A method and device for identifying faults of a fan engine dry-type transformer

Abstract

The application relates to the technical field of fault diagnosis, in particular to a fan engine room dry-type transformer fault identification method and device. The method comprises the following steps: acquiring a temperature change curve, a current change curve and a voltage change curve of a fan engine room dry-type transformer; based on the temperature change curve, the current change curve and the voltage change curve, temperature influence data, current influence data and voltage influence data are sequentially determined; the temperature influence data, the current influence data and the voltage influence data are subjected to feature extraction, and temperature features, current features and voltage features are sequentially obtained; the temperature features, the current features and the voltage features are fused according to an environmental influence proportion to obtain a fusion matrix; wherein the environmental influence proportion is determined according to environmental parameters of the fan engine room dry-type transformer; and the fusion matrix is input into a preset first mixed model to obtain a first fault identification result. In this way, the application can improve the fault identification precision of the fan engine room dry-type transformer.

Description

Technical Field

[0001] This invention relates to the field of fault diagnosis technology, and in particular to a fault identification method and device for a dry-type transformer in a wind turbine nacelle. Background Technology

[0002] Wind turbines are typically deployed in outdoor environments, where their nacelles operate under harsh conditions of high vibration, high dust levels, and drastic temperature and humidity fluctuations. As the component responsible for the conversion and transmission of electrical energy in wind turbines, the operational stability of the dry-type transformer directly determines the power generation efficiency, operational safety, and reliability of the unit.

[0003] Existing dry-type transformer fault diagnosis technologies mostly rely on offline detection methods (such as periodic power outage detection) and manual experience judgment, which have many insurmountable defects: First, the response lag is significant, and it is impossible to capture weak signals of early faults such as the initial stage of inter-turn short circuit, slight overheating of windings, and local heating of core in real time. Often, the fault can only be detected when it has developed to a serious stage (such as insulation breakdown and winding burnout), missing the best time for handling. Second, the anti-interference ability is weak. Strong mechanical vibration and complex electromagnetic environment in the nacelle can easily lead to distortion of detection signals, resulting in misjudgment or missed faults, increasing ineffective operation and maintenance costs.

[0004] Based on this, the present invention proposes a fault identification method and device for dry-type transformers in wind turbine nacelles to solve the above-mentioned technical problems. Summary of the Invention

[0005] This invention describes a fault identification method and apparatus for dry-type transformers in wind turbine nacelles, which can improve the accuracy of fault identification for dry-type transformers in wind turbine nacelles.

[0006] According to a first aspect, the present invention provides a fault identification method for a dry-type transformer in a wind turbine nacelle, comprising: Obtain the temperature change curve, current change curve, and voltage change curve of the dry-type transformer in the wind turbine nacelle; Based on the temperature change curve, the current change curve, and the voltage change curve, temperature influence data, current influence data, and voltage influence data are determined sequentially; wherein, the temperature influence data includes the temperature rise rate, peak temperature, and temperature fluctuation amplitude; the current influence data includes the current amplitude abrupt change value, harmonic distortion rate, and current imbalance; and the voltage influence data includes the voltage amplitude fluctuation amplitude, voltage drop duration, and three-phase voltage imbalance. The temperature influence data, the current influence data, and the voltage influence data are subjected to feature extraction to obtain temperature features, current features, and voltage features in sequence. The temperature feature, current feature, and voltage feature are fused according to the environmental impact ratio to obtain a fusion matrix; wherein the environmental impact ratio is determined based on the environmental parameters of the dry-type transformer in the wind turbine nacelle. The fusion matrix is ​​input into a preset first hybrid model to obtain the first fault identification result.

[0007] According to a second aspect, the present invention provides a fault identification device for a dry-type transformer in a wind turbine nacelle, comprising: The acquisition unit is configured to acquire the temperature change curve, current change curve, and voltage change curve of the dry-type transformer in the wind turbine nacelle. The first data processing unit is configured to determine temperature influence data, current influence data, and voltage influence data sequentially based on the temperature change curve, the current change curve, and the voltage change curve; wherein the temperature influence data includes the temperature rise rate, peak temperature, and temperature fluctuation amplitude; the current influence data includes the current amplitude abrupt change value, harmonic distortion rate, and current imbalance; and the voltage influence data includes the voltage amplitude fluctuation amplitude, voltage drop duration, and three-phase voltage imbalance. The second data processing unit is configured to extract features from the temperature influence data, the current influence data and the voltage influence data to obtain temperature features, current features and voltage features in sequence. The third data processing unit is configured to fuse the temperature feature, the current feature, and the voltage feature according to the environmental influence ratio to obtain a fusion matrix; wherein the environmental influence ratio is determined based on the environmental parameters of the dry-type transformer of the wind turbine nacelle. The fourth data processing unit is configured to input the fusion matrix into a preset first hybrid model to obtain a first fault identification result.

[0008] Thirdly, embodiments of this specification also provide an electronic device, including a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, it implements the method described in any embodiment of this specification.

[0009] Fourthly, embodiments of this specification also provide a computer-readable storage medium having a computer program stored thereon, which, when executed in a computer, causes the computer to perform the methods described in any embodiment of this specification.

[0010] According to the fault identification method and apparatus for a dry-type transformer in a wind turbine nacelle provided by the present invention, firstly, the temperature change curve, current change curve, and voltage change curve of the dry-type transformer in the wind turbine nacelle are simultaneously collected by sensing devices. Specifically, temperature data is collected using a fluorescent fiber optic temperature sensor deployed on heat-generating components such as the transformer windings and core to ensure the accuracy of temperature change monitoring; current and voltage data are collected using a Rogowski coil current sensor and a high-precision voltage sensor, respectively, to ensure the temporal integrity of the curves. For the temperature change curve, the temperature rise rate, peak temperature, and temperature fluctuation amplitude are calculated through time-series data analysis, constituting temperature influence data. This type of data directly reflects the thermal operating state of the transformer and is a characterization of faults such as winding overheating and abnormal core losses. For the current change curve, current amplitude mutation values, harmonic distortion rates, and current imbalance are extracted as current influence data through amplitude mutation detection, harmonic analysis, and three-phase balance calculation. This data is used to capture current anomalies caused by faults such as inter-turn short circuits and uneven loads. For the voltage change curve, voltage amplitude fluctuation amplitude, voltage drop duration, and three-phase voltage imbalance are determined as voltage influence data through fluctuation amplitude statistics, voltage drop duration, and three-phase voltage difference analysis. This data can effectively characterize voltage steady-state imbalance problems caused by faults such as interlayer breakdown and insulation aging. Subsequently, feature extraction is performed on the temperature influence data, current influence data, and voltage influence data to obtain temperature features, current features, and voltage features, respectively. Considering that complex environmental factors such as wind turbine nacelle vibration, low temperature, electromagnetic interference, and dust can significantly affect the reliability of fault characterization for various features, this invention calculates the environmental impact ratio based on real-time collected nacelle environmental parameters. Temperature, current, and voltage features are then dynamically weighted and fused according to this ratio to generate a fusion matrix adapted to the current environmental conditions. This allows for adaptive adjustment of feature weights under different environments, reducing the impact of environmental factors on fault identification results. Finally, the fusion matrix is ​​input into a preset first hybrid model, outputting the first fault identification result. This invention effectively reduces the interference of complex nacelle environments on feature characterization through an environmentally adaptive feature fusion strategy. Combined with the accurate identification capability of the first hybrid model, it significantly improves the accuracy of fault identification for dry-type transformers in wind turbine nacelles, providing data support for early fault warning of equipment. Attached Figure Description

[0011] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0012] Figure 1 A flowchart illustrating a fault identification method for a dry-type transformer in a wind turbine nacelle according to one embodiment is shown. Figure 2 A schematic block diagram of a fault identification device for a dry-type transformer in a wind turbine nacelle according to one embodiment is shown. Detailed Implementation

[0013] The solution provided by the present invention will now be described with reference to the accompanying drawings.

[0014] Figure 1 This diagram illustrates a fault identification method for a dry-type transformer in a wind turbine nacelle according to one embodiment. It is understood that this method can be executed by any device, equipment, platform, or cluster of devices with computing and processing capabilities. Figure 1 As shown, the method includes: Step 100: Obtain the temperature change curve, current change curve, and voltage change curve of the dry-type transformer in the wind turbine nacelle; Step 102: Based on the temperature change curve, current change curve, and voltage change curve, determine the temperature influence data, current influence data, and voltage influence data in sequence. Among them, the temperature influence data includes the temperature rise rate, peak temperature, and temperature fluctuation amplitude; the current influence data includes the current amplitude abrupt change value, harmonic distortion rate, and current imbalance; and the voltage influence data includes the voltage amplitude fluctuation amplitude, voltage drop duration, and three-phase voltage imbalance. Step 104: Extract features from the temperature effect data, current effect data, and voltage effect data to obtain temperature features, current features, and voltage features in sequence. Step 106: Merge the temperature characteristics, current characteristics, and voltage characteristics according to the environmental impact ratio to obtain a fusion matrix; whereby the environmental impact ratio is determined based on the environmental parameters of the dry-type transformer in the wind turbine nacelle. Step 108: Input the fusion matrix into the preset first hybrid model to obtain the first fault identification result.

[0015] In this embodiment, firstly, the temperature change curve, current change curve, and voltage change curve of the dry-type transformer in the wind turbine nacelle are simultaneously acquired using sensing devices. Specifically, temperature data is acquired using fluorescent fiber optic temperature sensors deployed on heat-generating components such as the transformer windings and core to ensure accurate temperature change monitoring. Current and voltage data are acquired using Rogowski coil current sensors and high-precision voltage sensors, respectively, to ensure the temporal integrity of the curves. For the temperature change curve, the temperature rise rate, peak temperature, and temperature fluctuation amplitude are calculated through time-series data analysis, constituting temperature influence data. This type of data directly reflects the thermal operating state of the transformer and is a characterization of faults such as winding overheating and abnormal core losses. For the current change curve, current amplitude mutation values, harmonic distortion rates, and current imbalance are extracted as current influence data through amplitude mutation detection, harmonic analysis, and three-phase balance calculation. This data is used to capture current anomalies caused by faults such as inter-turn short circuits and uneven loads. For the voltage change curve, voltage amplitude fluctuation amplitude, voltage drop duration, and three-phase voltage imbalance are determined as voltage influence data through fluctuation amplitude statistics, voltage drop duration, and three-phase voltage difference analysis. This data can effectively characterize voltage steady-state imbalance problems caused by faults such as interlayer breakdown and insulation aging. Subsequently, feature extraction is performed on the temperature influence data, current influence data, and voltage influence data to obtain temperature features, current features, and voltage features, respectively. Considering that complex environmental factors such as wind turbine nacelle vibration, low temperature, electromagnetic interference, and dust can significantly affect the reliability of fault characterization for various features, this invention calculates the environmental impact ratio based on real-time collected nacelle environmental parameters. Temperature, current, and voltage features are then dynamically weighted and fused according to this ratio to generate a fusion matrix adapted to the current environmental conditions. This allows for adaptive adjustment of feature weights under different environments, reducing the impact of environmental factors on fault identification results. Finally, the fusion matrix is ​​input into a preset first hybrid model to output the first fault identification result. This invention... By adopting an environment-adaptive feature fusion strategy, the interference of the complex nacelle environment on feature representation is effectively reduced. Combined with the accurate identification capability of the first hybrid model, the accuracy of fault identification of dry-type transformers in wind turbine nacelles is greatly improved, providing data support for early fault warning of equipment.

[0016] In one embodiment of the present invention, the preset first hybrid model includes a first input layer, a one-dimensional CNN spatial feature extraction layer, a BiLSTM temporal feature mining layer, a feature concatenation layer, and a first fully connected output layer connected in sequence. The first input layer is used to receive the fusion matrix, the one-dimensional CNN spatial feature extraction layer is used to extract the local spatial correlation features of the fault features in the fusion matrix, the BiLSTM temporal feature mining layer is used to extract the bidirectional temporal dependency features of the fault features in the fusion matrix, the feature concatenation layer receives the local spatial correlation features and the bidirectional temporal dependency features and outputs the spatiotemporal fusion feature matrix, and the first fully connected output layer receives the spatiotemporal fusion feature matrix and outputs the first fault identification result.

[0017] In this embodiment, the preset first hybrid model includes a first input layer, a one-dimensional CNN spatial feature extraction layer, a BiLSTM temporal feature mining layer, a feature concatenation layer, and a first fully connected output layer connected in sequence. The first input layer is used to receive the fusion matrix, and the output data of the first input layer is directly fed into the one-dimensional CNN spatial feature extraction layer. This layer adopts a multi-convolution kernel parallel computing design, and performs local feature scanning on the fusion matrix through convolution kernels of different sizes to extract local spatial correlation features of fault features, such as the spatial coupling pattern of temperature and current anomalies, the local correlation pattern of voltage fluctuations and other features, etc., which effectively enhances the local discriminative power of features. The output of the one-dimensional CNN spatial feature extraction layer is connected to the BiLSTM temporal feature mining layer. The BiLSTM layer comprehensively captures the bidirectional temporal dependency features of fault features in the fusion matrix through temporal calculations in both forward and reverse directions. This allows it to not only uncover the forward evolution trend of fault development but also trace the historical temporal origin information of fault occurrence, adapting to the temporal characteristics of the gradual evolution of early transformer faults. The output of the BiLSTM temporal feature mining layer is fed into the feature concatenation layer. This layer performs dimensional concatenation and information fusion on local spatial correlation features and bidirectional temporal dependency features, generating a spatiotemporal fusion feature matrix that combines spatial discriminability and temporal integrity. Finally, the spatiotemporal fusion feature matrix is ​​fed into the first fully connected output layer, where a multilayer perceptron performs fault pattern matching and classification on the fusion features, ultimately outputting the first fault identification result.

[0018] In one embodiment of the present invention, after inputting the fusion matrix into a preset first hybrid model to obtain a first fault identification result, the method further includes: The fusion matrix is ​​input into the preset second hybrid model to obtain the second fault identification result; Based on the first fault identification result and the second fault identification result, the third fault identification result is determined; Both the first fault identification result and the second fault identification result include the fault category and confidence level.

[0019] In this embodiment, firstly, the fusion matrix is ​​synchronously input into a preset second hybrid model. This model complements the recognition scenario of the first hybrid model, enabling secondary judgment of fault modes from the perspective of long-term time-series feature correlation, and ultimately outputting a second fault identification result. Subsequently, the first fault identification result and the second fault identification result are compared. Finally, a third fault identification result is determined. This invention, through the collaborative verification of dual models, effectively reduces the risk of misjudgment caused by feature capture bias in a single model, further ensuring the accuracy and stability of the fault identification results.

[0020] In one embodiment of the present invention, the preset second hybrid model includes a second input layer, a TCN long-time-series feature extraction layer, a multi-head attention weight allocation layer, a feature fusion layer, and a second fully connected output layer connected in sequence. The second input layer is used to receive the fusion matrix, the TCN long-time-series feature extraction layer is used to extract long-time-series causal correlation features of fault features in the fusion matrix, the multi-head attention weight allocation layer is used to allocate attention weights to the long-time-series causal correlation features to focus on key fault features, the feature fusion layer receives the weighted long-time-series causal correlation features and outputs a weighted fusion feature matrix, and the second fully connected output layer receives the weighted fusion feature matrix and outputs a second fault identification result.

[0021] In this embodiment, the pre-defined second hybrid model includes a second input layer, a TCN long-term feature extraction layer, a multi-head attention weight allocation layer, a feature fusion layer, and a second fully connected output layer connected in sequence. The second input layer receives the fusion matrix, and its output data is directly fed into the TCN long-term feature extraction layer. This layer employs a design combining causal convolution and dilated convolution to gradually expand the receptive field and capture long-term causal correlation features of fault features in the fusion matrix, such as the cumulative evolution of weak features in the early stage of a fault and the transmission relationship of fault features at different time stages, effectively compensating for the shortcomings of short-term models in capturing long-period features. The output of the TCN long-term feature extraction layer is connected to the multi-head attention weight allocation layer. This layer uses multiple attention heads to perform parallel computation and dynamically allocates attention weights to the extracted long-term causal correlation features, focusing on key features that play a decisive role in fault identification (such as temperature mutation features under extreme environments and voltage anomalies corresponding to insulation aging), while weakening the interference of irrelevant and redundant features. The output of the multi-head attention weight allocation layer is fed into the feature fusion layer. This layer integrates information and optimizes the dimensions of the weighted long-term causal correlation features, outputting a weighted fusion feature matrix that combines temporal integrity with the salience of key features. Finally, the weighted fusion feature matrix is ​​fed into the second fully connected output layer, where a multilayer perceptron performs accurate matching and classification of fault modes, ultimately outputting the second fault identification result.

[0022] In one embodiment of the present invention, determining a third fault identification result based on a first fault identification result and a second fault identification result includes: When the fault categories determined by the first fault identification result and the second fault identification result are the same, the first fault identification result or the second fault identification result is replaced by the third fault identification result. When the fault categories determined by the first fault identification result and the second fault identification result are different, the result with higher confidence between the first fault identification result and the second fault identification result shall be selected as the third fault identification result. The fault categories include inter-turn short circuit, inter-layer breakdown, local overheating of windings, abnormal heating of iron core, neutral point offset, and grounding imbalance.

[0023] In this embodiment, firstly, the consistency of the fault categories output by the first fault identification result and the second fault identification result is checked. If the check finds that the fault categories determined by the first fault identification result and the second fault identification result are exactly the same, then the consistent fault category is directly determined as the third fault identification result. This reduces the risk of misjudgment caused by the local feature capture deviation of the single model by leveraging the reliability of the dual model and the fault determination. If there is a difference in the fault categories output by the two models, the confidence levels of the first fault identification result and the second fault identification result are compared, and the identification result with the higher confidence level is selected as the third fault identification result. The first hybrid model is good at short-to-medium time-series spatiotemporal feature mining, while the second hybrid model has a greater advantage in long-term time-series causal feature capture. Through confidence quantification and filtering, the complementary optimization of the identification capabilities of the two models can be achieved, and the final output third fault identification result can more accurately match the actual fault state of the transformer.

[0024] In one embodiment of the present invention, the environmental impact ratio includes preliminary correction weights for temperature features, preliminary correction weights for current features, and preliminary correction weights for voltage features. The environmental impact ratio is determined using the following formula:

[0025] In the formula, The weights of temperature features were initially adjusted. As the baseline weight for temperature-related features, This is the weighting gain coefficient for vibration versus temperature characteristics. The vibration interference environmental factor, This represents the weighting attenuation coefficient of low temperature on temperature characteristics. The low-temperature interference environment coefficient, This represents the weighting gain coefficient for the temperature characteristics caused by electromagnetic interference. Electromagnetic interference environmental factor, Preliminary weighting of current-related features. As the benchmark weight for current-type features, This is the weighted attenuation coefficient of vibration on the current characteristic. This is the weighting attenuation coefficient for electromagnetic interference on the current characteristics. This is the weighting attenuation coefficient for voltage fluctuations on current characteristics. The voltage fluctuation environmental factor. Preliminary weighting of voltage-related features. As the benchmark weight for voltage-type features, This represents the weighting attenuation coefficient of voltage fluctuations on voltage characteristics. The dust fluctuation environmental coefficient. This represents the weighting gain coefficient for humidity on voltage characteristics. The humidity fluctuation environmental coefficient, This is the weighting attenuation coefficient of voltage fluctuation on voltage characteristics.

[0026] In this embodiment, the reference weights for temperature features, current features, and voltage features are all one-third. The weight gain coefficient for vibration on temperature features, used to characterize the degree of reliability improvement of temperature features when vibration interferes with the electrical signal, is set to 0.4. The weight attenuation coefficient for low temperature on temperature features, used to characterize the degree of weight reduction when low temperature causes sensor response delay, is set to 0.5. The weight gain coefficient for electromagnetic interference on temperature features, used to characterize the degree of reliability improvement of temperature features when electromagnetic interference distorts the electrical signal, is set to 0.3. The weight attenuation coefficient for vibration on current features, used to characterize the degree of weight reduction of current sampling jitter caused by vibration, is set to 0.45. The weight attenuation coefficient for electromagnetic interference on current features, used to characterize the degree of weight reduction of electromagnetic harmonic distortion of the current signal, is set to 0.5. The following parameters are used: 0.3 (weight attenuation coefficient for voltage fluctuation on current characteristics, used to characterize the degree of weight reduction in current sampling coupled with grid voltage fluctuation); 0.35 (weight gain coefficient for dust on voltage characteristics, used to characterize the degree of increase in the sensitivity of voltage characteristics when dust reduces insulation performance); 0.3 (weight gain coefficient for humidity on voltage characteristics, used to characterize the degree of increase in the sensitivity of voltage characteristics when humidity exacerbates insulation aging); and 0.25 (weight attenuation coefficient for voltage fluctuation on voltage characteristics, used to characterize the degree of weight reduction in sampling distortion caused by grid voltage fluctuation); and 0.4 (weight attenuation coefficient for voltage fluctuation on voltage characteristics, used to characterize the degree of weight reduction in sampling distortion caused by grid voltage fluctuation). Following the formula for calculating the environmental impact ratio, environmental parameters such as vibration and low temperature are bound to the sensing characteristics of the features (e.g., temperature features are detected using optical sensing, vibration does not interfere, hence a vibration gain term is set), rather than pure black-box data fitting. Multiple environmental factors such as vibration, electromagnetic fields, and dust are incorporated to adapt to the complex operating conditions of the wind turbine nacelle and avoid the limitations of single-environment correction. Environmental coefficients are updated in real time according to operating conditions, and weights can dynamically match the reliability of features under the current environment, making the fault identification model more accurate in complex scenarios.

[0027] In one embodiment of the present invention, the environmental factor is determined by the following formula:

[0028] In the formula, The vibration characteristic coupling coefficient, To measure vibration acceleration, As the reference vibration acceleration threshold, The coupling coefficient of the low-temperature sensor response. Based on ambient temperature, To measure the ambient temperature in the cabin, The electromagnetic coupling coefficient is... To measure the electromagnetic radiation intensity, The electromagnetic coupling coefficient is... The coupling coefficient for dust insulation performance. To measure the dust concentration, As the baseline dust concentration threshold, The coupling coefficient is the coefficient for humidity-induced insulation aging. To measure the relative humidity, The baseline relative humidity is The voltage fluctuation sampling accuracy coupling coefficient is... To measure the voltage fluctuation value of the power grid, This is the rated voltage of the transformer.

[0029] In this embodiment, the reference vibration acceleration threshold is 1g, the vibration characteristic coupling coefficient is 0.4, the reference ambient temperature is 25 degrees Celsius, the low temperature sensor response coupling coefficient is 0.5, the reference electromagnetic interference threshold is 10 volts per meter, the electromagnetic signal coupling coefficient is 0.5, the reference dust concentration threshold is 0.5 milligrams per cubic meter, the dust insulation performance coupling coefficient is 0.3, the reference relative humidity is 60%, the humidity insulation aging coupling coefficient is 0.25, and the voltage fluctuation sampling accuracy coupling coefficient is 0.4.

[0030] The foregoing has described specific embodiments of the invention. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps described in the claims may be performed in a different order than that shown in the embodiments and may still achieve the desired results. Furthermore, the processes depicted in the drawings do not necessarily require the specific or sequential order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0031] According to another embodiment, the present invention provides a fault identification device for a dry-type transformer in a wind turbine nacelle. Figure 2A schematic block diagram of a fault identification device for a dry-type transformer in a wind turbine nacelle according to one embodiment is shown. It will be understood that this device can be implemented by any device, equipment, platform, or cluster of devices with computing and processing capabilities. Figure 2 As shown, the device includes: an acquisition unit 200, a first data processing unit 202, a second data processing unit 204, a third data processing unit 206, and a fourth data processing unit 208. The main functions of each component are as follows: The acquisition unit 200 is configured to acquire the temperature change curve, current change curve and voltage change curve of the dry-type transformer in the wind turbine nacelle. The first data processing unit 202 is configured to determine temperature influence data, current influence data, and voltage influence data sequentially based on the temperature change curve, the current change curve, and the voltage change curve; wherein, the temperature influence data includes the temperature rise rate, peak temperature, and temperature fluctuation amplitude; the current influence data includes the current amplitude abrupt change value, harmonic distortion rate, and current imbalance; and the voltage influence data includes the voltage amplitude fluctuation amplitude, voltage drop duration, and three-phase voltage imbalance. The second data processing unit 204 is configured to extract features from the temperature influence data, the current influence data and the voltage influence data to obtain temperature features, current features and voltage features in sequence. The third data processing unit 206 is configured to fuse the temperature feature, the current feature, and the voltage feature according to the environmental influence ratio to obtain a fusion matrix; wherein the environmental influence ratio is determined based on the environmental parameters of the dry-type transformer of the wind turbine nacelle. The fourth data processing unit 208 is configured to input the fusion matrix into a preset first hybrid model to obtain a first fault identification result.

[0032] In one embodiment of the present invention, the preset first hybrid model includes a first input layer, a one-dimensional CNN spatial feature extraction layer, a BiLSTM temporal feature mining layer, a feature concatenation layer, and a first fully connected output layer connected in sequence; the first input layer is used to receive the fusion matrix, the one-dimensional CNN spatial feature extraction layer is used to extract local spatial correlation features of fault features in the fusion matrix, the BiLSTM temporal feature mining layer is used to extract bidirectional temporal dependency features of fault features in the fusion matrix, the feature concatenation layer receives the local spatial correlation features and the bidirectional temporal dependency features and outputs a spatiotemporal fusion feature matrix, and the first fully connected output layer receives the spatiotemporal fusion feature matrix and outputs the first fault identification result.

[0033] In one embodiment of the present invention, the apparatus further includes a fifth data processing unit, the fifth data processing unit being configured to perform the following operations: The fusion matrix is ​​input into a preset second hybrid model to obtain a second fault identification result; Based on the first fault identification result and the second fault identification result, a third fault identification result is determined. Both the first fault identification result and the second fault identification result include fault category and confidence level.

[0034] In one embodiment of the present invention, the preset second hybrid model includes a second input layer, a TCN long-time-series feature extraction layer, a multi-head attention weight allocation layer, a feature fusion layer, and a second fully connected output layer connected in sequence. The second input layer is used to receive the fusion matrix, the TCN long-time-series feature extraction layer is used to extract long-time-series causal correlation features of fault features in the fusion matrix, the multi-head attention weight allocation layer is used to allocate attention weights to the long-time-series causal correlation features to focus on key fault features, the feature fusion layer receives the weighted long-time-series causal correlation features and outputs a weighted fusion feature matrix, and the second fully connected output layer receives the weighted fusion feature matrix and outputs the second fault identification result.

[0035] In one embodiment of the present invention, the fifth data processing unit, when performing the operation of "determining a third fault identification result based on the first fault identification result and the second fault identification result", performs the following operation: When the fault categories determined by the first fault identification result and the second fault identification result are the same, the first fault identification result or the second fault identification result is replaced with the third fault identification result. When the fault categories determined by the first fault identification result and the second fault identification result are different, the result with higher confidence between the first fault identification result and the second fault identification result is selected as the third fault identification result. The fault categories include inter-turn short circuit, inter-layer breakdown, local overheating of windings, abnormal heating of core, neutral point offset, and grounding imbalance.

[0036] In one embodiment of the present invention, the proportion of environmental impact includes preliminary correction weights for temperature features, preliminary correction weights for current features, and preliminary correction weights for voltage features. The environmental impact ratio is determined using the following formula:

[0037] In the formula, The weights of the temperature-related features are initially adjusted. As the baseline weight for temperature-related features, This is the weighting gain coefficient for vibration versus temperature characteristics. The vibration interference environmental factor, This represents the weighting attenuation coefficient of low temperature on temperature characteristics. The low-temperature interference environment coefficient, This represents the weighting gain coefficient for the temperature characteristics caused by electromagnetic interference. Electromagnetic interference environmental factor, The weights of the current-type features are initially adjusted. As the benchmark weight for current-type features, This is the weighted attenuation coefficient of vibration on the current characteristic. This is the weighting attenuation coefficient for electromagnetic interference on the current characteristics. This is the weighting attenuation coefficient for voltage fluctuations on current characteristics. The voltage fluctuation environmental factor. The weights of the voltage-type features are initially adjusted. As the benchmark weight for voltage-type features, This represents the weighting attenuation coefficient of voltage fluctuations on voltage characteristics. The dust fluctuation environmental coefficient. This represents the weighting gain coefficient for humidity on voltage characteristics. The humidity fluctuation environmental coefficient, This is the weighting attenuation coefficient of voltage fluctuation on voltage characteristics.

[0038] In one embodiment of the present invention, the environmental coefficient is determined by the following formula:

[0039] In the formula, The vibration characteristic coupling coefficient, To measure vibration acceleration, As the reference vibration acceleration threshold, The coupling coefficient of the low-temperature sensor response. Based on ambient temperature, To measure the ambient temperature in the cabin, The electromagnetic coupling coefficient is... To measure the electromagnetic radiation intensity, The electromagnetic coupling coefficient is... The coupling coefficient for dust insulation performance. To measure the dust concentration, As the baseline dust concentration threshold, The coupling coefficient is the coefficient for humidity-induced insulation aging. To measure the relative humidity, The baseline relative humidity is The voltage fluctuation sampling accuracy coupling coefficient is... To measure the voltage fluctuation value of the power grid, This is the rated voltage of the transformer.

[0040] According to another embodiment, a computer-readable storage medium is also provided, on which a computer program is stored, which, when executed in a computer, causes the computer to perform a combination Figure 1 The method described.

[0041] According to another embodiment, an electronic device is also provided, including a memory and a processor, wherein the memory stores executable code, and when the processor executes the executable code, it implements a combination... Figure 1 The method described.

[0042] The various embodiments in this invention are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the apparatus embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions of the method embodiments.

[0043] Those skilled in the art will recognize that, in one or more of the examples above, the functions described in this invention can be implemented using hardware, software, firmware, or any combination thereof. When implemented in software, these functions can be stored in a computer-readable medium or transmitted as one or more instructions or code on a computer-readable medium.

[0044] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made on the basis of the technical solution of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for fault identification of a dry-type transformer of a fan engine, characterized in that, include: Obtain the temperature change curve, current change curve, and voltage change curve of the dry-type transformer in the wind turbine nacelle; Based on the temperature change curve, the current change curve, and the voltage change curve, temperature influence data, current influence data, and voltage influence data are determined sequentially; wherein, the temperature influence data includes the temperature rise rate, peak temperature, and temperature fluctuation amplitude; the current influence data includes the current amplitude abrupt change value, harmonic distortion rate, and current imbalance; and the voltage influence data includes the voltage amplitude fluctuation amplitude, voltage drop duration, and three-phase voltage imbalance. The temperature influence data, the current influence data, and the voltage influence data are subjected to feature extraction to obtain temperature features, current features, and voltage features in sequence. The temperature feature, current feature, and voltage feature are fused according to the environmental impact ratio to obtain a fusion matrix; wherein the environmental impact ratio is determined based on the environmental parameters of the dry-type transformer in the wind turbine nacelle. The fusion matrix is ​​input into a preset first hybrid model to obtain the first fault identification result.

2. The method according to claim 1, characterized in that, The preset first hybrid model includes a first input layer, a one-dimensional CNN spatial feature extraction layer, a BiLSTM temporal feature mining layer, a feature concatenation layer, and a first fully connected output layer connected in sequence. The first input layer is used to receive the fusion matrix. The one-dimensional CNN spatial feature extraction layer is used to extract the local spatial correlation features of the fault features in the fusion matrix. The BiLSTM temporal feature mining layer is used to extract the bidirectional temporal dependency features of the fault features in the fusion matrix. The feature concatenation layer receives the local spatial correlation features and the bidirectional temporal dependency features and outputs a spatiotemporal fusion feature matrix. The first fully connected output layer receives the spatiotemporal fusion feature matrix and outputs the first fault identification result.

3. The method according to claim 1, characterized in that, After inputting the fusion matrix into a preset first hybrid model to obtain a first fault identification result, the method further includes: The fusion matrix is ​​input into a preset second hybrid model to obtain a second fault identification result; Based on the first fault identification result and the second fault identification result, a third fault identification result is determined; Both the first fault identification result and the second fault identification result include fault category and confidence level.

4. The method according to claim 3, characterized in that, The preset second hybrid model includes a second input layer, a TCN long-time-series feature extraction layer, a multi-head attention weight allocation layer, a feature fusion layer, and a second fully connected output layer connected in sequence. The second input layer is used to receive the fusion matrix. The TCN long-time-series feature extraction layer is used to extract long-time-series causal correlation features of fault features in the fusion matrix. The multi-head attention weight allocation layer is used to allocate attention weights to the long-time-series causal correlation features to focus on key fault features. The feature fusion layer receives the long-time-series causal correlation features after weight allocation and outputs a weighted fusion feature matrix. The second fully connected output layer receives the weighted fusion feature matrix and outputs the second fault identification result.

5. The method according to claim 4, characterized in that, The step of determining a third fault identification result based on the first fault identification result and the second fault identification result includes: When the fault categories determined by the first fault identification result and the second fault identification result are the same, the first fault identification result or the second fault identification result is replaced with the third fault identification result. When the fault categories determined by the first fault identification result and the second fault identification result are different, the result with higher confidence between the first fault identification result and the second fault identification result is selected as the third fault identification result. The fault categories include inter-turn short circuit, inter-layer breakdown, local overheating of windings, abnormal heating of iron core, neutral point offset, and grounding imbalance.

6. The method according to claim 1, characterized in that, The environmental impact ratio includes preliminary correction weights for temperature-related features, current-related features, and voltage-related features. The environmental impact ratio is determined using the following formula: In the formula, The weights of the temperature-related features are initially adjusted. As the benchmark weight for temperature-related features, This is the weighting gain coefficient for the vibration-temperature characteristics. The vibration interference environmental factor, This represents the weighting attenuation coefficient of low temperature on temperature characteristics. The low-temperature interference environment coefficient, This represents the weighting gain coefficient for the temperature characteristics caused by electromagnetic interference. Electromagnetic interference environmental factor, The weights of the current-type features are initially adjusted. As the benchmark weight for current-type features, This is the weighted attenuation coefficient of vibration on the current characteristic. This is the weighting attenuation coefficient for electromagnetic interference on the current characteristics. This is the weighting attenuation coefficient for voltage fluctuations on current characteristics. The voltage fluctuation environmental factor. The weights of the voltage-type features are initially adjusted. As the benchmark weight for voltage-type features, This represents the weighting attenuation coefficient of voltage fluctuations on voltage characteristics. The dust fluctuation environmental coefficient. This represents the weighting gain coefficient for humidity on voltage characteristics. The humidity fluctuation environmental coefficient, This is the weighting attenuation coefficient of voltage fluctuation on voltage characteristics.

7. The method according to claim 6, characterized in that, The environmental coefficient is determined by the following formula: In the formula, The vibration characteristic coupling coefficient, To measure vibration acceleration, As the reference vibration acceleration threshold, The coupling coefficient of the low-temperature sensor response. Based on ambient temperature, To measure the ambient temperature in the cabin, The electromagnetic coupling coefficient is... To measure the electromagnetic radiation intensity, The electromagnetic coupling coefficient is... The coupling coefficient for dust insulation performance. To measure the dust concentration, As the baseline dust concentration threshold, The coupling coefficient is the coefficient for humidity-induced insulation aging. To measure the relative humidity, The baseline relative humidity is The voltage fluctuation sampling accuracy coupling coefficient is... To measure the voltage fluctuation value of the power grid, This is the rated voltage of the transformer.

8. A fault identification device for a dry-type transformer in a wind turbine nacelle, characterized in that, include: The acquisition unit is configured to acquire the temperature change curve, current change curve, and voltage change curve of the dry-type transformer in the wind turbine nacelle. The first data processing unit is configured to determine temperature influence data, current influence data, and voltage influence data sequentially based on the temperature change curve, the current change curve, and the voltage change curve; wherein the temperature influence data includes the temperature rise rate, peak temperature, and temperature fluctuation amplitude; the current influence data includes the current amplitude abrupt change value, harmonic distortion rate, and current imbalance; and the voltage influence data includes the voltage amplitude fluctuation amplitude, voltage drop duration, and three-phase voltage imbalance. The second data processing unit is configured to extract features from the temperature influence data, the current influence data and the voltage influence data to obtain temperature features, current features and voltage features in sequence. The third data processing unit is configured to fuse the temperature feature, the current feature, and the voltage feature according to the environmental influence ratio to obtain a fusion matrix; wherein the environmental influence ratio is determined based on the environmental parameters of the dry-type transformer of the wind turbine nacelle. The fourth data processing unit is configured to input the fusion matrix into a preset first hybrid model to obtain a first fault identification result.

9. An electronic device, characterized in that, It includes a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the method as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, It stores a computer program that, when executed in a computer, causes the computer to perform the method described in any one of claims 1-7.

Images