Method for rapidly calculating winding temperature of oil-immersed transformer based on convolutional neural network

Abstract

The invention relates to an oil-immersed transformer winding temperature rapid calculation method based on a convolutional neural network, and belongs to the technical field of transformer winding temperature control. Heat source density, oil flow speed and environment temperature of a transformer are used as input characteristic quantities, and winding temperature is used as output characteristic quantities; constructing a data set for convolutional neural network regression training; dividing the data set into a training set and a test set, and constructing a winding temperature rapid calculation model; inputting the data set into the winding temperature rapid calculation model for training and testing, and finally outputting the winding temperature rapid calculation model with the minimum calculation result error; and inputting the heat source density, the oil flow speed and the environment temperature into the winding temperature rapid calculation model with the minimum calculation result error, and calculating the temperature distribution of the oil-immersed transformer winding under different working conditions. According to the method, the real-time requirement of oil-immersed transformer winding temperature calculation in digital twinning application can be met.

Description

Technical Field

[0001] The invention relates to a method for quickly calculating the temperature of an oil-immersed transformer winding based on a convolutional neural network, and belongs to the technical field of transformer winding temperature control. Background Art

[0002] In the critical period of the integration and development of energy and digital revolution, the digital transformation of power system is very important. Among them, the safe and reliable operation of power transmission and transformation equipment is an important condition for maintaining the stable operation of power grid, and power transformer, as one of the key equipment of power transmission and transformation equipment, needs to ensure its continuous and stable operation. During the operation of oil-immersed transformer, the heat generated by the winding is mainly cooled by circulating transformer oil. If there are problems such as blockage of cooling circuit, the temperature distribution of winding is uneven, local overheating will occur, thus affecting the service life of transformer; if the winding temperature exceeds the allowable value, it will seriously affect the insulation life of transformer. Therefore, the calculation and analysis of transformer winding temperature is of vital importance to the design, maintenance and safe and reliable operation of transformer.

[0003] A lot of research has been carried out on transformer winding temperature at home and abroad. At present, the main methods for obtaining transformer winding temperature information are: experimental measurement method, thermal circuit model method and numerical calculation method. The experimental measurement method requires the installation of measuring thermocouples or distributed optical fiber sensors at the corresponding positions of the transformer; the thermal circuit model method mainly analyzes the internal thermal process of the transformer and establishes a thermal circuit model of the hot spots of the transformer winding, thereby calculating the hot spot characteristics of the transformer winding; the numerical calculation method generally models and divides the transformer entity model first, then applies the corresponding boundary conditions, and finally solves and post-processes.

[0004] Although the above-mentioned traditional transformer winding temperature calculation method can take into account various boundary conditions and parameters during the operation of the oil-immersed transformer, it also consumes a lot of time cost while obtaining high-precision data. For the complex structure of the oil-immersed transformer flow-heat coupling model, it may take several hours or even days, and each time it can only be calculated based on a specific working condition and parameters. Therefore, if the flow-heat coupling of the oil-immersed transformer winding under various working conditions is to be calculated, a large amount of model data needs to be calculated. The workload is too large and the efficiency is too low. It is impossible to monitor in real time or quickly predict and evaluate the internal temperature changes of the oil-immersed transformer, and it cannot meet the requirements of digital twin technology for real-time monitoring of the oil-immersed transformer. Summary of the invention

[0005] In order to solve the above technical problems, the present invention provides a method for quickly calculating the winding temperature of an oil-immersed transformer based on a convolutional neural network. The technical solution adopted by the present invention is as follows:

[0006] A fast calculation method for oil-immersed transformer winding temperature based on convolutional neural network includes the following steps:

[0007] The heat source density, oil flow velocity and ambient temperature of the oil-immersed transformer are used as input feature quantities, and the winding temperature of the oil-immersed transformer is used as the output feature quantity to construct a data set for convolutional neural network regression training.

[0008] The data set used for convolutional neural network regression training is divided into a training set and a test set, and a fast calculation model for oil-immersed transformer winding temperature based on convolutional neural network is constructed; the data set used for convolutional neural network training is input into the fast calculation model for oil-immersed transformer winding temperature for training and testing, and finally the fast calculation model for oil-immersed transformer winding temperature with the smallest calculation result error is output;

[0009] The load loss is calculated by actually measuring the current value of the oil-immersed transformer winding, and the heat source density is obtained by dividing the load loss of the oil-immersed transformer by the winding volume. The oil flow velocity and ambient temperature of the oil-immersed transformer are obtained through instrument monitoring. The heat source density, oil flow velocity and ambient temperature are input into a fast calculation model of the oil-immersed transformer winding temperature with the smallest calculation result error, and the temperature distribution of the oil-immersed transformer winding under different working conditions is calculated.

[0010] Preferably, when constructing a fast calculation model for the oil-immersed transformer winding temperature, the parameters of the convolutional neural network in the fast calculation model for the oil-immersed transformer winding temperature are set, the number of convolution layers is set to 1, 3, 5, 8 and 10, respectively, the activation function is set to Swish and ReLu functions, the number of iterations is set to 5, 8 and 10, respectively, the learning rates are set to 0.01, 0.03 and 0.05, respectively, and the optimization function is set to Adam and Nadam.

[0011] Beneficial effects of the present invention:

[0012] Compared with the finite element method for calculating transformer winding temperature, the winding temperature rapid calculation model of the present invention increases the winding temperature calculation speed by about 180 times. Compared with single-point temperature measurement, the oil-immersed transformer winding temperature rapid calculation model of the present invention can obtain the overall distribution state of the oil-immersed transformer winding temperature. Compared with distributed temperature measurement, the oil-immersed transformer winding temperature rapid calculation model of the present invention is more economical. The present invention can meet the real-time requirements of oil-immersed transformer winding temperature calculation in digital twin applications, and provides strong technical support for the evaluation of the operating status of oil-immersed transformers in digital twin applications. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0014] Figure 1 This is a flow chart of a method for quickly calculating winding temperature according to a first embodiment of the present invention;

[0015] Figure 2 It is a structural schematic diagram of an oil-immersed transformer according to a second embodiment of the present invention;

[0016] In the figure, 1 is the oil tank, 2 is the oil outlet, 3 is the oil inlet, 4 is the high-voltage coil, 5 is the low-voltage coil, and 6 is the iron core. DETAILED DESCRIPTION

[0017] The technical solution of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all of the embodiments.

[0018] Embodiment 1

[0019] like Figure 1 As shown, a fast calculation method for oil-immersed transformer winding temperature based on convolutional neural network is proposed. By constructing a fast calculation model for oil-immersed transformer winding temperature based on convolutional neural network under artificial intelligence algorithm, the winding temperature distribution of oil-immersed transformer under different working conditions is calculated. Specifically, the following steps are included:

[0020] Step 1.1: Construct a dataset for convolutional neural network regression training.

[0021] Firstly, a three-dimensional simulation model of the oil-immersed transformer is constructed for the oil-immersed transformer. The three-dimensional simulation model of structural parts including the iron core, high-voltage winding, low-voltage winding and transformer oil tank is established by using Creo software. The three-dimensional simulation model is imported into COMSOL software, and the thermal-fluid coupling calculation is performed on the three-dimensional simulation model by the finite element analysis method.

[0022] The internal heat transfer of oil-immersed transformers follows the conservation of mass, momentum and energy. The specific relationship is shown in formula 1-1:

[0023]

[0024] Where: is the Hamiltonian operator; t is time; v is the fluid velocity vector; ρ is the fluid density, the unit is kg / m 3 ; p is the fluid pressure, the unit is Pa; f is the external force per unit volume of fluid, the unit is N / m 3 ; η is the dynamic viscosity of the fluid, the unit is kg / (m·s); T is the fluid temperature, the unit is K; cp is the specific heat capacity of the fluid, in J / (kg·K); k is the thermal conductivity of the fluid, in W / (m·K); Φ is the heat source in the fluid; S h It is the part of fluid mechanical energy converted into heat energy under the action of fluid viscosity.

[0025] From the above formula, we can judge that the variables related to the temperature field of the oil-immersed transformer are fluid velocity, density, pressure, force, ambient temperature, specific heat capacity and thermal conductivity. When only analyzing the same oil-immersed transformer, its fluid density, pressure, force and other variables that change with the oil-immersed transformer can be regarded as constants. Therefore, considering the influence of various parameters on the winding temperature of the oil-immersed transformer, the heat source density, oil flow velocity and ambient temperature of the oil-immersed transformer are used as input feature quantities, and the winding temperature of the oil-immersed transformer is used as the output feature quantity to construct a data set for convolutional neural network regression training. The data set format is shown in Formula 1-2:

[0026] D={(x i ,T i )},i=1,2…n (1-2)

[0027] Where: x i is a set of input feature variables; T i is the output temperature value of the corresponding group.

[0028] The heat source density is equal to the oil-immersed transformer load loss divided by the winding volume.

[0029] Step 1.2: Construction of a rapid calculation model for oil-immersed transformer winding temperature.

[0030] According to the data set for convolutional neural network training obtained in step 1.1, the number of data sets is about 90,000. The data set is divided into a training set and a test set, and then a fast calculation model for the oil-immersed transformer winding temperature is constructed.

[0031] Firstly, the parameters of the convolutional neural network in the oil-immersed transformer winding temperature fast calculation model are set. The number of convolutional layers is set to 1, 3, 5, 8 and 10 layers respectively, the activation function is set to Swish and ReLu functions, the number of iterations is set to 5, 8 and 10 rounds respectively, the learning rate is set to 0.01, 0.03 and 0.05 respectively, and the optimization function is set to Adam and Nadam.

[0032] Then the data set used for convolutional neural network training is input into the oil-immersed transformer winding temperature rapid calculation model for training and testing. According to the calculation results of the oil-immersed transformer winding temperature rapid calculation model each time, the various parameter values ​​of the calculation model are continuously adjusted, and the calculation model performance of different parameter values ​​is tested to reduce the error of the model calculation results. Comprehensively analyze the parameter settings, and finally output the oil-immersed transformer winding temperature rapid calculation model with the smallest calculation result error.

[0033] Step 1.3: Fast calculation of oil-immersed transformer winding temperature based on convolutional neural network.

[0034] Combined with intelligent sensing technology, the load loss is calculated by the actual measured current value of the oil-immersed transformer winding, and the heat source per unit volume, i.e., the heat source density, is obtained by dividing the load loss by the volume of the oil-immersed transformer winding. The oil inlet flow value (i.e., oil flow velocity) and the external ambient temperature value (i.e., ambient temperature) of the oil-immersed transformer are monitored by the liquid turbine flowmeter and temperature data recorder of the oil-immersed transformer. The heat source density, oil inlet flow value, and external ambient temperature value are input into the oil-immersed transformer winding temperature fast calculation model based on the convolutional neural network, so as to quickly calculate the temperature distribution of the oil-immersed transformer winding under different working conditions.

[0035] Embodiment 2

[0036] The second embodiment of the present invention is an example of applying the oil-immersed transformer winding temperature rapid calculation method described in the first embodiment to calculate the winding temperature of a 110kV oil-immersed transformer. The structure of the 110kV oil-immersed transformer is as follows: Figure 2 As shown, the 110kV oil-immersed transformer includes an oil tank 1, and an oil outlet 2 and an oil inlet 3 are respectively arranged at the upper and lower parts of one side surface of the oil tank 1, and a high-voltage coil 4, a low-voltage coil 5 and an iron core 6 are installed inside the oil tank 1. The winding in the second embodiment of the present invention includes a high-voltage coil 4, a low-voltage coil 5 and an iron core 6.

[0037] Step 2.1, construct a data set for 110kV oil-immersed transformer for convolutional neural network regression training.

[0038] First, a three-dimensional simulation model of a 110kV oil-immersed transformer is constructed. The 110kV oil-immersed transformer uses a model of SZ20-100000 / 110, with a high-voltage and low-voltage side rated voltage ratio of 115 / 37kV, a high-voltage coil 4 rated current of 502A, a rated number of turns of 407, and a low-voltage coil 5 rated current of 900.9A, with a rated number of turns of 227. The internal dimensions of the oil tank 1 are 5080mm long, 1700mm wide, and 3220mm high, the diameter of the core 6 is 770mm, the window height is 1545mm, and the phase spacing is 1370mm. The high-voltage rated loss of the 110kV oil-immersed transformer is 150kW, the low-voltage rated loss is 123kW, and the no-load loss is 49kW. Considering the influence of various parameters on the winding temperature of 110kV oil-immersed transformer, the heat source density, oil flow velocity and ambient temperature of 110kV oil-immersed transformer are taken as input feature quantities, and the winding temperature of 110kV oil-immersed transformer is taken as output feature quantity, and a data set for convolutional neural network regression training is constructed.

[0039] Step 2.2: Construct a fast calculation model for the winding temperature of a 110kV oil-immersed transformer.

[0040] The data set calculated by the finite element method is divided into a training set and a test set, with the training set accounting for 75% and the test set accounting for 25%. For the training set type of the 110kV oil-immersed transformer winding temperature rapid calculation model, the number of iterations is set to 8 rounds, 2842 iterations per round, and a total of more than 20,000 iterations. According to the calculation results of each 110kV oil-immersed transformer winding temperature rapid calculation model, the convolutional neural network parameters are continuously adjusted to reduce the error of the model calculation results. For the data set of the second embodiment, the number of convolutional layers in the convolutional neural network model is set to 5 layers, the convolution kernel size is 3*3, the number of fully connected layers is 1 layer, the activation function is set to the Relu function, the optimization function is set to the Adam function, and the learning rate is set to 0.01.

[0041] Step 2.3: Rapid calculation of 110kV oil-immersed transformer winding temperature based on convolutional neural network.

[0042] According to the construction process of the 110kV oil-immersed transformer winding temperature rapid calculation model, the codes of each process are automatically written, and the codes of each process are synthesized and encapsulated to obtain the 110kV oil-immersed transformer winding temperature rapid calculation model based on convolutional neural network. Then, combined with intelligent sensing technology, the load loss is calculated by the actual measured winding current value, and the loss is divided by the winding volume to obtain the unit volume heat source, that is, the heat source density. The transformer oil inlet flow value and the external ambient temperature value are monitored by the liquid turbine flowmeter and temperature data recorder of the 110kV oil-immersed transformer. The heat source density, oil inlet flow value and external ambient temperature value are input into the 110kV oil-immersed transformer winding temperature rapid calculation model, so as to calculate the temperature distribution of the 110kV oil-immersed transformer winding under different working conditions.

[0043] Taking the high-voltage winding phase A of a 110kV oil-immersed transformer as an example, the temperature distribution of the high-voltage winding phase A is obtained by finite element method. Figure 1 , the temperature distribution of phase A of the high-voltage winding is obtained through the convolutional neural network fast calculation model Figure 2 , and obtain the temperature relative error distribution diagram 3 of the high-voltage winding A phase. By comparing the three figures, it can be seen that the calculation results of the oil-immersed transformer winding temperature fast calculation model based on convolutional neural network in Example 2 of the present invention are basically consistent with the finite element simulation results, the root mean square error (RMSE) of the convolutional neural network fast calculation model is 1.06, the model accuracy (R2) is 0.995, and the relative percentage error is within 2.5%. The calculation accuracy of the oil-immersed transformer winding temperature fast calculation model based on convolutional neural network in Example 2 of the present invention can meet the application requirements in actual engineering, and the calculation speed can meet the real-time requirements of oil-immersed transformer winding temperature calculation in digital twin applications.

[0044] In the embodiments of the present invention, technical features that are not described in detail are all existing technologies or conventional technical means and will not be described in detail here.

[0045] Finally, it should be noted that the above embodiments are only specific implementation methods of the present invention, which are used to illustrate the technical solutions of the present invention, rather than to limit them, and the protection scope of the present invention is not limited thereto. Those skilled in the art should understand that any technician familiar with the technical field can modify the technical solutions recorded in the above embodiments within the technical scope disclosed by the present invention, or can easily think of changes, or make equivalent replacements for some of the technical features therein; and these modifications, changes or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should be covered within the protection scope of the present invention.

Claims

1. A fast calculation method for oil-immersed transformer winding temperature based on convolutional neural network, characterized in that: The following steps are involved: The heat source density, oil flow velocity and ambient temperature of the oil-immersed transformer are used as input feature quantities, and the winding temperature of the oil-immersed transformer is used as the output feature quantity to construct a data set for convolutional neural network regression training. The data set used for convolutional neural network regression training is divided into a training set and a test set, and a fast calculation model for oil-immersed transformer winding temperature based on convolutional neural network is constructed; the data set used for convolutional neural network training is input into the fast calculation model for oil-immersed transformer winding temperature for training and testing, and finally the fast calculation model for oil-immersed transformer winding temperature with the smallest calculation result error is output; The load loss is calculated by actually measuring the current value of the oil-immersed transformer winding, and the heat source density is obtained by dividing the load loss of the oil-immersed transformer by the winding volume. The oil flow velocity and ambient temperature of the oil-immersed transformer are obtained through instrument monitoring. The heat source density, oil flow velocity and ambient temperature are input into a fast calculation model of the oil-immersed transformer winding temperature with the smallest calculation result error, and the temperature distribution of the oil-immersed transformer winding under different working conditions is calculated.

2. The method for quickly calculating the oil-immersed transformer winding temperature based on a convolutional neural network according to claim 1 is characterized in that: When building a fast calculation model for the oil-immersed transformer winding temperature, the parameters of the convolutional neural network in the fast calculation model for the oil-immersed transformer winding temperature are set. The number of convolutional layers is set to 1, 3, 5, 8 and 10 layers, the activation function is set to Swish and ReLu functions, the number of iterations is set to 5, 8 and 10, the learning rates are set to 0.01, 0.03 and 0.05, and the optimization function is set to Adam and Nadam.

3. The method for quickly calculating the oil-immersed transformer winding temperature based on a convolutional neural network according to claim 2 is characterized in that: A three-dimensional simulation model of an oil-immersed transformer including an iron core, a high-voltage winding, a low-voltage winding and a transformer oil tank is established by using Creo software, and the three-dimensional simulation model of the oil-immersed transformer is imported into COMSOL software, and a thermal-fluid coupling calculation is performed on the three-dimensional simulation model of the oil-immersed transformer by using a finite element analysis method.

4. The method for quickly calculating the oil-immersed transformer winding temperature based on a convolutional neural network according to claim 2 is characterized in that: The data set used for convolutional neural network regression training is: D={(x i ,T i )},i=1,2…n Where: x i is a set of input feature variables; T i is the output temperature value of the corresponding group.

5. The method for quickly calculating the oil-immersed transformer winding temperature based on a convolutional neural network according to claim 2 is characterized in that: The oil flow rate is measured by a liquid turbine flowmeter, and the ambient temperature is measured by a temperature data recorder.

6. The method for quickly calculating the oil-immersed transformer winding temperature based on a convolutional neural network according to claim 2 is characterized in that: The number of iteration rounds and the number of iterations per round are set according to the training set type. According to the calculation results of the oil-immersed transformer winding temperature fast calculation model each time, the convolutional neural network parameters are continuously adjusted to reduce the calculation result error of the oil-immersed transformer winding temperature fast calculation model.

Images