A method and system for detecting losses in a dry-type transformer of a wind turbine nacelle

Abstract

The present application relates to the technical field of dry-type transformer, and more particularly to a loss detection method for a dry-type transformer of a wind turbine cabin, which comprises the following steps: monitoring the equipment state of the dry-type transformer, and simultaneously obtaining a plurality of physical operation and maintenance parameters according to the equipment state; constructing a physical constraint network, performing deep learning on operation and maintenance historical data, and outputting rated no-load loss and rated load loss according to the physical operation and maintenance parameters; monitoring the cabin environment of the dry-type transformer to obtain cabin environment information; constructing a loss correction model, dynamically correcting the rated no-load loss and the rated load loss according to the cabin environment information, and obtaining corrected no-load loss and corrected load loss; and obtaining comprehensive monitoring loss according to the corrected no-load loss and the corrected load loss, so as to effectively realize accurate and dynamic monitoring of the transformer loss, support intelligent operation and maintenance decision-making, reduce operation and maintenance cost, and improve the energy efficiency of the wind power generation system and the reliability of equipment operation.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of dry-type transformers, and in particular to a loss detection method and system for a dry-type transformer of a wind turbine nacelle. BACKGROUND

[0002] The dry-type transformer of the wind turbine nacelle is one of the core devices in the wind power generation system, and is mainly used for connecting the wind turbine and the power grid to realize the conversion of voltage levels. Compared with the traditional oil-immersed transformer, the dry-type transformer uses air or other gases as the insulation and cooling medium, does not contain liquid insulation materials, has good fireproof performance, is easy to maintain and operate, is green and environmentally friendly, and is very suitable for application scenarios such as wind turbine nacelles with limited space and high environmental requirements. In the operation process of the transformer, the loss mainly includes the no-load loss and the load loss, which are respectively composed of the core loss and the winding copper loss. These losses not only directly affect the overall energy efficiency of the wind power system, but also relate to the safety and operation cost of the equipment. Therefore, accurate detection and dynamic monitoring of the loss of the dry-type transformer are crucial for improving the operation efficiency and equipment reliability of the wind farm.

[0003] At present, the detection of the loss of the dry-type transformer mainly relies on periodic offline detection and online monitoring. Offline detection is usually performed by maintenance personnel using special instruments during equipment downtime, and the data processing is relatively lagging. Online monitoring can obtain parameters such as current, voltage, and temperature through sensors during the operation of the transformer, and estimate the loss by combining mathematical models, but most of them are based on static models or empirical formulas, and lack real-time dynamic correction of environmental factors such as temperature, humidity, and dust in the nacelle, resulting in deviations between the estimated loss and the actual loss. In particular, in the scenario of the wind turbine nacelle with dramatic environmental changes and complex parameters, the existing methods are difficult to accurately reflect the actual situation of the loss changing with the working conditions in a timely manner, and are prone to large errors and lagging abnormal detection, which has been difficult to meet the efficient and accurate operation and maintenance needs of the intelligent wind farm.

[0004] The information disclosed in this BACKGROUND section is only intended to enhance the understanding of the general background of the present disclosure and is not intended to be recognized as prior art to the present disclosure. SUMMARY

[0005] The present application provides a loss detection method and system for a dry-type transformer of a wind turbine nacelle, which can effectively solve the problems in the background art.

[0006] In order to achieve the above-mentioned purpose, the technical solution adopted by the present application is:

[0007] A loss detection method for a dry-type transformer of a wind turbine nacelle, the method comprising:

[0008] The device state of the dry-type transformer is monitored, and meanwhile, a plurality of physical operation parameters are obtained according to the device state;

[0009] A physical constraint network is constructed, and rated no-load loss and rated load loss are output according to the physical operation parameters;

[0010] The cabin environment of the dry-type transformer is monitored to obtain cabin environment information;

[0011] A loss correction model is constructed, and the rated no-load loss and the rated load loss are dynamically corrected according to the cabin environment information to obtain corrected no-load loss and corrected load loss;

[0012] The comprehensive monitoring loss is obtained according to the corrected no-load loss and the corrected load loss.

[0013] Further, the physical operation parameters include a current signal parameter, and the current signal parameter is preprocessed, including:

[0014] The current signal parameter is obtained, and the current signal parameter is subjected to multi-scale decomposition to obtain a plurality of frequency band decomposition coefficients;

[0015] Fundamental wave components and harmonic wave components are extracted from the frequency band decomposition coefficients;

[0016] The fundamental wave components and the harmonic wave components are input into a pre-trained deep learning model to extract time-frequency features of each harmonic wave, and output fundamental wave current amplitude and harmonic wave current amplitude.

[0017] Further, the pre-trained deep learning model is constructed, including:

[0018] A high-frequency harmonic analysis branch is constructed, and time-frequency features of the each harmonic wave are extracted based on a convolutional neural network processing time-frequency features generated by wavelet decomposition;

[0019] A fundamental wave dynamic tracking branch is constructed, and load-related changes of the fundamental wave current amplitude are captured based on a long short-term memory network processing time series signals of wavelet approximation coefficients;

[0020] The convolutional neural network and the long short-term memory network constitute a double-branch hybrid model as the deep learning model, and output the fundamental wave current amplitude and the harmonic wave current amplitude.

[0021] Further, the physical constraint network is constructed, and the rated no-load loss and the rated load loss are obtained according to the physical operation parameter output, including:

[0022] The physical constraint network of the equivalent circuit is constructed based on the physical constraint principle, and the electromagnetic energy transmission and the thermal energy loss process are simulated;

[0023] determining the physical operation parameter and mapping the physical operation parameter as an input variable to the physical constraint network;

[0024] updating the device operation time, outputting the rated no-load loss and rated load loss.

[0025] Further, the comprehensive monitoring loss is obtained according to the corrected no-load loss and corrected load loss, comprising:

[0026] According to the real-time operation load rate of the dry-type transformer, the weight values of the corrected no-load loss and corrected load loss are determined, and the corrected no-load loss and corrected load loss are weighted according to the weight values to obtain the comprehensive monitoring loss.

[0027] Further, a loss correction model is constructed, and the rated no-load loss and rated load loss are dynamically corrected according to the cabin environment information to obtain the corrected no-load loss and corrected load loss, comprising:

[0028] A plurality of environment application correction models are constructed, and each environment application correction model corresponds to the application environment of one of the wind turbine cabins where the dry-type transformer is located.

[0029] The monitored cabin environment information is obtained, and the matched environment application correction model is selected according to the cabin environment information.

[0030] The historical data set corresponding to the scene is loaded to obtain the positive and negative correlation effects of the historical data set on the rated no-load loss and rated load loss and the influence fluctuation range.

[0031] The rated no-load loss and rated load loss are corrected according to the positive and negative correlation effects and influence fluctuation range.

[0032] Further, the environment application correction model is constructed, comprising:

[0033] The operation environment information of the dry-type transformer of the multi-sample wind turbine cabin is collected, and the operation environment information includes cabin temperature, cabin humidity, cabin altitude and dust concentration.

[0034] The historical data set is obtained, and each historical data set includes the rated no-load loss, rated load loss and comprehensive monitoring loss corresponding to the multi-sample.

[0035] The operation environment information is clustered to obtain a plurality of operation environment data sets, and each operation environment data set corresponds to a device operation environment.

[0036] The deviation of the rated no-load loss and the rated load loss in each sample from the comprehensive monitoring loss is calculated, and deep learning is performed on a plurality of the operation and maintenance environment data sets to obtain detection data correction corresponding to each of the device operation and maintenance environments.

[0037] Further, the device detection database is established based on the historical data set and the operation and maintenance environment data set, and the device detection database is updated in real time according to the physical operation and maintenance parameters and the cabin environment information obtained by monitoring.

[0038] A loss detection system of a dry-type transformer in a wind turbine cabin, the system comprising:

[0039] A physical operation and maintenance acquisition module for monitoring the device state of the dry-type transformer and simultaneously obtaining a plurality of physical operation and maintenance parameters according to the device state;

[0040] A rated loss output module for constructing a physical constraint network, performing deep learning on operation and maintenance historical data, and outputting rated no-load loss and rated load loss according to the physical operation and maintenance parameters;

[0041] A cabin environment detection module for monitoring the cabin environment of the dry-type transformer to obtain cabin environment information;

[0042] An environmental loss correction module for constructing a loss correction model and dynamically correcting the rated no-load loss and the rated load loss according to the cabin environment information to obtain corrected no-load loss and corrected load loss;

[0043] A comprehensive loss acquisition module for obtaining comprehensive monitoring loss according to the corrected no-load loss and the corrected load loss.

[0044] Further, the environmental loss correction module comprises:

[0045] A formal correction construction unit for constructing a plurality of environmental application formal corrections, each of which corresponds to the application environment of one of the wind turbine cabins where the dry-type transformer is located;

[0046] A formal correction matching unit for obtaining the monitored cabin environment information and selecting a matched environmental application formal correction according to the cabin environment information;

[0047] A fluctuation influence determination unit for loading a historical data set corresponding to a scene to obtain the positive and negative correlation influence of the historical data set on the rated no-load loss and the rated load loss and the influence fluctuation range;

[0048] A loss correction unit for correcting the rated no-load loss and the rated load loss according to the positive and negative correlation influence and the influence fluctuation range.

[0049] The technical scheme of the present application can achieve the following technical effects:

[0050] The present application can realize accurate and dynamic monitoring of transformer loss, and through the fusion of physical constraint network and deep learning model, combined with real-time collection and dynamic correction of cabin environment information, the problems of low loss estimation accuracy and inability to reflect environmental changes in the prior art are overcome. The technical scheme can significantly improve the accuracy and real-time performance of loss detection, timely detect abnormal conditions, support intelligent operation and maintenance decision-making, reduce operation and maintenance costs, and improve the energy efficiency of the wind power generation system and the reliability of equipment operation.

[0051] The above description is only a summary of the technical scheme of the present application. In order to more clearly understand the technical means of the present application, the specific embodiments of the present application can be implemented according to the content of the specification, and in order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the following specific embodiments of the present application are described. BRIEF DESCRIPTION OF DRAWINGS

[0052] In order to more clearly illustrate the technical scheme in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or the prior art description. Obviously, the drawings in the following description are only some embodiments described in the present application, and for those skilled in the art, other drawings can also be obtained without creative labor on the basis of these drawings.

[0053] Figure 1 Flowchart of the loss detection method for the dry-type transformer of the wind turbine cabin;

[0054] Figure 2 Flowchart of the pre-processing of the current signal parameters;

[0055] Figure 3 Flowchart of the construction of the loss correction model;

[0056] Figure 4 Flowchart of the construction of the environment application model. DETAILED DESCRIPTION

[0057] The technical scheme in the embodiments of the present application will be described in detail below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some embodiments of the present application, not all embodiments.

[0058] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used in the description of the application herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the application. As used herein, the term "and / or" includes any and all combinations of one or more of the associated listed items.

[0059] Embodiment one;

[0060] As Figure 1 The application provides a loss detection method of a wind turbine cabin dry-type transformer, the method comprising:

[0061] S10: monitoring the equipment state of the dry-type transformer, and simultaneously obtaining a plurality of physical operation parameters according to the equipment state;

[0062] S20: constructing a physical constraint network, and outputting rated no-load loss and rated load loss according to the physical operation parameters;

[0063] S30: monitoring the cabin environment of the dry-type transformer, and obtaining cabin environment information;

[0064] S40: constructing a loss correction model, dynamically correcting the rated no-load loss and the rated load loss according to the cabin environment information, and obtaining corrected no-load loss and corrected load loss;

[0065] S50: obtaining comprehensive monitoring loss according to the corrected no-load loss and the corrected load loss.

[0066] Specifically, in the wind turbine cabin, the dry-type transformer operates under complex electrical and environmental conditions. To accurately monitor its loss, the device state is first monitored, and voltage transformers and current transformers are optimally arranged on the primary side and the secondary side, respectively. Temperature sensors are installed in the core and winding area, and a load rate monitoring device is configured in the operating environment. Through these sensors, real-time physical operation and maintenance parameters, including primary side voltage, secondary side current, winding temperature, core temperature, load rate, and running time, are obtained. These parameters are uploaded to the edge computing node or monitoring center as raw data for loss calculation. Based on the equivalent circuit theory of the transformer and the energy conservation relationship, a physical constraint network is constructed to directly calculate and output the rated no-load loss and rated load loss based on the collected operation and maintenance parameters. The no-load loss is mainly composed of core loss, which is related to the core flux density, voltage, and core temperature. The physical constraint network can output the rated no-load loss by inputting the primary side voltage and core temperature. The load loss modeling is mainly composed of winding copper loss, which is related to the load current and winding temperature. The physical constraint network can output the rated load loss by inputting the secondary side current, winding temperature, and load rate. It should be noted that the constraint condition calculation results of some model construction need to satisfy the energy balance equation and electrical characteristic constraints, such as the no-load loss not fluctuating greatly with the current and the load loss being approximately proportional to the square of the current. Through sensor collection, such as cabin temperature, humidity, and dust concentration, the cabin environmental information is used as a correction factor to input the loss correction model. The loss correction model is used to dynamically correct the rated no-load loss and rated load loss based on the cabin environmental information. The model can represent the relationship between the environment and the loss through a regression equation or an empirical coefficient. After obtaining the corrected no-load loss and corrected load loss, the comprehensive monitoring loss is calculated through various methods, such as direct addition or weighted addition.

[0067] Through the technical solution of the present application, accurate and dynamic monitoring of transformer loss is effectively realized. By combining real-time collection and dynamic correction of cabin environmental information, the problems of low estimation accuracy and inability to reflect environmental changes in the prior art are overcome. This technical solution can significantly improve the accuracy and real-time performance of loss detection, timely detect abnormal operating conditions, support intelligent operation and maintenance decisions, reduce operation and maintenance costs, and improve the energy efficiency and reliability of the wind power generation system.

[0068] Further, as shown in Figure 2 The physical operation and maintenance parameters include current signal parameters, and the current signal parameters are preprocessed, including:

[0069] S11: Obtain the current signal parameters and perform multi-scale decomposition on the current signal parameters to obtain a plurality of frequency band decomposition coefficients;

[0070] S12: extracting fundamental component and harmonic component from the frequency band decomposition coefficient;

[0071] S13: inputting the fundamental component and the harmonic component into the pre-trained deep learning model, extracting the time-frequency features of each harmonic, and outputting the fundamental current amplitude and the harmonic current amplitude.

[0072] As a preferred embodiment of the above, high-precision current transformers are installed on the primary side and the secondary side of the dry-type transformer respectively for real-time acquisition of current signal parameters. The acquired current signals often contain fundamental components, harmonic components and random noise, which will cause large errors if directly used for loss calculation, and therefore need to be preprocessed. In order to effectively separate different frequency components, the embodiment adopts multi-scale decomposition technology such as wavelet decomposition, empirical mode decomposition or ensemble empirical mode decomposition, etc. to decompose the acquired current signals. Through multi-scale decomposition, the complex current signals can be divided into several frequency band decomposition coefficients, which respectively represent low frequency, fundamental frequency and high frequency characteristics. After obtaining the multiple frequency band decomposition coefficients, the fundamental component and each harmonic component in the current signal are extracted by using the method of frequency spectrum analysis and band-pass filtering. The fundamental component directly reflects the current amplitude of the transformer under rated operating conditions, while the harmonic component can be used for diagnosing system nonlinear distortion and potential abnormalities. Further, the fundamental component and the harmonic component are input into the pre-trained deep learning model for feature extraction. The model can accurately identify the amplitude, phase and change trend of different harmonics by extracting multi-layer features in time domain and frequency domain.

[0073] Further, the pre-trained deep learning model is constructed, including:

[0074] A high-frequency harmonic analysis branch is constructed, which processes the time-frequency features generated by wavelet decomposition based on a convolutional neural network to extract the time-frequency features of each harmonic;

[0075] A fundamental dynamic tracking branch is constructed, which processes the time series signal of the wavelet approximation coefficient based on a long short-term memory network to capture the load-related changes of the fundamental current amplitude;

[0076] The convolutional neural network and the long short-term memory network form a double-branch hybrid model as the deep learning model, and output the generated fundamental current amplitude and harmonic current amplitude.

[0077] On the basis of the above-mentioned embodiments, after wavelet decomposition of the current signal, a plurality of detail coefficients are obtained, which contain high-frequency information and harmonic components. In order to efficiently extract the time-frequency characteristics, a convolutional neural network is introduced in the high-frequency harmonic analysis branch, which is composed of several convolutional layers, pooling layers and fully connected layers. The convolutional layers are used to automatically extract local features in the time-frequency domain, such as harmonic energy distribution and instantaneous change pattern. The pooling layers are used to reduce dimensionality, retain main features and reduce noise interference. The fully connected layers map the convolutional features to the harmonic amplitude space, and output the amplitude estimation results of each harmonic. The approximation coefficients obtained by wavelet decomposition mainly contain the fundamental component, and the fundamental current is affected by factors such as load rate and temperature, showing significant time correlation. In order to capture this dynamic change feature, a long short-term memory network is used in the fundamental dynamic tracking branch, and through the input gate, forget gate and output gate structure, the long-time dependence of the fundamental current can be effectively captured. The output layer of the network gives the estimation of the fundamental current amplitude, and can track the change of the load working condition over time. The high-frequency harmonic analysis branch and the fundamental dynamic tracking branch are run in parallel to form a double-branch hybrid model. In the output layer stage, the results of the two branches are combined through a feature fusion strategy to obtain the harmonic current amplitudes from the convolutional neural network branch and the fundamental current amplitude from the long short memory network branch. This double-branch architecture not only ensures accurate capture of high-frequency harmonic features, but also realizes continuous tracking of the dynamic trend of the fundamental current.

[0078] Further, a physical constraint network is constructed, and rated no-load loss and rated load loss are obtained according to physical operation parameters output, including:

[0079] Based on the principle of physical constraints, a physical constraint network of the equivalent circuit is constructed to simulate the process of electromagnetic energy transmission and heat loss;

[0080] Determine the physical operation parameters, and map the physical operation parameters as input variables to the physical constraint network;

[0081] Update the device operation time, and output the rated no-load loss and the rated load loss.

[0082] In the embodiment, the physical constraint network is preferably constructed based on the equivalent circuit principle of the dry-type transformer, which at least includes the following parts: core magnetic circuit constraint for describing the establishment process of core magnetic flux under no-load working condition, considering hysteresis loss and eddy current loss, forming a no-load loss model; winding circuit constraint for describing the change rule of winding copper loss under load working condition, mainly related to current square and winding temperature; thermal energy loss constraint introducing heat conduction and convection heat dissipation equation, for reflecting the correction effect of temperature change on core and winding loss. Through at least the above three types of physical constraint relationships, a complete loss calculation framework is formed. After the physical constraint network is constructed, the operation and maintenance parameters monitored are input into the network as simulation variables, including primary side voltage, secondary side current, winding temperature, core temperature, load rate, running time, etc. By mapping these real-time parameters to the input nodes of the physical constraint network, the network can be driven to run simulation. Under the driving of input parameters, the physical constraint network runs simulation, and the following calculations are completed: rated no-load loss: under the condition of open circuit on the secondary side and rated voltage on the primary side, the core loss is calculated through the core magnetic circuit and thermal loss equation, and the rated no-load loss is output; rated load loss: under the condition of rated load current, the copper loss is calculated combined with the winding resistance and temperature correction formula, and the rated load loss is output.

[0083] Further, according to the corrected no-load loss and the corrected load loss, a comprehensive monitoring loss is obtained, including:

[0084] According to the real-time running load rate of the dry-type transformer, the weight values of the corrected no-load loss and the corrected load loss are determined, and the comprehensive monitoring loss is obtained by weighting the corrected no-load loss and the corrected load loss according to the weight values.

[0085] As a preferred embodiment of the above, the system first collects the running load rate of the transformer in real time. The load rate can be calculated by the ratio of the secondary side output power to the rated power. Then, the weight values of the no-load loss and the load loss are determined according to different load rate intervals. The weight values can be set by an empirical formula, or can be dynamically generated by using a piecewise linear function or an interpolation algorithm to ensure that the weight changes continuously with the load rate.

[0086] Further, as shown in Figure 3 , a loss correction model is constructed to dynamically correct the rated no-load loss and the rated load loss according to the cabin environment information, and to obtain the corrected no-load loss and the corrected load loss, including:

[0087] S21: Construct a plurality of environment application correction models, each of which corresponds to the application environment of one of the wind turbine cabins where the dry-type transformer is located;

[0088] S22: Obtain the monitored cabin environment information, and select the matched environment application correction model according to the cabin environment information.

[0089] S23: Load the historical data set of the corresponding scene, obtain the positive and negative correlation influence of the rated no-load loss and rated load loss and the influence fluctuation range;

[0090] S24: Correct the rated no-load loss and rated load loss according to the positive and negative correlation influence and the influence fluctuation range.

[0091] As a preferred embodiment of the above embodiment, according to the operating environment characteristics of different wind turbine cabins, a plurality of environmental application correction forms are constructed, each of which corresponds to a specific environmental scene, such as a high temperature and high humidity environment, a low temperature environment, a high dust environment, and a normal temperature and good ventilation environment. Each environmental correction form can be composed of an empirical formula, statistical analysis or a data-driven model, which is used to describe the correction relationship between the environmental factors (such as temperature, humidity, dust concentration) and the no-load loss and load loss. The temperature and humidity sensors, dust concentration sensors and other sensors arranged inside the cabin are used to obtain the cabin environment information in real time, and compare it with the preset environmental classification threshold, so as to select the most matched environmental application correction form. In order to avoid the deviation caused by a single correction form, the historical data set of the corresponding scene is loaded at the same time as the correction form is called. The historical data set includes the correlation data of the rated no-load loss, the rated load loss and the environmental factors collected in the long-term operation. Through statistical analysis or data mining, the system can obtain the positive or negative correlation influence relationship of the environmental factors on the loss and the fluctuation range of the influence. After selecting the correction form and combining the historical data set analysis, the system corrects the rated no-load loss and the rated load loss, and outputs the corrected no-load loss and the corrected load loss. The result is used for subsequent comprehensive monitoring loss calculation.

[0092] Further, as shown in Figure 4 , the environmental application correction form is constructed, including:

[0093] S211: Collect the operation and maintenance environment information of the dry-type transformer of the multi-sample wind turbine cabin, and the operation and maintenance environment information includes the cabin temperature, cabin humidity, cabin altitude and dust concentration;

[0094] S212: Obtain the historical data set, and each historical data set includes the rated no-load loss, the rated load loss and the comprehensive monitoring loss of the corresponding multi-sample;

[0095] S213: Cluster the operation and maintenance environment information to obtain a plurality of operation and maintenance environment data sets, and each operation and maintenance environment data set corresponds to a device operation and maintenance environment;

[0096] S214: Calculate the deviation of the rated no-load loss, the rated load loss and the comprehensive monitoring loss in each sample, and perform deep learning on the plurality of operation and maintenance environment data sets to obtain the detection data correction corresponding to each device operation and maintenance environment.

[0097] In the long-term operation of the wind turbine cabin in the embodiment, the dry-type transformer under different geographical locations and climate conditions tends to present differentiated operation characteristics. To construct a scientific and effective environmental application correction formal, first, the operation and maintenance environmental information from various samples of wind turbine cabins is collected, including: cabin temperature, cabin humidity, cabin altitude, dust concentration. In addition to environmental data, the historical operation data set of the corresponding transformer is also collected, each data set containing rated no-load loss, rated load loss, and comprehensive monitoring loss. These data form matching samples with corresponding environmental parameters, providing a basis for subsequent correction formal establishment. A large amount of operation and maintenance environmental information collected is input into a clustering algorithm, and the samples are classified to divide into several operation and maintenance environmental data sets, each representing a typical device operation and maintenance environment, such as "high temperature and high humidity", "low temperature and high altitude", "high dust and medium temperature", etc. In each sample data set, the deviation between rated no-load loss, rated load loss, and comprehensive monitoring loss is calculated. Subsequently, each type of environmental data set is input into a deep learning model (such as multi-layer perception MLP, convolutional neural network CNN, or graph neural network GNN, etc.), to learn the correlation between environmental factors and loss deviation, and to obtain correction functions under each environment. In actual application, the system determines the belonging environment cluster according to the real-time collected cabin environmental information, calls the corresponding correction function, and corrects the rated no-load loss and rated load loss to obtain the corrected no-load loss and corrected load loss.

[0098] Further, it also includes: establishing a device detection database based on the historical data set and the operation and maintenance environmental data set, and updating the device detection database in real time according to the monitored physical operation and maintenance parameters and cabin environmental information.

[0099] As a preferred embodiment of the above-mentioned embodiment, a device detection database is further established based on the loss correction model, which is used to uniformly manage and store the historical data set, the operation and maintenance environmental data set, and the correction formal parameters. The database is preferably constructed using a relational database or a time series database to support high-frequency data storage and fast retrieval. The establishment and update of the database not only support the real-time calling of the loss correction model, but also provide a basis for long-term operation trend analysis, provide a reference for anomaly detection and fault warning, and support cross-regional wind turbine comparison analysis to form a big data-based operation optimization scheme.

[0100] Embodiment two;

[0101] Based on the same inventive concept as the loss detection method of a dry-type transformer in a wind turbine cabin in the foregoing embodiment, the present application also provides a loss detection system for a dry-type transformer in a wind turbine cabin, which comprises:

[0102] The physical operation acquisition module monitors the equipment state of the dry-type transformer, and simultaneously obtains a plurality of physical operation parameters according to the equipment state;

[0103] The rated loss output module constructs a physical constraint network, performs deep learning on the operation and maintenance historical data, and outputs the rated no-load loss and the rated load loss according to the physical operation parameters;

[0104] The cabin environment detection module monitors the cabin environment of the dry-type transformer to obtain cabin environment information;

[0105] The environmental loss correction module constructs a loss correction model, dynamically corrects the rated no-load loss and the rated load loss according to the cabin environment information, and obtains the corrected no-load loss and the corrected load loss;

[0106] The comprehensive loss acquisition module obtains the comprehensive monitoring loss according to the corrected no-load loss and the corrected load loss.

[0107] Further, the environmental loss correction module comprises:

[0108] The formal construction unit constructs a plurality of environmental application formalities, and each environmental application formality corresponds to the application environment of one wind turbine cabin where the dry-type transformer is located;

[0109] The formal matching unit obtains the monitored cabin environment information, and selects a matched environmental application formality according to the cabin environment information;

[0110] The fluctuation influence determination unit loads a historical data set corresponding to the scene, and obtains the positive and negative correlation influence of the historical data set on the rated no-load loss and the rated load loss, and the influence fluctuation range;

[0111] The loss correction unit corrects the rated no-load loss and the rated load loss according to the positive and negative correlation influence and the influence fluctuation range.

[0112] The above-mentioned system in the application can effectively realize the loss detection method of the dry-type transformer in the wind turbine cabin, and the technical effects are as described in the above-mentioned embodiments, which will not be repeated here.

[0113] Although the present application has been described in conjunction with specific features and embodiments thereof, it is obvious that various modifications and combinations can be made thereto without departing from the spirit and scope of the application. Accordingly, the specification and drawings are to be regarded simply as illustrative of the present application defined by the appended claims, and it is intended to cover any and all modifications, variations, combinations or equivalents that fall within the scope of the present application. Obviously, those skilled in the art can make various modifications and variations to the present application without departing from the scope of the present application. Thus, if these modifications and variations of the present application belong to the scope of the present application and its equivalents, the present application is intended to include these modifications and variations.

Claims

1. A method of detecting losses in a dry-type transformer of a wind turbine nacelle, characterized in that, The method comprises: Monitoring the equipment state of the dry-type transformer, and simultaneously obtaining a plurality of physical operation parameters according to the equipment state; Constructing a physical constraint network, and outputting rated no-load loss and rated load loss according to the physical operation parameters; Monitoring the cabin environment of the dry-type transformer, and obtaining cabin environment information; Constructing a loss correction model, and dynamically correcting the rated no-load loss and rated load loss according to the cabin environment information to obtain corrected no-load loss and corrected load loss, comprising: Constructing a plurality of environmental application correction models, each of which corresponds to the application environment of one of the wind turbine cabins where the dry-type transformer is located; Obtaining the monitored cabin environment information, and selecting a matching environmental application correction model according to the cabin environment information; Loading a historical data set corresponding to the scene to obtain the positive and negative correlation effects of the historical data set on the rated no-load loss, rated load loss, and the influence fluctuation range; Correcting the rated no-load loss and rated load loss according to the positive and negative correlation effects and the influence fluctuation range; Constructing an environmental application correction model, comprising: Collecting operation environment information of a plurality of sample wind turbine dry-type transformers, the operation environment information including cabin temperature, cabin humidity, cabin altitude, and dust concentration; Obtaining the historical data set, and each of the historical data sets including the rated no-load loss, rated load loss, and comprehensive monitoring loss of the corresponding multiple samples; Clustering the operation environment information to obtain a plurality of operation environment data sets, and each of the operation environment data sets corresponding to a device operation environment; Calculating the deviation of the rated no-load loss, rated load loss, and the comprehensive monitoring loss in each sample, and performing deep learning on the plurality of operation environment data sets to obtain detection data correction corresponding to each device operation environment; Obtaining the comprehensive monitoring loss according to the corrected no-load loss and corrected load loss.

2. The loss detection method of a dry-type transformer for a wind power generator nacelle according to claim 1, characterized by, The physical operation parameters include current signal parameters, and the current signal parameters are preprocessed, comprising: Obtaining current signal parameters, and performing multi-scale decomposition on the current signal parameters to obtain a plurality of frequency band decomposition coefficients; Extracting fundamental wave components and harmonic wave components from the frequency band decomposition coefficients; Inputting the fundamental wave components and harmonic wave components into a pre-trained deep learning model to extract time-frequency features of each harmonic, and outputting fundamental wave current amplitude and harmonic wave current amplitude.

3. The method of claim 2, wherein the method further comprises: Constructing a pre-trained deep learning model, comprising: Constructing a high-frequency harmonic analysis branch to process time-frequency features generated by wavelet decomposition based on a convolutional neural network to extract time-frequency features of each harmonic; Constructing a fundamental wave dynamic tracking branch to process time series signals of wavelet approximation coefficients based on a long short-term memory network to capture load-related changes in the fundamental wave current amplitude; The convolutional neural network and long short-term memory network form a double-branch hybrid model as a deep learning model, and output the generated fundamental wave current amplitude and harmonic wave current amplitude.

4. The method of claim 1, wherein Constructing a physical constraint network, and outputting rated no-load loss and rated load loss according to the physical operation parameters, comprising: The physical constraint network is constructed based on a physical constraint principle to simulate and analogize electromagnetic energy transmission and heat energy loss processes; The physical operation parameter is determined and mapped to the physical constraint network as an input variable; The device operation time is updated, and the rated no-load loss and rated load loss are output.

5. The method of claim 1, wherein, The comprehensive monitoring loss is obtained according to the corrected no-load loss and corrected load loss, including: The weight values of the corrected no-load loss and corrected load loss are determined according to the real-time operation load rate of the dry-type transformer, and the corrected no-load loss and corrected load loss are weighted calculated according to the weight values to obtain the comprehensive monitoring loss.

6. The method of claim 1, wherein Further comprising: The device detection database is established based on the historical data set and operation environment data set, and the device detection database is updated in real time according to the physical operation parameter and cabin environment information obtained by monitoring.

7. A loss detection system for a dry-type transformer of a wind power generator nacelle, characterized in that, The loss detection method of the dry-type transformer in the wind turbine cabin is adopted, and the system comprises: The physical operation acquisition module monitors the device state of the dry-type transformer, and simultaneously obtains a plurality of physical operation parameters according to the device state; The rated loss output module constructs a physical constraint network, performs deep learning on operation history data, and outputs a rated no-load loss and a rated load loss according to the physical operation parameters; The cabin environment detection module monitors the cabin environment of the dry-type transformer to obtain cabin environment information; The environment loss correction module constructs a loss correction model, dynamically corrects the rated no-load loss and rated load loss according to the cabin environment information, and obtains a corrected no-load loss and a corrected load loss; The comprehensive loss acquisition module obtains a comprehensive monitoring loss according to the corrected no-load loss and corrected load loss; The environment loss correction module comprises: The correction formal construction unit constructs a plurality of environment application correction forms, each of which corresponds to the application environment of one of the wind turbine cabins where the dry-type transformer is located; The correction formal matching unit obtains the monitored cabin environment information, and selects a matched environment application correction form according to the cabin environment information; The fluctuation influence determination unit loads the historical data set corresponding to the scene to obtain the positive and negative correlation influences of the historical data set on the rated no-load loss and rated load loss and the influence fluctuation range; The loss correction unit corrects the rated no-load loss and rated load loss according to the positive and negative correlation influences and influence fluctuation range.

Images