Automatic monitoring system and method for electrical engineering

Abstract

The invention discloses an automatic monitoring system and method for electrical engineering, relates to the technical field of electrical engineering, and solves the technical problems of how to construct a physical connection topological graph based on collected data, and how to extract features in a time domain, a frequency domain and a time-frequency domain by using an advanced mathematical model so as to accurately reflect the running state of equipment. According to the invention, through comprehensive data acquisition and preprocessing, not only can various types of electrical equipment operation parameters be acquired, but also equipment state signals and environmental data can be collected, and the original data is optimized through means such as filtering, denoising and normalization, so that the basic data quality of subsequent analysis is ensured, and the analysis accuracy is improved. According to the method, wavelet denoising and self-adaptive piecewise linear fitting are adopted to retain abrupt change point features, features are extracted in multiple dimensions based on an equipment physical connection topological graph, so that equipment state evaluation is more accurate, and a multi-level and all-around early warning mechanism is realized through a mode of combining a static threshold, a dynamic threshold and a trend threshold.

Description

Technical Field

[0001] This invention relates to the field of electrical engineering technology, specifically to an automated monitoring system and method for electrical engineering. Background Technology

[0002] With the development of modern industry, the reliability and safety of electrical equipment have become particularly important. Traditional electrical equipment monitoring systems can usually only provide basic operating parameter monitoring, such as voltage, current, and power, lacking a comprehensive consideration of equipment status signals and environmental data.

[0003] According to patent application number CN202411186228.4, an automated monitoring system and method for electrical engineering is disclosed. The monitoring system includes: a data acquisition module for acquiring current signals from the electrical engineering; a preprocessing module for preprocessing the current signals to obtain valid current signals; and a first determination module for determining the loss value of the electrical engineering based on the valid current signals and comparing it with a preset loss value; and issuing an alarm when the determined loss value is greater than the preset loss value. Based on the detected current signals, calculations are performed using a simple algorithm, avoiding the need to process and analyze each data point, reducing the amount of data processing, lowering computational complexity, and improving the efficiency and accuracy of determining whether the electrical engineering is abnormal.

[0004] However, existing systems often lack effective preprocessing methods when processing collected data, making it difficult to ensure data quality and thus affecting the accuracy of subsequent analysis. Furthermore, in feature extraction and fault early warning, traditional methods often rely on fixed thresholds or simple mathematical models, which have limited ability to capture changes in equipment status under complex operating conditions, leading to untimely warnings or high false alarm rates. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides an automated monitoring system and method for electrical engineering, which solves the problem of how to construct a physical connection topology based on collected data and extract features in the time domain, frequency domain, and time-frequency domain using advanced mathematical models to accurately reflect the operating status of equipment.

[0006] To achieve the above objectives, the present invention provides the following technical solution: an electrical engineering automation monitoring system, comprising: The electrical equipment information acquisition module is used to collect the operating parameters, status signals and environmental data of electrical equipment, and to perform filtering, noise reduction, normalization and outlier removal on these raw data to generate preprocessed data. The central data processing module receives preprocessed data and performs structured transformation and feature extraction on it. It synchronizes data from different sensors based on a unified timestamp, automatically inserts event labels and expands the high sampling rate waveform window when a specific discrete event is detected, uses wavelet denoising and adaptive piecewise linear fitting techniques to preserve the features of abrupt change points, constructs graph structure data based on the physical connection topology of the monitored electrical equipment, and extracts features in the time domain, frequency domain, and time-frequency domain respectively. It also establishes a mathematical model to map various parameters into quantifiable monitoring indicators. The monitoring, analysis and early warning module is used to configure static thresholds, dynamic thresholds and trend thresholds based on monitoring indicators, and to cross-verify the final monitoring indicators based on different thresholds to determine whether to trigger different levels of early warning or generate normal monitoring signals. The monitoring information display module is used to show management personnel early warning information or normal monitoring information.

[0007] As a further aspect of the present invention, the data central processing module synchronizes the preprocessed data from different sensors according to a unified timestamp, and automatically inserts event tags and expands the high sampling rate waveform window when a switching action, protection trip, or voltage drop event is detected. Based on the physical connection topology of the monitored equipment, the preprocessed data is mapped into graph structure data, where nodes are equipment units, edges are electrical connection relationships, and node attributes are multi-dimensional time-series feature vectors. Statistical features, spectral features, and transient features are extracted in the time domain, frequency domain, and time-frequency domain respectively to form a multi-scale feature set.

[0008] As a further aspect of the present invention, the data central processing module combines multi-scale features with domain knowledge of equipment type and failure mode to construct a feature map with physical or functional meaning, calculates the SHAP importance score of each feature based on the XGBoost model, and filters pre-selected features that are higher than a preset threshold. Starting with pre-selected features, a recursive feature elimination method with 5-fold cross-validation is used. Under the condition of satisfying the lower limit of the number of features, the final strongly correlated feature subset is determined by combining SHAP sorting and RFE stability.

[0009] As a further embodiment of the present invention, the data central processing module is also used to perform weighted fusion of strongly correlated features to generate a final monitoring indicator, and transmit the indicator to the monitoring analysis and early warning module.

[0010] As a further aspect of the present invention, in the monitoring, analysis, and early warning module: Static thresholds are set based on equipment factory parameters, industry standards, or historical stable values. The dynamic threshold is based on historical data from the same period over the past 3 months. The mean μ and standard deviation σ are calculated using a 10-minute sliding window and set as [μ−3σ, μ+3σ]. The trend threshold is set by training an LSTM network to predict the indicator sequence for the next 15 minutes and setting a threshold for the slope of its first derivative.

[0011] As a further aspect of the present invention, the monitoring, analysis, and early warning module performs cross-verification of the final monitoring indicators based on different thresholds in the following manner: If a static anomaly, dynamic anomaly, and trend anomaly are all met simultaneously, a Level 1 warning is triggered. If a dynamic anomaly and trend anomaly, or a static anomaly and dynamic anomaly are all met, a Level 2 warning is triggered. If a single dynamic anomaly or trend anomaly is met, a Level 3 warning is triggered, and monitoring warning information is generated. If none of the final monitoring indicators meet the corresponding thresholds, a normal monitoring signal is generated.

[0012] As a further aspect of the present invention, it also includes an adaptive optimization module, which is used to receive normal monitoring signals and perform adaptive monitoring and analysis of electrical equipment based on the signals, and automatically label each normal monitoring signal with a multi-dimensional operating condition label, the operating condition label including load rate range, ambient temperature range and seasonal operating conditions. The labeled data is stored in a time-series database, and a dynamic baseline library is built based on the operating condition labels. The dynamic baseline library includes individual baselines, associated baselines, and group baselines.

[0013] As a further aspect of the present invention, the individual baseline calculates a Gaussian distribution of key indicators for each stable operating condition slice as a health fingerprint. The correlation baseline is based on the training of physical correlation models using normal data, including current-temperature regression models or three-phase current balance rules; The group baseline is formed by aggregating normal data from multiple devices of the same model to create the group performance distribution.

[0014] As a further aspect of the present invention, for individual baselines, the Mann-Kendall trend test is used to monitor the long-term monotonic changes of the baseline center value. When the changes are significant and the cumulative amount exceeds the threshold within N consecutive periods, the specific value of N is set by the operator to trigger a performance degradation warning. For the correlation baseline, when the absolute value of the model residual is greater than 3 times the historical standard deviation for 10 consecutive sampling points, a correlation anomaly warning is triggered. For the population baseline, when the target equipment indicator is in the bottom 10 percentile of the population for 24 consecutive hours, a population deviation warning is triggered.

[0015] An automated monitoring method for electrical engineering, the method specifically includes the following steps: Step 1: Collect the operating parameters, status signals and environmental data of electrical equipment, and perform preprocessing. Then, perform time synchronization, event labeling and feature extraction on the preprocessed data to obtain multi-dimensional features. Step 2: Combine multidimensional features with equipment knowledge to construct a feature map, select strongly correlated features and integrate them into the final monitoring indicators, and configure static, dynamic and trend thresholds for the final monitoring indicators; Step 3: Cross-verify the final monitoring indicators with thresholds at all levels, trigger different levels of early warning based on different abnormal combinations, use normal monitoring signals to automatically label the data with multi-dimensional working condition labels, and build a dynamic baseline library for individuals, associations and groups. Step 4: Perform trend analysis, correlation analysis, and group comparison analysis based on the dynamic baseline library to generate early warning information.

[0016] This invention provides an automated monitoring system and method for electrical engineering. Compared with the prior art, it has the following advantages: This invention, through comprehensive data acquisition and preprocessing, can not only collect operating parameters of various types of electrical equipment, but also collect equipment status signals and environmental data. It optimizes the raw data through filtering, denoising, and normalization to ensure the quality of the basic data for subsequent analysis. It uses wavelet denoising and adaptive piecewise linear fitting to retain the features of abrupt change points, and extracts features in multiple dimensions based on the physical connection topology of the equipment, making the equipment status assessment more accurate.

[0017] This invention differs from traditional fixed-threshold early warning methods. By combining static thresholds, dynamic thresholds, and trend thresholds, it achieves a multi-level, comprehensive early warning mechanism. In particular, the dynamic baseline library built based on normal monitoring signals, including individual, correlated, and group baselines, provides strong support for long-term tracking of equipment status, greatly improving the sensitivity and accuracy of early warnings. Attached Figure Description

[0018] Figure 1 This is a block diagram of the electrical engineering automation monitoring system of the present invention; Figure 2 This is a flowchart of the electrical engineering automation monitoring method of the present invention. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] First Embodiment Please see Figure 1 This application provides an electrical engineering automation monitoring system, comprising: The electrical equipment information acquisition module is used to collect multi-dimensional information, including electrical equipment operating parameters, equipment status signals, and environmental data. The operating parameters include electrical quantities such as voltage, current, power, and frequency. The equipment status signals include switch positions, alarm information, and fault codes. The environmental data includes temperature, humidity, and dust concentration. At the same time, the module performs filtering, noise reduction, normalization, and outlier removal on the collected raw data to obtain preprocessed data, which is then transmitted to the central data processing module.

[0021] The central data processing module is used for monitoring electrical equipment based on the acquired preprocessed data. It performs structured transformation and feature extraction on the preprocessed data, and maps various parameters into quantifiable monitoring indicators through an established mathematical model. The specific processing methods are as follows: Preprocessed data from different sensors are synchronized according to a unified timestamp reference. When discrete events such as switching action, protection tripping or voltage drop are detected, event tags are automatically inserted and the high sampling rate waveform window is extended by ±500ms before and after. Wavelet denoising and adaptive piecewise linear fitting are performed on continuous quantities to preserve the characteristics of abrupt change points. Based on the physical connection topology of the monitored electrical equipment, the preprocessed data is mapped into graph structure data. Specifically, the equipment unit is used as a node, the electrical connection relationship is used as an edge, and the multi-dimensional time-series feature vector of the equipment unit is used as the node attribute to obtain the physical connection topology. Then, features are extracted in the time domain, frequency domain, and time-frequency domain according to the obtained physical connection topology. The time domain features include statistical quantities such as mean, variance, peak factor, and kurtosis coefficient. The changing trend of equipment parameters in different time periods is calculated by using a sliding time window. The frequency domain features are converted into a spectrum diagram by using fast Fourier transform to extract the energy proportion, characteristic frequency points, and harmonic distortion rate of each frequency band. The time-frequency domain features are captured by short-time Fourier transform and wavelet transform to capture the transient features of the signal in a specific time-frequency region, such as the duration and frequency distribution of the impact pulse. Based on the obtained features, these features are combined with knowledge from areas such as equipment type and failure modes to form a feature map with clear physical or functional meaning. Methods such as recursive feature elimination or SHAP value analysis are then used to filter out the most critical and strongly correlated features for the monitored target from the map. The specific processing methods are as follows: The XGBoost baseline model is used to initially train the features. The average absolute value of the SHAP of each feature on all samples in the validation set is calculated as its importance score. Then, an importance threshold is set, and features with importance greater than the threshold are selected and recorded as pre-selected features. At the same time, starting from the selected pre-selected feature set, 5-fold cross-validation RFE is used. A lower limit for the number of features is set, and the features eliminated in each round of RFE are recorded. Combining SHAP ranking and RFE stability, the final feature subset is determined and the strongly correlated features are output. The obtained strongly correlated features are weighted and fused, and the final monitoring index is calculated according to the formula = The final monitoring indicators are calculated and then transmitted to the monitoring analysis and early warning module.

[0022] The monitoring, analysis, and early warning module is used to perform early warning analysis based on the acquired final monitoring indicators. The specific analysis and processing methods are as follows: Based on business characteristics, static thresholds, dynamic thresholds, and trend thresholds are configured for each monitoring indicator. The static threshold is set by collecting equipment factory parameters and industry standards, setting a benchmark range, and for non-equipment indicators, setting static thresholds based on historical stable values ​​of the business. The dynamic threshold is set by extracting historical data from the same period over the past three months to remove noise and outliers, using a 10-minute sliding window to calculate the mean μ and standard deviation σ of the data within the window in real time, and setting the dynamic threshold range to [μ-3σ, μ+3σ]. The trend threshold is set by using indicator data from the past 6 months to train an LSTM network with a sampling frequency of 1 minute / time. The input features include historical indicator values, timestamps, and associated indicators. The output is the indicator sequence for the next 15 minutes. The first derivative of the predicted sequence is used to obtain the slope of change, and the trend threshold is set accordingly. Next, the final monitoring indicators are cross-verified with the thresholds. If the static anomaly, dynamic anomaly, and trend anomaly are all met simultaneously, a level one warning is triggered. If the dynamic anomaly and trend anomaly, or the static anomaly and dynamic anomaly are both met, a level two warning is triggered. If only a dynamic anomaly or trend anomaly is met, a level three warning is triggered, and monitoring warning information is generated. If none of the final monitoring indicators meet the corresponding thresholds, a normal monitoring signal is generated, and the monitoring warning information is transmitted to the monitoring information display module.

[0023] The monitoring information display module is used to display the acquired monitoring and early warning information to the relevant management personnel.

[0024] Second Embodiment As a second embodiment of the present invention, it is implemented based on the first embodiment, and the difference from the first embodiment is as follows: The monitoring, analysis and early warning module transmits the generated normal monitoring signals to the adaptive optimization module. This module is used to monitor and analyze electrical equipment based on the normal monitoring signals, automatically label each normal data with multi-dimensional operating condition tags, such as load rate >80%, ambient temperature 25-30℃, summer operating condition, and store them in the time series database. At the same time, a dynamic baseline library is built based on the obtained multi-dimensional operating condition tags, which specifically includes individual baselines: for each stable operating condition slice, the Gaussian distribution of key indicators including temperature, vibration and other indicators is calculated and used as the health fingerprint of the equipment under that operating condition. Correlation baseline: Using a large amount of normal data, train correlation models between key indicators, such as current-temperature regression models and three-phase current balance rules; Group baseline: Aggregate normal data from multiple devices of the same model to form a group performance distribution for horizontal benchmarking; Based on the obtained dynamic baseline library, early warning analysis is performed. For individual baselines, the long-term trend of the center value of each working condition slice baseline is monitored. When the Mann-Kendall trend test is used to find that the index shows a statistically significant monotonically increasing trend over N consecutive periods (the specific value of N is set by the operator), and the cumulative change exceeds the threshold, a performance degradation early warning is triggered. For the correlation baseline, real-time data is input into the correlation model. When the model residuals continuously exceed the normal fluctuation range, such as when the absolute value of the residuals is greater than 3 times the historical standard deviation for 10 consecutive points, it indicates that the inherent physical relationship has been destroyed, triggering a correlation anomaly warning. For the population baseline, the indicator value of the target device is compared with the population baseline in real time. If the indicator value of the device continues to deviate from the population health range, such as being in the bottom 10 percentile of the population for 24 consecutive hours, a population deviation warning is triggered, indicating that it has become a weak link that needs to be focused on. Based on the above early warning analysis, normal monitoring information is generated and transmitted to the monitoring information display module.

[0025] The monitoring information display module is used to display the acquired normal monitoring information to the corresponding management personnel.

[0026] Third Embodiment As a third embodiment of the present invention, the focus is on combining the implementation processes of the first and second embodiments.

[0027] Fourth embodiment Please see Figure 2 This application provides an automated monitoring method for electrical engineering, which specifically includes the following steps: Step 1: Collect the operating parameters, status signals and environmental data of electrical equipment, and perform preprocessing. Then, perform time synchronization, event labeling and feature extraction on the preprocessed data to obtain multi-dimensional features. Step 2: Combine multidimensional features with equipment knowledge to construct a feature map, select strongly correlated features and integrate them into the final monitoring indicators, and configure static, dynamic and trend thresholds for the final monitoring indicators; Step 3: Cross-verify the final monitoring indicators with thresholds at all levels, trigger different levels of early warning based on different abnormal combinations, use normal monitoring signals to automatically label the data with multi-dimensional working condition labels, and build a dynamic baseline library for individuals, associations and groups. Step 4: Perform trend analysis, correlation analysis, and group comparison analysis based on the dynamic baseline library to generate early warning information.

[0028] Some of the data in the above formulas are numerical calculations with dimensions removed, and the contents not described in detail in this specification are all prior art known to those skilled in the art.

[0029] The above embodiments are only used to illustrate the technical methods of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical methods of the present invention without departing from the spirit and scope of the technical methods of the present invention.

Claims

1. An automated monitoring system for electrical engineering, characterized in that, include: The electrical equipment information acquisition module is used to collect the operating parameters, status signals and environmental data of electrical equipment, and to perform filtering, noise reduction, normalization and outlier removal on these raw data to generate preprocessed data. The central data processing module receives preprocessed data and performs structured transformation and feature extraction on it. It synchronizes data from different sensors based on a unified timestamp, automatically inserts event labels and expands the high sampling rate waveform window when a specific discrete event is detected, uses wavelet denoising and adaptive piecewise linear fitting techniques to preserve the features of abrupt change points, constructs graph structure data based on the physical connection topology of the monitored electrical equipment, and extracts features in the time domain, frequency domain, and time-frequency domain respectively. It also establishes a mathematical model to map various parameters into quantifiable monitoring indicators. The monitoring, analysis and early warning module is used to configure static thresholds, dynamic thresholds and trend thresholds based on monitoring indicators, and to cross-verify the final monitoring indicators based on different thresholds to determine whether to trigger different levels of early warning or generate normal monitoring signals. The monitoring information display module is used to show management personnel early warning information or normal monitoring information.

2. The electrical engineering automation monitoring system according to claim 1, characterized in that, The central data processing module synchronizes preprocessed data from different sensors with a unified timestamp, and automatically inserts event tags and expands the high sampling rate waveform window when a switch action, protection trip, or voltage drop event is detected. Based on the physical connection topology of the monitored equipment, the preprocessed data is mapped into graph structure data, where nodes are equipment units, edges are electrical connection relationships, and node attributes are multi-dimensional time-series feature vectors. Statistical features, spectral features, and transient features are extracted in the time domain, frequency domain, and time-frequency domain respectively to form a multi-scale feature set.

3. The electrical engineering automation monitoring system according to claim 2, characterized in that, The central data processing module combines multi-scale features with domain knowledge of equipment type and failure mode to construct a feature map with physical or functional meaning. Based on the XGBoost model, it calculates the SHAP importance score of each feature and filters out pre-selected features that are higher than a preset threshold. Starting with pre-selected features, a recursive feature elimination method with 5-fold cross-validation is used. Under the condition of satisfying the lower limit of the number of features, the final strongly correlated feature subset is determined by combining SHAP sorting and RFE stability.

4. The electrical engineering automation monitoring system according to claim 3, characterized in that, The central data processing module is also used to perform weighted fusion of strongly correlated features to generate the final monitoring indicator, and transmit the indicator to the monitoring analysis and early warning module.

5. An electrical engineering automation monitoring system according to claim 1, characterized in that, In the monitoring, analysis and early warning module: Static thresholds are set based on equipment factory parameters, industry standards, or historical stable values. The dynamic threshold is based on historical data from the same period over the past 3 months. The mean μ and standard deviation σ are calculated using a 10-minute sliding window and set as [μ−3σ, μ+3σ]. The trend threshold is set by training an LSTM network to predict the indicator sequence for the next 15 minutes and setting a threshold for the slope of its first derivative.

6. The electrical engineering automation monitoring system according to claim 1, characterized in that, The monitoring, analysis, and early warning module performs cross-verification of the final monitoring indicators based on different thresholds in the following manner: If a static anomaly, dynamic anomaly, and trend anomaly are all met simultaneously, a Level 1 warning is triggered. If a dynamic anomaly and trend anomaly, or a static anomaly and dynamic anomaly are all met, a Level 2 warning is triggered. If a single dynamic anomaly or trend anomaly is met, a Level 3 warning is triggered, and monitoring warning information is generated. If none of the final monitoring indicators meet the corresponding thresholds, a normal monitoring signal is generated.

7. An electrical engineering automation monitoring system according to claim 1, characterized in that, It also includes an adaptive optimization module, which receives normal monitoring signals and performs adaptive monitoring and analysis of electrical equipment based on the signals. It automatically labels each normal monitoring signal with multi-dimensional operating condition tags, including load rate range, ambient temperature range and seasonal operating conditions. The labeled data is stored in a time-series database, and a dynamic baseline library is built based on the operating condition labels. The dynamic baseline library includes individual baselines, associated baselines, and group baselines.

8. An electrical engineering automation monitoring system according to claim 7, characterized in that, The individual baselines calculate a Gaussian distribution of key indicators for each stable operating condition slice as a health fingerprint. The correlation baseline is based on the training of physical correlation models using normal data, including current-temperature regression models or three-phase current balance rules; The group baseline is formed by aggregating normal data from multiple devices of the same model to create the group performance distribution.

9. An electrical engineering automation monitoring system according to claim 1, characterized in that, For individual baselines, the Mann-Kendall trend test is used to monitor long-term monotonic changes in the baseline center value. When the change is significant and the cumulative amount exceeds the threshold within N consecutive periods, the specific value of N is set by the operator, triggering a performance degradation warning. For the correlation baseline, when the absolute value of the model residual is greater than 3 times the historical standard deviation for 10 consecutive sampling points, a correlation anomaly warning is triggered. For the population baseline, when the target equipment indicator is in the bottom 10 percentile of the population for 24 consecutive hours, a population deviation warning is triggered.

10. An automated monitoring method for electrical engineering, executed by the automated monitoring system for electrical engineering according to any one of claims 1-9, characterized in that, The method specifically includes the following steps: Step 1: Collect the operating parameters, status signals and environmental data of electrical equipment, and perform preprocessing. Then, perform time synchronization, event labeling and feature extraction on the preprocessed data to obtain multi-dimensional features. Step 2: Combine multidimensional features with equipment knowledge to construct a feature map, select strongly correlated features and integrate them into the final monitoring indicators, and configure static, dynamic and trend thresholds for the final monitoring indicators; Step 3: Cross-verify the final monitoring indicators with thresholds at all levels, trigger different levels of early warning based on different abnormal combinations, use normal monitoring signals to automatically label the data with multi-dimensional working condition labels, and build a dynamic baseline library for individuals, associations and groups. Step 4: Perform trend analysis, correlation analysis, and group comparison analysis based on the dynamic baseline library to generate early warning information.

Images