Converter transformer valve side winding oil flow temperature rise calculation method based on PCA-FCNN

Abstract

The invention belongs to the field of winding temperature control, and relates to a PCA-FCNN-based converter transformer valve side winding oil flow temperature rise calculation method, which analyzes a thermodynamic conduction process in a valve side winding, determines a characteristic variable having the highest correlation with a temperature field, performs high-correlation characteristic variable parameterization setting by using Latin hypercube sampling, and calculates the oil flow temperature rise of the valve side winding. A valve side winding flow-heat coupling calculation model is established, and finite element calculation software is used for flow-heat coupling calculation; constructing a data set of a calculation model by taking the parameterized characteristic variable as an input characteristic quantity and taking the valve side winding oil flow temperature rise calculated by a finite element as an output characteristic quantity, dividing the data set into a training set and a test set, and generating a valve side winding oil flow temperature rise rapid calculation model with relatively high precision by setting model training parameters; a validation set other than the model data set is set to validate the accuracy of the computational model. According to the method, the required calculation time is shortened by about 300 times compared with the calculation time of a finite element method, and the calculation speed of the oil flow temperature rise of the valve side winding is increased.

Description

Technical Field

[0001] The present invention relates to a method for calculating the oil flow temperature rise of a valve-side winding of a converter transformer based on PCA-FCNN, and belongs to the technical field of transformer winding temperature control. Background Art

[0002] In modern power systems, converter transformers are core components of high-voltage direct current (HVDC) transmission systems. The stability of their operating conditions is directly related to the reliability and safety of the entire power system. During operation, converter transformer windings experience losses due to the current flowing through them, leading to temperature increases. Excessively high winding temperatures not only accelerate the aging of insulation materials and reduce the lifespan of the equipment, but can also cause insulation failures, resulting in serious power outages and significant socioeconomic losses. Therefore, accurately and rapidly calculating converter transformer winding temperature rise is crucial for real-time monitoring of equipment operating status, optimizing equipment design, and developing appropriate maintenance strategies.

[0003] Traditional methods for calculating converter transformer winding temperature rise primarily include analytical methods and finite element methods. Analytical methods are typically based on simplified assumptions and offer faster calculation speeds, but their accuracy is often difficult to guarantee for complex converter transformer winding structures and actual operating conditions. The finite element method, by discretizing the solution region into a finite number of elements, can accurately simulate the complex geometry and boundary conditions of the windings, thereby obtaining more accurate temperature distribution results. However, the finite element method requires a large number of matrix operations and iterative solutions, resulting in a significant computational effort and long computation time, making it difficult to meet the requirements of real-time monitoring and rapid analysis. Summary of the Invention

[0004] To address the above technical issues, the present invention provides a method for calculating the oil flow temperature rise of the valve-side winding of a converter transformer based on PCA-FCNN. In the present invention, PCA refers to principal component analysis, an unsupervised dimensionality reduction method that projects high-dimensional data into a low-dimensional space through linear transformation, retaining the main variance information of the data and removing redundant features. The specific PCA algorithm can be directly called; FCNN refers to a fully connected neural network. The technical solutions adopted by the present invention are as follows:

[0005] The PCA-FCNN-based method for calculating the oil flow temperature rise of the valve-side winding of a converter transformer includes the following steps:

[0006] First, the internal thermodynamic conduction process of the converter transformer valve-side winding was analyzed to identify characteristic variables with high correlation with the temperature field: winding loss, oil flow velocity, and oil inlet temperature. Latin hypercube sampling was used to parameterize the highly correlated characteristic variables, and a flow-heat coupling calculation model for the converter transformer valve-side winding was established. Finite element methods were used to perform the flow-heat coupling calculations.

[0007] The finite element calculation results are then used as the data set for a PCA-FCNN-based calculation model for the oil flow temperature rise of the valve-side winding of a converter transformer. Winding loss, oil flow velocity, and oil inlet temperature are used as input feature quantities, and the oil flow temperature rise of the valve-side winding of the converter transformer is used as the output feature quantity. The data set of the calculation model is divided into a training set and a test set, and the calculation model parameters are optimized to finally generate a PCA-FCNN-based calculation model for the oil flow temperature rise of the valve-side winding of a converter transformer.

[0008] Finally, by setting new samples outside the calculation model data set as the validation set, the input feature quantities in the validation set are input into the calculation model, and the errors between the output feature quantities in the validation set and the calculated values ​​of the calculation model are compared to verify the accuracy of the converter transformer valve-side winding oil flow temperature rise calculation model based on PCA-FCNN.

[0009] Preferably, a two-dimensional simulation model including windings, oil channels, oil baffles, oil inlets and oil outlets is established using CAD software, the established two-dimensional simulation model is imported into COMSOL finite element calculation software, and the two-dimensional simulation model is subjected to flow-heat coupling calculation using the finite element calculation software.

[0010] Preferably, 20 sample operating conditions of the valve-side winding of the converter transformer are randomly obtained as input parameters through Latin hypercube sampling. These sample operating condition input parameters are subjected to flow-heat coupling calculation through finite element calculation software to obtain corresponding temperature rise distribution data for constructing a calculation model data set. The data set format is shown in Formula 1-2:

[0011] D=(x i ,T i ),i=1,2…n (1-2)

[0012] Where: x i is a set of 3-dimensional input feature variables, T i is the output temperature value of the corresponding group.

[0013] Preferably, new samples other than the 20 sample working conditions are set as the validation set to exclude the situation where the test set participates in the model training, resulting in high accuracy of the trained model but poor accuracy of the actual model; and the accuracy of the PCA-FCNN calculation model is quantified by evaluation indicators, including root mean square error rmse, model accuracy r 2 And the relative error percentage error.

[0014] Preferably, when constructing the computational model, the computational model parameters are set, the cumulative variance contribution rate in PCA is 95%, the PCNN includes an input layer, a hidden layer and an output layer, the input layer includes 3 nodes, the number of hidden layers is 3, and the activation function used is ReLU; in the first two hidden layers, L2 weight regularization and Dropout layer random inactivation are introduced respectively; the output layer uses a linear activation function, the computational model is trained by the Adam optimizer, the initial learning rate is set to 0.001, the loss function uses the mean square error, and the maximum number of training rounds is 500.

[0015] Beneficial effects of the present invention:

[0016] The method of the present invention combines finite element calculation, principal component analysis and fully connected neural network methods to obtain a calculation model for the oil flow temperature rise of the valve-side winding of the converter transformer. Under the same conditions, the calculation time required for the calculation model of the oil flow temperature rise of the valve-side winding of the converter transformer based on PCA-FCNN is nearly 300 times shorter than the calculation time of the finite element method, which effectively improves the speed of the calculation of the oil flow temperature rise of the valve-side winding of the converter transformer and can meet the real-time requirements of the oil flow temperature rise calculation of the valve-side winding of the converter transformer in digital twin applications. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] 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:

[0018] Figure 1 This is a flow chart of a method for calculating the oil flow temperature rise of a valve-side winding of a converter transformer according to an embodiment of the present invention;

[0019] Figure 2 Schematic diagram of the structure of the valve-side winding of the converter transformer according to an embodiment of the present invention. DETAILED DESCRIPTION

[0020] The technical solution of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, but not all of the embodiments.

[0021] like Figure 1 、 2 As shown in FIG, a fast calculation method for the oil flow temperature rise of the valve-side winding of a converter transformer based on PCA-FCNN is proposed. By constructing a fast calculation model for the oil flow temperature rise of the valve-side winding under a neural network algorithm, the oil flow temperature rise distribution of the valve-side winding of a converter transformer under different working conditions is calculated. Specifically, the method includes the following steps:

[0022] Step 1.1: Calculate the flow-heat coupling of the valve-side winding of the converter transformer using finite element method.

[0023] When establishing a coupled flow-heat finite element model for the valve-side winding of a converter transformer, an analysis of the thermodynamic conduction process within the valve-side winding revealed that winding loss (P0), oil flow velocity (v0), and oil inlet temperature (T0) were the three variables most highly correlated with the oil flow temperature rise in the valve-side winding. Based on data from the normal operating range of the converter transformer, the parameter ranges were determined as: v0 =

[510] kg / s, P0 = [200-250] kW, and T0 = [313.15-333.15] K. Latin hypercube sampling was used to randomly select 20 sample operating conditions within each parameter range. The finite element calculation software COMSOL was then used to perform a coupled flow-heat calculation for each sample condition, yielding the oil flow temperature rise distribution data for the valve-side winding of the converter transformer.

[0024] First, a 2D simulation model of the winding, oil channel, oil baffle, oil inlet, and oil outlet was created using CAD software. The model was then imported into COMSOL finite element calculation software. Material parameters were set for each component in the 2D simulation model. The thermal conductivity of the transformer oil was 0.11 W / (m·K), and the density was 870 kg / m 3 . Add solid and fluid heat transfer in the physical field, set the inflow and outflow boundaries of the heat source and fluid; add laminar flow in the physical field, set the inlet flow rate and outlet pressure in the fluid domain, where the outlet pressure is 0Pa, and the wall condition in the fluid domain is no-slip; add non-isothermal flow in the multi-physical field, and couple the two physical fields of solid and fluid heat transfer and laminar flow for analysis; set the grid sequence type to physical field controlled grid and the unit size to finer, perform flow-heat coupling calculation on the two-dimensional simulation model through the steady-state analysis in the study, and obtain the oil flow temperature distribution of the valve-side winding of the converter transformer.

[0025] During the thermodynamic conduction process, the Joule heat generated by the valve-side winding conductor of the converter transformer is transferred through the solid-fluid contact interface. The heat convection between the winding and the oil channel satisfies the conservation of mass, momentum, and energy. The specific relationship is shown in Equation 1-1:

[0026]

[0027] 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; c p 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 internal heat source of the fluid; Sh It is the part of the fluid mechanical energy converted into thermal energy under the action of fluid viscous force.

[0028] According to formula (1-1), the characteristic quantities related to the temperature rise of the oil flow in the valve side winding of the converter transformer include winding loss, thermal conductivity, specific heat capacity at constant pressure, density, pressure, force, dynamic viscosity, oil flow velocity, etc. Since the research object is the same valve side winding, the fluid density, pressure, force, thermal conductivity and specific heat capacity at constant pressure are all constant parameters. Therefore, the winding loss, oil inlet temperature and oil flow velocity have a direct impact on the flow-thermal coupling of the oil flow temperature rise of the valve side winding of the converter transformer. These three variables are selected as the input characteristic quantities for PCA-FCNN model training, and the oil flow temperature rise of the valve side winding of the converter transformer is used as the output characteristic quantity.

[0029] Based on the data within the normal operating range of the converter transformer, the parameter value range is determined, as shown in Table 1. P0 is the winding loss, v0 is the oil flow velocity, and T0 is the oil inlet temperature.

[0030] Table 1 Input characteristic parameter value range

[0031] <![CDATA[v0(kg / s)]]> <![CDATA[P0(kW)]]> <![CDATA[T0(K)]]> scope

[510] [200250] [313.15333.15]

[0032] Using Latin hypercube sampling (Latin hypercube sampling is a multidimensional spatial stratified sampling technique that efficiently and evenly covers the sample space. The specific algorithm is well-known and can be directly called in software such as MATLAB or Python), 20 sample operating conditions were randomly selected within the aforementioned parameter range. The specific operating condition data are shown in Table 2 below. The finite element calculation software COMSOL was then used to perform a flow-thermal coupling calculation for each sample condition, and the oil flow temperature rise distribution data for the valve-side winding of the converter transformer was obtained.

[0033] Table 220 sample working condition parameters

[0034]

[0035]

[0036] Step 1.2: Construct a dataset for the PCA-FCNN fast computation model.

[0037] When constructing a PCA-FCNN-based model for calculating the oil flow temperature rise of the valve-side winding of a converter transformer, 18 sample operating conditions were randomly selected as the training set, and the remaining two sample conditions were used as the test set. Both the training and test sets were standardized to eliminate the influence of different dimensions in the parameters. The orthogonal transformation method used in principal component analysis (PCA) was used to project high-dimensional data into a low-dimensional space. This method significantly reduced the number of output layer nodes and the number of model parameters while preserving key model information. The parameters of the fully connected neural network (FCNN) training model were then set, and model optimization methods such as regularization strategies, optimizer configuration, and callback functions were incorporated into the training model.

[0038] Based on the 20 sample operating condition input parameters and corresponding temperature rise distribution data of the converter transformer valve-side winding obtained in step 1.1, a PCA-FCNN fast calculation model dataset is constructed. The dataset format is shown in Equation 1-2:

[0039] D=(x i ,T i ),i=1,2…n (1-2)

[0040] 18 sample working conditions are randomly selected as the training set, and the remaining 2 sample working conditions are used as the test set. In order to eliminate the differences in magnitude and dimension between the feature quantities and improve the accuracy of the training results, the training set and the test set are standardized respectively. The standardization formula is as follows:

[0041]

[0042] Where: X is the data set (the set of all 20 samples is X), μ is the mean of the data set, σ is the standard deviation of the data set, X z Here, the training set and test set are standardized separately, but the standardization principle and formula are the same, so X is collectively referred to as the data set.

[0043] Step 1.3: Build a PCA-FCNN fast calculation model.

[0044] First, the cumulative variance contribution rate is set to 95% in the principal component analysis (PCA), the number of principal components to be retained is determined, and 5% of the low-variance components are discarded. This can filter out some data with poor correlation, maintain the key information of the fast calculation model, and significantly reduce the number of output layer nodes while reducing the number of parameters of the fast calculation model.

[0045] The parameters of a fully connected neural network (FCNN) were then set. The FCNN consists of an input layer, hidden layers, and an output layer. The input layer contains three nodes, corresponding to the winding loss P0, the oil inlet temperature T0, and the oil flow velocity v0, respectively. The number of hidden layers was set to three, with 256, 128, and 64 neurons in each hidden layer, respectively. A ReLU activation function was used to enhance nonlinear representation. L2 weight regularization and a Dropout layer with random dropout were introduced in the first two hidden layers. L2 weight regularization achieves weight decay and prevents overfitting; the Dropout layer randomly drops 20% of the nodes to enhance generalization. The output layer uses a linear activation function, with the number of nodes matching the number of principal components of the temperature field after PCA dimensionality reduction.

[0046] The fast computation model was trained using the Adam optimizer, with an initial learning rate of 0.001 and dynamic learning rate scheduling. After validation loss stagnated for five epochs, the learning rate was automatically decayed by a factor of 0.5, with a lower limit of 1e-6. The mean squared error (MSE) loss function was used, with a maximum number of training epochs of 500. An early stopping mechanism was implemented, halting training if no improvement in model accuracy was observed after 15 epochs.

[0047] Step 1.4: Accuracy evaluation of the PCA-FCNN fast calculation model.

[0048] First, the fast calculation model was evaluated using a test set. The input values ​​in the test set were input into the PCA-FCNN fast calculation model to generate results. The obtained results were restored to the original temperature field scale through inverse transformation and compared with the actual temperature values ​​in the test set. In addition, a new set of samples outside the 20 sample working conditions was set as a validation set. The input feature values ​​of the validation set were winding loss P0 = 222.846 [kW], oil flow velocity v0 = 7.25 [kg / s], and oil inlet temperature T0 = 330.25 [K]. The accuracy of the PCA-FCNN fast calculation model was quantified using evaluation indicators, including root mean square error (rmse), model accuracy r 2 And the relative error percentage error.

[0049] The calculation formula of the root mean square error rmse is as follows:

[0050]

[0051] Where: n is the number of samples; T tr is the true value; T pr is the predicted value.

[0052] Accuracy r 2 The calculation formula is as follows:

[0053] r 2=1-u / v (4)

[0054] Where u is the sum of squares of the difference between the true value and the predicted value, and v is the sum of squares of the difference between the true value and the true mean.

[0055] The relative error percentage calculation formula is as follows:

[0056]

[0057] Where: T tr is the true value; T pr is the predicted value.

[0058] After verification, the PCA-FCNN-based converter transformer valve side winding oil flow temperature rise rapid calculation model has a model accuracy (r 2 ) is 0.974, the root mean square error (RMSE) is 0.73, the maximum relative error is 0.55%, and the average relative error is 0.18%. This model can replace the finite element method to calculate the oil flow temperature rise distribution of the valve-side winding of a converter transformer. Compared with single-point temperature measurement, this rapid calculation model for the oil flow temperature rise of the valve-side winding of a converter transformer can shorten the calculation time by nearly 300 times and quickly determine the overall temperature distribution of the valve-side winding of the converter transformer.

[0059] In the embodiments of the present invention, technical features not described in detail are all existing technologies or conventional technical means and will not be described in detail here.

[0060] 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 person skilled in the art can modify or easily think of changes to the technical solutions described in the above embodiments within the technical scope disclosed by the present invention, or make equivalent replacements for some of the technical features therein; and these modifications, changes or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included in the protection scope of the present invention.

Claims

1. A method for calculating the oil flow temperature rise of the valve-side winding of a converter transformer based on PCA-FCNN is characterized by: The following steps are involved: First, the internal thermodynamic conduction process of the converter transformer valve-side winding was analyzed to identify characteristic variables with high correlation with the temperature field: winding loss, oil flow velocity, and oil inlet temperature. Latin hypercube sampling was used to parameterize the highly correlated characteristic variables, and a flow-heat coupling calculation model for the converter transformer valve-side winding was established. Finite element methods were used to perform the flow-heat coupling calculations. The finite element calculation results are then used as the data set for a PCA-FCNN-based calculation model for the oil flow temperature rise of the valve-side winding of a converter transformer. Winding loss, oil flow velocity, and oil inlet temperature are used as input feature quantities, and the oil flow temperature rise of the valve-side winding of the converter transformer is used as the output feature quantity. The data set of the calculation model is divided into a training set and a test set, and the calculation model parameters are optimized to finally generate a PCA-FCNN-based calculation model for the oil flow temperature rise of the valve-side winding of a converter transformer. Finally, by setting new samples outside the calculation model data set as the validation set, the input feature quantities in the validation set are input into the calculation model, and the errors between the output feature quantities in the validation set and the calculated values ​​of the calculation model are compared to verify the accuracy of the converter transformer valve-side winding oil flow temperature rise calculation model based on PCA-FCNN.

2. The PCA-FCNN-based method for calculating the oil flow temperature rise of the valve-side winding of a converter transformer according to claim 1 is characterized in that: A two-dimensional simulation model including windings, oil channels, oil baffles, oil inlets and oil outlets is established using CAD software, and the established two-dimensional simulation model is imported into COMSOL finite element calculation software, and fluid-thermal coupling calculation is performed on the two-dimensional simulation model using the finite element calculation software.

3. The PCA-FCNN-based method for calculating the oil flow temperature rise of the valve-side winding of a converter transformer according to claim 2 is characterized in that: Through Latin hypercube sampling, 20 sample operating conditions of the converter transformer valve-side winding are randomly obtained as input parameters. These sample operating condition input parameters are used for fluid-thermal coupling calculation using finite element calculation software to obtain the corresponding temperature rise distribution data for constructing the calculation model data set. The data set format is shown in Equation 1-2: D=(x i ,T i ),i=1,2…n (1-2) Where: x i is a set of 3-dimensional input feature variables, T i is the output temperature value of the corresponding group.

4. The method for calculating the oil-immersed transformer winding temperature based on a convolutional neural network according to claim 3, characterized in that: New samples other than the 20 sample working conditions are set as the validation set, and the accuracy of the PCA-FCNN calculation model is quantified by evaluation indicators, including root mean square error rmse, model accuracy r 2 And the relative error percentage error.

5. The PCA-FCNN-based method for calculating the oil flow temperature rise of the valve-side winding of a converter transformer according to claim 1 is characterized in that: When constructing the computational model, the computational model parameters were set. The cumulative variance contribution rate in PCA was 95%. The PCNN consisted of an input layer, a hidden layer, and an output layer. The input layer contained 3 nodes, the number of hidden layers was 3, and the activation function used was ReLU. In the first two hidden layers, L2 weight regularization and Dropout layer random inactivation were introduced respectively. The output layer used a linear activation function. The computational model was trained using the Adam optimizer, the initial learning rate was set to 0.001, the mean square error was used as the loss function, and the maximum number of training rounds was 500.

Images