Hydro-generator stator core temperature distribution analysis method

Abstract

The invention provides a hydro-generator stator core temperature distribution analysis method, which aims at other common analysis methods for the temperature of a large-scale generator core, including monthly maximum value year-on-year and ring-on-year methods, and further deep analysis is carried out when the difference between the year-on-year and ring-on-year is large. Meanwhile, the maximum temperature deviation value of all the bushes of the bearing at the same moment can be analyzed, and further deep analysis is carried out when the temperature deviation value is large. An existing large-scale generator bearing bush temperature analysis method is few in content, the temperature state of multiple bushes cannot be analyzed, and the state and trend of the bush temperature cannot be analyzed in the spatial dimension. The hydro-generator stator core temperature distribution analysis method has remarkable beneficial effects in the aspects of improving real-time performance and accuracy, comprehensively analyzing data, expanding bearing bush temperature analysis and the like, and safe and stable operation of a hydro-generator is guaranteed.

Description

Technical Field

[0001] The invention relates to a method for analyzing temperature distribution of a stator core of a hydro-generator, in particular to a method for analyzing temperature distribution of a stator core of a hydro-generator. Background Art

[0002] A hydro-turbine generator is a device that converts water flow energy into electrical energy. Its stator core is one of the core components of the generator. The temperature distribution of the stator core has an important influence on the operating stability and life of the generator. Therefore, the study of the analysis method of the temperature distribution of the stator core of the hydro-turbine generator is of great significance.

[0003] The traditional analysis method of temperature distribution of the stator core of a hydro-turbine generator is mainly based on the empirical formula method. The empirical formula is an empirical law derived from experimental data or theory, which can be used to preliminarily estimate the temperature distribution of the stator core. However, due to the complex and changeable working conditions of hydro-turbine generators, there may be large differences between different models of hydro-turbine generators, so the accuracy of the empirical formula has certain limitations.

[0004] Another commonly used method is finite element simulation. Finite element simulation is to establish a mathematical model, divide the stator core into multiple small units, and perform numerical calculations based on boundary conditions and material properties to obtain the temperature distribution of the stator core. This method can simulate the temperature distribution of the stator core more accurately, but it requires a lot of computing resources and time.

[0005] In recent years, with the development of sensor technology and data processing technology, the temperature distribution analysis method of the stator core of a hydro-turbine generator based on sensor data has gradually attracted attention. This method arranges multiple temperature sensors on the stator core to collect the temperature data of the stator core in real time. Through the industrial Internet platform, the time series data of all hydropower station equipment is aggregated into the database, and the data analysis algorithm provided by the technical center is used to process and analyze the data to obtain the temperature distribution of the stator core. This method has the advantages of real-time performance and high accuracy, and can better monitor and control the operating status of the hydro-turbine generator. Summary of the invention

[0006] The main purpose of the present invention is to provide a method for analyzing the temperature distribution of a stator core of a hydro-generator, so as to solve the problems of poor real-time performance and insufficient accuracy of the existing methods.

[0007] In order to solve the above technical problems, the technical solution adopted by the present invention is: a method for analyzing the temperature distribution of a stator core of a hydro-generator, comprising the following steps: S1. Arrange multiple temperature sensors on the stator core; S2, collecting temperature data of the stator core in real time through a data acquisition device; S3, processing and storing the collected temperature data; S4. Through the industrial Internet platform, the core temperature time series data is aggregated into the database, and then the unit's operating parameters and operation status parameters are pre-processed regularly; S5, analyzing the processed temperature data through a data analysis algorithm; S6. Generate a stator core temperature distribution diagram and a temperature change trend diagram according to the analysis results; S7. Diagnose and issue warning events and abnormal events based on the core temperature distribution and change trend; S8. Draw conclusions and treatment suggestions based on manual determination of early warning events and special analysis of abnormal events. Finally, based on the previous statistics and analysis, make equipment operation and maintenance suggestions, including equipment operation suggestions and equipment maintenance suggestions.

[0008] In a preferred embodiment, the temperature sensor is connected to the data acquisition device via a wired or wireless method.

[0009] In a preferred embodiment, the data processing step includes data filtering, denoising, and calibration processing steps.

[0010] In the preferred embodiment, the data analysis algorithm includes the steps of temperature distribution feature extraction and temperature change trend prediction and analysis; The core temperature of all units is analyzed by median; Using the Industrial Internet platform, calculate the median temperature of the cores of all units in the current analysis period, compare and analyze the median temperatures of the cores of multiple units, and add month-on-month and year-on-year analysis, so as to analyze the difference between the overall temperature of the core and other units of the same model / different models / adjacent / non-adjacent units in the power station; This analysis is not only helpful in discovering the overall temperature changes of components, but also in discovering the overall temperature differences between components of different units. For components with large differences, subsequent special analyses such as manual analysis can be carried out.

[0011] In the preferred embodiment, the analysis algorithm for extracting temperature distribution features is a method combining singular value decomposition and wavelet transform, and the steps include: A1. First, collect and organize the core temperature data of all units to form a temperature data set. ,in Indicates The temperature value set of each unit at different time points. is the total number of units, is the time point; For the temperature dataset The matrix formed Perform singular value decomposition, ,in and is an orthogonal matrix, is a diagonal matrix, and the elements on its diagonal are singular values. The left singular vectors and right singular vectors corresponding to the larger singular values ​​form a new matrix and , then the temperature distribution characteristic matrix ; A2. Then, for the temperature time series collected by each sensor, wavelet transform is used to further extract local features; let the temperature change over time be , the wavelet coefficient calculation formula is: ; in is the average temperature, is the wavelet basis function, are the wavelet coefficients, and They are scale and translation parameters respectively; A3. Combine the wavelet coefficients with the characteristic matrix obtained by singular value decomposition to obtain the final temperature distribution characteristic matrix ,in It is a weight function used to adjust the contribution of wavelet coefficients in the feature matrix.

[0012] In the preferred solution, the core temperature adopts median analysis, and the temperature median calculation method is: B1. Calculate the median temperature of all units in the current analysis period , first merge the temperature data of all units into one set ,in is the total number of all temperature data, Then To sort, if is an odd number, then the median ;like is an even number, then the median ; B2. For each unit, calculate its median temperature during the current analysis period. , Similarly, the temperature data of each unit is sorted, and the median is calculated according to the odd and even cases respectively; formula: , ;in, is the median temperature of all units, It is The median temperature of the units, Indicates that after sorting in ascending order, The temperature value of the location, is the total number of all temperature data, It is Total number of temperature data of each unit; The usefulness of the formula is that by calculating the median, it can more accurately reflect the central trend of the temperature data and resist the influence of outliers.

[0013] In the preferred solution, the core temperature is analyzed by median, and the median core temperatures of multiple units are compared and analyzed. The median comparison analysis method is: C1. Perform comparative analysis and calculate the difference value , and conduct both month-on-month and year-on-year analysis. The month-on-month analysis is to compare with the median temperature of the previous time period and calculate the month-on-month growth rate. ; Year-on-year analysis is to compare the median temperature with the same period last year and calculate the year-on-year growth rate ; C2. Use standardized difference to measure the difference between two sets of data. The calculation formula is: ; in It is The median of the group data, is the average of all medians, is the standard deviation; The usefulness of the above formula is to understand the degree of deviation of a unit from the average level and discover potential problem units through comparative analysis and standardized differences.

[0014] In the preferred solution, the overall temperature trend between components of different units is predicted according to the overall temperature change of the components. The prediction method includes: D1. Combining kernel-based support vector regression and ARIMA model to predict temperature change trend, using Gaussian kernel function ,in is the kernel parameter, which maps the temperature time series data to a high-dimensional feature space and performs linear regression in the high-dimensional feature space. D2. Algorithm construction optimization: , subject to , , ,in and is the regression coefficient, and is the slack variable, is the penalty parameter, is an insensitive loss parameter, Is to input data Functions that map to high-dimensional feature spaces; D3. By solving the optimization problem in step D2, we get the regression function , used to preliminarily predict the temperature value at a future time point; D4. At the same time, the ARIMA model is used to further predict the trend of the temperature series. The temperature series is , the expression of the ARIMA model is: ,in is the predicted temperature value, is the mean term, is the autoregressive part, is the difference order, is a white noise process; D5. Finally, the prediction results of the SVR and ARIMA models are weighted and fused to obtain the final temperature change trend prediction value.

[0015] In the preferred solution, the overall temperature change of the components is analyzed, and it is helpful to find the overall temperature difference between different unit components. The overall temperature difference analysis method includes: E1. The difference analysis is performed by combining principal component analysis with a custom difference measure. First, principal component analysis is used to identify the temperature distribution differences between different units. The temperature data matrix is , the data matrix after dimensionality reduction is obtained through PCA ,in is a matrix consisting of the first few principal components; E2. Then, define a custom difference measurement function , which is used to measure the difference in temperature distribution between the two units. This difference measurement function can combine multiple features of temperature distribution, including mean, variance, median, and principal component information obtained by PCA, and can be defined as: ; in and are the median temperatures of the two units, and are the temperature standard deviations of the two units, and The two units are respectively The projection coefficients on the principal components are is the number of principal components; The usefulness of the formula is to reduce the dimension and highlight the main variation direction of the data through PCA, and at the same time combine it with a custom difference measurement function to more comprehensively discover the significant differences in temperature distribution between different units.

[0016] In the preferred embodiment, the positions of the temperature sensors are reasonably arranged according to the structural characteristics and temperature distribution law of the stator core.

[0017] The present invention provides a method for analyzing the temperature distribution of the stator core of a hydro-turbine generator. By arranging multiple temperature sensors on the stator core and using a data acquisition device to collect temperature data in real time, the core temperature time series data is collected to a database for processing and analysis with the help of an industrial Internet platform, which effectively solves the problems of poor real-time performance and insufficient accuracy of the existing methods. It can timely and accurately reflect the temperature changes of the stator core, providing reliable protection for the safe operation of the equipment.

[0018] It not only uses a unique data analysis algorithm to analyze the processed temperature data and generate stator core temperature distribution diagrams and temperature change trend diagrams, but also can diagnose and issue warning events and abnormal events based on the core temperature distribution and change trend, and conduct manual determination and special analysis, give conclusions and treatment suggestions, and finally put forward equipment operation and maintenance suggestions. It realizes the full process management from data collection to analysis, warning, and processing.

[0019] For the core temperature of large generators, in addition to the core analysis steps of this method, the monthly maximum year-on-year and month-on-month methods are also combined. When the year-on-year and month-on-month differences are large, further in-depth analysis can more comprehensively grasp the changing trends and abnormal conditions of the core temperature.

[0020] The existing large generator bearing temperature analysis methods have relatively little content, while this method can not only analyze the maximum temperature deviation of all bearing bearings at the same time, and further analyze when the temperature deviation value is large, but also analyze the state and trend of bearing temperature in the spatial dimension, making up for the shortcomings of the existing methods and providing richer information for the overall operating status evaluation of the generator.

[0021] In summary, the proposed temperature distribution analysis method for the stator core of a hydro-turbine generator has significant beneficial effects in improving real-time performance and accuracy, comprehensive data analysis, and expanding bearing temperature analysis, which helps to ensure the safe and stable operation of the hydro-turbine generator. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] The present invention will be further described below in conjunction with the accompanying drawings and embodiments: Figure 1 It is the core temperature median analysis data table of the present invention; DETAILED DESCRIPTION Example 1 A method for analyzing temperature distribution of a stator core of a hydro-generator comprises the following steps: S1. Arrange multiple temperature sensors on the stator core; S2, collecting temperature data of the stator core in real time through a data acquisition device; S3, processing and storing the collected temperature data; S4. Through the industrial Internet platform, the core temperature time series data is aggregated into the database, and then the unit's operating parameters and operation status parameters are regularly pre-processed.

[0023] S5, analyzing the processed temperature data through a data analysis algorithm; S6. Generate a stator core temperature distribution diagram and a temperature change trend diagram according to the analysis results; S7. Diagnose and issue warning events and abnormal events based on the core temperature distribution and change trend; S8. Draw conclusions and treatment suggestions based on manual determination of early warning events and special analysis of abnormal events. Finally, based on the previous statistics and analysis, make equipment operation and maintenance suggestions, including equipment operation suggestions and equipment maintenance suggestions.

[0024] In a preferred embodiment, the temperature sensor is connected to the data acquisition device via a wired or wireless method.

[0025] In a preferred embodiment, the data processing step includes data filtering, denoising, and calibration processing steps.

[0026] In the preferred embodiment, the data analysis algorithm includes the steps of temperature distribution feature extraction and temperature change trend prediction and analysis; The core temperature of all units is analyzed by median; Using the Industrial Internet platform, calculate the median temperature of the cores of all units in the current analysis period, compare and analyze the median temperatures of the cores of multiple units, and add month-on-month and year-on-year analysis, so as to analyze the difference between the overall temperature of the core and other units of the same model / different models / adjacent / non-adjacent units in the power station; This analysis is not only helpful in discovering the overall temperature changes of components, but also in discovering the overall temperature differences between components of different units. For components with large differences, subsequent special analyses such as manual analysis can be carried out.

[0027] In the preferred embodiment, the analysis algorithm for extracting temperature distribution features is a method combining singular value decomposition and wavelet transform, and the steps include: A1. First, collect and organize the core temperature data of all units to form a temperature data set. ,in Indicates The temperature value set of each unit at different time points. is the total number of units, is the time point; For the temperature dataset The matrix formed Perform singular value decomposition, ,in and is an orthogonal matrix, is a diagonal matrix, and the elements on its diagonal are singular values. The left singular vectors and right singular vectors corresponding to the larger singular values ​​form a new matrix and , then the temperature distribution characteristic matrix ; A2. Then, for the temperature time series collected by each sensor, wavelet transform is used to further extract local features; let the temperature change over time be , the wavelet coefficient calculation formula is: ; in is the average temperature, is the wavelet basis function, are the wavelet coefficients, and They are scale and translation parameters respectively; A3. Combine the wavelet coefficients with the characteristic matrix obtained by singular value decomposition to obtain the final temperature distribution characteristic matrix ,in It is a weight function used to adjust the contribution of wavelet coefficients in the feature matrix.

[0028] formula: , , , .in, is a matrix consisting of temperature data, , and is the matrix and diagonal matrix obtained by singular value decomposition, Is the selected The matrix of left singular vectors corresponding to the larger singular values, is the temperature distribution characteristic matrix obtained by singular value decomposition, are the wavelet coefficients, is the wavelet basis function, and are the scale and translation parameters, is the final temperature distribution characteristic matrix, is the weight function. The usefulness of this formula is to extract the main and local features of temperature data by combining singular value decomposition and wavelet transform, reduce the data dimension, and capture the local and global characteristics of temperature distribution at the same time.

[0029] In the preferred solution, the core temperature adopts median analysis, and the temperature median calculation method is: B1. Calculate the median temperature of all units in the current analysis period , first merge the temperature data of all units into one set ,in is the total number of all temperature data, Then To sort, if is an odd number, then the median ;like is an even number, then the median ; B2. For each unit, calculate its median temperature during the current analysis period. , Similarly, the temperature data of each unit is sorted, and the median is calculated according to the odd and even cases respectively; formula: , ;in, is the median temperature of all units, It is The median temperature of the units, Indicates that the The temperature value of the location, is the total number of all temperature data, It is Total number of temperature data of each unit; The usefulness of the formula is that by calculating the median, it can more accurately reflect the central trend of the temperature data and resist the influence of outliers.

[0030] In the preferred solution, the core temperature is analyzed by median, and the median core temperatures of multiple units are compared and analyzed. The median comparison analysis method is: C1. Perform comparative analysis and calculate the difference value , and conduct both month-on-month and year-on-year analysis. The month-on-month analysis is to compare with the median temperature of the previous time period and calculate the month-on-month growth rate. ; Year-on-year analysis is to compare the median temperature with the same period last year and calculate the year-on-year growth rate ; C2. Use standardized difference to measure the difference between two sets of data. The calculation formula is: ; in It is The median of the group data, is the average of all medians, is the standard deviation; formula: , , , .in, is the difference value, and The month-on-month growth rate and year-on-year growth rate are respectively is the standardized difference, is the median temperature of all units, It is The median temperature of the units, is the average of all medians, is the standard deviation. The usefulness of this formula is to understand the degree of deviation of a unit from the average level and discover potential problem units through comparative analysis and standardized differences.

[0031] In the preferred solution, the overall temperature trend between components of different units is predicted according to the overall temperature change of the components. The prediction method includes: D1. Combining kernel-based support vector regression and ARIMA model to predict temperature change trend, using Gaussian kernel function ,in is the kernel parameter, which maps the temperature time series data to a high-dimensional feature space and performs linear regression in the high-dimensional feature space. D2. Algorithm construction optimization: , subject to , , ,in and is the regression coefficient, and is the slack variable, is the penalty parameter, is an insensitive loss parameter, Is to input data Functions that map to high-dimensional feature spaces; D3. By solving the optimization problem in step D2, we get the regression function , used to preliminarily predict the temperature value at a future time point; D4. At the same time, the ARIMA model is used to further predict the trend of the temperature series. The temperature series is , the expression of the ARIMA model is: ,in is the predicted temperature value, is the mean term, is the autoregressive part, is the difference order, is a white noise process; D5. Finally, the prediction results of the SVR and ARIMA models are weighted and fused to obtain the final temperature change trend prediction value.

[0032] formula: , , subject to , , , , .in, It is a kernel function used to map data into a high-dimensional feature space; the optimization problem is used to solve the regression coefficient; is the regression function of SVR, which is used to predict the temperature value; is the temperature value predicted by the ARIMA model, is the mean term, is the autoregressive part, is the difference order, is a white noise process. The usefulness of this formula is that by combining the SVR and ARIMA models, the advantages of the two methods are fully utilized to improve the accuracy of temperature change trend prediction.

[0033] In the preferred solution, the overall temperature change of the components is analyzed and it is helpful to find the overall temperature difference between different unit components. The overall temperature difference analysis method includes: E1. The difference analysis is performed by combining principal component analysis with a custom difference measure. First, principal component analysis is used to identify the temperature distribution differences between different units. The temperature data matrix is , the data matrix after dimensionality reduction is obtained through PCA ,in is a matrix consisting of the first few principal components; E2. Then, define a custom difference measurement function , which is used to measure the difference in temperature distribution between the two units. This difference measurement function can combine multiple features of temperature distribution, including mean, variance, median, and principal component information obtained by PCA, and can be defined as: ; in and are the median temperatures of the two units, and are the temperature standard deviations of the two units, and The two units are respectively The projection coefficients on the principal components are is the number of principal components; The usefulness of the formula is to reduce the dimension and highlight the main variation direction of the data through PCA, and at the same time combine it with a custom difference measurement function to more comprehensively discover the significant differences in temperature distribution between different units.

[0034] formula: , .in, is the temperature data matrix, is a matrix consisting of the first few principal components, is the data matrix after dimensionality reduction, is a custom difference metric function, and are the median temperatures of the two units, and are the temperature standard deviations of the two units, and The two units are respectively The projection coefficients on the principal components are is the number of principal components. The usefulness of this formula is to reduce the dimension and highlight the main variation direction of the data through PCA, and at the same time, combine the customized difference measurement function to more comprehensively discover the significant differences in temperature distribution between different units.

[0035] The algorithm combines a variety of advanced analysis methods, taking into account the local and global characteristics of temperature distribution, accurately calculating the median and performing comparative analysis. At the same time, through complex prediction methods and differential analysis, it provides a deeper understanding of the analysis of temperature distribution in the stator core of hydro-generators and more effective fault prediction and maintenance planning.

[0036] In the preferred embodiment, the positions of the temperature sensors are reasonably arranged according to the structural characteristics and temperature distribution law of the stator core.

[0037] Example 2 Further illustrate with reference to Example 1, Figure 1 The structure shown in the figure shows the maximum over-limit analysis of the stator core temperature of all units. Using the industrial Internet platform, the maximum stator core temperature of all units in all power stations this month is calculated, and the bearing components that exceed the stator core temperature threshold are listed. As the equipment with abnormal events, manual analysis and other special analysis can be carried out later.

[0038] Analysis of the median temperature of the stator core of all units. Using the Industrial Internet platform, calculate the median temperature of the stator core of all units in all power plants this month, compare and analyze the median temperature of the same stator core of multiple units in a power plant, and add month-on-month and year-on-year analysis, such as Figure 1 As shown, Figure 1In the table, 1F~32F represents the unit number, and the X coordinate represents the temperature value. This allows us to intuitively analyze the difference between the overall temperature of the stator core and other units of the same model or different models in the power station, and also intuitively analyze the difference between the overall temperature of the stator core and adjacent or non-adjacent units. When the difference is large, such as the temperature difference exceeds 5°C, manual analysis and other special analysis are performed for further in-depth analysis.

[0039] The above embodiments are only preferred technical solutions of the present invention and should not be regarded as limiting the present invention. The protection scope of the present invention shall be the technical solutions recorded in the claims, including equivalent replacement solutions of the technical features in the technical solutions recorded in the claims. That is, equivalent replacement improvements within this scope are also within the protection scope of the present invention.

Claims

1. A method for analyzing temperature distribution of a stator core of a hydro-generator, characterized in that: The following steps are involved: S1. Arrange multiple temperature sensors on the stator core; S2, collecting temperature data of the stator core in real time through a data acquisition device; S3, processing and storing the collected temperature data; S4. Through the industrial Internet platform, the core temperature time series data is aggregated into the database, and then the unit's operating parameters and operation status parameters are pre-processed regularly; S5, analyzing the processed temperature data through a data analysis algorithm; S6. Generate a stator core temperature distribution diagram and a temperature change trend diagram according to the analysis results; S7. Diagnose and issue warning events and abnormal events based on the core temperature distribution and change trend; S8. Draw conclusions and treatment suggestions based on manual determination of early warning events and special analysis of abnormal events. Finally, based on the previous statistics and analysis, make equipment operation and maintenance suggestions, including equipment operation suggestions and equipment maintenance suggestions.

2. A method for analyzing temperature distribution of a stator core of a hydro-generator according to claim 1, characterized in that: The temperature sensor is connected to the data acquisition device via wired or wireless means.

3. The method for analyzing temperature distribution of a stator core of a hydro-generator according to claim 1, characterized in that: The data processing steps include data filtering, denoising, and calibration processing steps.

4. The method for analyzing temperature distribution of a stator core of a hydro-generator according to claim 1, characterized in that: The data analysis algorithm includes the steps of temperature distribution feature extraction and temperature change trend prediction and analysis; The core temperature of all units is analyzed by median; Using the Industrial Internet platform, calculate the median temperature of the cores of all units in the current analysis period, compare and analyze the median temperatures of the cores of multiple units, and add month-on-month and year-on-year analysis, so as to analyze the difference between the overall temperature of the core and other units of the same model / different models / adjacent / non-adjacent units in the power station; This analysis is not only helpful in discovering the overall temperature changes of components, but also in discovering the overall temperature differences between components of different units. For components with large differences, subsequent special analyses such as manual analysis can be carried out.

5. A method for analyzing temperature distribution of a stator core of a hydro-generator according to claim 4, characterized in that: The analysis algorithm for extracting temperature distribution features is a method combining singular value decomposition and wavelet transform. The steps include: A1. First, collect and organize the core temperature data of all units to form a temperature data set. ,in Indicates The temperature value set of each unit at different time points. is the total number of units, is the time point; For the temperature dataset The matrix formed Perform singular value decomposition, ,in and is an orthogonal matrix, It is a diagonal matrix, and the elements on its diagonal are singular values. The left singular vectors and right singular vectors corresponding to the larger singular values ​​form a new matrix and , then the temperature distribution characteristic matrix ; A2. Then, for the temperature time series collected by each sensor, wavelet transform is used to further extract local features; let the temperature change over time be , the wavelet coefficient calculation formula is: ； in is the average temperature, is the wavelet basis function, are the wavelet coefficients, and They are scale and translation parameters respectively; A3. Combine the wavelet coefficients with the characteristic matrix obtained by singular value decomposition to obtain the final temperature distribution characteristic matrix ,in It is a weight function used to adjust the contribution of wavelet coefficients in the feature matrix.

6. A method for analyzing temperature distribution of a stator core of a hydro-generator according to claim 4, characterized in that: The core temperature adopts median analysis, and the temperature median calculation method is: B1. Calculate the median temperature of all units in the current analysis period , first merge the temperature data of all units into one set ,in is the total number of all temperature data, Then To sort, if is an odd number, then the median ;like is an even number, then the median ; B2. For each unit, calculate its median temperature during the current analysis period. , Similarly, the temperature data of each unit is sorted, and the median is calculated according to the odd and even cases respectively; formula: , ;in, is the median temperature of all units, It is The median temperature of the units, Indicates that the The temperature value of the location, is the total number of all temperature data, It is Total number of temperature data of each unit; The usefulness of the formula is that by calculating the median, it can more accurately reflect the central trend of the temperature data and resist the influence of outliers.

7. A method for analyzing temperature distribution of a stator core of a hydro-generator according to claim 5, characterized in that: The core temperature adopts median analysis, and the medians of the core temperatures of multiple units are compared and analyzed. The median comparison analysis method is: C1. Perform comparative analysis and calculate the difference value , and conduct both month-on-month and year-on-year analysis. The month-on-month analysis is to compare with the median temperature of the previous time period and calculate the month-on-month growth rate. ; Year-on-year analysis is to compare the median temperature with the same period last year and calculate the year-on-year growth rate ; C2. Use standardized difference to measure the difference between two sets of data. The calculation formula is: ； in It is The median of the group data, is the average of all medians, is the standard deviation; The usefulness of the above formula is to understand the degree of deviation of a unit from the average level and discover potential problem units through comparative analysis and standardized differences.

8. The method for analyzing temperature distribution of a stator core of a hydro-generator according to claim 4, characterized in that: The overall temperature trend between different unit components is predicted based on the overall temperature change of the components. The prediction methods include: D1. Combining kernel-based support vector regression and ARIMA model to predict temperature change trend, using Gaussian kernel function ,in is the kernel parameter, which maps the temperature time series data to a high-dimensional feature space and performs linear regression in the high-dimensional feature space. D2. Algorithm construction optimization: , subject to , , ,in and is the regression coefficient, and is the slack variable, is the penalty parameter, is an insensitive loss parameter, Is to input data Functions that map to high-dimensional feature spaces; D3. By solving the optimization problem in step D2, we get the regression function , used to preliminarily predict the temperature value at a future time point; D4. At the same time, the ARIMA model is used to further predict the trend of the temperature series. The temperature series is , the expression of ARIMA model is: ,in is the predicted temperature value, is the mean term, is the autoregressive part, is the difference order, is a white noise process; D5. Finally, the prediction results of the SVR and ARIMA models are weighted and fused to obtain the final temperature change trend prediction value.

9. A method for analyzing temperature distribution of a stator core of a hydro-generator according to claim 4, characterized in that: The overall temperature change is also helpful to find the overall temperature difference between different unit components. The overall temperature difference analysis method includes: E1. The difference analysis is performed by combining principal component analysis with a custom difference measure. First, principal component analysis is used to identify the temperature distribution differences between different units. The temperature data matrix is , the data matrix after dimensionality reduction is obtained through PCA ,in is a matrix consisting of the first few principal components; E2. Then, define a custom difference measurement function , which is used to measure the temperature distribution difference between the two units. This difference measurement function can combine multiple characteristics of temperature distribution, including mean, variance, median, and principal component information obtained by PCA, and can be defined as: ； in and are the median temperatures of the two units, and are the temperature standard deviations of the two units, and The two units are respectively The projection coefficients on the principal components are is the number of principal components; The usefulness of the formula is to reduce the dimension and highlight the main variation direction of the data through PCA, and at the same time combine it with a custom difference measurement function to more comprehensively discover the significant differences in temperature distribution between different units.

10. The method for analyzing temperature distribution of a stator core of a hydro-generator according to claim 1, characterized in that: The positions of the temperature sensors are reasonably arranged according to the structural characteristics of the stator core and the temperature distribution law.

Images