Mutual inductor state prediction method based on big data analysis

Abstract

The invention proposes a mutual inductor state prediction method based on big data analysis, and relates to the field of mutual inductor state prediction, and the method specifically comprises the steps: constructing a mutual inductor state data set, carrying out the segmented representation of the tensorized time series data in the data set, and carrying out the segmentation of the tensorized time series data through sine transformation and a tanh function. The method comprises the following steps: generating a mutual inductor state feature vector by combining linear projection and a cross-dimension difference item, introducing a disturbance correction kernel and combining softmax weighted fusion to generate a cross-time feature vector, extracting a global dependency relationship and local fluctuation information to form disturbance feature representation, then calculating total disturbance energy, fusing with the disturbance feature representation to form a fusion feature vector, and finally, calculating the total disturbance energy according to the fusion feature vector. The method comprises the following steps: constructing an overall curvature representation, extracting long-term trend and local peak information by using average pooling and maximum pooling, splicing and fusing feature vectors to form multi-scale fusion features, splicing and flattening all segmented multi-scale fusion features, and then outputting a transformer state prediction value through linear transformation and Sigmoid, thereby improving the accuracy and robustness of transformer state prediction.

Description

Technical Field

[0001] This invention relates to the field of current transformer state prediction, and in particular to a current transformer state prediction method based on big data analysis. Background Technology

[0002] As a core measurement and protection device in power systems, the operating status of instrument transformers directly affects the accuracy of voltage and current signal acquisition and the reliability of system protection actions. Instrument transformer condition monitoring plays an important role in fault diagnosis, operation assessment, and equipment maintenance. With the expansion of power grid scale and the increase in the complexity of the operating environment, instrument transformers exhibit nonlinear and multi-scale characteristics under the influence of factors such as temperature changes, load fluctuations, and insulation aging. Traditional methods relying on manual inspection or single signal monitoring are insufficient to comprehensively reflect the health status of instrument transformers. There is an urgent need to introduce big data analysis-based methods to model and predict the operating status of instrument transformers in order to improve the accuracy of fault early warning and condition assessment.

[0003] Currently, current transformer condition prediction relies heavily on monitoring single physical quantities or static modeling methods based on statistical features. These methods typically only consider limited dimensions of information such as current, voltage, or ambient temperature and humidity, lacking a comprehensive integration of electrical, insulation, and environmental quantities. This makes it difficult to fully characterize the operating characteristics of current transformers under complex conditions. Furthermore, while some methods introduce time series analysis or traditional machine learning models, they generally suffer from insufficient feature extraction capabilities and poor sensitivity to cross-time dependencies, resulting in limited accuracy of prediction results and insufficient adaptability to sudden disturbances or periodic changes.

[0004] The operating state of instrument transformers is affected by multiple sources of data, including electrical quantities, insulation quantities, and environmental quantities. At different time scales, it includes both long-term evolution trends and transient anomalies caused by partial discharge pulses or current surges. Its cross-time dependence and local disturbances exhibit high complexity. Therefore, a prediction method that integrates segmented representation, cross-time disturbance modeling, and global-local fusion is proposed. This method can achieve efficient extraction and dynamic modeling of multi-source features, thereby enabling accurate prediction of the instrument transformer state under complex operating conditions. Summary of the Invention

[0005] This invention proposes a method for predicting the state of current transformers based on big data analysis. Addressing the issues of large amounts of state data, unstable cross-time dependencies, and frequent local disturbances in the long-term operation of current transformers, this invention proposes a multi-stage prediction mechanism that integrates piecewise representation, cross-time dependency modeling, and multi-scale curvature fusion. The mechanism includes: acquiring current transformer state data and constructing a current transformer state dataset; calculating piecewise representations using sine transform and tanh function, combined with linear projection and cross-dimensional difference terms to generate current transformer state feature vectors; introducing a disturbance correction kernel, transforming it through softmax to obtain attention weights, and weighting and fusing the main trend and periodic disturbance information to generate cross-time feature vectors; forming disturbance feature representations based on global dependencies and local fluctuation information, and calculating the total disturbance energy to obtain a fused feature vector; combining overall curvature representation with average pooling and max pooling to extract long-term trend and local peak information to form multi-scale fused features; and finally, flattening the data and outputting the predicted current transformer state value through linear transformation and sigmoid mapping. This achieves unified modeling and accurate prediction of the global evolution and local disturbance characteristics of the current transformer's operating state.

[0006] A method for predicting the state of an instrument transformer based on big data analysis is as follows: S1. Obtain the state data of the current transformer, preprocess it and construct the current transformer state dataset; S2. Divide the tensor time series data in the dataset into segments, calculate the segmented representation, generate feature representation values ​​using sine transform and tanh function, and generate mutual inductor state feature vectors by combining linear projection and cross-dimensional difference terms. S3. Introduce perturbation correction kernels between different segments, obtain attention weights through weighted aggregation and softmax transformation, extract main trend and periodic perturbation information based on the mutual inductor state feature vector, and generate cross-time feature vectors by weighted fusion using attention weights. S4. Extract the global dependency and local fluctuation information of the cross-time feature vectors, and after weighted fusion, obtain the perturbation feature representation by layer normalization and ReLU. After square and absolute value correction, obtain the total perturbation energy, and fuse it with the perturbation feature representation to form a fused feature vector. S5. Construct an overall curvature representation based on the fusion feature vectors of adjacent segments, and use averaging and max pooling to extract long-term trend and local peak information, which are then concatenated with the fusion feature vector of the current segment to form a multi-scale fusion feature. S6. Flatten all segmented multi-scale fusion features, and output the transformer state prediction value after linear transformation and Sigmoid mapping. S7. Construct a current transformer state prediction model based on big data analysis. Input the current transformer state dataset, use weighted binary cross-entropy as the loss function, and iterate through steps S2 to S6 until convergence to complete the optimization training of the model.

[0007] Preferably, the construction process of the instrument transformer state dataset includes three stages: data acquisition, preprocessing, and structuring. The acquired instrument transformer state data includes electrical quantity data, environmental quantity data, and insulation quantity data. The electrical quantity data consists of the current and voltage signals output from the secondary side of the instrument transformer; the environmental quantity data includes the surface temperature of the instrument transformer body and the temperature and humidity of the operating environment; and the insulation quantity data includes partial discharge pulses, leakage current, and dielectric loss factor. In the preprocessing stage, the acquired data of various types is first time-aligned. Using the sampling time of the voltage signal as a reference, the timestamps of the electrical quantity data, environmental quantity data, and insulation quantity data are corrected by interpolation. In the structuring stage, the data is segmented according to a fixed time series length. Within the time series length, the electrical quantity data, environmental quantity data, and insulation quantity data are stacked in channels to form tensor time series data, and finally, the instrument transformer state dataset is constructed.

[0008] Preferably, in step S2, the tensor time series data in the dataset is divided into segments, the segmented representation is calculated, the feature representation values ​​are generated using sine transform and tanh function, and the mutual inductor state feature vector is generated by combining linear projection and cross-dimensional difference terms. Each dimension of the tensor-quantized time series data is divided into sections of length [length missing]. The segmentation is performed by squaring each value within a segment, subtracting a term proportional to the absolute value of that value, and then normalizing the average to obtain a segmented representation for each dimension. For the Duan Di The feature representation values ​​of the first dimension are linearly projected, and then a cross-dimensional difference term is introduced to transform the first dimension's feature representation values ​​into a linear projection. The squared difference between the feature representations of dimension 1 and all other dimensions is added to the projection result to obtain the 1st dimension. Duan Di Dimensional fusion and cross-dimensional relationship representation; The first The cross-dimensional relationship representations of all dimensions of the segment are fused and concatenated to obtain the mutual inductor state feature vector.

[0009] Furthermore, addressing the issues of large numerical scale differences, frequent local impact interference, and difficulty in characterizing cross-dimensional coupling relationships in the time-series data of electrical quantities, environmental quantities, and insulation quantities generated during long-term operation of current transformers, this invention proposes an embedding mechanism combining piecewise representation construction and cross-dimensional difference term enhancement. First, the tensed time-series data is segmented according to a preset segment length. For each dimension of the sequence, a correction operation is performed within the segment by subtracting the absolute value of the proportional term from the square. The segmented representation is then obtained through normalized averaging, thereby weakening abnormal peak values ​​caused by extreme disturbances such as current surges, partial discharge pulses, and temperature and humidity fluctuations, ensuring the robustness and stability of the segmented representation. Then, for the first... Duan Di The piecewise representation of the dimensional features is linearly projected to a unified space, eliminating the impact of inconsistent dimensions and numerical distributions of different physical quantities on modeling accuracy. Based on this, a cross-dimensional difference term is introduced for the dimensional... The squared difference of the feature representations of dimension 1 and other dimensions is calculated and fused with the linear projection result to obtain the 1st dimension. Duan Di The fusion of cross-dimensional relationship representations preserves single-dimensional feature trends while strengthening multi-source coupling effects. Finally, the fusion of cross-dimensional relationship representations of all dimensions within the same segment is concatenated to generate the mutual inductor state feature vector, providing high-quality input for subsequent modeling.

[0010] Preferably, in step S3, a perturbation correction kernel between different segments is introduced, and attention weights are obtained through weighted aggregation and softmax transformation. The main trend and periodic perturbation information are extracted based on the mutual inductor state feature vector, and the attention weights are used for weighted fusion to generate a cross-time feature vector. For the Section and the The section in The dimensional fusion of cross-dimensional relationships is represented by calculating the squared difference between the two, introducing a penalty term proportional to the absolute value of the difference, and generating the 1st... Section and the The section in Dimensional perturbation correction kernel; The first Section and the The perturbation correction kernels of all dimensions of the segment are weighted and aggregated, and then averaged to obtain the first segment. Section and the The cross-time perturbation weight of the segment; Using softmax, the weights of the time-varying perturbations are transformed into the first... Section and the Attention weights for segments; Based on the state feature vector of the mutual inductor, the main trend information is extracted using a linear projection matrix, and the periodic perturbation information is obtained through a sine function. The main trend information and the periodic perturbation information are then fused using attention weights to obtain the [missing information]. Segment-time feature vector.

[0011] Furthermore, addressing the issues of time-dependent instability and frequent transient interference in the collected transformer state data, including electrical quantities, insulation quantities, and environmental quantities, this invention proposes a time-dependent disturbance modeling method based on a disturbance correction kernel and an attention mechanism. Firstly, for the... Section and the The section in The dimensional fusion of cross-dimensional relationships involves calculating the squared difference between the two dimensions and introducing a penalty term proportional to the absolute value of the difference to form a perturbation correction kernel. This kernel weakens the impact of abnormal peaks such as current surges, partial discharge pulses, and sudden temperature and humidity changes on the determination of cross-time differences. Then, the perturbation correction kernels of each dimension are weighted, aggregated, and averaged to obtain the cross-time perturbation weight, thus comprehensively characterizing the overall differences between electrical quantities, insulation quantities, and environmental quantities at different time periods. Next, softmax is used to transform the perturbation weight into attention weights, enabling the model to adaptively highlight time-series segments highly relevant to the current segment and suppress noise segments with low correlation to the actual operating state. Finally, based on the transformer state feature vector, a linear projection matrix is ​​used to extract the main trend information reflecting long-term evolution. Simultaneously, a sine function is combined to generate periodic perturbation information, and the two types of features are weighted and fused through attention weights to form the second... The segment-time feature vector mechanism not only maintains the continuity of the operating trend, but also accurately captures periodic fluctuations and sudden disturbances in electrical, insulation and environmental signals, thereby achieving robust modeling and accurate characterization of the time-dependent relationship of the transformer.

[0012] Preferably, in step S4, the global dependency and local fluctuation information of the cross-time feature vector are extracted, and after weighted fusion, the perturbation feature representation is obtained by layer normalization and ReLU. The total perturbation energy is obtained by square and absolute value correction, and then fused with the perturbation feature representation to form a fused feature vector. Using the weight matrix to assess the first Perform a linear transformation on the feature vectors across time segments to extract the global dependencies between data, and calculate the first... Segment cross-time feature vector and the first The absolute value of the difference between the means of the feature vectors across time segments is used to extract local fluctuation information. The local fluctuation information, weighted by the global dependency and the perturbation enhancement factor, is then added and processed through layer normalization and ReLU activation to obtain the first... The perturbation characteristics of the segment are represented; The first The disturbance characteristics of segment 1 represent the first Squaring the eigenvalues ​​of dimension 1, and subtracting the weighted eigenvalues ​​of the perturbation suppression factor of dimension 2. The absolute value of the eigenvalues ​​of the dimension is used to obtain the perturbation energy, which is then used to determine the eigenvalues ​​of the dimensional features. The perturbation characteristics of a segment represent the sum of the perturbation energies in all dimensions to obtain the first segment. Total disturbance energy of the segment; The first The total perturbation energy of a segment is multiplied by the perturbation feature representation, compressed using tanh nonlinearity, and finally multiplied element-wise with the original perturbation feature representation to output a fused feature vector that integrates global and local information.

[0013] Furthermore, considering that the state data of current transformers during operation contains both long-term global trends and is susceptible to local disturbances caused by factors such as current surges, partial discharges, and sudden changes in temperature and humidity, this invention proposes a global-local disturbance fusion mechanism. First, a weight matrix is ​​used to... A linear transformation is performed on the segment-time feature vector to extract the global dependency between data. Simultaneously, the absolute value of the difference between this vector and its mean is calculated to extract local fluctuation information reflecting electrical quantity abrupt changes and environmental fluctuations. This is then weighted by a perturbation enhancement factor and added to the global dependency, followed by layer normalization and ReLU activation to obtain the first... The perturbation characteristics of the segment are represented to balance global stability and local sensitivity. Subsequently, the perturbation characteristics are represented in the first segment. The eigenvalues ​​in each dimension are squared, and the absolute value term weighted by the perturbation suppression factor is subtracted to obtain the perturbation energy. The perturbation energies of all dimensions are then summed to obtain the eigenvalue of the eigenvalue. The total disturbance energy of the segment is used to highlight significant changes while suppressing anomalous amplification caused by random noise. Finally, the total disturbance energy is multiplied by the disturbance feature representation, and extreme energy values ​​are suppressed by tanh nonlinear compression. Then, it is fused element-wise with the original disturbance feature representation to obtain a fused feature vector that integrates global dependence and local disturbance. This ensures the ability to characterize the long-term state evolution trend of the transformer and can also effectively capture and constrain transient interference caused by current peaks, partial discharge pulses and environmental abrupt changes, thereby improving the robustness of the model and the accuracy of prediction.

[0014] Preferably, in step S5, an overall curvature representation is constructed based on the fusion feature vectors of adjacent segments, and long-term trend and local peak information are extracted using averaging and max pooling, which are then concatenated with the fusion feature vector of the current segment to form a multi-scale fusion feature. Calculate the first Section and the The squared L2 norm of the difference between the fused feature vectors of the segments is used to obtain the forward change, and then the th segment is further calculated. Section and the The squared L2 norm of the difference in the fused feature vectors of the segments is used to obtain the backward variation. Subtracting the backward variation from the forward variation yields the difference in the overall direction of change, thus constructing the first... , , The second-order difference vectors of the three segments are calculated by squared their second norms and then added to a constant 1 as the denominator to obtain the first... The overall curvature representation of the segment; In the Take the past based on the segment The overall curvature representation sequence of the length is subjected to average pooling and max pooling to obtain smooth long-term trend information and local peak information of extreme perturbations, respectively. The two types of pooling results are then compared with the first... The segment fusion feature vectors are concatenated to form the first Multi-scale fusion features of segments.

[0015] Furthermore, considering that the operating data of current transformers contains both a smooth long-term trend and severe disturbances caused by current surges, partial discharges, or sudden environmental changes in the time series, this invention proposes a multi-scale fusion mechanism based on overall curvature representation, which calculates the first-order fusion based on the fused feature vector of adjacent segments. Section and the The forward variation is obtained by squared the L2 norm of the difference between segments; then the forward variation is calculated. Section and the The squared L2 norm of the difference between the segments yields the backward variation; subtracting the two gives the difference in the overall direction of change. Simultaneously, the first... , , The second-order difference vectors of the three are used to calculate the square of their second norms, and then the constant 1 is added as the denominator to obtain the first... The overall curvature representation of the segment, this design can highlight the dynamic characteristics brought about by local acceleration or abrupt changes while maintaining the smoothness of the change. Subsequently, in the... Take the past based on the segment The overall curvature representation sequence of the length is subjected to average pooling and max pooling respectively to extract long-term trend information and local peak information under extreme perturbations. The results of the two types of pooling are then compared with the first... The feature vectors of the segments are concatenated to generate the first segment. The multi-scale fusion feature of the segment can capture the smooth aging trend in the operation of the transformer, and can also keenly sense short-term shocks and sudden events, so as to realize multi-scale unified modeling of global trends and local disturbances.

[0016] Preferably, in step S6, the multi-scale fusion features of all segments are flattened, and the predicted value of the transformer state is output after linear transformation and Sigmoid mapping. The multi-scale fused features of all segments are stacked sequentially to form a comprehensive feature matrix. The comprehensive feature matrix is ​​flattened, multiplied by the learnable weight matrix, and a bias is added to obtain a linear output result. Then, the Sigmoid function is used to obtain the predicted value of the transformer state.

[0017] Furthermore, addressing the issue that transformer state prediction requires a final discriminative output based on all segments, this invention proposes a prediction output mechanism based on linear mapping and Sigmoid compression. The multi-scale fused features of all segments are stacked sequentially to form a comprehensive feature matrix, which is then flattened to unify the feature representations of different segments in the time dimension. Subsequently, the flattened comprehensive feature vector is multiplied by a learnable weight matrix and a bias is added to obtain a linear output result. The linear result is then mapped to a normalized transformer state prediction value using the Sigmoid function. This compresses the complex segmented representation into an intuitive probability output, thereby achieving accurate prediction of the transformer state.

[0018] Preferably, in step S7, a current transformer state prediction model based on big data analysis is constructed. The current transformer state dataset is input, and the weighted binary cross-entropy is used as the loss function. The model is iteratively trained through steps S2 to S6 until convergence, thus completing the optimized training of the model.

[0019] Furthermore, the model is implemented using the Python programming language and the PyTorch deep learning framework. During training, an NVIDIA 3090 24GB GPU is used. The loss function adopts weighted binary cross-entropy to balance the imbalance between normal and abnormal samples. The optimizer uses Adam for parameter updates, and the learning rate is dynamically adjusted using an exponential decay strategy.

[0020] The technical effects and advantages provided by the present invention in the above technical solution are as follows: 1. This invention introduces piecewise representation into the tensor time series of the current transformer state dataset. It uses square and absolute value terms to modify and combine sine transform and tanh function to generate feature representation values. Then, it combines linear projection and cross-dimensional difference terms to generate current transformer state feature vectors. This effectively weakens the extreme peak interference caused by operating conditions such as current surge, partial discharge pulse and temperature and humidity change, and ensures the stability of piecewise feature expression. At the same time, it explicitly describes the coupling relationship between electrical quantity, insulation quantity and environmental quantity, thereby improving the robust modeling ability and cross-dimensional correlation expression ability of the current transformer multi-source features. 2. This invention constructs a disturbance correction kernel, weighted aggregation, and combines it with a softmax attention mechanism to obtain attention weights. Based on the current transformer state feature vector, it extracts the main trend information and periodic disturbance information respectively, and uses the attention weights to perform weighted fusion to generate a cross-time feature vector. This suppresses transient noise in the running data and adaptively highlights key time segments with high correlation to the current segment. This ensures the continuity of the state trend expression and accurately reflects the periodic disturbance characteristics such as current and voltage signal fluctuations and temperature and humidity changes, thereby enhancing the stability and accuracy of cross-time dependency modeling. 3. This invention utilizes linear transformation and the absolute value of the mean difference to construct global dependencies and local fluctuation information. After weighted fusion, layer normalization and ReLU are applied to obtain a disturbance feature representation. The total disturbance energy is further calculated by squaring and absolute value correction, and fused with the original disturbance feature representation to obtain a robust fused feature vector. Subsequently, an overall curvature representation is constructed based on adjacent segments, and long-term trend and local peak information are extracted by combining average pooling and max pooling. Finally, a multi-scale fused feature is formed, and the transformer state prediction value is output through linear mapping and Sigmoid function. This approach takes into account both global evolution trends and local disturbance features, effectively capturing the smooth degradation law in the aging process and the extreme peak problem of transformer state data, thereby significantly improving the state prediction accuracy and reliability of transformers under complex operating conditions. Attached Figure Description

[0021] Figure 1 This is a flowchart of a current transformer state prediction method based on big data analysis provided by the present invention.

[0022] Figure 2 This is a diagram of the structure for generating the state feature vector of a mutual inductor provided by the present invention.

[0023] Figure 3 This is a diagram of the cross-time feature vector generation structure provided by the present invention.

[0024] Figure 4 This is the multi-scale fusion feature generation structure diagram provided by the present invention.

[0025] Figure 5 This is a graph showing the model training loss curve provided by the present invention.

[0026] Figure 6 This is a prediction effect diagram of the current transformer state prediction model based on big data analysis provided by the present invention. Detailed Implementation

[0027] This invention proposes a method for predicting the state of current transformers based on big data analysis. Addressing the issues of large amounts of state data, instability due to cross-time dependencies, and frequent local disturbances during long-term operation, this invention proposes a prediction mechanism that integrates piecewise representation, cross-time feature modeling, and multi-scale curvature fusion. This includes constructing a current transformer state feature vector based on piecewise representation and cross-dimensional difference; generating a cross-time feature vector by combining a disturbance correction kernel and an attention mechanism; forming a disturbance feature representation by fusing global dependencies and local fluctuation information and introducing an overall curvature representation to generate multi-scale fused features; and finally outputting the predicted current transformer state value through linear transformation and Sigmoid mapping. This achieves unified modeling and accurate prediction of the global evolution and local disturbance characteristics of the current transformer's operating state.

[0028] S1. Obtain the state data of the current transformer, and construct the current transformer state dataset after preprocessing.

[0029] Please see Figure 1 As shown in the figure, a current transformer state prediction method based on big data analysis is described in this application embodiment. The specific steps are as follows.

[0030] The construction process of the instrument transformer status dataset includes three stages: data acquisition, preprocessing, and structuring. The instrument transformer status data includes electrical quantity data, environmental quantity data, and insulation quantity data. The electrical quantity data consists of the current and voltage signals output from the secondary side of the instrument transformer. The environmental quantity data includes the surface temperature of the instrument transformer body and the temperature and humidity of the operating environment. The insulation quantity data includes partial discharge pulses, leakage current, and dielectric loss factor. In the data acquisition stage, the sampling rate of the current and voltage signals is set to 5kHz to ensure complete recording of harmonics and dynamic components. The surface temperature of the instrument transformer body and the temperature and humidity of the operating environment are sampled with a 1-minute period to reflect the gradual change process. Partial discharge... The pulse signal sampling rate is set to 100kHz to capture transient impact characteristics. Leakage current and dielectric loss factor are acquired hourly as key references for long-term insulation degradation. In the preprocessing stage, the collected data of various types are first time-aligned. Based on the sampling time of the voltage signal, the timestamps of electrical quantity data, environmental quantity data, and insulation quantity data are corrected by interpolation. In the structuring stage, the data is divided according to a fixed time series length, which is set to 60 minutes. Within the time series length, electrical quantity, environmental quantity, and insulation quantity data are stacked in channels to form tensor time series data, and finally, the current transformer state dataset is constructed.

[0031] S2. Divide the tensor time series data in the dataset into segments, calculate the segmented representation, generate feature representation values ​​using sine transform and tanh function, and generate mutual inductor state feature vectors by combining linear projection and cross-dimensional difference terms.

[0032] Furthermore, in step S2, the transformer state feature vector is generated, and the process is as follows: Figure 2 As shown, the specific steps include: S21, the tensor-quantized time series data is denoted as , ,in The length of the time series. The number of data types for the current transformer status. For the first The data collected at each time step, Within each dimension, the sequence is divided into segments of length [length missing]. The segments were obtained in total. The data is divided into segments. Each value within a segment is squared, and a term proportional to the absolute value of that value is subtracted. The result is then normalized and averaged to reduce the impact of extreme anomalies caused by current surges, partial discharge pulses, and sudden temperature and humidity changes, thus obtaining a robust segmented representation. The specific mathematical model is as follows: ; in, For the first Duan Di Piecewise representation of dimensions The segment length, For the first Time step The numerical value of dimension, For learnable dimension-dependent perturbation factors, This is for taking the absolute value; In this embodiment, Set to 60. Set to 5, and get The system is divided into segments to ensure stable segmentation while highlighting local variations. Initialized to 0.5 to suppress extreme outliers caused by transformer current surges and temperature / humidity fluctuations during large-scale data acquisition. The value is set to 8, which represents the number of data types for the current transformer status data. S22. Segmentation representation With sine transformation The summation is used to simulate the dynamic disturbance of the transformer state signal under periodic operating conditions. This disturbance is mapped to a finite interval using tanh. The specific mathematical model is as follows: ; in, For the first Duan Di Feature representation values ​​of dimension; S23, Feature Representation Perform a linear projection, then introduce a cross-dimensional difference term, and then... The squared difference between the feature representation values ​​of dimension 1 and all other dimensions is added to the projection result to explicitly encode the coupling relationship between the state data of different current inductors, resulting in a representation after fusing cross-dimensional relationships. The specific mathematical model is as follows: ; in, For the first Duan Di Dimensional fusion and cross-dimensional relationship representation. Specific projection weights for the learnable dimensions. For the learnable first Dimension and First Inter-dimensional correlation coefficients For the first Duan Di Feature representation values ​​of dimension; S24, the first The cross-dimensional relation representations of all segments are fused and concatenated to obtain the mutual inductor state feature vector. .

[0033] S3. Introduce perturbation correction kernels between different segments, obtain attention weights through weighted aggregation and softmax transformation, extract main trend and periodic perturbation information based on the mutual inductor state feature vector, and generate cross-time feature vectors by weighted fusion using attention weights.

[0034] Furthermore, in step S3, feature vectors across time are generated, and the process is as follows: Figure 3 As shown, the specific steps include: S31, Regarding the first Section and the Segments in the same dimension The fusion of cross-dimensional relationships is represented by calculating the squared difference between the two data points. A penalty term proportional to the absolute value of the difference is introduced to weaken the transient peak effects caused by current surges, temperature and humidity abrupt changes, etc., in the transformer state data, thereby generating a disturbance correction kernel. The specific mathematical model is as follows: ; in, For the first Section and the The section in Dimensional perturbation correction kernel, For the first Duan Di Dimensional fusion and cross-dimensional relationship representation. For learnable dimension-dependent perturbation factors; In this embodiment, Initialize it to 0.3 to reduce the interference of sudden noise, while retaining the difference signal of the actual operating condition change of the current transformer; S32, the first Section and the The perturbation correction kernels of all dimensions of the segment are weighted and aggregated, and then averaged to obtain the cross-time perturbation weight, so that the state data of different transformers contribute differently in the cross-time relationship. The specific mathematical model is as follows: ; in, For the first Section and the The cross-time perturbation weight of the segment, These are the learnable dimension weights; S33. Using softmax, the cross-time perturbation weights are transformed into attention weights to measure the temporal dependency between segments. The specific mathematical model is as follows: ; in, For the first Section and the The attention weight of a segment indicates that the two segments have a higher cross-time correlation in the transformer state data. For the first Section and the The cross-time perturbation weight of the segment; S34. Based on the state feature vector of the mutual inductor, the main trend information is extracted using the linear projection matrix, and the periodic perturbation information is obtained through the sine function, utilizing attention weights. By weighted fusion of main trend information and cyclical disturbance information, a cross-time feature vector is obtained. The specific mathematical model is as follows: ; in, For the first Segment-time feature vector, , For learnable projection matrices, For the first State feature vector of a segment mutual inductor.

[0035] S4. Extract the global dependency and local fluctuation information of the cross-time feature vectors, and after weighted fusion, obtain the perturbation feature representation through layer normalization and ReLU. After square and absolute value correction, obtain the total perturbation energy, and fuse it with the perturbation feature representation to form a fused feature vector.

[0036] Furthermore, in step S4, the fused feature vector is generated, which specifically includes the following steps: S41. Using the weight matrix to analyze cross-time eigenvectors Perform linear transformations to extract global dependencies between data and calculate... and The absolute value of the difference between means is used to extract local fluctuation information, which is used to capture transient disturbances such as current surges, partial discharge pulses, and sudden changes in temperature and humidity, and to combine global dependencies with disturbance enhancement factors. The weighted local fluctuation information is summed, and the perturbation feature representation is obtained through layer normalization and ReLU activation. The specific mathematical model is as follows: ; in, For the first The perturbation characteristics of the segment are represented. The weight matrix is ​​a learnable matrix. As a learnable perturbation enhancement factor, for The mean calculated from all dimensions For layer normalization, This is for taking the absolute value; In this embodiment, Initialize it to 1 to keep local fluctuation information and global dependencies on the same order of magnitude, so that the model can maintain a stable convergence speed at the beginning; S42, will No. The eigenvalues ​​of the dimension are squared to highlight large changes, and the weighted eigenvalues ​​of the perturbation suppression factor are subtracted. The absolute value of the dimensional eigenvalues ​​is used to obtain the perturbation energy, which is used to suppress excessive amplification caused by environmental noise or abnormal peaks. The total perturbation energy is obtained by summing the perturbation energies of all dimensions. The specific mathematical model is as follows: ; in, For the first The total disturbance energy of the segment, For the first Duan Di Dimensional perturbation eigenvalues It is a learnable perturbation suppression factor; In this embodiment, Initialized to 0.2 to adjust the effect of the absolute value term; S43. Multiply the total perturbation energy by the perturbation feature representation, and use tanh nonlinear compression to avoid abnormal energy dominating the feature distribution. Finally, multiply the total perturbation energy by the original perturbation feature representation element by element to output a fused feature vector that integrates global and local information. The specific mathematical model is as follows: ; in, For the first Segment fusion feature vector, This is for element-wise multiplication.

[0037] S5. Construct an overall curvature representation based on the fusion feature vectors of adjacent segments, and use averaging and max pooling to extract long-term trend and local peak information, which are then concatenated with the fusion feature vector of the current segment to form a multi-scale fusion feature.

[0038] Furthermore, in step S5, the process is as follows: Figure 4 As shown, calculate the first... Section and the The squared L2 norm of the difference between the fused feature vectors of the segments is used to obtain the forward change, and then the th segment is further calculated. Section and the The squared L2 norm of the difference in the fused feature vectors of the segments is used to obtain the backward variation. Subtracting the backward variation from the forward variation yields the difference in the overall direction of change, thus constructing the first... , , The second-order difference vectors of the three segments are used to calculate their second-norm squares. A constant 1 is added to the denominator to constrain the numerical range, resulting in the overall curvature representation. The specific mathematical model is as follows: ; in, For the first The overall curvature representation of the segment. , The first and Segment fusion feature vector, It is a 2-norm; In the Take the past based on the segment The overall curvature representation sequence of length ,in For the first The overall curvature of the segment is characterized, and average pooling and max pooling are performed to obtain smooth long-term trend information and local peak information of extreme perturbations, respectively. The two types of pooling results are then compared with the first... The segment fusion feature vectors are concatenated to form a multi-scale fusion feature. The specific mathematical model is as follows: ; in, For the first Multi-scale fusion features of segments This is the stitching operation for the last dimension. For average pooling operation, This is a max pooling operation.

[0039] S6. Flatten all segmented multi-scale fusion features, and output the transformer state prediction value after linear transformation and Sigmoid mapping.

[0040] Furthermore, in step S6, the multi-scale fused features of all segments are stacked sequentially to form a comprehensive feature matrix. , The comprehensive feature matrix The output is flattened, multiplied by the learnable weight matrix, and a bias is added to obtain a linear output. Then, the output is mapped to a normalized probability distribution using the Sigmoid function. The specific mathematical model is as follows: ; in, This represents the predicted state value of the current transformer. The higher the predicted value, the higher the risk of the current transformer malfunctioning. Let be a learnable weight matrix, where Let be the dimension of the weight matrix. , For bias, For the Sigmoid function, For flattening operation.

[0041] To address the characteristics of current transformers during long-term operation, such as large amounts of state data, unstable cross-time dependencies, and frequent local disturbances, this invention constructs a current transformer state prediction model based on big data analysis. The model inputs a current transformer state dataset, sequentially performs tensor quantization of time series data to segment and generate current transformer state feature vectors, introduces a disturbance correction kernel and attention mechanism to generate cross-time feature vectors, fuses global dependencies and local fluctuation information, calculates the total disturbance energy, and combines overall curvature representation and multi-scale pooling to generate multi-scale fused features. Finally, the predicted current transformer state value is output through linear transformation and sigmoid mapping. During training, a weighted binary cross-entropy loss function is used to update the model parameters end-to-end. Convergence is achieved by constraining the difference between the predicted results and the actual state labels, improving the model's stable response capability under extreme conditions such as current surges, partial discharge pulses, and sudden temperature and humidity changes, ultimately forming a current transformer state prediction model with high state recognition accuracy.

[0042] Furthermore, in step S7, the current transformer state prediction model proposed in this invention, based on big data analysis, is implemented using the Python programming language and the PyTorch deep learning framework. During training, an NVIDIA 3090 24GB GPU is used. The loss function is weighted binary cross-entropy, used to balance the imbalance between normal and abnormal samples. The optimizer is Adam for parameter updates, and the initial learning rate is set to... The batch size is set to 12 and dynamically adjusted using an exponential decay strategy.

[0043] Furthermore, in step S7, the transformer state dataset is input into the constructed transformer state prediction model based on big data analysis for processing. The training loss curve of the model is shown below. Figure 5 As shown in the figure, the loss value gradually decreases and eventually stabilizes with the increase of training epochs, indicating that the model can be effectively optimized during training; the prediction effect is as follows. Figure 6 As shown in the figure, the horizontal axis represents time in minutes, the vertical axis represents the transformer state value, the gray dashed line represents the actual evaluation value, and the black solid line represents the model prediction value. It can be seen from the figure that the predicted value and the actual evaluation value maintain a high degree of consistency in the overall trend, with small local fluctuation errors, which fully verifies the accuracy and reliability of the method in the transformer state prediction task.

[0044] The above are merely preferred embodiments of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the inventive concept of the present invention, and these all fall within the protection scope of the present invention.

Claims

1. A method for predicting the state of a current transformer based on big data analysis, characterized in that, Includes the following steps: S1. Obtain the state data of the current transformer, preprocess it and construct the current transformer state dataset; S2. Divide the tensor time series data in the dataset into segments, calculate the segmented representation, generate feature representation values ​​using sine transform and tanh function, and generate mutual inductor state feature vectors by combining linear projection and cross-dimensional difference terms. S3. Introduce perturbation correction kernels between different segments, obtain attention weights through weighted aggregation and softmax transformation, extract main trend and periodic perturbation information based on the mutual inductor state feature vector, and generate cross-time feature vectors by weighted fusion using attention weights. S4. Extract the global dependency and local fluctuation information of the cross-time feature vectors, and after weighted fusion, obtain the perturbation feature representation by layer normalization and ReLU. After square and absolute value correction, obtain the total perturbation energy, and fuse it with the perturbation feature representation to form a fused feature vector. S5. Construct an overall curvature representation based on the fusion feature vectors of adjacent segments, and use averaging and max pooling to extract long-term trend and local peak information, which are then concatenated with the fusion feature vector of the current segment to form a multi-scale fusion feature. S6. Flatten all segmented multi-scale fusion features, and output the transformer state prediction value after linear transformation and Sigmoid mapping. S7. Construct a current transformer state prediction model based on big data analysis. Input the current transformer state dataset, use weighted binary cross-entropy as the loss function, and iterate through steps S2 to S6 until convergence to complete the optimization training of the model.

2. The method for predicting the state of a current transformer based on big data analysis according to claim 1, characterized in that, The instrument transformer status data is collected, including electrical quantity data, environmental quantity data, and insulation quantity data. The electrical quantity data consists of the current and voltage signals output from the secondary side of the instrument transformer. The environmental quantity data includes the surface temperature of the instrument transformer body and the temperature and humidity of the operating environment. The insulation quantity data includes partial discharge pulses, leakage current, and dielectric loss factor. After preprocessing and structuring, tensor time series data are formed to construct the instrument transformer status dataset.

3. The method for predicting the state of a current transformer based on big data analysis according to claim 2, characterized in that, Each dimension of the tensor-quantized time series data is divided into sections of length [length missing]. The segmentation is performed by squaring each value within a segment, subtracting a term proportional to the absolute value of that value, and then normalizing the average to obtain a segmented representation for each dimension. For the first Duan Di The feature representation values ​​of the first dimension are linearly projected, and then a cross-dimensional difference term is introduced to transform the first dimension's feature representation values ​​into a linear projection. The squared difference between the feature representations of dimension 1 and all other dimensions is added to the projection result to obtain the 1st dimension. Duan Di Dimensional fusion and cross-dimensional relationship representation; The first The cross-dimensional relationship representations of all dimensions of the segment are fused and concatenated to obtain the mutual inductor state feature vector.

4. The method for predicting the state of a current transformer based on big data analysis according to claim 3, characterized in that, For the Section and the The section in The dimensional fusion of cross-dimensional relationships is represented by calculating the squared difference between the two, introducing a penalty term proportional to the absolute value of the difference, and generating the 1st... Section and the The section in Dimensional perturbation correction kernel; The first Section and the The perturbation correction kernels of all dimensions of the segment are weighted and aggregated, and then averaged to obtain the first segment. Section and the The cross-time perturbation weight of the segment; Using softmax, the weights of the time-varying perturbations are transformed into the first... Section and the Attention weights for segments; Based on the state feature vector of the mutual inductor, the main trend information is extracted using a linear projection matrix, and the periodic perturbation information is obtained through a sine function. The main trend information and the periodic perturbation information are then fused using attention weights to obtain the [missing information]. Segment-time feature vector.

5. The method for predicting the state of a current transformer based on big data analysis according to claim 4, characterized in that, Using the weight matrix to assess the first Perform a linear transformation on the feature vectors across time segments to extract the global dependencies between data, and calculate the first... Segment cross-time feature vector and the first The absolute value of the difference between the means of the feature vectors across time segments is used to extract local fluctuation information. The local fluctuation information, weighted by the global dependency and the perturbation enhancement factor, is then added and processed through layer normalization and ReLU activation to obtain the first... The perturbation characteristics of the segment are represented; The first The disturbance characteristics of segment 1 represent the first Squaring the eigenvalues ​​of dimension 1, and subtracting the weighted eigenvalues ​​of the perturbation suppression factor of dimension 2. The absolute value of the eigenvalues ​​of the dimension is used to obtain the perturbation energy, which is then used to determine the eigenvalues ​​of the dimensional features. The perturbation characteristics of a segment represent the sum of the perturbation energies in all dimensions to obtain the first segment. Total disturbance energy of the segment; The first The total perturbation energy of a segment is multiplied by the perturbation feature representation, compressed using tanh nonlinearity, and finally multiplied element-wise with the original perturbation feature representation to output a fused feature vector that integrates global and local information.

6. The method for predicting the state of a current transformer based on big data analysis according to claim 5, characterized in that, Calculate the first Section and the The squared L2 norm of the difference between the fused feature vectors of the segments is used to obtain the forward change, and then the th segment is further calculated. Section and the The squared L2 norm of the difference in the fused feature vectors of the segments is used to obtain the backward variation. Subtracting the backward variation from the forward variation yields the difference in the overall direction of change, thus constructing the first... , , The second-order difference vectors of the three segments are calculated by squared their second norms and then added to a constant 1 as the denominator to obtain the first... The overall curvature representation of the segment; In the Take the past based on the segment The overall curvature representation sequence of the length is subjected to average pooling and max pooling to obtain smooth long-term trend information and local peak information of extreme perturbations, respectively. The two types of pooling results are then compared with the first... The segment fusion feature vectors are concatenated to form the first Multi-scale fusion features of segments.

7. The method for predicting the state of a current transformer based on big data analysis according to claim 6, characterized in that, The multi-scale fused features of all segments are stacked sequentially to form a comprehensive feature matrix. The comprehensive feature matrix is ​​flattened, multiplied by the learnable weight matrix, and a bias is added to obtain a linear output result. Then, the Sigmoid function is used to obtain the predicted value of the mutual inductor state.

Images