A transformer temperature prediction method

Abstract

This invention discloses a method for predicting transformer temperature. The method inputs transformer temperature sequence data into an optimized temperature prediction model to obtain the predicted temperature result. The optimized temperature prediction model is obtained through the following steps: acquiring historical operating data of the transformer, including aligned temperature label sequences and sample temperature sequences; performing multi-scale decomposition on the temperature label sequences to obtain sequence trend, sequence fluctuation, and sequence residual terms; enhancing the dynamic capability features of the sample temperature sequences based on the temperature label sequences, sequence trend, sequence fluctuation, and sequence residual terms to obtain an enhanced feature set; and training and optimizing the initialized temperature prediction model based on the enhanced feature set to obtain the optimized temperature prediction model. This method can effectively improve the accuracy of transformer temperature prediction. This invention relates to the field of power equipment condition monitoring and prediction technology.

Description

Technical Field

[0001] This invention relates to the field of power equipment condition monitoring and prediction technology, and in particular to a method for predicting transformer temperature. Background Technology

[0002] In power transmission and distribution networks, power transformers, as core hubs for voltage transformation and energy transmission, directly determine the safety, stability, and reliability of the power grid. The transformer core, as one of the core components of the transformer, directly reflects its operating condition through temperature changes. Excessively high core temperatures accelerate the aging of insulation materials, shorten the transformer's lifespan, and in severe cases, cause core overheating faults, resulting in significant economic losses and power grid accidents. Therefore, achieving accurate prediction of transformer core temperature and anticipating temperature change trends is of great practical significance for condition-based maintenance, fault warning, and safe operation of transformers.

[0003] Currently, related technologies are usually based on algorithms such as neural networks and support vector machines to predict the core temperature in transformers. However, since these methods mostly use raw monitoring data directly, the models have difficulty capturing feature information, resulting in unsatisfactory accuracy in temperature prediction.

[0004] Therefore, the problems with the relevant technologies still need to be solved and optimized. Summary of the Invention

[0005] The purpose of this invention is to at least partially solve one of the technical problems existing in the related art.

[0006] Therefore, one objective of this invention is to provide a transformer temperature prediction method that can effectively improve the accuracy of transformer temperature prediction.

[0007] To achieve the above-mentioned technical objectives, the technical solutions adopted in the embodiments of this application include: In a first aspect, embodiments of this application provide a method for predicting transformer temperature, including: Obtain the temperature sequence data of the transformer; The temperature sequence data is input into the optimized temperature prediction model to obtain the temperature prediction result of the transformer; The optimized temperature prediction model is obtained through the following steps: Obtain the historical operating data of the transformer, which includes aligned temperature label sequences and sample temperature sequences; The temperature label sequence is decomposed into a multi-scale component to obtain a sequence trend term, a sequence fluctuation term, and a sequence residual term. Based on the temperature label sequence, the sequence trend term, the sequence fluctuation term, and the sequence residual term, the sample temperature sequence is dynamically enhanced to obtain an enhanced feature set. Based on the enhanced feature set, the initial temperature prediction model is trained and optimized to obtain the optimized temperature prediction model.

[0008] In addition, the method according to the above embodiments of this application may also have the following additional technical features: Furthermore, in one embodiment of this application, the step of performing multi-scale decomposition on the temperature label sequence to obtain a sequence trend term, a sequence fluctuation term, and a sequence residual term includes: Obtain a short-scale window and a long-scale window, wherein the window size of the long-scale window is larger than the window size of the short-scale window; Based on the short-scale window, a short-scale moving average is performed on the temperature label sequence to obtain the short-scale temperature term; Based on the long-scale window, a long-scale moving average is performed on the temperature label sequence to obtain the sequence trend term; The sequence fluctuation term and the sequence residual term are obtained based on the short-scale temperature term, the sequence trend term, and the temperature label sequence.

[0009] Further, in one embodiment of this application, obtaining the sequence fluctuation term and the sequence residual term based on the short-scale temperature term, the sequence trend term, and the temperature label sequence includes: The difference analysis between the short-scale temperature term and the sequence trend term is performed to obtain the sequence fluctuation term. Difference analysis is performed on the short-scale temperature term and the temperature label sequence to obtain the sequence residual term.

[0010] Furthermore, in one embodiment of this application, the step of performing dynamic capability feature enhancement on the sample temperature sequence based on the temperature label sequence, the sequence trend term, the sequence fluctuation term, and the sequence residual term to obtain an enhanced feature set includes: The temperature label sequence, the sequence trend term, the sequence fluctuation term, the sequence residual term, and the sample temperature sequence are fused using feature matrices to obtain a fused feature matrix; The fused feature matrix is ​​enhanced to obtain the enhanced feature set.

[0011] Further, in one embodiment of this application, the sample temperature sequence includes several temperature samples, each of which has a different sampling time point in the sample temperature sequence; the feature matrix fusion of the temperature label sequence, the sequence trend term, the sequence fluctuation term, the sequence residual term, and the sample temperature sequence to obtain a fused feature matrix includes: Obtain the current temperature sample and the embedded feature matrix; Based on the current sampling time point of the temperature sample, multi-dimensional feature extraction is performed on the temperature label sequence, the sequence trend item, the sequence fluctuation item, and the sequence residual item to obtain target temperature data corresponding to the temperature label sequence, target trend data corresponding to the sequence trend item, target fluctuation data corresponding to the sequence fluctuation item, and target residual data corresponding to the sequence residual item. Based on the target temperature data, the target trend data, the target fluctuation data, the target residual data, and the current temperature sample, the current embedding feature matrix is ​​updated to obtain the updated embedding feature matrix. The fused feature matrix is ​​obtained based on the updated embedded feature matrix.

[0012] Furthermore, in one embodiment of this application, the step of performing matrix feature enhancement on the fused feature matrix to obtain the enhanced feature set includes: The fused feature matrix is ​​subjected to difference analysis to obtain first-order change information; Perform the square root operation on the absolute value of each element in the fusion feature matrix to obtain the energy mapping feature; Based on the first-order change information and the energy mapping features, the fused feature matrix is ​​spliced ​​to obtain the enhanced feature set.

[0013] Further, in this embodiment of the application, the step of training and optimizing the initial temperature prediction model based on the enhanced feature set to obtain the optimized temperature prediction model includes: Obtain the initial population corresponding to the initialized temperature prediction model, wherein the initial population includes several parameter individuals; Based on the enhanced feature set, the initial population is optimized to obtain the optimal population individuals; Based on the individual hyperparameters of the optimal population individuals, the model coefficients of the initialized temperature prediction model are optimized to obtain the optimized temperature prediction model.

[0014] Further, in this embodiment of the application, the step of optimizing the initial population based on the enhanced feature set to obtain the optimal population individuals includes: The enhanced feature set is subjected to sample distance analysis to obtain a sample distance matrix; Based on the sample distance matrix and the enhanced feature set, an individual fitness analysis is performed on the initial population to obtain a first intermediate individual. The first intermediate individual is the parameter individual with the smallest fitness among all parameter individuals in the initial population. Based on the first intermediate individual and the sample distance matrix, the initial population is optimized to obtain the optimal population individual.

[0015] Further, the step of performing population optimization on the initial population based on the first intermediate individual and the sample distance matrix to obtain the optimal population individual includes: Obtain a second intermediate individual, which is either the first intermediate individual or the third intermediate individual in the previous optimization process; Individual optimization is performed on all the current parameter individuals to obtain several updated parameter individuals; Based on all the updated individual parameters, the second intermediate individual is optimized to obtain the third intermediate individual in the current optimization process; The optimal population individual is obtained based on the third intermediate individual in the current optimization process and all the updated parameter individuals.

[0016] Furthermore, in this embodiment of the application, the step of performing individual optimization on all current parameter individuals to obtain a plurality of updated parameter individuals includes: The current parameter individuals are subjected to balance-guided optimization to obtain the first optimized individual; The current parameter individual is optimized by fluctuation perturbation to obtain the second optimized individual; The first optimized individual and the second optimized individual are merged to obtain the third optimized individual; Individual fitness screening is performed on the third optimized individual and the current parameter individual to obtain the updated parameter individual.

[0017] Secondly, embodiments of this application provide a transformer temperature prediction system, comprising: The first processing unit is used to acquire the temperature sequence data of the transformer; The second processing unit is used to input the temperature sequence data into the optimized temperature prediction model to obtain the temperature prediction result of the transformer. The optimized temperature prediction model is obtained through the following steps: Obtain the historical operating data of the transformer, which includes aligned temperature label sequences and sample temperature sequences; The temperature label sequence is decomposed into a multi-scale component to obtain a sequence trend term, a sequence fluctuation term, and a sequence residual term. Based on the temperature label sequence, the sequence trend term, the sequence fluctuation term, and the sequence residual term, the sample temperature sequence is dynamically enhanced to obtain an enhanced feature set. Based on the enhanced feature set, the initial temperature prediction model is trained and optimized to obtain the optimized temperature prediction model.

[0018] Thirdly, embodiments of this application also provide an electronic device, including: At least one processor; At least one memory for storing at least one program; When the at least one program is executed by the at least one processor, the at least one processor performs the method described above.

[0019] Fourthly, embodiments of this application also provide a computer-readable storage medium storing a processor-executable program, which, when executed by the processor, is used to implement the above-described method.

[0020] Fifthly, embodiments of this application also provide a computer program product, which includes a computer program stored in a computer-readable storage medium. A processor of an electronic device reads the computer program from the computer-readable storage medium and executes the computer program, causing the electronic device to perform the method described above.

[0021] The advantages and beneficial effects of this application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this application: This application discloses a transformer temperature prediction method. The method acquires transformer temperature sequence data; inputs the temperature sequence data into an optimized temperature prediction model to obtain the transformer temperature prediction result. The optimized temperature prediction model is obtained through the following steps: acquiring historical operating data of the transformer, including aligned temperature label sequences and sample temperature sequences; performing multi-scale decomposition on the temperature label sequences to obtain sequence trend terms, sequence fluctuation terms, and sequence residual terms; enhancing the sample temperature sequences with dynamic capability features based on the temperature label sequences, the sequence trend terms, the sequence fluctuation terms, and the sequence residual terms to obtain an enhanced feature set; and training and optimizing the initialized temperature prediction model based on the enhanced feature set to obtain the optimized temperature prediction model. This method, by performing multi-scale decomposition on the temperature label sequences and then enhancing the decomposed trend terms, fluctuation terms, and residual terms into the sample temperature sequences to obtain an enhanced feature set, enables the optimized model to fully capture the feature information in the data, thereby improving the accuracy of temperature prediction. Attached Figure Description

[0022] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the following description is provided with accompanying drawings of the relevant technical solutions in the embodiments of this application or the prior art. It should be understood that the accompanying drawings described below are only for the purpose of clearly illustrating some embodiments of the technical solutions in this application. For those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.

[0023] Figure 1 A flowchart illustrating a transformer temperature prediction method provided in an embodiment of this application; Figure 2 A schematic diagram of the framework of a transformer temperature prediction system provided in an embodiment of this application; Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0024] The embodiments of this application are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain this application, and should not be construed as limiting this application. The step numbers in the following embodiments are set only for ease of explanation, and there is no limitation on the order between the steps. The execution order of each step in the embodiments can be adaptively adjusted according to the understanding of those skilled in the art.

[0025] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing embodiments of this application only and is not intended to limit this application.

[0026] Currently, related technologies typically rely on algorithms such as neural networks and support vector machines to predict the core temperature of transformers. However, since these methods mostly model raw monitoring data in sequence form directly, they fail to effectively extract multi-scale features from the sequence. This makes it difficult for the model to capture the patterns of transformer temperature changes, resulting in unsatisfactory accuracy in subsequent temperature predictions. Furthermore, these methods usually only use raw monitoring data as input features for the temperature prediction model, failing to fully integrate historical information and sequence decomposition features from the monitoring data. This insufficient feature representation capability also contributes to unsatisfactory accuracy in subsequent temperature predictions.

[0027] In addition, some related technologies use hyperparameter optimization algorithms such as grid search and random search to optimize the model. These technologies suffer from low optimization efficiency and are prone to getting trapped in local optima. They often fail to obtain the optimal combination of hyperparameters, which affects the model's prediction accuracy and stability, and consequently makes the accuracy of subsequent temperature predictions unsatisfactory.

[0028] It should be noted that the aforementioned related technologies are only used to assist in understanding the technical solutions of this application and do not mean that they belong to the publicly disclosed prior art.

[0029] In view of this, embodiments of this application provide a transformer temperature prediction method. This method decomposes the temperature label sequence into multiple scales, which can effectively separate the long-term trend, short-term fluctuations and high-frequency residuals (i.e., the trend term, fluctuation term and residual term) of the temperature sequence. It can fully capture the multi-scale dynamic features of temperature changes, improve the richness and accuracy of features, and thus help improve the accuracy of temperature prediction models.

[0030] Furthermore, this method enhances the features obtained from the decomposition, such as trend terms, fluctuation terms, and residual terms, into the sample temperature sequence to obtain an enhanced feature set. Specifically, it obtains first-order change information and energy mapping features through difference analysis and element square root operation on the fused feature matrix, and then concatenates the obtained first-order change information and energy mapping features into the fused feature matrix. This can further enrich the feature expression of the input features, effectively improve the model's sensitivity to temperature change trends, and thus help improve the accuracy of temperature prediction models in making temperature predictions.

[0031] Furthermore, this method trains and optimizes the temperature prediction model based on an enhanced feature set. Specifically, it combines the sample distance matrix and optimizes the population through two dimensions: balanced guided optimization and fluctuation perturbation optimization. This results in an optimized temperature prediction model that can suppress the model from getting trapped in local optima, improve the efficiency and effectiveness of parameter optimization, and effectively enhance the model's prediction accuracy and stability. Ultimately, this improves the accuracy of temperature prediction.

[0032] Reference Figure 1 In this application embodiment, a transformer temperature prediction method includes: Step 110: Obtain the temperature sequence data of the transformer; Step 120: Input the temperature sequence data into the optimized temperature prediction model to obtain the temperature prediction result of the transformer; The core components of a transformer include the core and several winding coils. In a first embodiment, the transformer's temperature sequence data can specifically be a combination of real-time acquired temperature sequences from several winding coils and the core temperature sequence; alternatively, in a second embodiment, the transformer's temperature sequence data can be a combination of real-time acquired temperature sequences from several winding coils. The temperature sequence data is input into an optimized temperature prediction model, which then predicts the temperature of the transformer and / or the transformer core to obtain the transformer's temperature prediction result.

[0033] The optimized temperature prediction model is obtained through the following steps: Step 130: Obtain the historical operating data of the transformer, which includes aligned temperature tag sequences and sample temperature sequences; In this embodiment, the historical operating data of the transformer can be a combination of temperature sequences of several winding groups and core temperature sequences of the transformer during a certain historical period. For ease of understanding, this embodiment takes a transformer whose core includes the core and three winding groups as an example. The historical operating data can specifically include the historical temperature sequence of winding group A, the historical temperature sequence of winding group B, the historical temperature sequence of winding group C, and the historical temperature sequence of the core. Among them, the temperature label sequence can be the historical temperature sequence of the core, and the sample temperature sequence can be the historical temperature sequence of winding group A, the historical temperature sequence of winding group B, and the historical temperature sequence of winding group C. The temperature label sequence and the sample temperature sequence are aligned on the time axis.

[0034] It is understandable that in practical applications, after obtaining the transformer's operating data (such as temperature series data or historical operating data), the transformer's operating data can be preprocessed. This data preprocessing can specifically include operations such as outlier removal, missing value imputation, and data normalization. There are already various specific data preprocessing methods, which will not be elaborated here.

[0035] The temperature label sequence is decomposed into a multi-scale component to obtain a sequence trend term, a sequence fluctuation term, and a sequence residual term. In the embodiments of this application, the temperature label sequence can be decomposed into multiple scales to obtain a sequence trend term that characterizes the long-term overall trend of the iron core temperature sequence, a sequence fluctuation term that characterizes the short-term fluctuations in the iron core temperature sequence, and a sequence residual term that characterizes the high-frequency noise or sudden changes in the iron core temperature sequence.

[0036] Step 140: In some embodiments, the multi-scale decomposition of the temperature label sequence to obtain a sequence trend term, a sequence fluctuation term, and a sequence residual term includes: Obtain a short-scale window and a long-scale window, wherein the window size of the long-scale window is larger than the window size of the short-scale window; Based on the short-scale window, a short-scale moving average is performed on the temperature label sequence to obtain the short-scale temperature term; Based on the long-scale window, a long-scale moving average is performed on the temperature label sequence to obtain the sequence trend term; In this embodiment of the application, a pre-set short-scale window and a long-scale window can be obtained. The specific window size of the short-scale window and the long-scale window can be set according to the actual situation, as long as the constraint condition that the window size of the long-scale window is greater than the window size of the short-scale window is met.

[0037] Short-scale moving averages can be achieved by using a short-scale window to average all temperature elements sorted along the time axis in a temperature label sequence, capturing short-term local fluctuations in the core temperature sequence and obtaining a short-scale temperature term. Long-scale moving averages are similar to short-scale moving averages, but the difference lies in that they are used to capture the long-term overall trend of the core temperature sequence, and the long-scale temperature term obtained from the long-scale moving average is determined as the sequence trend term.

[0038] The sequence fluctuation term and the sequence residual term are obtained based on the short-scale temperature term, the sequence trend term, and the temperature label sequence.

[0039] Further, the step of obtaining the sequence fluctuation term and the sequence residual term based on the short-scale temperature term, the sequence trend term, and the temperature label sequence includes: The difference analysis between the short-scale temperature term and the sequence trend term is performed to obtain the sequence fluctuation term. Difference analysis is performed on the short-scale temperature term and the temperature label sequence to obtain the sequence residual term.

[0040] In the embodiments of this application, for the sequence fluctuation term, the difference analysis can be performed by calculating the difference between each temperature element in the short-scale temperature term and the temperature element at the corresponding time point in the sequence trend term, thereby characterizing the sequence fluctuation term that represents the short-term fluctuation in the core temperature sequence; while the difference analysis of the sequence residual term is similar to the difference analysis of the sequence fluctuation term, and can be simply deduced to finally obtain the sequence residual term that characterizes the high-frequency noise and sudden changes in the core temperature sequence.

[0041] Step 150: Based on the temperature label sequence, the sequence trend term, the sequence fluctuation term, and the sequence residual term, perform dynamic capability feature enhancement on the sample temperature sequence to obtain an enhanced feature set; In the embodiments of this application, the feature information in the temperature label sequence, sequence trend term, sequence fluctuation term and sequence residual term can be dynamically enhanced into the sample temperature sequence to obtain an enhanced feature set.

[0042] In some embodiments, the step of performing dynamic capability feature enhancement on the sample temperature sequence based on the temperature label sequence, the sequence trend term, the sequence fluctuation term, and the sequence residual term to obtain an enhanced feature set includes: The temperature label sequence, the sequence trend term, the sequence fluctuation term, the sequence residual term, and the sample temperature sequence are fused using feature matrices to obtain a fused feature matrix; Further, the sample temperature sequence includes several temperature samples, each with a different sampling time point in the sample temperature sequence; the feature matrix fusion of the temperature label sequence, the sequence trend term, the sequence fluctuation term, the sequence residual term, and the sample temperature sequence to obtain a fused feature matrix includes: Obtain the current temperature sample and the embedded feature matrix; Based on the current sampling time point of the temperature sample, multi-dimensional feature extraction is performed on the temperature label sequence, the sequence trend item, the sequence fluctuation item, and the sequence residual item to obtain target temperature data corresponding to the temperature label sequence, target trend data corresponding to the sequence trend item, target fluctuation data corresponding to the sequence fluctuation item, and target residual data corresponding to the sequence residual item. Based on the target temperature data, the target trend data, the target fluctuation data, the target residual data, and the current temperature sample, the current embedding feature matrix is ​​updated to obtain the updated embedding feature matrix. The fused feature matrix is ​​obtained based on the updated embedded feature matrix.

[0043] In this embodiment, since the sample temperature sequence includes several coiled line groups whose temperatures are collected at multiple consecutive sampling time points, each temperature sample in the sample temperature sequence can be a combination of the temperatures of several coiled line groups at a certain same sampling time point. Feature matrix fusion can be performed cyclically, constructing a fused feature matrix from the temperature label sequence, sequence trend term, sequence fluctuation term, sequence residual term, and sample temperature sequence according to the sampling order on the time axis.

[0044] Specifically, for the first iteration, the embedded feature matrix can be an empty matrix, and the current temperature sample can be a combination of the temperatures of the winding line group at the first sampling time point in the sample temperature sequence. Multidimensional feature extraction can involve extracting the core temperature (i.e., target temperature data) at the sampling time point corresponding to the temperature label sequence, extracting the sequence trend value (i.e., target trend data) at the corresponding sampling time point in the sequence trend term, extracting the sequence fluctuation value (i.e., target fluctuation data) at the corresponding sampling time point in the sequence fluctuation term, and extracting the sequence residual value (i.e., target residual data) at the corresponding sampling time point in the sequence residual term.

[0045] The embedding update concatenates the target temperature data, target trend data, target fluctuation data, and target residual data into embedding features, and then stacks them row by row into the embedding feature matrix to obtain the updated embedding feature matrix.

[0046] For the second and subsequent iterations, the acquired embedding feature matrix can be the updated embedding feature matrix from the previous iteration, and the current temperature sample can be a combination of the temperatures of the coiled line group at the corresponding sampling time point in the sample temperature sequence. For example, for the second iteration, the current temperature sample can be a combination of the temperatures of the coiled line group at the second sampling time point in the sample temperature sequence; and for the third iteration, the current temperature sample can be a combination of the temperatures of the coiled line group at the second sampling time point in the sample temperature sequence.

[0047] As for the content of multidimensional feature extraction and embedding update in the second and subsequent iterations, it can be simply deduced by analogy with the content of multidimensional feature extraction and embedding update in the first iteration.

[0048] It should be noted that, for any given iteration, when obtaining the updated embedding feature matrix in the current iteration, it can be determined whether each temperature sample in the sample temperature sequence has participated in the multi-dimensional feature extraction and embedding update steps. Specifically, if at least one temperature sample in the sample temperature sequence has not participated in the multi-dimensional feature extraction and embedding update steps, the process can return to obtaining the current temperature sample and embedding feature matrix; or, if all temperature samples in the sample temperature sequence have participated in the multi-dimensional feature extraction and embedding update steps, the updated embedding feature matrix obtained in the current iteration can be determined as the fused feature matrix.

[0049] The fused feature matrix is ​​enhanced to obtain the enhanced feature set.

[0050] Further, the step of performing matrix feature enhancement on the fused feature matrix to obtain the enhanced feature set includes: The fused feature matrix is ​​subjected to difference analysis to obtain first-order change information; Perform the square root operation on the absolute value of each element in the fusion feature matrix to obtain the energy mapping feature; Based on the first-order change information and the energy mapping features, the fused feature matrix is ​​spliced ​​to obtain the enhanced feature set.

[0051] In this embodiment, differential analysis can employ differential computation to calculate the first-order rate of change of the fused feature matrix to capture the changing trend of the features, and the calculated first-order rate of change is recorded as first-order change information. The absolute value square root operation can be performed by taking the square root of the absolute value of each element in the fused feature matrix to obtain the energy mapping feature used to characterize the feature amplitude information. Feature concatenation can be performed by concatenating the fused feature matrix, first-order change information, and energy mapping feature column-wise to expand the feature dimension and enhance the feature representation capability, thereby obtaining an enhanced feature set.

[0052] Step 160: Based on the enhanced feature set, train and optimize the initialized temperature prediction model to obtain the optimized temperature prediction model.

[0053] In this embodiment of the application, the temperature prediction model can be trained and optimized based on the feature information in the enhanced feature set to obtain the optimized temperature prediction model.

[0054] In some embodiments, training and optimizing the initial temperature prediction model based on the enhanced feature set to obtain the optimized temperature prediction model includes: Obtain the initial population corresponding to the initialized temperature prediction model, wherein the initial population includes several parameter individuals; In this embodiment, the initial temperature prediction model can be a kernel ridge regression (KRR) prediction model, which employs a radial basis function (RBF) kernel function, which can be expressed as:

[0055] in, This is a functional representation of the radial basis kernel function; It is a natural constant; The squared Euclidean distance between samples; is the kernel width parameter in the radial basis kernel function.

[0056] Understandably, initializing the population can be achieved by pre-setting the parameters and hyperparameter search space, then randomly generating several parameter individuals within the search space through initialization, and finally determining the initial population based on all these parameter individuals. Specifically, for the pre-set parameters and hyperparameter search space, the hyperparameter dimension is 2, meaning the parameters to be optimized are... Lower bound of search space Upper bound of search space Population size 20 (i.e., the maximum number of individuals in the initial population is 20), maximum number of iterations 30 (i.e., the maximum number of population optimizations in the initial population is 30).

[0057] It should be noted that for any parameter individual in the initial population, this parameter individual is a 2-dimensional vector, which includes two pieces of information: log10(σ) and log10(λ), where λ is a regularization parameter used to balance the complexity and fitting accuracy of the temperature prediction model. Furthermore, the population size and maximum number of iterations in the examples provided in this application are merely optional examples for ease of understanding; the specific values ​​can be flexibly set according to actual circumstances.

[0058] Based on the enhanced feature set, the initial population is optimized to obtain the optimal population individuals; Further, the step of optimizing the initial population based on the enhanced feature set to obtain the optimal population individuals includes: The enhanced feature set is subjected to sample distance analysis to obtain a sample distance matrix; Based on the sample distance matrix and the enhanced feature set, an individual fitness analysis is performed on the initial population to obtain a first intermediate individual. The first intermediate individual is the parameter individual with the smallest fitness among all parameter individuals in the initial population. In this embodiment, before optimizing the initial population, the sample distance analysis can be performed by pre-calculating the squared Euclidean distance between each pair of samples in the enhanced feature set, and constructing a sample distance matrix based on the obtained squared Euclidean distances between all samples. Individual fitness analysis can be performed by calculating the fitness of each parameter individual in the initial population, and determining the parameter individual with the lowest fitness as the first intermediate individual.

[0059] For example, for any parameter individual in the initial population, a kernel matrix corresponding to the augmented feature set can be constructed based on the two pieces of information contained in the parameter individual and the pre-calculated sample distance matrix, and the current model coefficients of the temperature prediction model can be solved. Subsequently, the root mean square error (RSME) of the augmented feature set is calculated based on the current model coefficients of the temperature prediction model, and this RSME is recorded as the fitness of the parameter individual. After obtaining the fitness of all parameter individuals, the fitness of each parameter individual can be retained while comparing the fitness of each parameter individual, and the parameter individual with the smallest fitness is determined as the first intermediate individual.

[0060] Based on the first intermediate individual and the sample distance matrix, the initial population is optimized to obtain the optimal population individual.

[0061] Further, the step of performing population optimization on the initial population based on the first intermediate individual and the sample distance matrix to obtain the optimal population individual includes: Obtain a second intermediate individual, which is either the first intermediate individual or the third intermediate individual in the previous optimization process; In the embodiments of this application, each individual in the initial population can be optimized iteratively, and the optimal individual in the population is obtained after the population optimization process is completed. Specifically, for the first optimization process, the second intermediate individual can be the first intermediate individual; or, for the second and subsequent optimization processes, the second intermediate individual can be the third intermediate individual obtained in the previous optimization process. For example, for the second optimization process, the second intermediate individual can be the third intermediate individual obtained in the first optimization process; or, for the fourth optimization process, the second intermediate individual can be the third intermediate individual obtained in the third optimization process.

[0062] Individual optimization is performed on all the current parameter individuals to obtain several updated parameter individuals; Furthermore, the step of performing individual optimization on all current parameter individuals to obtain a plurality of updated parameter individuals includes: The current parameter individuals are subjected to balance-guided optimization to obtain the first optimized individual; The current parameter individual is optimized by fluctuation perturbation to obtain the second optimized individual; The first optimized individual and the second optimized individual are merged to obtain the third optimized individual; Individual fitness screening is performed on the third optimized individual and the current parameter individual to obtain the updated parameter individual.

[0063] In the embodiments of this application, for any one of the parameter individuals among all parameter individuals in a certain optimization process, denoted as the current parameter individual, the balance-guided optimization can be based on the balance-guided strategy in the Equilibrium Optimizer (EO) to optimize the kernel width parameter and regularization parameter in the current parameter individual to obtain the first optimized individual; while the wave perturbation optimization can be based on the wave optical perturbation strategy in the Wave Optics Optimizer (WOO) to optimize the kernel width parameter and regularization parameter in the current parameter individual to obtain the second optimized individual.

[0064] Understandably, the individual fusion process can begin by assigning a fusion weight to each of the first and second optimized individuals. For example, assigning a fusion weight of 0.6 to the first optimized individual and a fusion weight of 0.4 to the second optimized individual; or assigning a fusion weight of 0.3 to the first optimized individual and a fusion weight of 0.7 to the second optimized individual. Then, under the constraint of the upper and lower bounds of the search space, the first and second optimized individuals are fused based on all the fusion weights to ensure that the resulting new individual is within the search space, and this new individual is designated as the third optimized individual.

[0065] It should be noted that the individual fitness selection process can first involve calculating the fitness of the third optimized individual, which can be derived by analogy from the fitness of the aforementioned parameter individuals. Then, the fitness of the third optimized individual is compared with that of the current parameter individual to obtain the updated parameter individual. Specifically, if the fitness of the third optimized individual is less than that of the current parameter individual, the third optimized individual can be selected as the updated parameter individual; or, if the fitness of the third optimized individual is greater than or equal to that of the current parameter individual, the current parameter individual can be selected as the updated parameter individual.

[0066] Based on all the updated individual parameters, the second intermediate individual is optimized to obtain the third intermediate individual in the current optimization process; The optimal population individual is obtained based on the third intermediate individual in the current optimization process and all the updated parameter individuals.

[0067] In this embodiment of the application, for a certain optimization process, after obtaining all the updated parameter individuals of the optimization process, the fitness of the updated parameter individuals can be compared with the fitness of the second intermediate individual of the optimization process, and the individual with the smallest fitness among all individuals can be determined as the third intermediate individual.

[0068] It is understandable that, for a certain optimization process, after obtaining the third intermediate individual and all updated parameter individuals in the optimization process, if the optimization process is not the last optimization process, the step of obtaining the second intermediate individual can be returned based on the obtained third intermediate individual and all updated parameter individuals; or, if the optimization process is the last optimization process, the optimization process can be exited and the third intermediate individual can be determined as the optimal population individual.

[0069] Based on the individual hyperparameters of the optimal population individuals, the model coefficients of the initialized temperature prediction model are optimized to obtain the optimized temperature prediction model.

[0070] In this embodiment, the individual hyperparameters of the optimal population individual can specifically be the regularization parameter and kernel width parameter of the optimal population individual. Model coefficient optimization can be based on the individual hyperparameters and the pre-calculated sample distance matrix, recalculating the corresponding kernel matrix and solving the corresponding model coefficients, and then determining the optimized temperature prediction model based on the temperature prediction model initialized with the finally obtained model coefficient kernel.

[0071] Reference Figure 2 The transformer temperature prediction system proposed in this application includes: The first processing unit 101 is used to acquire the temperature sequence data of the transformer; The second processing unit 102 is used to input the temperature sequence data into the optimized temperature prediction model to obtain the temperature prediction result of the transformer. The optimized temperature prediction model is obtained through the following steps: Obtain the historical operating data of the transformer, which includes aligned temperature label sequences and sample temperature sequences; The temperature label sequence is decomposed into a multi-scale component to obtain a sequence trend term, a sequence fluctuation term, and a sequence residual term. Based on the temperature label sequence, the sequence trend term, the sequence fluctuation term, and the sequence residual term, the sample temperature sequence is dynamically enhanced to obtain an enhanced feature set. Based on the enhanced feature set, the initial temperature prediction model is trained and optimized to obtain the optimized temperature prediction model.

[0072] It is understood that the content of the above method embodiments is applicable to this system embodiment. The specific functions implemented in this system embodiment are the same as those in the above method embodiments, and the beneficial effects achieved are also the same as those achieved in the above method embodiments.

[0073] Reference Figure 3 This application also provides an electronic device, including: At least one processor 201; At least one memory 202 is used to store at least one program; When the at least one program is executed by the at least one processor 201, the at least one processor 201 implements the above-described method embodiments.

[0074] Similarly, it can be understood that the content of the above method embodiments is applicable to this device embodiment. The specific functions implemented by this device embodiment are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0075] This application also provides a computer-readable storage medium storing a program executable by a processor 201, which, when executed by the processor 201, is used to implement the above-described method embodiments.

[0076] Similarly, the content of the above method embodiments is applicable to the present computer-readable storage medium embodiments. The specific functions implemented by the present computer-readable storage medium embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0077] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0078] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above methods.

[0079] In some alternative embodiments, the functions / operations mentioned in the block diagrams may not occur in the order shown in the operation diagrams. For example, depending on the functions / operations involved, two consecutively shown blocks may actually be executed substantially simultaneously, or the blocks may sometimes be executed in reverse order. Furthermore, the embodiments presented and described in the flowcharts of this application are provided by way of example to provide a more comprehensive understanding of the technology. The disclosed methods are not limited to the operations and logic flows presented herein. Alternative embodiments are contemplated in which the order of various operations is changed and sub-operations described as part of a larger operation are executed independently.

[0080] Furthermore, although this application is described in the context of functional modules, it should be understood that, unless otherwise stated to the contrary, one or more of the functions and / or features may be integrated into a single physical device and / or software module, or one or more functions and / or features may be implemented in a separate physical device or software module. It is also understood that a detailed discussion of the actual implementation of each module is unnecessary for understanding this application. Rather, given the properties, functions, and internal relationships of the various functional modules in the apparatus disclosed herein, the actual implementation of the module will be understood within the scope of conventional technology for an engineer. Therefore, those skilled in the art can implement the application set forth in the claims using ordinary techniques without excessive experimentation. It is also understood that the specific concepts disclosed are merely illustrative and not intended to limit the scope of this application, which is determined by the full scope of the appended claims and their equivalents.

[0081] If a function is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods in the embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0082] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-including system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0083] More specific examples (a non-exhaustive list) of computer-readable media include: electrical connections (electronic devices) having one or more wires, portable computer disk drives (magnetic devices), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Furthermore, computer-readable media can even be paper or other suitable media on which programs can be printed, because programs can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in computer memory.

[0084] It should be understood that various parts of this application can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented using software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0085] In the foregoing description of this specification, the references to terms such as "one embodiment," "another embodiment," or "some embodiments," etc., indicate that a specific feature, structure, material, or characteristic described in connection with an embodiment or example is included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0086] Although embodiments of this application have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of this application, the scope of which is defined by the claims and their equivalents.

[0087] The above is a detailed description of the preferred embodiments of this application, but this application is not limited to the embodiments. Those skilled in the art can make various equivalent modifications or substitutions without departing from the spirit of this application, and these equivalent modifications or substitutions are all included within the scope defined by the claims of this application.

Claims

1. A method for predicting transformer temperature, characterized in that, include: Obtain the temperature sequence data of the transformer; The temperature sequence data is input into the optimized temperature prediction model to obtain the temperature prediction result of the transformer; The optimized temperature prediction model is obtained through the following steps: Obtain the historical operating data of the transformer, which includes aligned temperature label sequences and sample temperature sequences; The temperature label sequence is decomposed into a multi-scale component to obtain a sequence trend component, a sequence fluctuation component, and a sequence residual component. Based on the temperature label sequence, the sequence trend term, the sequence fluctuation term, and the sequence residual term, the sample temperature sequence is dynamically enhanced to obtain an enhanced feature set. Based on the enhanced feature set, the initial temperature prediction model is trained and optimized to obtain the optimized temperature prediction model.

2. The method according to claim 1, characterized in that, The multi-scale decomposition of the temperature label sequence to obtain the sequence trend term, sequence fluctuation term, and sequence residual term includes: Obtain a short-scale window and a long-scale window, wherein the window size of the long-scale window is larger than the window size of the short-scale window; Based on the short-scale window, a short-scale moving average is performed on the temperature label sequence to obtain the short-scale temperature term; Based on the long-scale window, a long-scale moving average is performed on the temperature label sequence to obtain the sequence trend term; The sequence fluctuation term and the sequence residual term are obtained based on the short-scale temperature term, the sequence trend term, and the temperature label sequence.

3. The method according to claim 2, characterized in that, The step of obtaining the sequence fluctuation term and the sequence residual term based on the short-scale temperature term, the sequence trend term, and the temperature label sequence includes: The difference analysis between the short-scale temperature term and the sequence trend term is performed to obtain the sequence fluctuation term. Difference analysis is performed on the short-scale temperature term and the temperature label sequence to obtain the sequence residual term.

4. The method according to claim 1, characterized in that, The step involves enhancing the dynamic capability features of the sample temperature sequence based on the temperature label sequence, the sequence trend term, the sequence fluctuation term, and the sequence residual term, resulting in an enhanced feature set, including: The temperature label sequence, the sequence trend term, the sequence fluctuation term, the sequence residual term, and the sample temperature sequence are fused using feature matrices to obtain a fused feature matrix; The fused feature matrix is ​​enhanced to obtain the enhanced feature set.

5. The method according to claim 4, characterized in that, The sample temperature sequence includes several temperature samples, each with a different sampling time point within the sample temperature sequence; the feature matrix fusion of the temperature label sequence, the sequence trend term, the sequence fluctuation term, the sequence residual term, and the sample temperature sequence yields a fused feature matrix, including: Obtain the current temperature sample and the embedded feature matrix; Based on the current sampling time point of the temperature sample, multi-dimensional feature extraction is performed on the temperature label sequence, the sequence trend item, the sequence fluctuation item, and the sequence residual item to obtain target temperature data corresponding to the temperature label sequence, target trend data corresponding to the sequence trend item, target fluctuation data corresponding to the sequence fluctuation item, and target residual data corresponding to the sequence residual item. Based on the target temperature data, the target trend data, the target fluctuation data, the target residual data, and the current temperature sample, the current embedding feature matrix is ​​updated to obtain the updated embedding feature matrix. The fused feature matrix is ​​obtained based on the updated embedded feature matrix.

6. The method according to claim 4, characterized in that, The step of performing matrix feature enhancement on the fused feature matrix to obtain the enhanced feature set includes: The fused feature matrix is ​​subjected to difference analysis to obtain first-order change information; Perform the square root operation on the absolute value of each element in the fusion feature matrix to obtain the energy mapping feature; Based on the first-order change information and the energy mapping features, the fused feature matrix is ​​spliced ​​to obtain the enhanced feature set.

7. The method according to any one of claims 1-6, characterized in that, The step of training and optimizing the initial temperature prediction model based on the enhanced feature set to obtain the optimized temperature prediction model includes: Obtain the initial population corresponding to the initialized temperature prediction model, wherein the initial population includes several parameter individuals; Based on the enhanced feature set, the initial population is optimized to obtain the optimal population individuals; Based on the individual hyperparameters of the optimal population individuals, the model coefficients of the initialized temperature prediction model are optimized to obtain the optimized temperature prediction model.

8. The method according to claim 7, characterized in that, The step of optimizing the initial population based on the enhanced feature set to obtain the optimal population individuals includes: The enhanced feature set is subjected to sample distance analysis to obtain a sample distance matrix; Based on the sample distance matrix and the enhanced feature set, an individual fitness analysis is performed on the initial population to obtain a first intermediate individual. The first intermediate individual is the parameter individual with the smallest fitness among all parameter individuals in the initial population. Based on the first intermediate individual and the sample distance matrix, the initial population is optimized to obtain the optimal population individual.

9. The method according to claim 8, characterized in that, The step of optimizing the initial population based on the first intermediate individual and the sample distance matrix to obtain the optimal population individuals includes: Obtain a second intermediate individual, which is either the first intermediate individual or the third intermediate individual in the previous optimization process; Individual optimization is performed on all the current parameter individuals to obtain several updated parameter individuals; Based on all the updated individual parameters, the second intermediate individual is optimized to obtain the third intermediate individual in the current optimization process; The optimal population individual is obtained based on the third intermediate individual in the current optimization process and all the updated parameter individuals.

10. The method according to claim 9, characterized in that, The step of performing individual optimization on all current parameter individuals to obtain a plurality of updated parameter individuals includes: The current parameter individuals are subjected to balance-guided optimization to obtain the first optimized individual; The current parameter individual is subjected to fluctuation perturbation optimization to obtain a second optimized individual; The first optimized individual and the second optimized individual are merged to obtain the third optimized individual; Individual fitness screening is performed on the third optimized individual and the current parameter individual to obtain the updated parameter individual.

Images