Method and device for constructing multiple neural network system for transformer thermal defect identification

Abstract

The invention discloses a method and device for constructing a multiple neural network system for transformer thermal defect identification, and the method comprises the steps: obtaining each excitation parameter set, a temperature value data set on each monitoring point, and a temperature value data set on each feature point of a transformer under each working condition, constructing a first neural network on the basis of each excitation parameter set and the temperature value data set on each feature point under each working condition, constructing a second neural network on the basis of the temperature value data set on each monitoring point and the temperature value data set on each feature point, and combining the first neural network and the second neural network to obtain a first neural network and a second neural network; and a multi-neural network system is obtained. Through the method and the device provided by the embodiment of the invention, a multiple neural network system can be obtained to realize rapid and high-precision transformer temperature field inversion and fault thermal defect identification.

Description

Technical Field

[0001] The present invention relates to the technical field of power equipment simulation, and more particularly, to a method and device for constructing a multiple neural network system for transformer thermal defect identification. Background Art

[0002] The complexity of the transformer structure and the difference in material parameters result in uneven internal temperature distribution. The highest internal temperature is the hot spot temperature. Excessive hot spot temperature will damage the insulation system and affect the life of the transformer. Due to the uncertainty of the hot spot temperature position of the transformer, the fiber optic measurement method often does not obtain the hot spot. The commonly used method in current production practice is the empirical formula method. By measuring the top oil temperature and then estimating the hot spot temperature according to the IEC guidelines. However, this method has large errors in cases of complex environmental factors, large load fluctuations, overload, etc. During the operation of the transformer, iron loss and copper loss will be generated. The losses are converted into heat energy and transferred outward, causing the transformer to continuously heat up and then the temperature to rise. The heat transfer process in the transformer is roughly as follows: The heat generated by the iron loss and copper loss first causes the temperature of the iron core and winding to gradually rise. As the temperature of the iron core and winding rises, the temperature difference between them and the transformer oil gradually increases. At this time, the winding and the iron core transfer a part of the heat to the transformer oil. The oil at the bottom of the transformer has a slower temperature rise rate than the oil at the top. Similar to the siphon effect, the top oil will circulate heat with the bottom oil, guiding the transformer oil to dissipate heat to the environment through the radiator. When the temperatures of the winding and the iron core gradually tend to be stable, all the heat generated by the two is dissipated to the environment through the oil flow in the transformer box and the radiator, and the transformer reaches a dynamic thermal equilibrium state. There are literature reports that the external shell temperature measurement points can be selected as characteristic quantities according to the streamline of the transformer oil flow.

[0003] The hot-spot temperature of the winding is an important indicator of oil-immersed power transformers, which affects the load capacity and the degree of insulation aging of the transformer. The hot-spot temperature is related to the structure, size, cooling method, external environmental parameters, and load operating conditions of the transformer. To accurately analyze and predict the hot-spot temperature of the winding, existing methods mainly include empirical thermal models, thermal-path equivalent models, artificial intelligence algorithms, and numerical simulation analysis methods. The thermal equivalent model method is used to convert the internal thermal model of the transformer into a circuit model for calculation. The differential equations corresponding to the empirical thermal model and the thermal-path equivalent model can be discretized into difference equations for rapid calculation, which can meet the requirements of real-time online monitoring. However, some parameter values of such models are difficult to accurately obtain, and the accuracy of calculating the hot-spot temperature is limited in practical applications. In addition, artificial intelligence algorithms and numerical simulation analysis are also effective methods for calculating the hot-spot temperature. Numerical simulation analysis can achieve accurate calculation of the hot-spot temperature by considering in detail the winding structure and loss distribution. However, numerical simulation analysis requires a large amount of calculation. Especially for large-capacity transformers under load fluctuations, it is difficult to achieve real-time tracking calculation of the hot-spot temperature, so this method is difficult to apply to the online monitoring system. Moreover, existing artificial intelligence algorithms have problems such as difficult to obtain accurate solutions when the number of observed values is small and poor model accuracy when the number of observed values is large. Summary of the Invention

[0004] In view of this, the present invention proposes a method and device for constructing a multiple neural network system for transformer thermal defect identification, aiming to solve one or more of the technical problems mentioned in the above background art.

[0005] In a first aspect, an embodiment of the present invention provides a method for constructing a multiple neural network system for transformer thermal defect identification, the method including: obtaining each set of excitation parameters, the temperature value data set at each monitoring point, and the temperature value data set at each feature point under each operating condition of the transformer; constructing a first neural network N based on each set of excitation parameters and the temperature value data set at each feature point under each operating condition of the transformer char-means , and constructing a second neural network N based on the temperature value data set at each monitoring point and the temperature value data set at each feature point input-char ; combining the first neural network N char-means and the second neural network N input-char to obtain a multiple neural network system for transformer thermal defect identification.

[0006] Further, the temperature value datasets at each monitoring point and the temperature value datasets at each characteristic point are obtained in the following manner: parametric modeling is performed on the electromagnetic-temperature-fluid multi-physics field coupling simulation model of the transformer winding, and based on the finite element method, the electromagnetic-temperature-fluid multi-physics field coupling simulation model of the transformer winding under each working condition is solved to obtain each sample solution; the proper orthogonal decomposition is used to perform modal decomposition on each sample solution to obtain each order of mode; based on each order of mode, the greedy selection algorithm is used to select each monitoring point and each characteristic point of the transformer; and the temperature value datasets at each monitoring point and each characteristic point are obtained.

[0007] Further, the parametric modeling of the electromagnetic-temperature-fluid multi-physics field coupling simulation model of the transformer winding and the solution of the electromagnetic-temperature-fluid multi-physics field coupling simulation model of the transformer winding under each working condition based on the finite element method to obtain each sample solution include: taking each layer of high-voltage and low-voltage windings and the iron core as heat sources respectively, performing parametric modeling on the electromagnetic-temperature-fluid multi-physics field coupling simulation model of the transformer winding, and the parametric space is represented as follows: ∑(T HAi ×T HBi ×T HCi ×T LAi ×T LBi ×T LCi ×T corei ); where T HAi 、T HBi 、T HCi respectively represent the i-th layer of high-voltage windings of phases A, B, and C, T LAi 、T LBi 、T LCi respectively represent the i-th layer of low-voltage windings of phases A, B, and C, and T corei represents the i-th sub-block heat source in the iron core region; based on Latin hypercube sampling, the electromagnetic-temperature-fluid multi-physics field coupling simulation model of the transformer winding is solved and the samples are verified to obtain each sample solution.

[0008] Further, the working conditions include normal conditions and fault conditions; the excitation parameters include: spatial position, current excitation, voltage excitation, ambient temperature, and fault point.

[0009] In a second aspect, an embodiment of the present invention further provides a device for constructing a multiple neural network system for transformer thermal defect identification, and the device includes: an acquisition unit, configured to acquire each set of excitation parameters, the temperature value datasets at each monitoring point, and the temperature value datasets at each characteristic point under each working condition of the transformer; a construction unit, configured to construct a first neural network N char-means, and a second neural network N is constructed based on the temperature value datasets at the respective monitoring points and the temperature value datasets at the respective characteristic points. input-char ; a combining unit for combining the first neural network N char-means and the second neural network N input-char to obtain a multi-neural network system for identifying thermal defects in a transformer.

[0010] Further, the temperature value datasets at the respective monitoring points and the temperature value datasets at the respective characteristic points are obtained in the following manner: parametric modeling is performed on the electromagnetic-temperature-fluid multi-physical field coupling simulation model of the transformer winding, and based on finite element method, the electromagnetic-temperature-fluid multi-physical field coupling simulation model of the transformer winding under each working condition is solved to obtain respective sample solutions; the proper orthogonal decomposition is used to perform modal decomposition on the respective sample solutions to obtain respective orders of modes; based on the respective orders of modes, the greedy selection algorithm is used to select the respective monitoring points and characteristic points of the transformer; and the temperature value datasets at the respective monitoring points and characteristic points are obtained.

[0011] Further, the parametric modeling of the electromagnetic-temperature-fluid multi-physical field coupling simulation model of the transformer winding and the solving of the electromagnetic-temperature-fluid multi-physical field coupling simulation model of the transformer winding under each working condition based on finite element method to obtain respective sample solutions include: taking the high-voltage and low-voltage windings of each layer and the iron core as heat sources respectively, performing parametric modeling on the electromagnetic-temperature-fluid multi-physical field coupling simulation model of the transformer winding, and the parametric space is represented as follows: ∑(T HAi × T HBi × T HCi × T LAi × T LBi × T LCi × T corei ); where, T HAi , T HBi , T HCi respectively represent the high-voltage i-th layer windings of phases A, B, and C, T LAi , T LBi , T LCi respectively represent the low-voltage i-th layer windings of phases A, B, and C, and T corei represents the i-th sub-block heat source in the iron core region; based on Latin hypercube sampling, the electromagnetic-temperature-fluid multi-physical field coupling simulation model of the transformer winding is solved and sample verification is performed to obtain respective sample solutions.

[0012] Further, the working conditions include normal working conditions and fault working conditions; the excitation parameters include: spatial position, current excitation, voltage excitation, ambient temperature, and fault point.

[0013] In a third aspect, an embodiment of the present invention further provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the methods provided in the above embodiments are implemented.

[0014] In a fourth aspect, an embodiment of the present invention further provides an electronic device, including: a processor; a memory for storing executable instructions of the processor; the processor is configured to read the executable instructions from the memory and execute the instructions to implement the methods provided in the above embodiments.

[0015] The method and device for constructing a multi-neural network system for transformer thermal defect identification provided by the embodiments of the present invention obtain various excitation parameter sets, temperature value data sets at each monitoring point, and temperature value data sets at each feature point under each operating condition of the transformer. Based on each excitation parameter set and the temperature value data sets at each feature point under each operating condition, a first neural network N char-means is constructed, and based on the temperature value data sets at each monitoring point and the temperature value data sets at each feature point, a second neural network N input-char is constructed, and the first neural network N char-means and the second neural network N input-char are combined to obtain a multi-neural network system. Based on the multi-neural network system constructed according to the above embodiments, fast and high-precision transformer temperature field inversion and fault thermal defect identification can be realized. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 FIG. shows an exemplary flowchart of a method for constructing a multi-neural network system for transformer thermal defect identification according to an embodiment of the present invention;

[0017] Figure 2 FIG. shows a schematic diagram of constructing a reduced-order model of a transformer according to an embodiment of the present invention;

[0018] Figures 3a - 3f FIG. shows a schematic diagram of the first to sixth order modes after constructing a reduced-order model of a transformer according to an embodiment of the present invention;

[0019] Figure 4 FIG. shows a schematic diagram of a multi-neural network model constructed with feature points as a link according to an embodiment of the present invention;

[0020] Figure 5 FIG. shows a schematic structural diagram of a device for constructing a multi-neural network system for transformer thermal defect identification according to an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0021] Reference is now made to the accompanying drawings to describe exemplary embodiments of the present invention. However, the present invention can be implemented in many different forms and is not limited to the embodiments described herein. These embodiments are provided to disclose the present invention in detail and completely, and to fully convey the scope of the present invention to those skilled in the art. The terms in the exemplary embodiments shown in the drawings are not intended to limit the present invention. In the drawings, the same units / components are denoted by the same reference numerals.

[0022] Unless otherwise specified, the terms used herein (including scientific and technical terms) have the ordinary meaning understood by those skilled in the art. Additionally, it can be understood that terms defined in commonly used dictionaries should be construed as having a meaning consistent with the context of their relevant fields, and should not be construed as having an idealized or overly formal meaning.

[0023] Currently, in the operation and maintenance of transformers, it is still very difficult to directly obtain the temperature distribution at each point inside the operating transformer windings through measurement. The winding temperature can be obtained through multi-physics field finite element numerical calculation, but the calculation accuracy and calculation speed still cannot meet the real-time monitoring and early warning requirements of operation and maintenance personnel.

[0024] Generally speaking, a transformer can obtain the current and voltage values on the high and low voltage sides. For those with an oil pump, the flow rate of the pump can be obtained, and for those with a radiator, the heat dissipation wind speed can be obtained. Transformers usually have a top-layer oil temperature sensor installed at the top. For environmental factors, the ambient temperature, wind speed, sunlight conditions, etc. can be obtained.

[0025] Aiming at the problem of difficult real-time monitoring of the temperature inside the transformer windings, the present invention proposes a method for constructing a transformer thermal defect identification model based on the fusion of proper orthogonal decomposition and multiple neural networks, and uses a multiple neural network model to learn a small amount of sensor observation data to reconstruct the transformer temperature field.

[0026] Figure 1 An exemplary flowchart of a method for constructing a multiple neural network system for transformer thermal defect identification according to an embodiment of the present invention is shown.

[0027] As Figure 1 shown, the method includes:

[0028] Step S101: Obtain the set of each excitation parameter, the temperature value dataset at each monitoring point, and the temperature value dataset at each feature point under each working condition of the transformer.

[0029] Furthermore, the working conditions include normal working conditions and fault working conditions;

[0030] The excitation parameters include: spatial position, current excitation, voltage excitation, ambient temperature, and fault point.

[0031] Furthermore, the temperature value datasets at each monitoring point and the temperature value datasets at each characteristic point are obtained in the following manner:

[0032] Perform parametric modeling on the electromagnetic-temperature-fluid multi-physics coupling simulation model of the transformer winding, and based on the finite element method, solve the electromagnetic-temperature-fluid multi-physics coupling simulation model of the transformer winding under each working condition to obtain each sample solution;

[0033] Perform modal decomposition on each sample solution using proper orthogonal decomposition to obtain each order of mode;

[0034] Based on each order of mode, use the greedy selection algorithm to select each monitoring point and each characteristic point of the transformer;

[0035] Obtain the temperature value datasets at each monitoring point and each characteristic point.

[0036] Furthermore, perform parametric modeling on the electromagnetic-temperature-fluid multi-physics coupling simulation model of the transformer winding, and based on the finite element method, solve the electromagnetic-temperature-fluid multi-physics coupling simulation model of the transformer winding under each working condition to obtain each sample solution, including:

[0037] Taking the high-voltage and low-voltage windings of each layer and the iron core as heat sources respectively, perform parametric modeling on the electromagnetic-temperature-fluid multi-physics coupling simulation model of the transformer winding, and the parametric space is expressed as follows:

[0038] ∑(T HAi ×T HBi ×T HCi ×T LAi ×T LBi ×T LCi ×T corei );

[0039] where T HAi 、T HBi 、T HCi respectively represent the high-voltage i-th layer windings of phases A, B, and C, T LAi 、T LBi 、T LCi respectively represent the low-voltage i-th layer windings of phases A, B, and C, and T corei represents the i-th sub-block heat source in the iron core region;

[0040] Based on Latin hypercube sampling, solve the electromagnetic-temperature-fluid multi-physics coupling simulation model of the transformer winding and verify the samples to obtain each sample solution.

[0041] Specifically, various typical defects of the transformer are sorted out, and different working conditions and different fault samples are set up. Parametric models are established for different working conditions and different fault types, and the full models under each working condition are solved based on the finite element method to obtain the solutions of each sample. It should be noted that the calculation models corresponding to different working conditions are different and need to be constructed one by one.

[0042] Based on the finite element method, a coupled simulation model of electromagnetic - temperature - fluid multi - physical fields of the transformer winding is constructed, and the effectiveness of the simulation model is verified through the temperature rise test; then, the influence laws of different overheating positions, the severity of thermal defects, and operating conditions of the winding on the temperature of the local area of the winding are simulated and analyzed; finally, a method for identifying the position and severity of the winding thermal defect is proposed by applying the multi - point temperature mutation analysis of the axial oil ducts of the winding.

[0043] Overheating faults of power transformers can be mainly divided into magnetic circuit faults, circuit faults, and overheating faults caused by other reasons. Magnetic circuit faults are mainly caused by multi - point grounding of the iron core; circuit faults mainly include inter - turn short - circuit faults and poor contact faults. In addition, overheating faults may be caused by local oil duct blockage resulting in poor heat dissipation, but the occurrence probability is relatively small. Regarding the simulation methods of transformer thermal faults, scholars at home and abroad have carried out certain research, proposed some feasible equivalent simulation methods and carried out relevant tests.

[0044] Considering current - induced thermal defects, heat sources are set for different windings on the high - voltage and low - voltage sides, and the three - dimensional electromagnetic field simulation model of the transformer body established is solved using the finite element method. The solutions of different heat sources applied to the coils of the full model are selected as sample snapshots using the finite element method, and then singular value decomposition is performed to obtain all modal solutions. Considering each layer of the high - voltage and low - voltage windings and the iron core as heat sources respectively, parametric modeling is carried out, and the parametric space is expressed as

[0045] Σ(T HAi ×T HBi ×T HCi ×T LAi ×T LBi ×T LCi ×T corei );

[0046] Among them, T HAi 、T HBi 、T HCi respectively represent the i - th layer winding of the high - voltage side of phases A, B, and C, T LAi 、T LBi 、T LCi respectively represent the i - th layer winding of the low - voltage side of phases A, B, and C, and T corei represents the i - th sub - block heat source in the iron core area.

[0047] When i takes values in [1, 4] and each parameter takes 5 groups of parameters, the number of samples reaches 5 35Considering that the full model needs to be used for each calculation, the Latin Hypercube Sampling (LHS) method is adopted for full model calculation.

[0048] Furthermore, proper orthogonal decomposition is used to perform modal decomposition on each sample solution to obtain each order of mode, including:

[0049] A reduced-order model of the target physical field distribution is established based on the thermally induced defect samples, including the basis and corresponding coefficients of the reduced-order model. Figure 2 The schematic diagram of the construction of the transformer reduced-order model according to an embodiment of the present invention is shown. Figures 3a - 3f The schematic diagrams of the first to sixth order modes after the construction of the transformer reduced-order model according to an embodiment of the present invention are shown.

[0050] For example, the reduced-order model of the temperature field can be expressed by the formula:

[0051]

[0052] where t(x) represents the temperature field, n is the number of reduced-order bases, Φ i (x) is the i-th reduced-order basis (i.e., mode), and a i is the coefficient corresponding to the i-th basis.

[0053] This reduced-order model does not necessarily need to be established by the proper orthogonal decomposition POD method and can be constructed by projection methods or the like.

[0054] The basis of the reduced-order model can reflect the main physical characteristics of the transformer.

[0055] Due to the complexity of the transformer, a large amount of calculation time and calculation resources are required. By using the reduced-order model, the calculation time is very short, and a large number of samples can be generated efficiently.

[0056] In the above embodiment, by using proper orthogonal decomposition to perform modal decomposition on each sample solution, the space of the sample library can be compressed.

[0057] Furthermore, based on each order of mode, a greedy selection algorithm is used to select each monitoring point and each characteristic point of the transformer, including:

[0058] (1) Selecting each monitoring point of the transformer based on the greedy selection algorithm

[0059] Consider the problem of placing monitoring sensors. The position of the sensors is selected to obtain the required flow information. In particular, we are concerned with using sensor data to determine the POD modes. In non-implantable measurement and monitoring methods, the monitoring points are located on the surface of the transformer housing. The selection of temperature measurement points has a great influence on the accuracy of temperature field reconstruction. The ideal method is to select characteristic measurement temperature points that can be measured outside the transformer and have a strong correlation with the hot spot temperature of the winding. From the perspective of matrices, consider the singularity or condition number of the inverse matrix. When the matrix is full rank, the modal coefficients can be calculated accurately. In the Gappy-POD framework, the problem is how to maintain the orthogonality of the corresponding modal vectors.

[0060] Based on the multi-physics simulation model of the transformer, starting from the eigenvectors obtained from the simulation, the selection of temperature measurement points is established. In non-implantable measurement and monitoring methods, the monitoring points are located on the surface of the transformer housing. The selection of temperature measurement points has a great influence on the accuracy of temperature field reconstruction. The ideal method is to select characteristic measurement temperature points that can be measured outside the transformer and have a strong correlation with the hot spot temperature of the winding. A technique for determining monitoring points is proposed based on POD through a greedy selection algorithm. This method utilizes the orthogonality of the eigenvectors after singular value decomposition (SVD) and uses the corresponding temperature points in the eigenvectors to determine the temperature measurement points. Based on the greedy algorithm, the temperature measurement points and the eigenvectors reconstruct the temperature field with the minimum error in a certain way.

[0061] (2) Determine each characteristic point inside the transformer based on the greedy selection algorithm

[0062] After calculating several samples, among various samples, spatial geometric partitioning is performed inside or on the boundary of the transformer, and points with large temperature gradients are selected in each region. Multiple values are appropriately taken in the heat concentration regions, and the obtained points are used as characteristic points.

[0063] Let the set of spatial positions of the characteristic points be represented as {L i}. Let the set of spatial positions of the monitoring points (i.e., the monitoring sensors during the actual operation of the transformer) be represented as {O i}.

[0064] Sample data can be obtained based on simulation. However, on the one hand, the numerical simulation calculation time is relatively long (it may take dozens of minutes or even dozens of hours to calculate one working condition), and there are a large number of different working conditions (such as different loads) and different faults (such as short circuit faults, oil circuit blockages, etc.). Therefore, it is difficult to cover all working conditions and the concerned faults within an acceptable time. In the above embodiments, the POD and gappy POD methods are used to construct samples, which greatly reduces the number of full model calculations and can ensure the accuracy.

[0065] Step S102: Based on each set of excitation parameters and the temperature value datasets at each feature point under each operating condition, construct the first neural network N char-means , and based on the temperature value datasets at each monitoring point and the temperature value datasets at each feature point, construct the second neural network N input-char .

[0066] Specifically, denote the normal operating conditions and various types of fault conditions of the transformer as {Cond j}}, and the normal operating condition can be denoted as Cond0; denote different excitation parameters (such as ambient temperature, phase currents, voltages, etc.) as {P k}}.

[0067] Thus, three datasets are obtained:

[0068] (1) Each set of excitation parameters {Cond j}}-{P k}} under each (parameterized) operating condition;

[0069] (2) The temperature values {T k-Li}} at the corresponding feature points;

[0070] (3) The temperature values {T k-Oi}} at the corresponding monitoring points.

[0071] Figure 4 shows a schematic diagram of a multi-neural network model constructed with feature points as the link according to an embodiment of the present invention. As Figure 4 shown, based on the construction process of the artificial neural network, the following are constructed:

[0072] (1) The network model associated with {Cond j}}-{P k}}→{T k-Li}, denoted as the second neural network N input-char ;

[0073] (2) The network model associated with {T k-Li}}→{T k-Oi}}, denoted as the first neural network N char-means .

[0074] The monitored quantity is usually the oil temperature sensor arranged on the upper part of the transformer, or the thermocouple temperature sensor arranged on the core. Transformer faults are usually caused by electromagnetism or blockage of the internal oil ducts. There is a time-scale difference in the physical field between the physical phenomena caused by electromagnetism and the information fed back by the top-layer oil temperature measurement. If the heat generated by electromagnetism is in the millisecond or second level, while the feedback to the top-layer measurement sensor is in the (half)-hour level, even the optical fiber arranged inside is in the minute level. In order to improve the inversion accuracy and eliminate the errors in the process of establishing the correlation between physical fields, the above embodiments propose to construct the temperature quantity at the characteristic point as an intermediate parameter, construct an artificial neural network model between the parameters of the working condition or fault condition and the temperature quantity at the characteristic point, and construct an artificial neural network model between the temperature quantity at the characteristic point and the oil temperature monitoring sensor.

[0075] In the above embodiments, the method of "working condition parameter data set - physical quantity (heat) data set at characteristic points - physical quantity (heat) data set at monitoring points" is adopted to closely associate the working condition with the spatial physical characteristics, retain the spatial physical distribution characteristics, greatly reduce the neglect of physical characteristics in the traditional method of "working condition parameter data set - physical quantity (heat) data set at monitoring points", improve the recognition accuracy, and endow it with physical interpretability.

[0076] Step S103: Combine the first neural network N char-means and the second neural network N input-char to obtain a multi-layer neural network system for transformer thermal defect recognition.

[0077] In the above embodiments, based on each set of excitation parameters and the temperature value data set at each characteristic point under each working condition, the first neural network N char-means is constructed, and based on the temperature value data set at each monitoring point and the temperature value data set at each characteristic point, the second neural network N input-char is constructed, and the first neural network N char-means and the second neural network N input-char are combined to obtain a multi-layer neural network system. Based on the multi-layer neural network system constructed according to the above embodiments, fast and high-precision inversion of the transformer temperature field and identification of fault thermal defects can be realized.

[0078] Figure 5 Fig. shows a schematic structural diagram of a device for constructing a multi-layer neural network system for transformer thermal defect recognition according to an embodiment of the present invention.

[0079] As Figure 5 shown, the device includes:

[0080] An acquisition unit 501, configured to acquire each set of excitation parameters, the temperature value data set at each monitoring point, and the temperature value data set at each characteristic point under each working condition of the transformer;

[0081] A building unit 502, configured to construct a first neural network N based on each set of excitation parameters under each operating condition and the temperature value dataset at each feature point char-means , and construct a second neural network N based on the temperature value dataset at each monitoring point and the temperature value dataset at each feature point input-char ;

[0082] A combining unit 503, configured to combine the first neural network N char-means and the second neural network N input-char to obtain a multiple neural network system for transformer thermal defect identification.

[0083] Furthermore, the temperature value dataset at each monitoring point and the temperature value dataset at each feature point are obtained in the following manner:

[0084] Perform parametric modeling on the transformer winding electromagnetic-temperature-fluid multi-physics field coupling simulation model, and based on the finite element method, solve the transformer winding electromagnetic-temperature-fluid multi-physics field coupling simulation model under each operating condition to obtain each sample solution;

[0085] Perform modal decomposition on each sample solution using proper orthogonal decomposition to obtain each order of mode;

[0086] Based on each order of mode, select each monitoring point and each feature point of the transformer using the greedy selection algorithm;

[0087] Obtain the temperature value dataset at each monitoring point and each feature point.

[0088] Furthermore, perform parametric modeling on the transformer winding electromagnetic-temperature-fluid multi-physics field coupling simulation model, and based on the finite element method, solve the transformer winding electromagnetic-temperature-fluid multi-physics field coupling simulation model under each operating condition to obtain each sample solution, including:

[0089] Taking each layer of high-voltage and low-voltage windings and the iron core as heat sources respectively, perform parametric modeling on the transformer winding electromagnetic-temperature-fluid multi-physics field coupling simulation model, and the parametric space is expressed as follows:

[0090] ∑(T HAi ×T HBi ×T HCi ×T LAi ×T LBi ×T LCi ×T corei );

[0091] where, T HAi , T HBi , T HCi respectively represent the high-voltage i-th layer windings of phases A, B, and C, TLAi , T LBi , T LCi respectively represent the low-voltage i-th layer windings of phases A, B, and C, and T corei represents the i-th sub-block heat source in the iron core area;

[0092] Based on Latin hypercube sampling, the electromagnetic-temperature-fluid multi-physical field coupling simulation model of the transformer winding is solved and the samples are verified to obtain the solutions of each sample.

[0093] Furthermore, the operating conditions include normal conditions and fault conditions;

[0094] The excitation parameters include: spatial position, current excitation, voltage excitation, ambient temperature, and fault point.

[0095] In the above embodiment, based on each set of excitation parameters under each operating condition and the temperature value data set at each characteristic point, the first neural network N char-means is constructed, and based on the temperature value data set at each monitoring point and the temperature value data set at each characteristic point, the second neural network N input-char is constructed, and the first neural network N char-means and the second neural network N input-char are combined to obtain a multi-neural network system. Based on the multi-neural network system constructed in the above embodiment, fast and high-precision inversion of the transformer temperature field and identification of fault thermal defects can be realized.

[0096] It should be noted that when the device provided in the above embodiment realizes its functions, only the above-mentioned division of each functional module is used for illustration. In actual application, the above functions can be allocated to different functional modules according to needs, that is, the internal structure of the device is divided into different functional modules to complete all or part of the functions described above. In addition, the device provided in the above embodiment and the method embodiment belong to the same concept, and the specific implementation process can be seen in the method embodiment, which will not be repeated here.

[0097] The embodiment of the present invention also provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the method for constructing a multi-neural network system for transformer thermal defect identification provided in each of the above embodiments is realized.

[0098] The embodiment of the present invention also provides an electronic device, including: a processor; a memory for storing processor-executable instructions; the processor is used to read the executable instructions from the memory and execute the instructions to realize the method for constructing a multi-neural network system for transformer thermal defect identification provided in each of the above embodiments.

[0099] The present invention has been described with reference to a few embodiments. However, as is well known to those skilled in the art, other embodiments equivalent to those disclosed above of the present invention equally fall within the scope of the present invention as defined by the appended patent claims.

[0100] Generally, all terms used in the claims are to be interpreted according to their ordinary meaning in the technical field, unless otherwise clearly defined therein. All references to "a / the [device, component, etc.]" are to be construed openly as at least one instance of the device, component, etc., unless otherwise expressly stated. The steps of any method disclosed herein need not be performed in the exact order disclosed, unless expressly stated.

[0101] Those skilled in the art will appreciate that embodiments of the present invention may be provided as a method, system, or computer program product. Accordingly, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) having computer-usable program code embodied therein.

[0102] The present invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each flow and / or block of the flowchart illustrations and / or block diagrams, and combinations of flows and / or blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to the processors of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing device to produce a machine, such that the instructions executed by the processors of the computer or other programmable data processing device create means for implementing the functions specified in Figure 1 one or more of the flows Figure 1 or blocks or combinations of blocks specified.

[0103] These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means that implement the functions specified in Figure 1 one or more of the flows Figure 1 or blocks or combinations of blocks specified.

[0104] These computer program instructions can also be loaded onto a computer or other programmable data processing apparatus, so that a series of operation steps are performed on the computer or other programmable apparatus to generate a computer-implemented process, and thus the instructions executed on the computer or other programmable apparatus provide steps for implementing the functions specified in one process or multiple processes and / or one block or multiple blocks. Figure 1 one process or multiple processes and / or Figure 1 steps for implementing the functions specified in one block or multiple blocks.

[0105] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them. Although the present invention has been described in detail with reference to the above embodiments, those of ordinary skill in the art should understand that: modifications or equivalent replacements can still be made to the specific implementation manners of the present invention, and any modifications or equivalent replacements that do not depart from the spirit and scope of the present invention shall be covered by the protection scope of the claims of the present invention.

Claims

1. A method for constructing a multi-neural network system for transformer thermal defect identification, characterized in that The method includes: Obtaining each set of excitation parameters, the temperature value data sets at each monitoring point, and the temperature value data sets at each characteristic point under each operating condition of the transformer; Construct the first neural network N based on each set of excitation parameters under each of the working conditions and the temperature value data set at each of the characteristic points char-means , and construct the second neural network N based on the temperature value data set at each of the monitoring points and the temperature value data set at each of the characteristic points input-char ; Combine the first neural network N char-means and the second neural network N input-char to obtain a multi-neural network system for transformer thermal defect identification.

2. The method according to claim 1, wherein The temperature value data sets at each monitoring point and the temperature value data sets at each characteristic point are obtained in the following manner: Perform parametric modeling on the electromagnetic-thermal-fluid multi-physics field coupling simulation model of the transformer winding, and solve the electromagnetic-thermal-fluid multi-physics field coupling simulation model of the transformer winding under each operating condition based on the finite element method to obtain each sample solution; Perform modal decomposition on the sample solutions using proper orthogonal decomposition to obtain each order of mode; Based on each order of mode, use the greedy selection algorithm to select each monitoring point and each characteristic point of the transformer; Obtain the temperature value data sets at each monitoring point and each characteristic point.

3. The method according to claim 2, wherein The parametric modeling of the electromagnetic-thermal-fluid multi-physics field coupling simulation model of the transformer winding and solving the electromagnetic-thermal-fluid multi-physics field coupling simulation model of the transformer winding under each operating condition based on the finite element method to obtain each sample solution includes: Taking each layer of the high-voltage and low-voltage windings and the iron core as heat sources respectively, perform parametric modeling on the electromagnetic-thermal-fluid multi-physics field coupling simulation model of the transformer winding, and the parametric space is represented as follows: ∑(T HAi ×T HBi ×T HCi ×T LAi ×T LBi ×T LCi ×T corei ); Among them, T HAi , T HBi , T HCi respectively represent the high-voltage i-th layer windings of phases A, B, and C, and T LAi , T LBi , T LCi respectively represent the low-voltage i-th layer windings of phases A, B, and C, and T corei represents the i-th sub-block heat source in the iron core region; Based on Latin hypercube sampling, solve the electromagnetic-thermal-fluid multi-physics field coupling simulation model of the transformer winding and verify the samples to obtain each sample solution.

4. The method according to claim 1, wherein The operating conditions include normal operating conditions and fault conditions; The excitation parameters include: spatial position, current excitation, voltage excitation, ambient temperature, and fault point.

5. An apparatus for constructing a multiple neural network system for transformer thermal defect identification, characterized in that, The device includes: An acquisition unit for obtaining each set of excitation parameters, the temperature value data sets at each monitoring point, and the temperature value data sets at each characteristic point under each operating condition of the transformer; A building unit, configured to build a first neural network N based on each set of excitation parameters under each working condition and the temperature value data set at each feature point char-means , and build a second neural network N based on the temperature value data set at each monitoring point and the temperature value data set at each feature point input-char ; Combined unit for combining the first neural network N char-means and the second neural network N input-char to obtain a multiple neural network system for identifying thermal defects of transformers.

6. The device according to claim 5, characterized in that The temperature value data sets at each monitoring point and the temperature value data sets at each characteristic point are obtained in the following manner: Perform parametric modeling on the electromagnetic-thermal-fluid multi-physics field coupling simulation model of the transformer winding, and solve the electromagnetic-thermal-fluid multi-physics field coupling simulation model of the transformer winding under each operating condition based on the finite element method to obtain each sample solution; Perform modal decomposition on the sample solutions using proper orthogonal decomposition to obtain each order of mode; Based on each order of mode, use the greedy selection algorithm to select each monitoring point and each characteristic point of the transformer; Obtain the temperature value data sets at each monitoring point and each characteristic point.

7. The device according to claim 6, characterized in that, The parametric modeling of the electromagnetic-thermal-fluid multi-physics field coupling simulation model of the transformer winding and solving the electromagnetic-thermal-fluid multi-physics field coupling simulation model of the transformer winding under each operating condition based on the finite element method to obtain each sample solution includes: Taking each layer of the high-voltage and low-voltage windings and the iron core as heat sources respectively, perform parametric modeling on the electromagnetic-thermal-fluid multi-physics field coupling simulation model of the transformer winding, and the parametric space is represented as follows: ∑(T HAi ×T HBi ×T HCi ×T LAi ×T LBi ×T LCi ×T corei ); Among them, T HAi , T HBi , T HCi respectively represent the i-th layer windings of the high voltage of phases A, B, and C. T LAi , T LBi , T LCi respectively represent the i-th layer windings of the low voltage of phases A, B, and C. T corei represents the i-th block heat source in the iron core area; Based on Latin hypercube sampling, solve the electromagnetic-thermal-fluid multi-physics field coupling simulation model of the transformer winding and verify the samples to obtain each sample solution.

8. The device according to claim 5, characterized in that The operating conditions include normal operating conditions and fault conditions; The excitation parameters include: spatial position, current excitation, voltage excitation, ambient temperature, and fault point.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method according to any one of claims 1-4.

10. An electronic device, comprising: A processor; A memory for storing executable instructions of the processor; The processor is configured to read the executable instructions from the memory and execute the instructions to implement the method according to any one of claims 1-4.

Images