is:

[0197]

[0198] In the formula, is the fluctuation degree of the wavelet coefficients in the j-th frequency band. A larger variance means that the signal fluctuates more in this frequency band, which may be related to abnormal changes; is the mean value of the wavelet coefficients in the j-th frequency band;

[0199] The data deviation degree, the specific expression is:

[0200]

[0201] In the formula, γ jRegarding the symmetry of the signal within the j-th frequency band, a large deviation of the data may indicate a large asymmetry in the signal, further providing a basis for error identification.

[0202] These statistical features are combined with the multi-level feature information composed of wavelet coefficients to construct a feature vector, enabling each data point in the original data sequence to provide information from different frequencies and statistical perspectives, enhancing the error identification model's ability to capture complex data patterns. These features can be provided as input data to the error identification model to improve the model's recognition accuracy for outliers, especially in electrical parameters where instantaneous current, voltage changes, etc. may occur. In this way, the feature vector not only contains the time-frequency information of the wavelet coefficients themselves but also combines the statistical properties of the signal, enabling the error identification model to analyze the data more comprehensively and ensuring higher accuracy and robustness.

[0203] In some embodiments, the error identification module 203 is specifically:

[0204] Then, the obtained feature vector and the original data sequence are input into the error identification model together, and the original data sequence is subjected to error identification through a preset error identification model.

[0205] The error identification model is jointly constructed by a support vector machine and a neural network. The support vector machine is a supervised learning method that classifies data by finding an optimal decision boundary and has strong generalization ability for high-dimensional data. To improve the performance of the support vector machine in circuit breaker data error identification, a radial basis function is used as the kernel function. The expression of the radial basis function is:

[0206]

[0207] In the formula, x and y are the feature vectors of the input data, and σ is the coefficient of the radial basis function, which determines the similarity between data points. By adjusting σ, the influence range between data points can be controlled. A smaller σ will result in a more local influence, while a larger σ will expand the influence range. Therefore, during the training process of the error identification model, the cross-validation method is used to optimize σ and the penalty factor C (the penalty factor determines the penalty degree for misclassification). Through the cross-validation method, the generalization ability of the model can be evaluated under different combinations of hyperparameters, so as to select the best parameters, avoid overfitting, and improve the accuracy of the model on the test set.

[0208] For the neural network algorithm, a multi-layer perceptron structure is selected. This is a feed-forward neural network with at least one hidden layer. The number of nodes in the input layer of the multi-layer perceptron structure is determined by the dimension of the feature vector. Usually, after feature extraction, the input data has a relatively high dimension. Therefore, the number of nodes in the input layer should be consistent with the dimension of the feature vector. The number of nodes in the middle hidden layer is usually determined by the trial-and-error method. A common approach is to start with a small number of nodes, gradually increase them, and evaluate the performance of the model. Generally, a hidden layer structure with 1 to 2 layers can effectively extract complex patterns in the data. Between each layer, the neural network non-linearly transforms the data through activation functions (such as ReLU or sigmoid) to enhance the learning ability of the model.

[0209] The neural network algorithm is trained using the backpropagation function. The core idea of the backpropagation function is to calculate the gradient of each weight and update the weights according to the gradient descent method to minimize the loss function. The loss function is usually selected as the mean squared error or cross-entropy, depending on the nature of the task. The weight update formula for the neural network is:

[0210]

[0211] where w ij is the weight from the i-th layer to the j-th layer, η is the learning rate, and E is the loss function. By continuously adjusting the weights, the neural network gradually learns the features of the data.

[0212] To ensure the stability of the neural network training process and prevent overfitting, the early stopping method is adopted. The early stopping method monitors the loss value on the validation set. If the validation set loss does not decrease over multiple training epochs, the training is stopped to avoid overfitting of the model on the training set and improve the generalization ability of the model. During the training process, parameters such as the learning rate and batch size are usually adjusted to accelerate convergence and ensure that the neural network can find the optimal solution in a relatively short time.

[0213] Through the above training strategies, both the support vector machine and the neural network can make full use of the feature information extracted from the wavelet transform for error identification. The support vector machine can handle complex non-linear classification problems through the radial basis kernel function, while the neural network automatically extracts complex features of the data through a multi-layer structure and uses backpropagation to optimize the network structure. Both combine the time-frequency features and statistical characteristics of the data, can effectively identify error patterns in the intelligent circuit breaker data, and provide accurate error type determination.

[0214] The error identification model is based on a large number of preset error labels, calculates the matching degree between the feature vector and the preset error labels, and identifies outliers and the trend change characteristics of low-frequency coefficients through the abnormal fluctuations of the high-frequency coefficients provided by the feature vector, and identifies the error identification results and the corresponding error types.

[0215] Among them, the error tags are generated from manual annotation or by combining physical models and historical circuit breaker data anomaly detection.

[0216] In some embodiments, the error correction module 204 is specifically:

[0217] The error types include the existence of outliers and the existence of trend errors. The judgment of outliers is identified by the abnormal fluctuation of the high-frequency coefficients provided by the eigenvectors, and the trend errors are obtained from the trend change characteristics of the low-frequency coefficients provided by the eigenvectors. In the data with trend errors, regular fluctuations will appear, such as long-term drift and periodic deviation. In the data with outliers, the outliers will appear in the form of discrete points.

[0218] In view of the characteristics of the above error data, the embodiments of the present application adopt different error correction methods to correct the error recognition results to ensure the accuracy and reliability of the data.

[0219] For the correction of outliers, the embodiments of the present application adopt the mean replacement method. According to the time series characteristics of the original data sequence, the mean value is calculated by selecting the data within a certain time window before and after the outlier, and this mean value is used as the replacement value to replace the outlier.

[0220] Specifically, assume that the original data sequence is {x1, x2,..., x n}}, if a certain data point x i is determined to be an outlier, then the data of w time steps before and after this point can be selected for mean calculation, as follows:

[0221]

[0222] In the formula, is the corrected value, w is the size of the time window, which determines the range of the front and back data used to calculate the mean. Through the mean replacement method, accidental noise and mutation values can be effectively eliminated, so that the corrected data maintains continuity and consistency.

[0223] In addition, in other embodiments, the interpolation correction method is used to correct the outliers in the original data sequence. The interpolation correction method generates reasonable estimated values by using the information of adjacent data points. Commonly used interpolation methods include cubic spline interpolation and Lagrange interpolation. The cubic spline interpolation method constructs a smooth polynomial function, making the function continuous at all nodes and having continuous derivatives, which can provide smoother and more accurate interpolation results. The expression of cubic spline interpolation is:

[0224] S(x) = a(x - x0) 3 + b(x - x0) 2 + c(x - x0) + d;

[0225] In the formula, a, b, c, and d are undetermined coefficients of the cubic spline interpolation formula, x0 is the abscissa of the interpolation point, and x is the point to be interpolated. By solving a set of equations, these coefficients can be determined, thus obtaining a smooth interpolation curve. Through cubic spline interpolation, interpolation is performed on the outliers in the original data sequence to obtain a smooth and corrected data sequence.

[0226] Lagrange interpolation rule constructs an interpolation polynomial using the values of all known data points in the original data sequence. The specific expression of the interpolation polynomial is:

[0227]

[0228] In the formula, y i is the value of the known data point, and x i is the abscissa of the known data point. Lagrange interpolation is suitable for interpolating data points in a small range, but for large data sets, the computational cost is relatively high. Whether it is cubic spline interpolation or Lagrange interpolation, they can effectively estimate reasonable interpolation results based on the surrounding data points, thereby correcting outliers.

[0229] For the correction of trend errors, the embodiment of the present application adopts the fitting prediction method. Trend errors usually manifest as regular fluctuations in the data. The future data values can be predicted by fitting the trend of historical data. Common fitting methods include polynomial fitting and exponential smoothing method. Polynomial fitting captures the trend of the data by fitting a high-degree polynomial function. It is assumed that the trend of the original data sequence can be represented by an m-degree polynomial, and the fitting function is:

[0230] f(x) = a m x m + a m-1 x m-1 + … + a1x + a0;

[0231] In the formula, a m , a m-1 , …, a0 are polynomial coefficients solved by methods such as the least squares method. Through this method, the future data values can be predicted, and these predicted values are used to replace the error data points. Therefore, in the embodiment of the present application, by performing fitting prediction on a segment of the data sequence before the error data point in the original data sequence, the predicted data values are obtained, and then the error data points are replaced with the predicted data values to obtain the corrected data sequence.

[0232] In addition, in other embodiments, the exponential smoothing method can also be used to correct the trend error. The exponential smoothing method is a prediction method based on weighted average. It smooths the data and predicts future values by assigning exponentially decaying weights to historical data. The basic formula of the exponential smoothing method is as follows:

[0233]

[0234] In the formula, is the predicted value, y t-1 is the actual value at the previous moment, is the predicted value at the previous moment, and α is the smoothing coefficient, with a value range of (0, 1). A larger α value gives more weight to recent historical data, while a smaller α value makes the influence of the long-term trend greater. Through this method, the long-term trend in the circuit breaker data can be effectively captured, and future data can be predicted, thereby correcting the anomalies caused by trend errors.

[0235] Therefore, the exponential smoothing method assigns exponentially decaying weights to the data before the error data points in the original data sequence to predict the corresponding future values, and uses the future values to correct the existing trend error data.

[0236] During the error identification process and data correction process, the physical meaning of the data and the actual operating conditions of the circuit breaker also need to be fully considered. For example, it is normal for parameters such as current and voltage to fluctuate within a certain range, but changes beyond a certain range may cause equipment damage. Therefore, the corrected data must conform to electrical laws and cannot deviate from the operating characteristics of the equipment. By introducing physical models and equipment characteristic constraints into the error identification method and correction method, the rationality and accuracy of the results can be further improved, ensuring that the data after error correction can truly reflect the operating state of the intelligent circuit breaker.

[0237] As an improvement to the above solution, a series of evaluations also need to be performed on the corrected data sequence to quantify the quality of the correction effect, so as to ensure that the corrected data has sufficient accuracy and reliability. The correction effect indicators used in the embodiments of this application include root mean square error, mean absolute error, and correlation coefficient, and these indicators can evaluate the quality of the data after correction from different perspectives.

[0238] The embodiments of this application mainly compare the historical data sequence of the circuit breaker with the corrected data sequence, calculate the corresponding root mean square error, mean absolute error, and correlation coefficient to evaluate the effect of this error correction. The historical data sequence is the true data of the circuit breaker obtained previously without errors.

[0239] Specifically, the root mean square error (RMSE) is one of the important indicators for evaluating data correction errors, which can reflect the difference between the corrected data and the actual data. The calculation formula for the root mean square error is as follows:

[0240]

[0241] In the formula, x i is the actual data value (referring to the data point in the historical data sequence in the embodiments of the present application), is the corrected data value, and n is the number of data points. The smaller the root mean square error, the closer the corrected data is to the actual data, and the better the correction effect. This indicator can intuitively reflect the overall error size of the data, especially suitable for measuring the uniformity of the differences between data points.

[0242] The mean absolute error (MAE) is another indicator for measuring the error size, which evaluates the correction effect by calculating the absolute value of the error of each data point and taking its average. The calculation formula for the mean absolute error is as follows:

[0243]

[0244] In the formula, x i is the actual data value (referring to the data point in the historical data sequence in the embodiments of the present application), is the corrected data value, and n is the number of data points. Different from the root mean square error, the mean absolute error pays more attention to the average level of errors and assigns the same weight to large and small errors. The smaller the mean absolute error, the more accurate the correction result, especially suitable for scenarios with fewer outliers and higher error tolerance. Through these two indicators, the difference between the corrected data and the actual data can be quantified, so as to determine the effect of the error correction method.

[0245] The correlation coefficient is used to measure the linear correlation degree between the corrected data and the actual data. The calculation formula for the correlation coefficient is as follows:

[0246]

[0247] In the formula, a and b are the means of the actual data and the corrected data respectively. The value range of the correlation coefficient is from -1 to 1. When the correlation coefficient approaches 1, it indicates that there is a strong positive correlation between the corrected data and the actual data, and the correction effect is good; when the correlation coefficient approaches -1, it indicates that there is a negative correlation between the corrected data and the actual data, and the correction effect is poor; when the correlation coefficient is close to 0, it indicates that there is no linear relationship between the two.

[0248] To ensure the effectiveness of the error correction method, the aforementioned correction performance indicators must meet preset thresholds. For example, if the root mean square error (RMS) exceeds a certain set value or the correlation coefficient falls below a set value, the correction effect is considered unsatisfactory. In this case, the system automatically returns to the error identification model training step, reselecting an appropriate model algorithm, adjusting model parameters, or increasing the amount of training data to ensure the model can better identify and correct errors in the data. During this process, the existing dataset may need to be expanded to include more labeled error data to enhance the model's generalization and accuracy. Furthermore, regular monitoring and evaluation of the correction effect is crucial to adapt to changes in the operating state of the smart circuit breaker and drift in data characteristics. As the operating state of the device changes, data characteristics may also drift. Therefore, the correction method needs to be continuously adjusted to ensure that it can effectively address emerging error patterns and ensure long-term stable system operation. This dynamic correction and evaluation mechanism can continuously improve the accuracy and adaptability of the error correction method, ensuring that the corrected data always meets the actual operating requirements of the smart circuit breaker.

[0249] The threshold requirement is related to the error correction accuracy requirement, the allowable error range of electrical parameters for safe operation of the equipment, and industry standards and specifications.

[0250] The implementation of the embodiments of the present application has the following beneficial effects:

[0251] The embodiment of the present application first collects the data on the change of electrical parameters of the circuit breaker during operation and the equipment status data generated by the operation in units of time to obtain the original data sequence of the error to be corrected. Then, the wavelet transform is used to decompose the original data sequence into different frequency segments, and the characteristic information of the circuit breaker data in different frequency segments is extracted, so that the data points in the original data sequence provide information at different frequencies and statistical angles, which can improve the accuracy of error identification in the subsequent error identification process and help the model capture more potential error content. After error identification, the error identification result and the corresponding error type are output, and then different error correction methods are used based on different error types to ensure the accuracy and reliability of the data, and obtain the corrected high-accuracy circuit breaker data related to the actual operating conditions, providing a guarantee for the long-term stable operation of the power system.

[0252] The specific embodiments described above further illustrate the purpose, technical solutions, and beneficial effects of this application. It should be understood that the above description is merely a specific embodiment of this application and is not intended to limit the scope of protection of this application. In particular, it should be noted that for those skilled in the art, any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of this application should be included in the scope of protection of this application.

Claims

1. A method for error identification and correction of circuit breaker data, characterized in that, Including: Collecting the electrical data and equipment status data of the circuit breaker during operation to obtain an original data sequence; Performing multi-scale decomposition in frequency on the original data sequence through wavelet transform to extract the feature vector of the original data sequence; Based on the feature vector, performing error identification on the original data sequence through a preset error identification model, and outputting an error identification result and the corresponding error type; Selecting a correction algorithm corresponding to the error type to correct the error identification result to obtain a corrected data sequence.

2. The error identification and correction method for circuit breaker data according to claim 1, characterized in that The performing multi-scale decomposition in frequency on the original data sequence through wavelet transform to extract the feature vector of the original data sequence is specifically: Based on a preset decomposition level, performing frequency segmentation on the original data sequence through wavelet transform, and extracting the feature information of the original data sequence in each frequency segment; wherein, the decomposition level is determined by the complexity and fluctuation of the original data sequence; According to the feature information, extracting the statistical features of each frequency segment from the original data sequence; wherein, the statistical features include signal energy, data amplitude, and data deviation degree; Constructing the feature vector of the original data sequence through the statistical features and the feature information.

3. The error identification and correction method for circuit breaker data according to claim 2, characterized in that, The statistical features are specifically: The signal energy, the specific expression is: where E j is the total energy of the signal in the j-th frequency band, n is the total number of the feature information, and c ji is the i-th feature information in the j-th frequency band; The data amplitude, the specific expression is: In the formula, is the degree of fluctuation of the feature information in the j-th frequency band, is the mean value of the feature information in the j-th frequency band; The data deviation degree, the specific expression is: where γ j is the symmetry of the signal within the j-th frequency band.

4. The error identification and correction method for circuit breaker data according to claim 2, characterized in that, Before performing multi-scale decomposition in frequency on the original data sequence through wavelet transform, it further includes: Successively performing outlier removal, data cleaning, and data standardization on the original data sequence to obtain a preprocessed original data sequence; Wherein, the outlier removal is to delete the data in the original data sequence that deviates from the preset normal range; the data cleaning is to delete the data in the original data sequence that exceeds the preset frequency; the data standardization is to map different electrical parameters in the original data sequence to a preset same standard.

5. The error identification and correction method for circuit breaker data according to claim 1, characterized in that, The performing error identification on the original data sequence through a preset error identification model based on the feature vector and outputting an error identification result and the corresponding error type is specifically: Inputting the feature vector into the error identification model, and identifying the errors existing in the original data sequence by calculating the matching degree between the feature vector and a preset error label to generate an error identification result; wherein, the error label is formed by error annotation of historical circuit breaker data; Determining the error type by identifying the error mode of the error identification result.

6. The method for error identification and correction of circuit breaker data according to claim 1, wherein The selecting a correction algorithm corresponding to the error type to correct the error identification result to obtain a corrected data sequence is specifically: If the error type is the existence of outliers, correcting the error identification result through the mean replacement method or the interpolation correction method; If the error type is the existence of trend errors, correcting the error identification result through the fitting prediction method; When the corrected error identification result meets the preset equipment characteristic constraints, obtaining the corrected data sequence through the corrected error identification result; wherein, the equipment characteristic constraints are determined by the physical characteristics of the circuit breaker and the normal fluctuation range of the electrical data.

7. The method for error identification and correction of circuit breaker data according to claim 6, characterized in that, The error recognition result is corrected by the fitting prediction method, specifically as follows: Select the first data sequence before the error data point in the error recognition result, and perform trend fitting on the first data sequence to obtain the predicted data value; Use the predicted data value to replace the error data point in the error recognition result to obtain the corrected error recognition result.

8. The method for error identification and correction of circuit breaker data according to any one of claims 1 to 7, characterized in that After obtaining the corrected data sequence, it further includes: Compare the corrected data sequence with the obtained historical data sequence, and quantify the effect of this error correction by calculating the correction effect index; wherein, the correction effect index includes the root mean square error, mean absolute error, and correlation coefficient between the corrected data sequence and the historical data sequence.

9. The method for error identification and correction of circuit breaker data according to claim 1, wherein The electrical data and equipment status data of the circuit breaker during operation are collected to obtain the original data sequence, specifically as follows: Collect the electrical data and equipment status data of the circuit breaker during operation at a preset sampling frequency, and simultaneously process the collected data using hardware filtering technology and signal correction algorithms; During the data collection process, store the collected data in the form of a time series to form the original data sequence.

10. An error identification and correction system for circuit breaker data, characterized in that It includes: A data sequence acquisition module, a feature extraction module, an error recognition module, and an error correction module; Among them, the data sequence acquisition module is used to collect the electrical data and equipment status data of the circuit breaker during operation to obtain the original data sequence; The feature extraction module is used to perform multi-scale decomposition in frequency on the original data sequence through wavelet transform to extract the feature vector of the original data sequence; The error recognition module is used to perform error recognition on the original data sequence based on the feature vector through a preset error recognition model, and output the error recognition result and the corresponding error type; The error correction module is used to select the correction algorithm corresponding to the error type to correct the error recognition result to obtain the corrected data sequence.

Images