Method for predicting transformer hot-spot temperature based on numerical calculation and linear regression model

Abstract

A kind of method for predicting transformer hot spot temperature based on numerical calculation and linear regression model, steps are: determining transformer design object;Construct three-dimensional electromagnetic-temperature-fluid multi-coupling field simulation model;Obtain the size of each temperature measuring point temperature and hot spot temperature of transformer under different load coefficients and ambient temperature;Establish the database used when training and verifying prediction model;Based on neural network, construct the prediction model of distribution transformer winding hot spot temperature;Select cuckoo search algorithm to optimize the initial weight and threshold of neural network model.The method for predicting hot spot temperature based on cuckoo search algorithm optimized neural network model is established, and the numerical size of distribution transformer hot spot temperature under different operating environments is obtained.The present application is beneficial to evaluate the operating state of the transformer, and provides reference data for the subsequent use and operation of distribution transformer.It has practical engineering significance for ensuring the safe operation of distribution transformer and improving the level of rural power supply.

Description

TECHNICAL FIELD

[0001] The present application belongs to the field of transformer hot spot temperature value calculation, and particularly relates to a method for predicting transformer hot spot temperature based on numerical calculation and linear regression model. BACKGROUND

[0002] During the operation of distribution transformers, the internal temperature distribution is uneven, and if the local temperature is abnormal, it will accelerate the thermal aging and affect the degradation of transformer oil paper insulation. When the transformer winding is overheated, the short-term overheating treatment measure is to reduce the load, and the long-term overheating needs to install an additional transformer. The temperature of the transformer needs to be predicted according to the predicted load, environmental temperature and cooling mode, and accurate temperature prediction can make full use of the capacity of the existing equipment, delay the installation of new transformers, thereby saving money for the power industry. In addition, temperature prediction can be used to determine the maximum load acceptable by the transformer under different load, environment and cooling mode conditions, and to evaluate its overload capacity. Real-time and accurate calculation of the temperature rise characteristics and hot spot temperature of the transformer is beneficial to the safe and stable operation of the distribution transformer.

[0003] Some people in the prior art have proposed a method for predicting transformer temperature based on a neural network model, for example, Chinese patent document CN 113705082A describes a transformer hot spot temperature prediction method based on an improved BP neural network, which is based on an improved BP neural network model. The load current, top layer oil temperature, environmental temperature and hot spot temperature under the running state of the transformer are collected in real time, and the load rate is calculated according to the load current, and a transformer hot spot temperature prediction method is proposed. However, this method cannot directly calculate the hot spot temperature at this time according to the existing geometric structure, physical parameters, environmental temperature and load coefficient of the transformer. It cannot select more optimized materials to be applied to the transformer, thereby improving the overload capacity and service life of the transformer. It also cannot provide data support for the design of the best design material transformer. There are defects in use and need to be improved. SUMMARY

[0004] In view of the technical problems existing in the background art, the method for predicting transformer hot spot temperature based on numerical calculation and linear regression model provided by the present application collects temperature distribution data of the transformer under different operating conditions through simulation calculation, and based on this, a neural network optimized by a cuckoo search algorithm is proposed to predict the hot spot temperature of the transformer winding under different operating conditions, thereby improving the accuracy and practicality of solving the temperature field distribution of the transformer.

[0005] In order to solve the above technical problems, the present application adopts the following technical solutions to realize:

[0006] A method for predicting transformer hot spot temperature based on numerical calculation and linear regression model, comprising the following steps:

[0007] S1, determining transformer design object: determining the material type of the core of the transformer, transformer oil, and copper wire for winding the winding;

[0008] S2, establishing a three-dimensional model: refining the winding distribution according to the actual structure of the transformer, dividing the grid by region, improving the iteration efficiency by using the one-way coupling method, and constructing a three-dimensional electromagnetic-temperature-fluid multi-physical field coupling model;

[0009] S3, verifying the rationality and accuracy of the three-dimensional model;

[0010] S4, solving the no-load loss size of the distribution transformer and the temperature rise distribution under the rated load;

[0011] S5, obtaining the temperature rise distribution of each structure of the transformer and the hot spot temperature of the winding by changing the load coefficient and the ambient temperature, and constructing a database of the hot spot prediction model;

[0012] S6, based on the neural network algorithm, the cuckoo search algorithm is used to optimize the defects of the neural network; based on the neural network, a prediction model of the hot spot temperature of the winding of the distribution transformer is constructed, the cuckoo algorithm is used to optimize the neural network, the neural network is used as the objective function, the weights and thresholds of the network are optimized through the intelligent algorithm, the parameter settings in the model are determined, and the optimal weights and biases are obtained;

[0013] S7, constructing a cuckoo-neural network model for predicting the hot spot temperature of the distribution transformer, and obtaining the hot spot temperature data value under multiple operating states;

[0014] S8, analyzing the prediction results.

[0015] Compared with the traditional neural network, the model has higher accuracy on the real winding hot spot temperature data set, and the work can provide a new idea for the thermal characteristic research of the distribution transformer.

[0016] Preferably, in step S2, the step of establishing a three-dimensional model is:

[0017] S2.1, determining the basis for establishing a three-dimensional model: constructing according to the geometric structure distribution and size of the actual transformer equipment;

[0018] S2.2, since the transformer can be approximated as a symmetrical structure, in order to improve the calculation efficiency, 1 / 4 of the middle phase winding of the transformer is taken as a symmetrical model in the model;

[0019] S2.3, using a regional grid division method, different regions and boundaries are divided into different grid shapes;

[0020] S2.4, based on the electromagnetic field model in COMSOL Multiphysics software to solve the core and winding loss, the results are coupled into the heat source of the heat transfer field grid by grid, and a three-dimensional electromagnetic-temperature-fluid multi-physical coupling model is constructed;

[0021] S2.5, accurately simulate the convection heat transfer process of the transformer, and realize the calculation of the temperature rise of the transformer components and the calculation of the hot spot temperature value of the winding.

[0022] Preferably, in step S2.3, the method of determining the grid division in modeling is: when constructing the model grid, the simulation model is divided into non-overlapping grids, and structured grid is used when constructing the grid. Since the temperature gradient at the fluid-solid interface is obtained by the temperature difference between the interface and the adjacent layer of grid points, the grid division precision here is higher.

[0023] Preferably, in step 3, the method for verifying the rationality and accuracy of the transformer simulation model is:

[0024] S3.1, in the short-circuit temperature rise test, first apply the rated total loss to the test transformer, measure the real-time temperature of the top layer of oil, and calculate the average temperature of all the oil inside the transformer;

[0025] S3.2, there should be no interruption during the temperature rise test, immediately reduce the current value in the winding after the top layer of oil temperature test is completed, and install temperature sensors at the temperature measuring points of the high and low voltage windings;

[0026] S3.3, when calculating the hot spot temperature value of the transformer and evaluating the cooling effect of the cooling medium, the influence of the top layer of oil temperature and the oil average temperature parameters should be considered. In the temperature rise test, the top layer of oil temperature, the oil average temperature and the winding hot spot temperature are measured.

[0027] Preferably, in step 4, the no-load loss of the distribution transformer and the temperature rise distribution under the rated load are solved as follows:

[0028]

[0029] Where D is the electric displacement, in C / m 2 ; B is the magnetic induction intensity, in T; p is the charge density, in C / m 3 ; E is the electric field intensity, in V / m; H is the magnetic field intensity, in A / m; J is the current density, in A / m 2 ; l is the magnetic path length of the primary winding, in m;

[0030] The energy conservation equation of the fluid is as follows:

[0031]

[0032] Where cp Cp is the specific constant pressure heat capacity, with the unit of J / (kg·℃); λ is the thermal conductivity, with the unit of W / (m·k); S T is the total heat source, with the unit of W, which is the sum of the transformer no-load loss and the load loss.

[0033] Preferably, in step S5, the method for calculating the internal temperature rise distribution of the distribution transformer and the size of the winding hot spot temperature is that, in the simulation, the temperature rise distribution curves of the transformer under different load coefficients are obtained by changing the load parameters and the ambient temperature, and the data sets required for regression calculation are arranged.

[0034] Preferably, in step S6, a prediction model of the winding hot spot temperature of the distribution transformer is constructed based on a neural network, the neural network is optimized based on a cuckoo search algorithm, the neural network is taken as an objective function (fitness function), the weights and thresholds of the network are optimized through an intelligent algorithm, the parameter settings in the model are determined, and the optimal weights and biases are obtained.

[0035] Preferably, the optimization formula of the cuckoo search algorithm is:

[0036] The nest position updating formula of the cuckoo is as follows:

[0037]

[0038] wherein, x t i represents the nest position of the i-th bird in the t-th generation; is point-to-point multiplication; α>0 is a step control quantity, which is valued according to specific conditions and is usually taken as 1; Leuy(λ) is a random search path:

[0039] Levy~u=t -λ (1<λ≤3);

[0040] The nest position is subject to a random number r∈[0, 1] uniformly distributed, after the position is updated, the new nest exchange old nest probability Pa of the host bird is compared with r; if r<Pa, the optimal nest position data is retained and output, otherwise, x i t+1 is randomly selected, and the calculation is continued.

[0041] Preferably, in step S7, the process of optimizing the neural network by the cuckoo search algorithm is divided into three parts: determining the neural network structure, obtaining the best weights and biases through the cuckoo search algorithm, and predicting the hot spot temperature value based on the neural network; in the temperature rise prediction model, the objective function is the error between the predicted value and the true value.

[0042] Preferably, the analysis process of step S8 is: error analysis is performed on the data by taking the average relative error, root mean square error as the performance characteristic value, the effectiveness of the cuckoo search algorithm optimized neural network is evaluated, and the error analysis formula is:

[0043]

[0044] Wherein, RMSE is the root mean square error; R 2 is a linear regression fitting coefficient; MSE is the mean square error; MAE is the mean absolute error; N is the sample number; T t is the actual hot spot temperature size, unit: ℃; T t * is the predicted value calculated by the model, unit: ℃.

[0045] The patent can achieve the following beneficial effects:

[0046] The COMSOL Multiphysics simulation software is used to solve the internal temperature rise distribution and hot spot temperature size of the distribution transformer under different load coefficients and environmental temperatures, and a data set containing transformer oil temperature distribution, load condition, environmental temperature and hot spot temperature is constructed. On this basis, the cuckoo search algorithm is used to optimize the neural network, the structure of the neural network is determined, the initial weight matrix is optimized, the parameters are set to minimize the error value of the predicted hot spot temperature of the distribution transformer, and a set of optimal parameters of the neural network model can be obtained. Compared with the traditional intelligent algorithm model, the results show that the proposed method significantly improves the convergence speed and hot spot temperature prediction accuracy of the algorithm model, thereby realizing the analysis of the thermal characteristics and load capacity of the transformer, which helps to efficiently and accurately evaluate the operation state of the transformer and promotes the rapid development of the transformer industry. BRIEF DESCRIPTION OF DRAWINGS

[0047] The application will be further described below in combination with the drawings and examples:

[0048] Figure 1 is a schematic diagram of the test distribution transformer of the application;

[0049] Figure 2 is a three-dimensional simulation geometry model and grid chart of the application;

[0050] Figure 3 is a transformer temperature rise distribution chart of the application;

[0051] Figure 4 is a temperature rise test principle diagram of the application;

[0052] Figure 5 is a neural network topology chart of the application;

[0053] Figure 6 is a comparison chart of the test results of the neural network modelFigure 1 ;

[0054] Figure 7 is a comparison chart of test results of the neural network model Figure 2 ;

[0055] Figure 8 is a comparison chart of test results of the neural network model Figure 3 ;

[0056] Figure 9 is a flow chart of the cuckoo-neural network model of the application

[0057] Figure 10 is the fitness curve of the optimized model of the application. DETAILED DESCRIPTION

[0058] As shown in Figures 1-10 , a method for predicting the hot spot temperature of a transformer based on numerical calculation and linear regression model comprises the following steps:

[0059] S1, determining the transformer design object: determining the material types of the transformer core, transformer oil and copper wire of the wound winding;

[0060] S2, establishing a three-dimensional model: refining the winding distribution according to the actual structure of the transformer, dividing the grid by region, improving the iteration efficiency by using the one-way coupling method, and constructing a three-dimensional electromagnetic-temperature-fluid multi-physical field coupling model; specifically:

[0061] S2.1, determining the basis for establishing a three-dimensional model: constructing according to the geometric structure distribution and size of the actual transformer equipment;

[0062] S2.2, since the transformer can be approximated as a symmetrical structure, in order to improve the calculation efficiency, 1 / 4 of the middle phase winding of the transformer in the model is taken as a symmetrical model;

[0063] S2.3, using a regional grid division method, different regions and boundaries are divided into different grid shapes;

[0064] The method for determining the grid division when modeling is: when constructing the model grid, the simulation model is divided into non-overlapping grids, and structured grid is used when constructing the grid. Since the temperature gradient at the fluid-solid interface is obtained by the temperature difference between the interface and the adjacent layer of grid points, the grid division precision at this place is higher.

[0065] S2.4, based on the electromagnetic field model in COMSOL Multiphysics software, the core and winding losses are solved, and the results are coupled into the heat source of the heat field by grid, and a three-dimensional electromagnetic-temperature-fluid multi-physical coupling model is constructed;

[0066] S2.5, accurately simulate the transformer convection heat transfer process, and realize the calculation of transformer component temperature rise and winding hot spot temperature value.

[0067] S3, verify the rationality and accuracy of the three-dimensional model; specifically:

[0068] S3.1, in the short-circuit temperature rise test, first apply the rated total loss to the test transformer, measure the real-time temperature of the top layer oil, and calculate the average temperature of all the oil inside the transformer;

[0069] S3.2, there should be no interruption during the temperature rise test process, immediately reduce the current value in the winding after the top layer oil temperature test is completed, and install temperature sensors at the temperature measurement points of the high and low voltage windings;

[0070] S3.3, when calculating the transformer hot spot temperature value and evaluating the cooling effect of the cooling medium, the influence of the top layer oil temperature and the oil average temperature parameters should be considered, and the top layer oil temperature, oil average temperature and winding hot spot temperature should be measured during the temperature rise test.

[0071] S4, solve the size of the distribution of the no-load loss of the distribution transformer and the temperature rise under the rated load;

[0072] Solve the size of the distribution of the no-load loss of the distribution transformer and the temperature rise under the rated load as follows:

[0073]

[0074]

[0075] Where D is the electric displacement, with units of C / m 2 ; B is the magnetic induction, with units of T; ρ is the charge density, with units of C / m 3 ; E is the electric field strength, with units of V / m; H is the magnetic field strength, with units of A / m; J is the current density, with units of A / m 2 ; l is the magnetic path length of the primary winding, with units of m;

[0076] The energy conservation equation of the fluid is as follows:

[0077]

[0078] Where c p is the constant-pressure specific heat capacity, with units of J / (kg·℃); λ is the thermal conductivity, with units of W / (m·k); S T is the total heat source, with units of W, which is the sum of the transformer no-load loss and the load loss.

[0079] S5, by changing the load coefficient and the ambient temperature, the temperature rise distribution of each structure of the transformer and the winding hot spot temperature are obtained, and a database of the hot spot prediction model is constructed;

[0080] In the simulation, the temperature rise distribution curve of the transformer under different load coefficients is obtained by changing the load parameters and the ambient temperature, and the data set required for regression calculation is arranged.

[0081] S6, based on the algorithm of neural network, the cuckoo search algorithm is used to optimize the defects of neural network; specifically:

[0082] Based on the neural network, a prediction model of the winding hot spot temperature of the distribution transformer is constructed, the neural network is optimized based on the cuckoo algorithm, the neural network is used as the objective function (fitness function), the weights and thresholds of the network are optimized through intelligent algorithm, and the parameter settings in the model are determined to obtain the optimal weights and biases.

[0083] The optimization formula of the cuckoo search algorithm is:

[0084] The nest position updating formula is as follows:

[0085]

[0086] Where, x t i Indicates the nest position of the i-th bird in the t-th generation; Point-to-point multiplication; alpha>0 is the step control quantity, which is usually taken as 1 according to the specific situation; Levy(λ) is a random search path:

[0087] Levy~u=t -λ (1<λ≤3);

[0088] The nest position is subject to a random number r in [0, 1] uniformly distributed, after updating the position, compare r with the new nest exchange old nest probability Pa of the host bird; if r<Pa, the optimal nest position data is retained and output, otherwise x i t+1 And continue to calculate.

[0089] S7, a cuckoo-neural network model for predicting the hot spot temperature of the distribution transformer is constructed, and the hot spot temperature data values under multiple operating conditions are obtained; specifically:

[0090] The process of optimizing the neural network by the cuckoo algorithm includes three parts: determining the neural network structure, obtaining the best weights and biases through the cuckoo algorithm, and predicting the hot spot temperature value based on the neural network; in the temperature rise prediction model, the objective function is the error between the predicted value and the true value.

[0091] S8, analyze the prediction results. Specifically:

[0092] The data is analyzed by average relative error and root mean square error as performance characteristic values to evaluate the effectiveness of the cuckoo algorithm optimized neural network, and the error analysis formula is:

[0093]

[0094] Wherein, RMSE is the root mean square error; R 2 is the linear regression fitting coefficient; MSE is the mean square error; MAE is the mean absolute error; N is the sample number; T t is the actual hot spot temperature size, unit: ℃; T t * is the predicted value calculated by the model, unit: ℃.

[0095] Compared with the traditional neural network, the proposed model has higher accuracy on the real winding hot spot temperature data set, and the work can provide a new idea for the thermal characteristic research of distribution transformers. The operation method of the present application will be described in the form of examples as follows:

[0096] Example 1

[0097] As Figure 1 described above, the high and low voltage winding of the test transformer is a pie-shaped winding wound by copper wire. In order to improve the overload bearing capacity, the turn insulation adopts H-grade heat-resistant level, and other structural insulation materials adopt different heat-resistant level materials. This distribution transformer is selected for research.

[0098] In the specific implementation, when constructing the three-dimensional model of the test transformer, the structure is optimized, the winding is refined into a pie-shaped structure according to the actual structure of the transformer, and the axial distribution characteristics of the oil flow in the horizontal oil channel of the transformer and the temperature distribution of the oil flow along the circumferential direction are considered, and the geometric model is shown in Figure 2 When the simulation model is calculated based on the finite element method, the calculation accuracy is proportional to the quality and quantity of the occupied unit grid, and high-quality grid makes the geometric model structure have high restoration degree. When the simulation model needs to reduce the calculation amount, the balance time and accuracy are obtained, and the results that meet the calculation accuracy and iteration time are obtained, and an economical finite element calculation model is sought. Therefore, when constructing the model grid, the simulation model is divided into non-overlapping grids, and structured grid is used when constructing the grid. Since the temperature gradient at the fluid-solid interface is obtained by the temperature difference between the interface and the adjacent layer of grid points, the grid division precision at this position is higher, and the grid of the geometric model after division is shown in Figure 2 .

[0099] Then the material property parameters existing in the transformer can be fitted in the origin software, and fitted as a mathematical function, as shown in Table 1.

[0100] Table 1 Material property parameters

[0101]

[0102] The no-load loss and load loss are obtained by the transformer temperature rise test. In order to calculate conveniently, the average loss density is uniformly applied in each heat source when the heat source is loaded. The heat generation rate of each heat generating part of the transformer is shown in Table 2.

[0103] Table 2 Loss and heat generation rate of distribution transformer

[0104]

[0105] The transformer operation process is a non-ideal state, and each structural part will generate loss and convert into heat and dissipate to the surrounding medium, causing internal temperature rise. The internal temperature distribution of the distribution transformer is shown in Figure 3 When the axial height increases, the temperature of the core and winding rises but is not uniform. The internal temperature of the transformer gradually rises due to the continuous heating of the loss such as iron loss and copper loss, and finally reaches a thermal equilibrium state. The transformer oil in the oil duct is heated and the temperature rises, and the buoyancy difference caused by the decrease of oil density causes the insulating oil to flow upward along the oil duct, and the outlet temperature of the transformer internal oil flow is higher than the inlet temperature. Since the heat dissipation effect of the lower end of the transformer is better than that of the upper end, the upper end temperature is higher than the lower end temperature. When the transformer is stably operated, the hot spot temperature is located at about 9 / 10 of the upper part of the low-voltage winding.

[0106] More specifically, in the simulation, the temperature rise distribution curve of the transformer under different load coefficients is obtained by changing the load parameters, the overload capacity of the amorphous alloy plant insulating oil distribution transformer is evaluated, and the hot spot temperature of the transformer under natural oil circulation cooling mode is shown in Table 3.

[0107] Table 3 Hot spot temperature of transformer under natural oil circulation cooling mode

[0108]

[0109] Subsequently, the short-circuit test method is used for temperature rise test to verify the rationality and accuracy of the multi-physical field coupling simulation model of the transformer. The actual test data is used as a standard and compared with the simulation results. The test wiring is shown in Figure 4 The temperature distribution results obtained by the transformer temperature rise test are compared with the simulation results. The temperature distribution trend of the simulation results and the test data is basically consistent, and the temperature rise prediction of the transformer is relatively accurate.

[0110] In order to establish a universal transformer winding hot-spot temperature prediction model, a feature data set containing various different operating states of the transformer must be constructed. The external environment can significantly affect the heat exchange between the transformer and the external space, changing the heat exchange efficiency and thus affecting the winding hot-spot temperature of the transformer, so the ambient temperature is selected as a characteristic variable. In addition, the load condition of the transformer will affect the heating of the core and coil, directly affecting the radiation and conduction processes of the heat source, so the overload factor of the load is also selected for constructing the database. The data is randomly divided into two parts, training set and test set. 80% of the total number of samples is used to train the hot-spot prediction model, and 20% is used to verify the feasibility of the model. Part of the training data set is shown in Table 4.

[0111] Table 4 Training sample data set

[0112]

[0113] Based on the neural network model, the overload factor and real-time ambient temperature are taken as input factors, and the hot-spot temperature is output to construct the hot-spot prediction model of the distribution transformer. The neural network topology diagram is shown in Figure 5 , and the training process is as follows:

[0114] (1) Network initialization: initialize the connection weights ω ij and ω jk between the input layer, hidden layer, and output layer neurons, and initialize the threshold values of the hidden layer i and the output layer k.

[0115] (2) Calculate the output of the hidden layer H j :

[0116]

[0117] In the formula, f is the activation function of the hidden layer; X i is the input variable of the neuron; a j is the threshold value or offset value of the hidden layer. The function selected in this paper is the Sigmoid function:

[0118]

[0119] (3) Calculate the output layer result O k :

[0120]

[0121] In the formula, b k is the threshold value or offset value of the output layer.

[0122] (4) Error calculation, according to the network prediction output O k and the expected output Y k , the network prediction error ek :

[0123] e k = Y k - O k ;

[0124] (5) Weight update, according to the network prediction error e k update network connection weights ω ij and ω jk :

[0125]

[0126] where η is the learning rate; x i is the input value.

[0127] (6) Threshold update: according to the network prediction error e k update network node threshold a j and b k .

[0128]

[0129] (7) Determine whether the iteration of the algorithm has ended or reached the required error range. After reaching, the model training is completed, and if not, return to step (2).

[0130] The neural network model training results are compared with the ideal as shown in Figures 6-8 . The fitting effect of the predicted value is good, and the training speed is fast and the calculation efficiency is high. However, the neural network model also has drawbacks in practical application. The initial weights and thresholds are randomly generated, leading to unstable algorithms. In addition, the gradient descent method has limitations, including long training time and weak local search ability. Since the cuckoo search algorithm has superior optimization ability, to address the above problems, this paper optimizes the neural network based on the cuckoo search algorithm, overcoming the instability of the algorithm caused by blind selection of initial values, and improving the reliability of the prediction results. The optimization method is shown in Figure 9 .

[0131] To reduce the number of iterations and ensure the practicality of the training results. When training the hot spot temperature, the number of hidden layers of the model is set to 5, and the confirmation check value is 6. The learning rate is selected as 0.01, and if the number of iterations reaches 1000 or the error rate is less than 10E-5, the iteration will stop. Table 5 shows the selection results of the initial weights and thresholds of the neural network model optimized by the cuckoo search algorithm. w H is the weight matrix from the input layer to the hidden layer, b H is the threshold of the hidden layer. w O is the weight matrix from the hidden layer to the output layer, b OThreshold value of output layer. According to the determined model structure, the model is trained by the data of the training set, and the model with strong universality and generalization is obtained through continuous iteration training.

[0132] Table 5 initial weight and threshold size of the neural network based on the cuckoo search algorithm optimization

[0133]

[0134] Table 6 comparison of training results before and after model improvement

[0135]

[0136] Based on the cuckoo search algorithm optimization, the optimization iteration results of the neural network model and the traditional neural network model are analyzed, and the prediction results of the hot spot temperature of the cuckoo algorithm before and after the improvement are compared and analyzed, as shown in Table 6. It can be seen that the error eigenvalue of the improved model is significantly reduced, the accuracy is improved, the sample iteration efficiency is improved, and the running time is significantly shortened, which has great advantages in the direction of hot spot temperature size prediction. Figure 10 In order to optimize the fitness curve of the model, the terminal number is 50, which indicates that the model can converge after 50 iterations, indicating that the convergence speed of the optimized model is improved.

[0137] The above embodiments are only preferred technical solutions of the present application, and should not be regarded as a limitation of the present application. The protection scope of the present application should be based on the technical solutions claimed in the claims, including the equivalent replacement solutions of the technical features claimed in the claims. That is, the equivalent replacement improvement within this range is also within the protection scope of the present application.

Claims

1. A method for predicting transformer hot spot temperature based on numerical calculation and linear regression model, characterized in that... Includes the following steps: S1. Determine the transformer design object: Determine the material type of the transformer core, transformer oil, and copper wire used for winding; S2. Establish a three-dimensional model: refine the winding distribution according to the actual structure of the transformer, divide the mesh into regions, use the one-way coupling method to improve the iteration efficiency, and construct a three-dimensional electromagnetic-temperature-fluid multi-physics coupling model. S3. Verify the rationality and accuracy of the 3D model; S4. Solve for the magnitude of the no-load loss of the distribution transformer and the temperature rise distribution under rated load; S5. By changing the load factor and ambient temperature, the temperature rise distribution of each structure of the transformer and the magnitude of the winding hot spot temperature are obtained, and a database of hot spot prediction models is constructed. S6. Based on the neural network algorithm, the Cuckoo Search algorithm is used to optimize the defects of the neural network; based on the neural network, a prediction model for the hot spot temperature of the distribution transformer winding is constructed, and the neural network is optimized based on the Cuckoo Search algorithm. The neural network is used as the objective function, and the weights and thresholds of the network are optimized through intelligent algorithms to determine the parameter settings in the model in order to obtain the optimal weights and biases. S7. Construct a cuckoo-neural network model for predicting the hot spot temperature of distribution transformers and obtain hot spot temperature data values ​​under various operating conditions. S8. Analyze the prediction results; In step S3, the method for verifying the rationality and accuracy of the transformer simulation model is as follows: S3.1 In the short-circuit temperature rise test, the rated total loss is first applied to the test transformer, the real-time temperature of the top oil is measured, and the average temperature of all oil inside the transformer is calculated. S3.2 The temperature rise test should not be interrupted. After the top oil temperature test is completed, the current value in the winding should be reduced immediately, and temperature sensors should be installed at the temperature measurement points of the high and low voltage windings. S3.3 When calculating the hot spot temperature of the transformer and evaluating the cooling effect of its cooling medium, the influence of the top oil temperature and the oil uniform temperature parameters should be considered. In the temperature rise test, the top oil temperature, the oil uniform temperature and the winding hot spot temperature should be measured. In step S5, when calculating the internal temperature rise distribution and winding hot spot temperature of the distribution transformer, the temperature rise distribution curve of the transformer under different load factors is obtained by changing the load parameters and ambient temperature in the simulation, and then organized into the dataset required for regression calculation.

2. The method for predicting transformer hotspot temperature based on numerical calculation and linear regression model according to claim 1, characterized in that: In step S2, the steps for establishing the 3D model are as follows: S2.

1. Basis for establishing the three-dimensional model: all models are constructed based on the geometric structure distribution and size of the actual transformer equipment; S2.2 To improve computational efficiency, 1 / 4 of the transformer's intermediate phase winding is taken as a symmetrical model in the three-dimensional model; S2.

3. Adopt a regional grid division method to divide different regions and boundaries into different grid shapes; S2.

4. Solve the core and winding losses based on the electromagnetic field model in COMSOL Multiphysics software, and couple the results to the heat source of the heat transfer field grid by grid to construct a three-dimensional electromagnetic-temperature-fluid multiphysics coupled model. S2.

5. Accurately simulates the convective heat transfer process of a transformer and calculates the temperature rise of transformer components and the temperature value of winding hot spots.

3. The method for predicting transformer hotspot temperature based on numerical calculation and linear regression model according to claim 2, characterized in that: In step S2.3, the method for determining the mesh generation during modeling is as follows: when constructing the model mesh, the simulation model is divided into non-overlapping meshes, and a structured mesh is used when constructing the mesh.

4. The method for predicting transformer hotspot temperature based on numerical calculation and linear regression model according to claim 1, characterized in that: In step S4, the magnitude of the no-load loss of the distribution transformer and the temperature rise distribution under rated load are calculated as follows: Where D is the electric displacement, with units of C / m. 2 B is the magnetic flux density, measured in tons (T); ρ is the charge density, measured in cubic centimeters (C / m³). 3 E is the electric field strength, in V / m; H is the magnetic field strength, in A / m; J is the current density, in A / m. 2 l represents the magnetic path length of the primary winding, in meters (m). The energy conservation equation for a fluid is shown below: Among them, c p λ is the specific heat capacity at constant pressure, in J / (kg·℃); λ is the thermal conductivity coefficient, in W / (m·K); S T The total heat source is represented by W, and its value is the sum of the transformer's no-load loss and load loss.

5. The method for predicting transformer hotspot temperature based on numerical calculation and linear regression model according to claim 1, characterized in that: The optimization formula for the cuckoo search algorithm is: The formula for updating the location of cuckoo nests is as follows: Where, x t i This indicates the position of the i-th bird's nest in generation t; This is point-to-point multiplication; α>0 is the step size control variable, which is set to 1 depending on the specific situation; Leuy(λ) is the random search path. Levy~u=t -λ (1<λ≤3); The random number \(r\) of the nest location follows a uniform distribution, \(r\in[0,1]\). After the location is updated, compare \(r\) with the probability \(P_a\) of the nest owner bird replacing the old nest with a new one; if \(r < P_a\), then retain and output the optimal nest location data, otherwise randomly select \(x\). i t+1 , and continue the calculation.

6. The method for predicting transformer hotspot temperature based on numerical calculation and linear regression model according to claim 1, characterized in that: In step S7, the process of optimizing the neural network using the Cuckoo algorithm is divided into three parts: determining the neural network structure, obtaining the optimal weights and biases through the Cuckoo algorithm, and predicting hotspot temperature values ​​based on the neural network; in the temperature rise prediction model, the objective function is to calculate the error between the predicted value and the actual value.

7. The method for predicting transformer hotspot temperature based on numerical calculation and linear regression model according to claim 1, characterized in that: The analysis process in step S8 is as follows: Error analysis is performed on the data using the mean relative error and root mean square error as performance feature values ​​to evaluate the effectiveness of the Cuckoo algorithm in optimizing the neural network. The error analysis formula is: Where RMSE is the root mean square error; R 2 The linear regression fitting coefficients are: MSE (mean squared error), MAE (mean absolute error), N (sample size), and T (time factor). t Actual hotspot temperature, in °C (°C); T t * These are the predicted values ​​calculated by the model, in °C.

Images