Capacitor pin state intelligent diagnosis system and method based on multi-source data fusion

Abstract

The application discloses a capacitor pin state intelligent diagnosis system and method based on multi-source data fusion, relates to the technical field of power equipment state monitoring, and collects sensing data of a pin area under the operation state of a capacitor, generates time-synchronized sensing data sequences, associates the sensing data sequences with spatial coordinates, and generates a sensing data synchronization set with space-time correspondence; basic physical features of a pin structure state are extracted, and sensing parameter coupling features are constructed based on the basic physical features; the basic physical features and the sensing parameter coupling features are input into a pin state recognition model generated through training to generate pin state recognition results; according to the pin state recognition results and the basic physical features, pin state evolution grade indexes are calculated, and corresponding maintenance decision instructions are output, thereby improving the accuracy and monitoring efficiency of capacitor pin diagnosis.

Description

Technical Field

[0001] This invention relates to the field of power equipment condition monitoring technology, specifically to an intelligent diagnostic system and method for capacitor pin status based on multi-source data fusion. Background Technology

[0002] During the operation of existing power equipment, capacitor pins, as important structural parts for electrical connection and mechanical support, are susceptible to thermal stress, mechanical vibration, environmental corrosion and contact aging under long-term operating conditions. This may lead to abnormal conditions such as loosening, poor connection, increased contact resistance or local overheating, which may cause equipment failure or even safety accidents. The main methods for detecting the status of capacitor pins include manual inspection, single sensor monitoring, and status judgment methods based on fixed rules. These methods have problems such as long detection cycles, strong subjectivity, and difficulty in achieving continuous monitoring. They also cannot fully reflect the true operating status of the pins under complex working conditions. Some existing intelligent diagnostic methods fail to effectively combine the correlation characteristics between different physical quantities, resulting in low identification accuracy in scenarios of environmental changes, load fluctuations, or early latent faults. The lack of spatial correlation and collaborative acquisition mechanisms between multi-source data makes it difficult for the collected data to form discriminative feature information, thus limiting the effectiveness of subsequent intelligent diagnostic methods. Therefore, how to construct a multi-source information collaborative acquisition mechanism based on the structural characteristics of capacitor pins, and how to achieve effective fusion of multi-dimensional operational information and intelligent status diagnosis by optimizing the sensing and detection structure and data feature extraction method, thereby improving the accuracy and real-time performance of anomaly identification and reducing reliance on human intervention, has become a technical problem that needs to be solved in this field. Summary of the Invention

[0003] The purpose of this invention is to provide an intelligent diagnostic system and method for capacitor pin status based on multi-source data fusion, so as to solve the problems raised in the prior art.

[0004] To address the aforementioned technical problems, this invention provides the following technical solution: an intelligent diagnostic method for capacitor pin status based on multi-source data fusion, the method comprising: Step S100: Based on the historical operation records of the capacitor pins, obtain the operation data during the operation process, analyze the operation data, identify the spatial distribution of abnormal areas, determine the key monitoring areas, establish the matching relationship between the sensors and the key monitoring areas, and configure the spatial distribution of the sensors. Step S200: Obtain the spatial structure position of the capacitor pins, and establish a spatial coordinate system for the pin area based on the spatial structure position. Based on the spatial distribution configuration, extract the spatial coordinates corresponding to each sensing data. Collect sensing data of the pin area while the capacitor is running. Time-align the sensing data to generate a time-synchronized sensing data sequence. Associate the sensing data sequence with the spatial coordinates to generate a sensing data synchronization set with spatiotemporal correspondence. Step S300: Based on the sensor data synchronization set, extract the basic physical features of the pin structure state, and construct the sensor parameter coupling features based on the basic physical features; Step S400: Input the basic physical features and sensing parameter coupling features into the trained pin state recognition model to generate pin state recognition results; Step S500: Based on the pin status identification results and basic physical characteristics, calculate the pin status evolution level index and output the corresponding maintenance decision command.

[0005] Furthermore, the spatial distribution configuration in step S100 includes: The operational data during operation includes temperature data, vibration data, and electrical signal data. The temperature data is arranged in a time series. Along the axial direction of the capacitor pins, the pins are divided into multiple spatial nodes. The temperature change gradient between adjacent spatial nodes is calculated as A = (a2 - a1) / b, where a1 and a2 represent the temperatures of the previous and next spatial nodes, respectively, and b is the axial distance between the two nodes. The temperature change sequence of each spatial node over time is summarized, and the instantaneous temperature rise rate of the spatial node is calculated as B = (a4 - a3) / c, where a3 and a4 represent the temperatures of the previous and current moments, respectively, and c represents the time interval. An axial temperature gradient threshold and a temperature rise rate threshold are set and compared with the temperature change gradient and instantaneous temperature rise rate of the spatial nodes. If the temperature change gradient of a spatial node is greater than the axial temperature gradient threshold or the instantaneous temperature rise rate of the spatial node is greater than the temperature rise rate threshold, the spatial node is marked as a temperature anomaly point. The temperature anomalies of all time nodes are summarized to obtain the temperature anomaly region. The collected vibration data is denoised and segmented according to the time series. Fourier transform is used to perform frequency domain transformation on the segmented vibration signals to calculate the amplitude characteristics and energy distribution characteristics of the vibration signals in each time period. The amplitude characteristic represents the maximum spectral amplitude of the vibration signal within the corresponding time period. The energy distribution characteristic represents the proportion of high-frequency energy to total energy, where high-frequency energy is the sum of squares of the amplitudes of all spectra within the high-frequency band, and total energy is the sum of squares of the amplitudes of all spectra within the entire frequency band. A threshold for amplitude characteristics and a threshold for energy distribution characteristics are set, and the amplitude characteristics and energy distribution characteristics of each time period are compared. When either the amplitude characteristic threshold or the energy distribution characteristic threshold is greater than the threshold, the corresponding time period is marked as a vibration anomaly segment. The vibration anomaly region is determined by combining the spatial mapping relationship of the vibration anomaly segments in each time period. The acquired electrical signal is filtered to extract the time-domain waveform feature parameters of the electrical signal. The time-domain waveform feature parameters include peak value, mean value, variance and fluctuation amplitude. A threshold range is set for each time-domain waveform feature parameter. When any time-domain waveform feature parameter exceeds the threshold range, the corresponding time period is marked as an abnormal area of ​​the electrical signal. Based on temperature anomaly regions, vibration anomaly regions, and electrical signal anomaly regions, historical operation records containing various anomaly regions are statistically analyzed. The frequency of anomaly occurrence, anomaly intensity index, and anomaly duration of each space node in the historical operation records are calculated. The frequency of anomaly occurrence is the number of times a space node is marked as an anomaly state in the historical operation records divided by the total number of times all space nodes are marked as an anomaly state. The anomaly intensity index is the average value of the difference between the anomaly feature parameter and the corresponding threshold. The duration of the anomaly is the length of time during which the anomaly occurs continuously. The frequency of anomalies, the intensity index of anomalies, and the duration of anomalies are normalized. The weights of the frequency of anomalies, the intensity index of anomalies, and the duration of anomalies are preset. The normalized frequency of anomalies, the intensity index of anomalies, and the duration of anomalies are weighted and summed to calculate the comprehensive anomaly evaluation index of spatial nodes. Based on the comprehensive anomaly evaluation index, all spatial nodes are sorted from high to low according to the comprehensive anomaly evaluation index, and the top few spatial regions in terms of comprehensive anomaly evaluation index are selected as key monitoring areas. The number of occurrences of each anomaly type within the key monitoring area is counted, and then divided by the total number of occurrences of all anomaly types within the key monitoring area to calculate the anomaly percentage corresponding to each anomaly type. Based on the proportion of each abnormality type in the key monitoring area, if the proportion of temperature abnormality type is higher than the preset threshold, a temperature sensor is configured; if the proportion of vibration abnormality type is higher than the preset threshold, a vibration sensor is configured; if the proportion of electrical signal abnormality is higher than the preset threshold, an electrical signal acquisition unit is configured; if the proportion of two or more abnormality types is higher than the preset threshold, multiple types of sensors are configured for collaborative monitoring to form a spatial distribution configuration scheme for multiple types of sensors. By comprehensively utilizing temperature data, vibration data, and electrical signal data, the pin operating status is jointly characterized from multiple dimensions, including thermal characteristics, mechanical state, and electrical characteristics. This avoids the limitations of analysis from a single data source, thereby significantly improving the accuracy and reliability of anomaly identification. By dividing the space nodes along the pin axis and combining the temperature gradient and temperature rise rate with time series analysis, the coupled analysis of "spatial distribution + time evolution" was realized; at the same time, the vibration anomaly segment was mapped to the spatial location, so that the anomaly detection result has a clear spatial orientation and effectively improves the positioning accuracy of the anomaly area. Key feature parameters are extracted for different types of data, and abnormal states are identified by threshold judgment, so that anomaly detection can reflect both mutation characteristics and capture the gradual deterioration process, thereby improving the ability to identify early anomalies. By normalizing and weighting the frequency of anomalies, anomaly intensity indicators, and anomaly duration, a comprehensive anomaly evaluation index for spatial nodes is constructed, enabling a quantitative description of anomaly states and transforming anomaly analysis from "whether it is anomaly" to "anomaly degree assessment". Based on comprehensive anomaly evaluation indicators, spatial nodes are sorted and key monitoring areas are selected, so that monitoring resources can be concentrated in high-risk areas, avoiding the waste of resources caused by uniform global deployment and improving the overall monitoring efficiency of the system.

[0006] Furthermore, generating the sensor data synchronization set in step S200 includes: The three-dimensional spatial structure of the capacitor pins is obtained, the spatial position information of each part of the pins is determined, and a spatial coordinate system of the pin area is established based on the spatial position information. The spatial coordinate system takes the preset pin endpoint as the origin and the axial and radial directions of the pins as coordinate axes. Based on the spatial distribution configuration of the sensors, the spatial coordinates corresponding to the sensing data collected by each sensor are extracted. When the capacitor is in operation, each sensor is activated to start collecting sensing data from the pin area. Using timestamp synchronization technology, the sensing data collected by different sensors are time-aligned, and the sensing data collected by each sensor are arranged in chronological order to form a time-synchronized sensing data sequence. By establishing a data mapping relationship, the sensor data sequence is paired with the corresponding spatial coordinates to generate a sensor data synchronization set with spatiotemporal correspondence. By acquiring the three-dimensional spatial structure of the capacitor pins and constructing a unified spatial coordinate system, the spatial position information of each part of the pins is standardized and expressed, so that the data collected by each sensor can be accurately mapped to the specific spatial position, thereby significantly improving the spatial positioning accuracy of the data. A spatial coordinate system is established with the pin endpoint as the origin and the axial and radial axes as the coordinate axes, providing a unified reference benchmark for sensors of different types and locations, effectively avoiding data deviation problems caused by differences in installation positions, and improving the consistency and comparability of multi-sensor data; By introducing timestamp synchronization technology, data collected by different sensors are time-aligned and a time-synchronized sensor data sequence is constructed to ensure that multi-source data are analyzed on the same time scale, thereby improving the accuracy of subsequent data fusion and anomaly analysis. By establishing a mapping relationship between sensor data and spatial coordinates, a synchronized set of sensor data with spatiotemporal correspondence is generated, enabling the data to possess not only temporal attributes but also clear spatial attributes, thereby realizing the transformation from single time series analysis to "time-space" multidimensional analysis.

[0007] Furthermore, the construction of sensing parameter coupling features in step S300 includes: Feature extraction is performed on the temperature data, vibration data and electrical signal data of each spatial node in the sensor data synchronization set under each time series to construct a basic physical feature parameter set. The basic physical features of the temperature data include the average temperature, the maximum temperature, the minimum temperature and the temperature change rate. The basic physical characteristics of the vibration data include the average vibration amplitude, the peak vibration value, and the frequency band energy distribution. The basic physical characteristics of the electrical signal data include peak voltage, average voltage, and voltage fluctuation amplitude. The basic physical features of each spatial node under the same time series are normalized, and the normalized basic physical features are spliced ​​together according to temperature, vibration and electrical signal to construct a cross-modal feature combination vector. Based on the cross-modal feature combination vector, sensor parameter coupling is constructed according to the sensor parameter type. The sensor parameter coupling includes temperature-vibration coupling term, temperature-electrical signal coupling term and vibration-electrical signal coupling term. For each coupling term, based on the value of the corresponding feature sequence in the time dimension, the Pearson correlation coefficient between the two corresponding sensing parameters is calculated. The Pearson correlation coefficient is obtained by dividing the covariance of the coupling feature vectors of the two parameters by the product of their respective standard deviations. The Pearson correlation coefficient is used as the correlation index of the coupling of sensing parameters. Based on the aforementioned correlation index, the characteristics of the sensing parameter coupling are determined. The characteristic of the temperature-vibration coupling term is the Pearson correlation coefficient between temperature data and vibration data; the characteristic of the temperature-electrical signal coupling term is the Pearson correlation coefficient between temperature data and electrical signal data; and the characteristic of the vibration-electrical signal coupling term is the Pearson correlation coefficient between vibration data and electrical signal data. By extracting the basic physical features of temperature, vibration and electrical signal data of each spatial node in the synchronous collection of sensor data under each time series, and normalizing each feature, different types of sensor data can be analyzed at a unified scale, thus providing a reliable foundation for cross-modal fusion. By splicing together the normalized basic physical features of temperature, vibration and electrical signals from multiple sources to form a cross-modal feature combination vector, a comprehensive representation of multiple sensor information is achieved, which improves the richness of data expression and the comprehensiveness of analysis. By constructing coupling terms for sensing parameters such as temperature-vibration, temperature-electrical signal, and vibration-electrical signal, and calculating the Pearson correlation coefficient of each coupling term, the correlation between different sensing parameters can be quantified, revealing their inherent coupling laws and providing a scientific basis for subsequent state analysis, anomaly diagnosis, and system optimization.

[0008] Furthermore, the training process of the pin status recognition model in step S400 includes: Based on the sensor data synchronization set, the pin status marked in the historical operation record is obtained. The pin status includes normal status, slightly abnormal status and severely abnormal status. The basic physical features and sensor parameter coupling features under the corresponding time series are matched with the pin status to construct a labeled training dataset. The training dataset is preprocessed, including interpolating and completing missing data, removing or correcting outliers, and normalizing the basic feature parameters. Using the aforementioned basic physical characteristics and sensing parameter coupling characteristics as model input, the pin state is assigned adjacent positive integer values. These pin state assignments serve as supervision labels. By training the deep learning model, a pin state recognition model is constructed. The training process involves initializing the network structure and parameters of each layer of the deep learning model, pre-setting several hidden layers in the deep learning model, and inputting the basic physical features and sensing parameter coupling features into the deep learning model. After layer-by-layer forward propagation, the output value of the last hidden layer is obtained. Based on the output value and pin state assignment, the corresponding classification error is calculated. The errors of each category are weighted and summed according to preset weights to obtain the total loss function. Based on the total loss function, the partial derivative of the total loss function with respect to the prediction result of the output layer is calculated starting from the output layer. The calculation is performed layer by layer from back to front according to the network connection relationship, and the error signal of each hidden layer is calculated in turn. Based on the error signal of each hidden layer, the parameter gradient of each hidden layer is calculated in combination with the corresponding layer input value, and the network weights and bias parameters are updated using the gradient descent method. Repeat the forward propagation, classification error calculation, loss function construction, backpropagation, and parameter update process until training ends when any of the following stopping conditions are met: the total loss function value is less than the preset loss threshold; the change in loss between two adjacent iterations is less than the preset convergence threshold; the number of training rounds reaches the preset maximum number of iterations. After training is completed, the pin status recognition model is obtained. By combining the labeled pin states in historical operation records, the basic physical features and sensing parameter coupling features are matched with the state labels to construct a labeled training dataset, enabling the model to fully learn the feature distribution patterns under different operating states, thereby improving the targeting and effectiveness of model training. By performing missing data interpolation to complete the training dataset, removing or correcting outliers, and normalizing features, data noise and abnormal interference are effectively reduced, data quality is improved, and the model's adaptability and robustness to real-world working conditions are enhanced. By using both basic physical characteristics and sensing parameter coupling characteristics as model inputs, the model can comprehensively utilize single parameter information and multi-parameter correlation information, enabling it to characterize the pin's operating state from multiple dimensions, thereby significantly improving the accuracy of state recognition. By constructing a deep learning model with multiple hidden layers, and performing layer-by-layer nonlinear mapping and feature extraction on the input features, it is possible to effectively uncover the deep correlation between different sensing parameters under complex working conditions and improve the ability to identify complex states. The trained pin status recognition model can automatically identify the operating status of capacitor pins and distinguish between normal, slightly abnormal, and severely abnormal states, thereby providing strong support for equipment operation monitoring, fault warning, and maintenance decision-making.

[0009] Furthermore, the maintenance decision instructions output in step S500 include: During capacitor operation, based on the pin state recognition model, the sensor data collected in real time by each sensor is continuously received, and a preset time window is set. For the sensor data in each time window, the basic physical features and sensor parameter coupling features are extracted and input into the pin state recognition model to obtain the pin state recognition result in the corresponding time window. Based on the time window, a pin state sequence under continuous time series is generated. Based on the pin state sequence, the state changes between the next time window and the previous time window are calculated, a state transition sequence is constructed, and the duration, transition frequency, and transition direction of each pin state within the two time windows are statistically analyzed. The transition frequency is calculated by dividing the number of transitions between different states by the duration of the two time windows. The transition direction includes a degradation direction and a recovery direction. The degradation direction is the change of the pin state from a normal state to an abnormal state or from a slight abnormality to a severe abnormality; the recovery direction is the change of the pin state from an abnormal state to a normal state or from a severe abnormality to a slight abnormality. Based on the state transition sequence and basic physical characteristics, the state deterioration rate and the anomaly accumulation intensity are calculated. The state deterioration rate is calculated by counting the number of deterioration directions within two time windows and dividing by the duration of the two time windows. The anomaly accumulation intensity is the cumulative value of the deviation between the basic physical characteristics under the abnormal state and the normal threshold within two time windows. The duration, transition frequency, state deterioration rate, and anomaly accumulation intensity are normalized and then weighted and summed according to preset weights to calculate the pin state evolution level index. When the pin status evolution level index is lower than the first preset threshold, it is determined to be a stable operating state and a regular monitoring command is output; when the pin status evolution level index is between the first preset threshold and the second preset threshold, it is determined to be a warning state and an enhanced monitoring and preventive maintenance command is output; when the pin status evolution level index is higher than the second preset threshold, it is determined to be a high-risk state and a maintenance decision command for repairing the pin is output. By introducing a time window mechanism during capacitor operation, real-time sensing data is continuously collected and segmented for analysis. Combined with the pin status recognition model, a pin status sequence under continuous time series is generated, realizing dynamic and continuous monitoring of the device's operating status and overcoming the problem of poor timeliness of traditional static detection methods. By constructing state transition sequences between adjacent time windows and systematically modeling state changes, the evolution of pin states at different time scales can be effectively characterized, transforming state assessment from single-point judgment to process analysis and improving the ability to identify potential fault evolution trends. By comprehensively statistically analyzing the duration, transfer frequency, and transfer direction of pin states, and combining the distinction between degradation and recovery directions, the system can perform quantitative analysis of its operating state from multiple dimensions such as time characteristics, change frequency, and evolution trend, thereby improving the comprehensiveness and refinement of state assessment. By defining two key indicators, the rate of state deterioration and the intensity of anomaly accumulation, respectively reflecting the speed of state degradation and the cumulative effect of anomaly severity, the system can not only identify current anomalies, but also assess the trend and severity of anomaly development, thereby significantly improving its sensitivity to early failures. By setting multiple threshold levels, the pin operating status is divided into stable operating status, early warning status, and high-risk status. Corresponding monitoring or maintenance instructions are output for different statuses, realizing closed-loop management from status identification to decision output, and improving the automation and intelligence level of equipment operation and maintenance.

[0010] To better implement the above methods, a capacitor pin status intelligent diagnostic system based on multi-source data fusion is proposed. The system includes a spatial distribution configuration module, a sensor data synchronization set module, a sensor parameter coupling feature module, a pin status identification model module, and a maintenance decision instruction module. Spatial distribution configuration module: Based on the historical operation records of the capacitor pins, it acquires the operation data during the operation process, analyzes the operation data, identifies the spatial distribution of abnormal areas, determines key monitoring areas, establishes the matching relationship between sensors and key monitoring areas, and configures the spatial distribution of sensors. Sensor data synchronization set module: acquires the spatial structure position of capacitor pins, establishes a spatial coordinate system for the pin area based on the spatial structure position, extracts the spatial coordinates corresponding to each sensor data based on the spatial distribution configuration, collects sensor data of the pin area in the capacitor operating state, performs time alignment on the sensor data, generates a time-synchronized sensor data sequence, associates the sensor data sequence with the spatial coordinates, and generates a sensor data synchronization set with spatiotemporal correspondence. Sensing parameter coupling feature module: Based on the sensing data synchronization set, extract the basic physical features of the pin structure state, and construct sensing parameter coupling features based on the basic physical features; Pin status recognition model module: Input the basic physical features and sensing parameter coupling features into the trained pin status recognition model to generate pin status recognition results; Maintenance decision instruction module: Based on the pin status identification results and basic physical characteristics, calculate the pin status evolution level index and output the corresponding maintenance decision instruction.

[0011] Furthermore, the sensor data synchronization set module includes a spatial coordinate unit and a sensor data synchronization set generation unit: Establish spatial coordinate unit: Obtain the three-dimensional spatial structure of capacitor pins, determine the spatial position information of each part of the pins, and establish a spatial coordinate system for the pin area based on the spatial position information. The spatial coordinate system takes the preset pin endpoint as the origin and the axial and radial directions of the pins as coordinate axes. Based on the spatial distribution configuration of the sensors, extract the spatial coordinates corresponding to the sensing data collected by each sensor. Generating a synchronized sensor data set unit: When the capacitor is in operation, each sensor is started to collect sensor data from the pin area. Using timestamp synchronization technology, the sensor data collected by different sensors is time-aligned, and the sensor data collected by each sensor is arranged in chronological order to form a time-synchronized sensor data sequence. By establishing a data mapping relationship, the sensor data sequence is paired with the corresponding spatial coordinates to generate a sensor data synchronization set with spatiotemporal correspondence.

[0012] Furthermore, the sensing parameter coupling feature module includes a cross-modal feature combination vector unit and a feature unit for determining sensing parameter coupling: Construct a cross-modal feature combination vector unit: extract features from the temperature data, vibration data and electrical signal data of each spatial node in the synchronous set of the sensing data under each time series, and construct a set of basic physical feature parameters. The basic physical features of the temperature data include the average temperature, the maximum temperature, the minimum temperature and the rate of temperature change. The basic physical characteristics of the vibration data include the average vibration amplitude, the peak vibration value, and the frequency band energy distribution. The basic physical characteristics of the electrical signal data include peak voltage, average voltage, and voltage fluctuation amplitude. The basic physical features of each spatial node under the same time series are normalized, and the normalized basic physical features are spliced ​​together according to temperature, vibration and electrical signal to construct a cross-modal feature combination vector. Determine the feature units of sensing parameter coupling: Based on the cross-modal feature combination vector, construct sensing parameter coupling according to the sensing parameter type. The sensing parameter coupling includes temperature-vibration coupling term, temperature-electrical signal coupling term and vibration-electrical signal coupling term. For each coupling term, based on the value of the corresponding feature sequence in the time dimension, the Pearson correlation coefficient between the two corresponding sensing parameters is calculated. The Pearson correlation coefficient is obtained by dividing the covariance of the coupling feature vectors of the two parameters by the product of their respective standard deviations. The Pearson correlation coefficient is used as the correlation index of the coupling of sensing parameters. Based on the aforementioned correlation index, the characteristics of the sensing parameter coupling are determined. The characteristic of the temperature-vibration coupling term is the Pearson correlation coefficient between temperature data and vibration data; the characteristic of the temperature-electrical signal coupling term is the Pearson correlation coefficient between temperature data and electrical signal data; and the characteristic of the vibration-electrical signal coupling term is the Pearson correlation coefficient between vibration data and electrical signal data.

[0013] Furthermore, the maintenance decision instruction module includes a unit for calculating pin state evolution level indicators and a unit for determining maintenance decision instructions: The unit for calculating the pin state evolution level index is as follows: During the operation of the capacitor, based on the pin state recognition model, it continuously receives the sensing data collected in real time by each sensor, and presets a time window. For the sensing data in each time window, it extracts the basic physical features and sensing parameter coupling features, and inputs them into the pin state recognition model to obtain the pin state recognition result in the corresponding time window. Based on the time window, it generates a pin state sequence in a continuous time series. Based on the pin state sequence, the state changes between the next time window and the previous time window are calculated, a state transition sequence is constructed, and the duration, transition frequency, and transition direction of each pin state within the two time windows are statistically analyzed. The transition frequency is calculated by dividing the number of transitions between different states by the duration of the two time windows. The transition direction includes a degradation direction and a recovery direction. The degradation direction is the change of the pin state from a normal state to an abnormal state or from a slight abnormality to a severe abnormality; the recovery direction is the change of the pin state from an abnormal state to a normal state or from a severe abnormality to a slight abnormality. Based on the state transition sequence and basic physical characteristics, the state deterioration rate and the anomaly accumulation intensity are calculated. The state deterioration rate is calculated by counting the number of deterioration directions within two time windows and dividing by the duration of the two time windows. The anomaly accumulation intensity is the cumulative value of the deviation between the basic physical characteristics under the abnormal state and the normal threshold within two time windows. The duration, transition frequency, state deterioration rate, and anomaly accumulation intensity are normalized and then weighted and summed according to preset weights to calculate the pin state evolution level index. The maintenance decision instruction unit determines the following: when the pin status evolution level index is lower than the first preset threshold, it is determined to be in a stable operating state, and a routine monitoring instruction is output; when the pin status evolution level index is between the first preset threshold and the second preset threshold, it is determined to be in an early warning state, and an enhanced monitoring and preventive maintenance instruction is output; when the pin status evolution level index is higher than the second preset threshold, it is determined to be in a high-risk state, and a maintenance decision instruction for repairing the pin is output.

[0014] Compared with the prior art, the beneficial effects of the present invention are: by fusing and analyzing multi-source information such as temperature data, vibration data and electrical signal data, the problem of one-sided information caused by traditional single sensor parameter monitoring is avoided; by analyzing historical operating data, the frequency of anomalies, the intensity of anomalies and the duration of anomalies are calculated, and a comprehensive anomaly evaluation index is formed; spatial nodes are sorted to determine key monitoring areas; and temperature sensors, vibration sensors and electrical signal acquisition units are adaptively configured according to the proportion of anomaly types. Establish a spatial coordinate system for the pin area and time-align data from different sensors using timestamp synchronization technology. This enables the sensor data and spatial coordinates to form a synchronized set of sensor data with a spatiotemporal correspondence. Through a unified spatiotemporal data structure, the operating status of the equipment can be analyzed simultaneously in both spatial and temporal dimensions, thereby improving the accuracy of fault location and status assessment. The deep learning model is trained based on historical labeled data, and the basic physical features and coupling features are used as input to the model to achieve automatic identification of normal state, slightly abnormal state and severely abnormal state. Compared with human experience judgment or simple threshold method, it has higher diagnostic accuracy. Attached Figure Description

[0015] Figure 1 This is a schematic diagram of the intelligent diagnostic system for capacitor pin status based on multi-source data fusion, as described in this invention. Detailed Implementation

[0016] 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.

[0017] Please see Figure 1 This invention provides a technical solution: an intelligent diagnostic method for capacitor pin status based on multi-source data fusion, the method comprising: Step S100: Based on the historical operation records of the capacitor pins, obtain the operation data during the operation process, analyze the operation data, identify the spatial distribution of abnormal areas, determine the key monitoring areas, establish the matching relationship between the sensors and the key monitoring areas, and configure the spatial distribution of the sensors. The spatial distribution configuration in step S100 includes: The operational data during operation includes temperature data, vibration data, and electrical signal data. The temperature data is arranged in a time series. Along the axial direction of the capacitor pins, the pins are divided into multiple spatial nodes. The temperature change gradient between adjacent spatial nodes is calculated as A = (a2 - a1) / b, where a1 and a2 represent the temperatures of the previous and next spatial nodes, respectively, and b is the axial distance between the two nodes. The temperature change sequence of each spatial node over time is summarized, and the instantaneous temperature rise rate of the spatial node is calculated as B = (a4 - a3) / c, where a3 and a4 represent the temperatures of the previous and current moments, respectively, and c represents the time interval. An axial temperature gradient threshold and a temperature rise rate threshold are set and compared with the temperature change gradient and instantaneous temperature rise rate of the spatial nodes. If the temperature change gradient of a spatial node is greater than the axial temperature gradient threshold or the instantaneous temperature rise rate of the spatial node is greater than the temperature rise rate threshold, the spatial node is marked as a temperature anomaly point. The temperature anomalies of all time nodes are summarized to obtain the temperature anomaly region. The collected vibration data is denoised and segmented according to the time series. Fourier transform is used to perform frequency domain transformation on the segmented vibration signals to calculate the amplitude characteristics and energy distribution characteristics of the vibration signals in each time period. The amplitude characteristic represents the maximum spectral amplitude of the vibration signal within the corresponding time period. The energy distribution characteristic represents the proportion of high-frequency energy to total energy, where high-frequency energy is the sum of squares of the amplitudes of all spectra within the high-frequency band, and total energy is the sum of squares of the amplitudes of all spectra within the entire frequency band. A threshold for amplitude characteristics and a threshold for energy distribution characteristics are set, and the amplitude characteristics and energy distribution characteristics of each time period are compared. When either the amplitude characteristic threshold or the energy distribution characteristic threshold is greater than the threshold, the corresponding time period is marked as a vibration anomaly segment. The vibration anomaly region is determined by combining the spatial mapping relationship of the vibration anomaly segments in each time period. The acquired electrical signal is filtered to extract the time-domain waveform feature parameters of the electrical signal. The time-domain waveform feature parameters include peak value, mean value, variance and fluctuation amplitude. A threshold range is set for each time-domain waveform feature parameter. When any time-domain waveform feature parameter exceeds the threshold range, the corresponding time period is marked as an abnormal area of ​​the electrical signal. Based on temperature anomaly regions, vibration anomaly regions, and electrical signal anomaly regions, historical operation records containing various anomaly regions are statistically analyzed. The frequency of anomaly occurrence, anomaly intensity index, and anomaly duration of each space node in the historical operation records are calculated. The frequency of anomaly occurrence is the number of times a space node is marked as an anomaly state in the historical operation records divided by the total number of times all space nodes are marked as an anomaly state. The anomaly intensity index is the average value of the difference between the anomaly feature parameter and the corresponding threshold. The duration of the anomaly is the length of time during which the anomaly occurs continuously. The frequency of anomalies, the intensity index of anomalies, and the duration of anomalies are normalized. The weights of the frequency of anomalies, the intensity index of anomalies, and the duration of anomalies are preset. The normalized frequency of anomalies, the intensity index of anomalies, and the duration of anomalies are weighted and summed to calculate the comprehensive anomaly evaluation index of spatial nodes. Based on the comprehensive anomaly evaluation index, all spatial nodes are sorted from high to low according to the comprehensive anomaly evaluation index, and the top few spatial regions in terms of comprehensive anomaly evaluation index are selected as key monitoring areas. The number of occurrences of each anomaly type within the key monitoring area is counted, and then divided by the total number of occurrences of all anomaly types within the key monitoring area to calculate the anomaly percentage corresponding to each anomaly type. Based on the proportion of each abnormality type in the key monitoring area, if the proportion of temperature abnormality type is higher than the preset threshold, a temperature sensor is configured; if the proportion of vibration abnormality type is higher than the preset threshold, a vibration sensor is configured; if the proportion of electrical signal abnormality is higher than the preset threshold, an electrical signal acquisition unit is configured; if the proportion of two or more abnormality types is higher than the preset threshold, multiple types of sensors are configured for collaborative monitoring to form a spatial distribution configuration scheme for multiple types of sensors. For example, a capacitor has a total lead length of 10, divided into 5 nodes: S1-S5, with an adjacent spacing of 2, a sampling interval of 1, and time points: t1-t5; The following is an example of raw temperature data: Time t1 corresponds to S1 = 30, S2 = 31, S3 = 32, S4 = 33, and S5 = 34; Time t2 corresponds to S1 as 31, S2 as 33, S3 as 35, S4 as 36, and S5 as 37; Time t3 corresponds to S1 as 32, S2 as 36, S3 as 40, S4 as 42, and S5 as 43; Time t4 corresponds to S1 as 33, S2 as 38, S3 as 45, S4 as 48, and S5 as 50; Time t5 corresponds to S1 as 34, S2 as 40, S3 as 50, S4 as 55, and S5 as 58; Calculate the axial temperature gradient, taking time t3 as an example: S2 is (36-32) / 2=2; S3 is (40-36) / 2=2; S4 is (42-40) / 2=1; Calculate the instantaneous temperature rise rate, taking S3 as an example: t3=(40-35) / 1=5; t4=(45-40) / 1=5; t5=(50-45) / 1=5; The temperature gradient threshold is set to 1.5, the temperature rise rate threshold is set to 4, and the temperature anomaly regions are S3, S4, and S5.

[0018] Step S200: Obtain the spatial structure position of the capacitor pins, and establish a spatial coordinate system for the pin area based on the spatial structure position. Based on the spatial distribution configuration, extract the spatial coordinates corresponding to each sensing data. Collect sensing data of the pin area while the capacitor is running. Time-align the sensing data to generate a time-synchronized sensing data sequence. Associate the sensing data sequence with the spatial coordinates to generate a sensing data synchronization set with spatiotemporal correspondence. The generation of the sensor data synchronization set in step S200 includes: The three-dimensional spatial structure of the capacitor pins is obtained, the spatial position information of each part of the pins is determined, and a spatial coordinate system of the pin area is established based on the spatial position information. The spatial coordinate system takes the preset pin endpoint as the origin and the axial and radial directions of the pins as coordinate axes. Based on the spatial distribution configuration of the sensors, the spatial coordinates corresponding to the sensing data collected by each sensor are extracted. When the capacitor is in operation, each sensor is activated to start collecting sensing data from the pin area. Using timestamp synchronization technology, the sensing data collected by different sensors are time-aligned, and the sensing data collected by each sensor are arranged in chronological order to form a time-synchronized sensing data sequence. By establishing a data mapping relationship, the sensor data sequence is paired with the corresponding spatial coordinates to generate a sensor data synchronization set with spatiotemporal correspondence.

[0019] Step S300: Based on the sensor data synchronization set, extract the basic physical features of the pin structure state, and construct the sensor parameter coupling features based on the basic physical features; The construction of sensing parameter coupling features in step S300 includes: Feature extraction is performed on the temperature data, vibration data and electrical signal data of each spatial node in the sensor data synchronization set under each time series to construct a basic physical feature parameter set. The basic physical features of the temperature data include the average temperature, the maximum temperature, the minimum temperature and the temperature change rate. The basic physical characteristics of the vibration data include the average vibration amplitude, the peak vibration value, and the frequency band energy distribution. The basic physical characteristics of the electrical signal data include peak voltage, average voltage, and voltage fluctuation amplitude. The basic physical features of each spatial node under the same time series are normalized, and the normalized basic physical features are spliced ​​together according to temperature, vibration and electrical signal to construct a cross-modal feature combination vector. Based on the cross-modal feature combination vector, sensor parameter coupling is constructed according to the sensor parameter type. The sensor parameter coupling includes temperature-vibration coupling term, temperature-electrical signal coupling term and vibration-electrical signal coupling term. For each coupling term, based on the value of the corresponding feature sequence in the time dimension, the Pearson correlation coefficient between the two corresponding sensing parameters is calculated. The Pearson correlation coefficient is obtained by dividing the covariance of the coupling feature vectors of the two parameters by the product of their respective standard deviations. The Pearson correlation coefficient is used as the correlation index of the coupling of sensing parameters. Based on the aforementioned correlation index, the characteristics of the sensing parameter coupling are determined. The characteristic of the temperature-vibration coupling term is the Pearson correlation coefficient between temperature data and vibration data; the characteristic of the temperature-electrical signal coupling term is the Pearson correlation coefficient between temperature data and electrical signal data; and the characteristic of the vibration-electrical signal coupling term is the Pearson correlation coefficient between vibration data and electrical signal data. For example, the raw sensor data of node S3 is shown below: The time t1 coordinate is (1, 0, 4), the temperature is 32, the vibration amplitude is 2.0, the high frequency energy ratio is 0.2, the voltage peak value is 5, the voltage average value is 3, and the fluctuation amplitude is 2. The coordinates of time t2 are (1, 0, 4), the temperature is 35°C, the vibration amplitude is 3.0, the high-frequency energy ratio is 0.35, the voltage peak value is 6, the voltage average value is 3.5, and the fluctuation amplitude is 2.5. The time t3 coordinate is (1, 0, 4), the temperature is 40, the vibration amplitude is 5.2, the high frequency energy ratio is 0.57, the voltage peak value is 10, the voltage average value is 6, and the fluctuation amplitude is 4. The time t4 coordinate is (1, 0, 4), the temperature is 45, the vibration amplitude is 6.0, the high frequency energy ratio is 0.69, the voltage peak value is 12, the voltage average value is 7, and the fluctuation amplitude is 5. The time t5 coordinate is (1, 0, 4), the temperature is 50, the vibration amplitude is 5.5, the high-frequency energy ratio is 0.60, the voltage peak value is 9, the voltage average value is 5, and the fluctuation amplitude is 3.5. Basic physical characteristics of temperature: The average temperature is (32+35+40+45+50) / 5=40.4; the maximum temperature is 50; the minimum temperature is 32; the rate of temperature change is (50-32) / 4=4.5. Vibration fundamental physical characteristics: The average vibration amplitude is (2.0+3.0+5.2+6.0+5.5) / 5=4.34; the peak vibration amplitude is 6.0; the average high-frequency energy distribution is (0.2+0.35+0.57+0.69+0.6); / 5=0.482 Basic physical characteristics of electrical signals: peak voltage is 12; average voltage is (3+3.5+6+7+5) / 5=4.9; average fluctuation amplitude is (2+2.5+4+5+3.5) / 5=3.4; Taking temperature-vibration coupling as an example, the temperature normalized sequence is [0.4667, 0.5667, 0.8889, 1, 1.0556], and the vibration normalized sequence is [0.333, 0.5, 0.867, 1, 0.917]; The calculated Pearson correlation coefficient is 0.98.

[0020] Step S400: Input the basic physical features and sensing parameter coupling features into the trained pin state recognition model to generate pin state recognition results; The training process of the pin status recognition model in step S400 includes: Based on the sensor data synchronization set, the pin status marked in the historical operation record is obtained. The pin status includes normal status, slightly abnormal status and severely abnormal status. The basic physical features and sensor parameter coupling features under the corresponding time series are matched with the pin status to construct a labeled training dataset. The training dataset is preprocessed, including interpolating and completing missing data, removing or correcting outliers, and normalizing the basic feature parameters. Using the aforementioned basic physical characteristics and sensing parameter coupling characteristics as model input, the pin state is assigned adjacent positive integer values. These pin state assignments serve as supervision labels. By training the deep learning model, a pin state recognition model is constructed. The training process involves initializing the network structure and parameters of each layer of the deep learning model, pre-setting several hidden layers in the deep learning model, and inputting the basic physical features and sensing parameter coupling features into the deep learning model. After layer-by-layer forward propagation, the output value of the last hidden layer is obtained. Based on the output value and pin state assignment, the corresponding classification error is calculated. The errors of each category are weighted and summed according to preset weights to obtain the total loss function. Based on the total loss function, the partial derivative of the total loss function with respect to the prediction result of the output layer is calculated starting from the output layer. The calculation is performed layer by layer from back to front according to the network connection relationship, and the error signal of each hidden layer is calculated in turn. Based on the error signal of each hidden layer, the parameter gradient of each hidden layer is calculated in combination with the corresponding layer input value, and the network weights and bias parameters are updated using the gradient descent method. Repeat the forward propagation, classification error calculation, loss function construction, backpropagation, and parameter update process until training ends when any of the following stopping conditions are met: the total loss function value is less than the preset loss threshold; the change in loss between two adjacent iterations is less than the preset convergence threshold; the number of training rounds reaches the preset maximum number of iterations. After training is completed, the pin status recognition model is obtained.

[0021] Step S500: Based on the pin status identification results and basic physical characteristics, calculate the pin status evolution level index and output the corresponding maintenance decision command; The maintenance decision instructions output in step S500 include: During capacitor operation, based on the pin state recognition model, the sensor data collected in real time by each sensor is continuously received, and a preset time window is set. For the sensor data in each time window, the basic physical features and sensor parameter coupling features are extracted and input into the pin state recognition model to obtain the pin state recognition result in the corresponding time window. Based on the time window, a pin state sequence under continuous time series is generated. Based on the pin state sequence, the state changes between the next time window and the previous time window are calculated, a state transition sequence is constructed, and the duration, transition frequency, and transition direction of each pin state within the two time windows are statistically analyzed. The transition frequency is calculated by dividing the number of transitions between different states by the duration of the two time windows. The transition direction includes a degradation direction and a recovery direction. The degradation direction is the change of the pin state from a normal state to an abnormal state or from a slight abnormality to a severe abnormality; the recovery direction is the change of the pin state from an abnormal state to a normal state or from a severe abnormality to a slight abnormality. Based on the state transition sequence and basic physical characteristics, the state deterioration rate and the anomaly accumulation intensity are calculated. The state deterioration rate is calculated by counting the number of deterioration directions within two time windows and dividing by the duration of the two time windows. The anomaly accumulation intensity is the cumulative value of the deviation between the basic physical characteristics under the abnormal state and the normal threshold within two time windows. The duration, transition frequency, state deterioration rate, and anomaly accumulation intensity are normalized and then weighted and summed according to preset weights to calculate the pin state evolution level index. When the pin status evolution level index is lower than the first preset threshold, it is determined to be in a stable operating state, and a routine monitoring command is output; when the pin status evolution level index is between the first preset threshold and the second preset threshold, it is determined to be in an early warning state, and an enhanced monitoring and preventive maintenance command is output; when the pin status evolution level index is higher than the second preset threshold, it is determined to be in a high-risk state, and a maintenance decision command for repairing the pin is output.

[0022] To better implement the above method, a smart diagnostic method for capacitor pin status based on multi-source data fusion is also proposed. The system includes a spatial distribution configuration module, a sensor data synchronization set module, a sensor parameter coupling feature module, a pin status identification model module, and a maintenance decision instruction module. Spatial distribution configuration module: Based on the historical operation records of the capacitor pins, it acquires the operation data during the operation process, analyzes the operation data, identifies the spatial distribution of abnormal areas, determines key monitoring areas, establishes the matching relationship between sensors and key monitoring areas, and configures the spatial distribution of sensors. Sensor data synchronization set module: acquires the spatial structure position of capacitor pins, establishes a spatial coordinate system for the pin area based on the spatial structure position, extracts the spatial coordinates corresponding to each sensor data based on the spatial distribution configuration, collects sensor data of the pin area in the capacitor operating state, performs time alignment on the sensor data, generates a time-synchronized sensor data sequence, associates the sensor data sequence with the spatial coordinates, and generates a sensor data synchronization set with spatiotemporal correspondence. The sensor data synchronization set module includes a spatial coordinate establishment unit and a sensor data synchronization set generation unit. Establish spatial coordinate unit: Obtain the three-dimensional spatial structure of capacitor pins, determine the spatial position information of each part of the pins, and establish a spatial coordinate system for the pin area based on the spatial position information. The spatial coordinate system takes the preset pin endpoint as the origin and the axial and radial directions of the pins as coordinate axes. Based on the spatial distribution configuration of the sensors, extract the spatial coordinates corresponding to the sensing data collected by each sensor. Generating a synchronized sensor data set unit: When the capacitor is in operation, each sensor is started to collect sensor data from the pin area. Using timestamp synchronization technology, the sensor data collected by different sensors is time-aligned, and the sensor data collected by each sensor is arranged in chronological order to form a time-synchronized sensor data sequence. By establishing a data mapping relationship, the sensor data sequence is paired with the corresponding spatial coordinates to generate a sensor data synchronization set with spatiotemporal correspondence.

[0023] Sensing parameter coupling feature module: Based on the sensing data synchronization set, extract the basic physical features of the pin structure state, and construct sensing parameter coupling features based on the basic physical features; The sensing parameter coupling feature module includes a cross-modal feature combination vector unit and a feature unit for determining sensing parameter coupling. Construct a cross-modal feature combination vector unit: extract features from the temperature data, vibration data and electrical signal data of each spatial node in the synchronous set of the sensing data under each time series, and construct a set of basic physical feature parameters. The basic physical features of the temperature data include the average temperature, the maximum temperature, the minimum temperature and the rate of temperature change. The basic physical characteristics of the vibration data include the average vibration amplitude, the peak vibration value, and the frequency band energy distribution. The basic physical characteristics of the electrical signal data include peak voltage, average voltage, and voltage fluctuation amplitude. The basic physical features of each spatial node under the same time series are normalized, and the normalized basic physical features are spliced ​​together according to temperature, vibration and electrical signal to construct a cross-modal feature combination vector. Determine the feature units of sensing parameter coupling: Based on the cross-modal feature combination vector, construct sensing parameter coupling according to the sensing parameter type. The sensing parameter coupling includes temperature-vibration coupling term, temperature-electrical signal coupling term and vibration-electrical signal coupling term. For each coupling term, based on the value of the corresponding feature sequence in the time dimension, the Pearson correlation coefficient between the two corresponding sensing parameters is calculated. The Pearson correlation coefficient is obtained by dividing the covariance of the coupling feature vectors of the two parameters by the product of their respective standard deviations. The Pearson correlation coefficient is used as the correlation index of the coupling of sensing parameters. Based on the aforementioned correlation index, the characteristics of the sensing parameter coupling are determined. The characteristic of the temperature-vibration coupling term is the Pearson correlation coefficient between temperature data and vibration data; the characteristic of the temperature-electrical signal coupling term is the Pearson correlation coefficient between temperature data and electrical signal data; and the characteristic of the vibration-electrical signal coupling term is the Pearson correlation coefficient between vibration data and electrical signal data.

[0024] Pin status recognition model module: Input the basic physical features and sensing parameter coupling features into the trained pin status recognition model to generate pin status recognition results; Maintenance decision instruction module: Based on the pin status identification results and basic physical characteristics, calculate the pin status evolution level index and output the corresponding maintenance decision instruction; The maintenance decision instruction module includes a unit for calculating pin state evolution level indicators and a unit for determining maintenance decision instructions. The unit for calculating the pin state evolution level index is as follows: During the operation of the capacitor, based on the pin state recognition model, it continuously receives the sensing data collected in real time by each sensor, and presets a time window. For the sensing data in each time window, it extracts the basic physical features and sensing parameter coupling features, and inputs them into the pin state recognition model to obtain the pin state recognition result in the corresponding time window. Based on the time window, it generates a pin state sequence in a continuous time series. Based on the pin state sequence, the state changes between the next time window and the previous time window are calculated, a state transition sequence is constructed, and the duration, transition frequency, and transition direction of each pin state within the two time windows are statistically analyzed. The transition frequency is calculated by dividing the number of transitions between different states by the duration of the two time windows. The transition direction includes a degradation direction and a recovery direction. The degradation direction is the change of the pin state from a normal state to an abnormal state or from a slight abnormality to a severe abnormality; the recovery direction is the change of the pin state from an abnormal state to a normal state or from a severe abnormality to a slight abnormality. Based on the state transition sequence and basic physical characteristics, the state deterioration rate and the anomaly accumulation intensity are calculated. The state deterioration rate is calculated by counting the number of deterioration directions within two time windows and dividing by the duration of the two time windows. The anomaly accumulation intensity is the cumulative value of the deviation between the basic physical characteristics under the abnormal state and the normal threshold within two time windows. The duration, transition frequency, state deterioration rate, and anomaly accumulation intensity are normalized and then weighted and summed according to preset weights to calculate the pin state evolution level index. The maintenance decision instruction unit determines the following: when the pin status evolution level index is lower than the first preset threshold, it is determined to be in a stable operating state, and a routine monitoring instruction is output; when the pin status evolution level index is between the first preset threshold and the second preset threshold, it is determined to be in an early warning state, and an enhanced monitoring and preventive maintenance instruction is output; when the pin status evolution level index is higher than the second preset threshold, it is determined to be in a high-risk state, and a maintenance decision instruction for repairing the pin is output.

[0025] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

Claims

1. A method for intelligent diagnosis of capacitor pin status based on multi-source data fusion, characterized in that, The methods include: Step S100: Based on the historical operation records of the capacitor pins, obtain the operation data during the operation process, analyze the operation data, identify the spatial distribution of abnormal areas, determine the key monitoring areas, establish the matching relationship between the sensors and the key monitoring areas, and configure the spatial distribution of the sensors. Step S200: Obtain the spatial structure position of the capacitor pins, and establish a spatial coordinate system for the pin area based on the spatial structure position. Based on the spatial distribution configuration, extract the spatial coordinates corresponding to each sensing data. Collect sensing data of the pin area while the capacitor is running. Time-align the sensing data to generate a time-synchronized sensing data sequence. Associate the sensing data sequence with the spatial coordinates to generate a sensing data synchronization set with spatiotemporal correspondence. Step S300: Based on the sensor data synchronization set, extract the basic physical features of the pin structure state, and construct the sensor parameter coupling features based on the basic physical features; The step S300 of constructing the sensing parameter coupling features includes the following steps: Feature extraction is performed on the temperature data, vibration data and electrical signal data of each spatial node in the sensor data synchronization set under each time series to construct a basic physical feature parameter set. The basic physical features of the temperature data include the average temperature, the maximum temperature, the minimum temperature and the temperature change rate. The basic physical characteristics of the vibration data include the average vibration amplitude, the peak vibration value, and the frequency band energy distribution. The basic physical characteristics of the electrical signal data include peak voltage, average voltage, and voltage fluctuation amplitude. The basic physical features of each spatial node under the same time series are normalized, and the normalized basic physical features are spliced ​​together according to temperature, vibration and electrical signal to construct a cross-modal feature combination vector. Based on the cross-modal feature combination vector, sensor parameter coupling is constructed according to the sensor parameter type. The sensor parameter coupling includes temperature-vibration coupling term, temperature-electrical signal coupling term and vibration-electrical signal coupling term. For each coupling term, based on the value of the corresponding feature sequence in the time dimension, the Pearson correlation coefficient between the two corresponding sensing parameters is calculated. The Pearson correlation coefficient is obtained by dividing the covariance of the coupling feature vectors of the two parameters by the product of their respective standard deviations. The Pearson correlation coefficient is used as the correlation index of the coupling of sensing parameters. Based on the aforementioned correlation index, the characteristics of the sensing parameter coupling are determined. The characteristic of the temperature-vibration coupling term is the Pearson correlation coefficient between temperature data and vibration data; the characteristic of the temperature-electrical signal coupling term is the Pearson correlation coefficient between temperature data and electrical signal data; and the characteristic of the vibration-electrical signal coupling term is the Pearson correlation coefficient between vibration data and electrical signal data. Step S400: Input the basic physical features and sensing parameter coupling features into the trained pin state recognition model to generate pin state recognition results; Step S500: Based on the pin status identification results and basic physical characteristics, calculate the pin status evolution level index and output the corresponding maintenance decision command.

2. The intelligent diagnostic method for capacitor pin status based on multi-source data fusion according to claim 1, characterized in that, The spatial distribution configuration in step S100 includes the following steps: The operational data during operation includes temperature data, vibration data, and electrical signal data. The temperature data is arranged in a time series. Along the axial direction of the capacitor pins, the pins are divided into multiple spatial nodes. The temperature change gradient between adjacent spatial nodes is calculated as A = (a2 - a1) / b, where a1 and a2 represent the temperatures of the previous and next spatial nodes, respectively, and b is the axial distance between the two nodes. The temperature change sequence of each spatial node over time is summarized, and the instantaneous temperature rise rate of the spatial node is calculated as B = (a4 - a3) / c, where a3 and a4 represent the temperatures of the previous and current moments, respectively, and c represents the time interval. An axial temperature gradient threshold and a temperature rise rate threshold are set and compared with the temperature change gradient and instantaneous temperature rise rate of the spatial nodes. If the temperature change gradient of a spatial node is greater than the axial temperature gradient threshold or the instantaneous temperature rise rate of the spatial node is greater than the temperature rise rate threshold, the spatial node is marked as a temperature anomaly point. The temperature anomalies of all time nodes are summarized to obtain the temperature anomaly region. The collected vibration data is denoised and segmented according to the time series. Fourier transform is used to perform frequency domain transformation on the segmented vibration signals to calculate the amplitude characteristics and energy distribution characteristics of the vibration signals in each time period. The amplitude characteristic represents the maximum spectral amplitude of the vibration signal within the corresponding time period. The energy distribution characteristic represents the proportion of high-frequency energy to total energy, where high-frequency energy is the sum of squares of the amplitudes of all spectra within the high-frequency band, and total energy is the sum of squares of the amplitudes of all spectra within the entire frequency band. A threshold for amplitude characteristics and a threshold for energy distribution characteristics are set, and the amplitude characteristics and energy distribution characteristics of each time period are compared. When either the amplitude characteristic threshold or the energy distribution characteristic threshold is greater than the threshold, the corresponding time period is marked as a vibration anomaly segment. The vibration anomaly region is determined by combining the spatial mapping relationship of the vibration anomaly segments in each time period. The acquired electrical signal is filtered to extract the time-domain waveform feature parameters of the electrical signal. The time-domain waveform feature parameters include peak value, mean value, variance and fluctuation amplitude. A threshold range is set for each time-domain waveform feature parameter. When any time-domain waveform feature parameter exceeds the threshold range, the corresponding time period is marked as an abnormal area of ​​the electrical signal. Based on temperature anomaly regions, vibration anomaly regions, and electrical signal anomaly regions, historical operation records containing various anomaly regions are statistically analyzed. The frequency of anomaly occurrence, anomaly intensity index, and anomaly duration of each space node in the historical operation records are calculated. The frequency of anomaly occurrence is the number of times a space node is marked as an anomaly state in the historical operation records divided by the total number of times all space nodes are marked as an anomaly state. The anomaly intensity index is the average value of the difference between the anomaly feature parameter and the corresponding threshold. The duration of the anomaly is the length of time during which the anomaly occurs continuously. The frequency of anomalies, the intensity index of anomalies, and the duration of anomalies are normalized. The weights of the frequency of anomalies, the intensity index of anomalies, and the duration of anomalies are preset. The normalized frequency of anomalies, the intensity index of anomalies, and the duration of anomalies are weighted and summed to calculate the comprehensive anomaly evaluation index of spatial nodes. Based on the comprehensive anomaly evaluation index, all spatial nodes are sorted from high to low according to the comprehensive anomaly evaluation index, and the top few spatial regions in terms of comprehensive anomaly evaluation index are selected as key monitoring areas. The number of occurrences of each anomaly type within the key monitoring area is counted, and then divided by the total number of occurrences of all anomaly types within the key monitoring area to calculate the anomaly percentage corresponding to each anomaly type. Based on the proportion of each anomaly type within the key monitoring area, if the proportion of temperature anomalies is higher than a preset threshold, a temperature sensor is configured; if the proportion of vibration anomalies is higher than a preset threshold, a vibration sensor is configured; if the proportion of electrical signal anomalies is higher than a preset threshold, an electrical signal acquisition unit is configured; if the proportion of two or more anomalies is higher than a preset threshold, multiple types of sensors are configured for collaborative monitoring, forming a spatial distribution configuration scheme for multiple types of sensors.

3. The intelligent diagnostic method for capacitor pin status based on multi-source data fusion according to claim 1, characterized in that, The step S200 of generating the sensor data synchronization set includes the following steps: The three-dimensional spatial structure of the capacitor pins is obtained, the spatial position information of each part of the pins is determined, and a spatial coordinate system of the pin area is established based on the spatial position information. The spatial coordinate system takes the preset pin endpoint as the origin and the axial and radial directions of the pins as coordinate axes. Based on the spatial distribution configuration of the sensors, the spatial coordinates corresponding to the sensing data collected by each sensor are extracted. When the capacitor is in operation, each sensor is activated to start collecting sensing data from the pin area. Using timestamp synchronization technology, the sensing data collected by different sensors are time-aligned, and the sensing data collected by each sensor are arranged in chronological order to form a time-synchronized sensing data sequence. By establishing a data mapping relationship, the sensor data sequence is paired with the corresponding spatial coordinates to generate a sensor data synchronization set with spatiotemporal correspondence.

4. The intelligent diagnostic method for capacitor pin status based on multi-source data fusion according to claim 1, characterized in that, The training process of the pin status recognition model in step S400 includes the following steps: Based on the sensor data synchronization set, the pin status marked in the historical operation record is obtained. The pin status includes normal status, slightly abnormal status and severely abnormal status. The basic physical features and sensor parameter coupling features under the corresponding time series are matched with the pin status to construct a labeled training dataset. The training dataset is preprocessed, including interpolating and completing missing data, removing or correcting outliers, and normalizing the basic feature parameters. Using the aforementioned basic physical characteristics and sensing parameter coupling characteristics as model input, the pin state is assigned adjacent positive integer values. These pin state assignments serve as supervision labels. By training the deep learning model, a pin state recognition model is constructed. The training process involves initializing the network structure and parameters of each layer of the deep learning model, pre-setting several hidden layers in the deep learning model, and inputting the basic physical features and sensing parameter coupling features into the deep learning model. After layer-by-layer forward propagation, the output value of the last hidden layer is obtained. Based on the output value and pin state assignment, the corresponding classification error is calculated. The errors of each category are weighted and summed according to preset weights to obtain the total loss function. Based on the total loss function, the partial derivative of the total loss function with respect to the prediction result of the output layer is calculated starting from the output layer. The calculation is performed layer by layer from back to front according to the network connection relationship, and the error signal of each hidden layer is calculated in turn. Based on the error signal of each hidden layer, the parameter gradient of each hidden layer is calculated in combination with the corresponding layer input value, and the network weights and bias parameters are updated using the gradient descent method. Repeat the forward propagation, classification error calculation, loss function construction, backpropagation, and parameter update process until training ends when any of the following stopping conditions are met: the total loss function value is less than the preset loss threshold; the change in loss between two adjacent iterations is less than the preset convergence threshold; the number of training rounds reaches the preset maximum number of iterations. After training is completed, the pin status recognition model is obtained.

5. The intelligent diagnostic method for capacitor pin status based on multi-source data fusion according to claim 1, characterized in that, The maintenance decision instruction output in step S500 includes the following steps: During capacitor operation, based on the pin state recognition model, the sensor data collected in real time by each sensor is continuously received, and a preset time window is set. For the sensor data in each time window, the basic physical features and sensor parameter coupling features are extracted and input into the pin state recognition model to obtain the pin state recognition result in the corresponding time window. Based on the time window, a pin state sequence under continuous time series is generated. Based on the pin state sequence, the state changes between the next time window and the previous time window are calculated, a state transition sequence is constructed, and the duration, transition frequency, and transition direction of each pin state within the two time windows are statistically analyzed. The transition frequency is calculated by dividing the number of transitions between different states by the duration of the two time windows. The transition direction includes a degradation direction and a recovery direction. The degradation direction is the change of the pin state from a normal state to an abnormal state or from a slight abnormality to a severe abnormality; the recovery direction is the change of the pin state from an abnormal state to a normal state or from a severe abnormality to a slight abnormality. Based on the state transition sequence and basic physical characteristics, the state deterioration rate and the anomaly accumulation intensity are calculated. The state deterioration rate is calculated by counting the number of deterioration directions within two time windows and dividing by the duration of the two time windows. The anomaly accumulation intensity is the cumulative value of the deviation between the basic physical characteristics under the abnormal state and the normal threshold within two time windows. The duration, transition frequency, state deterioration rate, and anomaly accumulation intensity are normalized and then weighted and summed according to preset weights to calculate the pin state evolution level index. When the pin status evolution level index is lower than the first preset threshold, it is determined to be in a stable operating state, and a routine monitoring command is output; when the pin status evolution level index is between the first preset threshold and the second preset threshold, it is determined to be in an early warning state, and an enhanced monitoring and preventive maintenance command is output; when the pin status evolution level index is higher than the second preset threshold, it is determined to be in a high-risk state, and a maintenance decision command for repairing the pin is output.

6. A capacitor pin status intelligent diagnostic system based on multi-source data fusion, used to implement the capacitor pin status intelligent diagnostic method based on multi-source data fusion as described in any one of claims 1-5, characterized in that, The system includes a spatial distribution configuration module, a sensor data synchronization set module, a sensor parameter coupling feature module, a pin status identification model module, and a maintenance decision instruction module. The spatial distribution configuration module: based on the historical operation records of the capacitor pins, acquires the operation data during the operation process, analyzes the operation data, identifies the spatial distribution of abnormal areas, determines key monitoring areas, establishes a matching relationship between sensors and key monitoring areas, and configures the spatial distribution of sensors. The sensor data synchronization set module: acquires the spatial structure position of the capacitor pins, establishes a spatial coordinate system for the pin area based on the spatial structure position, extracts the spatial coordinates corresponding to each sensor data based on the spatial distribution configuration, collects sensor data of the pin area in the capacitor operating state, performs time alignment on the sensor data, generates a time-synchronized sensor data sequence, associates the sensor data sequence with the spatial coordinates, and generates a sensor data synchronization set with a spatiotemporal correspondence. The sensing parameter coupling feature module: extracts the basic physical features of the pin structure state according to the sensing data synchronization set, and constructs sensing parameter coupling features based on the basic physical features; The pin status recognition model module: inputs the basic physical features and sensing parameter coupling features into the trained pin status recognition model to generate pin status recognition results; The maintenance decision instruction module calculates the pin state evolution level index based on the pin state identification results and basic physical characteristics, and outputs the corresponding maintenance decision instruction.

7. The intelligent diagnostic system for capacitor pin status based on multi-source data fusion according to claim 6, characterized in that, The sensor data synchronization set module includes a spatial coordinate unit and a sensor data synchronization set generation unit: The establishment of the spatial coordinate unit involves: acquiring the three-dimensional spatial structure of the capacitor pins, determining the spatial position information of each part of the pins, and establishing a spatial coordinate system for the pin area based on the spatial position information. The spatial coordinate system takes the preset pin endpoint as the origin and the axial and radial directions of the pins as coordinate axes. Based on the spatial distribution configuration of the sensors, the spatial coordinates corresponding to the sensing data collected by each sensor are extracted. The sensor data synchronization set generation unit: when the capacitor is in operation, it starts each sensor to start collecting sensor data in the pin area, uses timestamp synchronization technology to time-align the sensor data collected by different sensors, and arranges the sensor data collected by each sensor in chronological order to form a time-synchronized sensor data sequence. By establishing a data mapping relationship, the sensor data sequence is paired with the corresponding spatial coordinates to generate a sensor data synchronization set with spatiotemporal correspondence.

8. The intelligent diagnostic system for capacitor pin status based on multi-source data fusion according to claim 6, characterized in that, The sensing parameter coupling feature module includes a cross-modal feature combination vector unit and a feature unit for determining sensing parameter coupling: The construction of the cross-modal feature combination vector unit involves extracting features from the temperature data, vibration data, and electrical signal data of each spatial node in the synchronous set of sensor data under each time series, and constructing a set of basic physical feature parameters. The basic physical features of the temperature data include the average temperature, the maximum temperature, the minimum temperature, and the rate of temperature change. The basic physical characteristics of the vibration data include the average vibration amplitude, the peak vibration value, and the frequency band energy distribution. The basic physical characteristics of the electrical signal data include peak voltage, average voltage, and voltage fluctuation amplitude. The basic physical features of each spatial node under the same time series are normalized, and the normalized basic physical features are spliced ​​together according to temperature, vibration and electrical signal to construct a cross-modal feature combination vector. The feature unit for determining the coupling of sensing parameters is: based on the cross-modal feature combination vector, sensing parameter coupling is constructed according to the sensing parameter type, and the sensing parameter coupling includes temperature-vibration coupling term, temperature-electrical signal coupling term and vibration-electrical signal coupling term; For each coupling term, based on the value of the corresponding feature sequence in the time dimension, the Pearson correlation coefficient between the two corresponding sensing parameters is calculated. The Pearson correlation coefficient is obtained by dividing the covariance of the coupling feature vectors of the two parameters by the product of their respective standard deviations. The Pearson correlation coefficient is used as the correlation index of the coupling of sensing parameters. Based on the aforementioned correlation index, the characteristics of the sensing parameter coupling are determined. The characteristic of the temperature-vibration coupling term is the Pearson correlation coefficient between temperature data and vibration data; the characteristic of the temperature-electrical signal coupling term is the Pearson correlation coefficient between temperature data and electrical signal data; and the characteristic of the vibration-electrical signal coupling term is the Pearson correlation coefficient between vibration data and electrical signal data.

9. The intelligent diagnostic system for capacitor pin status based on multi-source data fusion according to claim 6, characterized in that, The maintenance decision instruction module includes a unit for calculating pin state evolution level indicators and a unit for determining maintenance decision instructions. The calculation pin state evolution level index unit: During the operation of the capacitor, based on the pin state recognition model, it continuously receives the sensing data collected in real time by each sensor, and presets a time window. For the sensing data in each time window, it extracts the basic physical features and sensing parameter coupling features, and inputs them into the pin state recognition model to obtain the pin state recognition result in the corresponding time window. Based on the time window, it generates a pin state sequence under continuous time series. Based on the pin state sequence, the state change between the next time window and the previous time window is calculated, a state transition sequence is constructed, and the duration, transition frequency and transition direction of each pin state within the two time windows are statistically analyzed. The transition frequency is the number of transitions between different states divided by the duration of the two time windows. The transition direction includes the degradation direction and the recovery direction. The degradation direction is the change of the pin state from a normal state to an abnormal state or from a slight abnormality to a severe abnormality. The recovery direction is the change of the pin state from an abnormal state to a normal state or from a severe abnormality to a slight abnormality. Based on the state transition sequence and basic physical characteristics, the state deterioration rate and the anomaly accumulation intensity are calculated. The state deterioration rate is calculated by counting the number of deterioration directions within two time windows and dividing by the duration of the two time windows. The anomaly accumulation intensity is the cumulative value of the deviation between the basic physical characteristics under the abnormal state and the normal threshold within two time windows. The duration, transition frequency, state deterioration rate, and anomaly accumulation intensity are normalized and then weighted and summed according to preset weights to calculate the pin state evolution level index. The maintenance decision instruction unit determines the following: when the pin status evolution level index is lower than the first preset threshold, it is determined to be in a stable operating state and outputs a routine monitoring instruction; when the pin status evolution level index is between the first preset threshold and the second preset threshold, it is determined to be in an early warning state and outputs an enhanced monitoring and preventive maintenance instruction; when the pin status evolution level index is higher than the second preset threshold, it is determined to be in a high-risk state and outputs a maintenance decision instruction for repairing the pin.

Images