Power equipment state detection method and system based on artificial intelligence

Abstract

The invention relates to the technical field of data processing, in particular to a power equipment state detection method and system based on artificial intelligence, and the method comprises the steps: dividing a monitoring data sequence of target power equipment under a target monitoring index into at least two periodic sequences according to a preset main period, and combining the periodic sequences into a periodic phase matrix; according to data distribution in the periodic phase matrix and the dominant change direction between the row vectors, obtaining a periodic feature sequence of a preset main period; according to the rank of the remainder matrix of each data in the periodic phase matrix and the difference between the periodic feature sequence and each row vector in the periodic phase matrix, obtaining a weighted preliminary seasonal factor of each column vector in the periodic phase matrix, and further obtaining a seasonal component by using an STL decomposition algorithm; and the trend component and the residual component in the monitoring data sequence are extracted according to the seasonal component and are used for detecting the state of the target power equipment, so that the accuracy of the detection result of the state of the power equipment is improved.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a method and system for detecting the condition of power equipment based on artificial intelligence. Background Technology

[0002] Condition monitoring of power equipment (such as transformers, switchgear, and lines) is a crucial step in ensuring the safe, reliable, and efficient operation of power systems. Power equipment condition data is typically non-stationary, exhibiting complex fluctuations. Therefore, Standard Trajectory Layer (STL) decomposition is commonly used to break down the acquired data into three components: trend, seasonal, and residual, to better understand and analyze the equipment's operating status. Among the three components extracted by STL, the seasonal component is the core description of the "normal operating pattern" and serves as the "background" for constructing power equipment condition monitoring. Its extraction quality directly determines the reliability of the trend and residual component extractions, and is a prerequisite for the reliability of power equipment condition monitoring.

[0003] For the extraction of seasonal components, the traditional method divides the time series data sequence of power equipment status into several subsequences according to a fixed period. Each subsequence corresponds to a phase. Loess smoothing is performed on each subsequence. Finally, the mean or center point prediction value of each smoothed subsequence is used as the preliminary seasonal factor for each phase. Then, based on the preliminary seasonal factor, the principal periodic component of the time series data sequence of power equipment status is obtained. Subsequently, seasonal components, trend components, and residual components are gradually extracted to monitor the operating status of power equipment based on the trend components and residual components.

[0004] However, the load patterns of power equipment are significantly affected by user electricity consumption behavior, such as weekdays versus weekends, holidays, sudden temperature changes, and switching of production conditions. This can cause the data to exhibit a mixed characteristic of multiple periodic components. The seasonal components extracted based on a fixed period may be affected by secondary periods. That is, the data in the obtained subsequences may not represent the same periodic component. When the mean or center point prediction value of each Loess-smoothed subsequence is used as the preliminary seasonal factor for the corresponding phase, the main periodic features may be masked or distorted by secondary periodic components. This results in the obtained preliminary seasonal factors failing to accurately reflect the typical pattern of the main period, thus blurring the main periodic features. In other words, the main periodic components generated by the preliminary seasonal factors are inaccurate, which in turn leads to inaccurate seasonal components. For example, abnormal signals such as equipment start-up and shutdown and inrush current may be incorrectly absorbed into the seasonal components, thus being weakened or lost in the residuals, ultimately causing the detection results of the power equipment status to be distorted.

[0005] Therefore, improving the accuracy of extracting seasonal components from power equipment status data, and thus improving the accuracy of power equipment status detection results, has become an urgent problem to be solved. Summary of the Invention

[0006] In view of this, embodiments of the present invention provide an artificial intelligence-based method and system for detecting the condition of power equipment, in order to solve the problem of how to improve the accuracy of extracting seasonal components from the condition data of power equipment, thereby improving the accuracy of the detection results of the condition of power equipment.

[0007] In a first aspect, embodiments of the present invention provide a power equipment status detection method based on artificial intelligence, the method comprising the following steps: The monitoring data of the target power equipment under the target monitoring indicators are obtained in the current time period. All monitoring data are centralized to obtain the monitoring data sequence. In the process of decomposing the monitoring data sequence using the STL decomposition algorithm, the monitoring data sequence is divided into at least two periodic sequences according to the preset main period. Each periodic sequence is used as a row vector and combined into a periodic phase matrix. According to the data distribution in the periodic phase matrix and the dominant change direction between each row vector, the periodic feature sequence of the preset main period is obtained. Based on the rank of the cofactor matrix of each data point in the periodic phase matrix, the contribution of each data point in the periodic phase matrix to the periodic phase matrix is ​​obtained. Based on the difference between the periodic feature sequence and each row vector in the periodic phase matrix, the periodic matching degree between each data point in the periodic phase matrix and the preset main period is obtained. Based on the contribution and period matching degree of each data point in the periodic phase matrix, the weighted preliminary seasonal factor of each column vector in the periodic phase matrix is ​​obtained. Based on the weighted preliminary seasonal factor of each column vector, the seasonal component in the monitoring data sequence is obtained using the STL decomposition algorithm. Based on the seasonal component, the trend component and residual component in the monitoring data sequence are extracted for use in detecting the status of the target power equipment.

[0008] Preferably, obtaining the periodic feature sequence of the preset main period based on the data distribution in the periodic phase matrix and the dominant change direction between each row vector includes: Cluster all the data in the periodic phase matrix to obtain the target cluster with the largest number of members. Form a sub-matrix by combining the row vectors corresponding to the data in the target cluster. Calculate the average value of all data in each column vector of the sub-matrix. Form a first representative sequence of the preset main period by combining the average values ​​corresponding to each column vector of the sub-matrix. Principal component analysis is performed on the row vectors in the periodic phase matrix to obtain the first principal component vector, which is denoted as the second representative sequence of the preset principal period. Based on each data position in the first representative sequence and the second representative sequence, the two data with the same position are added together to obtain the periodic feature sequence of the preset main period.

[0009] Preferably, obtaining the contribution of each data point in the periodic phase matrix to the periodic phase matrix based on the rank of the cofactor matrix of each data point in the periodic phase matrix includes: For any data in the periodic phase matrix, the rank of the cofactor matrix of any data is subtracted from the rank of the periodic phase matrix to obtain the rank difference between the cofactor matrix of any data and the periodic phase matrix. The rank difference is then linearly normalized to obtain the first importance of any data. Based on the data in the column vector where any data is located in the periodic phase matrix, calculate the deviation from the mean of any data, perform linear normalization on the deviation from the mean to obtain the degree of deviation between any data and its column vector, and subtract the degree of deviation from the constant 1 to obtain the second importance of any data. Calculate the average between the first importance level and the second importance level to obtain the contribution of any data to the periodic phase matrix.

[0010] Preferably, the step of obtaining the periodic matching degree between each data point in the periodic phase matrix and the preset main period based on the difference between the periodic feature sequence and each row vector in the periodic phase matrix includes: For any data in any row vector of the periodic phase matrix, based on the position of the data in the row vector, obtain data with the same position as the data in the periodic feature sequence, and record it as reference data. Calculate the absolute value of the difference between the data and the reference data to obtain the principal periodic component offset of the data. Calculate the principal periodic component offset of each data in any row vector, sum the principal periodic component offsets of each data in any row vector to obtain the total principal periodic component offset of any row vector, calculate the proportion of the principal periodic component offset of any data in the total principal periodic component offset to obtain the degree of principal periodic component offset of any data, and subtract the degree of principal periodic component offset from the constant 1 to obtain the period matching degree between any data and the preset principal period.

[0011] Preferably, the step of obtaining the weighted preliminary seasonal factor for each column vector in the periodic phase matrix based on the contribution degree and periodic matching degree of each data point in the periodic phase matrix includes: For any data in any column vector of the periodic phase matrix, calculate the sum between the contribution degree of the data and the periodic matching degree to obtain the comprehensive weight of the data. Calculate the comprehensive weight of each data point in any column vector, and calculate the proportion of the comprehensive weight of any data point in the sum of the comprehensive weights of all data points in any column vector to obtain the preliminary seasonal factor weight of any data point. Loess smoothing is performed on any column vector to obtain smoothed data for each data point in the column vector. Based on the preliminary seasonal factor weights of each data point in the column vector, the smoothed data for each data point in the column vector are weighted and summed to obtain the weighted preliminary seasonal factor of the column vector.

[0012] Secondly, embodiments of the present invention also provide an artificial intelligence-based power equipment condition detection system, including a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the computer program, it implements an artificial intelligence-based power equipment condition detection method as described in the first aspect.

[0013] The beneficial effects of the embodiments of the present invention compared with the prior art are as follows: This invention acquires monitoring data of target power equipment under target monitoring indicators within the current time period. All monitoring data is centrally processed to obtain a monitoring data sequence. During the decomposition of the monitoring data sequence using the STL decomposition algorithm, the monitoring data sequence is divided into at least two periodic sequences according to a preset principal period. Each periodic sequence is used as a row vector and combined to form a periodic phase matrix. Based on the data distribution in the periodic phase matrix and the dominant change direction between each row vector, the periodic feature sequence of the preset principal period is obtained. The contribution of each data point in the periodic phase matrix to the periodic phase matrix is ​​obtained based on the rank of the cofactor matrix of each data point. The periodic matching degree between each data point in the periodic phase matrix and the preset principal period is obtained based on the difference between the periodic feature sequence and each row vector in the periodic phase matrix. Based on the contribution degree and periodic matching degree of each data point in the periodic phase matrix, a weighted preliminary seasonal factor is obtained for each column vector in the periodic phase matrix. Based on the weighted preliminary seasonal factor of each column vector, the seasonal component in the monitoring data sequence is obtained using the STL decomposition algorithm. The trend component and residual component in the monitoring data sequence are extracted based on the seasonal component and used for detecting the status of the target power equipment. The process involves constructing a periodic phase matrix based on the monitoring data sequence of the target power equipment during the current time period and a preset main period. The contribution of each data point in the periodic phase matrix to the matrix, along with the periodic matching degree of each data point, comprehensively reflects the degree of interference from secondary periods, i.e., the reliability of each data point in the periodic phase matrix as a periodic component of the preset main period. This reliability is used to assign weights to each data point in each column vector of the periodic phase matrix, obtaining a preliminary weighted seasonal factor for each column vector. Then, the STL decomposition algorithm is used to extract the seasonal components from the monitoring data sequence, making the extracted seasonal components closer to the true periodic pattern. Based on the seasonal components, trend components and residual components are gradually extracted, enabling the extracted trend components and residual components to more accurately reflect the operating status of the target power equipment and improve the accuracy of condition monitoring of the power equipment. Attached Figure Description

[0014] To more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0015] Figure 1 This is a flowchart of a power equipment condition detection method based on artificial intelligence, provided in Embodiment 1 of the present invention. Detailed Implementation

[0016] Embodiments of this disclosure are described in detail below, examples of which are illustrated in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain this disclosure, and should not be construed as limiting this disclosure.

[0017] It should be noted that the terms "first," "second," etc., used in this disclosure and the accompanying drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this disclosure described herein can be implemented in orders other than those illustrated or described herein. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this disclosure. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this disclosure.

[0018] To illustrate the technical solution of the present invention, specific embodiments are described below.

[0019] See Figure 1 This is a flowchart of a power equipment condition detection method based on artificial intelligence provided in Embodiment 1 of the present invention. Figure 1 As shown, the method may include: Step S101: Obtain the monitoring data of the target power equipment under the target monitoring indicators in the current time period, centralize all monitoring data to obtain the monitoring data sequence, and in the process of decomposing the monitoring data sequence using the STL decomposition algorithm, divide the monitoring data sequence into at least two periodic sequences according to the preset main period, and use each periodic sequence as a row vector to form a periodic phase matrix. According to the data distribution in the periodic phase matrix and the dominant change direction between each row vector, obtain the periodic feature sequence of the preset main period.

[0020] Typically, monitoring indicators such as current, voltage, harmonics, and power quality are used to detect the status of power equipment. Since these indicators are correlated, any one can be chosen to monitor the equipment. In this embodiment of the invention, any power equipment to be monitored is designated as the target power equipment, current is used as the target monitoring indicator, and the 14 days prior to the current moment are designated as the current time period. Current sensors are installed at the substation or equipment site, and the sampling frequency is set to once per minute. The monitoring data of the target power equipment under the target monitoring indicator is acquired minute by minute within the current time period to detect the operating status of the target power equipment during the current time period. There are no restrictions here; the implementer can set the duration of the current time period and the sampling frequency according to the specific scenario. To remove dimensions and facilitate subsequent data analysis, the minute-by-minute monitoring data of the target power equipment under the target monitoring indicator within the current time period is centrally processed to obtain a monitoring data sequence. Centralization processing is existing technology and will not be elaborated here.

[0021] Traditionally, the STL decomposition algorithm is used to decompose monitoring data sequences into three components: seasonal, trend, and residual. The trend and residual components are then used to analyze the operating status of target power equipment. Among the three components extracted by the STL decomposition algorithm, the seasonal component is the core description of the "normal operating pattern," and its extraction quality directly determines the reliability of the trend and residual component extraction, which is a prerequisite for ensuring the reliability of the target power equipment's condition monitoring results. For the extraction of the seasonal component, the STL decomposition algorithm divides the monitoring data sequence into several subsequences according to a fixed period, with each subsequence corresponding to a phase. Each subsequence is then smoothed using Loess. Finally, the mean or predicted center point value of each smoothed subsequence is used as the preliminary seasonal factor for each phase. Based on the preliminary seasonal factor, the principal periodic component of the monitoring data sequence is obtained, and the trend and residual components are gradually extracted to detect the operating status of the target power equipment.

[0022] However, the load patterns of target power equipment are significantly affected by user electricity consumption behavior, such as weekdays versus weekends, holidays, sudden temperature changes, and switching of production conditions. This can cause the data to exhibit a mixed characteristic of multiple periodic components. Consequently, the seasonal components extracted based on a fixed period may be affected by secondary periods. That is, the data in each subsequence may not represent the same periodic component. When the mean or center point prediction value of each Loess-smoothed subsequence is used as the preliminary seasonal factor for the corresponding phase, the main periodic features may be masked or distorted by secondary periodic components. This results in the obtained preliminary seasonal factors failing to accurately reflect the typical pattern of the main period, thus blurring the main periodic features. In other words, the main periodic components generated by the preliminary seasonal factors are inaccurate, which in turn leads to inaccurate seasonal components. For example, abnormal signals such as equipment start-up and shutdown and inrush current may be incorrectly absorbed into the seasonal components, thus being weakened or lost in the residuals, ultimately causing distortion in the detection results of the power equipment status.

[0023] Therefore, in this embodiment of the invention, by analyzing the degree of secondary periodic interference in the data of the subsequence corresponding to each phase, a weight is assigned to each data in the subsequence corresponding to each phase to obtain the weighted preliminary seasonal factor of each phase, thereby improving the accuracy of extracting seasonal components and making the extracted seasonal components closer to the real periodic pattern, thereby improving the accuracy of extracting trend components and residual components and enhancing the accuracy of condition detection of power equipment.

[0024] In the process of decomposing the monitoring data sequence using the STL decomposition algorithm, the preset main period is set to 24 hours, which is 1440 data points. The Loess smoothing parameter uses a relative ratio and is set to 0.6. There are no restrictions here. Implementers can set the various parameters in the STL algorithm according to the scenario.

[0025] According to a preset main period, the monitoring data sequence is divided into at least two periodic sequences. Since the preset main period in this embodiment is 24 hours, and the monitoring data sequence contains data from the target power equipment under the target monitoring index for the 14 days prior to the current moment, the number of periodic sequences is 14. Each periodic sequence is used as a row vector to form a periodic phase matrix. In the periodic phase matrix, each column vector represents a phase.

[0026] The monitoring data sequence itself exhibits obvious periodicity due to users' electricity consumption behavior, but may be subject to interference such as secondary periodic components, making it impossible to directly obtain the typical value of each phase within the preset main period. Therefore, in this embodiment of the invention, based on the data distribution in the periodic phase matrix and the dominant change direction between each row vector, the periodic feature sequence of the preset main period is obtained. Each data in the periodic feature sequence represents an approximate representation of each phase within the preset main period, that is, an approximate representation of the periodic components of the preset main period. Based on the periodic feature sequence, the degree to which each data in each column vector of the periodic phase matrix is ​​affected by secondary periodic interference can be analyzed.

[0027] The specific method for obtaining the periodic feature sequence is as follows: The DBSCAN clustering algorithm is used to cluster all data in the periodic phase matrix to filter out the influence of extreme data such as inrush current and equipment start-up and shutdown in each periodic sequence, and obtain the target cluster with the largest number of members. It is believed that the data in the target cluster best reflects the periodic components of the preset main period. Furthermore, the row vectors corresponding to the data in the target cluster are used to form a sub-matrix, and the average value of all data in each column vector of the sub-matrix is ​​calculated. The average value corresponding to each column vector of the sub-matrix is ​​used to form the first representative sequence of the preset main period. Principal component analysis (PCA) is performed on the row vectors in the periodic phase matrix to obtain the first principal component vector. The general steps are as follows: (1) Calculate the covariance matrix of the periodic phase matrix, (2) Obtain all eigenvalues ​​and eigenvectors of the covariance matrix, (3) Obtain the largest eigenvalue of the covariance matrix and its corresponding eigenvector. The eigenvector corresponding to the largest eigenvalue is the first principal component vector, which is used to reflect the dominant change direction between different periodic sequences. The first principal component vector is used as the second representative sequence of the preset main period. It is worth noting that the length of the first principal component vector is determined by the number of features. In this embodiment of the invention, principal component analysis is performed on the row vectors in the periodic phase matrix, so each column vector in the periodic phase matrix is ​​a feature. That is, the length of the first principal component vector is equal to the number of column vectors in the periodic phase matrix. In other words, the number of data in the second representative sequence is equal to the number of column vectors in the periodic phase matrix. Based on each data position in the first and second representative sequences, two data points with the same position are added together to obtain a periodic feature sequence with a preset principal period, denoted as Y. The DBSCAN clustering algorithm and principal component analysis are existing technologies and will not be described in detail here.

[0028] Step S102: Based on the rank of the cofactor matrix of each data in the periodic phase matrix, obtain the contribution degree of each data in the periodic phase matrix to the periodic phase matrix; based on the difference between the periodic feature sequence and each row vector in the periodic phase matrix, obtain the periodic matching degree between each data in the periodic phase matrix and the preset main period.

[0029] Based on the periodic feature sequence obtained in step S101, the degree of interference of each data in the periodic phase matrix by the secondary period is analyzed, that is, the credibility of each data in the periodic phase matrix as a periodic component of the preset main period. This is used to assign weights to each data in the periodic phase matrix, and obtain the weighted preliminary seasonal factor of each phase (that is, each column vector in the periodic phase matrix), thereby improving the accuracy of seasonal component extraction and thus improving the accuracy of condition detection of power equipment.

[0030] When a data point belongs to the main periodic component, it carries the typical pattern of the main period within the periodic phase matrix. That is, this data point maintains a high degree of consistency with other data points of the same phase, reflecting the main form of the periodic pattern. Conversely, if this data point belongs to a secondary period or is an outlier, it deviates from the main periodic pattern and contributes little or nothing to the overall structure of the periodic pattern. Taking the data in the i-th row and j-th column of the periodic phase matrix as an example, let's denote the data in the i-th row and j-th column as... ,if Data belonging to the periodic components of the preset main period. If the row and column vectors occupy a large projection along the principal direction of the periodic phase matrix, then... The cofactor matrix (removed from the periodic phase matrix) The rank of the matrix formed after the row and column it is in will be significantly lower than the rank of the periodic phase matrix, which indicates that... It is indispensable in linear structures, that is Its contribution to the periodic components of the preset main period is relatively large, therefore, according to the periodic phase matrix... The rank of the cofactor matrix is ​​obtained. The degree of contribution to the periodic phase matrix is ​​used to reflect... To assess the reliability of the periodic components of the preset main period, specifically: Subtract the rank of the periodic phase matrix The rank of the cofactor matrix is ​​obtained. The rank difference between the cofactor matrix and the periodic phase matrix is ​​linearly normalized to obtain... The first degree of importance, of which linear normalization is the existing technology and will not be elaborated here; According to the periodic phase matrix Calculate the data in the column vector. The deviation from the mean is then linearly normalized to obtain... The degree of deviation between it and its column vector is obtained by subtracting the degree of deviation from the constant 1. The second most important; Calculate the mean between the first importance level and the second importance level to obtain The degree of contribution to the periodic phase matrix.

[0031] In one implementation, The formula for calculating the contribution of the periodic phase matrix is ​​as follows: ; in, express The degree of contribution to the periodic phase matrix, where R represents the rank of the periodic phase matrix. express The rank of the cofactor matrix is ​​given by n, which represents the total number of data points in the monitored data sequence, i.e., the total number of data points in the periodic phase matrix, and s represents the total number of data points in each periodic sequence, i.e., the total number of data points in each row vector of the periodic phase matrix, i.e., the total number of column vectors in the periodic phase matrix. This represents the number of all periodic sequences, which is also the number of row vectors in the periodic phase matrix. This represents the data in the i-th row and j-th column of the periodic phase matrix. This represents the data in the u-th row and j-th column of the periodic phase matrix. This represents the linear normalization function.

[0032] It should be noted that, The larger, the more it means The rank of the cofactor matrix is ​​significantly lower than that of the periodic phase matrix. The more significant the structural influence on the periodic phase matrix, the more... The larger, The greater the contribution of the periodic component in the preset main period, The greater the likelihood that a component belongs to the preset main cycle; The smaller, the better. The smaller the difference between it and other data in its column vector, the better. The smaller the difference between the data in phase, The greater the probability that a periodic component belongs to a pre-defined main period, the more likely it is to be true. The larger, The greater the contribution of the periodic components in the preset main period, the greater the contribution.

[0033] The above steps obtained The degree of contribution to the periodic phase matrix is ​​based on The importance of the periodic phase matrix as a whole needs to be determined based on... The difference between the periodic feature sequence of the preset main period obtained in step S101 and the actual periodic feature sequence is obtained. The degree of period matching between the current period and the preset master period, to reflect... The degree of interference from secondary cycles, and then combined with The degree of contribution and the degree of cycle matching comprehensively reflect To assess the reliability of the periodic components of the preset master period.

[0034] Then obtain The specific steps for matching the cycle degree with the preset master cycle are as follows: according to Given its position in the row vector (the i-th row in the periodic phase matrix), obtain the corresponding value in the periodic feature sequence. Data with the same position, i.e., the j-th data in the periodic feature sequence, is denoted as the reference data, and the calculation is performed. The absolute value of the difference between the data and the reference data is obtained. The principal periodic component offset; Calculate separately The offset of the principal periodic component of each data point in the row vector (the i-th row in the periodic phase matrix) will be... The principal periodic component offsets of each data point in the row vector are summed to obtain... Calculate the total offset of the principal periodic components of the row vector. The proportion of the principal periodic component offset in the total principal periodic component offset is obtained. The degree of shift of the principal periodic component is calculated by subtracting the degree of shift of the principal periodic component from the constant 1, resulting in... The degree of period matching with the preset master period.

[0035] In one implementation, The formula for calculating the period matching degree with the preset master period is: ; in, express The degree of period matching with the preset master cycle, This represents the data in the i-th row and j-th column of the periodic phase matrix. This represents the j-th data point in a periodic feature sequence. This represents the data in the i-th row and v-th column of the periodic phase matrix. This represents the v-th data point in the periodic feature sequence, and s represents the number of column vectors in the periodic phase matrix. Represents the absolute value symbol.

[0036] It should be noted that, since each data point in the periodic feature sequence represents an approximate representation of each phase of the preset main period, i.e., the periodic component of the preset main period, therefore The smaller, the better. The smaller the difference between the approximate representation of its corresponding phase, the better. The less it is affected by secondary periodic interference, that is... The more pronounced the cyclical characteristics of the preset master cycle, the better. The larger, The greater the likelihood that a component belongs to a pre-defined main cycle, the more likely it is to be a cyclical component.

[0037] Thus, we have obtained Similarly, the contribution level and period matching degree of each data in the periodic phase matrix are obtained to reflect the credibility of each data in the periodic phase matrix as a periodic component of the preset main period. Then, each data in the periodic phase matrix is ​​assigned a weight to obtain the weighted preliminary seasonal factor of each phase (that is, each column vector in the periodic phase matrix), thereby improving the accuracy of seasonal component extraction.

[0038] Accuracy of extraction.

[0039] Step S103: Based on the contribution degree and period matching degree of each data in the periodic phase matrix, obtain the weighted preliminary seasonal factor of each column vector in the periodic phase matrix. Based on the weighted preliminary seasonal factor of each column vector, use the STL decomposition algorithm to obtain the seasonal component in the monitoring data sequence. Based on the seasonal component, extract the trend component and residual component in the monitoring data sequence for use in detecting the status of the target power equipment.

[0040] After obtaining the contribution level and period matching degree of each data point in the periodic phase matrix through step S102, the preliminary seasonal factor weights of each data point in the periodic phase matrix are obtained, specifically: Using the data in the i-th row and j-th column of the periodic phase matrix For example, calculate The sum of the contribution level and the periodic matching degree yields the following: The overall weight; Calculate separately The comprehensive weight of each data point in the column vector (the j-th column in the periodic phase matrix) is calculated. The overall weight is the proportion of the cumulative overall weight of all data in column j, which is obtained as a whole. Preliminary seasonal factor weights.

[0041] In one implementation, The formula for calculating the preliminary seasonal factor weights is as follows: ; in, express Preliminary seasonal factor weights for (data in the i-th row and j-th column of the periodic phase matrix). express The degree of contribution to the periodic phase matrix, express The periodicity matching degree with the preset master period, where n represents the total number of data in the periodic phase matrix, and s represents the number of column vectors in the periodic phase matrix. This represents the number of row vectors in the periodic phase matrix. express The degree of contribution of (the data in the u-th row and j-th column of the periodic phase matrix) to the periodic phase matrix. express The period matching degree between the data in the u-th row and j-th column of the periodic phase matrix and the preset main period.

[0042] It should be noted that, The larger, The larger, the more it means The more significant the influence on the structure of the periodic phase matrix, The less it is affected by secondary cycle interference, The greater the probability that a periodic component belongs to a pre-defined main period, the more likely it is to be true. The larger, the more The greater the weight assigned.

[0043] Similarly, to obtain the periodic phase matrix Preliminary seasonal factor weights for each data point in the column vector (the j-th column in the periodic phase matrix).

[0044] Furthermore, Loess smoothing (a standard step in the STL decomposition algorithm) is performed on the data in the j-th column vector to obtain smoothed data for each data point in the j-th column vector. Based on the preliminary seasonal factor weights of each data point in the j-th column vector, the smoothed data for each data point in the j-th column vector are weighted and summed to obtain the weighted preliminary seasonal factor of the j-th column vector.

[0045] Similarly, the weighted preliminary seasonal factor of each column vector in the periodic phase matrix is ​​obtained. Based on the weighted preliminary seasonal factor of each column vector, the STL decomposition algorithm is used to extract the seasonal component of the monitoring data sequence of the target power equipment in the current time period under the target monitoring index, making the extracted seasonal component closer to the true periodic pattern. Furthermore, the trend component and residual component of the monitoring data sequence are gradually extracted based on the seasonal component, so that the extracted trend component and residual component can more realistically reflect the operating status of the target power equipment, and the operating status of the target power equipment in the current time period is detected based on the trend component and residual component. Among them, the STL decomposition algorithm and the detection of the operating status of the target power equipment based on the trend component and residual component are existing technologies, and will not be described in detail here.

[0046] In summary, this invention acquires monitoring data of target power equipment under target monitoring indicators within the current time period, centralizes all monitoring data to obtain a monitoring data sequence, and, during the decomposition of the monitoring data sequence using the STL decomposition algorithm, divides the monitoring data sequence into at least two periodic sequences according to a preset principal period, and uses each periodic sequence as a row vector to form a periodic phase matrix. Based on the data distribution in the periodic phase matrix and the dominant change direction between each row vector, the periodic feature sequence of the preset principal period is obtained; based on the rank of the cofactor matrix of each data in the periodic phase matrix, the contribution of each data in the periodic phase matrix to the periodic phase matrix is ​​obtained; based on the difference between the periodic feature sequence and each row vector in the periodic phase matrix, the periodic matching degree between each data in the periodic phase matrix and the preset principal period is obtained; based on the contribution degree and periodic matching degree of each data in the periodic phase matrix, the weighted preliminary seasonal factor of each column vector in the periodic phase matrix is ​​obtained; based on the weighted preliminary seasonal factor of each column vector, the seasonal component in the monitoring data sequence is obtained using the STL decomposition algorithm; and based on the seasonal component, the trend component and residual component in the monitoring data sequence are extracted for use in detecting the status of the target power equipment. The process involves constructing a periodic phase matrix based on the monitoring data sequence of the target power equipment during the current time period and a preset main period. The contribution of each data point in the periodic phase matrix to the matrix, along with the periodic matching degree of each data point, comprehensively reflects the degree of interference from secondary periods, i.e., the reliability of each data point in the periodic phase matrix as a periodic component of the preset main period. This reliability is used to assign weights to each data point in each column vector of the periodic phase matrix, obtaining a preliminary weighted seasonal factor for each column vector. Then, the STL decomposition algorithm is used to extract the seasonal components from the monitoring data sequence, making the extracted seasonal components closer to the true periodic pattern. Based on the seasonal components, trend components and residual components are gradually extracted, enabling the extracted trend components and residual components to more accurately reflect the operating status of the target power equipment and improve the accuracy of condition monitoring of the power equipment.

[0047] Based on the same inventive concept as the above method, this embodiment of the invention also provides an artificial intelligence-based power equipment condition detection system, including a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the computer program, it implements the steps of any one of the above-described artificial intelligence-based power equipment condition detection methods.

[0048] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A method for monitoring the condition of power equipment based on artificial intelligence, characterized in that, The artificial intelligence-based power equipment condition detection method includes: The monitoring data of the target power equipment under the target monitoring indicators are obtained in the current time period. All monitoring data are centralized to obtain the monitoring data sequence. In the process of decomposing the monitoring data sequence using the STL decomposition algorithm, the monitoring data sequence is divided into at least two periodic sequences according to the preset main period. Each periodic sequence is used as a row vector and combined into a periodic phase matrix. According to the data distribution in the periodic phase matrix and the dominant change direction between each row vector, the periodic feature sequence of the preset main period is obtained. Based on the rank of the cofactor matrix of each data point in the periodic phase matrix, the contribution of each data point in the periodic phase matrix to the periodic phase matrix is ​​obtained. Based on the difference between the periodic feature sequence and each row vector in the periodic phase matrix, the periodic matching degree between each data point in the periodic phase matrix and the preset main period is obtained. Based on the contribution and period matching degree of each data point in the periodic phase matrix, the weighted preliminary seasonal factor of each column vector in the periodic phase matrix is ​​obtained. Based on the weighted preliminary seasonal factor of each column vector, the seasonal component in the monitoring data sequence is obtained using the STL decomposition algorithm. Based on the seasonal component, the trend component and residual component in the monitoring data sequence are extracted for use in detecting the status of the target power equipment.

2. The method for detecting the condition of power equipment based on artificial intelligence according to claim 1, characterized in that, The step of obtaining the periodic feature sequence of the preset main period based on the data distribution in the periodic phase matrix and the dominant change direction between each row vector includes: Cluster all the data in the periodic phase matrix to obtain the target cluster with the largest number of members. Form a sub-matrix by combining the row vectors corresponding to the data in the target cluster. Calculate the average value of all data in each column vector of the sub-matrix. Form a first representative sequence of the preset main period by combining the average values ​​corresponding to each column vector of the sub-matrix. Principal component analysis is performed on the row vectors in the periodic phase matrix to obtain the first principal component vector, which is denoted as the second representative sequence of the preset principal period. Based on each data position in the first representative sequence and the second representative sequence, the two data with the same position are added together to obtain a periodic feature sequence with a preset main period.

3. The method for detecting the condition of power equipment based on artificial intelligence according to claim 1, characterized in that, The step of obtaining the contribution of each data point in the periodic phase matrix to the periodic phase matrix based on the rank of the cofactor matrix of each data point in the periodic phase matrix includes: For any data in the periodic phase matrix, the rank of the cofactor matrix of any data is subtracted from the rank of the periodic phase matrix to obtain the rank difference between the cofactor matrix of any data and the periodic phase matrix. The rank difference is then linearly normalized to obtain the first importance of any data. Based on the data in the column vector where any data is located in the periodic phase matrix, calculate the deviation from the mean of any data, perform linear normalization on the deviation from the mean to obtain the degree of deviation between any data and its column vector, and subtract the degree of deviation from the constant 1 to obtain the second importance of any data. Calculate the average between the first importance level and the second importance level to obtain the contribution of any data to the periodic phase matrix.

4. The method for detecting the condition of power equipment based on artificial intelligence according to claim 1, characterized in that, The step of obtaining the periodic matching degree between each data point in the periodic phase matrix and the preset main period based on the difference between the periodic feature sequence and each row vector in the periodic phase matrix includes: For any data in any row vector of the periodic phase matrix, based on the position of the data in the row vector, obtain data with the same position as the data in the periodic feature sequence, and record it as reference data. Calculate the absolute value of the difference between the data and the reference data to obtain the principal periodic component offset of the data. Calculate the principal periodic component offset of each data in any row vector, sum the principal periodic component offsets of each data in any row vector to obtain the total principal periodic component offset of any row vector, calculate the proportion of the principal periodic component offset of any data in the total principal periodic component offset to obtain the degree of principal periodic component offset of any data, and subtract the degree of principal periodic component offset from the constant 1 to obtain the period matching degree between any data and the preset principal period.

5. The method for detecting the condition of power equipment based on artificial intelligence according to claim 1, characterized in that, The preliminary weighted seasonal factor for each column vector in the periodic phase matrix is ​​obtained based on the contribution level and periodic matching degree of each data point in the periodic phase matrix, including: For any data in any column vector of the periodic phase matrix, calculate the sum between the contribution degree of the data and the periodic matching degree to obtain the comprehensive weight of the data. Calculate the comprehensive weight of each data point in any column vector, and calculate the proportion of the comprehensive weight of any data point in the sum of the comprehensive weights of all data points in any column vector to obtain the preliminary seasonal factor weight of any data point. Loess smoothing is performed on any column vector to obtain smoothed data for each data point in the column vector. Based on the preliminary seasonal factor weights of each data point in the column vector, the smoothed data for each data point in the column vector are weighted and summed to obtain the weighted preliminary seasonal factor of the column vector.

6. An artificial intelligence-based power equipment condition monitoring system, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the artificial intelligence-based power equipment condition detection method as described in any one of claims 1-5.

Images