High-voltage bushing fault grading alarm online monitoring method and device and storage medium

Abstract

The invention relates to a high-voltage bushing fault grading alarm online monitoring method and device and a storage medium, and aims to realize high-voltage bushing fault monitoring. The method comprises the steps of obtaining time sequence detection data; a time sequence oil pressure parameter, a time sequence temperature parameter and a time sequence humidity parameter in the time sequence detection data are subjected to normalization processing and then input into a pre-trained high-voltage bushing fault detection model to calculate a deviation parameter of an actual neighborhood interval difference mean value; determining whether a neighborhood interval difference mean value has a fault according to the deviation parameter and a measurement threshold value, and determining a fault level according to the degree of deviation from a normal range; wherein the measurement threshold value is obtained through a high-voltage bushing fault detection model in the training process. According to the method, time sequence modeling is carried out on time sequence detection data in a normal operation state by adopting a self-supervision mode, and distribution of a neighborhood interval difference mean value of a prediction result is modeled by utilizing a Gaussian mixture model, so that the extreme degree of the neighborhood interval difference mean value under a normal condition is determined, and high-voltage bushing fault monitoring is carried out.

Description

Technical Field

[0001] The present invention relates to the technical field of high-voltage bushing fault detection, and in particular to an online monitoring method, device and storage medium for high-voltage bushing fault grading and alarming. Background Art

[0002] The high-voltage bushing is an important electrical equipment in power facilities. The bushing in long-term operation may fail due to various reasons. Therefore, the condition detection of the bushing is a very important task for the power department. At present, the condition detection of the bushing mainly relies on the power-off test method. By measuring the insulation resistance, dielectric loss, capacitance and frequency-domain dielectric response, etc., to judge its condition, which has a great impact on production and operation. At the same time, the power-off test is often carried out only once every few years, and the development of defects during operation cannot be grasped, which may lead to the occurrence of dangerous accidents. Therefore, the overhaul test cycle should not be too short, and an overly long test cycle is difficult to timely grasp the operation status of the equipment. The continuity of operation and the timeliness of maintenance constitute a pair of contradictions. In order to detect internal faults of the bushing and give early warnings, relevant research has been carried out on the online monitoring method of transformer bushings, such as neutral point monitoring method, dielectric loss monitoring method, partial discharge monitoring, etc., which are easily affected by factors such as strong electromagnetic fields, corona discharges, grounding systems, carrier communications and climate environments of surrounding equipment. To sum up, the existing mainstream monitoring methods have not achieved good results and have not been widely applied. Most domestic transformer bushings are still in the blind area of monitoring, and bushing explosion accidents occur from time to time. Summary of the Invention

[0003] In order to solve the above technical problems or at least partially solve the above technical problems, the present invention provides an online monitoring method, device and storage medium for high-voltage bushing fault grading and alarming.

[0004] In a first aspect, the present invention provides an online monitoring method for high-voltage bushing fault grading and alarming, including:

[0005] Collect the oil pressure, temperature and humidity of any-phase high-voltage bushing through a pressure sensor, a temperature sensor and a humidity sensor, and obtain the time series of the oil pressure, temperature and humidity of any-phase high-voltage bushing;

[0006] Obtain the timing detection data for high-voltage bushing fault detection according to the time series of the oil pressure, temperature and humidity of any-phase high-voltage bushing;

[0007] Normalize the timing oil pressure parameter, the timing temperature parameter and the timing humidity parameter in the timing detection data respectively and input them into the pre-trained high-voltage bushing fault detection model; wherein, the high-voltage bushing fault detection model includes:

[0008] A bidirectional recurrent neural network for performing time series modeling on time series detection data under normal conditions, and a Gaussian mixture model for modeling the mean value of the neighborhood interval difference between real time series detection data and time series detection data predicted by the bidirectional recurrent neural network under normal conditions; using the bidirectional recurrent neural network to predict time series detection data, calculating the mean value of the corresponding neighborhood interval difference from the real time series detection data, and calculating the deviation parameter of the actual mean value of the neighborhood interval difference using the cumulative distribution function of the Gaussian mixture model;

[0009] Determine whether the mean value of the neighborhood interval difference is faulty according to the deviation parameter and the measurement threshold, and determine the fault level according to the degree of deviation of the deviation parameter from the measurement threshold; wherein, the measurement threshold is obtained through a high-voltage bushing fault detection model during the training process.

[0010] Furthermore, the time series detection data integrates the differences between any two parameters of the oil pressure, temperature, and humidity data of the three-phase high-voltage bushing, and is expressed as:

[0011] The time series detection data is expressed as:

[0012] where P is the time series oil pressure parameter, P = [p1, p2,... p t ... p T , where the data at time t in the time series oil pressure parameter is are the oil pressures of the three-phase high-voltage bushings at time t respectively, are the oil pressure differences between the first and second, first and third, and second and third phases of the high-voltage bushing at time t respectively;

[0013] TEM is the time series temperature parameter, TEM = [tem1, tem2,... tem t ... tem T , where the data at time t in the time series temperature parameter is are the temperatures of the three-phase high-voltage bushings at time t respectively, are the temperature differences between the first and second, first and third, and second and third phases of the high-voltage bushing at time t respectively;

[0014] H is the time series humidity parameter, H = [h1, h2,... h t ... h T , where the data at time t in the time series humidity parameter is are the humidities of the three-phase high-voltage bushings at time t respectively, are the humidity differences between the first and second, first and third, and second and third phases of the high-voltage bushing at time t respectively.

[0015] Further, the bidirectional recurrent neural network is trained using the training time series detection data under normal operation of the high-voltage bushing. The training method is as follows: The training time series detection data is divided according to a set time window. The time series detection data within any time window is input into the bidirectional recurrent neural network, and the bidirectional recurrent time network predicts the training time series detection data of the history and the future set time within this time window. The training in the above training process uses the squared L2 norm to calculate the loss between the predicted time series detection data and the true time series detection data.

[0016] Further, the training process of the Gaussian mixture model is as follows:

[0017] During the training process, the normalized training time series detection data is input into the trained bidirectional recurrent neural network to obtain the predicted training time series detection data;

[0018] Calculate the mean of the neighborhood interval differences between the true training time series detection data and the training time series detection data predicted by the bidirectional recurrent neural network;

[0019] Determine the number of Gaussian components K, and construct a Gaussian mixture model for the mean of the neighborhood interval differences of the training time series detection data:

[0020]

[0021] where, π j is the weight of the j-th Gaussian component, is the normal density with the mean of the neighborhood interval differences being test μ j and the variance being The Gaussian mixture model fits the distribution of the mean of the neighborhood interval differences through the superposition of multiple Gaussian distributions;

[0022] Use the EM algorithm to perform iterative adjustment training on the Gaussian mixture model.

[0023] Further, the calculation formula for the mean of the neighborhood interval differences between the true training time series detection data and the training time series detection data predicted by the bidirectional recurrent neural network is as follows:

[0024]

[0025] where, L is the half length of the neighborhood interval, test x i , are respectively the actual training time series detection data and the training time series detection data predicted by the bidirectional recurrent neural network within the neighborhood interval at time t.

[0026] Further, the process of using the EM algorithm to perform iterative adjustment training on the Gaussian mixture model is as follows:

[0027] Calculate the mean of the neighborhood interval differences of the training time series detection data test e t Posterior probability belonging to the k-th Gaussian component:

[0028]

[0029] In the formula, the numerator is the weight π of the k-th Gaussian component k Multiplied by the normal density generated by this Gaussian component test e t The denominator is the weighted sum of the normal densities generated by all Gaussian components test e t ;

[0030] Update the parameters with the goal of maximizing the log-likelihood of the complete data based on the current posterior probability γ tk :

[0031]

[0032] Take the derivative of Q with respect to test μ k , π k Set the derivative to zero, and we get:

[0033] Mean test μ k The update formula is as follows:

[0034]

[0035] Among them, the formula means to calculate the weighted average of the mean of the neighborhood interval differences of all training time series detection data with γ tk as the weight;

[0036] Variance The update formula is as follows:

[0037]

[0038] Among them, the formula means to calculate the weighted variance of the mean of the neighborhood interval differences of all training time series detection data with γ tk as the weight;

[0039] Weight π k The update formula is as follows:

[0040]

[0041] Among them, the formula means that the weight of the k-th Gaussian component is the average of the membership degrees of all neighborhood interval differences to it;

[0042] Stop adjusting the parameters of the Gaussian mixture model when the increase in log-likelihood is less than the set threshold.

[0043] Furthermore, the process of obtaining the measurement threshold includes: inputting the training time series detection data into the trained high-voltage bushing fault detection model, using a bidirectional recurrent neural network to predict the training time series detection data, and using the cumulative distribution function of the Gaussian mixture model in the high-voltage bushing fault detection model to calculate the extreme degree of the mean difference in the neighborhood interval between the predicted training time series detection data and the true training time series detection data under normal conditions, and taking the extreme degree as the measurement threshold.

[0044] Furthermore, the cumulative distribution function of the Gaussian mixture model for calculating the measurement threshold and the deviation parameter is:

[0045]

[0046] where is the cumulative distribution function of the standard normal distribution; are the predicted mean and standard deviation of the mean difference in the neighborhood interval respectively, and z taking test or r represents calculating the corresponding parameters according to the training time series detection data or the time series detection data;

[0047] Then the maximum value of 1 - F( test e t ) is the measurement threshold for the upper limit of the normal range, and the maximum value of F( test e t ) is the measurement threshold for the lower limit of the normal range;

[0048] The upper and lower limit deviation parameters are 1 - F( r e t ) and F( r e t ) respectively. Use 1 - F( r e t ) and the upper limit measurement threshold to measure whether the mean difference in the neighborhood interval is higher than the normal range, and use F( r e t ) of the fault detection and the lower limit measurement threshold to measure whether the mean difference in the neighborhood interval r e t is lower than the normal range, so as to determine whether the high-voltage bushing has a fault; determine the fault level according to the degree of deviation of the deviation parameter from the measurement threshold.

[0049] In a second aspect, the present invention provides an on-line monitoring device for high-voltage bushing fault classification and alarm, including: at least one processing unit, the processing unit is connected to a storage unit and a collection unit through a bus unit, the storage unit stores computer programs and the data collected by the collection unit, and when the computer program is executed by the processing unit, it realizes any of the on-line monitoring methods for high-voltage bushing fault classification and alarm.

[0050] In a third aspect, the present invention provides a computer-readable storage medium storing a computer program which, when executed by a processor, implements the high-voltage bushing fault grading alarm online monitoring method as described above.

[0051] The above technical solutions provided by the embodiments of the present invention have the following advantages compared with the prior art:

[0052] The bidirectional recurrent neural network of the high-voltage bushing fault detection model in this application uses a self-supervised method to perform time series modeling on the time series detection data under normal operating conditions. The trained bidirectional recurrent neural network can predict the time series detection data according to the normal operating conditions. And the Gaussian mixture model is used to model the distribution of the mean difference of the neighborhood intervals of the prediction results to determine the extreme degree of the mean difference of the neighborhood intervals under normal conditions, and this calculated degree under normal conditions is used as a measurement threshold to compare and analyze the deviation parameters in the actual detection process for high-voltage bushing fault monitoring. So that the high-voltage bushing fault detection model trained with normal operating data realizes fault monitoring. In this application, the training of the high-voltage bushing fault detection model adopts a self-supervised method and uses the training time series detection data under normal operating conditions, which can effectively model and avoid the problem of ineffective modeling in the case of sparse and scarce data. This application models the complex dependencies among the temperature, humidity, pressure and their differences of the high-voltage bushing fault detection model through the learning method of the high-voltage bushing fault detection model, and better realizes fault monitoring. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] The drawings here are incorporated into the description and form a part of this description, showing embodiments in line with the present invention and used together with the description to explain the principles of the present invention.

[0054] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, for those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0055] Figure 1 It is a flowchart of a high-voltage bushing fault grading alarm online monitoring method provided by an embodiment of the present invention;

[0056] Figure 2 It is a schematic diagram of the model architecture of the high-voltage bushing fault grading alarm online monitoring method provided by an embodiment of the present invention;

[0057] Figure 3 It is a schematic diagram of the acquisition unit provided by an embodiment of the present invention;

[0058] Figure 4 Schematic diagram of the on-line monitoring device for high-voltage bushing fault grading alarm provided by the embodiment of the present invention.

[0059] Reference numerals and meanings in the figure: 1. Combined valve; 2. Sensor assembly. Specific implementation manner

[0060] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0061] It should be noted that in this document, the terms "include", "comprise", or any other variant thereof are intended to cover a non-exclusive inclusion, such that a process, method, article, or device including a series of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or device. Without further limitation, an element defined by the phrase "including a..." does not exclude the presence of additional identical elements in the process, method, article, or device including the element.

[0062] Embodiment 1

[0063] As Figure 1 shown, the present invention realizes an on-line monitoring method for high-voltage bushing fault grading alarm, including:

[0064] Collect the oil pressure, temperature, and humidity of any-phase high-voltage bushing through a pressure sensor, a temperature sensor, and a humidity sensor. Obtain the time series of the oil pressure, temperature, and humidity of any-phase high-voltage bushing.

[0065] Obtain the time-series detection data for high-voltage bushing fault detection according to the time series of the oil pressure, temperature, and humidity of any-phase high-voltage bushing; wherein, the time-series detection data integrates the differences between the parameters of any two of the oil pressure, temperature, and humidity data of the three-phase high-voltage bushing.

[0066] The time-series detection data is expressed as:

[0067] where P is the time-series oil pressure parameter, P = [p1, p2,... p t ... p T , where the data at time t in the time-series oil pressure parameter is respectively the oil pressures of the three-phase high-voltage bushings at time t, They are the oil pressure differences between the first and second, the first and third, and the second and third phase high-voltage bushings at time t, respectively.

[0068] TEM is the time-series temperature parameter, TEM = [tem1, tem2,... tem t ... tem T , where the data at time t in the time-series temperature parameter is They are the temperatures of the three-phase high-voltage bushings at time t, respectively. They are the temperature differences between the first and second, the first and third, and the second and third phase high-voltage bushings at time t, respectively.

[0069] H is the time-series humidity parameter, H = [h1, h2,... h t ... h T , where the data at time t in the time-series humidity parameter is They are the humidities of the three-phase high-voltage bushings at time t, respectively. They are the humidity differences between the first and second, the first and third, and the second and third phase high-voltage bushings at time t, respectively.

[0070] Normalize the time-series oil pressure parameter, the time-series temperature parameter, and the time-series humidity parameter in the time-series detection data, respectively.

[0071] For the online monitoring method for high-voltage bushing fault classification and alarm of the present application, the problem to be solved is to predict the corresponding time-series high-voltage bushing faults and their grades Y = [y1, y2,... y t ... y T based on the time-series detection data X. To achieve the above functions, the online monitoring method for high-voltage bushing fault classification and alarm of the present application constructs a high-voltage bushing fault detection model.

[0072] In the specific implementation process, the high-voltage bushing fault detection model includes: a bidirectional recurrent neural network for performing time-series modeling on the time-series detection data under normal conditions and a Gaussian mixture model for modeling the mean of the neighborhood interval differences between the real time-series detection data and the time-series detection data predicted by the bidirectional recurrent neural network under normal conditions.

[0073] The bidirectional recurrent neural network is trained using the training time series detection data under normal operation of the high-voltage bushing. The content included in the training time series detection data is the same as that of the time series detection data. The training method of the bidirectional recurrent neural network is as follows: The training time series detection data is divided according to a set time window, and the time series detection data within any time window is input into the bidirectional recurrent neural network. The bidirectional recurrent time network predicts the training time series detection data of the history and the future set time within this time window. The training in the above training process uses the squared L2 norm to calculate the loss between the predicted time series detection data and the real time series detection data, aiming to train the recurrent neural network to model the dynamic evolution relationship of the oil pressure, temperature, humidity and the difference between any two parameters in the time series detection data under the normal state of the high-voltage bushing.

[0074] In the actual scenario, the proportion of the normal operation data of the high-voltage bushing is very large compared with the abnormal operation data, that is, the abnormal operation data is very sparse. If the abnormal data, faults and their levels are used for labeled training, there will be a problem of insufficient data volume, and it is difficult to accurately model the association between the abnormal data, faults and their levels. In the training process of this application for training the bidirectional recurrent neural network, the training time series detection data under the normal operation state is used, and no data labeling is involved in the training process, which belongs to an unsupervised training process and supports modeling through a large amount of time series detection data under normal operation. It can effectively model the relationship between various parameters in the time series detection data under normal conditions.

[0075] The training process of the Gaussian mixture model is as follows:

[0076] During the training process, the normalized training time series detection data is input into the trained bidirectional recurrent neural network to obtain the predicted training time series detection data. The mean value of the neighborhood interval difference is calculated for the real training time series detection data and the training time series detection data predicted by the bidirectional recurrent neural network. The formula is as follows:

[0077]

[0078] where L is the half length of the neighborhood interval, test x i , are the actual training time series detection data and the training time series detection data predicted by the bidirectional recurrent neural network within the neighborhood interval at time t, respectively.

[0079] Determine the number of Gaussian components K, and construct a Gaussian mixture model for the mean value of the neighborhood interval difference of the training time series detection data:

[0080]

[0081] where π j is the weight of the j-th Gaussian component, is the mean of the mean of the neighborhood interval differences, test μ j , and the variance is . The Gaussian mixture model fits the distribution of the mean of the neighborhood interval differences by superimposing multiple Gaussian distributions.

[0082] The EM algorithm is used for iterative adjustment of the Gaussian mixture model, and the process is as follows:

[0083] Calculate the mean of the neighborhood interval differences of the training time series detection data test e t The posterior probability belonging to the k-th Gaussian component:

[0084]

[0085] In the formula, the numerator is the weight π of the k-th Gaussian component k multiplied by the normal density generated by this Gaussian component test e t . The denominator in the formula is the weighted sum of the normal densities generated by all Gaussian components test e t . The posterior probability reflects test e t the degree of membership to the k-th Gaussian component.

[0086] Based on the current posterior probability γ tk , update the parameters to maximize the log-likelihood of the following complete data:

[0087]

[0088] Take the derivative of Q with respect to test μ k , π k respectively, and set the derivative to zero, we can get:

[0089] The mean test μ k is updated as follows:

[0090]

[0091] denotes the weighted average of the means of the neighborhood interval differences of all training time series detection data with γ tk as the weight;

[0092] The variance is updated as follows:

[0093]

[0094] denotes with γ tkTaking \(\pi\) as the weight, calculate the weighted variance of the mean of the neighborhood interval differences of all training time series detection data;

[0095] The weight \(\pi\) k The update formula is as follows:

[0096]

[0097] It means that the weight of the \(k\)-th Gaussian component is the average of the membership degrees of all neighborhood interval difference means to it.

[0098] When the increase in log-likelihood is less than the set threshold, stop adjusting the parameters of the Gaussian mixture model.

[0099] Similarly, the training process of the Gaussian mixture model is also an unsupervised training process. By modeling and predicting the normal deviation distribution of the mean of the neighborhood interval differences through unsupervised training, a measurement threshold is obtained. The process of obtaining the measurement threshold for measuring faults in the present application is as follows:

[0100] Input the training time series detection data into the trained high-voltage bushing fault detection model, use the bidirectional recurrent neural network to predict the training time series detection data, use the cumulative distribution function of the Gaussian mixture model in the high-voltage bushing fault detection model to calculate the extreme degree of the mean of the neighborhood interval differences between the predicted training time series detection data and the real training time series detection data under normal conditions, and take the said extreme degree as the measurement threshold.

[0101] In the specific implementation process, the cumulative distribution function of the Gaussian mixture model is:

[0102]

[0103] Where is the cumulative distribution function of the standard normal distribution; are respectively the predicted mean and standard deviation of the mean of the neighborhood interval differences of the training time series detection data.

[0104] Then the maximum value of \(1 - F( test e t ) is the measurement threshold for the upper limit of the normal range, and the maximum value of \(F( test e t ) is the measurement threshold for the lower limit of the normal range.

[0105] During fault detection, input the time series detection data into the high-voltage bushing fault detection model, use the bidirectional recurrent neural network to predict the time series detection data, and calculate the corresponding mean of the neighborhood interval differences; use the cumulative distribution function of the Gaussian mixture model to calculate the deviation parameter of the actual mean of the neighborhood interval differences.

[0106] The cumulative distribution function of the Gaussian mixture model for calculating the deviation parameter is:

[0107]

[0108] Among them, is the cumulative distribution function of the standard normal distribution; are respectively the mean and standard deviation of the prediction of the mean of the neighborhood interval difference of the time series detection data, r e t and test e t are calculated in the same way, the difference is that they are calculated using the actually collected time series detection data.

[0109] The upper and lower limit deviation parameters are 1 - F( r e t ) and F( r e t ), use 1 - F( r e t ) and the upper limit measurement threshold to measure whether the mean of the neighborhood interval difference is higher than the normal range, and use F( r e t ) of the fault detection and the lower limit measurement threshold to measure the mean of the neighborhood interval difference r e t whether it is lower than the normal range, so as to determine whether the high-voltage bushing fails; determine the fault level according to the degree of deviation of the deviation parameter from the measurement threshold. Determine the fault level according to the degree of deviation from the normal range: Specifically:

[0110] The degree of deviation of the upper limit deviation parameter from the upper limit measurement threshold is

[0111] The degree of deviation of the lower limit deviation parameter from the lower limit measurement threshold is

[0112] An exemplary fault degree level is divided into three levels in total, and the alarm level gradually becomes more serious; the first level is normal, indicated by a green indicator light, the second level is a minor defect, alarmed by a red indicator light, and the last level is a serious defect, alarmed by a buzzer.

[0113] Example 2

[0114] Refer to Figure 4As shown in the figure, an online monitoring device for high-voltage bushing fault grading and alarming provided by an embodiment of the present invention includes: at least one processing unit, the processing unit is connected to a storage unit and a collection unit through a bus unit, and the storage unit is used as a computer-readable storage medium and can be used to store software programs, computer-executable programs, and modules, such as software programs, computer-executable programs, and modules corresponding to a high-voltage bushing fault grading and alarming method in an embodiment of the present invention. The processing unit realizes the above-mentioned high-voltage bushing fault grading and alarming method by running the software programs, computer-executable programs, and modules stored in the storage unit, including:

[0115] Collect the oil pressure, temperature, and humidity of any-phase high-voltage bushing through a pressure sensor, a temperature sensor, and a humidity sensor, and obtain the time series of the oil pressure, temperature, and humidity of any-phase high-voltage bushing;

[0116] Obtain the time-series detection data for high-voltage bushing fault detection according to the time series of the oil pressure, temperature, and humidity of any-phase high-voltage bushing;

[0117] Normalize the time-series oil pressure parameter, time-series temperature parameter, and time-series humidity parameter in the time-series detection data respectively and input them into a pre-trained high-voltage bushing fault detection model; wherein, the high-voltage bushing fault detection model includes: a bidirectional recurrent neural network for time-series modeling of time-series detection data under normal conditions, and, under normal conditions, a Gaussian mixture model for modeling the mean of the neighborhood interval difference between the real time-series detection data and the time-series detection data predicted by the bidirectional recurrent neural network; use the bidirectional recurrent neural network to predict the time-series detection data, calculate the corresponding mean of the neighborhood interval difference with the real time-series detection data, and calculate the deviation parameter of the actual mean of the neighborhood interval difference using the cumulative distribution function of the Gaussian mixture model;

[0118] Determine whether the mean of the neighborhood interval difference is faulty according to the deviation parameter and the measurement threshold, and determine the fault level according to the degree of deviation of the deviation parameter from the measurement threshold; wherein, the measurement threshold is obtained through the high-voltage bushing fault detection model during the training process.

[0119] Certainly, the computer program stored in the storage unit of the online monitoring device for high-voltage bushing fault grading and alarming provided by an embodiment of the present invention is not limited to the method operations described above, and can also execute related operations in a high-voltage bushing fault grading and alarming method provided by any embodiment of the present invention.

[0120] Such as Figure 3As shown in the figure, the acquisition unit consists of a sensor assembly 2 formed by a temperature sensor, a humidity sensor, and a pressure sensor, and a combination valve 1. The combination valve is a three-way valve. One of the three interfaces of the combination valve is connected to the oil extraction port on the casing flange; one interface is connected to the sensor assembly 2, and the other interface can be used for oil extraction, sensor assembly calibration, and live oil replenishment operations. Among them, the pressure sensor uses a graphene pressure sensor. When the pressure is within 0 - 10 kPa, the highest sensitivity of the graphene pressure sensor can reach 15.6 kPa -1 , and the response speed is at the millisecond level. Its working response time is 130 ms, and the recovery response time is 160 ms. The self-resistance of the graphene material changes under the action of pressure.

[0121] In the specific implementation process, the high-voltage casing fault grading alarm online monitoring device further includes a prompting unit connected to the processing unit through a bus unit. The prompting unit includes warning lights and buzzers configured for different levels of situations.

[0122] Embodiment 3

[0123] The embodiment of the present invention provides a computer-readable storage medium. The computer-readable storage medium stores a computer program. When the computer program is executed, the high-voltage casing fault grading alarm online monitoring method is implemented, including:

[0124] Collect the oil pressure, temperature, and humidity of any phase of the high-voltage casing through a pressure sensor, a temperature sensor, and a humidity sensor, and obtain the time series of the oil pressure, temperature, and humidity of any phase of the high-voltage casing;

[0125] Obtain the timing detection data for high-voltage casing fault detection according to the time series of the oil pressure, temperature, and humidity of any phase of the high-voltage casing;

[0126] Normalize the timing oil pressure parameter, the timing temperature parameter, and the timing humidity parameter in the timing detection data respectively and input them into a pre-trained high-voltage casing fault detection model; among them, the high-voltage casing fault detection model includes: a bidirectional recurrent neural network for performing timing modeling on the timing detection data under normal conditions, and, under normal conditions, a Gaussian mixture model for modeling the mean of the neighborhood interval differences between the real timing detection data and the timing detection data predicted by the bidirectional recurrent neural network; use the bidirectional recurrent neural network to predict the timing detection data, calculate the corresponding neighborhood interval difference mean with the real timing detection data, and calculate the deviation parameter of the actual neighborhood interval difference mean using the cumulative distribution function of the Gaussian mixture model;

[0127] Determine whether there is a fault in the mean difference of the neighborhood interval according to the deviation parameter and the measurement threshold, and determine the fault level according to the degree of deviation of the deviation parameter from the measurement threshold; wherein, the measurement threshold is obtained through a high-voltage bushing fault detection model during the training process.

[0128] The computer-readable storage medium provided by the embodiment of the present invention stores computer programs that are not limited to the method operations described above, and can also execute related operations in a high-voltage bushing fault classification and alarm online monitoring method provided by any embodiment of the present invention.

[0129] In the embodiments provided by the present invention, it should be understood that the disclosed structure and method can be implemented in other ways. For example, the structural embodiments described above are merely illustrative. For example, the division of the units is only a logical function division. In actual implementation, there may be other division methods. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed couplings or direct couplings or communication connections between each other can be through some interfaces, and the indirect couplings or communication connections of structures or units can be in electrical, mechanical or other forms.

[0130] The units described as separate components may or may not be physically separated. The components displayed as units may or may not be physical units, that is, they can be located in one place, or they can be distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0131] In addition, each functional unit in various embodiments of the present invention can be integrated in a processing unit, or each unit exists physically alone, or two or more units can be integrated in one unit. The above-mentioned integrated units can be implemented in the form of hardware or in the form of software functional units.

[0132] The above are only specific embodiments of the present invention, which enable those skilled in the art to understand or implement the present invention. Various modifications to these embodiments will be obvious to those skilled in the art. The general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features claimed herein.

Claims

1. An online monitoring method for high-voltage bushing fault classification and alarm, characterized in that, Including: Collect the oil pressure, temperature, and humidity of any-phase high-voltage bushing through a pressure sensor, a temperature sensor, and a humidity sensor, and obtain the time series of the oil pressure, temperature, and humidity of any-phase high-voltage bushing; Obtain the time-series detection data for high-voltage bushing fault detection according to the time series of the oil pressure, temperature, and humidity of any-phase high-voltage bushing; After normalizing the time-series oil pressure parameter, time-series temperature parameter, and time-series humidity parameter in the time-series detection data respectively, input them into the pre-trained high-voltage bushing fault detection model; wherein, the high-voltage bushing fault detection model includes: a bidirectional recurrent neural network for performing time-series modeling on the time-series detection data under normal conditions, and, under normal conditions, a Gaussian mixture model for modeling the mean of the neighborhood interval difference between the real time-series detection data and the time-series detection data predicted by the bidirectional recurrent neural network; use the bidirectional recurrent neural network to predict the time-series detection data, calculate the corresponding mean of the neighborhood interval difference with the real time-series detection data, and calculate the deviation parameter of the actual mean of the neighborhood interval difference using the cumulative distribution function of the Gaussian mixture model; Determine whether the mean of the neighborhood interval difference is faulty according to the deviation parameter and the measurement threshold, and determine the fault level according to the degree of deviation of the deviation parameter from the measurement threshold; wherein, the measurement threshold is obtained through the high-voltage bushing fault detection model during the training process.

2. The online monitoring method for high-voltage bushing fault classification and alarm according to claim 1, characterized in that, The time-series detection data integrates the differences between any two parameters of the oil pressure, temperature, and humidity data of the three-phase high-voltage bushing, and is expressed as: The timing detection data is represented as: Among them, P is the sequential oil pressure parameter, P = [p1, p2,... p t ... p T , where the data at time t in the sequential oil pressure parameter is respectively the oil pressures of the three-phase high-voltage bushings at time t, respectively the oil pressure differences between the first and second, the first and third, and the second and third phase high-voltage bushings at time t; TEM is the timing temperature parameter, TEM = [tem1, tem2,... tem t ... tem T , where the data at time t in the timing temperature parameter is the temperatures of the three-phase high-voltage bushing at time t respectively, the temperature differences between the first and second, the first and third, and the second and third phases of the high-voltage bushing at time t respectively; H is the time-sequence humidity parameter, H = [h1, h2,... h t ... h T , where the data at time t in the time-sequence humidity parameter is the humidity of the three-phase high-voltage bushing at time t respectively, the humidity differences between the first and second, the first and third, and the second and third phase high-voltage bushings at time t respectively.

3. The online monitoring method for high-voltage bushing fault classification and alarm according to claim 1, characterized in that, The bidirectional recurrent neural network is trained using the training time-series detection data under normal operating conditions of the high-voltage bushing. The training method is: divide the training time-series detection data according to a set time window, input the time-series detection data within any time window into the bidirectional recurrent neural network, and the bidirectional recurrent time network predicts the training time-series detection data of the history and future set time within this time window; the training of the above training process uses the squared L2 norm to calculate the loss between the predicted time-series detection data and the real time-series detection data.

4. The on-line monitoring method for high-voltage bushing fault grading alarm according to claim 1, characterized in that, The training process of the Gaussian mixture model is as follows: During the training process, input the normalized training time-series detection data into the trained bidirectional recurrent neural network to obtain the predicted training time-series detection data; Calculate the mean of the neighborhood interval difference between the real training time-series detection data and the training time-series detection data predicted by the bidirectional recurrent neural network; Determine the number of Gaussian components K, and construct a Gaussian mixture model for the mean of the neighborhood interval difference of the training time-series detection data: where, π j is the weight of the j-th Gaussian component, is the mean of the mean of the neighborhood interval differences, test μ j and the variance is The normal density, and the Gaussian mixture model fits the distribution of the mean of the neighborhood interval differences through the superposition of multiple Gaussian distributions; Use the EM algorithm to perform iterative adjustment training on the Gaussian mixture model.

5. The on-line monitoring method for high-voltage bushing fault classification and alarm according to claim 4, characterized in that The formula for calculating the mean of the neighborhood interval difference for the real training time-series detection data and the training time-series detection data predicted by the bidirectional recurrent neural network is as follows: where L is the half length of the neighborhood interval, are the actual training time series detection data and the training time series detection data predicted by the bidirectional recurrent neural network within the neighborhood interval at time t, respectively.

6. The online monitoring method for high-voltage bushing fault classification and alarm according to claim 4, characterized in that, The process of using the EM algorithm to perform iterative adjustment training on the Gaussian mixture model is as follows: Calculate the mean of the neighborhood interval differences of the training time series detection data test e t Posterior probability belonging to the k-th Gaussian component: where the numerator is the weight π of the k-th Gaussian component k multiplied by the Gaussian component to generate test e t of the normal density, where the denominator is the weighted sum of the normal densities generated by all Gaussian components test e t ; Based on the current posterior probability γ tk , update the parameters with the goal of maximizing the log-likelihood of the following complete data: Differentiate Q with respect to test μ k and π k and set the derivative equal to zero, we get: Mean value test μ k The update formula is as follows: Among them, the formula represents the weighted average of the mean of the neighborhood interval differences of all training time series detection data calculated with γ tk as the weight; Variance The update formula is as follows: Among them, the formula represents that with γ tk as the weight, calculate the weighted variance of the mean of the neighborhood interval differences of all training time series detection data; Weight π k The update formula is as follows: Among them, the formula indicates that the weight of the k-th Gaussian component is the average of the membership degrees of all neighborhood interval differences to it; Stop adjusting the parameters of the Gaussian mixture model when the increase in log-likelihood is less than the set threshold.

7. The online monitoring method for high-voltage bushing fault classification alarm according to claim 1, wherein The process of obtaining the measurement threshold includes: inputting the training time-series detection data into the trained high-voltage bushing fault detection model, using a bidirectional recurrent neural network to predict the training time-series detection data, using the cumulative distribution function of the Gaussian mixture model in the high-voltage bushing fault detection model to calculate the extreme degree of the mean of the neighborhood interval difference between the predicted training time-series detection data and the true training time-series detection data under normal conditions, and using the extreme degree as the measurement threshold.

8. The online monitoring method for high-voltage bushing fault grading alarm according to claim 1, wherein, The cumulative distribution function of the Gaussian mixture model for calculating the measurement threshold and the deviation parameter is: wherein, is the cumulative distribution function of the standard normal distribution; are the mean and standard deviation of the prediction of the mean of the neighborhood interval differences respectively, and z taking test or r represents calculating the corresponding parameters according to the training time series detection data or the time series detection data. Then 1 - F( test e t ) has a maximum value as the measurement threshold for the upper limit of the normal range, and the maximum value of F( test e t ) is the measurement threshold for the lower limit of the normal range; The upper and lower limit deviation parameters are 1 - F( r e t ) and F( r e t ). Use 1 - F( r e t ) and the upper limit measurement threshold to measure whether the mean value of the neighborhood interval difference is higher than the normal range. Use the F( r e t ) of the fault detection and the lower limit measurement threshold to measure the mean value of the neighborhood interval difference r e t whether it is lower than the normal range, so as to determine that the high-voltage bushing has a fault; determine the fault level according to the degree of deviation of the deviation parameter from the measurement threshold.

9. An on-line monitoring device for high-voltage bushing fault classification alarm, characterized in that, including: At least one processing unit, the processing unit is connected to a storage unit and an acquisition unit through a bus unit, the storage unit stores computer programs and the data acquired by the acquisition unit, and when the computer program is executed by the processing unit, the on-line monitoring method for high-voltage bushing fault classification and alarm as described in any one of claims 1-8 is realized.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, the on-line monitoring method for high-voltage bushing fault classification and alarm as described in any one of claims 1-8 is realized.

Images