Transformer comprehensive monitoring and early warning method and system

Abstract

The invention discloses a comprehensive monitoring and early warning method and system for a transformer, and relates to the field of transformer monitoring, and the method comprises the steps: collecting the state data of a target transformer in real time, the state data comprising oil chromatography data, a partial discharge signal, a grounding current, local temperature data and a local vibration signal; the current state data is preprocessed; performing multi-angle evaluation on the state of the target transformer according to the preprocessed current state data; if the state of the target transformer obtained through evaluation from at least one angle is abnormal, performing early warning on each abnormal state of the target transformer; accurate monitoring and early warning of the state of the target transformer are achieved, and the problems that an existing transformer monitoring method is low in monitoring precision and insufficient in fault early warning capacity are solved.

Description

Technical Field

[0001] The present application relates to the field of transformer monitoring, and in particular to a transformer comprehensive monitoring and early warning method and system. Background Art

[0002] Monitoring the health status of power transformers is one of the key links to ensure the stable operation of the power system. By monitoring the operating status of the transformer, potential faults can be discovered in a timely manner, and preventive measures can be taken to avoid power outages and other economic losses caused by equipment failures. In the prior art, when a transformer has suspected faults and defects but the specific nature of the fault has not yet been determined, certain monitoring equipment and measures will be used to allow the transformer to continue to operate for a period of time in order to collect data to further evaluate the impact of the fault on the transformer performance and whether it poses a threat to the safety of the power grid.

[0003] However, the traditional transformer monitoring method uses a single or a few sensors to collect the single state data of the transformer, and compares it with the preset threshold, and then issues an early warning based on the comparison result, or uses a few state data for weighted summation, and compares the weighted summation result with the preset threshold, and then issues an early warning based on the comparison result to achieve transformer state monitoring. This transformer monitoring method has low monitoring accuracy and insufficient fault warning capability. Summary of the invention

[0004] The purpose of this application is to provide a transformer comprehensive monitoring and early warning method and system to solve the problems of low monitoring accuracy and insufficient fault early warning capability in existing transformer monitoring methods.

[0005] To achieve the above objectives, this application provides the following solutions:

[0006] In a first aspect, the present application provides a transformer comprehensive monitoring and early warning method, comprising:

[0007] Collecting status data of the target transformer in real time, the status data including oil chromatography data, local discharge signal, grounding current, local temperature data and local vibration signal;

[0008] Preprocessing the current state data;

[0009] According to the preprocessed current state data, the state of the target transformer is evaluated from multiple angles;

[0010] If at least one angle evaluation shows that the state of the target transformer is abnormal, an early warning is issued for each abnormal state of the target transformer.

[0011] Optionally, before performing multi-angle evaluation on the state of the target transformer according to the preprocessed current state data, the method further includes:

[0012] Construct a first machine learning model, and use the particle swarm algorithm and InputData1 and T1 to optimize the parameters of the first machine learning model, where InputData1 is a plurality of local temperature data of the sample transformer, T1 is the overall temperature of the sample transformer, InputData1 is used as the input of the first machine learning model, and T1 is used as the output of the first machine learning model. The optimal parameter combination is found through iterative search:

[0013] θ opt =argmin θ Error(f ML (InputData1,θ j ));

[0014] Among them, f ML represents the first machine learning model, θ j represents the first influencing factor, wherein j={1,2,…,n1}, n1 represents the number of the first influencing factors, the first influencing factor refers to the factor affecting T1, Error represents the prediction error function of the first machine learning model, argmin θ Error represents the parameter combination of the first machine learning model when the prediction error Error of the first machine learning model is the smallest, θ opt Represents the optimal parameter combination found by the particle swarm algorithm;

[0015] The multi-angle evaluation of the state of the target transformer according to the pre-processed current state data includes:

[0016] According to multiple second influencing factors θ i and the current InputData2, using the parameter combination θ opt The first machine learning model predicts the current overall temperature T2 of the target transformer; wherein i={1,2,…,n2}, n2 represents the number of the second influencing factors, the second influencing factors refer to factors affecting T2, and InputData2 represents multiple local temperature data of the target transformer;

[0017] Calculate the temperature rise rate Δv of the target transformer within Δt time T :

[0018]

[0019] Wherein, T'2 represents the overall temperature of the target transformer predicted last time, Δt=t2-t1, t2 represents the acquisition time point of the current InputData2, and t1 represents the acquisition time point of the previous InputData2;

[0020] If Δv T If the temperature rise rate is greater than a preset threshold value, it indicates that the temperature rise of the target transformer is abnormal.

[0021] Optionally, the step of evaluating the state of the target transformer from multiple angles according to the preprocessed current state data further includes:

[0022] In the method according to the plurality of second influencing factors θ i and the current InputData2, using the parameter combination θ opt After the first machine learning model predicts the current overall temperature T2 of the target transformer, according to a plurality of second influencing factors θ i and the current InputData2, based on the Bayesian model, determine P(T2>T threshold ,θ i |InputData2):

[0023]

[0024] Among them, P(InputData2|T2>T threshold ,θ i ) indicates that when the threshold T2 and multiple second influencing factors θ i The likelihood of observing InputData2 under this condition, P(T2>T threshold ,θ i ) means greater than T threshold T2 and multiple second influencing factors θ i The prior probability of InputData2 is P(InputData2), which represents the marginal likelihood of InputData2, and P(T2>T threshold ,θ i |InputData2) indicates that it is greater than T threshold T2 and multiple second influencing factors θ i The posterior probability of threshold is the set temperature threshold;

[0025] If P(T2>T threshold ,θ i |InputData2) is greater than the set probability threshold, indicating that the temperature of the target transformer is abnormal.

[0026] Optionally, the preprocessing of the current state data specifically includes:

[0027] Perform time domain analysis, frequency domain analysis and time-frequency analysis on a single discharge pulse of the current partial discharge signal to extract the time domain features, frequency domain features and time-frequency features of the partial discharge signal;

[0028] The state of the target transformer is evaluated from multiple angles according to the preprocessed current state data, including:

[0029] Inputting the extracted time domain features, frequency domain features and time-frequency features of the partial discharge signal into a constructed insulation fault assessment model, the insulation fault assessment model outputting a corresponding insulation threat level;

[0030] If the insulation threat level output by the insulation fault assessment model is greater than a preset insulation threat level threshold, it indicates that the insulation performance of the target transformer is abnormal.

[0031] Optionally, the state of the target transformer is evaluated from multiple angles according to the preprocessed current state data, including:

[0032] If, among the current plurality of local vibration signals, at least some of the local vibration signals have frequencies that are integer multiples of 50 Hz and amplitudes that are greater than a preset first amplitude threshold, it indicates that the winding of the target transformer is loose;

[0033] If among the current plurality of local vibration signals, at least some of the local vibration signals have a frequency greater than 20 Hz and less than 50 Hz and an amplitude greater than a preset second amplitude threshold, it indicates that the core of the target transformer is loose;

[0034] According to the grounding current, a first fault identification model based on an attention convolutional neural network or reinforcement learning is used to identify the grounding fault and insulation damage fault of the target transformer.

[0035] Optionally, the state of the target transformer is evaluated from multiple angles according to the preprocessed current state data, including:

[0036] According to the current oil chromatogram data, a second machine learning model is used to identify the overheating fault or discharge fault and the fault location of the target transformer, wherein the second machine learning model includes a GA-SVM model or a CEABC-WNN model; wherein the GA-SVM model refers to the final SVM obtained by training the parameter combination of the support vector machine SVM optimized by the genetic algorithm GA; the CEABC-WNN model refers to the final WNN model obtained by training the initial parameter combination of the wavelet neural network WNN model optimized by the chaos enhanced artificial bee colony algorithm CEABC.

[0037] Optionally, the state of the target transformer is evaluated from multiple angles according to the preprocessed current state data, including:

[0038] Calculate the correlation coefficient between the current state data after preprocessing;

[0039] According to the correlation coefficients between the state data, a correlation coefficient matrix is ​​constructed;

[0040] Based on the correlation coefficient matrix, an undirected graph network is constructed, wherein the names of the nodes of the undirected graph network are the names of the state data, the attribute values ​​of the nodes are the corresponding state data values, and the edges are the connecting lines between the significantly correlated state data pairs; wherein, if the correlation coefficient of two state data is greater than a preset correlation coefficient threshold, they are defined as a group of the significantly correlated state data pairs;

[0041] identifying independent communities in the undirected graph network by a community detection algorithm;

[0042] According to the identified independent community, an abnormal state of the target transformer is identified.

[0043] Optionally, the state of the target transformer is evaluated from multiple angles according to the preprocessed current state data, including:

[0044] Clustering the pre-processed current state data using a hierarchical clustering model based on a bottom-up aggregation strategy;

[0045] Comprehensive fault diagnosis is performed based on the clustering result to identify the fault of the target transformer, wherein the fault includes at least one of a partial discharge fault, a mechanical fault, a grounding fault, an insulation fault and an overheating fault.

[0046] Optionally, the transformer comprehensive monitoring and early warning method further includes:

[0047] Constructing a digital twin model of the target transformer;

[0048] Synchronously simulating the operation process of the target transformer through the digital twin model, and collecting state data of the digital twin model in real time;

[0049] Inputting an input sequence into a temporal difference learning model, and predicting future state data of the target transformer through the temporal difference learning model; wherein the input sequence includes: multiple pairs of input data, each pair of input data includes state data of the digital twin model and the target transformer at the same time;

[0050] The future state data of the target transformer is used as the current state data of the target transformer, and preprocessing and subsequent steps are performed on the current state data to issue an early warning for each future abnormal state of the target transformer.

[0051] In a second aspect, the present application provides a transformer comprehensive monitoring and early warning system, comprising:

[0052] An acquisition module, used for real-time acquisition of current status data of the target transformer, wherein the status data includes oil chromatography data, partial discharge signal, ground current, local temperature data and local vibration signal;

[0053] A data processing module, used for preprocessing the current state data;

[0054] A multi-angle evaluation module, used to evaluate the state of the target transformer from multiple angles according to the pre-processed current state data;

[0055] The early warning module is used to issue an early warning for each current abnormal state of the target transformer if at least one angle evaluation shows that the state of the target transformer is abnormal.

[0056] According to the specific embodiments provided in this application, this application discloses the following technical effects:

[0057] The present application provides a method and system for comprehensive monitoring and early warning of transformers. By real-time collection of the current status data of the target transformer, the status data includes oil chromatography data, local discharge signal, grounding current, local temperature data and local vibration signal. Compared with the prior art which only uses a single indicator or a few indicators to monitor the status of the transformer, the indicators used cover a wider range, avoiding the blind spots of using a single or a few indicators to monitor the current status of the transformer, and being able to more accurately monitor the current status of the transformer and provide abnormal early warnings; by preprocessing the current status data, the quality of the status data is improved, and the accuracy and reliability of multi-angle evaluation of the status of the target transformer are improved; by performing multi-angle evaluation of the status of the target transformer based on the preprocessed current status data, compared with the prior art, the multi-angle The evaluation can improve the accuracy of the current state evaluation of the transformer; when the state of the target transformer is abnormal as evaluated from at least one angle, an early warning is issued for each abnormal state of the target transformer. Since the accuracy of the state evaluation of the target transformer is improved, the early warning capability of the abnormal state (fault) of the target transformer is further improved; in summary, the present application uses the current oil chromatography data, local discharge signal, ground current, local temperature data and local vibration signal of the target transformer after preprocessing to perform a multi-angle evaluation of the state of the target transformer, and when the state of the target transformer is abnormal as evaluated from at least one angle, an early warning is issued for each abnormal state of the target transformer, thereby achieving accurate monitoring and early warning of the state of the target transformer, and solving the problems of low monitoring accuracy and insufficient fault early warning capability in the existing transformer monitoring method. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0059] Figure 1 A flowchart of a transformer comprehensive monitoring and early warning method provided in one embodiment of the present application;

[0060] Figure 2 A schematic diagram of functional modules of a transformer comprehensive monitoring and early warning system provided in one embodiment of the present application;

[0061] Figure 3 A schematic diagram of the structure of a computer device provided in one embodiment of the present application. DETAILED DESCRIPTION

[0062] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0063] In order to make the above-mentioned objects, features and advantages of the present application more obvious and easy to understand, the present application is further described in detail below with reference to the accompanying drawings and specific implementation methods.

[0064] In an exemplary embodiment, Figure 1 As shown, a transformer comprehensive monitoring and early warning method is provided, including the following steps 101 to 104. Among them:

[0065] Step 101 , collecting status data of a target transformer in real time, the status data including oil chromatography data, local discharge signal, ground current, local temperature data and local vibration signal.

[0066] In the embodiment of the present application, there is no specific limitation on the sensor used to collect the partial discharge signal, and it can be selected according to actual needs. For example, an oil valve type UHF high-frequency sensor is used to collect the partial discharge signal of the target transformer. The temperature includes the oil temperature and the shell temperature. There is no specific limitation on the sensor used to collect the oil temperature and the shell temperature, and it can be selected according to actual needs. For example, a fiber grating temperature sensor set in the insulating oil tank of the target transformer is used to collect the oil temperature of the target transformer. For example, a PT100 temperature sensor set outside the insulating oil tank of the target transformer is used to collect the shell temperature of the target transformer. There is no specific limitation on the sensor used to collect the grounding current of the iron core of the target transformer, and it can be selected according to actual needs. For example, a high-frequency current transformer is used to collect the grounding current of the iron core of the target transformer, and the high-frequency current transformer can use an open-and-closed iron core sensor. A vibration sensor is used to collect the local vibration signal of the target transformer. The oil chromatographic data refers to the gas components and concentration data of each gas in the insulating oil of the target transformer. The insulating oil of the target transformer is introduced into the degasser by an inlet and outlet method through the oil extraction valve on the target transformer body. The circulation of the insulating oil can be guaranteed when the target transformer works normally. The degasser adopts the headspace degassing method to separate the gas in the insulating oil when the gas and liquid of the insulating oil reach equilibrium. The separated mixed gas is separated into each single component gas through a chromatographic column, and then the concentration data of each gas is obtained through a gas sensor. In the embodiment of the present application, the components such as H2, CH4, CO, C2H6, CO2, C2H4, C2H2, hydrocarbon gas in the transformer insulating oil are detected in real time online.

[0067] Step 102: pre-process the current status data.

[0068] Step 103: Based on the preprocessed current status data, the status of the target transformer is evaluated from multiple angles.

[0069] Step 104: If at least one angle evaluation shows that the state of the target transformer is abnormal, an early warning is issued for each abnormal state of the target transformer.

[0070] Implement the above-mentioned steps 101 to 104, and collect the current status data of the target transformer in real time. The status data includes oil chromatography data, local discharge signal, ground current, local temperature data and local vibration signal. Compared with the prior art that only uses a single indicator or a few indicators to monitor the status of the transformer, the indicators used cover a wider range, avoiding the blind spots of using a single or a few indicators to monitor the current status of the transformer, and can more accurately monitor the current status of the transformer and warn of abnormalities; by preprocessing the current status data, the quality of the status data is improved, and the accuracy and reliability of the multi-angle evaluation of the status of the target transformer are improved; by performing a multi-angle evaluation of the status of the target transformer based on the preprocessed current status data, compared with the prior art, the multi-angle evaluation can Improve the accuracy of the current state assessment of the transformer; when the state of the target transformer is abnormal as assessed from at least one angle, warn of each abnormal state of the target transformer. Since the accuracy of the state assessment of the target transformer is improved, the warning capability of the abnormal state (fault) of the target transformer is further improved; in summary, the present application uses the current oil chromatogram data, local discharge signal, ground current, local temperature data and local vibration signal of the target transformer after preprocessing to perform multi-angle assessment of the state of the target transformer, and when the state of the target transformer is abnormal as assessed from at least one angle, warn of each abnormal state of the target transformer, thereby achieving accurate monitoring and warning of the state of the target transformer, and solving the problems of low monitoring accuracy and insufficient fault warning capability in the existing transformer monitoring method.

[0071] It should be noted that in the embodiment of the present application, the state of the target transformer is evaluated from multiple angles according to the current state data after preprocessing, including evaluating the state of the target transformer according to the current oil chromatogram data, partial discharge signal, ground current, local temperature data and local vibration signal after preprocessing, thereby obtaining the multi-angle evaluation result of the state of the target transformer. The following is a detailed description of the evaluation of the state of the target transformer according to the current oil chromatogram data, partial discharge signal, ground current, local temperature data and local vibration signal after preprocessing.

[0072] Before evaluating the state of the target transformer based on the pre-processed current local temperature data, it also includes:

[0073] Construct a first machine learning model, and use the particle swarm algorithm and InputData1 and T1 to optimize the parameters of the first machine learning model, where InputData1 is multiple local temperature data of the sample transformer, T1 is the overall temperature of the sample transformer, InputData1 is used as the input of the constructed first machine learning model, and T1 is used as the output of the constructed first machine learning model. The optimal parameter combination is found through iterative search:

[0074] θ opt =argmin θ Error(f ML (InputData1,θ j ));

[0075] Among them, f ML represents the first machine learning model, θ j represents the first influencing factor, where j = {1, 2, ..., n1}, n1 represents the number of the first influencing factors, the first influencing factor refers to the factor affecting T1, Error represents the prediction error function of the first machine learning model, argmin θ Error represents the parameter combination of the first machine learning model when the prediction error Error of the first machine learning model is the smallest, θ opt Represents the optimal parameter combination found by the particle swarm algorithm.

[0076] In the embodiment of the present application, the first machine learning model preferably uses a neural network model, and uses a PSO (particle swarm) algorithm to optimize the parameters of the first machine learning model, and finds the optimal parameter combination through iterative search, thereby effectively improving the prediction accuracy of the first machine learning model. The overall temperature of the sample transformer can be estimated by an empirical formula based on past experience historical operation data.

[0077] According to the pre-processed current local temperature data, the state of the target transformer is evaluated, including the following steps 201 to 203.

[0078] Step 201: based on a plurality of second influencing factors θ i And the current InputData2, using the parameter combination of θ opt The first machine learning model predicts the current overall temperature T2 of the target transformer; wherein, i={1,2,…,n2}, n2 represents the number of second influencing factors, the second influencing factors refer to the factors affecting T2, and InputData2 represents multiple local temperature data of the target transformer.

[0079] In the embodiment of the present application, the first influencing factor and the second influencing factor refer to factors that affect the temperature of the transformer, such as load, ambient temperature, etc., which mainly reflect external conditions and control parameters. The local temperature includes oil temperature (insulating oil temperature) and casing temperature.

[0080] Step 202, calculate the temperature rise rate Δv of the target transformer within the time Δt T :

[0081]

[0082] Wherein, T'2 represents the overall temperature of the target transformer predicted last time, Δt=t2-t1, t2 represents the acquisition time point of the current InputData2, and t1 represents the acquisition time point of the previous InputData2.

[0083] Step 203, if Δv T If the temperature rise rate is greater than the preset threshold, it means that the temperature rise of the target transformer is abnormal.

[0084] In the embodiment of the present application, the target transformer temperature (such as oil temperature, casing temperature) is an important indicator reflecting the operating status and health status of the target transformer, and is also an early reflection of problems such as insulation aging and heat dissipation system failure of the target transformer. Considering the influence of various influencing factors on the target transformer temperature, the temperature change of the target transformer can be predicted more accurately. Through the above steps 201 to 203, monitoring and early warning of abnormal temperature rise are achieved, thereby improving the accuracy of abnormal diagnosis and early warning of the target transformer.

[0085] According to the pre-processed current local temperature data, the state of the target transformer is evaluated, and after the above step 201, the following steps 301 to 302 are also included. Among them:

[0086] Step 301, based on multiple influencing factors θ i and the current InputData2, based on the Bayesian model, determine P(T2>T threshold ,θ i |InputData2):

[0087]

[0088] Among them, P(InputData2|T2>T threshold ,θ i ) indicates that when the threshold T2 and multiple second influencing factors θ i The likelihood of observing InputData under this condition, P(T2>T threshold ,θ i ) means greater than Tthreshold T2 and multiple second influencing factors θ i The prior probability of InputData2 is P(InputData2), which represents the marginal likelihood of InputData2, and P(T2>T threshold ,θ i |InputData2) indicates that it is greater than T threshold T2 and multiple second influencing factors θ i The posterior probability of threshold is the set temperature threshold.

[0089] Step 302: If P(T2>T threshold ,θ i |InputData2) is greater than the set probability threshold, indicating that the temperature of the target transformer is abnormal.

[0090] In the embodiment of the present application, through steps 301 to 302, a dynamic risk assessment of the overall temperature T2 of the target transformer is implemented.

[0091] Preprocess the current status data, including:

[0092] The single discharge pulse of the current partial discharge signal is subjected to time domain analysis, frequency domain analysis and time-frequency analysis to extract the time domain features, frequency domain features and time-frequency features of the partial discharge signal.

[0093] In the embodiment of the present application, the time domain characteristics of the partial discharge signal include but are not limited to the amplitude, frequency, phase and energy density of the partial discharge signal, and the frequency domain characteristics include but are not limited to the center frequency, harmonic components, power spectrum density, spectrum shape, bandwidth, etc. The time-frequency characteristics include but are not limited to the time-frequency distribution diagram, instantaneous frequency and instantaneous amplitude, marginal spectrum, average power spectrum density, time-frequency entropy, etc.

[0094] According to the pre-processed current partial discharge signal, the state of the target transformer is evaluated, including the following steps 401 to 402. Among them:

[0095] Step 401: input the extracted time domain features, frequency domain features and time-frequency features of the partial discharge signal into the constructed insulation fault assessment model, and the insulation fault assessment model outputs the corresponding insulation threat level.

[0096] In an embodiment of the present application, the time domain characteristics, frequency domain characteristics and time-frequency characteristics of the partial discharge signal sample can be directly associated with the insulation threat level, and the critical values ​​(thresholds) of different insulation threat levels can be set in combination with standards (such as IEEE C57.104 or IEC 60270), and each partial discharge signal sample can be divided into different insulation threat levels. Then, the insulation fault assessment model is trained using the partial discharge signal samples with insulation threat levels.

[0097] Step 402: If the insulation threat level output by the insulation fault assessment model is greater than a preset insulation threat level threshold, it indicates that the insulation performance of the target transformer is abnormal.

[0098] In the embodiment of the present application, through the above steps 401 to 402, potential threats to the insulation performance of the target transformer can be evaluated and warned, thereby improving the accuracy of abnormal diagnosis and warning of the target transformer.

[0099] According to the pre-processed current local vibration signal, the state of the target transformer is evaluated, including the following steps 501 to 502. Among them:

[0100] Step 501: If, among the current multiple local vibration signals, the frequencies of at least some local vibration signals are integer multiples of 50 Hz and the amplitudes are greater than a preset first amplitude threshold, it indicates that the winding of the target transformer is loose.

[0101] In the embodiment of the present application, the above-mentioned integer multiple refers to a multiple of 50 Hz and the multiple is an integer, for example, 1, 2 or 3 times, etc., corresponding to the frequency of at least part of the local vibration signal being 50 Hz, 100 Hz or 150 Hz, etc.

[0102] Step 502: If, among the current multiple local vibration signals, at least some of the local vibration signals have a frequency greater than 20 Hz and less than 50 Hz and an amplitude greater than a preset second amplitude threshold, it indicates that the core of the target transformer is loose.

[0103] In the embodiment of the present application, a plurality of vibration sensors are arranged at the top of the winding, near the core and on the surface of the box of the target transformer, and vibration signals at different positions, i.e., local vibration signals, are collected. Winding loose faults are usually more significant at the vibration sensors close to the windings, while core loose faults are more significant at the vibration sensors close to the cores. Vibration spectrum analysis shows that the vibration signal of winding looseness shows abnormally high amplitude near 50Hz or its integer multiple frequency, while core looseness mainly shows enhanced vibration peaks in lower frequency bands (usually 20Hz to 50Hz). Therefore, through experiments, the first amplitude threshold used to judge the abnormality of the winding of the transformer and the second amplitude threshold used to judge the abnormality of the core of the transformer can be determined. Through the above steps, the potential threats of winding looseness and core looseness can be realized, and the accuracy of abnormal diagnosis and early warning of the target transformer can be improved.

[0104] Based on the pre-processed current ground current, the status of the target transformer is evaluated, including:

[0105] According to the current grounding current, the first fault identification model based on attention convolutional neural network or reinforcement learning is used to identify the grounding fault and insulation damage fault of the target transformer.

[0106] In an embodiment of the present application, the convolutional neural network can extract local and global features of complex spatiotemporal ground current data through convolutional layers and pooling layers. Introducing the attention mechanism in the convolutional neural network can make the convolutional neural network pay more attention to the key features in the ground current data, thereby improving the accuracy of fault prediction. The main function of reinforcement learning is dynamic adaptation, which continuously adjusts the prediction strategy through learning to adapt to the complexity and uncertainty of the fault mode of the power system. Through continuous data learning and optimization, it is possible to more keenly identify minor anomalies in the ground current, such as weak harmonics, non-periodic fluctuations, etc., and improve the accuracy of ground current monitoring. Thereby improving the accuracy of the entire comprehensive monitoring and early warning method to a certain extent.

[0107] Based on the pre-processed current oil chromatogram data, the status of the target transformer is evaluated, including:

[0108] According to the current oil chromatography data, a second machine learning model is used to identify the overheating fault or discharge fault of the target transformer and its fault location, wherein the second machine learning model includes a GA-SVM model or a CEABC-WNN model, the GA-SVM model refers to a support vector machine SVM obtained by training by optimizing parameter combination using a genetic algorithm GA, and the CEABC-WNN model refers to a WNN model obtained by training by optimizing the initial parameter combination of a wavelet neural network WNN model using a chaos enhanced artificial bee colony algorithm CEABC.

[0109] In the embodiment of the present application, the oil chromatogram real-time monitoring is combined with the GA-SVM model, and the genetic algorithm GA is used to optimize the parameter combination of the support vector machine SVM. The SVM classifies and identifies overheating faults and discharge faults according to the components and concentrations of the dissolved gas in the target transformer oil monitored by the oil chromatogram in real time, so as to improve the accuracy and generalization ability of the early warning. CEABC improves the global search ability and convergence speed of the algorithm by introducing chaos theory to optimize the search process. The WNN model automatically extracts fault features by using the time-frequency localization characteristics of the wavelet transform. First, the initial weights and thresholds of the WNN model are optimized by CEABC to reduce the influence of the artificially set parameters, which can speed up the training process and help the WNN model better converge to the global optimal solution or a better local optimal solution. Subsequently, the WNN model is trained using the initial weights and thresholds optimized by CEABC so that it can accurately map the input features to the output fault types (thermal / electrical). During the training process, by comparing the fault types output by the WNN model with the actual fault labels of the input features, the parameters of the WNN model are continuously adjusted until the classification accuracy of the thermal faults and electrical faults is satisfactory, and the final WNN model is trained.

[0110] Optionally, in other embodiments of the present application, preprocessing the current state data further includes:

[0111] The current partial discharge signal is sequentially subjected to first amplification, mixing, first filtering, second amplification and second filtering. Then, a single discharge pulse of the partial discharge signal after the second filtering is subjected to time domain analysis, frequency domain analysis and time-frequency analysis to extract the time domain features, frequency domain features and time-frequency features of the partial discharge signal.

[0112] In the embodiment of the present application, the function of frequency mixing is to convert the local discharge signal to an intermediate frequency (IF) or a lower frequency for subsequent processing. The mixed signal usually contains multiple frequency components. The noise and interference signals introduced in the mixing process can be removed by the first filtering to improve the signal-to-noise ratio. The filter used for the first filtering is not specifically limited and can be selected according to actual needs. For example, a program-controlled filter (such as wavelet denoising, adaptive filtering, narrowband interference elimination, etc.) can be used to dynamically adjust the parameters of the program-controlled filter as needed, thereby selectively extracting the useful frequency band in the local discharge signal after frequency mixing. The purpose of performing the second amplification after the first filtering is to further improve the signal-to-noise ratio of the local discharge signal, making the local discharge signal clearer and more reliable. The filter used for the second filtering is not specifically limited and can be selected according to actual needs. For example, a bandpass filter is used to ensure the extraction of the useful frequency band in the local discharge signal after the second amplification.

[0113] The current ground current is sequentially subjected to third filtering, smoothing, data offset adjustment and normalization.

[0114] In ground current monitoring, due to differences in the characteristics of the monitoring device (such as sensitivity, measurement range) and changes in environmental conditions (such as temperature, humidity, and electromagnetic interference), the collected data often have problems with offset and scale inconsistency. In order to adjust the data offset and unify the data scale, data calibration is required, including offset correction and scale normalization. Offset correction aims to eliminate the offset of the measurement reference point due to inherent deviations of the device or environmental factors. The mean or median of the monitoring data of multiple monitoring points is calculated, and the original data is adjusted according to the mean or median. For example, the original data is subtracted from the mean or median to eliminate systematic errors. The Z-score method is used for scale normalization to convert the data into a standard normal distribution form, which is suitable for situations where the data range may change.

[0115] The current local temperature data is subjected to wavelet denoising and error correction in turn to eliminate the influence of environmental interference and the characteristics of the temperature monitoring device itself on the measurement results.

[0116] The median replacement and K nearest neighbor interpolation methods are used to clean the current oil chromatography data (concentration data of each gas component) to eliminate noise and outliers.

[0117] The current local vibration signal is subjected to the fourth filtering and blind source separation processing in sequence to improve the signal-to-noise ratio of the local vibration signal.

[0118] Optionally, in other embodiments of the present application, the collected partial discharge signal is mixed, including frequency reduction (local oscillator signal), sampling, comparison (comparison with a reference signal), and interference elimination in sequence.

[0119] In the embodiment of the present application, frequency reduction refers to multiplying the local discharge signal RF by the local oscillator signal (local oscillator signal) LO, and the generated signal includes a sum frequency (RF+LO) and a difference frequency (|RF-LO|). The local oscillator signal LO is not specifically limited and can be selected according to actual needs.

[0120] Optionally, in other embodiments of the present application, a sliding average filtering method is used to smooth the current ground current to reduce data fluctuations and improve data stability.

[0121] Optionally, in other embodiments of the present application, the third filtering includes at least one of low-pass digital filtering and Kalman filtering to remove high-frequency noise and random interference.

[0122] Optionally, in other embodiments of the present application, the fourth filtering adopts an adaptive filtering algorithm.

[0123] Optionally, evaluating the state of the target transformer according to the preprocessed current partial discharge signal further includes:

[0124] The pre-processed current multiple partial discharge signals collected by the constructed multi-dimensional sensor network are used to locate multiple discharge fault points and display them in three-dimensional space;

[0125] Using dynamic tracking algorithms, the discharge position, displacement, intensity change, and discharge type change of multiple discharge fault points are tracked and recorded in real time to ensure comprehensive control of the discharge phenomenon.

[0126] In the embodiments of the present application, multi-discharge fault point positioning, that is, a multi-dimensional sensor network constructed by combining multiple sensors such as electro-acoustic, electric-electric, and acoustic-acoustic sensors, realizes the three-dimensional spatial positioning of multiple discharge fault points through technologies such as signal time difference positioning and beam forming. The electro-acoustic sensor combination is the integrated use of sound sensors (microphones) and electrical sensors (such as current and voltage sensors). The electric inductor combination is the joint use of different types of electrical sensors (such as voltage, current, power, etc.). The acoustic-acoustic sensor combination is the joint use of multiple sound sensors (such as microphone arrays).

[0127] Optionally, in other embodiments of the present application, preprocessing the current state data specifically includes:

[0128] The local vibration signal after blind source separation is subjected to time domain analysis, frequency domain analysis and time-frequency analysis to extract the frequency complexity FCA, vibration stability DET, vibration correlation MPC and energy similarity EDR of the local vibration signal, and calculate the permutation entropy PE of the state data of the target transformer.

[0129] In the embodiment of the present application, time domain analysis refers to directly analyzing the local vibration signal in the time domain, focusing on the time domain characteristics of the signal, such as amplitude, root mean square (RMS), kurtosis, peak factor and other characteristics, and quickly detecting abnormal vibration. Frequency domain analysis refers to converting the local vibration signal from the time domain to the frequency domain using Fourier transform (FFT) to obtain a spectrum diagram. Through the spectrum diagram, specific frequency components of loose cores, such as low-frequency vibrations, can be identified. Frequency domain analysis can reveal the frequency components and distribution of vibration signals. Time-frequency analysis refers to analyzing vibration signals in the time domain and frequency domain using time-frequency analysis methods such as wavelet transform, short-time Fourier transform STFT and Hilbert-Huang transform (HHT) to obtain a time-frequency diagram. The results of time domain analysis, frequency domain analysis and time-frequency analysis can be combined to perform a comprehensive diagnosis of mechanical faults. For example, time domain analysis can quickly detect abnormal vibrations, frequency domain analysis can identify specific vibration fault frequencies, and time-frequency analysis can capture the non-stationary characteristics of vibration signals. Through the fusion of multiple methods, the location and type of vibration faults can be more accurately located. Frequency complexity reflects the richness of different frequency components in the signal and is an important indicator for judging the internal mechanical state of the target transformer. Changes in vibration stability may indicate looseness or wear of internal mechanical components. By calculating the vibration correlation of vibration signals at different locations, that is, different local vibration signals, the interaction and coordination between the internal mechanical components of the target transformer can be judged. Energy similarity is used to evaluate the similarity of the energy distribution of local vibration signals in different time periods. By comparing the energy distribution of local vibration signals, the mechanical stability of the target transformer in different states can be judged.

[0130] Optionally, in other embodiments of the present application, evaluating the state of the target transformer according to the preprocessed current local vibration signal further includes:

[0131] Before step 501, if at least one of the following occurs: the FCA of the current local vibration signal is greater than a preset FCA threshold, the DET is less than a preset DET threshold, the MPC is less than a preset MPC threshold, the EDR is greater than a preset EDR threshold, and the PE is greater than a preset PE threshold, it indicates that the target transformer may have a mechanical fault, and the mechanical fault is further determined according to steps 501 to 502.

[0132] In the embodiment of the present application, the FCA fault threshold, DET fault threshold, MPC fault threshold, EDR fault threshold and PE fault threshold are set according to historical data and experience. For example, according to historical data and experience, the normal range of FCA is 0.5-1.0, the normal range of DET is 0.7-1.0, the normal range of MPC is 0.8-1.0, the normal range of EDR is 0.5-0.8, and the normal range of PE is 0.5-0.8. Then, the FCA fault threshold is set to be greater than 1.0, the DET fault threshold is set to be less than 0.7, the MPC fault threshold is set to be less than 0.8, the EDR fault threshold is set to be less than 0.5, and the PE fault threshold is set to be greater than 0.8. If FCA exceeds the FCA threshold, it indicates that the frequency complexity of the local vibration signal is high, and a mechanical fault may exist. If DET is lower than the DET threshold, it indicates that the certainty and predictability of the local vibration signal are poor, and a mechanical fault may exist. If MPC is lower than the MPC threshold, it indicates that the mutual prediction ability between local vibration signals is poor, and a mechanical fault may exist. If the EDR exceeds the EDR threshold, it indicates that the energy distribution of the local vibration signal in different time periods is quite different, and there may be a mechanical fault. If the PE exceeds the threshold, it indicates that the complexity and randomness of the state data of the target transformer are high, and there may be a mechanical fault. By comprehensively analyzing the time-frequency characteristics of the local vibration signal and the permutation entropy value of the state data of the target transformer, the mechanical stability of the target transformer can be effectively evaluated and faults can be identified. Combining FCA, DET, MPC, EDR and PE, the state of the target transformer can be more comprehensively evaluated, potential mechanical faults can be discovered in a timely manner, and the reliable operation of the target transformer can be ensured.

[0133] Optionally, in other embodiments of the present application, preprocessing the current state data specifically includes:

[0134] After cleaning the current oil chromatographic data (concentration data of each gas component) by using the median replacement and K-nearest neighbor interpolation methods, the oil chromatographic sample data is subjected to DGA analysis to extract the key fault features of the oil chromatographic sample data.

[0135] In the embodiment of the present application, the key fault characteristics of the oil chromatogram sample data extracted by the DGA analysis method refer to the concentration and concentration ratio of the dissolved gas in the transformer oil extracted by the DGA analysis method. The dissolved gas in the transformer oil includes but is not limited to H2, CH4, CO, C2H6, CO2, C2H4, C2H2 and other gases, and the concentration ratio of the dissolved gas in the transformer oil includes but is not limited to C2H2 / C2H4 (acetylene to ethylene concentration ratio), CH4 / H2 (methane to hydrogen concentration ratio) and C2H4 / C2H6 (ethylene to ethane concentration ratio).

[0136] According to the current oil chromatogram data, a second machine learning model is used to identify the overheating fault or discharge fault of the target transformer and its fault location, including:

[0137] The key fault features of the extracted oil chromatogram sample data are input into the second machine learning model to identify the overheating fault or discharge fault of the target transformer and its fault location.

[0138] On the basis of adopting the second machine learning model to explicitly distinguish between thermal faults and electrical faults, the corresponding DGA (oil chromatography analysis) is combined to extract the key fault features of the oil chromatography sample data to more accurately determine the location of the thermal fault and electrical fault machines.

[0139] Optionally, in other embodiments of the present application, the process of optimizing the initial parameters of the WNN model using CEABC includes the following steps (1) to (11).

[0140] Step (1), define the structure of the WNN model, including the number of nodes in the input layer, wavelet hidden layer and output layer.

[0141] Step (2), initialize the parameters of the CEABC algorithm, which include the number of honey populations and the number of iterations.

[0142] Step (3), initialize the weights and thresholds (initial values ​​of each node) of the WNN model. The weights are the parameters connecting the hidden layer (wavelet basis function layer) and the output layer. These weights determine the influence of the hidden layer output on the final output.

[0143] Step (4), encode the weights and thresholds of the WNN model into food sources in the CEABC algorithm.

[0144] Step (5), for each food source, calculate the loss function of its corresponding WNN model on the training data, and use the loss function as the fitness.

[0145] In step (6), each employed bee searches for a new food source (i.e., a new weight and threshold) near its corresponding food source, and evaluates the fitness corresponding to the searched new food source.

[0146] Step (7), if the fitness of the new food source is better than that of the old food source, the old food source is replaced.

[0147] Step (8), follower bees are allocated according to the fitness ratio of the food source, and the follower bees select food sources with high fitness for local search.

[0148] Step (9), if a food source has not been improved after multiple iterations, it is considered to be exhausted (trapped in a local optimum). At this time, the scout bee is called to randomly select a new food source to replace the exhausted food source, that is, the scout bee is called to explore the new food source again.

[0149] In step (10), chaotic mapping is used to update the location of the food source to increase the randomness of the search.

[0150] Step (11), repeat the above steps (5) to (10) until the preset termination condition (such as the maximum number of iterations) is reached, output the optimal food source and its fitness value, and set the optimal food source as the initial weight and threshold of the optimized WNN.

[0151] In the embodiment of the present application, the iteration of CEABC (repeated calculation of step (5) to step (10)) and the optimization process are combined with multi-strategy fusion and adaptive adjustment, and adaptive adjustment is performed according to the actual situation in the search process, making the algorithm more flexible and efficient.

[0152] Optionally, in other embodiments of the present application, evaluating the state of the target transformer according to the preprocessed current grounding current further includes:

[0153] Based on the preprocessed historical ground current data and the preprocessed current ground current data of the target transformer, the Bi-LSTM algorithm is used to predict the future ground current;

[0154] Compare the future ground current with the currently collected ground current to determine the deviation between the two;

[0155] If the deviation between the future grounding current and the currently acquired grounding current exceeds the set value, it indicates that the target transformer is grounded abnormally.

[0156] In the embodiment of the present application, through the above steps, potential grounding faults can be evaluated and warned, thereby improving the accuracy of monitoring and warning of the state of the target transformer.

[0157] Optionally, in other embodiments of the present application, the state of the target transformer is evaluated from multiple angles according to the preprocessed current state data, further comprising the following steps 601 to 605. Among them:

[0158] Step 601, calculating the correlation coefficient between the pre-processed current state data.

[0159] In the embodiment of the present application, calculating the correlation coefficient between each state data is to calculate the correlation coefficient between any two variables. There is no specific limitation on the type of the correlation coefficient, which can be selected according to actual needs.

[0160] Step 602: construct a correlation coefficient matrix according to the correlation coefficients between the state data.

[0161] Step 603, based on the correlation coefficient matrix, construct an undirected graph network, wherein the name of the node of the undirected graph network is the name of each state data (such as oil chromatography data, local discharge signal, ground current, local temperature data and local vibration signal), the attribute value of the node is the corresponding state data value, and the edge is the connection line between the significantly correlated state data pairs; wherein, if the correlation coefficient of two state data is greater than a preset correlation coefficient threshold, they are defined as a group of the significantly correlated state data pairs.

[0162] In the embodiment of the present application, the correlation coefficient threshold is not specifically limited and can be set according to actual needs. For example, the correlation coefficient threshold is set to 0.5.

[0163] Step 604: Identify independent communities in the undirected graph network using a community detection algorithm.

[0164] Step 605 , identifying the abnormal state of the target transformer according to the identified independent community.

[0165] In the embodiment of the present application, the abnormal state of the target transformer can be identified according to the identified independent community through community structure analysis. For example, the connectivity and stability within each independent community are checked. If an abnormal disconnection or a large number of new edges occur within an independent community, it may indicate that there is a local problem in the independent community. Among them, the abnormal disconnection refers to that at least one node within the independent community has lost all the edges connected to it, and the addition of a large number of edges refers to that the number of new edges connected to it by at least one node within the independent community exceeds the set edge number threshold. The embodiment of the present application does not specifically limit the edge number threshold, which can be set according to actual needs. The abnormal state of the target transformer can be identified according to the identified independent community through the change analysis of the attribute value of the node. For example, the change of the attribute value of the node of each independent community over time is observed. If the attribute value of some nodes exceeds the preset safe operation threshold (such as a local temperature data value exceeds the safe operation threshold), it means that these nodes have failed. The attribute value of the node of each independent community can also be compared with the attribute value of the node of the independent community during historical normal operation. The difference obtained by comparison is greater than the preset difference threshold, which means that the node with a difference greater than the preset difference threshold has a corresponding fault.

[0166] Optionally, in other embodiments of the present application, the state of the target transformer is evaluated from multiple angles according to the preprocessed current state data, further comprising:

[0167] A hierarchical clustering model based on bottom-up aggregation strategy is used to cluster the preprocessed current state data;

[0168] Comprehensive fault diagnosis is performed based on the clustering result to identify the fault of the target transformer, which includes at least one of a partial discharge fault, a mechanical fault, a grounding fault, an insulation fault and an overheating fault.

[0169] In the embodiment of the present application, the state data includes oil chromatography data, local discharge signal, ground current, local temperature data and local vibration signal. First, the sample data set is cleaned and standardized, and then each data point (each monitoring data) is used as the initial clustering, and the similarity between clusters is calculated using the Euclidean distance, and the most similar clusters are gradually merged to form a higher level clustering structure; this process is iterated until the number of clusters reaches the optimal value or the similarity between clusters is lower than the threshold. In the clustering process, the Calinski-Harabasz index evaluation index is combined to monitor the clustering effect, and the parameters of the hierarchical clustering model are continuously adjusted to optimize the clustering results. Taking each data point as the initial clustering, although it avoids the subjectivity of artificially presetting the number of clusters or cluster centers, it also provides rich initial information for subsequent hierarchical clustering, which helps to discover the potential structure and pattern in the data. The hierarchical clustering model can merge data points with similar characteristic faults into the same cluster through step-by-step iterative merging, Euclidean distance calculation, and Calinski-Harabasz evaluation, thereby realizing rapid identification and positioning of fault categories.

[0170] Optionally, in other embodiments of the present application, the above transformer comprehensive monitoring and early warning method further includes the following steps 701 to 704. Among them:

[0171] Step 701, construct a digital twin model of the target transformer.

[0172] In an embodiment of the present application, a digital twin model of the target transformer is established using multi-physical field simulation (multi-source simulation data of the transformer under different working conditions obtained through finite element analysis (FEA) and other methods (oil chromatography data, local discharge signal, local temperature data, local vibration signal, grounding current)), measured multi-source data of the transformer, and historical operation data of the transformer (including operation records, maintenance records, fault records and other historical operation data) as the original data pool.

[0173] Step 702: The operation process of the target transformer is synchronously simulated through the digital twin model, and the status data of the digital twin model is collected in real time.

[0174] Step 703: input the input sequence into the temporal difference learning model, and predict the future state data of the target transformer through the temporal difference learning model; wherein the input sequence includes: multiple pairs of input data, each pair of input data includes the state data of the digital twin model and the target transformer at the same time.

[0175] For example, suppose there are n pairs of input data in the input sequence, and the current time is T m , the input sequence may include:

[0176] Data(T z ),Data'(T z ),Data(T z +1),Data'(T z +1),...Data(T m ),Data'(T m ), where Data(T i ) indicates that the target transformer is at T i Status data at the moment, Data'(T i ) indicates that the digital twin model is in T i The status data at a certain moment, i is an integer not less than z and not greater than m, z=m-n+1.

[0177] Step 704, taking the future state data of the target transformer as the current state data of the target transformer, performing preprocessing on the current state data and subsequent steps to issue an early warning for each future abnormal state of the target transformer.

[0178] In the embodiment of the present application, during the training of the time difference learning model, the mean square error MSE is used as the loss function to evaluate the difference between the predicted value and the actual value, and the model parameters are optimized by the back propagation algorithm. When the digital model is dynamically updated, the time difference learning model is retrained or fine-tuned according to the newly acquired measured data or simulation results, so as to continuously train and improve the time difference learning model, so that its generalization ability is increasingly strong, thereby improving the prediction accuracy, and the retrained or fine-tuned time difference learning model is used to predict the dynamics of the target transformer.

[0179] Optionally, in other embodiments of the present application, the transformer comprehensive monitoring and early warning method further includes:

[0180] A status assessment report is generated based on the evaluation results obtained from the multi-angle evaluation of the status of the target transformer.

[0181] In the embodiment of the present application, the form of the generated status assessment report can be selected by oneself, including but not limited to tables, charts, text descriptions, interactive reports, etc., and the above forms can also be combined to ensure the comprehensiveness and readability of the information. The status assessment report contains the status data of the target transformer collected in real time, the multi-angle assessment results of the target transformer, early warning measures and maintenance suggestions.

[0182] The development of maintenance recommendations includes:

[0183] 1) Priority sorting: prioritize multiple faults (abnormalities) to ensure that critical faults are handled first;

[0184] 2) Provide specific maintenance plans for different types of faults, according to relevant standards or maintenance manuals;

[0185] 3) Risk assessment: evaluate the risks that may be encountered during the maintenance process and provide corresponding preventive measures based on relevant standards or maintenance manuals.

[0186] Based on the same inventive concept, the embodiment of the present application also provides a transformer comprehensive monitoring and early warning system for implementing the above-mentioned transformer comprehensive monitoring and early warning method. The implementation scheme for solving the problem provided by the system is similar to the implementation scheme recorded in the above-mentioned method, so the specific limitations in one or more transformer comprehensive monitoring and early warning system embodiments provided below can refer to the above-mentioned limitations on the transformer comprehensive monitoring and early warning method, and will not be repeated here.

[0187] In an exemplary embodiment, Figure 2 As shown, a transformer comprehensive monitoring and early warning system 80 is provided, comprising:

[0188] The acquisition module 801 is used to acquire the status data of the target transformer in real time, and the status data includes oil chromatography data, local discharge signal, ground current, local temperature data and local vibration signal;

[0189] The data processing module 802 is used to pre-process the current state data;

[0190] A multi-angle evaluation module 803 is used to evaluate the state of the target transformer from multiple angles according to the pre-processed current state data;

[0191] The early warning module 804 is used to issue an early warning for each abnormal state of the target transformer if at least one angle evaluation shows that the state of the target transformer is abnormal.

[0192] In the embodiment of the present application, the data processing module 802 may be a hardware chip with data processing functions, and the multi-angle evaluation module 803 may be a server or terminal host connected to the data processing module 802 via a switch, and the switch may also include a backup switch. The transformer comprehensive monitoring and early warning system also includes a data storage system, which is connected to the switch and is at the same layer as the terminal host. The stored data is as follows:

[0193] Save the local discharge data, ground current data, local vibration data, oil chromatography data, and local temperature data collected within the past two years. When the local discharge data is normal, the storage interval is ≤1min, and when the ground current data is normal, the data storage interval is ≤1min. When an abnormality occurs, all data should be recorded; when the local vibration data is normal, the storage interval is ≤1min, and when an abnormality occurs, all data should be recorded; the local temperature data detection interval is ≤1h, and the oil chromatography data storage interval is ≤1day, and all stored data contain time information, accurate to seconds.

[0194] The process of performing multi-angle evaluation on the status of the target transformer is described in detail in the above-mentioned transformer comprehensive monitoring and early warning method embodiment, and will not be repeated here.

[0195] Optionally, in other embodiments of the present application, the data processing module 802 is further used to:

[0196] The current partial discharge signal is sequentially subjected to first amplification, mixing, first filtering, second amplification and second filtering. Then, a single discharge pulse of the partial discharge signal after the second filtering is subjected to time domain analysis, frequency domain analysis and time-frequency analysis to extract the time domain features, frequency domain features and time-frequency features of the partial discharge signal;

[0197] The current ground current is sequentially subjected to third filtering, smoothing, data offset adjustment and normalization;

[0198] Perform wavelet denoising and error correction on the current local temperature data in turn to eliminate the influence of environmental interference and the characteristics of the temperature monitoring device itself on the measurement results;

[0199] The current oil chromatography data (concentration data of each gas component) is cleaned by using median replacement and K nearest neighbor interpolation methods to eliminate noise and outliers;

[0200] The current local vibration signal is subjected to the fourth filtering and blind source separation processing in sequence to improve the signal-to-noise ratio of the local vibration signal.

[0201] Optionally, in other embodiments of the present application, the multi-angle evaluation module 803 is further used to:

[0202] The current multiple partial discharge signals collected by the pre-processed multi-dimensional sensor network are located at multiple discharge fault points and displayed in three-dimensional space;

[0203] Using dynamic tracking algorithms, the discharge position, displacement, intensity change, and discharge type change of multiple discharge fault points are tracked and recorded in real time to ensure comprehensive control of the discharge phenomenon.

[0204] Optionally, in other embodiments of the present application, the data processing module 802 is further used to:

[0205] The local vibration signal after blind source separation is subjected to time domain analysis, frequency domain analysis and time-frequency analysis to extract the frequency complexity FCA, vibration stability DET, vibration correlation MPC and energy similarity EDR of the local vibration signal, and calculate the permutation entropy PE of the state data of the target transformer.

[0206] Accordingly, the multi-angle evaluation module 803 is also used for:

[0207] Before step 501, if at least one of the following occurs: the FCA of the current local vibration signal is greater than a preset FCA threshold, the DET is less than a preset DET threshold, the MPC is less than a preset MPC threshold, the EDR is greater than a preset EDR threshold, and the PE is greater than a preset PE threshold, it indicates that the target transformer may have a mechanical fault, and the mechanical fault is further determined according to steps 501 to 502.

[0208] Optionally, in other embodiments of the present application, the data processing module 802 is further used to:

[0209] After cleaning the current oil chromatographic data (concentration data of each gas component) by using the median replacement and K-nearest neighbor interpolation methods, the oil chromatographic sample data is subjected to DGA analysis to extract the key fault features of the oil chromatographic sample data.

[0210] Accordingly, the multi-angle evaluation module 803 is also used for:

[0211] The key fault features of the extracted oil chromatogram sample data are input into the second machine learning model to identify the overheating fault or discharge fault of the target transformer and its fault location.

[0212] Optionally, in other embodiments of the present application, the multi-angle evaluation module 803 is further used to:

[0213] The initial parameters of the WNN model are optimized using CEABC according to the following process:

[0214] Define the structure of the WNN model, including the number of nodes in the input layer, wavelet hidden layer, and output layer;

[0215] Initialize the parameters of the CEABC algorithm, including the number of honey populations and the number of iterations;

[0216] Initialize the weights and thresholds (initial values ​​of each node) of the WNN model. The weights are the parameters connecting the hidden layer (wavelet basis function layer) and the output layer. These weights determine the influence of the hidden layer output on the final output.

[0217] The weights and thresholds of the WNN model are encoded into food sources in the CEABC algorithm;

[0218] For each food source, calculate the loss function of its corresponding WNN model on the training data, and use the loss function as the fitness;

[0219] Each employed bee searches for a new food source (i.e., a new weight and threshold) near its corresponding food source and evaluates the fitness corresponding to the searched new food source;

[0220] If the fitness of the new food source is better than that of the old food source, the old food source is replaced;

[0221] Follower bees are assigned according to the fitness ratio of food sources, and follower bees select food sources with high fitness for local search;

[0222] If a food source has not been improved after multiple iterations, it is considered to have been exhausted (trapped in a local optimum). At this time, the scout bees are called to randomly select a new food source to replace the exhausted food source, that is, the scout bees are called to explore the new food source again;

[0223] Use a chaotic map to update the location of the food source to increase the randomness of the search;

[0224] Repeat the above steps for each food source, calculate the loss function of its corresponding WNN model on the training data, and use the loss function as the fitness to update the position of the food source using chaotic mapping until the preset termination condition (such as the maximum number of iterations) is reached, output the optimal food source and its fitness value, and use the optimal food source as the initial weight and threshold of the optimized WNN.

[0225] Optionally, in other embodiments of the present application, the multi-angle evaluation module 803 is further used to:

[0226] Based on the historical grounding current data of the target transformer and the currently collected grounding current data, the Bi-LSTM algorithm is used to predict the future grounding current;

[0227] Compare the future ground current with the currently collected ground current to determine the deviation between the two;

[0228] If the deviation between the future grounding current and the currently acquired grounding current exceeds the set value, it indicates that the target transformer is grounded abnormally.

[0229] Optionally, in other embodiments of the present application, the multi-angle evaluation module 803 is further used to:

[0230] For the preprocessed current state data, calculate the correlation coefficient between each state data;

[0231] According to the correlation coefficients between the state data, a correlation coefficient matrix is ​​constructed;

[0232] Based on the correlation coefficient matrix, an undirected graph network is constructed, wherein the names of the nodes of the undirected graph network are the names of the state data (such as oil chromatography data, local discharge signal, ground current, local temperature data and local vibration signal), the attribute values ​​of the nodes are the corresponding state data values, and the edges are the connecting lines between the significantly correlated state data pairs; wherein, if the correlation coefficient of two state data is greater than a preset correlation coefficient threshold, they are defined as a group of significantly correlated state data pairs;

[0233] Identify independent communities in undirected graph networks through community detection algorithms;

[0234] Based on the identified independent communities, the abnormal status of the target transformer is identified.

[0235] Optionally, in other embodiments of the present application, the multi-angle evaluation module 803 is further used to:

[0236] A hierarchical clustering model based on bottom-up aggregation strategy is used to cluster the preprocessed current state data;

[0237] Comprehensive fault diagnosis is performed based on the clustering result to identify the fault of the target transformer, which includes at least one of a partial discharge fault, a mechanical fault, a grounding fault, an insulation fault and an overheating fault.

[0238] Optionally, in other embodiments of the present application, the multi-angle evaluation module 803 is further used to:

[0239] Build a digital twin model of the target transformer;

[0240] The operation process of the target transformer is synchronously simulated through the digital twin model, and the status data of the digital twin model is collected in real time;

[0241] Inputting the input sequence into the temporal difference learning model, and predicting the future state data of the target transformer through the temporal difference learning model; wherein the input sequence includes: multiple pairs of input data, each pair of input data includes the state data of the digital twin model and the target transformer at the same time, and the relevant introduction is detailed in the description of the above method embodiment;

[0242] The future state data of the target transformer is used as the current state data of the target transformer, and preprocessing of the current state data and subsequent steps are performed to warn of each future abnormal state of the target transformer.

[0243] Optionally, in other embodiments of the present application, the above-mentioned transformer comprehensive monitoring and early warning system further includes:

[0244] The report generation module 805 is used to generate a status assessment report according to the assessment results obtained by performing multi-angle assessment on the status of the target transformer.

[0245] Optionally, in other embodiments of the present application, the acquisition module 801 includes:

[0246] The temperature acquisition module 8011 is used to acquire the temperature of the target transformer in real time, including the oil temperature and the shell temperature;

[0247] The partial discharge acquisition module 8012 is used to acquire partial discharge signals of the target transformer in real time;

[0248] The ground current acquisition module 8013 is used to acquire the ground current of the target transformer in real time;

[0249] The local vibration acquisition module 8014 is used to acquire the local vibration signal of the target transformer in real time;

[0250] The oil chromatogram acquisition module 8015 is used to acquire the oil chromatogram data of the target transformer in real time. The oil chromatogram data refers to the gas components in the target transformer oil and the concentration data of each gas.

[0251] In an exemplary embodiment, a computer device is provided. The computer device may be a server or a terminal. The internal structure diagram thereof may be as follows: Figure 3 As shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, referred to as I / O) and a communication interface. The processor, the memory and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store processing data for comprehensive monitoring and early warning of transformers. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a method for comprehensive monitoring and early warning of transformers is implemented.

[0252] Those skilled in the art will understand that Figure 3 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0253] In an exemplary embodiment, a computer device is further provided, including a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the steps in the above-mentioned method embodiments when executing the computer program.

[0254] In an exemplary embodiment, a computer-readable storage medium is provided, storing a computer program, and when the computer program is executed by a processor, the steps in the above method embodiments are implemented.

[0255] In an exemplary embodiment, a computer program product is provided, including a computer program, and when the computer program is executed by a processor, the steps in the above method embodiments are implemented.

[0256] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant regulations.

[0257] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM may be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM).

[0258] The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. The non-relational database may include a distributed database based on blockchain, etc., but is not limited thereto. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but is not limited thereto.

[0259] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0260] This article uses specific examples to illustrate the principles and implementation methods of this application. The description of the above embodiments is only used to help understand the method and core ideas of this application. At the same time, for those skilled in the art, according to the ideas of this application, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting this application.

Claims

1. A transformer comprehensive monitoring and early warning method, characterized in that: The transformer comprehensive monitoring and early warning method comprises: Collecting status data of the target transformer in real time, the status data including oil chromatography data, local discharge signal, grounding current, local temperature data and local vibration signal; Preprocessing the current state data; According to the preprocessed current state data, the state of the target transformer is evaluated from multiple angles; If at least one angle evaluation shows that the state of the target transformer is abnormal, an early warning is issued for each abnormal state of the target transformer.

2. The transformer comprehensive monitoring and early warning method according to claim 1 is characterized in that: Before evaluating the state of the target transformer from multiple angles according to the preprocessed current state data, the method further includes: Construct a first machine learning model, and use the particle swarm algorithm and InputData1 and T1 to optimize the parameters of the first machine learning model, where InputData1 is a plurality of local temperature data of the sample transformer, T1 is the overall temperature of the sample transformer, InputData1 is used as the input of the first machine learning model, and T1 is used as the output of the first machine learning model. The optimal parameter combination is found through iterative search: θ opt =argmin θ Error(f ML (InputData1,θ j )); Among them, f ML represents the first machine learning model, θ j represents the first influencing factor, wherein j={1,2,…,n1}, n1 represents the number of the first influencing factors, the first influencing factor refers to the factor affecting T1, Error represents the prediction error function of the first machine learning model, argmin θ Error represents the parameter combination of the first machine learning model when the prediction error Error of the first machine learning model is the smallest, θ opt Represents the optimal parameter combination found by the particle swarm algorithm; The multi-angle evaluation of the state of the target transformer according to the pre-processed current state data includes: According to multiple second influencing factors θ i and the current InputData2, using the parameter combination θ opt The first machine learning model predicts the current overall temperature T2 of the target transformer; wherein i={1,2,…,n2}, n2 represents the number of the second influencing factors, the second influencing factors refer to factors affecting T2, and InputData2 represents multiple local temperature data of the target transformer; Calculate the temperature rise rate Δv of the target transformer within Δt time T : Wherein, T'2 represents the overall temperature of the target transformer predicted last time, Δt=t2-t1, t2 represents the acquisition time point of the current InputData2, and t1 represents the acquisition time point of the previous InputData2; If Δv T If the temperature rise rate is greater than a preset threshold value, it indicates that the temperature rise of the target transformer is abnormal.

3. The transformer comprehensive monitoring and early warning method according to claim 1 is characterized in that: The method of evaluating the state of the target transformer from multiple angles according to the preprocessed current state data also includes: In the method according to the plurality of second influencing factors θ i and the current InputData2, using the parameter combination θ opt After the first machine learning model predicts the current overall temperature T2 of the target transformer, according to a plurality of second influencing factors θ i and the current InputData2, based on the Bayesian model, determine P(T2>T threshold ,θ i |InputData2): Among them, P(InputData2|T2>T threshold ,θ i ) indicates that when the threshold T2 and multiple second influencing factors θ i The likelihood of observing InputData2 under this condition, P(T2>T threshold ,θ i ) means greater than T threshold T2 and multiple second influencing factors θ i The prior probability of InputData2 is P(InputData2), which represents the marginal likelihood of InputData2, and P(T2>T threshold ,θ i |InputData2) indicates that it is greater than T threshold T2 and multiple second influencing factors θ i The posterior probability of threshold is the set temperature threshold; If P(T2>T threshold ,θ i |InputData2) is greater than the set probability threshold, indicating that the temperature of the target transformer is abnormal.

4. The transformer comprehensive monitoring and early warning method according to claim 1 is characterized in that: The preprocessing of the current state data specifically includes: Perform time domain analysis, frequency domain analysis and time-frequency analysis on a single discharge pulse of the current partial discharge signal to extract the time domain features, frequency domain features and time-frequency features of the partial discharge signal; The state of the target transformer is evaluated from multiple angles according to the preprocessed current state data, including: Inputting the extracted time domain features, frequency domain features and time-frequency features of the partial discharge signal into a constructed insulation fault assessment model, the insulation fault assessment model outputting a corresponding insulation threat level; If the insulation threat level output by the insulation fault assessment model is greater than a preset insulation threat level threshold, it indicates that the insulation performance of the target transformer is abnormal.

5. The transformer comprehensive monitoring and early warning method according to claim 1 is characterized in that: The state of the target transformer is evaluated from multiple angles according to the preprocessed current state data, including: If, among the current plurality of local vibration signals, at least some of the local vibration signals have frequencies that are integer multiples of 50 Hz and amplitudes that are greater than a preset first amplitude threshold, it indicates that the winding of the target transformer is loose; If among the current plurality of local vibration signals, at least some of the local vibration signals have a frequency greater than 20 Hz and less than 50 Hz and an amplitude greater than a preset second amplitude threshold, it indicates that the core of the target transformer is loose; According to the grounding current, a first fault identification model based on an attention convolutional neural network or reinforcement learning is used to identify the grounding fault and insulation damage fault of the target transformer.

6. The transformer comprehensive monitoring and early warning method according to claim 1 is characterized in that: The state of the target transformer is evaluated from multiple angles according to the preprocessed current state data, including: According to the current oil chromatogram data, a second machine learning model is used to identify the overheating fault or discharge fault and the fault location of the target transformer, wherein the second machine learning model includes a GA-SVM model or a CEABC-WNN model; wherein the GA-SVM model refers to the final SVM obtained by training the parameter combination of the support vector machine SVM optimized by the genetic algorithm GA; the CEABC-WNN model refers to the final WNN model obtained by training the initial parameter combination of the wavelet neural network WNN model optimized by the chaos enhanced artificial bee colony algorithm CEABC.

7. The transformer comprehensive monitoring and early warning method according to claim 1 is characterized in that: The state of the target transformer is evaluated from multiple angles according to the preprocessed current state data, including: Calculate the correlation coefficient between the current state data after preprocessing; According to the correlation coefficients between the state data, a correlation coefficient matrix is ​​constructed; Based on the correlation coefficient matrix, an undirected graph network is constructed, wherein the names of the nodes of the undirected graph network are the names of the state data, the attribute values ​​of the nodes are the corresponding state data values, and the edges are the connecting lines between the significantly correlated state data pairs; wherein, if the correlation coefficient of two state data is greater than a preset correlation coefficient threshold, they are defined as a group of the significantly correlated state data pairs; identifying independent communities in the undirected graph network by a community detection algorithm; According to the identified independent community, an abnormal state of the target transformer is identified.

8. The transformer comprehensive monitoring and early warning method according to claim 1 is characterized in that: The state of the target transformer is evaluated from multiple angles according to the preprocessed current state data, including: Clustering the pre-processed current state data using a hierarchical clustering model based on a bottom-up aggregation strategy; Comprehensive fault diagnosis is performed based on the clustering result to identify the fault of the target transformer, wherein the fault includes at least one of a partial discharge fault, a mechanical fault, a grounding fault, an insulation fault and an overheating fault.

9. The transformer comprehensive monitoring and early warning method according to any one of claims 1 to 8, characterized in that: Also includes: Constructing a digital twin model of the target transformer; Synchronously simulating the operation process of the target transformer through the digital twin model, and collecting state data of the digital twin model in real time; Inputting an input sequence into a temporal difference learning model, and predicting future state data of the target transformer through the temporal difference learning model; wherein the input sequence includes: multiple pairs of input data, each pair of input data includes state data of the digital twin model and the target transformer at the same time; The future state data of the target transformer is used as the current state data of the target transformer, and preprocessing and subsequent steps are performed on the current state data to issue an early warning for each future abnormal state of the target transformer.

10. A transformer comprehensive monitoring and early warning system, characterized in that: The transformer comprehensive monitoring and early warning system includes: An acquisition module, used for real-time acquisition of current status data of the target transformer, wherein the status data includes oil chromatogram data, partial discharge signal, ground current, local temperature data and local vibration signal; A data processing module, used for preprocessing the current state data; A multi-angle evaluation module, used to evaluate the state of the target transformer from multiple angles according to the pre-processed current state data; The early warning module is used to issue an early warning for each current abnormal state of the target transformer if at least one angle evaluation shows that the state of the target transformer is abnormal.

Images