Power transformer temperature prediction method and system based on multi-feature extraction and fusion

Abstract

The invention discloses a power transformer temperature prediction method and system based on multi-feature extraction and fusion, and the method comprises the steps: carrying out the time-space decoupling standardization and causal emergence clustering of all temperature variables in data, and obtaining a clustering result; extracting a common pattern from each class of the clustering result to obtain an information complemented multivariable time sequence, and performing down-sampling on the information complemented multivariable time sequence to obtain a plurality of subsequences; performing feature extraction on the multivariable time sequence and the sub-sequence after information completion through a dynamic time-frequency domain MLP mechanism to obtain overall features and local features, and fusing the overall features and the local features to obtain time sequence features; acquiring an external environment covariable, inputting the external environment covariable into the self-attention mechanism network to obtain a covariable self-feature, and performing feature fusion on the covariable self-feature and the time sequence feature through a covariable-driven causal convolution fusion mechanism to obtain a fusion feature; and inputting the fusion feature into a dynamic attention feature weighted prediction model for prediction to obtain a prediction result. And the precision and reliability of temperature prediction of the power transformer are obviously improved.

Description

Technical Field

[0001] The present invention relates to the technical field of equipment status monitoring, and in particular to a method and system for predicting temperature of a power transformer based on multi-feature extraction and fusion. Background Art

[0002] The statements in this section merely provide background information related to the present disclosure and do not necessarily constitute prior art.

[0003] With the in-depth advancement of smart grid construction, the power system has an increasingly urgent need for key equipment status monitoring and fault warning. As the core equipment of the power grid, the operating temperature of the power transformer directly reflects the health status and load capacity of the equipment. Accurately predicting the temperature change of the transformer not only helps to optimize the equipment operation and maintenance strategy and extend the service life, but also effectively prevents power outages caused by overheating failures, which is of great significance to ensuring power supply reliability and promoting energy digital transformation.

[0004] The temperature of power transformers is affected by the dynamic coupling of multi-dimensional features such as winding temperature, oil temperature, radiator temperature, and ambient temperature, and has significant nonlinearity and time series correlation. Existing data-driven temperature prediction methods still have obvious limitations in feature mining and modeling, which are mainly manifested in the following problems: (1) When dealing with multiple temperature variables, some existing methods usually simply superimpose all other temperature variables without considering the correlation between variables. This can easily lead to noise introduced by interactions between irrelevant variables, thus affecting the prediction effect.

[0005] (2) Previous methods failed to fully characterize the dynamic interaction mechanism between multi-source features. They mainly captured the characteristics of the internal temperature variables of the power transformer, but did not introduce the dynamic interaction between external environmental covariates, resulting in bottleneck limitations in the prediction effect.

[0006] (3) Common time series modeling methods (such as long short-term memory neural networks (LSTM) and temporal convolutional neural networks (TCN)) are insufficient in terms of time series dependency and complex feature extraction capabilities, making it difficult to effectively capture and extract complex feature relationships. Summary of the invention

[0007] In order to overcome the above-mentioned deficiencies of the prior art, the present invention provides a method and system for predicting the temperature of a power transformer based on multi-feature extraction and fusion, which significantly improves the accuracy and reliability of the temperature prediction of the power transformer.

[0008] To achieve the above objectives, one or more embodiments of the present invention provide the following technical solutions: In a first aspect, the present invention provides a method for predicting temperature of a power transformer based on multi-feature extraction and fusion, comprising: Obtain multivariate time series data consisting of temperature variables for power transformers; Performing spatiotemporal decoupling standardization on the multivariate time series data, and performing causal emergence clustering on each temperature variable in the standardized multivariate time series data to obtain a clustering result; Extracting common patterns from each class of the clustering results, splicing and fusing the common patterns into various classes of sequences, and obtaining a multivariate time series with complete information; Downsampling the information-completed multivariate time series to obtain multiple subsequences; extracting features of the information-completed multivariate time series through a dynamic time-frequency domain MLP mechanism to obtain overall features, extracting features of the subsequences through a dynamic time-frequency domain MLP mechanism to obtain local features, and fusing the overall features and local features to obtain time series features; Obtain external environmental covariates and input them into the self-attention mechanism network to obtain the covariate's own features, and fuse the covariate's own features with the time series features through the covariate-driven causal convolution fusion mechanism to obtain fused features; The fused features are input into the dynamic attention feature weighted prediction model for prediction to obtain a prediction result. Further technical solutions, the clustering results are as follows: Standardize multivariate time series data by temporal and spatial decoupling; Calculating the mutual information between each temperature variable in the standardized multivariate time series data, and converting the mutual information into a similarity matrix; A symmetric Plaus matrix is ​​obtained based on the similarity matrix, and eigenvalue decomposition is performed on the symmetric Plaus matrix to obtain a characteristic matrix; The causal emergence clustering algorithm is used to cluster the feature matrix to obtain a clustering result.

[0009] A further technical solution is to cluster the feature matrix using a causal emergence clustering algorithm as follows: According to the feature matrix, the causal DAG structure between variables is learned by optimizing the objective function to construct a causal graph; Inputting the feature matrix and the causal graph into a graph neural network for feature extraction to obtain causal features; Calculating the cosine similarity of the causal features, and obtaining the causal similarity based on the cosine similarity; The causal similarity is used as a clustering criterion, and causal constraints are incorporated into the hierarchical clustering process to obtain a clustering result.

[0010] A further technical solution to obtain a multivariate time series with complete information is as follows: Adaptive convolutional pooling is used to obtain the significant features of all temperature variables within each category, and random pooling is used to obtain the overall features of all temperature variables within each category; The significant features and overall features of each class are fused separately, and then spliced ​​to each temperature variable in the corresponding class; The long short-term memory network is used to fuse the features of each temperature variable after splicing to obtain a multivariate time series with complete information.

[0011] In a further technical solution, both the overall features and the local features are extracted by a dynamic time-frequency domain MLP mechanism, specifically:

[0012]

[0013]

[0014] in, represents the input sequence, represents the frequency domain MLP modeling results, represents continuous wavelet transform, represents the sequence converted back to the time domain, represents the inverse continuous wavelet transform, represents the sparse MLP, Represents the output features after the dynamic time-frequency domain MLP mechanism.

[0015] A further technical solution is to obtain the fusion features as follows: based on the characteristics of the covariate itself, a dynamic convolution kernel is obtained for each time step, the characteristics of the covariate itself and the time series characteristics are aligned according to the time step, and causal convolution fusion is performed according to the dynamic convolution kernel at each time step.

[0016] As a further technical solution, the dynamic attention feature weighted prediction model embeds a single-layer lightweight attention network into linear regression to achieve dynamic adjustment at the feature level.

[0017] In a second aspect, the present invention provides a power transformer temperature prediction system based on multi-feature extraction and fusion, comprising: A data acquisition module is configured to: acquire multivariate time series data consisting of temperature variables of the power transformer; A similarity clustering module is configured to: perform spatiotemporal decoupling standardization on the multivariate time series data, perform causal emergence clustering on each temperature variable in the standardized multivariate time series data, and obtain a clustering result; An intra-class common pattern extraction module is configured to: extract common patterns from each class of the clustering results, and splice and fuse the common patterns into various classes of sequences to obtain a multivariate time series with complete information; A time series feature modeling module is configured to: downsample the information-completing multivariate time series to obtain multiple subsequences; extract features of the information-completing multivariate time series through a dynamic time-frequency domain MLP mechanism to obtain overall features, extract features of the subsequences through a dynamic time-frequency domain MLP mechanism to obtain local features, and fuse the overall features and local features to obtain time series features; A covariate interaction module is configured to: obtain external environment covariates and input them into the self-attention mechanism network to obtain covariate features, and fuse the covariate features with time series features through a covariate-driven causal convolution fusion mechanism to obtain fused features; The multi-step prediction module is configured to: input the fusion feature into the dynamic attention feature weighted prediction model for prediction to obtain a prediction result.

[0018] In a third aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps in the power transformer temperature prediction method based on multi-feature extraction and fusion as described in the first aspect.

[0019] In a fourth aspect, the present invention provides a computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the steps in the method for predicting temperature of a power transformer based on multi-feature extraction and fusion as described in the first aspect are implemented.

[0020] One or more of the above technical solutions have the following beneficial effects: The present invention adopts a causal emergent clustering algorithm based on mutual information to cluster the correlations of various temperature variables, effectively reducing the influence of irrelevant information; extracts the common patterns of variables by combining random pooling and adaptive convolution pooling, ensures the completion of missing or ambiguous information, and enhances the stability of the prediction model; utilizes the covariate-driven causal convolution fusion mechanism and self-attention mechanism to realize the deep feature fusion between external environmental covariates and temperature variables, significantly improving the accuracy of multi-source feature interaction; finally, a dynamic attention feature weighted prediction model is used to perform multi-step temperature prediction, significantly improving the accuracy and reliability of power transformer temperature prediction.

[0021] When standardizing multivariate time series data of power transformers composed of temperature variables, the present invention considers the dimensional differences between variables and introduces a time-space decoupling standardization technology to decouple time and space standardization respectively, thereby solving the multivariate coupling interference problem.

[0022] The present invention designs a causal emergent clustering algorithm that combines causal discovery with hierarchical clustering to perform correlation clustering on temperature variables, identify the causal dependencies between variables in the data, and use this causal information to guide the clustering process, so that the formation of clusters is not only based on data similarity, but also reflects the causal mechanism between variables.

[0023] The present invention proposes a dynamic time-frequency domain MLP mechanism (TFMLP) for feature extraction. Compared with conventional MLP, it can not only process frequency domain features, but also reduce the complexity of traditional MLP through sparsification, thereby achieving full modeling of time-frequency domain features.

[0024] When the present invention predicts the temperature of a power transformer, a single-layer lightweight attention network is embedded in a linear regression to achieve dynamic adjustment of the feature level, thereby improving the prediction capability for complex scenarios containing multiple temperature variables. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] The accompanying drawings in the specification, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0026] Figure 1 is a flow chart of data correlation clustering in the temperature prediction method according to an embodiment of the present invention; Figure 2 is a common pattern extraction flow chart of the temperature prediction method according to an embodiment of the present invention; Figure 3 is a flow chart of time series feature modeling in a temperature prediction method according to an embodiment of the present invention; Figure 4 It is a flow chart of dynamic time-frequency domain MLP modeling in the temperature prediction method of an embodiment of the present invention; Figure 5 is a flow chart of covariate feature interaction in the temperature prediction method according to an embodiment of the present invention; Figure 6 It is a module diagram of the temperature prediction system according to an embodiment of the present invention. DETAILED DESCRIPTION

[0027] It should be noted that the following detailed descriptions are exemplary and are intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present invention belongs.

[0028] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprising" and / or "including" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof.

[0029] In the absence of conflict, the embodiments of the present invention and the features of the embodiments may be combined with each other.

[0030] Embodiment 1 This embodiment discloses a method for predicting the temperature of a power transformer based on multi-feature extraction and fusion, and the method comprises the following steps: S1: Obtain multivariate time series data consisting of temperature variables of power transformers; In this embodiment, the temperature of the power transformer is affected by the dynamic coupling of multi-dimensional features such as winding temperature, oil temperature, radiator temperature, and ambient temperature. Each temperature variable index of the power transformer, such as oil temperature, winding temperature, radiator temperature, etc., can be regarded as a variable, so each index data constitutes multivariate time series data, defined as ,in represents the quantity of power transformer temperature variable, Indicates the length of the time series data.

[0031] S2: performing spatiotemporal decoupling standardization on the multivariate time series data, and performing causal emergence clustering on each temperature variable in the standardized multivariate time series data to obtain a clustering result; In this embodiment, in order to avoid the large difference in the numerical range between different variables affecting the clustering effect, the multivariate time series data is standardized. In data standardization, the current standardization only considers the difference characteristics of the time dimension in the time series data, but does not consider the dimensional difference between the variables. Therefore, the present invention introduces a time-space decoupling standardization technology to perform time-space decoupling standardization on the multivariate time series data, decouple time and space standardization respectively, and solve the problem of multivariate coupling interference. Specifically, first standardize in the time dimension to eliminate the time series drift of a single variable: the expression is: (1) in, represents the result of normalization in the time dimension, Representation data The mean in the time dimension, Representation data Variance in the time dimension.

[0032] Then standardize in the spatial dimension to eliminate the dimensional differences between variables: (2) (3) (4) (5) (6) in, represents the number of power transformer temperature variables, Indicates Power transformer temperature variables, express No. The variable in The data at the time step, Representation data Each variable in time The mean value of In the The result of spatial normalization at the time step, represents the result of coupling time normalization and space normalization, Representation data Each variable in time The variance of represents the scaling factor, Represents the offset. The scaling factor and offset are an affine transformation added after normalization to enhance the expressiveness of the model.

[0033] In this embodiment, if Figure 1 As shown in Figure 1, the mutual information clustering technique is used to cluster variables based on their correlation, and highly correlated variables are clustered together. This method first calculates the mutual information between variables, replacing the traditional Pearson correlation coefficient to capture the nonlinear dependency between variables. Define mutual information ,in represents the number of temperature variables of the power transformer, and the mutual information calculation formula is as follows: (7) in, represents the mutual information matrix OK, represents the mutual information matrix List, Indicates Variable data, Indicates Variable data, Indicates The variable The value of the time step, Indicates The variable The value of the time step, is the joint / marginal probability distribution of the variable.

[0034] In order to enhance the similarity distinction, the exponential function Gaussian kernel is used to transform the mutual information value to obtain the similarity matrix , expressed as: (8) in, Is a positive constant used to control the decay rate of similarity. This will make the similarity more sensitive to changes in the mutual information value.

[0035] Next, according to the similarity matrix Constructing the degree matrix , It is a diagonal matrix, and the elements are the degrees of each node; then the Laplacian matrix is ​​obtained according to the degree matrix , according to the Laplace matrix, we can get the symmetric Laplace matrix ; Then perform eigenvalue decomposition on the symmetric Laplace matrix to obtain the previous The eigenvectors corresponding to the smallest eigenvalues ​​form the characteristic matrix The above calculation expression is: (9) (10) (11) (12) in, represents the eigenvalue, Represents the number of power transformer temperature variables. Characteristic matrix Each row of is considered as a low-dimensional representation of a variable.

[0036] For the clustering method of feature matrix, the present invention designs a causal emergence clustering algorithm that combines causal discovery and hierarchical clustering, and uses the causal emergence clustering algorithm to cluster the feature matrix to obtain clustering results. Specifically, according to the feature matrix, the causal DAG structure between variables is learned by optimizing the objective function to construct a causal graph; the feature matrix and the causal graph are input into the graph neural network for feature extraction to obtain causal features; the cosine similarity of the causal features is calculated, and the causal similarity is calculated based on the cosine similarity; the causal similarity is used as the clustering standard, and the causal constraints are incorporated into the hierarchical clustering process to obtain the clustering results.

[0037] In traditional clustering methods, the causal dependency between variables is not considered, which will affect the clustering effect. Therefore, the present invention identifies the causal dependency between variables in the data and uses this causal information to guide the clustering process, so that the formation of clusters is not only based on data similarity, but also reflects the causal mechanism between variables.

[0038] Specifically, according to the feature matrix , by optimizing the objective function to learn the causal DAG structure between variables: (13) in, The adjacency matrix of the causal graph represents the causal weights between variables, that is, the variables For variables the causal strength of Represents the reconstruction loss based on the graph structure; represents the constraints of a directed acyclic graph, , Dynamic weights balance sparsity and acyclicity to avoid overfitting; represents the Frobenius norm, The calculation method of is the square root of the sum of the squares of the absolute values ​​of all elements of the matrix; represents the sparsity regularization term, which controls the complexity (sparseness) of the causal graph.

[0039] Next, based on the causal graph (i.e., adjacency matrix ) Extract causal features and enhance the interpretability of clustering: (14) in, Represents causal characteristics, represents a graph neural network, represents the activation function, represents the degree matrix, represents the feature matrix, represents the learnable parameters, Represents the adjacency matrix with self-connection added.

[0040] Finally, define causal similarity as the cluster merging criterion: (15) in, Indicates causal similarity; represents the balance weight between data similarity and causal path contribution; Representing causal characteristics The cosine similarity of is used to measure the variables and variables similarity in causal feature space; Representation variables To variable The set of all causal paths of Indicates the path causal effect.

[0041] In the process of hierarchical clustering, causal constraints are incorporated to form a macro cluster structure and obtain the final clustering result. This introduction of causal information guidance in hierarchical clustering can break through the limitations of traditional data-driven and further improve the accuracy of clustering.

[0042] Through the above technical solution, the correlation of various temperature variables of the power transformer is clustered through mutual information clustering, and highly correlated variables are clustered together to avoid the influence of irrelevant noise.

[0043] S3: extracting common patterns from each class of the clustering results, splicing and fusing the common patterns into various classes of sequences, and obtaining a multivariate time series with complete information; In this embodiment, if Figure 2 As shown in the figure, after clustering, the common patterns include significant features and overall features. The random pooling combined with the adaptive convolution pooling method is used to aggregate the features of each variable in the class. Specifically, random pooling extracts the overall features within the class from a global perspective, while adaptive convolution pooling performs convolution on the variables within the class to extract the significant features within the class. This method first uses adaptive convolution pooling to obtain the most significant features of all variables in each class, and at the same time, uses random pooling to obtain the overall features of all variables in each class. The calculation formula is as follows: (16) (17) in, Represents the overall characteristics within the class, Represents the salient features within a class, Indicates All variables of a class, represents random pooling, Represents adaptive convolution. Both random pooling and adaptive convolution pooling have adaptive characteristics and have stronger generalization capabilities than traditional average pooling and maximum pooling.

[0044] Subsequently, the two types of features, the significant features and the overall features of each class, are fused and spliced ​​to the sequence of each variable in the class. The features are then fused through a shared LSTM to obtain a multivariate time series with complete information. , Feature fusion refers to the fusion of each variable in a class with the fused significant features and overall features. That is, after each variable in a class is spliced ​​with the corresponding fused significant features and overall features, each variable is fused separately to incorporate the significant features and overall features of the class into the variable.

[0045] The modeling of LSTM (Long Short-Term Memory Network) is as follows: (18) (19) (20) (twenty one) (twenty two) (twenty three) in, represents the forget gate, represents the sigmoid activation function, , , , Represent the weight matrices of the forget gate, input gate, candidate memory unit and output gate respectively, , , Represent the bias items of the forget gate, input gate, and candidate memory unit respectively, Indicates the hidden state at the current moment, Represents the input at the current moment, represents the input gate, represents the candidate memory unit, Represents the memory unit at the current moment, represents the output gate, Represents a multiplication operation.

[0046] The purpose of this is to achieve effective interaction of common patterns between highly correlated variables within a class, while avoiding interactions between low-correlated variables that introduce irrelevant noise.

[0047] Through the above technical solution, a combination of random pooling and adaptive convolution pooling is used to extract common patterns from each class, and these common patterns are added to the original sequence of variables within the class through splicing and fusion to complete the missing or ambiguous information. The missing or ambiguous information here refers to the variables in the cluster. In S2, related temperature variables are clustered together. A variable may lack the information contained in other variables. By integrating the common patterns within the class into this variable, the information of other variables can be supplemented, that is, information completion.

[0048] S4: downsampling the information-completing multivariate time series to obtain multiple subsequences; extracting features of the information-completing multivariate time series through a dynamic time-frequency domain MLP mechanism to obtain overall features, extracting features of the subsequences through a dynamic time-frequency domain MLP mechanism to obtain local features, and fusing the overall features and local features to obtain time series features; In this embodiment, if Figure 3 As shown in FIG. 1 , the power transformer temperature data (original time series) as time series data has an inherent low-rank property. Therefore, the present invention is useful for multivariate time series with information complementation. Using uniform step size downsampling, multiple subsequences with time patterns very similar to the sequence before downsampling are obtained. ,in , Represents the number of intervals. Feature extraction for different subsequences can be performed to more deeply explore multivariate time series with complete information. In this regard, the present invention proposes a dynamic time-frequency domain MLP mechanism (TFMLP). Figure 4 As shown in the figure, the mechanism first uses wavelet transform to transfer the sequence to the frequency domain, and performs MLP (multi-layer perceptron) processing on the amplitude and phase in the frequency domain, and then returns to the time domain through inverse wavelet transform; further, sparse MLP is used in the time domain to capture time domain features. Sparse MLP means that only Top-K neurons are activated and the rest are set to zero. The specific steps of the dynamic time-frequency domain MLP mechanism are: (twenty four) (25) (26) in, represents the input sequence, represents the frequency domain MLP modeling results, represents continuous wavelet transform, represents the sequence converted back to the time domain, represents the inverse continuous wavelet transform, represents the sparse MLP, Represents the output features of the dynamic time-frequency domain MLP mechanism. Compared with conventional MLP, the dynamic time-frequency domain MLP mechanism (TFMLP) can not only process frequency domain features, but also reduce the complexity of traditional MLP through sparsification, thus achieving full modeling of time-frequency domain features.

[0049] Multivariate time series with complete information After a layer of TFMLP modeling, the overall features are obtained , the subsequence is modeled by a layer of TFMLP to obtain local features .Will After splicing according to the position during downsampling, Add and fuse to get the time series features The expression is: (27) (28) in, Represents the overall characteristics, Indicates A local feature.

[0050] S5: Obtain external environmental covariates and input them into the self-attention mechanism network to obtain the covariate's own characteristics, and fuse the covariate's own characteristics with the timing characteristics through the covariate-driven causal convolution fusion mechanism to obtain fused characteristics; apply the covariate's own characteristics as "convolution weights" to causal convolution, so that the convolution operation has dynamic adaptability in timing.

[0051] Specifically, the covariate's own characteristics generate convolution kernel parameters through a network at each time step, so that the "receptive field" of the convolution operation of the main variable at each time step can be adaptively adjusted according to the state of the covariate. This method not only retains the structural information of the time series features, but also makes the impact of the covariate on the main variable more flexible and dynamic.

[0052] In this embodiment, if Figure 5 As shown, the external environment covariate data is defined as , the self-attention mechanism is applied to the external environment covariate data to model the covariate's own characteristics . External environmental covariates include but are not limited to external factors that dynamically affect the heat dissipation efficiency, such as ambient temperature, air humidity, wind speed, sunshine intensity, and transformer load current. These covariates are collected in real time by sensors and aligned with the temperature variable time series to form multi-source feature inputs. Taking external covariates into full consideration will further improve the accuracy of the prediction. The self-attention mechanism modeling process is expressed as: (29) (30) (31) (32) in, represents the external environmental covariate, represents the query vector, , , represents the learnable weight parameters, represents the key vector, Represented as a vector of values, represents the dot product of the query vector and the key vector, Indicates the dimension of the key vector.

[0053] Then, the designed covariate-driven causal convolution fusion mechanism is used to fuse the covariate features into the time series features. First, the covariate feature data of the current time step is input into the fully connected neural network to obtain the dynamic convolution kernel: (33) in, Indicates The convolution kernel weights for time steps, represents a nonlinear activation function, The first time steps, Indicates time steps of learnable weight parameters, Indicates time steps.

[0054] The timing characteristics and the covariate characteristics themselves Align according to the time step to get the alignment feature . At each time step, Then, the obtained dynamic convolution kernel is used to perform causal convolution fusion on the time series features and covariate features contained in each time step, that is, the covariate information is fused into the time series features using causal convolution, which is expressed as: (34) in, represents the final fusion feature, represents causal convolution.

[0055] Through the above technical scheme, the uniform step size downsampling feature extraction method is used to model the time series characteristics of each temperature variable of the power transformer, and the external environment covariates are fused with the temperature variables through the covariate driven causal convolution fusion mechanism and self-attention mechanism.

[0056] S6: Input the fusion feature into the dynamic attention feature weighted prediction model for prediction to obtain a prediction result.

[0057] In this embodiment, compared with single-step prediction and traditional linear regression multi-step prediction, the present invention introduces a dynamic attention feature weighted prediction model (also called a multi-step prediction model), embeds a single-layer lightweight attention network into linear regression, realizes dynamic adjustment of feature level, and improves the prediction ability of complex scenes containing multiple temperature variables.

[0058] Specifically, first, according to the fusion results of each feature at the current time step , calculate the hidden layer features: (35) in, represents the hidden layer features, represents a nonlinear activation function, represents the weight matrix of the hidden layer, represents the bias term of the hidden layer, and Respectively The mean and variance of .

[0059] Next, generate attention-enhanced weights based on the hidden layer features: (36) in, Indicates time The attention weight, represents the weight matrix of the query part, represents the weight matrix of the key part, Represents the dimension of the input feature, used to scale the attention score; Represents the weight matrix of the output part of the attention mechanism, represents the bias term of the output part of the attention mechanism, Represents the nonlinear activation function for calculating the attention weights.

[0060] Finally, the weighted output is performed according to the attention enhancement weight: (37) in, is the prediction result of the power transformer temperature, represents the weight matrix of the weighted output, Represents the bias term for the weighted output.

[0061] In order to better demonstrate the effect of the method of the present invention, the present invention uses several common regression problem evaluation indicators to evaluate the results. These indicators include mean square error (MSE), mean absolute error (MAE), root mean square error (RMSE) and continuous ranking probability score (CRPS). MSE, MAE and RMSE are point estimation evaluation indicators for a single prediction result, while CRPS is used to evaluate the confidence probability estimate of multiple prediction results. The smaller the value of these indicators, the closer the prediction result is to the true value and the better the prediction effect.

[0062] (38) (39) (40) (41) in, Indicates the power transformer at time The true temperature value, The method of the present invention is The predicted temperature value, is the predicted length, Represents the probability distribution of temperature values ​​for multiple point predictions.

[0063] The method of the present invention was compared with three advanced methods: long short-term memory neural network (LSTM), temporal convolutional neural network (TCN), and multi-head attention mechanism network (Multi-Head Attention). The method of the present invention has the best effect among all the methods and the lowest error value of the evaluation index, which verifies the effectiveness of the proposed method.

[0064] Embodiment 2 like Figure 6 As shown, this embodiment discloses a power transformer temperature prediction system based on multi-feature extraction and fusion, including: A data acquisition module is configured to: acquire multivariate time series data consisting of temperature variables of the power transformer; A similarity clustering module is configured to: perform spatiotemporal decoupling standardization on the multivariate time series data, perform causal emergence clustering on each temperature variable in the standardized multivariate time series data, and obtain a clustering result; An intra-class common pattern extraction module is configured to: extract common patterns from each class of the clustering results, and splice and fuse the common patterns into various classes of sequences to obtain a multivariate time series with complete information; A time series feature modeling module is configured to: downsample the information-completing multivariate time series to obtain multiple subsequences; extract features of the information-completing multivariate time series through a dynamic time-frequency domain MLP mechanism to obtain overall features, extract features of the subsequences through a dynamic time-frequency domain MLP mechanism to obtain local features, and fuse the overall features and local features to obtain time series features; A covariate interaction module is configured to: obtain external environment covariates and input them into the self-attention mechanism network to obtain covariate features, and fuse the covariate features with time series features through a covariate-driven causal convolution fusion mechanism to obtain fused features; The multi-step prediction module is configured to: input the fusion feature into the dynamic attention feature weighted prediction model for prediction to obtain a prediction result.

[0065] Embodiment 3 The purpose of this embodiment is to provide a computing device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the method of embodiment 1 when executing the program.

[0066] Embodiment 4 The purpose of this embodiment is to provide a computer-readable storage medium, a computer-readable storage medium having a computer program stored thereon, and when the program is executed by a processor, the steps of the method of embodiment 1 are performed.

[0067] The steps involved in the apparatus of the above embodiments 3 and 4 correspond to the method embodiment 1, and the specific implementation method can refer to the relevant description part of embodiment 1. The term "computer-readable storage medium" should be understood as a single medium or multiple media including one or more instruction sets; it should also be understood to include any medium that can store, encode or carry an instruction set for execution by a processor and enable the processor to execute any method in the present invention.

[0068] Those skilled in the art should understand that the modules or steps of the present invention described above can be implemented by a general-purpose computer device, or alternatively, they can be implemented by a program code executable by a computing device, so that they can be stored in a storage device and executed by the computing device, or they can be made into individual integrated circuit modules, or multiple modules or steps therein can be made into a single integrated circuit module for implementation. The present invention is not limited to any specific combination of hardware and software.

[0069] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

[0070] Although the above describes the specific implementation mode of the present invention in conjunction with the accompanying drawings, it is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art on the basis of the technical solution of the present invention without creative work are still within the scope of protection of the present invention.

Claims

1. A power transformer temperature prediction method based on multi-feature extraction and fusion, characterized in that: include: Obtain multivariate time series data consisting of temperature variables for power transformers; Performing spatiotemporal decoupling standardization on the multivariate time series data, and performing causal emergence clustering on each temperature variable in the standardized multivariate time series data to obtain a clustering result; Extracting common patterns from each class of the clustering results, splicing and fusing the common patterns into various classes of sequences, and obtaining a multivariate time series with complete information; Downsampling the information-completed multivariate time series to obtain multiple subsequences; extracting features of the information-completed multivariate time series through a dynamic time-frequency domain MLP mechanism to obtain overall features, extracting features of the subsequences through a dynamic time-frequency domain MLP mechanism to obtain local features, and fusing the overall features and local features to obtain time series features; Obtain external environmental covariates and input them into the self-attention mechanism network to obtain the covariate's own features, and fuse the covariate's own features with the time series features through the covariate-driven causal convolution fusion mechanism to obtain fused features; The fused features are input into the dynamic attention feature weighted prediction model for prediction to obtain a prediction result.

2. The power transformer temperature prediction method based on multi-feature extraction and fusion according to claim 1 is characterized in that: The clustering results are as follows: Standardize multivariate time series data by temporal and spatial decoupling; Calculating the mutual information between each temperature variable in the standardized multivariate time series data, and converting the mutual information into a similarity matrix; A symmetric Plaus matrix is ​​obtained based on the similarity matrix, and eigenvalue decomposition is performed on the symmetric Plaus matrix to obtain a characteristic matrix; The causal emergence clustering algorithm is used to cluster the feature matrix to obtain a clustering result.

3. The power transformer temperature prediction method based on multi-feature extraction and fusion as claimed in claim 2 is characterized in that: The causal emergence clustering algorithm is used to cluster the feature matrix as follows: According to the feature matrix, the causal DAG structure between variables is learned by optimizing the objective function to construct a causal graph; Inputting the feature matrix and the causal graph into a graph neural network for feature extraction to obtain causal features; Calculating the cosine similarity of the causal features, and obtaining the causal similarity based on the cosine similarity; The causal similarity is used as a clustering criterion, and causal constraints are incorporated into the hierarchical clustering process to obtain a clustering result.

4. The power transformer temperature prediction method based on multi-feature extraction and fusion according to claim 1 is characterized in that: The multivariate time series with completed information is as follows: Adaptive convolutional pooling is used to obtain the significant features of all temperature variables within each category, and random pooling is used to obtain the overall features of all temperature variables within each category; The significant features and overall features of each class are fused separately, and then spliced ​​to each temperature variable in the corresponding class; The long short-term memory network is used to fuse the features of each temperature variable after splicing to obtain a multivariate time series with complete information.

5. The power transformer temperature prediction method based on multi-feature extraction and fusion according to claim 1 is characterized in that: The overall features and local features are extracted through the dynamic time-frequency domain MLP mechanism, specifically: in, represents the input sequence, represents the frequency domain MLP modeling results, represents continuous wavelet transform, represents the sequence converted back to the time domain, represents the inverse continuous wavelet transform, represents the sparse MLP, Represents the output features after the dynamic time-frequency domain MLP mechanism.

6. The power transformer temperature prediction method based on multi-feature extraction and fusion according to claim 1 is characterized in that: The fusion features are obtained specifically as follows: based on the characteristics of the covariate itself, the dynamic convolution kernel of each time step is obtained, the characteristics of the covariate itself and the time series characteristics are aligned according to the time step, and causal convolution fusion is performed according to the dynamic convolution kernel at each time step.

7. The power transformer temperature prediction method based on multi-feature extraction and fusion according to claim 1 is characterized in that: The dynamic attention feature weighted prediction model embeds a single-layer lightweight attention network into linear regression to achieve dynamic adjustment at the feature level.

8. The power transformer temperature prediction system based on multi-feature extraction and fusion is characterized by: include: A data acquisition module is configured to: acquire multivariate time series data consisting of temperature variables of the power transformer; A similarity clustering module is configured to: perform spatiotemporal decoupling standardization on the multivariate time series data, perform causal emergence clustering on each temperature variable in the standardized multivariate time series data, and obtain a clustering result; An intra-class common pattern extraction module is configured to: extract common patterns from each class of the clustering results, and splice and fuse the common patterns into various classes of sequences to obtain a multivariate time series with complete information; A time series feature modeling module is configured to: downsample the information-completing multivariate time series to obtain multiple subsequences; extract features of the information-completing multivariate time series through a dynamic time-frequency domain MLP mechanism to obtain overall features, extract features of the subsequences through a dynamic time-frequency domain MLP mechanism to obtain local features, and fuse the overall features and local features to obtain time series features; A covariate interaction module is configured to: obtain external environment covariates and input them into the self-attention mechanism network to obtain covariate features, and fuse the covariate features with time series features through a covariate-driven causal convolution fusion mechanism to obtain fused features; The multi-step prediction module is configured to: input the fusion feature into the dynamic attention feature weighted prediction model for prediction to obtain a prediction result.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps in the power transformer temperature prediction method based on multi-feature extraction and fusion as described in any one of claims 1 to 7 are implemented.

10. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the steps in the power transformer temperature prediction method based on multi-feature extraction and fusion as described in any one of claims 1 to 7 are implemented.

Images