Oil-immersed reactor hot spot inversion method based on random forest

Abstract

The invention discloses an oil-immersed reactor hot spot inversion method based on a random forest, and the method comprises the following steps: firstly building a reactor flow field-temperature field model, and calculating the temperature field data of a reactor; preliminarily screening oil tank wall temperature measurement points according to the internal oil flow characteristics of the reactor in the calculation result, and then obtaining sample data sets under different working conditions by using a Latin hypercube sampling method; by taking the sample data set as a sample through a random forest algorithm, re-screening the preliminarily screened oil tank wall measuring points, and sorting the oil tank wall measuring points according to the feature importance of the measuring points; a GA-BP neural network model is established, oil tank wall measuring points under different measuring point combinations serve as input, and a sample obtained through calculation serves as a data set for training; and determining an optimal oil tank wall temperature measurement point combination according to the error index obtained by training. The defects in the prior art can be overcome, optimal selection of the temperature measuring points on the outer wall of the oil tank is achieved, the inversion precision of the temperature field is improved, and a reliable temperature monitoring means is provided for safe operation of the reactor.

Description

Technical Field

[0001] The present invention belongs to the technical field of temperature monitoring and fault diagnosis of power equipment, and in particular, to a method for inverting hot spots of oil-immersed reactors based on random forest. Background Technique

[0002] In modern power systems, oil-immersed reactors, as indispensable reactive power regulation equipment, play a crucial role in maintaining the stability of the system and power quality. Whether they operate normally is directly related to the safe and reliable power supply of the power system. Once a fault occurs, it may trigger a chain reaction, resulting in serious consequences such as large-scale power outages. The insulation performance of the reactor is crucial for its lifespan, and the hot spot temperature is one of the key indicators for measuring insulation performance. However, in the actual operation process, it is difficult to directly measure the hot spot temperature inside the reactor due to many difficulties such as harsh measurement environments and limited sensor arrangements. Therefore, indirectly predicting and calculating the hot spot temperature has become a research focus.

[0003] Currently, the existing methods for calculating the hot spot temperature of windings have their own advantages and disadvantages. The empirical formula method is simple and easy to use, but its accuracy is insufficient in complex environments; the accuracy of the thermal circuit equivalent model has been improved, but it is difficult to cope with complex and changeable operating conditions; although artificial intelligence algorithms have been applied in predicting the hot spot temperature of transformer windings, it is difficult to measure input parameters and obtain real-time experimental data. In addition, when selecting measuring points on the outer wall of the oil tank based on streamline characteristics, the number of measuring points is large and subjective, and it is difficult to determine the optimal combination of measuring points, which affects the inversion accuracy. Therefore, there is an urgent need for a new method to improve the accuracy and reliability of the temperature field inversion of oil-immersed reactors. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to provide a method for inverting hot spots of oil-immersed reactors based on random forest, which can solve the deficiencies of the existing technology, realize the optimal selection of temperature measuring points on the outer wall of the oil tank, improve the accuracy of temperature field inversion, and provide a reliable temperature monitoring means for the safe operation of the reactor.

[0005] To solve the above technical problem, the technical solutions adopted by the present invention are as follows.

[0006] A method for inverting hot spots of oil-immersed reactors based on random forest, comprising the following steps:

[0007] First, establish the reactor flow field-temperature field model and calculate the reactor temperature field data; preliminarily screen the temperature measurement points on the tank wall according to the oil flow characteristics inside the reactor in the calculation results, and then use the Latin hypercube sampling method to obtain the sample data sets under different working conditions; use the random forest algorithm to take the sample data sets as samples, re-screen the temperature measurement points on the tank wall initially screened, and sort the temperature measurement points on the tank wall according to the importance of the characteristics of the temperature measurement points; establish a GA-BP neural network model, use the temperature measurement points on the tank wall under different measurement point combinations as inputs, and use the calculated samples as the data set for training; determine the optimal combination of temperature measurement points on the tank wall based on the error index obtained from the training.

[0008] Preferably, establish the reactor flow field-temperature field model and calculate the reactor temperature field data, including

[0009] Introduce non-uniform losses into the fluid-temperature field coupling, construct the weak coupling relationship between the magnetic field and the fluid-temperature field, describe the flow of transformer oil based on the Navier-Stokes equation, calculate the convective heat transfer coefficient according to the actual wind speed, and establish the flow field-temperature field coupling calculation model;

[0010] Set the initial temperature and the ambient temperature to ensure that the initial values of the oil temperature in the tank, the tank wall, the oil temperature in the radiator, and the radiator wall are consistent with the ambient temperature, the initial fluid flow velocity is 0, set the tank wall and the inner core surface to be in a no-slip state, determine the convective heat transfer coefficients of the top, side wall, and fins of the tank under windless conditions, clarify the heat source type and the characteristics of the fluid field, and set the direction and magnitude of gravity;

[0011] Add the core and winding losses as heat sources to the fluid-thermal coupling physical field, calculate the temperature field distribution of the reactor under rated voltage, standard capacity, specific ambient temperature, and wind speed, and obtain the hot spot temperature and the maximum temperature rise data; compare with the actual value of the reactor temperature rise to verify the effectiveness of the reactor temperature field data; analyze the temperature distribution law of the tank and the radiator fins.

[0012] Preferably, preliminarily screen the temperature measurement points on the tank wall according to the oil flow characteristics inside the reactor in the calculation results, and then use the Latin hypercube sampling method to obtain the sample data sets under different working conditions, including using the three factors of the external ambient temperature, load current, and external wind speed as variables to simulate different working conditions, and using the Latin hypercube sampling method to obtain the sample data sets under different working conditions.

[0013] Preferably, use the random forest algorithm to take the sample data sets as samples, re-screen the temperature measurement points on the tank wall initially screened, and sort the temperature measurement points on the tank wall according to the importance of the characteristics of the temperature measurement points, including

[0014] First, use the sparrow optimization algorithm to optimize the number and depth of decision trees in the random forest model. Then, use the random forest model to screen the temperature measurement points on the fuel tank wall. Finally, use the immune algorithm to perform a secondary screening on the temperature measurement points on the fuel tank wall output by the random forest model, and assign the corresponding importance index to each temperature measurement point on the fuel tank wall screened out for the second time.

[0015] Preferably, using the sparrow optimization algorithm to optimize the number and depth of decision trees in the random forest model includes the following steps:

[0016] Set the sparrow population size, use the number and depth of decision trees in the random forest model to form the solution space, randomly assign a position in the solution space to each sparrow, use the output separation of different decision trees as the fitness function, and set the maximum number of iterations;

[0017] Arrange the sparrows in descending order according to the current fitness, regard the top 20% of the sparrows as discoverers, the bottom 20% of the sparrows as vigilant ones, and the remaining sparrows as joiners;

[0018] The update method of the discoverer is where is the position of the i-th discoverer in the t-th iteration, is the position of the i-th discoverer in the (t - 1)-th iteration, and λ is the perturbation coefficient;

[0019] The update method of the joiner is where is the position of the i-th joiner in the t-th iteration, is the position of the i-th joiner in the (t - 1)-th iteration, is the clustering center position of all discoverers in the t-th iteration;

[0020] The update method of the vigilant one is to randomly select a position in the solution space as the new position of the vigilant one, and there are no discoverers and joiners at this position;

[0021] Perform cyclic iteration on the sparrow population until the maximum number of iterations is reached or the optimal solution is obtained and the iteration stops. The position of the sparrow with the highest fitness when the iteration stops is used as the number and depth of decision trees in the random forest model.

[0022] Preferably, a random forest model is used to screen the temperature measurement points on the fuel tank wall. Each random tree in the random forest model outputs a screening result. All the screening results are used as the initial antibody population, and the optimization objective function and constraints are used as antigens. The affinity between each antibody and the antigen is calculated. Antibodies with an affinity greater than the upper threshold are cloned, antibodies with an affinity less than the lower threshold are mutated, and the remaining antibodies are used for peptide chain crossover. The above cloning, mutation, and crossover operations are repeated until the termination condition is met. At this time, the antibody with the highest affinity to the antigen is used as the result of the secondary screening. The importance index of each temperature measurement point in the secondary screening result is proportional to the number of times the temperature measurement point appears in the screening results output by the random tree.

[0023] Preferably, the average deviation rate η between the secondary screening result and the screening results output by all random trees is calculated, and the perturbation coefficient λ is updated according to the average deviation rate η. where λ′ is the updated perturbation coefficient.

[0024] Preferably, based on the error index obtained from training, the optimal combination of temperature measurement points on the fuel tank wall is determined, including

[0025] Starting from the double temperature measurement points, a GA-BP neural network model is used to predict the hot spot temperature of the reactor. The temperature measurement points are sorted in descending order according to the importance of the temperature measurement points. The temperature measurement point with the highest importance in the current sequence is added to the GA-BP neural network model, and the hot spot temperature of the reactor is predicted again. The prediction is repeated several times. The root mean square error, mean absolute error, and mean absolute percentage error are used as error indices, and the error indices under the current temperature measurement point combination are recorded after each prediction. By analyzing the error indices, training convergence curves, and fitness curves, the training effect of the GA-BP neural network model for different measurement point combinations is evaluated, redundant measurement points are removed, and the optimal measurement point combination is determined.

[0026] An oil-immersed reactor hot spot inversion device based on random forest includes

[0027] A first model establishment module for establishing a reactor flow field-temperature field model and calculating reactor temperature field data;

[0028] A sample data set generation module for preliminarily screening the temperature measurement points on the fuel tank wall according to the internal oil flow characteristics of the reactor in the calculation results, and then using the Latin hypercube sampling method to obtain sample data sets under different working conditions;

[0029] A sample secondary screening module for re-screening the preliminarily screened temperature measurement points on the fuel tank wall using the random forest algorithm with the sample data set as a sample, and sorting the temperature measurement points on the fuel tank wall according to the importance of the temperature measurement point characteristics;

[0030] The second model establishment module is used to establish a GA-BP neural network model, taking the oil tank wall measurement points under different measurement point combinations as inputs and using the calculated samples as a data set for training;

[0031] The optimal temperature measurement point calculation module is used to determine the optimal oil tank wall temperature measurement point combination based on the error index obtained from training.

[0032] Preferably, the first model establishment module introduces non-uniform loss into the fluid-temperature field coupling, constructs a weak coupling relationship between the magnetic field and the fluid-temperature field, describes the flow of transformer oil based on the Navier-Stokes equation, calculates the convective heat transfer coefficient according to the actual wind speed, and establishes a coupled calculation model of the flow field-temperature field;

[0033] Set the initial temperature and the ambient temperature to ensure that the initial values of the oil temperature in the oil tank, the oil tank wall, the oil temperature in the radiator, and the radiator wall are consistent with the ambient temperature. The initial fluid flow velocity is 0. Set the oil tank wall and the inner iron core surface to a no-slip state, determine the convective heat transfer coefficients of the top, side walls, and fins of the oil tank under windless conditions, clarify the heat source type and the characteristics of the fluid field, and set the direction and magnitude of gravity;

[0034] Add the iron core and winding losses as heat sources to the fluid-thermal coupling physical field, calculate the temperature field distribution of the reactor under rated voltage, standard capacity, specific ambient temperature, and wind speed, and obtain the hot spot temperature and the maximum temperature rise data; compare with the actual reactor temperature rise value to verify the effectiveness of the reactor temperature field data; analyze the temperature distribution laws of the oil tank and the fins.

[0035] Preferably, the sample data set generation module takes three factors, namely the external ambient temperature, the load current, and the external wind speed, as variables to simulate different working conditions, and uses the Latin hypercube sampling method to obtain the sample data set under different working conditions.

[0036] Preferably, the sample secondary screening module first uses the sparrow optimization algorithm to optimize the number and depth of the decision trees of the random forest model, then uses the random forest model to screen the oil tank wall temperature measurement points, and finally uses the immune algorithm to perform secondary screening on the oil tank wall temperature measurement points output by the random forest model and assigns an importance index corresponding to each secondarily screened oil tank wall temperature measurement point.

[0037] Preferably, the sample secondary screening module sets the number of sparrow populations, uses the number and depth of the decision trees of the random forest model to form a solution space, randomly assigns a position in the solution space to each sparrow, uses the output separation degree of different decision trees as the fitness function, and sets the maximum number of iterations;

[0038] Arrange the sparrows in descending order according to the current fitness, regard the top 20% of the sparrows as discoverers, regard the bottom 20% of the sparrows as vigilants, and the remaining sparrows as joiners;

[0039] The update method of the discoverers is where is the position of the \(i\)-th discoverer in the \(t\)-th iteration, is the position of the \(i\)-th discoverer in the \((t - 1)\)-th iteration, and \(\lambda\) is the perturbation coefficient;

[0040] The update method of the joiners is where is the position of the \(i\)-th joiner in the \(t\)-th iteration, is the position of the \(i\)-th joiner in the \((t - 1)\)-th iteration, is the cluster center position of all discoverers in the \(t\)-th iteration;

[0041] The update method of the sentinels is to randomly select a position in the solution space as the new position of the sentinels, and there are no discoverers and joiners at this position;

[0042] The sparrow population is iterated cyclically until the maximum number of iterations is reached or the optimal solution is obtained to stop the iteration. The position of the sparrow with the highest fitness at the end of the iteration is used as the number and depth of the decision trees of the random forest model.

[0043] Preferably, the sample secondary screening module uses the random forest model to screen the temperature measurement points on the fuel tank wall. Each random tree in the random forest model outputs a screening result. All the screening results are used as the initial antibody population, and the optimization objective function and constraints are used as antigens. Calculate the affinity between each antibody and the antigen, clone the antibodies with affinity greater than the upper threshold, mutate the antibodies with affinity less than the lower threshold, and use the remaining antibodies for peptide chain crossover. Repeat the above cloning, mutation, and crossover operations until the termination condition is met. At this time, the antibody with the greatest affinity with the antigen is used as the result of the secondary screening. The importance index of each temperature measurement point in the secondary screening result is proportional to the number of times the temperature measurement point appears in the screening results output by the random trees.

[0044] Preferably, the sample secondary screening module calculates the average deviation rate \(\eta\) between the secondary screening result and the screening results output by all random trees, and updates the perturbation coefficient \(\lambda\) according to the average deviation rate \(\eta\), where \(\lambda'\) is the updated perturbation coefficient.

[0045] Preferably, the optimal temperature measurement point calculation module starts from the dual temperature measurement points, uses the GA-BP neural network model to predict the hot spot temperature of the reactor, arranges the temperature measurement points in descending order according to the importance of the temperature measurement points, adds the temperature measurement point with the highest importance in the current sequence to the GA-BP neural network model, and predicts the hot spot temperature of the reactor again. After several cycles of prediction, the root mean square error, mean absolute error, and mean absolute percentage error are used as error indicators, and the error indicators under the current temperature measurement point combination are recorded after each prediction. By analyzing the error indicators, training convergence curve, and fitness curve, the training effect of the GA-BP neural network model for different measurement point combinations is evaluated, redundant measurement points are removed, and the optimal temperature measurement point combination is determined.

[0046] A computer device, comprising a processor and a memory, the memory is used to store at least one segment of computer program, and the at least one segment of computer program is loaded and executed by the processor to perform the above-mentioned method for inverting the hot spot of an oil-immersed reactor based on a random forest.

[0047] A computer-readable storage medium, which is used to store at least one segment of computer program, and the at least one segment of computer program is used to execute the above-mentioned method for inverting the hot spot of an oil-immersed reactor based on a random forest.

[0048] A computer program product, comprising a computer program, and when the computer program is executed by a processor, it implements the method for inverting the hot spot of an oil-immersed reactor based on a random forest as described above.

[0049] The beneficial effects brought by adopting the above technical solutions are as follows:

[0050] 1. The present invention uses the importance ranking of random forest features to optimize the measurement point combination, overcomes the subjectivity of traditional measurement point selection methods, improves the scientificity and rationality of measurement point selection, reduces redundant measurement points, and reduces measurement costs and data uncertainty.

[0051] 2. The present invention introduces the sparrow optimization algorithm to optimize the parameters of the random forest model. By improving the update methods of the discoverer, joiner, and vigilant in the sparrow optimization algorithm, the optimization efficiency of the random forest model parameters is accelerated. By comparing the secondary screening results of the immune algorithm with the output results of the random tree, the perturbation coefficient λ can be updated, which can further optimize the convergence speed of the sparrow optimization algorithm.

[0052] 3. The present invention introduces the immune algorithm to perform secondary screening on the output of the random forest model, and assigns importance indices to the temperature measurement points using the results of the two screenings, which is convenient for the prediction of the subsequent neural network model.

[0053] 4. The inversion model training based on the GA-BP neural network combines the advantages of the genetic algorithm and the BP neural network, improves the model training efficiency and prediction accuracy, enables the inversion model to better adapt to complex operating conditions, and achieves accurate prediction of the internal hot spot temperature of the reactor. Description of the Drawings

[0054] Figure 1 This is the principle flowchart of the present invention. Detailed Embodiments

[0055] Embodiment 1

[0056] Refer to Figure 1 , the inversion process of this embodiment includes the following steps.

[0057] Perform simulation modeling and calculation on the 22 kV experimental reactor. Introduce non-uniform loss into the fluid-temperature field coupling, construct the weak coupling relationship between the magnetic field and the fluid-temperature field, describe the flow of transformer oil according to the Navier-Stokes equation, calculate the convective heat transfer coefficient according to the actual wind speed, and establish the coupled calculation model of the flow field-temperature field.

[0058] Set the initial temperature and the ambient temperature to ensure that the initial values of the oil temperature in the oil tank, the oil tank wall, the oil temperature in the radiator and the radiator wall are consistent with the ambient temperature. The initial fluid flow velocity is 0. Set the oil tank wall and the surface of the internal iron core to be in a no-slip state. Determine the convective heat transfer coefficients of the top, side walls and fins of the oil tank under windless conditions, clarify the heat source type and the characteristics of the fluid field, and set the direction and magnitude of gravity.

[0059] Add the core and winding losses as heat sources to the fluid-thermal coupling physical field, calculate the temperature field distribution of the reactor under the rated voltage (22 kV), standard capacity (6.7 Mvar), specific ambient temperature (16 °C) and wind speed (0 m / s), and obtain the hot spot temperature and the maximum temperature rise data. Compare with the actual temperature rise value of the 22 kV experimental reactor to verify the effectiveness of the reactor temperature field data. Analyze the temperature distribution laws of the oil tank and the fins.

[0060] Preliminarily screen the temperature measurement points on the oil tank wall according to the internal oil flow characteristics of the 22 kV experimental reactor in the temperature field calculation results. Take the three factors of the external ambient temperature, load current and external wind speed as variables to simulate different working conditions, and use the Latin hypercube sampling method to obtain the sample data set under different working conditions.

[0061] First, use the sparrow optimization algorithm to optimize the number and depth of the decision trees of the random forest model. Then, use the random forest model to screen the temperature measurement points on the oil tank wall. Finally, use the immune algorithm to perform secondary screening on the temperature measurement points on the oil tank wall output by the random forest model, and assign the corresponding importance index to each temperature measurement point on the oil tank wall screened secondarily, and sort the temperature measurement points on the oil tank wall according to the importance of the temperature measurement points.

[0062] Optimizing the number and depth of decision trees in a random forest model using the sparrow optimization algorithm includes the following steps:

[0063] Set the sparrow population size. Use the number and depth of decision trees in the random forest model to form the solution space. Randomly assign a position in the solution space to each sparrow. Use the separation degree of the outputs of different decision trees as the fitness function, and set the maximum number of iterations.

[0064] Arrange the sparrows in descending order according to the current fitness. Take the top 20% of the sparrows as discoverers, the bottom 20% of the sparrows as vigilant ones, and the remaining sparrows as joiners.

[0065] The update method for discoverers is where is the position of the i-th discoverer in the t-th iteration, is the position of the i-th discoverer in the (t - 1)-th iteration, and λ is the perturbation coefficient.

[0066] The update method for joiners is where is the position of the i-th joiner in the t-th iteration, is the position of the i-th joiner in the (t - 1)-th iteration, is the clustering center position of all discoverers in the t-th iteration;

[0067] The update method for vigilant ones is to randomly select a position within the solution space as the new position of the vigilant one, where there are no discoverers and joiners at this position.

[0068] Perform cyclic iteration on the sparrow population until the maximum number of iterations is reached or the optimal solution is obtained and the iteration stops. The position of the sparrow with the highest fitness at the end of the iteration is used as the number and depth of decision trees in the random forest model.

[0069] Use the random forest model to screen the temperature measurement points on the fuel tank wall. Each random tree in the random forest model outputs a screening result. Take all the screening results as the initial antibody population. Use the optimization objective function and constraint conditions as the antigen. Calculate the affinity between each antibody and the antigen. Clone the antibodies with an affinity greater than the upper threshold, mutate the antibodies with an affinity less than the lower threshold, and perform peptide chain crossover using the remaining antibodies. Repeat the above cloning, mutation, and crossover operations until the termination condition is met. At this time, the antibody with the highest affinity to the antigen is used as the result of the secondary screening. The importance index of each temperature measurement point in the secondary screening result is proportional to the number of times the temperature measurement point appears in the screening results output by the random trees.

[0070] Calculate the average deviation rate η between the secondary screening result and the screening results output by all random trees, and update the perturbation coefficient λ according to the average deviation rate η. where λ′ is the updated perturbation coefficient.

[0071] Build a GA-BP neural network model and then train it with a sample data set.

[0072] Starting from the double temperature measurement points, use the GA-BP neural network model to predict the hot spot temperature of the reactor. Arrange the temperature measurement points in descending order according to the importance of the temperature measurement points. Add the temperature measurement point with the highest importance in the current sequence to the GA-BP neural network model, and predict the hot spot temperature of the reactor again. Repeat the prediction several times. Use the root mean square error, mean absolute error, and mean absolute percentage error as error indicators, and record the error indicators under the current temperature measurement point combination after each prediction. By analyzing the error indicators, training convergence curve, and fitness curve, evaluate the training effect of the GA-BP neural network model for different measurement point combinations, eliminate redundant measurement points, and determine the optimal temperature measurement point combination.

[0073] Aiming at the problems of strong subjectivity in the selection of measurement points, low inversion accuracy, and difficulty in adapting to complex operating conditions in the existing hot spot temperature prediction method for oil-immersed reactors, this invention combines a simulation model to accurately calculate the temperature field data of oil-immersed reactors under different working conditions, conducts a random forest feature importance analysis on the measured point temperature and hot spot temperature on the tank wall, combines an immune algorithm to find the measurement points with high importance, combines them for measurement point combination, and inversely finds the measurement points with high inversion accuracy to achieve high-precision hot spot inversion.

[0074] Embodiment 2

[0075] An oil-immersed reactor hot spot inversion device based on random forest for Embodiment 1, comprising

[0076] The first model establishment module is used to establish a reactor flow field-temperature field model and calculate the reactor temperature field data;

[0077] The sample data set generation module is used to preliminarily screen the temperature measurement points on the tank wall according to the internal oil flow characteristics of the reactor in the calculation results, and then use the Latin hypercube sampling method to obtain the sample data set under different working conditions;

[0078] The sample secondary screening module is used to re-screen the preliminarily screened temperature measurement points on the tank wall with the sample data set as the sample by using the random forest algorithm, and sort the temperature measurement points on the tank wall according to the feature importance of the temperature measurement points;

[0079] The second model establishment module is used to establish a GA-BP neural network model, and use the temperature measurement points on the tank wall under different measurement point combinations as inputs and the calculated samples as data sets for training;

[0080] The optimal temperature measurement point calculation module is used to determine the optimal temperature measurement point combination on the tank wall based on the error indicators obtained from training.

[0081] Specifically,

[0082] The first model establishment module introduces non-uniform loss into the fluid-temperature field coupling, constructs the weak coupling relationship between the magnetic field and the fluid-temperature field, describes the flow of transformer oil according to the Navier-Stokes equation, calculates the convective heat transfer coefficient according to the actual wind speed, and establishes a coupled calculation model of the flow field-temperature field;

[0083] Set the initial temperature and the ambient temperature to ensure that the initial values of the oil temperature in the oil tank, the oil tank wall, the oil temperature in the radiator and the radiator wall are the same as the ambient temperature. The initial fluid flow velocity is 0. Set the oil tank wall and the surface of the internal iron core to be in a no-slip state. Determine the convective heat transfer coefficients of the top, side walls and fins of the oil tank under windless conditions, clarify the heat source type and the characteristics of the fluid field, and set the direction and magnitude of gravity;

[0084] Take the core and winding losses as heat sources and add them to the fluid-thermal coupling physical field. Calculate the temperature field distribution of the reactor under rated voltage, standard capacity, specific ambient temperature and wind speed, and obtain the hot spot temperature and maximum temperature rise data; Compare with the actual value of the reactor temperature rise to verify the effectiveness of the reactor temperature field data; Analyze the temperature distribution law of the oil tank and the fins.

[0085] The sample data set generation module takes three factors, namely the external ambient temperature, the load current, and the external wind speed, as variables to simulate different working conditions, and uses the Latin hypercube sampling method to obtain the sample data set under different working conditions.

[0086] The sample secondary screening module first uses the sparrow optimization algorithm to optimize the number and depth of the decision trees of the random forest model, then uses the random forest model to screen the temperature measurement points on the oil tank wall, and finally uses the immune algorithm to perform secondary screening on the temperature measurement points on the oil tank wall output by the random forest model, and assigns the corresponding importance index to each temperature measurement point on the oil tank wall after secondary screening.

[0087] The sample secondary screening module sets the number of sparrow populations, uses the number and depth of the decision trees of the random forest model to form the solution space, randomly assigns a position in the solution space to each sparrow, uses the separation degree of the output of different decision trees as the fitness function, and sets the maximum number of iterations;

[0088] Arrange the sparrows in descending order according to the current fitness, take the top 20% of the sparrows as discoverers, take the bottom 20% of the sparrows as vigilants, and the remaining sparrows as joiners;

[0089] The update method of the discoverer is where is the position of the i-th discoverer in the t-th iteration, is the position of the i-th discoverer in the (t - 1)-th iteration, and λ is the perturbation coefficient;

[0090] The update method of the joiner is where is the position of the i-th joiner at the t-th iteration, is the position of the i-th joiner at the (t - 1)-th iteration, is the cluster center position of all discoverers at the t-th iteration;

[0091] The update method of the vigilant is to randomly select a position in the solution space as the new position of the vigilant, and there are no discoverers and joiners at this position;

[0092] The sparrow population is iterated cyclically until the maximum number of iterations is reached or the optimal solution is obtained to stop the iteration. The position of the sparrow with the highest fitness at the end of the iteration is used as the number and depth of decision trees of the random forest model.

[0093] The sample secondary screening module uses the random forest model to screen the temperature measurement points on the fuel tank wall. Each random tree in the random forest model outputs a screening result. All the screening results are used as the initial antibody population, and the optimization objective function and constraint conditions are used as antigens. Calculate the affinity between each antibody and the antigen, clone the antibodies with affinity greater than the upper threshold, mutate the antibodies with affinity less than the lower threshold, and use the remaining antibodies for peptide chain crossover. Repeat the above cloning, mutation, and crossover operations until the termination condition is met. At this time, the antibody with the greatest affinity with the antigen is used as the result of the secondary screening. The importance index of each temperature measurement point in the secondary screening result is proportional to the number of times the temperature measurement point appears in the screening result output by the random tree.

[0094] The sample secondary screening module calculates the average deviation rate η between the secondary screening result and the screening results output by all random trees, and updates the perturbation coefficient λ according to the average deviation rate η, where λ′ is the updated perturbation coefficient.

[0095] The optimal temperature measurement point calculation module starts from the double temperature measurement points, uses the GA-BP neural network model to predict the hot spot temperature of the reactor, sorts the temperature measurement points in descending order according to the importance of the temperature measurement points, adds the temperature measurement point with the highest importance in the current sequence to the GA-BP neural network model, and predicts the hot spot temperature of the reactor again. Predict cyclically for several times, and use the root mean square error, mean absolute error, and mean absolute percentage error as error indicators. Record the error indicators under the current temperature measurement point combination after each prediction; by analyzing the error indicators, training convergence curve, and fitness curve, evaluate the training effect of the GA-BP neural network model for different measurement point combinations, eliminate redundant measurement points, and determine the optimal temperature measurement point combination.

[0096] Example 3

[0097] A computer device, comprising a processor and a memory, where the memory is used to store at least one computer program, and the at least one computer program is loaded and executed by the processor to implement the method for inverting hot spots of oil-immersed reactors based on random forests described in Embodiment 1.

[0098] Embodiment 4

[0099] A computer-readable storage medium, which is used to store at least one computer program, and the at least one computer program is used to execute the method for inverting hot spots of oil-immersed reactors based on random forests described in Embodiment 1.

[0100] Embodiment 5

[0101] A computer program product, including a computer program, which implements the method for inverting hot spots of oil-immersed reactors based on random forests described in Embodiment 1 when executed by a processor.

[0102] In the description of the present invention, it should be understood that the orientation or positional relationship indicated by the terms "longitudinal", "transverse", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", etc. is based on the orientation or positional relationship shown in the drawings, and is only for the convenience of describing the present invention, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and thus cannot be understood as a limitation to the present invention.

[0103] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited by the above embodiments, and what is described in the above embodiments and the specification only illustrates the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention will have various changes and improvements, and these changes and improvements all fall within the scope of the present invention claimed. The scope of protection claimed by the present invention is defined by the appended claims and their equivalents.

Claims

1. A hot spot inversion method for oil-immersed reactor based on random forest, characterized in that The following steps are involved: Establish the reactor flow field-temperature field model and calculate the reactor temperature field data; The oil tank wall temperature measurement points were preliminarily selected based on the reactor internal oil flow characteristics in the reactor temperature field data, and the Latin hypercube sampling method was used to obtain sample data sets under different working conditions. The sample data set is used as a sample by using a random forest algorithm to screen the initially screened fuel tank wall measuring points again, and the fuel tank wall temperature measuring points are sorted according to the importance of the temperature measuring point features; A GA-BP neural network model is established, with the tank wall measurement points under different measurement point combinations as input and the calculated samples as the data set for training; based on the error index obtained from the training, the optimal tank wall temperature measurement point combination is determined.

2. The hot spot inversion method for oil-immersed reactor based on random forest according to claim 1, characterized in that: Establish the reactor flow field-temperature field model and calculate the reactor temperature field data, including: The non-uniform loss introduces fluid-temperature field coupling, constructs the weak coupling relationship between magnetic field and fluid-temperature field, describes the transformer oil flow according to the Navier-Stokes equation, calculates the convection heat transfer coefficient according to the actual wind speed, and establishes the flow field-temperature field coupling calculation model; Set the initial temperature and ambient temperature, ensure that the initial values ​​of the oil temperature in the tank, the tank wall, the oil temperature in the radiator and the radiator wall are consistent with the ambient temperature, the initial fluid flow velocity is 0, set the tank wall and the internal iron core surface to a non-slip state, determine the convection heat transfer coefficient of the tank top, side wall and heat sink under windless conditions, clarify the heat source type and fluid field characteristics, and set the gravity direction and magnitude; The core and winding losses are added as heat sources into the flow-heat coupling physical field, and the temperature field distribution of the reactor is calculated under rated voltage, standard capacity, specific ambient temperature and wind speed, and the hot spot temperature and maximum temperature rise data are obtained. The data are compared with the actual value of the reactor temperature rise to verify the validity of the reactor temperature field data; and the temperature distribution law of the oil tank and heat sink is analyzed.

3. The hot spot inversion method for oil-immersed reactor based on random forest according to claim 2 is characterized in that: According to the oil flow characteristics inside the reactor in the calculation results, the temperature measurement points on the oil tank wall are preliminarily selected, and then the Latin hypercube sampling method is used to obtain sample data sets under different working conditions, including taking the external ambient temperature, load current, and external wind speed as variables to simulate different working conditions, and using the Latin hypercube sampling method to obtain sample data sets under different working conditions.

4. The hot spot inversion method for oil-immersed reactor based on random forest according to claim 3 is characterized in that: The random forest algorithm is used to select the sample data set, and the initially selected fuel tank wall measurement points are screened again, and the fuel tank wall temperature measurement points are sorted according to the importance of the temperature measurement point features. include, First, the sparrow optimization algorithm is used to optimize the number and depth of decision trees of the random forest model, and then the random forest model is used to screen the temperature measurement points on the tank wall. Finally, the immune algorithm is used to perform secondary screening on the temperature measurement points on the tank wall output by the random forest model, and each secondary screened temperature measurement point on the tank wall is assigned a corresponding importance index.

5. The hot spot inversion method for oil-immersed reactor based on random forest according to claim 4 is characterized in that: Using the sparrow optimization algorithm to optimize the number and depth of decision trees in the random forest model includes the following steps: Set the number of sparrow populations, use the number and depth of random forest model decision trees to form the solution space, randomly assign a position in the solution space to each sparrow, use the output separation of different decision trees as the fitness function, and set the maximum number of iterations; Arrange the sparrows in descending order according to their current fitness, and use the first 20% of the sparrows as discoverers, the last 20% of the sparrows as guards, and the rest of the sparrows as joiners; The update method of the discoverer is in is the position of the i-th discoverer in the t-th iteration, is the position of the i-th discoverer in the t-1th iteration, and λ is the perturbation coefficient; The update method for joiners is in is the position of the i-th joiner in the t-th iteration, is the position of the i-th joiner in the t-1th iteration, is the cluster center position of all discoverers in the tth iteration; The update method of the alerter is to randomly select a position in the solution space as the position of the alerter in the new round. There are no discoverers or joiners at this position. The sparrow population is iterated cyclically until the maximum number of iterations is reached or the optimal solution is obtained, and the iteration is stopped. The position of the sparrow with the highest fitness when the iteration is stopped is used as the number and depth of decision trees of the random forest model.

6. The hot spot inversion method for oil-immersed reactor based on random forest according to claim 5 is characterized in that: The random forest model is used to screen the temperature measuring points on the tank wall. Each random tree in the random forest model outputs a screening result. All the screening results are used as the initial antibody population, and the optimization objective function and constraints are used as the antigen. The affinity of each antibody to the antigen is calculated, and the antibodies with affinity greater than the upper threshold are cloned, and the antibodies with affinity less than the lower threshold are mutated. The remaining antibodies are used for peptide chain crossover. The above cloning, mutation and crossover operations are repeated until the termination condition is met. At this time, the antibody with the largest affinity to the antigen is used as the result of the secondary screening. The importance index of each temperature measuring point in the secondary screening result is proportional to the number of times the temperature measuring point appears in the screening result output by the random tree.

7. The hot spot inversion method for oil-immersed reactor based on random forest according to claim 6 is characterized in that: Calculate the average deviation rate η between the secondary screening results and the screening results output by all random trees, and update the disturbance coefficient λ according to the average deviation rate η. Where λ′ is the updated perturbation coefficient.

8. The hot spot inversion method for oil-immersed reactor based on random forest according to claim 7, characterized in that: Based on the error index obtained from training, the optimal combination of tank wall temperature measurement points is determined, including: Starting from the dual temperature measurement points, the GA-BP neural network model is used to predict the hot spot temperature of the reactor. The temperature measurement points are arranged in descending order based on the importance of the temperature measurement points. The temperature measurement point with the highest importance in the current sequence is added to the GA-BP neural network model, and the hot spot temperature of the reactor is predicted again. The prediction is repeated several times, and the root mean square error, mean absolute error, and mean absolute percentage error are used as error indicators. After each prediction, the error indicators under the current temperature measurement point combination are recorded; by analyzing the error indicators, training convergence curves, and fitness curves, the training effect of the GA-BP neural network model for different measurement point combinations is evaluated, redundant measurement points are eliminated, and the optimal temperature measurement point combination is determined.

9. A hot spot inversion device for oil-immersed reactor based on random forest, characterized in that: include, The first model building module is used to build a reactor flow field-temperature field model and calculate the reactor temperature field data; The sample data set generation module is used to preliminarily select the temperature measurement points on the oil tank wall according to the oil flow characteristics inside the reactor in the calculation results, and then use the Latin hypercube sampling method to obtain sample data sets under different working conditions; The sample secondary screening module is used to re-screen the initially screened fuel tank wall measuring points using the sample data set through the random forest algorithm, and sort the fuel tank wall temperature measuring points according to the importance of the temperature measuring point features; The second model building module is used to build a GA-BP neural network model, using the tank wall measurement points under different measurement point combinations as input and the calculated samples as data sets for training; The optimal temperature measurement point calculation module is used to determine the optimal combination of tank wall temperature measurement points based on the error index obtained through training.

10. The hot spot inversion device for oil-immersed reactor based on random forest according to claim 9, characterized in that: The first model building module introduces non-uniform loss into the fluid-temperature field coupling, constructs the weak coupling relationship between the magnetic field and the fluid-temperature field, describes the transformer oil flow according to the Navier-Stokes equation, calculates the convection heat transfer coefficient according to the actual wind speed, and establishes the flow field-temperature field coupling calculation model; Set the initial temperature and ambient temperature, ensure that the initial values ​​of the oil temperature in the tank, the tank wall, the oil temperature in the radiator and the radiator wall are consistent with the ambient temperature, the initial fluid flow velocity is 0, set the tank wall and the internal iron core surface to a non-slip state, determine the convection heat transfer coefficient of the tank top, side wall and heat sink under windless conditions, clarify the heat source type and fluid field characteristics, and set the gravity direction and magnitude; The core and winding losses are added as heat sources into the flow-heat coupling physical field, and the temperature field distribution of the reactor is calculated under rated voltage, standard capacity, specific ambient temperature and wind speed, and the hot spot temperature and maximum temperature rise data are obtained. The data are compared with the actual value of the reactor temperature rise to verify the validity of the reactor temperature field data; and the temperature distribution law of the oil tank and heat sink is analyzed.

11. The hot spot inversion device for oil-immersed reactor based on random forest according to claim 10, characterized in that: The sample data set generation module uses external ambient temperature, load current, and external wind speed as variables to simulate different working conditions, and uses the Latin hypercube sampling method to obtain sample data sets under different working conditions.

12. The hot spot inversion device for oil-immersed reactor based on random forest according to claim 11, characterized in that: The sample secondary screening module first uses the sparrow optimization algorithm to optimize the number and depth of decision trees of the random forest model, and then uses the random forest model to screen the temperature measurement points on the fuel tank wall. Finally, the immune algorithm is used to perform secondary screening on the fuel tank wall temperature measurement points output by the random forest model, and each secondary screened fuel tank wall temperature measurement point is assigned a corresponding importance index.

13. The hot spot inversion device for oil-immersed reactor based on random forest according to claim 12, characterized in that: The sample secondary screening module sets the number of sparrow populations, uses the number and depth of decision trees in the random forest model to form a solution space, randomly assigns a position in the solution space to each sparrow, uses the output separation of different decision trees as the fitness function, and sets the maximum number of iterations; Arrange the sparrows in descending order according to their current fitness, and use the first 20% of the sparrows as discoverers, the last 20% of the sparrows as guards, and the rest of the sparrows as joiners; The update method of the discoverer is in is the position of the i-th discoverer in the t-th iteration, is the position of the i-th discoverer in the t-1th iteration, and λ is the perturbation coefficient; The update method for joiners is in is the position of the i-th joiner in the t-th iteration, is the position of the i-th joiner in the t-1th iteration, is the cluster center position of all discoverers in the tth iteration; The update method of the alerter is to randomly select a position in the solution space as the position of the alerter in the new round. There are no discoverers or joiners at this position. The sparrow population is iterated cyclically until the maximum number of iterations is reached or the optimal solution is obtained, and the iteration is stopped. The position of the sparrow with the highest fitness when the iteration is stopped is used as the number and depth of decision trees of the random forest model.

14. The hot spot inversion device for oil-immersed reactor based on random forest according to claim 13, characterized in that: The sample secondary screening module uses the random forest model to screen the temperature measuring points on the tank wall. Each random tree in the random forest model outputs a screening result. All screening results are used as the initial antibody population, and the optimization objective function and constraints are used as the antigen. The affinity of each antibody to the antigen is calculated, and the antibodies with affinity greater than the upper threshold are cloned, and the antibodies with affinity less than the lower threshold are mutated. The remaining antibodies are used for peptide chain crossover, and the above cloning, mutation and crossover operations are repeated until the termination condition is met. At this time, the antibody with the largest affinity with the antigen is used as the result of the secondary screening. The importance index of each temperature measuring point in the secondary screening result is proportional to the number of times the temperature measuring point appears in the screening result output by the random tree.

15. The hot spot inversion device for oil-immersed reactor based on random forest according to claim 14, characterized in that: The sample secondary screening module calculates the average deviation rate η between the secondary screening results and the screening results output by all random trees, and updates the disturbance coefficient λ according to the average deviation rate η. Where λ′ is the updated perturbation coefficient.

16. The hot spot inversion device for oil-immersed reactor based on random forest according to claim 15, characterized in that: The optimal temperature measurement point calculation module starts from the dual temperature measurement points, uses the GA-BP neural network model to predict the hot spot temperature of the reactor, arranges the temperature measurement points in descending order based on the importance of the temperature measurement points, adds the most important temperature measurement point in the current sequence to the GA-BP neural network model, and predicts the hot spot temperature of the reactor again. The prediction is repeated several times, and the root mean square error, mean absolute error, and mean absolute percentage error are used as error indicators. After each prediction, the error indicators under the current temperature measurement point combination are recorded; by analyzing the error indicators as well as the training convergence curve and the fitness curve, the training effect of the GA-BP neural network model for different measurement point combinations is evaluated, the redundant measurement points are eliminated, and the optimal temperature measurement point combination is determined.

17. A computer device, characterized in that: The computer device includes a processor and a memory, the memory is used to store at least one computer program, and the at least one computer program is loaded by the processor and executes the hot spot inversion method of an oil-immersed reactor based on random forest as described in any one of claims 1 to 8.

18. A computer-readable storage medium, characterized in that: The computer-readable storage medium is used to store at least one computer program, and the at least one computer program is used to execute the hot spot inversion method of an oil-immersed reactor based on random forest as described in any one of claims 1 to 8.

19. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the random forest-based hot spot inversion method for an oil-immersed reactor as described in any one of claims 1 to 8 is implemented.

Images