Construction method of PT fusing fault diagnosis model in plateau environment

Abstract

The invention relates to a construction method of a PT fusing fault diagnosis model in a plateau environment, and relates to the field of intelligent power distribution systems, and the method comprises the steps: building and training an electrical fault diagnosis sub-model, an environment fault diagnosis sub-model and a state fault diagnosis sub-model; obtaining structure information and operation environment information of the voltage transformer; generating a cascade strategy corresponding to the voltage transformer based on the structure information and the operation environment information of the voltage transformer; constructing a PT fusing fault diagnosis model based on a cascade strategy corresponding to the voltage transformer, the electrical fault diagnosis sub-model, the environment fault diagnosis sub-model and the state fault diagnosis sub-model; acquiring electrical parameters of the voltage transformer; acquiring environmental parameters of the voltage transformer; collecting the physical state of the voltage transformer; the PT fusing fault reason is determined through the PT fusing fault diagnosis model based on the electrical parameters, the environmental parameters and the physical state of the voltage transformer, and the method has the advantage of realizing accurate analysis of the PT fusing fault reason.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent power distribution system, and particularly relates to a method for constructing a PT (Potential Transformer) fuse failure diagnosis model in a plateau environment. BACKGROUND

[0002] A potential transformer (PT) is similar to a transformer and is used to transform voltage. However, the purpose of voltage transformation by a transformer is to facilitate power transmission, and therefore the capacity of the transformer is very large, generally calculated in kilovolt-amperes or megavolt-amperes. The purpose of voltage transformation by a potential transformer is mainly to supply power to measuring instruments and relay protection devices, to measure the voltage, power and energy of a line, or to protect valuable equipment, motors and transformers in the line in the event of a fault in the line. The basic structure of a potential transformer is similar to that of a transformer, and it also has two windings, one called a primary winding and one called a secondary winding. Both windings are mounted or wound on a core. There is insulation between the two windings and between the windings and the core, so that the two windings and the windings and the core are electrically isolated. When a potential transformer is in operation, the primary winding N1 is connected in parallel to a line, and the secondary winding N2 is connected in parallel to an instrument or a relay. Therefore, when measuring the voltage on a high-voltage line, although the primary voltage is very high, the secondary voltage is low, which can ensure the safety of operating personnel and instruments. There are many causes of PT fuse failure, including system causes, voltage transformer causes, fuse causes, operation failure causes and environmental causes.

[0003] At present, the means for analyzing the causes of a faulty potential transformer is to disassemble and observe the faulty potential transformer. If burn marks are observed on the primary winding of the potential transformer, it indicates that there is a problem of excessive transient excitation current, and it is believed that during the recovery period of a ground fault, due to the existence of low-frequency inrush current, the potential transformer is subjected to continuous saturation of magnetic flux under single-phase voltage, which is the cause of the fault, and the greater the ground capacitance, the more serious the fault. However, it is not possible to qualitatively determine the cause of the fuse failure, which affects the development of maintenance and operation strategies and leads to repeated faults.

[0004] Therefore, it is necessary to provide a method for constructing a PT fuse failure diagnosis model in a plateau environment, which can accurately analyze the causes of PT fuse failure. SUMMARY

[0005] The application provides a method for constructing a PT fuse failure diagnosis model in a plateau environment, comprising: establishing and training an electrical fault diagnosis submodel, an environmental fault diagnosis submodel, and a state fault diagnosis submodel; obtaining structure information and operating environment information of a voltage transformer; generating a corresponding cascade strategy of the voltage transformer based on the structure information and the operating environment information of the voltage transformer; constructing a PT fuse failure diagnosis model based on the corresponding cascade strategy of the voltage transformer, the electrical fault diagnosis submodel, the environmental fault diagnosis submodel, and the state fault diagnosis submodel; collecting electrical parameters of the voltage transformer, wherein the electrical parameters at least include primary side voltage, secondary side voltage, and secondary side current; collecting environmental parameters of the voltage transformer, wherein the environmental parameters at least include temperature, humidity, and air pressure; collecting a physical state of the voltage transformer, wherein the physical state at least includes vibration of multiple positions of the voltage transformer, discharge activity of multiple positions, and temperature of multiple positions of the fuse; determining a PT fuse failure cause based on the electrical parameters, the environmental parameters, and the physical state of the voltage transformer by the PT fuse failure diagnosis model.

[0006] Further, the electrical fault diagnosis submodel comprises a first input unit, a first feature extraction unit, and an electrical fault diagnosis unit; wherein the first input unit is used for inputting and preprocessing the primary side voltage sequence, the secondary side voltage sequence, and the secondary side current sequence; the first feature extraction unit comprises a first electrical feature extraction layer, a second electrical feature extraction layer, a third electrical feature extraction layer, and an electrical feature fusion layer, wherein the first electrical feature extraction unit is used for extracting time domain features, frequency domain features, and voltage harmonic features of the primary side voltage sequence, the second electrical feature extraction layer is used for extracting time domain features, frequency domain features, and voltage harmonic features of the secondary side voltage sequence, the third electrical feature extraction layer is used for extracting time domain features, frequency domain features, and current harmonic features of the secondary side current sequence, and the electrical feature fusion layer is used for fusing outputs of the first electrical feature extraction layer, the second electrical feature extraction layer, and the third electrical feature extraction layer to generate an electrical feature vector; and the electrical fault diagnosis unit is used for determining occurrence probabilities of multiple electrical faults based on the electrical feature vector.

[0007] Further, the environment fault diagnosis sub-model comprises a second input unit, a second feature extraction unit and an environment guided fault diagnosis unit; the second input unit is configured to input and preprocess temperature sequence, humidity sequence and air pressure sequence; the second feature extraction unit comprises an environment temperature feature extraction layer, a humidity feature extraction layer, an air pressure extraction layer and an environment feature fusion layer, the environment temperature feature extraction layer is configured to extract time domain features and frequency domain features of the temperature sequence, the humidity feature extraction layer is configured to extract time domain features and frequency domain features of the humidity sequence, the air pressure extraction layer is configured to extract time domain features and frequency domain features of the air pressure sequence, and the environment feature fusion layer is configured to fuse outputs of the environment temperature feature extraction layer, the humidity feature extraction layer and the air pressure extraction layer to generate an environment feature vector; the environment guided fault diagnosis unit is configured to determine occurrence probabilities of multiple environment guided faults based on the environment feature vector.

[0008] Further, the state fault diagnosis sub-model comprises a third input unit, a third feature extraction unit and a state fault diagnosis unit; the third input unit is configured to input and preprocess vibration sequence of multiple positions of the voltage transformer, discharge activity sequence of multiple positions of the fuse and temperature sequence of multiple positions of the fuse; the third feature extraction unit comprises a vibration feature extraction layer, a discharge activity extraction layer, a fuse temperature feature extraction layer and a physical state feature fusion layer, the vibration feature extraction layer is configured to extract and fuse time domain features and frequency domain features of the vibration sequence of each position, the discharge activity extraction layer is configured to extract and fuse time domain features and frequency domain features of the discharge activity sequence of each position, the fuse temperature feature extraction layer is configured to extract and fuse time domain features and frequency domain features of the temperature sequence of each position, and the physical state feature fusion layer is configured to fuse outputs of the vibration feature extraction layer, the discharge activity extraction layer and the fuse temperature feature extraction layer to generate a physical state feature vector; the state fault diagnosis unit is configured to determine occurrence probabilities of multiple physical faults based on the physical state feature vector.

[0009] Further, based on the structure information and the operation environment information of the voltage transformer, a cascade strategy corresponding to the voltage transformer is generated, comprising: establishing a sample database, wherein the sample database comprises cascade strategies of multiple sample voltage transformers in multiple sample operation environments; determining a target sample voltage transformer based on the structure information of the voltage transformer and the sample database; determining a target sample operation environment based on the operation environment information of the voltage transformer and the multiple sample operation environments; and generating the cascade strategy corresponding to the voltage transformer based on the cascade strategy of the target sample voltage transformer in the target sample operation environment.

[0010] Further, the cascade strategy of each sample voltage transformer in each sample operation environment is determined, including: obtaining fault information of the sample voltage transformer in a plurality of historical time periods, wherein the fault information includes occurrence times of electrical faults, environment-induced faults and physical faults; determining weights of the electrical faults, the environment-induced faults and the physical faults based on the fault information of the sample voltage transformer in the plurality of historical time periods; determining correlation coefficients of any two of the electrical faults, the environment-induced faults and the physical faults based on the fault information of the sample voltage transformer in the plurality of historical time periods; and determining the cascade strategy of the sample voltage transformer in the sample operation environment based on the weights of the electrical faults, the environment-induced faults and the physical faults and the correlation coefficients of any two of the electrical faults, the environment-induced faults and the physical faults.

[0011] Further, the cascade strategy of each sample voltage transformer in each sample operation environment is determined based on the weights of the electrical faults, the environment-induced faults and the physical faults and the correlation coefficients of any two of the electrical faults, the environment-induced faults and the physical faults, including: calculating a comprehensive correlation coefficient of the electrical faults, the environment-induced faults and the physical faults based on the correlation coefficients of any two of the electrical faults, the environment-induced faults and the physical faults; calculating a determination priority of the electrical faults, the environment-induced faults and the physical faults based on the weights of the electrical faults, the environment-induced faults and the physical faults and the comprehensive correlation coefficient of the electrical faults, the environment-induced faults and the physical faults; determining a first-level fault type based on the determination priority of the electrical faults, the environment-induced faults and the physical faults; and determining the cascade strategy of the sample voltage transformer in the sample operation environment based on the correlation coefficients of any two of the electrical faults, the environment-induced faults and the physical faults and the first-level fault type.

[0012] Further, a three-dimensional model of the voltage transformer is established; a finite element model of the voltage transformer is established based on operation environment information of the voltage transformer and the three-dimensional model of the voltage transformer, finite element analysis is performed based on the finite element model of the voltage transformer, electromagnetic force distribution and mechanical stress distribution inside the voltage transformer are determined, and a plurality of positions for vibration monitoring are determined according to the electromagnetic force distribution and the mechanical stress distribution inside the voltage transformer; and vibration sequences of the plurality of positions of the voltage transformer are collected according to the plurality of positions for vibration monitoring.

[0013] Further, the discharge activity sequences of the plurality of positions are collected, including: the electromagnetic force distribution inside the voltage transformer is determined based on the finite element analysis performed based on the finite element model of the voltage transformer, a plurality of positions for discharge activity monitoring are determined according to the electromagnetic force distribution inside the voltage transformer; and the discharge activity sequences of the plurality of positions are collected according to the plurality of positions for discharge activity monitoring.

[0014] Further, the temperature sequence of the plurality of positions of the fuse is collected, including: performing finite element analysis based on the finite element model of the voltage transformer to determine the temperature distribution inside the fuse, and determining the plurality of positions for temperature monitoring according to the temperature distribution inside the fuse; and collecting the temperature sequence of the plurality of positions of the fuse according to the plurality of positions for temperature monitoring.

[0015] Compared with the prior art, the method for constructing a PT fuse fault diagnosis model in a highland environment provided by the present specification has at least the following beneficial effects:

[0016] 1. Electrical parameters (voltage, current) reflect the running state of the equipment, environmental parameters (temperature, humidity, air pressure) reveal external influencing factors, and physical state (vibration, discharge, temperature) reveals internal damage mechanism. If only electrical parameters are monitored, the fuse may be misjudged as overload; combined with environmental parameters (low air pressure leading to a decrease in insulation strength) and physical state (enhanced partial discharge), the insulation breakdown caused by the environment can be accurately diagnosed.

[0017] 2. The corresponding cascade strategy is generated based on the structure information and running environment information of the voltage transformer, and the cascade strategy is combined with a plurality of sub-models to construct a PT fuse fault diagnosis model. The application of this cascade strategy can flexibly adjust the diagnosis mode of the model according to the characteristics and running environment of different voltage transformers, improving the adaptability and accuracy of the model. The traditional method needs to manually check the electrical, environmental and physical state one by one, which takes several hours; the model can complete the diagnosis within a few minutes. In the actual fault diagnosis scene, if all the sub-models corresponding to the fault types are run and calculated, a large amount of time and computing resources will be consumed. The scheme first runs the sub-model corresponding to the first level of fault type, and only when the probability of each fault output by it is less than the probability threshold, the sub-models corresponding to the second and third levels of fault type are run in turn, thereby saving a large amount of computing time and resources.

[0018] 3. The electrical fault diagnosis sub-model extracts and fuses the characteristics of the primary side and secondary side voltage and current respectively, avoids information loss, can diagnose multiple electrical faults (such as short circuit, overload, resonance, insulation breakdown, etc.) at the same time, and significantly improves the diagnosis accuracy of PT fuse electrical faults in highland environment through multi-dimensional feature extraction, hierarchical fusion and probability output.

[0019] 4. Traditional methods may randomly arrange sensors, but this method can accurately locate high-stress areas (such as iron cores and winding connections), key locations for discharge activity monitoring, and key locations for temperature monitoring, thereby improving monitoring efficiency. The large temperature difference between day and night on the plateau may cause thermal expansion and contraction of materials, triggering changes in mechanical stress. Accurate monitoring can capture such precursors to faults. Through finite element analysis of the electric field intensity distribution, the key locations for monitoring discharge activity are determined. High electric field areas (such as winding ends and weak insulation points) are more prone to local discharge, and accurate monitoring can improve the fault detection rate. Under low pressure, the heat dissipation efficiency is reduced, and the fuse is more likely to overheat. The temperature series can reflect the current-carrying capacity and contact status of the fuse. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] This specification will be further described in the form of exemplary embodiments, which will be described in detail with reference to the accompanying drawings. These embodiments are not limiting, and in these embodiments, like numbers represent like structures, wherein:

[0021] Figure 1 This is a flow chart of a method for constructing a PT fuse fault diagnosis model in a plateau environment, shown in one embodiment of the present application;

[0022] Figure 2 is a structural diagram of an electrical fault diagnosis sub-model shown in an embodiment of the present application;

[0023] Figure 3 is a structural diagram of an environmental fault diagnosis sub-model shown in an embodiment of the present application;

[0024] Figure 4 It is a structural diagram of the state fault diagnosis sub-model shown in one embodiment of the present application. DETAILED DESCRIPTION

[0025] In order to more clearly illustrate the technical solutions of the embodiments of this specification, the following briefly introduces the drawings required for describing the embodiments.

[0026] Figure 1 This is a flow chart of a method for constructing a PT fuse fault diagnosis model in a plateau environment shown in an embodiment of the present application. This method belongs to an intelligent power distribution system, such as Figure 1 As shown, a method for constructing a PT fuse fault diagnosis model in a plateau environment may include the following steps.

[0027] Step 110 : Establish and train an electrical fault diagnosis sub-model, an environmental fault diagnosis sub-model, and a state fault diagnosis sub-model.

[0028] Figure 2 This is a structural diagram of the electrical fault diagnosis sub-model shown in an embodiment of the present application, such as Figure 2As shown, in some embodiments, the electrical fault diagnosis sub-model includes a first input unit, a first feature extraction unit, and an electrical fault diagnosis unit.

[0029] The first input unit is configured to input and preprocess the primary side voltage sequence, the secondary side voltage sequence, and the secondary side current sequence.

[0030] The first feature extraction unit includes a first electrical feature extraction layer, a second electrical feature extraction layer, a third electrical feature extraction layer, and an electrical feature fusion layer. The first electrical feature extraction layer is configured to extract time domain features, frequency domain features, and voltage harmonic features of the primary side voltage sequence. The second electrical feature extraction layer is configured to extract time domain features, frequency domain features, and voltage harmonic features of the secondary side voltage sequence. The third electrical feature extraction layer is configured to extract time domain features, frequency domain features, and current harmonic features of the secondary side current sequence. The electrical feature fusion layer is configured to fuse the outputs of the first electrical feature extraction layer, the second electrical feature extraction layer, and the third electrical feature extraction layer to generate an electrical feature vector.

[0031] The electrical fault diagnosis unit is configured to determine the occurrence probability of a plurality of electrical faults (e.g., single-phase ground fault, secondary side short circuit, operating overvoltage, etc.) based on the electrical feature vector.

[0032] Specifically, the first input unit serves as the data inlet of the entire electrical fault diagnosis sub-model, responsible for receiving and processing electrical parameter sequences related to the voltage transformer, providing appropriate data formats for subsequent feature extraction and fault diagnosis. The first input unit receives the primary side voltage sequence, the secondary side voltage sequence, and the secondary side current sequence. These sequences are time-ordered electrical parameter values that can reflect the operating state of the voltage transformer at different time points. Since the actual collected electrical parameter sequences may be disturbed by various noises such as electromagnetic interference and measurement errors, a filtering algorithm (such as mean filtering or median filtering) is used to remove noise and improve data quality during preprocessing. The dimensions and numerical ranges of different electrical parameters differ greatly, and in order to eliminate the influence of such differences on subsequent model training, sequence data needs to be normalized. For example, the values of each sequence are mapped to the interval [0, 1] or [-1, 1], so that the data has the same scale. The primary side voltage sequence, the secondary side voltage sequence, and the secondary side current sequence are aligned to ensure that they are synchronized in time, i.e., the data corresponding to the same time point is collected simultaneously, so that the relationship between them can be accurately analyzed subsequently.

[0033] Significant features are extracted from the preprocessed electrical parameter sequences, which can represent the operating state and potential fault information of the voltage transformer. By extracting features from the primary voltage, secondary voltage, and secondary current sequences respectively, and fusing these features, a comprehensive electrical feature vector is generated, providing more abundant information for fault diagnosis.

[0034] The time-domain features of the primary voltage sequence can include the mean, variance, and peak value, etc. The first electrical feature extraction layer calculates the average value of the primary voltage sequence within a certain period of time, reflecting the overall level of the voltage. If the average value is abnormal, it may imply that there is a voltage offset problem in the primary side. The variance is used to measure the dispersion degree of the primary voltage sequence, the larger the variance, the more intense the voltage fluctuation. For example, if the primary voltage variance suddenly increases, it may indicate that the primary side is disturbed or has a fault. The peak value is the maximum value of the primary voltage sequence, and the change of the peak value can reflect the transient characteristics of the voltage. For example, if the primary voltage peak value abnormally increases, it may mean that the primary side has an overvoltage situation.

[0035] The first electrical feature extraction layer converts the time-domain sequence of the primary voltage into a frequency-domain sequence, obtaining the frequency spectrum of the voltage. By analyzing the frequency spectrum, the amplitude and phase information of different frequency components can be observed, and the main frequency components and existing harmonics in the voltage can be identified. The content of each harmonic in the primary voltage is calculated, and too high harmonic content may lead to voltage waveform distortion.

[0036] The second electrical feature extraction layer and the third electrical feature extraction layer have similar structures to the first electrical feature extraction layer, which will not be described here.

[0037] The outputs of the first electrical feature extraction layer, the second electrical feature extraction layer, and the third electrical feature extraction layer are spliced to generate a comprehensive electrical feature vector. Through feature fusion, the mutual relationship between different electrical parameters can be considered comprehensively, improving the accuracy and reliability of fault diagnosis. For example, there may be some correlation between the harmonic features of the primary voltage and the harmonic features of the secondary current, and the fused feature vector can better reflect this correlation, which helps to more accurately determine the fault type.

[0038] The electrical fault diagnosis unit can include a multi-layer perceptron neural network, and a large number of historical electrical parameter sequences and corresponding fault label data are used to train the model. During the training process, by adjusting the parameters of the electrical fault diagnosis unit (such as the weights and biases of the neural network), the electrical fault diagnosis unit can accurately learn the relationship between electrical features and fault types. Cross-validation, regularization, and other methods are used to optimize the electrical fault diagnosis unit to prevent overfitting and improve the generalization ability of the electrical fault diagnosis unit. The trained electrical fault diagnosis unit can output the occurrence probability of multiple electrical faults according to the input electrical feature vector. For example, it outputs "the occurrence probability of primary side short circuit fault is 70%, the occurrence probability of secondary side short circuit fault is 20%, and the occurrence probability of ferroresonance is 10%".

[0039] The electrical fault diagnosis sub-model training process can include:

[0040] Initialize model parameters: Set initial parameters for the electrical fault diagnosis sub-model, such as the weights and biases of the neural network.

[0041] Forward propagation: Input the training data into the electrical fault diagnosis sub-model, and output the probability of each fault type.

[0042] Loss calculation: Calculate the value of the loss function according to the fault probability output by the electrical fault diagnosis sub-model and the actual fault label, and the loss function includes a cross-entropy loss function.

[0043] Backpropagation and parameter update: Update the parameters of the electrical fault diagnosis sub-model according to the gradient information of the loss function, so that the performance of the electrical fault diagnosis sub-model on the training data is continuously optimized.

[0044] Iterative training: Repeat the process of forward propagation, loss calculation, and parameter update until the electrical fault diagnosis sub-model converges or reaches the preset number of training rounds.

[0045] Figure 3 is the structural diagram of the environmental fault diagnosis sub-model shown in an embodiment of the present application, as shown in some embodiments, the environmental fault diagnosis sub-model includes a second input unit, a second feature extraction unit, and an environmental guide fault diagnosis unit; Figure 3

[0046] Among them, the second input unit is used to input and preprocess the temperature sequence, the humidity sequence, and the air pressure sequence;

[0047] ​The second feature extraction unit includes an ambient temperature feature extraction layer, a humidity feature extraction layer, an air pressure extraction layer, and an ambient feature fusion layer. The ambient temperature feature extraction layer is used to extract the time domain features and frequency domain features of the temperature sequence. The humidity feature extraction layer is used to extract the time domain features and frequency domain features of the humidity sequence. The air pressure extraction layer is used to extract the time domain features and frequency domain features of the air pressure sequence. The ambient feature fusion layer is used to fuse the outputs of the ambient temperature feature extraction layer, the humidity feature extraction layer, and the air pressure extraction layer to generate an ambient feature vector.

[0048] The environment-induced fault diagnosis unit is used to determine the occurrence probability of various environment-induced faults (for example, humid environment causing degradation of PT insulation performance, triggering surface discharge or flashover, environment-induced ferromagnetic resonance, etc.) based on the environment feature vector.

[0049] The structure and principle of the environmental fault diagnosis sub-model are similar to those of the electrical fault diagnosis sub-model and will not be described in detail here.

[0050] Figure 4 This is a structural diagram of the state fault diagnosis sub-model shown in an embodiment of the present application, such as Figure 4 As shown, in some embodiments, the state fault diagnosis sub-model includes a third input unit, a third feature extraction unit, and a state fault diagnosis unit;

[0051] The third input unit is used to input vibration sequences of multiple positions of the voltage transformer, discharge activity sequences of multiple positions of the fuse, and temperature sequences of multiple positions of the fuse and perform preprocessing;

[0052] The third feature extraction unit includes a vibration feature extraction layer, a discharge activity extraction layer, a fuse temperature feature extraction layer and a physical state feature fusion layer. The vibration feature extraction layer is used to extract the time domain features and frequency domain features of the vibration sequence of each position and fuse them. The discharge activity extraction layer is used to extract the time domain features and frequency domain features of the discharge activity sequence of each position and fuse them. The fuse temperature feature extraction layer is used to extract the time domain features and frequency domain features of the temperature sequence of each position and fuse them. The physical state feature fusion layer is used to fuse the outputs of the vibration feature extraction layer, the discharge activity extraction layer and the fuse temperature feature extraction layer to generate a physical state feature vector.

[0053] The state fault diagnosis unit is used to determine the probability of occurrence of various physical faults (such as winding deformation, insulation damage, contact / terminal overheating, etc.) based on the physical state feature vector.

[0054] The structure and principle of the state fault diagnosis sub-model are similar to those of the electrical fault diagnosis sub-model and will not be described in detail here.

[0055] Step 120: Acquire structural information and operating environment information of the voltage transformer.

[0056] Specifically, the structure information of the voltage transformer can be obtained from design drawings, technical specifications, installation manuals, and other documents of the voltage transformer. These documents record detailed information about the overall structure of the voltage transformer, the size, material, connection method, and electrical parameters of each component, etc. For example, the design drawings will mark the number of turns of the winding, the shape and material of the core, the insulation structure, and other key information.

[0057] An environmental monitoring device can be installed near the installation location of the voltage transformer to monitor environmental parameters such as temperature, humidity, and air pressure in real time. For example, temperature and humidity sensors are used to monitor changes in environmental temperature and humidity, and a barometer is used to measure air pressure.

[0058] Step 130, based on the structure information and operating environment information of the voltage transformer, generating the corresponding cascade strategy of the voltage transformer.

[0059] Specifically, it includes:

[0060] Establish a sample database, wherein the sample database includes cascade strategies of a plurality of sample voltage transformers in a plurality of sample operating environments;

[0061] Based on the structure information of the voltage transformer and the sample database, determine the target sample voltage transformer;

[0062] Based on the operating environment information of the voltage transformer and the plurality of sample operating environments, determine the target sample operating environment;

[0063] Based on the cascade strategy of the target sample voltage transformer in the target sample operating environment, generate the corresponding cascade strategy of the voltage transformer.

[0064] Specifically, the plurality of sample voltage transformers can be voltage transformers of different manufacturers and different models. Considering a plurality of typical operating environments, including different temperature ranges, humidity conditions, altitudes, and electromagnetic interference strengths, as sample operating environments.

[0065] The cascade strategy of the sample operating environment can reflect the execution logic of fault determination based on the electrical fault diagnosis sub-model, the environmental fault diagnosis sub-model, and the state fault diagnosis sub-model.

[0066] In some embodiments, determining the cascade strategy of each sample voltage transformer in each sample operating environment includes:

[0067] Obtain the fault information of the sample voltage transformer in multiple historical time periods, wherein the fault information includes the occurrence times of electrical faults, environment-induced faults and physical faults. Specifically, for each historical time period, the occurrence times of each type of electrical fault in the historical time period can be counted, and the sum of the occurrence times of each type of electrical fault is taken as the occurrence times of electrical faults. The determination of the occurrence times of environment-induced faults and physical faults is similar and will not be repeated here.

[0068] Based on the fault information of the sample voltage transformer in multiple historical time periods, the weights of electrical faults, environment-induced faults and physical faults are determined.

[0069] Based on the fault information of the sample voltage transformer in multiple historical time periods, the correlation coefficients of any two of electrical faults, environment-induced faults and physical faults are determined. Specifically, the correlation coefficients of any two of electrical faults, environment-induced faults and physical faults can be determined based on the occurrence times of electrical faults, environment-induced faults and physical faults of the sample voltage transformer in multiple historical time periods according to the correlation coefficient (for example, Spearman rank correlation coefficient) calculation formula, wherein the value range of the correlation coefficient is [-1, 1].

[0070] Based on the weights of electrical faults, environment-induced faults and physical faults and the correlation coefficients of any two of electrical faults, environment-induced faults and physical faults, the cascade strategy of the sample voltage transformer in the sample operating environment is determined.

[0071] Specifically, the weights of electrical faults, environment-induced faults and physical faults can be determined according to the following formula:

[0072]

[0073] wherein w i is the weight of the i-th type of fault, N (i,t) is the occurrence times of the i-th type of fault in the t-th historical time period, n (j,t) is the occurrence times of the j-th type of fault in the t-th historical time period, and T is the total number of historical time periods.

[0074] It can be understood that the total occurrence times of the i-th type of fault are divided by the total occurrence times of the three types of faults to obtain the weight w i of the i-th type of fault. This weight reflects the relative importance of the i-th type of fault in the entire fault system. The greater the weight, the higher the frequency of occurrence of this type of fault in the entire historical time period, and the greater the possibility of causing PT fusing.

[0075] In some embodiments, based on the weights of the electrical fault, the environment-induced fault and the physical fault and the correlation coefficients of any two of the electrical fault, the environment-induced fault and the physical fault, a cascading strategy of the sample voltage transformer in the sample operating environment is determined, including:

[0076] Based on the correlation coefficients of any two of the electrical fault, the environment-induced fault and the physical fault, a comprehensive correlation coefficient of the electrical fault, the environment-induced fault and the physical fault is calculated, for example, for the electrical fault, the absolute values of the correlation coefficient of the electrical fault and the environment-induced fault and the correlation coefficient of the electrical fault and the physical fault are averaged as the comprehensive correlation coefficient of the electrical fault, and the comprehensive correlation coefficients of the environment-induced fault and the physical fault are calculated in a similar manner, which will not be described here;

[0077] Based on the weights of the electrical fault, the environment-induced fault and the physical fault and the comprehensive correlation coefficients of the electrical fault, the environment-induced fault and the physical fault, a determination priority value of the electrical fault, the environment-induced fault and the physical fault is calculated.

[0078] Based on the determination priority values of the electrical fault, the environment-induced fault and the physical fault, a first-level fault type is determined.

[0079] Based on the correlation coefficients of any two of the electrical fault, the environment-induced fault and the physical fault and the first-level fault type, a cascading strategy of the sample voltage transformer in the sample operating environment is determined.

[0080] Specifically, the determination priority value can be calculated according to the following formula:

[0081] P i =w i ×r i

[0082] Wherein, p i is the determination priority value of the i-th fault, and r i is the comprehensive correlation coefficient of the i-th fault.

[0083] It can be understood that w i in the formula is the weight of the i-th fault, which is calculated based on the fault information of the sample voltage transformer in multiple historical time periods, and reflects the relative importance of this type of fault in the entire fault system. The greater the weight, the higher the frequency of this type of fault in historical operation, and the greater the possible impact on the operation of the voltage transformer, which needs more attention in fault diagnosis. r iis the comprehensive correlation coefficient of the i-th fault, which is obtained by calculating the average of the absolute values of the correlation coefficients between the i-th fault and the other two faults, and reflects the degree of association between the i-th fault and the other two faults. The greater the comprehensive correlation coefficient is, the closer the association between the i-th fault and the other two faults is, and the i-th fault and the other two faults may influence each other or appear together when the faults occur. The product of the weight and the comprehensive correlation coefficient is the determination priority, which comprehensively considers the importance and association of the faults. The determination priority calculated in this way can more comprehensively reflect the priority of each fault in fault diagnosis, and avoid the diagnosis deviation caused by considering only a single factor.

[0084] The fault with the maximum determination priority can be taken as the first-level fault type.

[0085] The second-level fault type can be determined according to the correlation coefficients of the remaining fault types and the first-level fault type, for example, the fault type with a larger correlation coefficient with the first-level fault type is taken as the second-level fault type, and the remaining fault types are taken as the third-level fault types. The cascade strategy can be: when performing fault diagnosis, the sub-model corresponding to the first-level fault type is first run, when the probabilities of each fault output by the sub-model corresponding to the first-level fault type are all less than the probability threshold, the sub-model corresponding to the second-level fault type is run, and when the probabilities of each fault output by the sub-model corresponding to the second-level fault type are all less than the probability threshold, the sub-model corresponding to the third-level fault type is run. The probability threshold can be set through experience or experimental data, for example, the probability threshold is set to 30%.

[0086] In actual fault diagnosis scenarios, if the sub-models corresponding to all fault types are run and calculated, a large amount of time and computing resources will be consumed. The scheme first runs the sub-model corresponding to the first-level fault type, and only when the probabilities of each fault output by the sub-model are all less than the probability threshold, the sub-models corresponding to the second-level and third-level fault types are sequentially run, thereby saving a large amount of computing time and resources.

[0087] The sample database can also include structural information of various sample voltage transformers, such as winding type (for example, single winding, double winding, or multi-winding, etc.), insulation medium (for example, oil-immersed, dry type, SF6 gas insulation, etc.), rated voltage of primary side and secondary side, current value, structural form (for example, electromagnetic type, capacitive type, photoelectric type, etc.). The structural information of the sample voltage transformer is represented as a structural feature vector, which facilitates subsequent similarity calculation. For example, the winding type can be represented by numerical coding (such as single winding = 1, double winding = 2, and multi-winding = 3), and the insulation medium type can be represented by numerical coding (such as oil-immersed = 1, dry type = 2, and SF6 gas insulation = 3). The coding values are spliced to generate the structural feature vector of the sample voltage transformer. Similarly, the structural information of the voltage transformer is represented as a structural feature vector. The Euclidean distance between the structural feature vector of the current voltage transformer and the structural feature vector of the sample voltage transformer is calculated to obtain the structural feature vector similarity. The sample voltage transformer with the maximum structural feature vector similarity is taken as the target sample voltage transformer.

[0088] An environmental feature vector of a sample operating environment can be generated, which can include specific values of temperature, humidity, and air pressure. Similarly, an environmental feature vector of the operating environment information of the current voltage transformer can be generated. The Euclidean distance between the environmental feature vector of the sample operating environment and the environmental feature vector of the current voltage transformer can be calculated to obtain the environmental feature vector similarity. The sample operating environment with the maximum environmental feature vector similarity is taken as the target sample operating environment.

[0089] The cascading strategy of the target sample voltage transformer in the target sample operating environment can be taken as the corresponding cascading strategy of the voltage transformer.

[0090] In step 140, a PT fuse fault diagnosis model is constructed based on the corresponding cascading strategy of the voltage transformer, the electrical fault diagnosis sub-model, the environmental fault diagnosis sub-model, and the state fault diagnosis sub-model.

[0091] In step 150, electrical parameters of the voltage transformer are collected.

[0092] The electrical parameters at least include primary side voltage, secondary side voltage, and secondary side current.

[0093] In step 160, environmental parameters of the voltage transformer are collected.

[0094] The environmental parameters at least include temperature, humidity, and air pressure.

[0095] In step 170, physical states of the voltage transformer are collected.

[0096] The physical states at least include vibration of multiple positions of the voltage transformer, discharge activity of multiple positions, and temperature of multiple positions of the fuse.

[0097] In some embodiments, a vibration sequence of a plurality of positions of the voltage transformer is collected, including:

[0098] A three-dimensional model of the voltage transformer is established;

[0099] Based on the operating environment information of the voltage transformer and the three-dimensional model of the voltage transformer, a finite element model of the voltage transformer is established, finite element analysis is performed based on the finite element model of the voltage transformer, the electromagnetic force distribution and the mechanical stress distribution inside the voltage transformer are determined, and the plurality of positions for vibration monitoring are determined according to the electromagnetic force distribution and the mechanical stress distribution inside the voltage transformer;

[0100] According to the plurality of positions for vibration monitoring, a vibration sequence of a plurality of positions of the voltage transformer is collected.

[0101] Specifically, using CAD software, a detailed three-dimensional model is created according to the actual size and structure of the voltage transformer. The three-dimensional model should include all key components, such as windings, cores, insulation structures, housings, connecting parts, etc. Each component in the model is given correct material properties, including density, elastic modulus, Poisson's ratio, magnetic permeability, dielectric constant, electrical conductivity, breakdown field strength, thermal conductivity, etc., to ensure that the physical behavior of the model is consistent with the actual situation. The three-dimensional model is divided into a finite number of small units (meshes), and the density and type of the meshes should be optimized according to the analysis requirements. According to the actual operating current and voltage of the voltage transformer, the corresponding electromagnetic excitation is applied, the magnetic field distribution inside the voltage transformer is calculated through electromagnetic field analysis, and then the electromagnetic force is calculated. According to the operating environment information of the voltage transformer, thermal stress, pressure load and boundary conditions are applied. Use finite element software (such as ANSYS Maxwell, COMSOL Multiphysics, etc.) to perform electromagnetic field analysis and calculate the electromagnetic force distribution inside the voltage transformer. The electromagnetic force is applied as a load to the structural model for structural mechanics analysis to calculate the mechanical stress distribution and deformation. Electromagnetic force is one of the main excitation sources causing voltage transformer vibration. Through electromagnetic field analysis, the area with concentrated electromagnetic force is determined as the plurality of positions for vibration monitoring. For example, the end of the winding: the end of the magnetic field is larger, and the electromagnetic force is obvious. The core joint: the magnetic flux density is uneven, which may cause local electromagnetic force to increase. The gap between the winding and the core: the electromagnetic force may cause the relative movement between the winding and the core. The area with large mechanical stress is prone to deformation or fatigue, thereby causing vibration. Through structural mechanics analysis, the area with concentrated mechanical stress is determined as the plurality of positions for vibration monitoring. For example, the winding support structure: the support part may bear a large mechanical stress. The core fixing point: the connection between the core and the shell may have stress concentration. Insulation structure: the insulation material may deform slightly under the action of mechanical stress, affecting the vibration characteristics.

[0102] A sensor can be set at each position for vibration monitoring to collect vibration sequences of the positions.

[0103] In some embodiments, sequences of discharge activities of multiple positions are collected, including:

[0104] Finite element analysis is performed based on the finite element model of the voltage transformer to determine the distribution of electric field intensity inside the voltage transformer, and multiple positions for monitoring discharge activities are determined according to the distribution of electric field intensity inside the voltage transformer.

[0105] Sequences of discharge activities of the multiple positions are collected according to the multiple positions for monitoring discharge activities.

[0106] Specifically, a static electric field solver (such as the Electrostatic module of ANSYS Maxwell) is used to calculate the steady-state electric field distribution, or a transient electric field solver is used to analyze the electric field change under transient overvoltage, and a distribution cloud map of electric field intensity is output. The areas where the electric field intensity exceeds the material tolerance threshold are prone to partial discharge, for example, the edges of the winding end insulating paperboard; the air gap between the core and the winding. Geometric shape mutations (such as sharp corners, notches) or material inhomogeneity cause electric field concentration, for example. The tiny gap of inter-turn insulation of the winding; the surface contamination or cracks of the insulator. The multiple positions for monitoring discharge activities are arranged in the areas where the electric field intensity exceeds 80% of the material tolerance threshold, and the multiple positions for monitoring discharge activities are arranged on the path where the discharge may propagate along the insulating surface or gas gap, for example: the surface discharge path from the winding end to the shell; the discharge channel of the SF6 gas gap in the gas insulated switch (GIS).

[0107] A sensor can be set at each position for discharge activity monitoring to collect sequences of discharge activities of the positions.

[0108] In some embodiments, sequences of temperatures of multiple positions of the fuse are collected, including:

[0109] Finite element analysis is performed based on the finite element model of the voltage transformer to determine the distribution of temperature inside the fuse, and multiple positions for monitoring temperature are determined according to the distribution of temperature inside the fuse.

[0110] Sequences of temperatures of the multiple positions of the fuse are collected according to the multiple positions for monitoring temperature.

[0111] Specifically, the temperature distribution is calculated using a steady-state or transient heat analysis solver (such as ANSYS Mechanical, COMSOL Multiphysics), and a temperature distribution cloud chart is output. Temperature monitoring can be performed at multiple locations in the region where the temperature is greater than the temperature threshold or the temperature gradient is greater than the temperature gradient threshold, where the temperature threshold and the temperature gradient threshold can be set according to artificial experience. For example, the area with the largest melt resistance loss (such as the cross-section mutation and the connection point) is prone to become a high-temperature point, which may cause the fuse to melt prematurely or heat aging. The area with a large temperature gradient in the arc extinguishing medium (such as the melt around) may affect the arc extinguishing performance, causing arc reignition. Local overheating caused by contact resistance may cause connection failure or uneven melting of the fuse.

[0112] A sensor can be provided at each temperature monitoring location to collect a temperature sequence of the location.

[0113] Step 180, determining the PT fuse failure cause based on the electrical parameters, environmental parameters and physical state of the voltage transformer through the PT fuse failure diagnosis model.

[0114] For example, assuming that the cascade strategy determined in step 130 is:

[0115] First-level fault type: electrical fault

[0116] Second-level fault type: environmental fault

[0117] Third-level fault type: state fault.

[0118] When applied:

[0119] Input data: real-time electrical parameters, environmental parameters and state parameters of the voltage transformer.

[0120] First-level diagnosis: run the electrical fault diagnosis sub-model to output the probability of electrical fault.

[0121] If the probability is ≥ 30%, diagnose as an electrical fault.

[0122] If the probability is < 30%, proceed to the second-level diagnosis.

[0123] Second-level diagnosis: run the environmental fault diagnosis sub-model to output the probability of environmental fault.

[0124] If the probability is ≥ 30%, diagnose as an environmental fault.

[0125] If the probability is < 30%, proceed to the third-level diagnosis.

[0126] Third-level diagnosis: run the state fault diagnosis sub-model to output the probability of state fault, and make the final diagnosis.

[0127] Finally, it should be understood that the embodiments described herein are only given by way of example and that other modifications can occur to persons skilled in the art. Therefore, the scope of the present description is not intended to be limited to the embodiments described herein but is only limited by the claims that follow.

Claims

1. A method for constructing a PT fuse fault diagnosis model in a plateau environment, characterized in that: include: Establish and train electrical fault diagnosis sub-model, environmental fault diagnosis sub-model and state fault diagnosis sub-model; Obtaining structural information and operating environment information of the voltage transformer; Generate a cascade strategy corresponding to the voltage transformer based on the structure information and operating environment information of the voltage transformer; Based on the cascade strategy corresponding to the voltage transformer, the electrical fault diagnosis sub-model, the environmental fault diagnosis sub-model and the state fault diagnosis sub-model, a PT fuse fault diagnosis model is constructed; Collecting electrical parameters of the voltage transformer, wherein the electrical parameters include at least primary side voltage, secondary side voltage and secondary side current; Collecting environmental parameters of the voltage transformer, wherein the environmental parameters include at least temperature, humidity and air pressure; Collecting a physical state of the voltage transformer, wherein the physical state includes at least vibrations at multiple locations of the voltage transformer, discharge activities at multiple locations, and temperatures at multiple locations of the fuse; The PT fuse fault diagnosis model is used to determine the cause of the PT fuse fault based on the electrical parameters, environmental parameters and physical status of the voltage transformer.

2. The method for constructing a PT fuse fault diagnosis model under a plateau environment according to claim 1, characterized in that: The electrical fault diagnosis sub-model includes a first input unit, a first feature extraction unit and an electrical fault diagnosis unit; Wherein, the first input unit is used to input the primary side voltage sequence, the secondary side voltage sequence and the secondary side current sequence and perform preprocessing; The first feature extraction unit includes a first electrical feature extraction layer, a second electrical feature extraction layer, a third electrical feature extraction layer and an electrical feature fusion layer, wherein the first electrical feature extraction unit is used to extract the time domain features, frequency domain features and voltage harmonic features of the primary side voltage sequence, the second electrical feature extraction layer is used to extract the time domain features, frequency domain features and voltage harmonic features of the secondary side voltage sequence, the third electrical feature extraction layer is used to extract the time domain features, frequency domain features and current harmonic features of the secondary side current sequence, and the electrical feature fusion layer is used to fuse the outputs of the first electrical feature extraction layer, the second electrical feature extraction layer and the third electrical feature extraction layer to generate an electrical feature vector; The electrical fault diagnosis unit is used to determine the occurrence probability of various electrical faults based on the electrical characteristic vector.

3. The method for constructing a PT fuse fault diagnosis model under a plateau environment according to claim 1, characterized in that: The environmental fault diagnosis sub-model includes a second input unit, a second feature extraction unit and an environmental guidance fault diagnosis unit; Wherein, the second input unit is used to input a temperature sequence, a humidity sequence and an air pressure sequence and perform preprocessing; The second feature extraction unit includes an ambient temperature feature extraction layer, a humidity feature extraction layer, an air pressure extraction layer, and an ambient feature fusion layer. The ambient temperature feature extraction layer is used to extract the time domain features and frequency domain features of the temperature sequence. The humidity feature extraction layer is used to extract the time domain features and frequency domain features of the humidity sequence. The air pressure extraction layer is used to extract the time domain features and frequency domain features of the air pressure sequence. The ambient feature fusion layer is used to fuse the outputs of the ambient temperature feature extraction layer, the humidity feature extraction layer, and the air pressure extraction layer to generate an ambient feature vector. The environment-induced fault diagnosis unit is used to determine the occurrence probabilities of various environment-induced faults based on the environment feature vector.

4. The method for constructing a PT fuse fault diagnosis model in a plateau environment according to claim 1, characterized in that: The state fault diagnosis sub-model includes a third input unit, a third feature extraction unit and a state fault diagnosis unit; The third input unit is used to input vibration sequences of multiple positions of the voltage transformer, discharge activity sequences of multiple positions of the fuse, and temperature sequences of multiple positions of the fuse and perform preprocessing; The third feature extraction unit includes a vibration feature extraction layer, a discharge activity extraction layer, a fuse temperature feature extraction layer and a physical state feature fusion layer. The vibration feature extraction layer is used to extract the time domain features and frequency domain features of the vibration sequence of each position and fuse them. The discharge activity extraction layer is used to extract the time domain features and frequency domain features of the discharge activity sequence of each position and fuse them. The fuse temperature feature extraction layer is used to extract the time domain features and frequency domain features of the temperature sequence of each position and fuse them. The physical state feature fusion layer is used to fuse the outputs of the vibration feature extraction layer, the discharge activity extraction layer and the fuse temperature feature extraction layer to generate a physical state feature vector. The state fault diagnosis unit is used to determine the occurrence probability of multiple physical faults based on the physical state feature vector.

5. The method for constructing a PT fuse fault diagnosis model in a plateau environment according to any one of claims 1 to 4, characterized in that: Based on the voltage transformer's structural information and operating environment information, a cascade strategy corresponding to the voltage transformer is generated, including: Establishing a sample database, wherein the sample database includes cascading strategies of multiple sample voltage transformers in multiple sample operating environments; Determining a target sample voltage transformer based on the structure information of the voltage transformer and a sample database; Determine a target sample operating environment based on the operating environment information of the voltage transformer and multiple sample operating environments; Based on the cascade strategy of the target sample voltage transformer in the target sample operating environment, a cascade strategy corresponding to the voltage transformer is generated.

6. The method for constructing a PT fuse fault diagnosis model in a plateau environment according to claim 5, characterized in that: Determine the cascade strategy for each sample voltage transformer in each sample operating environment, including: Acquiring fault information of a sample voltage transformer in multiple historical time periods, wherein the fault information includes the number of occurrences of electrical faults, environmental faults, and physical faults; Determine the weights of electrical faults, environmental faults, and physical faults based on fault information of sample voltage transformers in multiple historical time periods; Determine the correlation coefficient between any two of electrical faults, environmentally induced faults, and physical faults based on fault information of sample voltage transformers in multiple historical time periods; A cascade strategy of the sample voltage transformer in the sample operating environment is determined based on the weights of the electrical fault, the environmentally induced fault, and the physical fault and the correlation coefficient between any two of the electrical fault, the environmentally induced fault, and the physical fault.

7. The method for constructing a PT fuse fault diagnosis model in a plateau environment according to claim 6, characterized in that: Based on the weights of electrical faults, environmental faults, and physical faults and the correlation coefficients between any two of the electrical faults, environmental faults, and physical faults, a cascading strategy for the sample voltage transformer in the sample operating environment is determined, including: Calculate the comprehensive correlation coefficient of electrical faults, environmental induced faults and physical faults based on the correlation coefficients of any two of the electrical faults, environmental induced faults and physical faults; Calculate the priority values ​​of electrical faults, environmental faults and physical faults based on their weights and their comprehensive correlation coefficients; Determine the first-level fault type based on the judgment priority values ​​of electrical faults, environmental faults and physical faults; Based on the correlation coefficient and the first-level fault type of any two of the electrical fault, the environmentally induced fault and the physical fault, a cascade strategy of the sample voltage transformer in the sample operating environment is determined.

8. The method for constructing a PT fuse fault diagnosis model in a plateau environment according to claim 4, characterized in that: Acquire vibration sequences at multiple locations on the voltage transformer, including: Establish a three-dimensional model of the voltage transformer; Based on the operating environment information of the voltage transformer and the three-dimensional model of the voltage transformer, a finite element model of the voltage transformer is established. Finite element analysis is performed based on the finite element model of the voltage transformer to determine the electromagnetic force distribution and mechanical stress distribution inside the voltage transformer. Based on the electromagnetic force distribution and mechanical stress distribution inside the voltage transformer, multiple locations for vibration monitoring are determined; According to the multiple locations for vibration monitoring, vibration sequences of multiple locations of the voltage transformer are collected.

9. The method for constructing a PT fuse fault diagnosis model in a plateau environment according to claim 8, characterized in that: Capture sequences of discharge activity at multiple locations, including: Performing finite element analysis based on a finite element model of the voltage transformer to determine the electric field strength distribution inside the voltage transformer, and determining multiple locations for discharge activity monitoring based on the electric field strength distribution inside the voltage transformer; According to the multiple locations where discharge activity monitoring is performed, discharge activity sequences of the multiple locations are collected.

10. The method for constructing a PT fuse fault diagnosis model in a plateau environment according to claim 8, characterized in that: Acquire temperature series at multiple locations on the fuse, including: Perform finite element analysis based on the finite element model of the voltage transformer to determine the temperature distribution inside the fuse. Based on the temperature distribution inside the fuse, determine multiple locations for temperature monitoring. According to the multiple positions for temperature monitoring, temperature sequences of the multiple positions of the fuse are collected.

Images