Method for predicting thermal life of main transformer based on multivariate load modeling

Abstract

The application relates to a main transformer thermal life prediction method based on multivariate load modeling, relates to the technical field of transformer thermal life prediction, and comprises the following steps: acquiring typical load data of a target main transformer; analyzing the typical load data based on multivariate load factors, classifying modeling to acquire a load mode set; traversing the load mode set, evaluating the physical property adaptation degrees of multiple load modes, and dividing the load mode set into a first-class load mode group and a second-class load mode group; respectively constructing multivariate thermal life consumption rate evaluation models based on the first-class load mode group and the second-class load mode group; acquiring life cycle load data of the target main transformer; and combining the two load mode groups and the multivariate thermal life consumption rate evaluation models to perform thermal life prediction on the target main transformer. The application solves the problems in traditional main transformer thermal life prediction, such as inaccurate prediction results, limited applicability and difficulty in fully meeting actual application requirements.

Description

Technical Field

[0001] This application relates to the field of transformer thermal life prediction, and in particular to a method for predicting the thermal life of main transformers based on multi-variable load modeling. Background Technology

[0002] With the continuous development of power systems, the prediction of the thermal life of main transformers is of crucial significance for ensuring the safe operation of equipment and optimizing life management. At present, traditional methods for predicting the thermal life of main transformers mostly adopt a single processing approach, which is difficult to adapt to complex and ever-changing operating conditions and is prone to deviations between the prediction results and actual needs.

[0003] Existing forecasting methods fail to fully incorporate various dynamic factors in actual operation, resulting in insufficient targeting and effectiveness of forecasts. This not only reduces the application value of thermal life forecasting but also increases the complexity and cost of equipment maintenance decisions. Summary of the Invention

[0004] This application provides a method for predicting the thermal life of main transformers based on multi-variable load modeling, which improves the poor prediction results and limited applicability of traditional predictions due to lack of adaptability and failure to fully correlate with actual operating conditions.

[0005] The embodiments of this application disclose the following technical solutions:

[0006] This application provides a method for predicting the thermal life of a main transformer based on multi-variable load modeling. The method includes:

[0007] Interact with the target area to obtain typical load data of the target main transformer;

[0008] Based on preset multi-factor load analysis, the typical load data is classified and modeled to obtain a load pattern set.

[0009] Traverse the load pattern set, evaluate the physical property fit of multiple load patterns, and divide the load pattern set into a first-class load pattern group and a second-class load pattern group according to the physical property fit.

[0010] Based on the first type of load mode group and the second type of load mode group respectively, a multivariate thermal lifetime consumption rate assessment model is constructed, wherein the multivariate thermal lifetime consumption rate assessment model includes a prior data assessment model and a data fitting assessment model.

[0011] The life cycle load data of the target main transformer is obtained, and the thermal life of the target main transformer is predicted by combining the first type of load mode group, the second type of load mode group and the multi-element thermal life consumption rate assessment model.

[0012] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0013] This application proposes a method for predicting the thermal life of main transformers based on multi-variable load modeling. By dividing the load into Class I and Class II load pattern groups, constructing a multi-variable thermal life consumption rate assessment model, and combining lifecycle load data for thermal life prediction, accurate prediction of the main transformer's thermal life is achieved. First, based on multi-variable load factors, Class I and Class II load pattern groups are divided. For Class I load pattern groups, a local simulation model is established by combining the simulation domain and environmental boundary conditions, constructing a temperature-thermal stress binary life mapping model and integrating it into a priori data evaluation model. For Class II load pattern groups, a data-driven consumption rate prediction model is trained through cluster analysis, forming a data fitting evaluation model. Then, lifecycle load data is acquired, and corresponding discrete load data is extracted and input into the multi-variable thermal life consumption rate assessment model. Through temperature sequence simulation, thermal life consumption rate mapping, sequence filtering, and integral calculation, the most unfavorable controlled location is determined, and finally, the remaining thermal life is calculated.

[0014] The technical solution of this application solves the problems of low prediction accuracy and poor reliability in traditional main transformer thermal life prediction caused by the complexity of load characteristics and the insufficient adaptability of a single model through multiple steps, such as integrating load mode classification, evaluation system combining prior and data-driven models, and accurate analysis of full life cycle data. It realizes targeted prediction of thermal life and provides technical support for ensuring the safe and stable operation of main transformers. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 A flowchart illustrating the main transformer thermal life prediction method based on multi-variable load modeling provided in this application embodiment;

[0017] Figure 2 This is a flowchart illustrating how a load pattern set is divided into a first-class load pattern group and a second-class load pattern group based on physical property adaptability, as provided in an embodiment of this application. Detailed Implementation

[0018] This application provides a method for predicting the thermal life of main transformers based on multi-variable load modeling, which is used to solve the technical problems of insufficient accuracy, limited applicability, and difficulty in meeting actual operating requirements in the existing technology for predicting the thermal life of main transformers.

[0019] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0020] In the description of this application, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the stated features. In the description of this application, "multiple" means two or more, unless otherwise explicitly specified.

[0021] In the description of this application, the term "for example" is used to mean "used as an example, illustration, or description." Any embodiment described as "for example" in this application is not necessarily to be construed as being more preferred or advantageous than other embodiments. The following description is provided to enable any person skilled in the art to make and use the invention. Details are set forth in the following description for purposes of explanation. It should be understood that those skilled in the art will recognize that the invention can be made without using these specific details. In other instances, well-known structures and processes will not be described in detail to avoid obscuring the description of the invention with unnecessary detail. Therefore, the invention is not intended to be limited to the embodiments shown, but is consistent with the broadest scope of the principles and features disclosed in this application.

[0022] Examples, as shown in the appendix Figure 1 As shown, this application provides a method for predicting the thermal life of a main transformer based on multi-variable load modeling. The method includes the following steps:

[0023] S110: Interact with the target area to obtain typical load data of the target main transformer;

[0024] In this embodiment of the application, in order to fully capture the load characteristics of the target main transformer, it is necessary to interactively determine the target area and collect typical load data to provide high-quality basic information for subsequent classification modeling, so as to adapt to the thermal life prediction requirements under different operating scenarios.

[0025] Specifically, the target area for interaction is first defined as the power system region associated with the operating status of the target main transformer. The delineation of this target area must take into account the power supply range of the target main transformer, the network topology, and the characteristics of the load distribution.

[0026] Meanwhile, typical load data obtained from interactive channels may include detailed operating parameters such as hourly load current data at different times, harmonic content ratio at different times, power factor change curves, and load fluctuation records, so as to comprehensively capture the operating characteristics of the target main transformer under different time dimensions and different load intensities.

[0027] For example, the hourly load current data of a certain 220kV main transformer during the morning peak hours of 7-9 am on weekdays are 1200A, 1450A, and 1380A, respectively; the proportion of the third harmonic content during the midday hours of 12-2 pm is 1.8%, 2.1%, and 1.9%, respectively; the power factor change curve shows that it remains in the range of 0.82-0.85 from 0-5 am, while it rises to 0.90-0.93 from 5-8 pm.

[0028] Furthermore, in the load fluctuation records, at 10:00 AM on a certain workday, due to the startup of a large motor, the load current surged from 1100A to 1500A within 10 minutes, a fluctuation of 36.4%, before stabilizing at around 1400A. These data present the load characteristics of the main transformer under various scenarios from different dimensions, providing specific references for subsequent analysis.

[0029] Furthermore, the initial typical load data is preprocessed. Specifically, load current data that deviates significantly from the normal range are removed after verification to confirm that the sensor is faulty, so as to avoid abnormal data interfering with the accuracy of subsequent load feature extraction and classification modeling.

[0030] In addition, for power factor data missing for a few hours, a linear interpolation method is used. Based on the effective power factor data before and after the missing period, the missing values ​​are filled by calculating the linear trend, thus ensuring the continuity of the power factor data sequence.

[0031] At the same time, the current data, which was originally in "kiloamperes", was uniformly converted to "amperes" to ensure that the data format was consistent.

[0032] For example, in the initial typical load data of a target main transformer, it was found that the load current suddenly increased to 2500A between 9-10 am on a certain workday, while the rated current of the main transformer is 1200A. Moreover, there were no corresponding abnormal fluctuations in the voltage and power data during the same period. After checking the substation operation and maintenance records, it was confirmed that the current sensor malfunctioned during that period. Therefore, the data set was removed to avoid affecting the accuracy of subsequent extraction of load current characteristics.

[0033] Meanwhile, if the power factor data of the main transformer is missing between 2:00 AM and 5:00 AM on Wednesday, the power factor at 1:00 AM is 0.86 and the power factor at 6:00 AM is 0.88. Using linear interpolation, the values ​​at 2:00 AM, 3:00 AM, 4:00 AM, and 5:00 AM can be calculated to be 0.864, 0.868, 0.872, and 0.876, respectively. This will fill in the missing data and ensure the completeness and continuity of the power factor sequence.

[0034] Ultimately, the preprocessed typical load data possesses temporal continuity, data integrity, and format uniformity, laying a reliable foundation for subsequent classification modeling based on multiple load factors.

[0035] S120: Based on preset multi-factor load analysis, analyze the typical load data for classification and modeling to obtain a load pattern set;

[0036] In this embodiment of the application, in the scenario of predicting the thermal life of the main transformer, in order to accurately capture the load characteristics under different operating conditions, it is necessary to classify and model typical load data through multiple load factors in order to construct a load pattern set that fits reality.

[0037] In the method provided in this application embodiment, the multi-element load factor includes at least load current, harmonic content, power factor, load timing fluctuation period, and load fluctuation entropy.

[0038] Furthermore, load features are extracted from typical load data based on these multi-factor load factors. This involves extracting key features related to each load factor from the typical load data, such as the peak load current at different times, the range of harmonic content, the fluctuation range of power factor, the duration of load time-series fluctuation cycle, and the calculation results of load fluctuation entropy.

[0039] Furthermore, a load feature space is established, and the extracted load features are mapped into this load feature space. Then, cluster analysis is used to group and classify these load feature data, aggregating feature data with similar load characteristics into one category.

[0040] Meanwhile, based on the results of cluster analysis, multiple load patterns are defined, each containing a typical feature set and a pattern boundary, thereby constructing a load pattern set.

[0041] This step, by extracting load characteristics from multiple dimensions and performing cluster analysis, can effectively distinguish different types of load states, laying the foundation for subsequent evaluation of the physical property adaptability of each load mode and the construction of corresponding thermal life consumption rate evaluation models, ensuring that the model can better adapt to the actual operating conditions of the main transformer.

[0042] Step S120 in the method provided in this application embodiment includes:

[0043] Load features are extracted from the typical load data based on the aforementioned multivariate load factors;

[0044] Establish a load feature space, and map the load feature extraction results to the load feature space for cluster analysis;

[0045] Multiple load patterns are defined based on the clustering analysis results, and a load pattern set is obtained. Each load pattern includes a typical feature set and a pattern boundary.

[0046] In this embodiment of the application, in order to achieve accurate classification of the load characteristics of the main transformer, it is necessary to extract key features from typical load data through multivariate load factors, construct a load pattern set after cluster analysis, and provide basic support for the subsequent thermal life prediction model adaptability assessment and construction.

[0047] Specifically, load characteristics are first extracted from typical load data based on multiple load factors. These multiple load factors include at least load current, harmonic content, power factor, load time-series fluctuation period, and load fluctuation entropy. The extraction process requires extracting representative characteristic parameters from typical load data for each load factor.

[0048] For example, for load current, features such as peak current, average current, and duration are extracted for different time periods; for harmonic content, features such as the content ratio and distortion rate of each harmonic are extracted; for power factor, features such as its fluctuation range, average current, and stable duration are extracted; for load time-series fluctuation cycle, features such as the fluctuation cycle duration during peak periods, the frequency of load fluctuations within the cycle, and the phase shift of fluctuations are extracted; and the load fluctuation entropy is obtained by calculating the sequence complexity to obtain the corresponding feature values, forming a comprehensive set of load features.

[0049] Furthermore, a load feature space is established, and the extracted load feature extraction results are mapped into this load feature space.

[0050] The load characteristic space is a multi-dimensional mathematical space, with each dimension corresponding to a load characteristic parameter. Through coordinate mapping, the abstract load characteristics can be transformed into specific points in the space, so that the differences between different load states can be intuitively reflected in the space.

[0051] For example, if a certain set of load characteristics includes a peak load current of 800A, a third harmonic content of 2.5%, a power factor of 0.9, a load timing fluctuation period of 5 minutes, and a load fluctuation entropy of 0.3, then it can correspond to a specific coordinate point (800, 2.5%, 0.9, 5, 0.3) in the five-dimensional load characteristic space.

[0052] Furthermore, cluster analysis is performed on the mapped feature points within the load feature space to aggregate feature points with similar load characteristics into the same category, providing a basis for subsequent definition of load patterns.

[0053] Specifically, the cluster analysis uses the K-means algorithm in existing technology, which measures similarity by calculating the Euclidean distance between feature points, and groups feature points that are close to each other into one class to form multiple clusters.

[0054] Meanwhile, during the clustering process, the optimal number of clusters needs to be determined using the silhouette coefficient method. For example, when the number of clusters is 5, the silhouette coefficient reaches its maximum value of 0.85, indicating that the clustering results at this time can best distinguish different load states.

[0055] Furthermore, multiple load patterns are defined based on the clustering analysis results, and a load pattern set is obtained. Each load pattern includes a set of typical features and a pattern boundary.

[0056] Specifically, the typical feature set of a load mode is the representative characteristic value of each load factor under this load mode. For example, the typical feature set of a nonlinear load mode may be a load current average of 1200A, harmonic content ≤3%, power factor ≥0.88, load time-series fluctuation period of 5-6 minutes, and load fluctuation entropy ≤0.2. The mode boundary is the load characteristic threshold range that distinguishes different load modes. For example, when the load current is >1000A, the load time-series fluctuation period is ≤7 minutes and the fluctuation entropy is <0.3, it is determined to be the "metro heavy load mode".

[0057] Ultimately, these load patterns together constitute a load pattern set. The typical characteristics and mode boundaries of each load pattern are clearly distinguishable, which can accurately reflect the load characteristics of the main transformer under different scenarios. This provides a precise classification basis for subsequent division of mode groups based on physical property adaptability and construction of targeted thermal life consumption rate assessment models.

[0058] S130: Traverse the load pattern set, evaluate the physical property fit of multiple load patterns, and divide the load pattern set into a first-class load pattern group and a second-class load pattern group according to the physical property fit.

[0059] In this embodiment of the application, in the scenario of predicting the thermal life of the main transformer, in order to ensure that different load modes can be matched with suitable prediction models, it is necessary to evaluate the physical property compatibility of each load mode to achieve a scientific division of the load mode set, so as to provide a basis for the subsequent construction of a targeted thermal life consumption rate assessment model.

[0060] Specifically, we first traverse each load pattern in the load pattern set, using a multivariate load factor as the independent variable and the heat lifetime consumption rate as the dependent variable to perform a lightweight fitting model. Through simple functional relationships, we initially explore the correlation between load characteristics and heat lifetime consumption, laying the foundation for subsequent property fit assessment.

[0061] Furthermore, a fitting evaluation index is calculated based on the lightweight fitting modeling results. This fitting evaluation index includes at least one of the following: coefficient of determination, root mean square error, and mean absolute error.

[0062] Specifically, the coefficient of determination measures the extent to which the model interprets the data; the closer its value is to 1, the better the fit. The root mean square error and mean absolute error reflect the degree of deviation between the predicted and actual values; the smaller the values, the higher the fitting accuracy.

[0063] Furthermore, the fitting evaluation index is comprehensively evaluated in conjunction with the preset fusion rules to obtain the physical property fit of each load mode.

[0064] Among them, the fusion rule can assign different weights to each indicator according to their importance, and the comprehensive score, namely the property suitability, is obtained through weighted calculation. The higher the value, the more suitable the load pattern is for prediction using the prior data model.

[0065] Finally, based on the comparison between the physical property fit degree and the preset fit degree threshold, the load pattern set is divided into two categories: load patterns whose physical property fit degree meets the preset fit degree threshold are classified into the first category of load pattern group, while those that do not meet the preset fit degree threshold are added to the second category of load pattern group, so as to achieve accurate differentiation of load patterns with different fit characteristics.

[0066] This step allows for precise differentiation of the physical property adaptation characteristics of different load modes, ensuring that the subsequently constructed multivariate thermal life consumption rate assessment model can match the characteristics of each load mode, thereby improving the accuracy and efficiency of thermal life prediction.

[0067] Step S130 in the method provided in this application embodiment includes:

[0068] Traverse each load pattern in the load pattern set, and perform lightweight fitting modeling with the multivariate load factor as the independent variable and the heat lifetime consumption rate as the dependent variable.

[0069] The fitting evaluation index is calculated by iterating through the lightweight fitting modeling results. The fitting evaluation index includes at least one of the following: coefficient of determination, root mean square error, and mean absolute error.

[0070] The fitting evaluation index is comprehensively evaluated by combining the preset fusion rules to obtain the property fit degree of each load mode.

[0071] The load patterns whose physical property fitness meets the fitness threshold are added to the first type of load pattern group;

[0072] Load patterns whose physical property fit does not meet the fit threshold are added to the second type of load pattern group.

[0073] In this embodiment of the application, in order to ensure that different load modes can be matched with suitable thermal life prediction models, it is necessary to systematically evaluate the physical property adaptability of each load mode to achieve a scientific division of the load mode set, and provide a basis for the subsequent construction of targeted evaluation models.

[0074] First, we iterate through each load pattern in the load pattern set and perform lightweight fitting modeling with multivariate load factors (load current, harmonic content, power factor, load time-series fluctuation period, and load fluctuation entropy) as independent variables and heat lifetime consumption rate as dependent variable.

[0075] For example, for linear load patterns, such as "residential daily load patterns", a multiple linear regression model can be used to construct a functional relationship such as "heat life consumption rate = a × load current + b × harmonic content + c × power factor + d × load time-series fluctuation period + e × load fluctuation entropy + f", where a, b, c, d, and e are regression coefficients, and f is a constant term. The coefficients are solved by the least squares method to complete the model construction.

[0076] Furthermore, a fitting evaluation index is calculated based on the lightweight fitting modeling results. This fitting evaluation index includes at least one of the following: coefficient of determination, root mean square error, and mean absolute error.

[0077] Specifically, the coefficient of determination (R²) 2 The metric is used to measure the model's ability to interpret data; a value closer to 1 indicates a better fit. The root mean square error (RMSE) reflects the square root of the mean of the sum of squares of the deviations between predicted and actual values; a smaller value indicates higher fitting accuracy. The mean absolute error (MAE) is the average of the absolute deviations between predicted and actual values; similarly, a smaller value is preferred. By calculating these multi-dimensional indicators, the adaptability of the lightweight model to various load patterns can be comprehensively evaluated.

[0078] For example, for a certain nonlinear load pattern, such as the "metro heavy load pattern", the coefficient of determination of its lightweight fitting model is 0.92, indicating that the model can explain 92% of the heat life consumption rate variation; the root mean square error is 0.03 (unit: % / h) and the mean absolute error is 0.02 (unit: % / h), indicating that the deviation between the model's predicted value and the actual value is small and the fitting effect is ideal.

[0079] Furthermore, the fitting evaluation index is comprehensively evaluated by combining the preset fusion rules to obtain the physical property fit of each load mode.

[0080] The fusion rule can assign different weights to each indicator based on their importance. For example, the coefficient of determination can be assigned a weight of 0.5, while the root mean square error and the mean absolute error can each be assigned a weight of 0.25. This balances the influence of different fitting evaluation indicators in the comprehensive evaluation, and the comprehensive value obtained through weighted calculation is used as the property fit.

[0081] Specifically, the formula for calculating the property fit can be expressed as "property fit = coefficient of determination × 0.5 + root mean square error conversion value × 0.25 + mean absolute error conversion value × 0.25".

[0082] The root mean square error transformation value and the mean absolute error transformation value are calculated by "1 - (error value / maximum error value)" to uniformly map error indicators of different magnitudes to the range of 0-1, ensuring that each indicator has the same magnitude of comparability in the comprehensive evaluation.

[0083] In addition, the maximum error refers to the largest value among the root mean square error or mean absolute error calculated from all sample data during the lightweight fitting modeling process corresponding to the current load mode.

[0084] For example, if the determination coefficient of the "residential daily load pattern (linear load pattern)" is 0.88, and the maximum root mean square error (RMSE) calculated for all samples during the lightweight fitting modeling process is 0.4 (unit: % / h), and the RMSE of a certain sample is 0.04, then its transformed RMSE value is 1 - (0.04 / 0.4) = 0.9; the maximum mean absolute error (MAE) is 0.3 (unit: % / h), and the MAE of this sample is 0.03, then its transformed MAE value is 1 - (0.03 / 0.3) = 0.9. Therefore, the property fit of this load pattern is 0.88 × 0.5 + 0.9 × 0.25 + 0.9 × 0.25 = 0.89.

[0085] Furthermore, based on the comparison results between the physical property fit degree and the preset fit degree threshold, the load pattern set is divided into two categories to achieve scientific classification of load patterns with different fit characteristics, providing a basis for subsequent selection of appropriate thermal life prediction models.

[0086] The adaptation threshold needs to be set based on the actual operating scenario of the main transformer, historical data performance, and prediction accuracy requirements, and determined through experience adjustments.

[0087] Specifically, load modes whose physical property fit meets the preset fit threshold are classified into Category I load mode group, while those that do not meet the preset fit threshold are added to Category II load mode group. For example, if the preset fit threshold is 0.8, the physical property fit of the "Metro Heavy Load Mode (Nonlinear Load Mode)" is 0.935, which meets the preset fit threshold requirement and is classified into Category I load mode group; the physical property fit of the "Fault Impact Mode" is 0.65, which does not meet the preset fit threshold, and is therefore added to Category II load mode group.

[0088] This step allows for precise differentiation of the adaptation characteristics of different load modes, ensuring that the subsequently constructed multivariate thermal lifetime consumption rate assessment model can match the characteristics of each load mode, thereby improving the accuracy and efficiency of thermal lifetime prediction.

[0089] S140: Construct a multivariate thermal lifetime consumption rate assessment model based on the first type of load mode group and the second type of load mode group respectively, wherein the multivariate thermal lifetime consumption rate assessment model includes a prior data assessment model and a data fitting assessment model.

[0090] In this embodiment of the application, in the scenario of predicting the thermal life of the main transformer, in order to provide an accurate assessment method for the thermal life consumption rate for load modes with different adaptation characteristics, it is necessary to construct corresponding assessment models for the first type of load mode group and the second type of load mode group, and construct a multivariate thermal life consumption rate assessment model.

[0091] For a group of load patterns, multiple load patterns are first identified as targets, and multiple simulation domains are defined in conjunction with multivariate load factors. Simultaneously, environmental boundary conditions are determined. These environmental boundary conditions include at least ambient temperature, wind speed, and solar radiation, providing comprehensive scene parameters for simulation analysis.

[0092] Furthermore, by combining the structural information of multiple simulation domains and the target main transformer, a local simulation model and simulation call relationship are established.

[0093] The local simulation model uses multiple hot spots of the target main transformer as local simulation targets to evaluate the temperature sequence at these controlled locations. The simulation call relationship is matched with the model call based on the typical feature set of each type of load mode to ensure that the corresponding simulation function can be called under different modes.

[0094] Furthermore, a temperature-thermal stress binary lifetime mapping model is constructed. This model takes temperature and material properties as inputs and outputs the thermal lifetime consumption rate, realizing the quantitative conversion from temperature to thermal lifetime consumption.

[0095] Finally, the local simulation model, simulation call relationships, and temperature-thermal stress binary life mapping model are integrated to obtain prior data for model evaluation. The output of the local simulation model is connected to the input of the temperature-thermal stress binary life mapping model, forming a complete evaluation chain.

[0096] In addition, for the Class II load mode group, the typical load data is first extracted by traversing the Class II load mode group as an index to obtain the total training dataset, which covers the multivariate load factors and actual thermal lifetime consumption rate under the corresponding mode.

[0097] Furthermore, the elbow method is used to perform cluster analysis on the total training dataset, dividing it into multiple binary training data subsets, so that the data in each subset have higher similarity and improve the model training effect.

[0098] Meanwhile, based on multiple subsets of binary training data, multiple data-driven consumption rate prediction models are constructed and trained, and a mapping relationship between the binary load pattern groups and these prediction models is established to obtain a data fitting evaluation model.

[0099] This step involves constructing a priori data evaluation model and a data fitting evaluation model, respectively, enabling the multivariate thermal lifetime consumption rate evaluation model to adapt to load patterns with different characteristics, thus providing an accurate rate evaluation basis for subsequent thermal lifetime prediction based on life cycle load data.

[0100] Step S140 in the method provided in this application embodiment includes:

[0101] With multiple Class I load patterns as targets, multiple simulation domains and environmental boundary conditions are defined in combination with the multivariate load factors, wherein the environmental boundary conditions include at least ambient temperature, wind speed, and solar radiation.

[0102] By combining the structural information of multiple simulation domains and the target main transformer, a local simulation model and simulation call relationship are established.

[0103] A temperature-thermal stress binary lifetime mapping model is constructed, wherein the temperature-thermal stress binary lifetime mapping model takes temperature and material properties as inputs and outputs the thermal lifetime consumption rate.

[0104] The local simulation model, the simulation call relationship, and the temperature-thermal stress binary life mapping model are integrated to obtain prior data to evaluate the model. The output of the local simulation model is connected to the input of the temperature-thermal stress binary life mapping model.

[0105] In this embodiment of the application, in order to construct an accurate thermal lifetime consumption rate assessment model for a certain type of load mode group, it is necessary to form an assessment system based on a priori model by defining a simulation domain, establishing a local simulation model, constructing a lifetime mapping model and integrating them, so as to achieve the scientific quantification of thermal lifetime consumption rate.

[0106] Specifically, multiple load patterns of type 1 are first targeted, and multiple simulation domains are defined in combination with multivariate load factors to accurately simulate the heat generation and dissipation processes of the main transformer under different load characteristics. At the same time, environmental boundary conditions, including ambient temperature, wind speed, and solar radiation, are determined to provide comprehensive scene parameter support for subsequent simulations.

[0107] In the method provided in this application embodiment, the simulation domain includes at least a current-generated heat domain, a harmonic-generated heat domain, and a heat dissipation domain, wherein the heat dissipation domain includes a heat conduction path, a heat convection path, and a heat radiation path.

[0108] Specifically, the current-generated heat domain is mainly related to the load current magnitude and power factor, and is used to simulate the process of heat generation in the main transformer due to resistive losses when current passes through the windings; the harmonic-generated heat domain is related to the harmonic content, and is used to simulate the heat generated by eddy current losses and stray losses caused by harmonic currents.

[0109] Meanwhile, the heat dissipation domain is used to simulate the heat transfer and dissipation process inside the main transformer. The heat conduction path reflects the heat transfer between solid components such as the core and windings, the heat convection path reflects the heat exchange brought about by fluid flow such as hot oil circulation, and the heat radiation path characterizes the heat dissipation radiated from the transformer surface to the surrounding environment. The three together determine the temperature distribution of the main transformer.

[0110] Furthermore, by combining the obtained structural information of multiple simulation domains and the target main transformer, a local simulation model and simulation call relationship are established to accurately simulate the temperature change process of key parts of the main transformer under different load modes.

[0111] In the method provided in this application embodiment, "establishing a local simulation model and simulation call relationship by combining multiple simulation domains and the structural information of the target main transformer" includes:

[0112] The prior model is invoked by traversing multiple simulation domains to generate prior model invocation results, which include current heat generation model group, harmonic heat generation model group and heat dissipation model group.

[0113] Multiple typical controlled locations of the target main transformer are obtained as local simulation targets, and the local simulation model is constructed by combining the results of the prior model call.

[0114] Model call matching is performed based on the typical feature set of each type of load mode, and the simulation call relationship is established accordingly.

[0115] In this embodiment of the application, in order to accurately simulate the temperature change process of the main transformer under a certain load mode, it is necessary to construct a local simulation model by calling a prior model and focusing on key locations, and to establish a simulation calling relationship adapted to the load mode, so as to improve the pertinence and efficiency of the simulation.

[0116] Specifically, the process begins by traversing multiple simulation domains to invoke prior models. By invoking these prior models, corresponding prior model invocation results are generated, providing the basic computational tools for subsequent simulations.

[0117] Among them, the a priori models are formulas or functions based on physical properties. For example, the current-generated heat model group includes heat generation calculation formulas related to load current and power factor, the harmonic-generated heat model group covers functions that reflect the relationship between harmonic content and eddy current loss and stray loss, and the heat dissipation model group includes physical formulas for heat transfer such as heat conduction, heat convection, and heat radiation.

[0118] Furthermore, several typical controlled locations of the target main transformer are obtained as local simulation targets. These typical controlled locations are high-heat locations determined based on experimental or actual operating experience, such as winding hot spots and core clamps, which are prone to having their lifespan affected by excessive temperature.

[0119] Meanwhile, by combining the results of prior model calls, the heat generation and heat dissipation models are applied to these typical controlled locations to construct local simulation models, so that the simulation focuses on key areas and reduces unnecessary computation.

[0120] Furthermore, model call matching is performed based on the typical feature set of each load mode to establish simulation call relationships.

[0121] Specifically, if the typical feature set of a certain load mode shows that its harmonic content is below the significance threshold, indicating that harmonic heat generation has a very small impact on the overall temperature, then the harmonic heat generation model group can be ignored under this mode, and only the current heat generation model group and the heat dissipation model group can be called.

[0122] The significance threshold is a critical value set based on the main transformer design parameters, operating experience, and the accuracy requirements of thermal life prediction. For example, it can be set to 1% of the rated current. When the proportion of harmonic content is lower than this value, the additional losses generated by it have a negligible impact on the temperature field distribution of the main transformer.

[0123] Conversely, for load modes with high harmonic content, it is necessary to activate the current-generated heat, harmonic-generated heat, and heat dissipation model groups simultaneously. This targeted model matching ensures both simulation accuracy and effectively improves computational efficiency.

[0124] For example, the typical feature set of linear load patterns such as "residential daily load pattern" shows that its harmonic content is consistently below 0.5%, far below the significance threshold of 1%. Therefore, when simulating this load pattern, the local simulation model only calls the current heat generation model and the heat dissipation model, without needing to enable the harmonic heat generation model.

[0125] However, for nonlinear load modes such as "metro heavy load mode", which contain a large number of nonlinear loads, the harmonic content often reaches more than 3%. In this case, the simulation call relationship will simultaneously activate the current heat generation model, the heat dissipation model and the harmonic heat generation model to comprehensively calculate its heat generation and heat dissipation process.

[0126] The method provided in this application, which "constructs a multivariate thermal lifetime consumption rate assessment model based on a type I load mode group and the type II load mode group respectively", further includes:

[0127] Using the two types of load pattern groups as indexes, the typical load data are traversed to extract data and obtain the total training dataset.

[0128] Cluster analysis of the total training dataset is performed using the elbow method to obtain multiple binary training data subsets.

[0129] Based on multiple subsets of the two types of training data, multiple data-driven consumption rate prediction models are constructed and trained respectively, and a mapping relationship between the two types of load pattern groups and the multiple consumption rate prediction models is established to obtain the data fitting evaluation model.

[0130] In this embodiment of the application, for the second type of load mode group whose physical property compatibility does not meet the preset compatibility threshold, it is necessary to construct a consumption rate prediction model through a data-driven approach in order to accurately capture the relationship between its complex load characteristics and thermal lifetime consumption rate, and make up for the limitations of the prior data model in the second type of load mode.

[0131] Specifically, the two types of load pattern groups are first used as indexes to traverse typical load data for targeted data extraction. This extraction includes multivariate load factors and corresponding heat lifetime consumption rates corresponding to the two types of load pattern groups, which are then integrated to form a total training dataset, providing sufficient sample support for model training.

[0132] Furthermore, the elbow method is used to perform cluster analysis on the resulting total training dataset, which is then divided into multiple binary training data subsets.

[0133] Among them, the elbow method calculates the sum of squared errors under different numbers of clusters. When the rate of decrease of the sum of squared errors slows down significantly (i.e., the "elbow" inflection point is formed), the corresponding number of clusters is the optimal partitioning scheme.

[0134] For example, if the total training dataset contains samples with different load characteristics such as "short-term impact load" and "nonlinear fluctuation load", and the optimal number of clusters is determined to be 3 by elbow method analysis, the dataset can be divided into 3 binary training data subsets, so that the samples in each subset have more similar load characteristics and consumption patterns, thereby improving the training effect of the model.

[0135] Furthermore, based on the divided subsets of training data of different types, multiple data-driven consumption rate prediction models are constructed and trained to accurately capture the complex nonlinear relationship between multivariate load factors and thermal lifetime consumption rate under different load modes.

[0136] Specifically, algorithms such as random forest, gradient boosting regression, and neural networks can be used to train the model with the multivariate load factors of each subset as input features and the thermal lifetime consumption rate as the output label.

[0137] During training, the 5-fold cross-validation method is used to evaluate the model performance. The model accuracy is optimized by adjusting hyperparameters such as decision tree depth, learning rate, and number of neurons until the prediction error on the validation set converges to a preset range, thus obtaining a data-driven consumption rate prediction model adapted to the corresponding subset of training data.

[0138] Simultaneously, a mapping relationship is established between the two types of load modes and multiple consumption rate prediction models. That is, each type of load mode corresponds to an optimal prediction model, thereby obtaining data fitting and evaluation models. For example, the "electric arc furnace impact load mode" corresponds to a gradient boosting regression model trained on its dedicated training subset, while the "frequency converter nonlinear load mode" corresponds to a neural network model, ensuring that different load modes can call the most suitable model for prediction.

[0139] This step, through the construction of a data fitting evaluation model, can specifically adapt to the complex characteristics of the two types of load modes, complementing the prior data evaluation model and together forming a multivariate thermal lifetime consumption rate evaluation model. This provides a comprehensive and accurate rate evaluation tool for subsequent thermal lifetime prediction based on full life cycle load data.

[0140] S150: Obtain the life cycle load data of the target main transformer, and combine the first type of load mode group, the second type of load mode group and the multi-element thermal life consumption rate assessment model to predict the thermal life of the target main transformer.

[0141] In this embodiment of the application, in the scenario of predicting the thermal life of the main transformer, in order to comprehensively consider the impact of different load modes on the thermal life of the main transformer, it is necessary to complete the overall prediction of the thermal life of the main transformer based on the life cycle load data, combined with the corresponding load mode group and the multi-element thermal life consumption rate evaluation model.

[0142] Specifically, for the lifecycle load data corresponding to a load pattern group, the first discrete load data is extracted. This first discrete load data is then input into the local simulation model of the multivariate thermal lifecycle consumption rate assessment model, and simulation analysis is performed in conjunction with the established simulation call relationship to obtain the temperature sequences of multiple typical controlled locations.

[0143] Furthermore, the temperature-thermal stress binary lifetime mapping model is activated to map the temperature sequence into a thermal lifetime consumption rate sequence. Based on these thermal lifetime consumption rate sequences, integral calculations are performed to obtain the thermal lifetime consumption at several typical controlled locations, and the most unfavorable controlled location is determined.

[0144] Meanwhile, for the lifecycle load data corresponding to the second type of load mode group, the corresponding discrete load data is also extracted, and the input data is used to fit the corresponding consumption rate prediction model in the evaluation model to obtain the thermal life consumption rate sequence.

[0145] Furthermore, the thermal lifetime consumption of each typical controlled location is obtained through integral calculation, thereby determining the most unfavorable controlled location under the two types of load modes.

[0146] Finally, by combining the thermal lifetime consumption of the most unfavorable controlled location obtained under the two types of load modes, the thermal lifetime of the target main transformer is calculated, and the final thermal lifetime prediction result is output.

[0147] This step, through load-sharing mode assessment and integration of key location data, ensures that thermal life prediction can comprehensively reflect the actual loss of the main transformer throughout its entire life cycle, providing a reliable basis for equipment operation, maintenance, and replacement.

[0148] Step S150 in the method provided in this application embodiment includes:

[0149] Based on the lifecycle load data, extract the first discrete load data corresponding to the first type of load pattern group;

[0150] The first discrete load data is input into the local simulation model of the multivariate thermal lifetime consumption rate evaluation model, and simulation analysis is performed in combination with the simulation call relationship to obtain the temperature sequence of multiple typical controlled locations.

[0151] Activate the temperature-thermal stress binary lifetime mapping model to map multiple temperature sequences into thermal lifetime consumption rate sequences;

[0152] Based on the integration operation of multiple thermal lifetime consumption rate sequences, the thermal lifetime consumption of multiple typical controlled locations is obtained, and the most unfavorable controlled location is defined accordingly.

[0153] The thermal life of the target main transformer is calculated based on the most unfavorable controlled location, and the output is the thermal life prediction result.

[0154] In this embodiment of the application, in order to achieve accurate prediction of the thermal life of the target main transformer throughout its entire life cycle, it is necessary to combine life cycle load data with a multivariate thermal life consumption rate assessment model. This is achieved by analyzing discrete load data step by step, simulating temperature changes, mapping thermal life consumption rate, quantifying life loss, and focusing on key locations to complete a comprehensive assessment of the thermal life.

[0155] Specifically, firstly, based on the life cycle load data of the target main transformer, the first discrete load data corresponding to a type of load mode group is extracted.

[0156] Among them, these first discrete load data cover key parameters such as load current, harmonic content, power factor and load fluctuation entropy at different time periods under this type of load mode group, providing basic input for simulation analysis.

[0157] Furthermore, the first discrete load data is input into the local simulation model of the multivariate thermal life consumption rate assessment model, and simulation analysis is performed in conjunction with the established simulation call relationship.

[0158] During the simulation, the corresponding heat generation and heat dissipation models are matched according to the typical feature set of a load mode, and the temperature changes at multiple typical controlled locations are accurately calculated, thereby outputting a temperature sequence.

[0159] For example, for the first discrete load data of a linear load pattern such as "residential daily load pattern", the simulation model only calls the current heat generation model and the heat dissipation model to obtain the temperature fluctuation sequence of the winding hot spot within 24 hours as 35℃-58℃-42℃.

[0160] Furthermore, the temperature-thermal stress binary lifetime mapping model is activated, and the thermal lifetime consumption rate sequence is obtained by mapping and transforming the temperature value at each moment in the temperature sequence one by one.

[0161] The temperature-thermal stress binary lifetime mapping model is based on material properties and calculates the lifetime consumption rate at different temperatures using a preset formula. For example, when the hot spot temperature of the winding is 58℃ at a certain moment in the temperature sequence, the corresponding thermal lifetime consumption rate obtained by mapping through the temperature-thermal stress binary lifetime mapping model is 0.002% / h.

[0162] Furthermore, the obtained thermal lifetime consumption rate sequence is integrated to obtain the thermal lifetime consumption corresponding to a typical controlled location.

[0163] In the method provided in this application embodiment, before "performing an integral operation based on multiple heat lifetime consumption rate sequences", it further includes:

[0164] Calculate the average slope of the multiple thermal lifetime consumption rate sequences;

[0165] Serialize multiple thermal lifetime consumption rate sequences based on the mean slope of the sequence, and retain the first preset number of thermal lifetime consumption rate sequences.

[0166] In this embodiment of the application, in order to improve the efficiency and specificity of thermal lifetime assessment and avoid performing indiscriminate integration on the thermal lifetime consumption rate sequence of all controlled locations, it is necessary to screen out the thermal lifetime consumption rate sequence that has a more significant impact on thermal lifetime through sequence slope analysis, and focus on the lifetime loss assessment of key locations.

[0167] Specifically, the mean slope of multiple thermal lifetime decay rate sequences is first calculated. The sequence slope reflects the trend of thermal lifetime decay rate over time; a larger slope indicates that the decay rate at that location increases faster and has a more significant impact on the overall lifespan of the main transformer.

[0168] Specifically, during the calculation, for each typical controlled location's thermal lifetime consumption rate sequence, a linear fitting method is used to determine its slope value, and then the average of all slope values ​​is taken as a reference benchmark for subsequent sequence selection.

[0169] For example, the slopes of the heat lifetime consumption rate sequence corresponding to five typical controlled locations of a main transformer are 0.001% / h. 2 0.003% / h 2 0.002% / h 2 0.005% / h 2 0.004% / h 2 Therefore, the mean slope of the sequence is (0.001+0.003+0.002+0.005+0.004) / 5=0.003% / h 2 .

[0170] Furthermore, the multiple thermal lifetime consumption rate sequences are serialized and sorted based on the mean slope of the sequences. That is, the slope value of each sequence is compared with the mean, and the sequences are sorted in descending order of slope, so that the sequences with larger slopes (i.e., the positions with faster lifetime consumption rates) are placed at the top.

[0171] For example, in the same example above, the five heat lifetime consumption rates are ordered from largest to smallest by slope as 0.005% / h. 2 0.004% / h 2 0.003% / h 2 0.002% / h 2 0.001% / h 2 .

[0172] Furthermore, a preset number of thermal life consumption rate sequences are retained. The preset number can be set according to the structural characteristics of the main transformer, operating requirements, and assessment accuracy requirements. For example, it can be set to 3, that is, retaining the first 3 sequences with the larger slope in the above sorting.

[0173] This screening process can eliminate locations with slow lifespan consumption and minimal impact on overall lifespan, thereby reducing the computational load of subsequent integration calculations. At the same time, it ensures that the evaluation results focus on key locations that play a decisive role in thermal lifespan, laying a reliable foundation for finally determining the most unfavorable controlled location and calculating the thermal lifespan of the main transformer.

[0174] Furthermore, integral calculations are performed based on the multiple thermal lifetime consumption rate sequences retained after screening.

[0175] Specifically, the integration operation uses time as the integration variable to accumulate the thermal lifetime consumption rate of each thermal lifetime consumption rate sequence over the entire life cycle, thereby obtaining the total thermal lifetime consumption corresponding to each typical controlled location.

[0176] For example, the three thermal lifetime consumption rate sequences with relatively large slopes (slopes of 0.005% / h) were retained. 2 0.004% / h 2 0.003% / h 2 If the cumulative consumption rates over a 10-year lifespan are 85%, 72%, and 60%, respectively, then the corresponding thermal lifetime consumption amounts are 85%, 72%, and 60%.

[0177] Furthermore, the most unfavorable controlled location is defined. This involves comparing the thermal lifetime consumption at multiple typical controlled locations and identifying the location with the highest thermal lifetime consumption as the most unfavorable controlled location. This location represents the "weak link" affecting the overall thermal lifetime of the main transformer. For example, in the above example, the location with 85% thermal lifetime consumption is the most unfavorable controlled location.

[0178] Finally, the thermal life of the target main transformer is calculated based on the most unfavorable controlled location, and the predicted thermal life is then output.

[0179] In specific calculations, the design service life of the main transformer is used as the benchmark, and the thermal life consumption of the most unfavorable controlled position is combined to determine the remaining thermal life. The calculation formula can be expressed as "remaining thermal life = design service life × (1 - percentage of thermal life consumption of the most unfavorable controlled position)".

[0180] For example, if the main transformer is designed to have a service life of 30 years and the thermal life consumption at the most unfavorable controlled location is 85%, then the remaining thermal life is 30×(1-85%)=4.5 years, which is used as the final thermal life prediction result.

[0181] This step extracts key information from the life cycle load data, combines a local simulation model with a temperature-thermal stress binary life cycle mapping model to quantify thermal life consumption, then filters the thermal life consumption rate sequence and performs integral calculations to focus on the most unfavorable controlled location, and finally calculates the remaining thermal life based on the design life and the consumption at the short board location, thus achieving accurate and efficient prediction of the thermal life of the main transformer throughout its entire life cycle.

[0182] The embodiments of this application, through the specific implementation methods described above, achieve the following technical effects:

[0183] This application proposes a method for predicting the thermal life of a main transformer based on multivariate load modeling. First, typical load data of the target main transformer is obtained based on the interactive target region. After preprocessing, classification modeling is performed based on preset multivariate load factors. Load features are extracted, and a load pattern set containing multiple load patterns is obtained through cluster analysis. Next, the load pattern set is traversed, and the property fit of each load pattern is evaluated, dividing the load pattern set into a first-class load pattern group and a second-class load pattern group. Subsequently, multivariate thermal life consumption rate assessment models are constructed for the two types of pattern groups. For the first-class load pattern group, a priori data assessment model is constructed, including a local simulation model, simulation call relationships, and a temperature-thermal stress binary life mapping model. For the second-class load pattern group, a data fitting assessment model based on cluster analysis and a data-driven model is constructed. Finally, the life cycle load data of the target main transformer is obtained. Combining the two types of load pattern groups and the multivariate thermal life consumption rate assessment model, the thermal life of the target main transformer is predicted through steps such as extracting discrete load data, simulating temperature sequences, mapping thermal life consumption rate sequences, and determining the most unfavorable controlled location through integral calculations.

[0184] The method provided in this application, through the technical solution of "load data acquisition and preprocessing - load mode classification and division - differential evaluation model construction - full life cycle thermal life prediction", solves the problems of low prediction accuracy and poor reliability caused by the complexity of load characteristics and insufficient adaptability of a single model in the traditional main transformer thermal life prediction. It achieves accurate adaptation to different load modes and efficient evaluation of thermal life, and provides reliable technical support for ensuring the safe and stable operation of the main transformer.

[0185] It should be noted that the order of the embodiments described above is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, the above description focuses on specific embodiments of this specification. Additionally, the processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some implementations, multitasking and parallel processing are possible or may be advantageous.

[0186] The above description is only a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

[0187] This specification and accompanying drawings are merely illustrative examples of this application and are intended to cover any and all modifications, variations, combinations, or equivalents within the scope of this application. Clearly, those skilled in the art can make various alterations and modifications to this application without departing from its scope. Therefore, if such modifications and variations fall within the scope of this application and its equivalents, this application intends to include such modifications and variations.

Claims

1. A method for predicting the thermal life of a main transformer based on multi-variable load modeling, characterized in that, include: Interact with the target area to obtain typical load data of the target main transformer; Based on preset multi-factor load analysis, the typical load data is classified and modeled to obtain a load pattern set. Traverse the load pattern set, evaluate the physical property fit of multiple load patterns, and divide the load pattern set into a first-class load pattern group and a second-class load pattern group according to the physical property fit. Based on the first type of load mode group and the second type of load mode group respectively, a multivariate thermal lifetime consumption rate assessment model is constructed, wherein the multivariate thermal lifetime consumption rate assessment model includes a prior data assessment model and a data fitting assessment model. Acquire the life cycle load data of the target main transformer, and combine the first type of load mode group, the second type of load mode group and the multi-element thermal life consumption rate assessment model to predict the thermal life of the target main transformer. Specifically, the load pattern set is traversed, the physical property fit of multiple load patterns is evaluated, and the load pattern set is divided into a first-class load pattern group and a second-class load pattern group based on the physical property fit, including: Traverse each load pattern in the load pattern set, and perform lightweight fitting modeling with the multivariate load factor as the independent variable and the heat lifetime consumption rate as the dependent variable. The fitting evaluation index is calculated by iterating through the lightweight fitting modeling results. The fitting evaluation index includes at least one of the following: coefficient of determination, root mean square error, and mean absolute error. The fitting evaluation index is comprehensively evaluated by combining the preset fusion rules to obtain the property fit degree of each load mode. The load patterns whose physical property fitness meets the fitness threshold are added to the first type of load pattern group; The load patterns whose physical property fit does not meet the fit threshold are added to the second type of load pattern group; Among them, a multivariate heat lifetime consumption rate assessment model is constructed based on the first type of load mode group and the second type of load mode group, respectively, including: With multiple Class I load patterns as targets, multiple simulation domains and environmental boundary conditions are defined in combination with the multivariate load factors, wherein the environmental boundary conditions include at least ambient temperature, wind speed, and solar radiation. By combining the structural information of multiple simulation domains and the target main transformer, a local simulation model and simulation call relationship are established. A temperature-thermal stress binary lifetime mapping model is constructed, wherein the temperature-thermal stress binary lifetime mapping model takes temperature and material properties as inputs and outputs the thermal lifetime consumption rate. Integrate the local simulation model, the simulation call relationship, and the temperature-thermal stress binary life mapping model to obtain prior data to evaluate the model, wherein the output of the local simulation model is connected to the input of the temperature-thermal stress binary life mapping model; The multivariate heat lifetime consumption rate assessment model, constructed based on the first type of load mode group and the second type of load mode group respectively, also includes: Using the two types of load pattern groups as indexes, the typical load data are traversed to extract data and obtain the total training dataset. Cluster analysis of the total training dataset is performed using the elbow method to obtain multiple binary training data subsets. Based on multiple subsets of the two types of training data, multiple data-driven consumption rate prediction models are constructed and trained respectively, and a mapping relationship between the two types of load pattern groups and the multiple consumption rate prediction models is established to obtain the data fitting evaluation model.

2. The method for predicting the thermal life of a main transformer based on multi-variable load modeling as described in claim 1, characterized in that, The multi-factor load factors include at least load current, harmonic content, power factor, load timing fluctuation period, and load fluctuation entropy.

3. The method for predicting the thermal life of a main transformer based on multi-variable load modeling as described in claim 1, characterized in that, Based on preset multivariate load factors, the typical load data is analyzed for classification and modeling to obtain a load pattern set, including: Load features are extracted from the typical load data based on the aforementioned multivariate load factors; Establish a load feature space, and map the load feature extraction results to the load feature space for cluster analysis; Multiple load patterns are defined based on the clustering analysis results, and a load pattern set is obtained. Each load pattern includes a typical feature set and a pattern boundary.

4. The method for predicting the thermal life of a main transformer based on multi-variable load modeling as described in claim 1, characterized in that, The simulation domain includes at least a current-generated heat domain, a harmonic-generated heat domain, and a heat dissipation domain, wherein the heat dissipation domain includes heat conduction paths, heat convection paths, and heat radiation paths.

5. The method for predicting the thermal life of a main transformer based on multi-variable load modeling as described in claim 4, characterized in that, By combining the structural information of multiple simulation domains and the target main transformer, a local simulation model and simulation call relationship are established, including: The prior model is invoked by traversing multiple simulation domains to generate prior model invocation results, which include current heat generation model group, harmonic heat generation model group and heat dissipation model group. Multiple typical controlled locations of the target main transformer are obtained as local simulation targets, and the local simulation model is constructed by combining the results of the prior model call. Model call matching is performed based on the typical feature set of each type of load mode, and the simulation call relationship is established accordingly.

6. The method for predicting the thermal life of a main transformer based on multi-variable load modeling as described in claim 5, characterized in that, Acquire the lifecycle load data of the target main transformer, and combine it with the first type of load mode group, the second type of load mode group, and the multi-element thermal life consumption rate assessment model to predict the thermal life of the target main transformer, including: Based on the lifecycle load data, extract the first discrete load data corresponding to the first type of load pattern group; The first discrete load data is input into the local simulation model of the multivariate thermal lifetime consumption rate evaluation model, and simulation analysis is performed in combination with the simulation call relationship to obtain the temperature sequence of multiple typical controlled locations. Activate the temperature-thermal stress binary lifetime mapping model to map multiple temperature sequences into thermal lifetime consumption rate sequences; Based on the integration operation of multiple thermal lifetime consumption rate sequences, the thermal lifetime consumption of multiple typical controlled locations is obtained, and the most unfavorable controlled location is defined accordingly. The thermal life of the target main transformer is calculated based on the most unfavorable controlled location, and the output is the thermal life prediction result.

7. The method for predicting the thermal life of a main transformer based on multi-variable load modeling as described in claim 6, characterized in that, Before performing integration based on multiple heat lifetime consumption rate sequences, the process also includes: Calculate the average slope of the multiple thermal lifetime consumption rate sequences; Serialize multiple thermal lifetime consumption rate sequences based on the mean slope of the sequence, and retain the first preset number of thermal lifetime consumption rate sequences.

Images