Switchgear insulation status monitoring and management system
Through the switch cabinet insulation state monitoring and management system, data fusion and feature screening combined with time series modeling and dynamic early warning mechanisms are used to solve the subjectivity and unreliability of traditional monitoring methods, and the accurate monitoring of the insulation state of switch cabinets is achieved and Early warning ensures the safe and stable operation of the power system.
Patent Information
- Application Number
- CN202410996335.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-24
- Publication Date
- 2024-11-15
- Estimated Expiration
- 2044-07-24
AI Technical Summary
The traditional switch cabinet insulation state monitoring method has strong subjectivity, poor real-time performance and low detection accuracy, which is difficult to meet the needs of modern power systems for high efficiency and accuracy. The information volume of a single data source is limited, making it difficult to fully reflect the insulation state of the switch cabinet. Moreover, monitoring data in complex operating environments is susceptible to interference, resulting in unreliable detection results.
Adopt the switch cabinet insulation status monitoring and management system, through data fusion and feature screening, combined with time series modeling and dynamic early warning mechanism, multi-source data fusion and dynamic weight adjustment, extract important features with high distinction ability and high stability, and build diverse The time series data set adopts multi-layer LSTM network model and statistical process control method to realize dynamic monitoring and abnormal detection, adjust the early warning threshold in real time, generate accurate early warning signals, and provide scientific and reasonable maintenance and maintenance strategies based on historical maintenance data. .
Improve the detection accuracy and fault warning capabilities of switch cabinet insulation status monitoring, ensure the safety and stability of the power system, significantly improve the performance and intelligence level of the monitoring system, and provide scientific and reasonable for power equipment status monitoring. Evaluation criteria.
Smart Images

Figure CN118536048B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of switch cabinet insulation state monitoring, and more specifically, to a switch cabinet insulation state monitoring and management system. Background Art
[0002] As a key equipment in the power system, the insulation state of the switchgear directly affects the safety and reliability of power transmission. During the long-term operation of the power system, due to the complex and changeable external environment and the natural aging of the insulation materials, the insulation state of the switchgear may deteriorate, resulting in operational failures of the power equipment. In order to ensure the stable operation of the power system, it is crucial to monitor the insulation state of the switchgear in real time, accurately identify abnormal conditions and issue timely warnings. However, traditional monitoring methods mainly rely on manual inspections and offline tests, which have defects such as strong subjectivity, poor real-time performance and low detection accuracy, and are difficult to meet the requirements of modern power systems for high efficiency and accuracy.
[0003] In the existing technology, although a variety of detection methods based on a single data source have been proposed for monitoring the insulation status of switchgear, the limited amount of information in a single data source makes it difficult to fully reflect the insulation status of the switchgear. In addition, in complex operating environments, the monitoring data is easily interfered with, resulting in unreliable detection results. In addition, in terms of abnormality detection and early warning, the existing methods usually rely on fixed thresholds and lack a dynamic adjustment mechanism. They cannot adapt to the actual needs under different operating conditions and are prone to false alarms and missed alarms, affecting the accuracy and timeliness of fault early warnings. Therefore, there is an urgent need for a more intelligent and accurate switchgear insulation status monitoring method to improve the safety and stability of the power system.
[0004] In order to solve the above problems, a technical solution is now provided. Summary of the invention
[0005] In order to overcome the above-mentioned defects of the prior art, the embodiments of the present invention provide a switch cabinet insulation status monitoring and management system, which ensures the accurate extraction of important features through data fusion and feature screening, and combines time series modeling and dynamic early warning mechanism to accurately predict and warn the insulation degradation trend in real time. Specifically, multi-source data fusion and dynamic weight adjustment are used to improve the integrity and representativeness of the data; through feature screening, important features with high discrimination ability and high stability are extracted; data enhancement and sliding window technology are used to construct a diversified time series data set to ensure the effectiveness and generalization ability of model training; through a multi-layer LSTM network model and a statistical process control method, dynamic monitoring and anomaly detection of the insulation state are realized, and the early warning threshold can be adjusted in real time to generate accurate early warning signals. In addition, combined with historical maintenance data, a scientific and reasonable maintenance and overhaul strategy is provided to ensure the safe and stable operation of the power system to solve the problems raised in the above background technology.
[0006] To achieve the above object, the present invention provides the following technical solutions:
[0007] The switch cabinet insulation status monitoring and management system includes: data aggregation module, feature identification module, sequence insight module and strategy matching module;
[0008] The data aggregation module collects switchgear data, performs multi-source data fusion and dynamic weight adjustment through principal component analysis, generates a fused data matrix, reconstructs and restores it to extract features, and passes the generated fused data matrix and extracted features to the feature screening module;
[0009] The feature identification module analyzes the effectiveness of each feature in distinguishing normal and abnormal states to obtain the feature value distinguishing ability index, and combines the stability of the feature under different time periods and environmental conditions to obtain the feature steady-state separation index. The feature goodness index is further obtained through comprehensive analysis to screen out important features, and the selected important features are passed to the sequence insight module;
[0010] The sequence insight module uses data enhancement technology to build a diverse time series data set based on the selected important features, and uses sliding window technology to segment the data and build a multi-layer LSTM network model. It performs dynamic anomaly detection and early warning through residual analysis and EWMA control charts, adjusts the early warning threshold in real time, generates early warning signals, and transmits the generated early warning signals to the strategy matching module.
[0011] The strategy matching module matches the warning results with the historical maintenance methods and issues corresponding maintenance and repair strategies.
[0012] In a preferred embodiment, the data aggregation module includes the following processing contents:
[0013] S1.1, collect different types of data from various sensors of the switchgear to ensure that the data covers different aspects of the switchgear operation;
[0014] S1.2, preprocess each type of data;
[0015] S1.3, construct a data matrix by chronologically arranging the standardized multi-source data , each column represents a data source, and each row represents data at a time point;
[0016] S1.4, Calculation of data matrix The covariance matrix of , the formula is: ,in, is the sample size, where Represents the data matrix The transpose of
[0017] S1.5, calculate the eigenvalues and eigenvectors of the covariance matrix, sort the eigenvalues in descending order, and select the first The eigenvector corresponding to the largest eigenvalue is constructed to construct the eigenvector matrix ;
[0018] S1.6, project the data matrix onto the eigenvector matrix to obtain the fused data matrix , the formula is: ;
[0019] S1.7, introduce a dynamic weight adjustment mechanism to dynamically adjust the weight of each data source in the fusion process according to the changes in real-time monitoring data;
[0020] S1.8, reconstruct the fused principal component data to restore it to the scale of the original data. The reconstruction formula is: , Represents the data reconstruction matrix after principal component analysis fusion and projection.
[0021] In a preferred embodiment, the process of obtaining the characteristic value discrimination capability index is as follows:
[0022] A1, divide the data set into normal state data and abnormal state data, extract the features of each sample, and construct a feature matrix, where The feature matrix representing the normal state data, A feature matrix representing abnormal state data;
[0023] A2, marked as characteristic momentum is , calculated by multiplying the eigenvalue and its changing speed, the characteristic momentum calculation formula is: ;
[0024] in, Representation characteristics In the The speed of change in a sample is defined as the rate of change of the characteristic value between two consecutive time points;
[0025] A3, standardize the characteristic momentum to eliminate dimension and scale differences. Subtract the mean of all characteristic momentums from the characteristic momentum and divide it by the standard deviation. The standardization formula is: ;
[0026] in, and Respectively represent the mean and standard deviation of all characteristic momentums;
[0027] A4, using information entropy to evaluate the distinguishing ability of feature momentum, calculate the entropy value of each feature momentum, the entropy value is obtained by calculating the probability distribution of the feature momentum in different states; then, calculate the entropy weight index of the feature momentum, the entropy weight index is the inverse of the entropy value;
[0028] A5, combine the standardized feature momentum and entropy weight index by multiplication, and calculate the feature discrimination ability index by combining the standardized feature momentum and entropy weight index .
[0029] In a preferred embodiment, the acquisition logic of the characteristic steady-state separation index is as follows:
[0030] B1, collect continuous monitoring data over a period of time, including characteristic values and corresponding environmental conditions;
[0031] Divide the data into several time periods, each of which contains data under the same environmental conditions or different environmental conditions;
[0032] B2, standardize each feature and environmental condition;
[0033] In each time period Calculate the coefficient of variation of environmental conditions : ;
[0034] in, Indicates an environmental condition. and Respectively represent environmental conditions In time period standard deviation and mean within ;
[0035] B3, calculation features Environmental conditions affect the correction value: ;
[0036] in, It is the time period Coefficient of variation of the main environmental conditions within the
[0037] B4, calculate the characteristic steady-state separation index by combining the coefficient of variation between time periods : ;
[0038] in, is the number of time periods; and Respectively in the time period and Internal, feature The coefficient of variation was corrected for the influence of environmental conditions.
[0039] In a preferred embodiment, the characteristic discrimination ability index and the characteristic steady-state separation index are standardized and converted to the same scale;
[0040] The standardized feature discrimination ability index and feature steady-state separation index are used as the coordinate axes of the two-dimensional space to construct the feature space. The coordinates of each feature in the corresponding space are ;
[0041] Assume that the coordinates of the ideal point in the feature space are (1,1), indicating that the feature is optimal in both distinguishing ability and stability;
[0042] Calculate the Euclidean distance between each feature point and the ideal feature point, then invert the Euclidean distance to calculate the feature goodness index, which is calculated using the following formula: ;
[0043] in, represents the feature goodness index; Represents the Euclidean distance between feature points and rational feature points;
[0044] According to the calculated feature goodness index, all features are sorted from large to small according to the index value;
[0045] Features whose feature goodness index is higher than the goodness threshold are selected as important features.
[0046] In a preferred embodiment, the sequence insight module includes the following processing contents:
[0047] S3.1, organize and standardize the historical data and real-time data of the selected important features to construct a time series dataset;
[0048] S3.2, divide the time series data set into training set, validation set and test set, and use sliding window technology to split the time series data into subsequences of fixed length for input into LSTM model, and each window contains a fixed number of time steps;
[0049] S3.3, construct a multi-layer LSTM network model, including an input layer, a multi-layer LSTM layer, a fully connected layer, and an output layer;
[0050] S3.4, after the model training is completed, the model is applied to real-time data prediction, and for each time step, the difference between the model prediction value and the actual value is calculated to obtain the residual;
[0051] S3.5, use an exponentially weighted moving average control chart to monitor the residuals, set control limits, and determine whether there are abnormalities based on the relationship between the residuals and the control limits;
[0052] S3.6, when the residual exceeds the control limit, it is marked as an outlier;
[0053] S3.7, as time goes by and the residual changes dynamically, the weighted moving average residual value and residual standard deviation are updated in real time, and then the warning threshold is adjusted. The adjustment formula of the warning threshold is: ;
[0054] in, Represents the residual, which is the smoothed residual value calculated by the exponentially weighted moving average method; is the residual standard deviation, which is used to assess the dispersion of the residuals; is the adjustment coefficient, which is used to adjust the sensitivity of the warning threshold;
[0055] S3.8, when the residual exceeds the warning threshold, a warning signal is generated.
[0056] In a preferred embodiment, the strategy matching module includes the following processing contents:
[0057] First, the real-time warning signal is matched with the fault type in the historical maintenance database, and the frequency and amplitude of the warning signal are analyzed using a Bayesian network to determine the fault type. Then, the optimal maintenance strategy is recommended based on the successful maintenance and overhaul cases in the historical data using a machine learning model. Detailed overhaul steps, required spare parts and labor time estimates are automatically generated, and a maintenance work order is generated. The work order is sent to on-site maintenance personnel through IoT devices. Finally, the maintenance results and new data are incorporated into the system through a feedback mechanism.
[0058] The technical effects and advantages of the switch cabinet insulation status monitoring and management system of the present invention are as follows:
[0059] 1. The present invention can intuitively evaluate the effectiveness of each feature in distinguishing normal and abnormal states and its stability under different time periods and environmental conditions by standardizing the feature distinguishing ability index and the feature steady-state separation index and constructing a two-dimensional feature space. By determining the ideal feature point and calculating the Euclidean distance between each feature point and the ideal feature point, and then calculating the feature goodness index by reversing the Euclidean distance, the comprehensive performance of each feature is effectively quantified. The larger the feature goodness index, the better the comprehensive performance of the feature in distinguishing ability and stability. Conversely, the smaller the goodness index, the poorer the comprehensive performance. By sorting and screening the feature goodness index, important features with high distinguishing ability and high stability can be scientifically selected, thereby improving the detection accuracy and fault warning capability of the switch cabinet insulation state monitoring system. The reliability and consistency of the selected features in practical applications are ensured, which helps to discover the degradation trend and potential faults of insulating materials at an early stage, ensure the safety and stability of the power system, significantly improve the performance and intelligence level of the monitoring system, and provide a scientific and reasonable evaluation standard for power equipment state monitoring.
[0060] 2. Through data fusion and feature screening, the accurate extraction of important features is ensured, and combined with time series modeling and dynamic early warning mechanism, the insulation degradation trend is accurately predicted and warned in real time. Specifically, the integrity and representativeness of the data are improved by using multi-source data fusion and dynamic weight adjustment; through feature screening, important features with high discrimination ability and high stability are extracted; data enhancement and sliding window technology are used to construct a diversified time series data set to ensure the effectiveness and generalization ability of model training; through multi-layer LSTM network model and statistical process control method, dynamic monitoring and anomaly detection of insulation status are realized, and the warning threshold can be adjusted in real time to generate accurate warning signals. In addition, combined with historical maintenance data, a scientific and reasonable maintenance and overhaul strategy is provided to ensure the safe and stable operation of the power system. BRIEF DESCRIPTION OF THE DRAWINGS
[0061] Figure 1 It is a flow chart of the switch cabinet insulation status monitoring and management system of the present invention;
[0062] Figure 2 The figure is a schematic diagram of the steps of the sequence insight module of the switch cabinet insulation status monitoring and management system of the present invention. DETAILED DESCRIPTION
[0063] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0064] Embodiment 1: Figure 1 The present invention provides a switch cabinet insulation state monitoring and management system, including: a data aggregation module, a feature identification module, a sequence insight module and a strategy matching module;
[0065] The data aggregation module collects switchgear data, performs multi-source data fusion and dynamic weight adjustment through principal component analysis, generates a fused data matrix, reconstructs and restores it to extract features, and passes the generated fused data matrix and extracted features to the feature screening module;
[0066] The feature identification module analyzes the effectiveness of each feature in distinguishing normal and abnormal states to obtain the feature value distinguishing ability index, and combines the stability of the feature under different time periods and environmental conditions to obtain the feature steady-state separation index. The feature goodness index is further obtained through comprehensive analysis to screen out important features, and the selected important features are passed to the sequence insight module;
[0067] The sequence insight module uses data enhancement technology to build a diverse time series data set based on the selected important features, and uses sliding window technology to segment the data and build a multi-layer LSTM network model. It performs dynamic anomaly detection and early warning through residual analysis and EWMA control charts, adjusts the early warning threshold in real time, generates early warning signals, and transmits the generated early warning signals to the strategy matching module.
[0068] The strategy matching module matches the warning results with the historical maintenance methods and issues corresponding maintenance and repair strategies.
[0069] In the insulation status monitoring of switchgear, comprehensive and accurate data collection is the basis for efficient monitoring and fault warning. As the core equipment of the power system, the insulation status of the switchgear is directly related to the safety and stability of power transmission. The traditional single data source monitoring method is difficult to fully reflect the complex operating status due to the single data dimension and limited information. Therefore, the multi-source data fusion technology is introduced to collect multiple parameters such as temperature, humidity, leakage current, partial discharge and vibration, so as to fully obtain the operating information of the equipment. In the data acquisition stage, advanced sensor technology and industrial Internet of Things (IIoT) technology are used to achieve high-precision synchronous acquisition to ensure the time consistency and integrity of various types of data. The dimensional differences are eliminated through standardization processing, and the data is fused by the improved principal component analysis (PCA) method. The dynamic weight adjustment mechanism is introduced to flexibly adjust the importance of each data source according to the changes in the real-time monitoring data, so as to extract representative principal components and improve the data expression ability and analysis efficiency. Finally, the fused data is reconstructed and the scale of the original data is restored to ensure that the data can accurately reflect the operating status of the equipment, providing a high-quality data foundation for subsequent feature extraction, dynamic evaluation and fault prediction. This innovative method of data collection and processing not only enhances the expressiveness and explanatory power of important features, but also lays a solid foundation for improving the accuracy and reliability of the monitoring system.
[0070] The data aggregation module includes the following processing contents:
[0071] S1.1, collect different types of data from various sensors of the switch cabinet, including temperature, humidity, leakage current, partial discharge, vibration and other parameters, to ensure that these data cover different aspects of the switch cabinet operation.
[0072] S1.2, preprocess each type of data, including denoising, missing value filling and normalization. Denoising can be done by filtering, missing value filling can be done by interpolation or statistical methods, and normalization standardizes the data to the same dimension.
[0073] S1.3, construct a data matrix by chronologically arranging the standardized multi-source data , each column represents a data source, and each row represents data at a time point.
[0074] S1.4, Calculation of data matrix The covariance matrix of , the formula is: ,in, is the sample size, where Represents the data matrix The transpose of .
[0075] S1.5, calculate the eigenvalues and eigenvectors of the covariance matrix. Arrange the eigenvalues in descending order and select the first The eigenvector corresponding to the largest eigenvalue is constructed to construct the eigenvector matrix .
[0076] S1.6, project the data matrix onto the eigenvector matrix to obtain the fused data matrix , the formula is: ;
[0077] Through the improved PCA method, multi-source data can be fused into a small number of representative principal components, reducing the dimension of the data and highlighting the main features of the data.
[0078] S1.7, in the improved principal component analysis PCA process, a dynamic weight adjustment mechanism is introduced to dynamically adjust the weight of each data source in the fusion process according to the changes in real-time monitoring data.
[0079] In the process of dynamic weight adjustment, by real-time monitoring of data changes, we first calculate the relevance and importance indicators of each data source at different time points, such as mutual information or Pearson correlation coefficient, and then dynamically allocate weights based on these indicators. Data sources with higher weights will contribute more to the final fusion result during the fusion process. Specifically, the weights are updated at each time point, and the weighted moving average method is used to smooth the weight changes to avoid instability caused by instantaneous fluctuations, ultimately ensuring that the data fusion process can flexibly respond to changes in real-time data and improve the accuracy and reliability of the fused data.
[0080] Dynamic weight adjustment can flexibly adjust the importance of different data sources according to actual conditions, thereby improving the adaptability and accuracy of data fusion.
[0081] S1.8, reconstruct the fused principal component data to restore it to the scale of the original data. The reconstruction formula is: , Represents the data reconstruction matrix after principal component analysis (PCA) fusion and projection.
[0082] By analyzing the reconstructed data, important features are extracted for insulation status evaluation.
[0083] The reconstruction process ensures that the fused data can still reflect the main features of the original data, which is convenient for subsequent analysis and feature extraction.
[0084] In the data collection stage, multi-source data is fused to improve data quality and consistency and reduce the complexity of subsequent processing. The fused data has higher expressive power, can more accurately reflect the insulation status of the switchgear, and improve the detection accuracy and fault warning capability of the monitoring system. In addition, the dynamic weight adjustment mechanism makes the method highly adaptable and robust, and can cope with different operating environments and changes in conditions.
[0085] The feature identification module includes the following:
[0086] In the insulation status monitoring of switchgear, accurately identifying the characteristics of normal and abnormal states is the key to ensuring the safe operation of the power system. As an important equipment in the power system, the internal insulation status of the switchgear directly affects the stability and reliability of power transmission. Due to the complex and changeable operating environment of the switchgear, the insulation material will gradually age over time, causing failures. Therefore, real-time monitoring and accurate identification of the insulation status of the switchgear are crucial. However, traditional monitoring methods have defects such as strong subjectivity and poor real-time performance due to their reliance on manual inspections and offline testing, and are difficult to meet the requirements of modern power systems for efficiency and accuracy.
[0087] In order to improve the performance of the monitoring system, it is of great significance to introduce the feature discrimination ability index as an indicator for quantitatively evaluating the discrimination ability of features under different states. The feature discrimination ability index can accurately quantify the performance difference of features under normal and abnormal states by comprehensively considering the change speed and information amount of the feature value. Specifically, the index captures the dynamic changes of features through the concept of momentum, and quantifies the information amount of features through the entropy weight method, thereby comprehensively reflecting the discrimination ability of features under different states. The larger the index value, the more significant the performance difference of the feature under different states, and the higher the discrimination ability. Conversely, the smaller the index value, the weaker the feature discrimination ability.
[0088] In practical applications, selecting features with high feature discrimination index can significantly improve the detection accuracy and fault warning capability of the switch cabinet insulation status monitoring system. Features with high discrimination ability can detect the degradation trend and potential faults of insulation materials earlier, realize real-time monitoring and accurate identification of insulation status, reduce the failure rate of power system, and ensure the safety and reliability of power transmission. Therefore, calculating the feature discrimination index is not only an optimization of traditional monitoring methods, but also an important means to improve the intelligence level and operation safety of power systems. The introduction of this index provides a scientific and reasonable evaluation standard for power equipment status monitoring, which has broad application prospects and important practical significance.
[0089] The process of obtaining the characteristic value discrimination capability index is as follows:
[0090] A1, divide the data set into normal state data and abnormal state data, extract the features of each sample, and construct the feature matrix. The feature matrix representing the normal state data, A feature matrix representing abnormal state data.
[0091] The acquisition process of normal state data and abnormal state data can be achieved through a combination of historical data records, laboratory simulations, online monitoring and real-time labeling, data-driven automatic classification, and synthetic data. This includes using the operation records and fault reports of the switchgear to mark normal and abnormal state data, generating corresponding data through fault simulation tests in a laboratory environment, installing an online monitoring system to collect and label data in real time, using unsupervised learning and anomaly detection algorithms to automatically identify abnormal state data, and generating more diverse normal and abnormal state data through data enhancement and numerical simulation technology, thereby building a comprehensive and accurate data set.
[0092] A2, marked as characteristic momentum is , calculated by multiplying the characteristic value by its changing speed. The characteristic momentum calculation formula is: ;
[0093] in, Representation characteristics In the The speed of change in a sample is defined as the rate of change of the characteristic value at two consecutive time points.
[0094] A3, standardize the characteristic momentum to eliminate dimension and scale differences. This is to subtract the mean of all characteristic momentum from the characteristic momentum and then divide it by the standard deviation. The standardization formula is: ;
[0095] in, and denote the mean and standard deviation of all feature momentums, respectively.
[0096] A4, using information entropy to evaluate the distinguishing ability of feature momentum, calculate the entropy value of each feature momentum, the entropy value is obtained by calculating the probability distribution of the feature momentum in different states. Then, calculate the entropy weight index of the feature momentum, which is the inverse of the entropy value and reflects the information content of the feature momentum.
[0097] A5, combine the standardized feature momentum and entropy weight index by multiplication, and calculate the feature discrimination ability index by combining the standardized feature momentum and entropy weight index .
[0098] The feature discrimination capability index is used to quantitatively evaluate the ability of a feature to distinguish between normal and abnormal states in switchgear insulation condition monitoring. The larger the index, the stronger the feature's ability to distinguish between normal and abnormal states, that is, the performance of the feature in different states is significantly different, and it can more effectively identify and predict insulation faults; conversely, the smaller the index, the weaker the feature's ability to distinguish between different states, and the difference in the feature value between normal and abnormal states is not obvious, which is not conducive to accurately identifying changes in insulation state. In switchgear insulation condition monitoring, selecting features with a high feature discrimination capability index can improve the detection accuracy and fault warning capability of the monitoring system, thereby ensuring the safe and reliable operation of the power system.
[0099] In the insulation status monitoring of switchgear, accurately identifying and distinguishing the characteristics of normal and abnormal states is the core element to ensure the safe operation of the power system. As a key component of the power system, the insulation status of the switchgear directly affects the stability and reliability of power transmission. Due to the complex and changeable operating environment of the switchgear, the insulating materials will gradually age over time, leading to potential failure risks. Therefore, real-time monitoring and accurate identification of the insulation status of the switchgear is particularly important. However, traditional monitoring methods have limitations such as strong subjectivity and poor real-time performance due to their reliance on manual inspections and offline testing, and are difficult to meet the requirements of modern power systems for efficiency and accuracy.
[0100] To solve these problems, the Feature Stabilization Divergence Index (FSDI) is introduced as an indicator to quantitatively evaluate the distinguishing ability of features in different states, which is of great significance. The Feature Stabilization Divergence Index can accurately quantify the performance difference of features in normal and abnormal states by comprehensively considering the change speed and information amount of the eigenvalues. Specifically, the index captures the dynamic changes of features through the concept of momentum, and quantifies the information amount of features through the entropy weight method, thereby comprehensively reflecting the distinguishing ability of features in different states. The larger the index value, the more significant the performance difference of the feature in different states, and the higher the distinguishing ability; conversely, the smaller the index value, the weaker the distinguishing ability of the feature.
[0101] In practical applications, selecting features with high characteristic steady-state separation index can significantly improve the detection accuracy and fault warning capability of the switchgear insulation state monitoring system. Features with high distinguishing ability can detect the degradation trend and potential faults of insulation materials earlier, realize real-time monitoring and accurate identification of insulation status, thereby reducing the failure rate of power systems and ensuring the safety and reliability of power transmission. Therefore, calculating the characteristic steady-state separation index not only optimizes traditional monitoring methods, but is also an important means to improve the intelligence level and operational safety of power systems. The introduction of this index provides a scientific and reasonable evaluation standard for power equipment status monitoring, which has broad application prospects and important practical significance.
[0102] The logic for obtaining the characteristic steady-state separation index is as follows:
[0103] B1, collect continuous monitoring data over a period of time, including characteristic values and corresponding environmental conditions (such as temperature, humidity, etc.).
[0104] The data is divided into several time periods, each of which contains data under the same environmental conditions or different environmental conditions.
[0105] B2, each characteristic and environmental condition is standardized to eliminate the effects of different dimensions and magnitudes.
[0106] In each time period Calculate the coefficient of variation of environmental conditions : ;
[0107] in, Indicates an environmental condition. and Respectively represent environmental conditions In time period The standard deviation and mean within .
[0108] B3, calculation features Environmental conditions affect the correction value: ;
[0109] in, It is the time period The coefficient of variation of the main environmental conditions within the
[0110] By dividing by the coefficient of variation of the environmental conditions, the coefficient of variation of the characteristic is corrected to reduce the impact of environmental factors on the stability of the characteristic.
[0111] B4, calculate the characteristic steady-state separation index by combining the coefficient of variation between time periods : ;
[0112] in, is the number of time periods; and Respectively in the time period and Internal, feature The coefficient of variation was corrected for the influence of environmental conditions.
[0113] The feature steady-state separation index quantifies the fluctuation of features between different time periods and provides an intuitive stability evaluation indicator.
[0114] The feature steady-state separation index is used to quantitatively evaluate the ability of a feature to distinguish between normal and abnormal states in switchgear insulation state monitoring. The larger the index, the stronger the ability of the feature to distinguish between normal and abnormal states, that is, the performance of the feature in different states is significantly different, and it can more effectively identify and predict insulation faults; conversely, the smaller the index, the weaker the ability of the feature to distinguish between different states, and the difference in the feature value between normal and abnormal states is not obvious, which is not conducive to accurately identifying changes in insulation state. In switchgear insulation state monitoring, selecting features with high feature steady-state separation index can improve the detection accuracy and fault warning capability of the monitoring system, thereby ensuring the safe and reliable operation of the power system.
[0115] The characteristic discrimination ability index and characteristic steady-state separation index were standardized and converted to the same scale.
[0116] The standardized feature discrimination ability index and feature steady-state separation index are used as the coordinate axes of the two-dimensional space to construct the feature space. The coordinates of each feature in the corresponding space are ;
[0117] By constructing the feature space, the comprehensive performance of each feature in terms of discrimination ability and stability can be intuitively represented
[0118] Determine the ideal feature point and set the coordinates of the ideal point in the feature space to (1,1), which means that the feature is optimal in both distinguishing ability and stability.
[0119] The ideal feature points provide a reference standard for evaluating how close each feature is to the ideal state.
[0120] The Euclidean distance between each feature point and the ideal feature point is calculated, and then the Euclidean distance is inverted to calculate the feature goodness index. For example, it can be calculated by the following calculation formula: ;
[0121] in, represents the feature goodness index; Represents the Euclidean distance between feature points and rational feature points.
[0122] The feature goodness index is used to quantitatively evaluate the comprehensive performance of each feature in terms of distinguishing ability and stability. The larger the value, the closer the feature is to the ideal state and the better the overall performance. That is, the feature is effective in distinguishing normal and abnormal states, and shows high stability in different time periods and environmental conditions. Conversely, the smaller the feature goodness index, the greater the deviation of the feature from the ideal state, the poorer the overall performance, which is not conducive to accurately identifying and predicting the insulation state changes of the switchgear.
[0123] According to the calculated feature goodness index, all features are sorted and arranged from large to small according to the index value.
[0124] Features whose feature goodness index is higher than the goodness threshold are selected as important features.
[0125] The present invention can intuitively evaluate the effectiveness of each feature in distinguishing normal and abnormal states and its stability under different time periods and environmental conditions by standardizing the feature distinguishing ability index and the feature steady-state separation index and constructing a two-dimensional feature space. By determining the ideal feature point and calculating the Euclidean distance between each feature point and the ideal feature point, and then calculating the feature goodness index by reversing the Euclidean distance, the comprehensive performance of each feature is effectively quantified. The larger the feature goodness index, the better the comprehensive performance of the feature in distinguishing ability and stability. Conversely, the smaller the goodness index, the poorer the comprehensive performance. By sorting and screening the feature goodness index, important features with high distinguishing ability and high stability can be scientifically selected, thereby improving the detection accuracy and fault warning capability of the switch cabinet insulation state monitoring system. The reliability and consistency of the selected features in practical applications are ensured, which helps to discover the degradation trend and potential faults of insulating materials at an early stage, ensure the safety and stability of the power system, significantly improve the performance and intelligence level of the monitoring system, and provide a scientific and reasonable evaluation standard for power equipment state monitoring.
[0126] like Figure 2 , the sequence insight module includes the following processing content:
[0127] S3.1, organize and standardize the historical data and real-time data of the selected important features to build a time series data set. Then, use data enhancement techniques (such as time series slicing, random noise addition, etc.) to generate more diverse training data to improve the robustness and generalization ability of the model.
[0128] S3.2, divide the time series data set into training set, validation set and test set. Use sliding window technique to split the time series data into subsequences of fixed length for input into LSTM model. Each window contains a fixed number of time steps to capture short-term and long-term dependencies.
[0129] S3.3, build a multi-layer LSTM network model, including input layer, multi-layer LSTM layer, fully connected layer and output layer. Use historical data to train the model and optimize the model parameters so that it can accurately capture the dynamic characteristics in the time series. During the training process, use the back propagation algorithm and gradient descent optimizer, such as Adam optimizer, to adjust the model weights and biases.
[0130] S3.4, after the model training is completed, the model is applied to real-time data prediction. For each time step, the difference between the model prediction value and the actual value is calculated to obtain the residual. The residual is used to evaluate the accuracy of the model prediction, and a larger residual may indicate a potential abnormal situation.
[0131] S3.5, use statistical process control methods, such as the exponentially weighted moving average (EWMA) control chart, to monitor residuals. The EWMA control chart smoothes short-term fluctuations and highlights long-term trends by weighted averaging the historical data of residuals. Set control limits and determine whether there are abnormalities based on the relationship between the residuals and the control limits.
[0132] S3.6, when the residual exceeds the control limit, it is marked as an outlier. The control limit is set based on the statistical properties of historical data, such as mean and standard deviation, to ensure reasonable sensitivity and accuracy. The detection of outliers helps to detect potential insulation degradation in a timely manner.
[0133] S3.7, as time goes by and the residual changes dynamically, the weighted moving average (EWMA) residual value and residual standard deviation are updated in real time to adjust the warning threshold. The adjustment formula of the warning threshold is: ;
[0134] in, Represents the residual, which is the smoothed residual value calculated by the exponentially weighted moving average (EWMA) method; is the residual standard deviation, which reflects the fluctuation range of the residual and is used to evaluate the dispersion of the residual; is the adjustment factor, which is used to adjust the sensitivity of the warning threshold.
[0135] The adjustment of the warning threshold is based on the statistical process control results, taking into account the fluctuation range and trend of the residual. By dynamically adjusting the warning threshold, the sensitivity and accuracy of the warning system under different operating conditions are ensured to avoid false alarms and missed alarms.
[0136] S3.8, when the residual exceeds the warning threshold, a warning signal is generated.
[0137] The present invention uses data enhancement technology and sliding window technology to construct a diversified time series data set, combines a multi-layer LSTM network model to capture time dynamic characteristics, and performs dynamic anomaly detection and early warning through residual analysis and EWMA control chart, realizing adaptive adjustment of the early warning threshold and generation of early warning signals, thereby significantly improving the detection accuracy and fault early warning capability of the switch cabinet insulation state monitoring system. Through comprehensive analysis of historical data and real-time data, the model can accurately predict the trend of insulation degradation, timely discover potential faults, and generate corresponding early warning signals. The adaptive early warning threshold ensures the sensitivity and accuracy of the early warning system under different operating conditions, avoids false alarms and missed alarms, and combines multi-level early warning signals to help operators quickly take corresponding maintenance and repair measures, effectively ensuring the safe and reliable operation of the power system. The overall process not only improves the intelligence level and decision-making ability of the monitoring system, but also continuously optimizes the model performance through real-time monitoring and feedback mechanisms, ensuring the long-term stability and reliability of the system in a complex operating environment.
[0138] The strategy matching module includes the following processing contents:
[0139] First, the real-time warning signal is matched with the fault type in the historical maintenance database, and the frequency and amplitude of the warning signal are analyzed using a Bayesian network or fuzzy logic model to determine the most likely fault type; then, combined with successful maintenance and overhaul cases in historical data, a rule-based expert system or machine learning model (such as random forest or support vector machine) is used to recommend the optimal maintenance strategy; the system automatically generates detailed overhaul steps, required spare parts and man-hour estimates, and generates maintenance work orders; the work orders are sent to on-site maintenance personnel through IoT devices, and real-time guidance and monitoring are provided to ensure the accuracy and timeliness of maintenance operations; finally, the maintenance results and new data are incorporated into the system through a feedback mechanism, and the decision model is continuously updated and optimized to achieve closed-loop management. This method innovatively combines historical data analysis, intelligent decision-making systems, and real-time feedback mechanisms, significantly improving the efficiency and accuracy of maintenance and overhaul, and ensuring the reliable operation of the power system.
[0140] The above formulas are all dimensionless and numerical calculations. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters in the formula are set by technicians in this field according to actual conditions.
[0141] The above description is only by way of illustration of certain exemplary embodiments of the present invention. It is undoubted that those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the above drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.
[0142] It should be noted that, in this article, if there are relational terms such as first and second, etc., they are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "including a..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.
[0143] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.
Claims
1. The switch cabinet insulation status monitoring and management system is characterized by: include: Data aggregation module, feature identification module, sequence insight module and strategy matching module; The data aggregation module collects switchgear data, performs multi-source data fusion and dynamic weight adjustment through principal component analysis, generates a fused data matrix, reconstructs and restores it to extract features, and passes the generated fused data matrix and extracted features to the feature identification module; The feature identification module analyzes the effectiveness of each feature in distinguishing normal and abnormal states to obtain the feature value distinguishing ability index, and combines the stability of the feature under different time periods and environmental conditions to obtain the feature steady-state separation index. The feature goodness index is further obtained through comprehensive analysis to screen out important features, and the selected important features are passed to the sequence insight module; The sequence insight module uses data enhancement technology to build a diverse time series data set based on the selected important features, and uses sliding window technology to segment the data and build a multi-layer LSTM network model. It performs dynamic anomaly detection and early warning through residual analysis and EWMA control charts, adjusts the early warning threshold in real time, generates early warning signals, and transmits the generated early warning signals to the strategy matching module. The strategy matching module matches the warning results with the historical maintenance methods and issues corresponding maintenance and repair strategies; The process of obtaining the characteristic value discrimination capability index is as follows: A1, divide the data set into normal state data and abnormal state data, extract the features of each sample, and construct a feature matrix, where The feature matrix representing the normal state data, A feature matrix representing abnormal state data; A2, marked as characteristic momentum is , calculated by multiplying the eigenvalue and its changing speed, the characteristic momentum calculation formula is: ; in, Representation characteristics In the The speed of change in a sample is defined as the rate of change of the characteristic value between two consecutive time points; A3, standardize the characteristic momentum to eliminate dimension and scale differences. Subtract the mean of all characteristic momentums from the characteristic momentum and divide it by the standard deviation. The standardization formula is: ; in, and Respectively represent the mean and standard deviation of all characteristic momentums; A4, using information entropy to evaluate the distinguishing ability of feature momentum, calculate the entropy value of each feature momentum, the entropy value is obtained by calculating the probability distribution of the feature momentum in different states; then, calculate the entropy weight index of the feature momentum, the entropy weight index is the inverse of the entropy value; A5, combine the standardized feature momentum and entropy weight index by multiplication, and calculate the feature discrimination ability index by combining the standardized feature momentum and entropy weight index ; The logic for obtaining the characteristic steady-state separation index is as follows: B1, collect continuous monitoring data over a period of time, including characteristic values and corresponding environmental conditions; Divide the data into several time periods, each of which contains data under the same environmental conditions or different environmental conditions; B2, standardize each feature and environmental condition; In each time period Calculate the coefficient of variation of environmental conditions : ; in, Indicates an environmental condition. and Respectively represent environmental conditions In time period standard deviation and mean within ; B3, calculation features Environmental conditions affect the correction value: ; in, It is the time period Coefficient of variation of the main environmental conditions within the B4, calculate the characteristic steady-state separation index by combining the coefficient of variation between time periods : ; in, is the number of time periods; and Respectively indicate the time period and Internal, feature The coefficient of variation was corrected for the influence of environmental conditions.
2. The switch cabinet insulation status monitoring and management system according to claim 1 is characterized in that: The data aggregation module includes the following processing contents: S1.1, collect different types of data from various sensors of the switchgear to ensure that the data covers different aspects of the switchgear operation; S1.2, preprocess each type of data; S1.3, construct a data matrix by chronologically arranging the standardized multi-source data , each column represents a data source, and each row represents data at a time point; S1.4, Calculation of data matrix The covariance matrix of , the formula is: ,in, is the sample size, where Represents the data matrix The transpose of S1.5, calculate the eigenvalues and eigenvectors of the covariance matrix, sort the eigenvalues in descending order, and select the first The eigenvector corresponding to the largest eigenvalue is constructed to construct the eigenvector matrix ; S1.6, project the data matrix onto the eigenvector matrix to obtain the fused data matrix , the formula is: ; S1.7, introduce a dynamic weight adjustment mechanism to dynamically adjust the weight of each data source in the fusion process according to the changes in real-time monitoring data; S1.8, reconstruct the fused principal component data to restore it to the scale of the original data. The reconstruction formula is: , Represents the data reconstruction matrix after principal component analysis fusion and projection.
3. The switch cabinet insulation status monitoring and management system according to claim 2 is characterized in that: The characteristic discrimination ability index and characteristic steady-state separation index were standardized and converted to the same scale; The standardized feature discrimination ability index and feature steady-state separation index are used as the coordinate axes of the two-dimensional space to construct the feature space. The coordinates of each feature in the corresponding space are ; Assume that the coordinates of the ideal point in the feature space are (1,1), indicating that the feature is optimal in both distinguishing ability and stability; Calculate the Euclidean distance between each feature point and the ideal feature point, then invert the Euclidean distance to calculate the feature goodness index, which is calculated using the following formula: ; in, represents the feature goodness index; Represents the Euclidean distance between feature points and rational feature points; According to the calculated feature goodness index, all features are sorted from large to small according to the index value; Features whose feature goodness index is higher than the goodness threshold are selected as important features.
4. The switch cabinet insulation status monitoring and management system according to claim 3 is characterized in that: The sequence insight module includes the following processing content: S3.1, organize and standardize the historical data and real-time data of the selected important features to construct a time series dataset; S3.2, divide the time series data set into training set, validation set and test set, and use sliding window technology to split the time series data into subsequences of fixed length for input into LSTM model, and each window contains a fixed number of time steps; S3.3, construct a multi-layer LSTM network model, including an input layer, a multi-layer LSTM layer, a fully connected layer and an output layer; S3.4, after the model training is completed, the model is applied to real-time data prediction, and for each time step, the difference between the model prediction value and the actual value is calculated to obtain the residual; S3.5, use an exponentially weighted moving average control chart to monitor the residuals, set control limits, and determine whether there are abnormalities based on the relationship between the residuals and the control limits; S3.6, when the residual exceeds the control limit, it is marked as an outlier; S3.7, as time goes by and the residual changes dynamically, the weighted moving average residual value and residual standard deviation are updated in real time, and then the warning threshold is adjusted. The adjustment formula of the warning threshold is: ; in, Represents the residual, which is the smoothed residual value calculated by the exponentially weighted moving average method; is the residual standard deviation, which is used to assess the dispersion of the residuals; is the adjustment coefficient, which is used to adjust the sensitivity of the warning threshold; S3.8, when the residual exceeds the warning threshold, a warning signal is generated.
5. The switch cabinet insulation status monitoring and management system according to claim 4, characterized in that: The strategy matching module includes the following processing contents: First, the real-time warning signal is matched with the fault type in the historical maintenance database, and the frequency and amplitude of the warning signal are analyzed using a Bayesian network to determine the fault type. Then, the optimal maintenance strategy is recommended based on the successful maintenance and overhaul cases in the historical data using a machine learning model. Detailed overhaul steps, required spare parts and labor time estimates are automatically generated, and a maintenance work order is generated. The work order is sent to on-site maintenance personnel through IoT devices. Finally, the maintenance results and new data are incorporated into the system through a feedback mechanism.
Citation Information
Cited By
Artificial intelligence model training method considering business scene requirements
CN119849664A
Power distribution room switch cabinet safety guarantee system integrated with multi-dimensional monitoring
CN122512662A
