Abnormal data analysis method for monitoring energy consumption of transformer

Abstract

The invention discloses an abnormal data analysis method for monitoring the energy consumption of a transformer, and relates to the technical field of power monitoring, and the method comprises the analysis steps: S1, collecting the operation data of the transformer in real time based on a sensor, carrying out the real-time collection of the data, carrying out the preprocessing of the collected data, and storing the data; and S2, performing analysis learning on the data based on historical transformer data, extracting indexes capable of reflecting energy consumption characteristics of the transformer, screening the extracted indexes based on a correlation analysis method, reducing data dimensions, and extracting trend characteristics and periodic characteristics of the data based on a time sequence analysis method. A complete system is formed, data acquisition and fault processing are linked, energy consumption abnormity of the transformer can be found and solved in time, operation efficiency is improved, energy consumption and cost are reduced, reliable power supply of the electric power system can be guaranteed, defects of a traditional detection method are avoided, and the method has great value in the aspect of improving the overall economic benefit and stability of the electric power system.

Description

Technical Field

[0001] The present invention relates to the technical field of power monitoring, and specifically relates to an abnormal data analysis method for monitoring the energy consumption of transformers. Background Technique

[0002] As an indispensable key device in the power system, the energy consumption of transformers is directly related to the efficiency and cost of power transmission. Accurately monitoring the energy consumption of transformers and promptly detecting abnormalities are of great significance for ensuring the safe and stable operation of the power system and reducing energy losses. At present, most traditional transformer energy consumption monitoring methods rely on simple threshold judgments, only setting a fixed energy consumption threshold. When the monitored energy consumption value exceeds this threshold, it is determined that the transformer is abnormal. However, this method has many limitations. On the one hand, the energy consumption of transformers is affected by a variety of complex factors, such as the dynamic changes of the load, the fluctuations of environmental temperature and humidity, and the aging degree of the equipment itself. Fixed thresholds cannot adapt to these dynamically changing factors, easily leading to misjudgments or missed judgments. For example, in a high-temperature environment in summer, the energy consumption of the transformer may slightly increase due to poor heat dissipation conditions, but this normal fluctuation may be misjudged as abnormal.

[0003] On the other hand, existing monitoring methods often only focus on a single indicator of transformer energy consumption, ignoring the rich internal correlations between energy consumption data, as well as the trends and periodic characteristics of the data over time. In fact, a large amount of important information about the operating state of the transformer is contained in these correlations, trends, and periodic characteristics, which is crucial for accurately judging the abnormal energy consumption of the transformer. In response to this, we propose an abnormal data analysis method for monitoring the energy consumption of transformers. Summary of the Invention

[0004] To solve the above technical problems, an abnormal data analysis method for monitoring the energy consumption of transformers is provided, and this technical solution solves the above problems.

[0005] To achieve the above objectives, the technical solution adopted by the present invention is: an abnormal data analysis method for monitoring the energy consumption of transformers, and the analysis steps are as follows:

[0006] S1. Based on sensors, the operating data of the transformer is collected in real time, the data is collected in real time, and the collected data is preprocessed and stored after that.

[0007] S2. Based on historical transformer data, the data is analyzed and learned, indicators that can reflect the energy consumption characteristics of the transformer are extracted, based on the correlation analysis method, the extracted indicators are screened to reduce the data dimension, and based on the time series analysis method, the trend characteristics and periodic characteristics of the data are extracted.

[0008] S3. Build an anomaly detection model, input the real-time collected data into the model, analyze the data, and determine the reasons for abnormal transformer energy consumption;

[0009] S4. For the analysis results, the staff detect and troubleshoot the abnormal energy consumption equipment of the transformer to eliminate potential hazards.

[0010] Preferably, in step S1, the sensors include voltage sensors, current sensors, temperature sensors and vibration sensors. By deploying different sensors at different positions of the transformer, the parameters during the operation of the transformer are collected in real time; the data preprocessing includes: data cleaning, processing of outliers and data normalization, and the preprocessed data is stored.

[0011] Preferably, the transformer index extraction method in step S2 is based on the Pearson correlation coefficient for extraction, calculate the correlation coefficients between different variables, and find out the variables related to the transformer energy consumption; measure the linear correlation degree between two variables, and the value range is between -1 and 1. When the correlation coefficient is 1, it means that the two variables are completely positively correlated; when it is -1, it means completely negatively correlated; when it is 0, it means that there is no linear correlation between the two variables. By calculating the Pearson correlation coefficients between different variables and the transformer energy consumption, find out the variables with strong correlation as the indicators reflecting the characteristics of the transformer energy consumption.

[0012] Preferably, the specific calculation steps of the Pearson correlation coefficient are as follows: assume that the historical transformer data contains several variables, let the transformer energy consumption data be variable Y, and the other relevant variables be X1, X2, …, X n , each variable has m observations, that is, Y = y1, y2, …, y m , X i = x i1 , x i2 , …, x im , where i = 1, 2, …, n; for variables X and Y, their Pearson correlation coefficient r XY is calculated as follows:

[0013]

[0014] where x i and y i are the i-th observations of variables X and Y respectively; is the mean value of variable X; is the mean value of variable Y; for each variable X i , calculate the two terms in the numerator and the denominator and

[0015] Calculate each variable X based on the above formula i The Pearson correlation coefficient between the transformer energy consumption variable Y

[0016]

[0017] After calculating the Pearson correlation coefficients between all variables and the transformer energy consumption, set a correlation threshold, and screen out the variables whose absolute value of the correlation coefficient is greater than this threshold. The screened variables are the variables related to the transformer energy consumption and are used as indicators reflecting the characteristics of the transformer energy consumption.

[0018] Preferably, the index screening step in step S2 is: define an indicator vector s = s1, s2, …, s p T , where s j ∈{0, 1}, j = 1, 2, …, p; s j = 1 means to retain the j-th index, s j = 0 means to remove the j-th index;

[0019] The screening process can be achieved through the following steps: initialize the s vector, all elements are 1, that is, initially retain all indicators; for each element r in the correlation matrix R jk j < k: if |r jk | > τ, then choose to remove one of the indicators, choose the indicator with the smaller variance to remove. Let VarX j represent the variance of the indicator X j , if VarX j ≤VarX k , then let s j = 0; otherwise let s k = 0;

[0020] Get the screened result:

[0021] According to the indicator vector s, select the retained indicators from the original data set X to obtain the screened data set X′. X′ is an n×p′ matrix, where is the number of retained indicators after screening, and is represented by matrix operation as:

[0022] X′ = A’·diags

[0023] where diags is the diagonal matrix composed of the vector s, and A’ is the data set. The extracted indicators are screened through the above steps.

[0024] Preferably, in step S2, the trend features and periodic features of the data are extracted. During the extraction of the trend features, the moving average method is used to smooth the data, linear regression is used for trend fitting, the correlation coefficient is determined by the least squares method, and finally the goodness of fit is calculated to evaluate the trend fitting effect. The closer the goodness of fit is to 1, the better the effect is.

[0025] During the extraction of the periodic features, autocorrelation analysis is carried out, the autocorrelation coefficients of different lag orders are calculated, the function graph is plotted to find the peak value to judge the periodicity, the power spectrum is calculated by discrete Fourier transform, the frequencies corresponding to the power spectrum peak values are found to determine the main frequency components and periods, and the confidence interval of spectral analysis is used to verify the significance of the periodic features. If it exceeds the confidence interval, the period is considered significant.

[0026] Preferably, the steps for constructing the anomaly detection model in step S3 are as follows: data collection and preparation, collecting transformer data including multiple dimensions from multiple channels, cleaning the data, removing noise, missing values and outliers, and then performing normalization processing to eliminate the influence of dimensions; selecting deep learning as the model framework, then determining the specific architecture and parameters, and optimizing with cross-validation; dividing the training set and the test set, training the model with the training set, and adjusting the parameters to minimize the loss function; evaluating the model with the test set, measuring the performance, and optimizing according to the results. When overfitting or underfitting occurs, corresponding measures are taken; constructing an anomaly cause analysis module, using association rule mining to find factor associations and causal analysis to determine the causal relationship.

[0027] Preferably, the real-time collected data in step S3 includes power, voltage, temperature, humidity and load rate. The collected data is input into the constructed anomaly detection model, the real-time data is compared with the historical normal mode to capture changes, association rule mining is used to find the relationship between the abnormal data and other factors, and when the energy consumption increases, the changes of relevant parameters are analyzed to find the factor combination; the cause is judged by mining historical data to improve the analysis results.

[0028] Preferably, the steps for using association rule mining to find the relationship between abnormal data and other factors are as follows: Let Av represent the event of increased energy consumption, which is defined as the energy consumption value E exceeding a certain set energy consumption threshold E t h res h old , that is, Av: E > E t h res h old Let B k represent the events of other relevant factors. k = 1, 2,..., s, where s is the number of relevant factors. That is, B1 represents too high voltage, and B2 represents too high load rate. The goal of association rule mining is to find the rules that satisfy the support Support and confidence Confidence. The support calculation formula is:

[0029]

[0030] The confidence calculation formula is:

[0031]

[0032] When ConfidenceAv→B k Exceeds the set confidence threshold α and SupportAv→B k When the support threshold β is exceeded, it is considered that the energy consumption increase and factor B are found. k There is a connection between them, that is, there is an association rule Av→B k ;

[0033] When energy consumption increases, analyze the changes in related parameters to find factor combinations. When it is determined that there is an energy consumption increase event Av, for those that meet the association rule Av→B k The various factors B k , find the factor combination by analyzing the impact of different combinations of factor values ​​on energy consumption. Suppose factor B k There is k Different value states, construct a factor combination space By analyzing the energy consumption data under different combinations, we can find out the combination of factors that have the greatest impact on the increase in energy consumption.

[0034] Preferably, after locating the abnormal energy consumption and causes of the transformer through data analysis in step S4, the staff conducts an investigation, performs electrical performance testing, compares standard values ​​and historical data, judges the electrical performance, pays attention to the operating environment, checks whether there are adverse conditions such as moisture, dust, and corrosive gases, and formulates a maintenance and rectification plan based on the results.

[0035] Compared with the prior art, the present invention has the following beneficial effects:

[0036] In the data collection link, the present invention comprehensively collects data through multiple sensors, ensures data quality through preprocessing, lays a solid foundation for subsequent analysis, starts with historical data, uses the Pearson correlation coefficient to find energy consumption-related variables, screens indicators through correlation analysis, and also uses time series analysis to mine trends and periodic characteristics, fully analyzes the law of energy consumption changes, builds a model with deep learning, and through data training and optimization, accurately determines the cause of the abnormality through comparison with historical patterns, association rule mining and causal analysis, and greatly reduces the misjudgment and missed judgment rate; the method forms a complete system, which is closely linked from data collection to fault handling, can not only timely discover and solve the abnormal energy consumption of transformers, improve operating efficiency, reduce energy consumption and costs, but also ensure reliable power supply of the power system, avoid the drawbacks of traditional detection methods, and is of great value in improving the overall economic benefits and stability of the power system. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1This is the flowchart of the abnormal data analysis steps of the present invention. Detailed implementation manners

[0038] The following description is used to disclose the present invention so that those skilled in the art can implement the present invention. The preferred embodiments in the following description are only examples, and other obvious variations can be thought of by those skilled in the art.

[0039] Refer to Figure 1 As shown, an abnormal data analysis method for monitoring the energy consumption of a transformer, the analysis steps are as follows:

[0040] S1. Based on sensors, the operation data of the transformer is collected in real time, the data is collected in real time, and the collected data is preprocessed and stored;

[0041] S2. Based on historical transformer data, the data is analyzed and learned, indicators that can reflect the energy consumption characteristics of the transformer are extracted, based on the correlation analysis method, the extracted indicators are screened to reduce the data dimension, and based on the time series analysis method, the trend characteristics and periodic characteristics of the data are extracted;

[0042] S3. An abnormal detection model is constructed, and the data collected in real time is input into the model for data analysis to determine the cause of abnormal energy consumption of the transformer;

[0043] S4. For the analysis results, the staff detects and checks the equipment with abnormal energy consumption of the transformer to eliminate potential hazards.

[0044] In this application, a variety of sensors are deployed at different positions of the transformer to collect data in real time, enabling comprehensive acquisition of various parameters during the operation of the transformer. The collected data is preprocessed and stored, ensuring data quality and providing an accurate and reliable data basis for subsequent analysis, avoiding analysis deviations caused by data problems. Based on historical data, indicators reflecting the energy consumption characteristics of the transformer are extracted, the Pearson correlation coefficient is used to find variables related to energy consumption, and then the indicators are screened through correlation analysis to reduce dimensions and avoid interference from redundant information. At the same time, time series analysis methods are used to extract trends and periodic characteristics, which can deeply explore the internal laws of the data, better understand the energy consumption change pattern of the transformer, and provide rich information for anomaly detection. The anomaly detection model constructed in step S3 is based on deep learning. It is trained and optimized by combining data collected from multiple channels and preprocessed. By comparing real-time data with historical normal patterns and using association rule mining and causal analysis methods, it can accurately capture data changes and determine the causes of abnormal energy consumption. This intelligent analysis method greatly improves the accuracy and reliability of anomaly detection compared with traditional simple threshold judgment, reducing false positives and false negatives. In step S4, the staff detects and troubleshoots abnormal energy consumption equipment based on the analysis results. They not only conduct electrical performance tests to compare with standard values and historical data but also pay attention to the operating environment. This comprehensive and systematic troubleshooting method can deeply find the root causes of anomalies, formulate targeted repair and rectification plans, effectively eliminate potential hazards, avoid further damage to the equipment, and ensure the stable operation of the transformer.

[0045] In step S1, the sensors include voltage sensors, current sensors, temperature sensors, and vibration sensors. By deploying different sensors at different positions of the transformer, the parameters during the operation of the transformer are collected in real time. The data preprocessing includes data cleaning, outlier handling, and data normalization, and the preprocessed data is stored.

[0046] In this application, data preprocessing is a prior art and will not be elaborated here. Using a variety of sensors such as voltage, current, temperature, and vibration, and deploying them at different positions of the transformer can obtain the operation parameters of the transformer in all directions. Voltage and current sensors can monitor electrical performance, temperature sensors can reflect the heating situation of the equipment, and vibration sensors can capture the mechanical operation state. These data combined can comprehensively present the working state of the transformer, avoiding judgment errors caused by parameter missing. Data cleaning can remove noise and incorrect data, outlier handling can avoid interference with the analysis results, and data normalization makes data with different dimensions comparable, jointly ensuring the high quality and accuracy of the data. After removing the noise in the temperature data, the relationship between temperature and energy consumption can be analyzed more accurately, providing a reliable basis for subsequent analysis.

[0047] In step S2, the transformer index extraction method is based on the Pearson correlation coefficient for extraction. Calculate the correlation coefficients between different variables to find the variables related to transformer energy consumption. Measure the linear correlation degree between two variables, and the value range is between -1 and 1. When the correlation coefficient is 1, it means that the two variables are completely positively correlated; when it is -1, it means that they are completely negatively correlated; when it is 0, it means that there is no linear correlation between the two variables. By calculating the Pearson correlation coefficients between different variables and transformer energy consumption, find the variables with strong correlation as the indicators reflecting the characteristics of transformer energy consumption.

[0048] This application can accurately find the variables with strong correlation with transformer energy consumption from numerous variables, eliminate the interference of irrelevant or weakly correlated variables, making the indicators reflecting the characteristics of transformer energy consumption more representative and targeted. It helps to deeply analyze the essence of the energy consumption problem, measure the linear correlation degree between variables with specific values between -1 and 1, enabling analysts to clearly and intuitively understand the degree of tight association between variables and transformer energy consumption, facilitating quantitative analysis and comparison, and providing accurate data support for subsequent research and decision-making.

[0049] The specific calculation steps of the Pearson correlation coefficient are as follows: Assume that the historical transformer data contains several variables. Let the transformer energy consumption data be variable Y, and the other relevant variables be X1, X2, …, X n , and each variable has m observations, that is, Y = y1, y2, …, y m , X i = x i1 , x i2 , …, x im , where i = 1, 2, …, n;

[0050] For variables X and Y, the Pearson correlation coefficient r XY between them is calculated as follows:

[0051]

[0052] Where x i and y i are the i-th observations of variables X and Y respectively; is the mean value of variable X; is the mean value of variable Y; For each variable X i , calculate the two terms in the numerator and the denominator and

[0053] Based on the above formula, calculate the Pearson correlation coefficient i between each variable X

[0054]

[0055] After calculating the Pearson correlation coefficients between all variables and the transformer energy consumption, a correlation threshold is set, and the variables with the absolute value of the correlation coefficient greater than this threshold are screened out. The screened variables are the variables related to the transformer energy consumption and are used as indicators reflecting the characteristics of the transformer energy consumption.

[0056] This application can accurately identify the variables closely related to the transformer energy consumption from numerous variables, avoiding misjudgments that may be caused by human judgment or other inaccurate methods. It can accurately determine which factors have a greater impact on the transformer energy consumption and provide key objects of concern for subsequent analysis. The linear correlation degree between variables and the transformer energy consumption is quantified by specific values between -1 and 1, enabling analysts to clearly understand the strength and direction of the relationship between variables, facilitating intuitive comparison of the correlation degrees of different variables with energy consumption, providing clear data support for further decision-making. By setting a threshold to screen variables, the data dimension can be effectively reduced, variables with weak correlation with energy consumption can be removed, the complexity of data processing can be reduced while retaining key information, and the efficiency and accuracy of subsequent analysis and model construction can be improved, making the analysis process more efficient and the results more reliable.

[0057] The index screening step in step S2 is as follows: Define an indicator vector s = s1, s2, …, s p T , where s j ∈{0, 1}, j = 1, 2, …, p; s j = 1 indicates retaining the j-th index, and s j = 0 indicates removing the j-th index;

[0058] The screening process can be achieved through the following steps: Initialize the s vector with all elements being 1, that is, initially retain all indicators; for each element r jk j < k in the correlation matrix R: If |r jk | > τ, then choose to remove one of the indicators, and choose to remove the indicator with the smaller variance. Let VarX j represent the variance of the indicator X j . If VarX j ≤ VarX k , then set s j = 0; otherwise set s k = 0;

[0059] Obtain the screened result:

[0060] According to the indicator vector s, select the retained indicators from the original data set X to obtain the screened data set X′. X′ is an n×p′ matrix, where is the number of indicators retained after screening, and is expressed in matrix operations as:

[0061] X′ = A’·diags

[0062] Where diags is a diagonal matrix formed by the vector s, and A’ is the data set. The extracted metrics are screened through the above steps.

[0063] This application analyzes the elements of the correlation matrix, removes the metrics with strong correlation and small variance, effectively reducing the number of metrics in the original data set. This helps to reduce the complexity of data processing, avoid the "curse of dimensionality" problem caused by too high data dimensionality, make the subsequent analysis and modeling processes more efficient, improve the computing efficiency and resource utilization rate. Metrics with strong correlation often contain similar information. By screening and removing one of them, it is possible to avoid repeatedly considering similar information in the analysis, so that the retained metrics can more accurately reflect different aspects of transformer energy consumption, improve the effectiveness of the data and the reliability of the analysis results. Two electrical parameters related to energy consumption are highly correlated. Only retaining the one with larger variance and more representative information can streamline the data without losing key information.

[0064] In step S2, the trend characteristics and periodic characteristics of the data are extracted. During the extraction of trend characteristics, the moving average method is used to smooth the data, linear regression is used for trend fitting, the correlation coefficient is determined by the least squares method, and finally the goodness of fit is calculated to evaluate the trend fitting effect. The closer the goodness of fit is to 1, the better the effect.

[0065] During the extraction of periodic characteristics, autocorrelation analysis is carried out, the autocorrelation coefficients of different lag orders are calculated, a function graph is plotted to find the peak value to judge periodicity, the power spectrum is calculated by discrete Fourier transform, the frequencies corresponding to the peak values of the power spectrum are found to determine the main frequency components and periods, and the confidence interval of spectral analysis is used to verify the significance of periodic characteristics. If it exceeds the confidence interval, the period is considered significant.

[0066] The moving average method of this application can effectively smooth the random fluctuations and noises in the data, making the data smoother and more regular, which helps to more clearly observe the overall trend of the data. The transformer energy consumption data may have some irregular fluctuations due to short-term equipment fluctuations or environmental interferences. The moving average method can reduce the influence of these interferences on trend judgment. Using linear regression for trend fitting and combining the least squares method to determine the correlation coefficient can find the best fitting line of the data and accurately describe the long-term change trend of the data. This is of great significance for predicting the future energy consumption change trend of the transformer and helps to make energy planning and equipment maintenance arrangements in advance.

[0067] In the anomaly detection model construction step of S3, the steps are as follows: data collection and preparation, collecting transformer data including multiple dimensions from multiple channels, cleaning the data, removing noise, missing values and outliers, and then performing normalization processing to eliminate the influence of dimensions; selecting deep learning as the model framework, then determining the specific architecture and parameters, and optimizing with cross-validation; dividing the training set and the test set, training the model with the training set, and adjusting the parameters to minimize the loss function; evaluating the model with the test set, measuring the performance, and optimizing according to the results, taking corresponding measures when overfitting or underfitting; constructing an anomaly cause analysis module, using association rule mining to find factor associations and causal analysis to determine causal relationships.

[0068] In this application, multi-dimensional transformer data is collected from multiple channels, and the data is prepared through cleaning, handling missing and outlier values, and normalization to eliminate the influence of dimensions; a deep learning framework is adopted, and the specific architecture and parameters are determined through cross-validation; the training set and the test set are divided, the training set is used to train and adjust the parameters to minimize the loss function, and then the test set is used to evaluate the model performance, and optimization is carried out for overfitting or underfitting; an anomaly cause analysis module is constructed, and association rule mining and causal analysis are used to explore the reasons; multi-channel collection and preprocessing improve the data quality and lay the foundation for model learning; the deep learning framework combined with cross-validation optimizes the model performance to adapt to the data characteristics; dividing the data set for training and evaluation ensures the reliability and stability of the model; the anomaly cause analysis module accurately locates the anomaly factors and provides a direction for solving problems; the whole process forms a system, comprehensively and effectively supports the operation and maintenance of transformers, and has important application value.

[0069] The real-time collected data in S3 step includes power, voltage, temperature, humidity and load rate. The collected data is input into the constructed anomaly detection model, the real-time data is compared with the historical normal mode to capture changes, association rule mining is used to find the connection between the abnormal data and other factors, and when the energy consumption increases, the relevant parameter changes are analyzed to find the factor combination; the cause is judged through historical data mining to improve the analysis result.

[0070] This application processes and analyzes the multi-dimensional data of power, voltage, temperature, humidity and load rate collected in real time, which has many benefits. On the one hand, the multi-dimensional data can comprehensively reflect the operation state of the transformer and provide rich information for judging anomalies; comparing with the historical normal mode can timely detect abnormal changes. On the other hand, association rule mining can deeply explore the connection between abnormal data and other factors, analyze the parameter changes when the energy consumption increases to find the key factor combination, and combining historical data mining can accurately locate the root cause of the anomaly. In addition, continuously mining historical data can continuously improve the understanding of the anomaly cause, optimize the anomaly detection model and analysis method, and improve the reliability and effectiveness of the system.

[0071] The steps of using association rule mining to find the relationship between abnormal data and other factors are as follows: Let Av represent the event of increased energy consumption, which is defined as the energy consumption value E exceeding a certain set energy consumption threshold E t h res h old , that is, Av: E > E t h res h old Let B k represent the events of other relevant factors. k = 1, 2, …, s, where s is the number of relevant factors. That is, B1 represents too high voltage, B2 represents too high load rate. The goal of association rule mining is to find rules that satisfy the support Support and confidence Confidence. The formula for support is:

[0072]

[0073] The formula for confidence is:

[0074]

[0075] When ConfidenceAv→B k exceeds the set confidence threshold α and SupportAv→B k exceeds the set support threshold β, it is considered that the relationship between increased energy consumption and factor B k is found, that is, there is an association rule Av→B k ;

[0076] When analyzing the change of relevant parameters to find factor combinations during increased energy consumption, when it is determined that there is an event of increased energy consumption Av, for each factor B k that satisfies the association rule Av→B k , the factor combination is found by analyzing the influence of different value combinations of the factors on energy consumption. Let factor B k have t k different value states, and a factor combination space is constructed By analyzing the energy consumption data under different combinations, the factor combination that has a great influence on the increase of energy consumption is found.

[0077] By setting energy consumption increase events and related factor events and screening using support and confidence formulas, this application can accurately identify the factors that are truly related to the increase in energy consumption from numerous factors, avoiding the inclusion of irrelevant or accidentally related factors in the consideration, improving the accuracy and pertinence of the analysis of the causes of abnormal energy consumption, and helping the staff to concentrate on investigating the key factors that really affect energy consumption; the calculation of support and confidence provides a quantitative index for the degree of association between factors and the increase in energy consumption. Support reflects the frequency of the factor combination appearing in the data, and confidence reflects the possibility of another factor appearing when one factor appears. This quantitative method makes the analysis results more persuasive and facilitates the staff to formulate corresponding monitoring and treatment strategies according to the strength of the association.

[0078] After locating the abnormal energy consumption of the transformer and its causes through data analysis in step S4, the staff carry out investigations, conduct electrical performance tests, compare the standard values and historical data, judge the electrical performance, pay attention to the operating environment, check whether there are adverse conditions such as humidity, a lot of dust, and corrosive gases, and formulate a maintenance and rectification plan according to the results.

[0079] Through the preliminary data analysis, this application can accurately locate the specific problems and causes of the abnormal energy consumption of the transformer. On this basis, the investigation work is carried out, enabling the staff to conduct targeted inspections, avoiding blind investigations, improving work efficiency, and saving time and labor costs. If the data analysis shows that the abnormal energy consumption may be caused by abnormal winding resistance, the staff can focus on detecting the components related to the winding.

[0080] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art of this industry should understand that the present invention is not limited by the above embodiments. What is described in the above embodiments and the specification is only the principle of the present invention. Without departing from the spirit and scope of the present invention, the present invention will have various changes and improvements, and these changes and improvements all fall within the scope of the present invention claimed.

Claims

1. A method for analyzing abnormal data for monitoring transformer energy consumption, characterized in that: The analysis steps are: S1. Collecting transformer operation data in real time based on sensors, collecting data in real time, preprocessing the collected data and storing it; S2. Analyze and learn the data based on historical transformer data, extract indicators that can reflect the energy consumption characteristics of transformers, screen the extracted indicators based on the correlation analysis method, reduce the data dimension, and extract the trend characteristics and periodicity characteristics of the data based on the time series analysis method; S3. Build an anomaly detection model, input the real-time collected data into the model, analyze the data, and determine the cause of abnormal transformer energy consumption; S4. Based on the analysis results, the staff will inspect and check the abnormal energy consumption equipment of the transformer to eliminate hidden dangers.

2. The abnormal data analysis method for monitoring transformer energy consumption according to claim 1 is characterized in that: The sensors in step S1 include voltage sensors, current sensors, temperature sensors and vibration sensors. Different sensors are deployed at different positions of the transformer to collect the working parameters of the transformer in real time. Data preprocessing includes: data cleaning, processing outliers and data normalization, and storing the preprocessed data.

3. The abnormal data analysis method for monitoring transformer energy consumption according to claim 1 is characterized in that: The transformer index extraction method in step S2 is based on the Pearson correlation coefficient, and the correlation coefficients between different variables are calculated to find out the variables related to transformer energy consumption; the linear correlation between two variables is measured, and the value range is between -1 and 1. When the correlation coefficient is 1, it means that the two variables are completely positively correlated; when it is -1, it means that the two variables are completely negatively correlated; When it is 0, it means that there is no linear correlation between the two variables. By calculating the Pearson correlation coefficient between different variables and transformer energy consumption, the variables with strong correlation are found as indicators reflecting the transformer energy consumption characteristics.

4. The abnormal data analysis method for monitoring transformer energy consumption according to claim 3 is characterized in that: The specific calculation steps of the Pearson correlation coefficient are as follows: assume that the historical transformer data contains several variables, set the transformer energy consumption data as variable Y, and the other related variables are X1, X2,…, X n , each variable has m observations, that is, Y = [y1, y2, ..., y m ],X i =[x i1 ,x i2 ,…,x im ], where i = 1, 2, ..., n; for variables X and Y, the Pearson correlation coefficient between them is r XY The calculation formula is as follows: where x i and i are the i-th observation values ​​of variables X and Y respectively; is the mean of variable X; is the mean of variable Y; for each variable X i , calculate the molecule and the two terms in the denominator and Calculate each variable X based on the above formula i Pearson correlation coefficient between the transformer energy consumption variable Y After calculating the Pearson correlation coefficient between all variables and transformer energy consumption, a correlation threshold is set to screen out the variables whose absolute values ​​of correlation coefficients are greater than the threshold. The screened variables are variables related to transformer energy consumption and serve as indicators reflecting the characteristics of transformer energy consumption.

5. The abnormal data analysis method for monitoring transformer energy consumption according to claim 1 is characterized in that: The indicator screening step in step S2 is: define an indicator vector s = (s1, s2, ..., s p ) T , where s j ∈{0,1},j=1,2,…,p;s j =1 means retaining the jth index, s j =0 means removing the jth index; The screening process can be achieved through the following steps: Initialize the s vector with all elements being 1, that is, all indicators are retained initially; for each element r in the correlation matrix R jk (j < k): If |r jk | > τ, then choose to remove one of the indicators, and choose the indicator with the smaller variance to remove. Let Var(X j ) represent the variance of the indicator X j . If Var(X j ) ≤ Var(X k ), then set s j = 0; otherwise set s k = 0; The filtered results are: According to the indicator vector s, the retained indicators are selected from the original data set X to obtain the filtered data set X′, which is an n×p′ matrix, where is the number of indicators retained after screening, expressed as matrix operations: X′=A'·diag(s) Where diag(s) is a diagonal matrix composed of vector s, A' is a data set, and the extracted indicators are screened through the above steps.

6. The abnormal data analysis method for monitoring transformer energy consumption according to claim 1 is characterized in that: In step S2, the trend characteristics and periodic characteristics of the data are extracted. In the process of trend feature extraction, the moving average method is used to smooth the data, linear regression is used for trend fitting, the correlation coefficient is determined by the least squares method, and finally the goodness of fit is calculated to evaluate the trend fitting effect. The closer the goodness of fit is to 1, the better the effect; In the process of extracting periodic features, autocorrelation analysis is performed, autocorrelation coefficients of different lag orders are calculated, function graphs are drawn to find peak values ​​to determine periodicity, power spectrum is calculated through discrete Fourier transform, the frequency corresponding to the peak value of the power spectrum is found to determine the main frequency components and period, and the confidence interval of spectrum analysis is used to verify the significance of periodic features. If it exceeds the confidence interval, the period is considered significant.

7. The abnormal data analysis method for monitoring transformer energy consumption according to claim 1 is characterized in that: The steps for building the anomaly detection model in step S3 are: data collection and preparation, collecting transformer data including multi-dimensional data from multiple channels, cleaning the data, removing noise, missing values ​​and outliers, and then normalizing the data to eliminate the impact of dimension; the model selects deep learning as the model framework, and then determines the specific architecture and parameters, and uses cross-validation to optimize; divides the training set and test set, trains the model with the training set, and adjusts the parameters to minimize the loss function; uses the test set to evaluate the model, measures the performance, optimizes according to the results, and takes corresponding measures in case of overfitting or underfitting; builds an abnormal cause analysis module, uses association rules to mine factor associations, and uses causal analysis to determine causal relationships.

8. The abnormal data analysis method for monitoring transformer energy consumption according to claim 1 is characterized in that: The data collected in real time in step S3 include power, voltage, temperature, humidity and load rate. The collected data is input into the constructed anomaly detection model, and the real-time data is compared with the historical normal mode to capture changes. Association rules are mined to find the connection between abnormal data and other factors. When energy consumption increases, the changes in related parameters are analyzed to find the combination of factors; the causes are determined through historical data mining to improve the analysis results.

9. The abnormal data analysis method for monitoring transformer energy consumption according to claim 8, characterized in that: The steps of mining association rules to find abnormal data and other factors are as follows: Let Av represent the event of increased energy consumption, which is defined as the energy consumption value E exceeding a certain set energy consumption threshold E th resh old , that is, Av:E>E th resh old , let B k Represents events of other related factors, k = 1, 2, ..., s, s is the number of related factors, that is, B1 represents voltage is too high, B2 represents load rate is too high, the goal of association rule mining is to find rules that meet support and confidence, support calculation formula is: The confidence calculation formula is: When Confidence(Av→B k ) exceeds the set confidence threshold α and Support(Av→B k ) exceeds the set support threshold β, it is considered that the energy consumption increase and factor B are found. k There is a connection between them, that is, there is an association rule Av→B k ; When energy consumption increases, analyze the changes in related parameters to find factor combinations. When it is determined that there is an energy consumption increase event Av, for those that meet the association rule Av→B k The various factors B k , find the factor combination by analyzing the impact of different combinations of factor values ​​on energy consumption. Suppose factor B k There is k Different value states, construct a factor combination space By analyzing the energy consumption data under different combinations, we can find out the combination of factors that have the greatest impact on the increase in energy consumption.

10. The abnormal data analysis method for monitoring transformer energy consumption according to claim 1, characterized in that: After locating the abnormal energy consumption of the transformer and its causes through data analysis in step S4, the staff conducts an investigation, performs electrical performance tests, compares standard values ​​and historical data, judges the electrical performance, pays attention to the operating environment, checks for adverse conditions such as moisture, dust, and corrosive gases, and formulates a maintenance and rectification plan based on the results.

Images