District new energy consumption capability big data analysis system and method

Abstract

The invention discloses a transformer area new energy consumption capability big data analysis system and method, and belongs to the technical field of new energy data analysis. The method is used for solving the technical problems that new energy power generation and absorption big data analysis in an existing scheme can only cover key nodes, and fine monitoring cannot be carried out on the interior of an array. Dual targets of global coverage and local refinement are realized through dynamic grid collaborative adaptive interpolation, data density and neighborhood adaptive adjustment; through coupling of space distance attenuation and time dynamic trend, cross validation and parameter feedback, it is ensured that the precision of a situation field is controllable; a health degree thermodynamic diagram is obtained through modular processing and combination, visual presentation of health degree assessment from a local unit to a global array is achieved, the limitation of data islands and privacy leakage of a traditional centralized model can be broken through, and big data analysis of global coverage, unit-level fineness and space-time coupling in a new energy power generation array is achieved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of new energy data analysis, in particular to a new energy consumption capacity big data analysis system and method for a transformer area. BACKGROUND

[0002] The new energy consumption capacity big data analysis for a transformer area is a scientific evaluation and prediction analysis process for the capacity of a distribution network transformer area to accept new energy in a power system by using big data technology. The transformer area usually refers to a power supply area served by a distribution transformer, which is the terminal link directly connected to the user. With the large access of roof photovoltaic and other distributed energy, how to ensure that the transformer area power grid safely, stably and efficiently digests these new energy power becomes a key problem. The traditional analysis method is usually static and has low precision, while the big data analysis can provide a more dynamic and accurate solution.

[0003] The existing technical solution can only cover the key nodes when implemented, and cannot perform fine monitoring on the array inside the new energy power generation, which has a blind area of big data analysis. SUMMARY

[0004] The purpose of the present application is to provide a new energy consumption capacity big data analysis system and method for a transformer area, which solves the technical problem that the new energy power generation and consumption big data analysis of the existing solution can only cover the key nodes and cannot perform fine monitoring on the array inside.

[0005] The technical solution of the present application is: a new energy consumption capacity big data analysis system for a transformer area, comprising: A multi-source data acquisition and processing module is used to deploy a distributed optical fiber sensing array, acquire real-time operation parameters and environmental micro-disturbance data of each unit inside the new energy power generation array, and generate a high-density situation field matrix. A data set gridding reconstruction module is used to perform gridding reconstruction on the high-density situation field matrix based on an adaptive space-time interpolation algorithm to obtain a fine situation field covering the entire array. A space-time coupling processing module is used to extract micro-disturbance features and macro-trend features of the fine situation field by multi-scale feature decomposition, and construct a space-time coupling feature matrix. An internal health perception analysis module is used to input the space-time coupling feature matrix into a federated learning model for distributed training, output real-time health degree evaluation results of each unit inside the array, and display the health state differences of each region of the new energy power generation array.

[0006] Preferably, the data set gridding reconstruction module comprises: performing space-time weight calculation, jointly calculating and normalizing the calculated spatial weight and time weight to obtain a joint weight, and performing grid node value interpolation calculation using the joint weight to obtain the situation value of the target grid node.

[0007] Preferably, the situation values ​​of all target grid nodes are arranged according to spatial coordinates to form a refined situation field of a three-dimensional tensor.

[0008] Preferably, the refined situation field is preprocessed by spatiotemporal standardization, and a standardized and denoised spatiotemporal matrix is ​​output; when performing multi-scale feature decomposition, VMD decomposition is performed for the time series of each spatial grid of the spatiotemporal matrix.

[0009] Preferably, the center frequency of each modal component is calculated. If the center frequency is greater than the frequency threshold, the corresponding component is marked as the micro-perturbation mode set, and the remaining components are marked as the macro-trend mode set.

[0010] Preferably, micro-perturbation feature extraction and macro-trend feature extraction are performed separately. The micro-feature vectors and macro-feature vectors extracted from each spatial grid are concatenated according to their dimensions to obtain a spatiotemporal feature vector. The spatiotemporal feature vectors of all spatial grids are arranged into a four-dimensional tensor according to spatial coordinates and time windows, and then reduced to a three-dimensional matrix through tensor decomposition to obtain a spatiotemporal coupling feature matrix.

[0011] Preferably, based on the physical topology of the new energy power generation array, an improved K-means algorithm is used to divide it into K edge computing nodes; The central server receives the local gradients and local validation set accuracy uploaded by each edge computing node, and calculates the node contribution weights.

[0012] Preferably, the central server aggregates gradients weighted by contribution, uses the aggregated gradients to update the global model parameters, and distributes the updated global model to each edge computing node for iterative training until the loss function converges.

[0013] Preferably, each edge computing node loads a global model, infers the real-time spatiotemporal coupling characteristics, outputs the unit health status, and generates a health status heatmap by grid unit, which intuitively displays the differences in health status of each region of the new energy power generation array through color depth.

[0014] Preferably, the expression related to the spatiotemporal coupling feature matrix is: Where core is the core tensor; The projection matrix is ​​the spatial dimension. This is the projection matrix along the time dimension; This refers to the multiplication operation between tensors and matrices.

[0015] A big data analysis method for the renewable energy absorption capacity of a power distribution area includes: Deploy a distributed fiber optic sensor array to collect real-time operating parameters and environmental micro-disturbance data of each unit inside the new energy power generation array, and generate a high-density situation field matrix. The high-density situation field matrix is ​​reconstructed by a gridded method based on an adaptive spatiotemporal interpolation algorithm to obtain a refined situation field covering the entire array domain. By extracting the micro-perturbation features and macro-trend features of the refined situation field through multi-scale feature decomposition, a spatiotemporal coupled feature matrix is ​​constructed. The spatiotemporal coupling feature matrix is ​​input into the federated learning model for distributed training, and the real-time health assessment results of each unit within the array are output, and the differences in health status of each region of the new energy power generation array are displayed.

[0016] The present invention has the following beneficial effects: This invention breaks through the traditional uniform grid plus fixed interpolation mode by using dynamic grid collaborative adaptive interpolation. It achieves the dual goals of full coverage and local fine-grained interpolation by adaptively adjusting data density and neighborhood. By coupling spatial distance attenuation with temporal dynamic trends, it can solve the static distortion problem caused by single-dimensional interpolation. Through cross-validation and parameter feedback, it ensures that the accuracy of the situation field is controllable and avoids the unreliability of black-box interpolation.

[0017] This invention compresses high-dimensional features into a computable coupled matrix by fusing microscopic and macroscopic feature vectors and performing spatiotemporal dimension expansion and tensor dimensionality reduction, which can provide input for subsequent federated learning models. By concatenating microscopic and macroscopic features and performing tensor decomposition, the data dimensionality is reduced while preserving spatiotemporal correlation information, thus solving the curse of dimensionality problem in high-resolution situation fields. The constructed spatiotemporal coupled feature matrix contains both local perturbation details and global trend features, providing multi-scale, coupled feature representation for refined situation assessment within the array.

[0018] This invention achieves a visualized presentation of health assessment from local units to the entire array by modularly processing and combining health heat maps. It breaks through the limitations of traditional centralized models, such as data silos and privacy leaks, and realizes big data analysis with full coverage, unit-level fine detail, and spatiotemporal coupling within the new energy power generation array, providing key technical support for early fault warning and operation and maintenance optimization. Attached Figure Description

[0019] Figure 1 This is a block diagram of a big data analysis system for the renewable energy absorption capacity of a transformer substation, according to the present invention. Figure 2 This is a flowchart of a big data analysis method for the renewable energy absorption capacity of a transformer substation according to the present invention. Detailed Implementation

[0020] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] like Figure 1 As shown, this invention is a big data analysis system for the renewable energy consumption capacity of a transformer substation, comprising: A multi-source data acquisition and processing module is used to deploy a distributed fiber optic sensor array to collect real-time operating parameters and environmental micro-disturbance data of each unit within the new energy power generation array, generating a high-density situation field matrix; the specific steps include: A distributed fiber optic grating sensor is selected as the core sensing element, integrating three sensing modes: vibration, strain, and temperature. Embedding micro IoT modules at fiber optic nodes enables communication with edge computing units via the LoRaWAN protocol, with data transmission latency controlled within 50ms. The fiber optic sensor is fixed to the surface of the power generation unit, such as the back panel of a photovoltaic panel or the inner wall of a wind turbine tower, using an embedded installation method with a bending radius of ≥50mm to avoid signal attenuation; Optical time domain reflectometers are used to detect the continuity of fiber optic links to ensure that the optical power attenuation at each sensing point is ≤0.2dB / km; The sensor was calibrated using a standard resistance box and a constant temperature bath, generating a calibration coefficient matrix K: ;in, These are vibration sensitivity, strain coefficient, and temperature coefficient, with specific values ​​of 0.98 mV / g, 1.2 pm / με, and 10 pm / ℃, respectively. Based on the deployed fiber optic sensor array, during multi-parameter synchronous acquisition and data preprocessing, operational data and environmental micro-disturbance data are collected. Among them, the operating parameters are: the real-time power P of each unit is collected through the current transformer and voltage sensor built into the fiber optic sensor; Environmental micro-disturbance data: Vibration acceleration a, strain ε, and temperature T are collected through integrated vibration and temperature modules. The units for different data are g, με, and ℃, respectively. Timestamp alignment algorithms, such as cross-correlation function method, are used to bind multi-source data to a unified time axis to form original data tuples; The original data tuples are preprocessed, including wavelet threshold denoising of vibration signals and moving average filtering of temperature and strain signals. This is a conventional technical solution, and the specific implementation steps will not be elaborated here. In addition, the original data tuples are modified, involving the following expressions: ; in, These are the vibration acceleration, strain, and temperature before correction; These are the corrected vibration acceleration, strain, and temperature, respectively. The preprocessed data is reconstructed using a grid, with 10cm×10cm as the grid unit. The corrected data is then interpolated to the three-dimensional grid nodes to generate a high-density situation field matrix. Where m×n is the number of grids and l is the length of the time series.

[0022] It should be noted that the high-density situation field matrix output in this step will serve as the input for subsequent multi-scale eigenvalue decomposition, and its grid resolution directly determines the level of detail in big data analysis. Furthermore, by suppressing noise and reconstructing the grid, discrete sensor data is transformed into a continuous global situation field, which can solve the blind spot problem in traditional key node monitoring; a 10cm-level grid resolution can be accurate to the local area of ​​a single photovoltaic cell or wind turbine blade, providing support for capturing microscopic disturbances within the array.

[0023] The dataset gridding reconstruction module is used to reconstruct the high-density situation field matrix using an adaptive spatiotemporal interpolation algorithm, obtaining a refined situation field covering the entire array domain; the specific steps include: The high-density situation field matrix is ​​time-stamped and outlier detected. Outliers are then repaired using spatiotemporal weighted interpolation to obtain a repaired data matrix. Timestamp synchronization and outlier detection are both existing conventional technical solutions, and the specific implementation steps will not be elaborated here. When performing physical space meshing, a uniform mesh is initialized based on the physical dimensions of the new energy power generation array. Grid cell size The coordinates of the entire region are [0,L]×[0,W]; the physical dimensions of the new energy power generation array include length L and width W. Calculate the sensor node density within the initial grid. , The number of sensor nodes within grid g; The meshes corresponding to sensor node densities exceeding a density threshold are marked as target meshes, and the target meshes are then refined using a bisection method; where the sub-mesh size... k represents the encryption level, with a maximum value of 3, meaning the minimum grid size is 0.125m × 0.125m; Output encrypted dynamic mesh set , where each target grid Including coordinate range and encryption layers ; q represents different target grids, and Q represents the total number of target grids; It should be noted that the dynamic grid set provides a spatial framework for subsequent adaptive interpolation, and the grid refinement result directly determines the resolution of the interpolation calculation. By using data density-driven dynamic refinement, a larger grid is maintained in sparse data regions, while a finer grid is generated in dense data regions, achieving well-balanced full-domain coverage. This can solve the problems of excessive redundancy or undersampling in key areas of traditional uniform grids.

[0024] For the target grid center coordinates In repairing the data matrix Searching for spatiotemporal neighborhood ;in, The neighborhood radius, , For grid size, This is the space expansion factor, with a default value of 3; For time window, , Here, j is the sampling point; j is the index of the time sampling point. When calculating spatiotemporal weights, spatial weights are calculated using inverse distance weighting, and the relevant expressions are as follows: ;in, This is the distance attenuation factor, with a default value of 2; For spatial distance, ; The time weighting incorporates a time decay factor in its calculation, and the relevant expression is as follows: ;in, The current moment; The spatial and temporal weights are jointly calculated and normalized to obtain the joint weight. The expression involved is: ; ; The expression involved in interpolating grid node values ​​using joint weights is as follows: ;in, For the target grid Current situational values, such as temperature and vibration amplitude; It should be noted that by calculating the situation value of each grid node through spatiotemporal weighted interpolation, a refined situation field can be directly output. By combining adaptive neighborhood selection with spatiotemporal weights, the interpolation results can preserve local spatial features, such as temperature gradients, and capture dynamic temporal trends, such as power fluctuations, thus solving the situation distortion problem caused by spatiotemporal fragmentation in traditional interpolation algorithms. The combination of dynamic grids and weights can effectively improve the interpolation accuracy in key areas.

[0025] When performing quality assessment and optimization of the refined situation field, 10% of the sensor node data is randomly selected as the validation set, and the root mean square error (RMSE) between the interpolation result and the true value is calculated: Where p represents different validation samples; P represents the number of validation samples; These are interpolated predicted values; This is the actual sensor value; If the root mean square error exceeds the error threshold, adjust the spatial spread factor or time window and re-perform the interpolation; for example, increase the spatial spread factor to 4, or expand the time window to... ; The situation values ​​of all target grid nodes Arranged according to spatial coordinates, forming a refined state field of three-dimensional tensors. Where H×W is the grid resolution. This represents the length of the time series.

[0026] It should be noted that the quality assessment results are fed back to the interpolation parameter adjustment, and the optimized refined situation field serves as the input for subsequent steps. Through cross-validation and dynamic optimization, the interpolation accuracy of the situation field can be ensured to meet the requirements of refined monitoring. The refined situation field in the form of a three-dimensional tensor can intuitively reflect the spatiotemporal distribution characteristics inside the array, such as local hot spot diffusion and vibration wave propagation, which can provide a quantitative basis for the full-domain perception of the power generation situation.

[0027] In this embodiment of the invention, dynamic grid collaborative adaptive interpolation breaks through the traditional uniform grid plus fixed interpolation mode. By adaptively adjusting data density and neighborhood, it achieves the dual goals of full coverage and local fine detail. By coupling spatial distance attenuation with temporal dynamic trends, it can solve the static distortion problem caused by single-dimensional interpolation. Through cross-validation and parameter feedback, it ensures that the accuracy of the situation field is controllable and avoids the unreliability of black-box interpolation.

[0028] The above steps can solve the technical bottleneck of insufficient coverage of key nodes and inadequate fine-grained monitoring in big data analysis of new energy power generation arrays, and provide high-resolution data support for early warning of micro-disturbances within the array.

[0029] The spatiotemporal coupling processing module is used to extract refined micro-perturbation features and macro-trend features of the situation field through multi-scale feature decomposition, and to construct a spatiotemporal coupling feature matrix; the specific steps include: The refined situation field is preprocessed using spatiotemporal standardization, and the standardized and denoised spatiotemporal matrix is ​​output. Spatiotemporal standardization preprocessing includes spatial dimension standardization and temporal dimension smoothing and denoising, both of which are existing conventional technical solutions. The specific implementation steps will not be elaborated here. It should be noted that by standardizing in both spatiotemporal dimensions and unifying the data distribution characteristics, we can avoid strong local signals from masking the global trend. At the same time, Kalman filtering preserves the dynamic changes of the situation field, laying a high-quality data foundation for subsequent multi-scale decomposition.

[0030] When performing multi-scale eigenvalue decomposition, for each spatial grid time series of the spatiotemporal matrix... Perform VMD decomposition, which is a variational mode decomposition. The objective function is: ;in, The set of modal components obtained from the decomposition; It is the set of center frequencies of each mode; Let be the center frequency of the k-th mode; k represents different modes; Here, j is the Dirac function; j is the imaginary unit. It is a complex exponential function; t is a time variable; The square of the L2 norm; the total number of modes K is set to 5; Calculate the center frequency of each modal component If the center frequency is greater than the frequency threshold, the corresponding component is marked as a micro-perturbation mode set. The remaining components are then labeled as macro trend modal sets. ; It should be noted that by adaptively separating modal components at different scales through variational mode decomposition, diverse modal inputs can be provided for subsequent feature extraction. Variational mode decomposition decomposes the situation field into modal components with sparse characteristics through iterative optimization, which can effectively separate micro-perturbations from macro-trends and solve the problem that single-scale analysis cannot take into account both local details and global trends.

[0031] When extracting micro-perturbation features, a set of micro-perturbation modes is obtained. The perturbation amplitude, kurtosis coefficient, and energy entropy of each mode; For adjacent grids (h, w) and (h+1, w), where h is the row index in the vertical direction (e.g., the height of the array) and w is the column index in the horizontal direction (e.g., the width of the array), calculate the spatial cross-correlation coefficient of the micro-features: ;in, For covariance; Microscopic features; These are the standard deviations of the current grid's micro-features and the standard deviations of the micro-features of adjacent grids, respectively. Output the micro-feature vector of each grid point , These are the perturbation amplitude, kurtosis coefficient, and energy entropy of the first micromodal component, respectively. These are the perturbation amplitude, kurtosis coefficient, and energy entropy of the second micromodal component, respectively. It should be noted that by extracting temporal statistical features and spatial correlation features, the intensity, distribution and propagation characteristics of local micro-disturbances within the array can be quantified. By characterizing the nonlinear characteristics of micro-disturbances through multi-dimensional indicators and capturing the propagation path of disturbances within the array by combining spatial cross-correlation, the technical pain point of traditional monitoring being unable to detect local weak anomalies can be solved.

[0032] When extracting macro trend features, the macro trend mode set is... For each mode, the trend term is fitted using the sliding window least squares method: ;in, , ; The slope; The intercept; Calculate the spatial gradient matrix of macro trends: ;in, The rate of change of the trend slope along the x-direction reflects how fast the macro trend changes in the horizontal direction; The slope of the trend along the y-direction represents the rate of change, reflecting how quickly the macro trend changes in the vertical direction; spatial gradient matrix. It comprehensively reflects the degree of spatial unevenness of macro trends; Output the macroscopic feature vector of each grid point , These are the trend slope, trend intercept, and principal period of the third macroscopic modal component, respectively. These are the trend slope and main period of the fourth macroscopic modal component, respectively.

[0033] It should be noted that by using trend fitting and spatial gradient analysis, macroscopic features reflecting the overall operating status of the array are extracted, which complement the microscopic features obtained in the previous processing. The slope and periodic features are used to quantify the dynamic changes of macroscopic trends, such as the increase in power output, the decay rate, and the day-night cycle. Spatial gradients are used to capture the impact of large-scale environmental disturbances, such as uneven temperature field distribution, so as to achieve an accurate characterization of the global status of the array.

[0034] When constructing the spatiotemporal coupled feature matrix, the micro-feature vector and macro-feature vector of each spatial grid (h,w) are concatenated according to their dimensions to obtain the spatiotemporal feature vector: ; The spatiotemporal eigenvectors of all spatial grids are arranged into a four-dimensional tensor according to spatial coordinates (h, w) and time window, and then reduced to a three-dimensional matrix through tensor decomposition to obtain the spatiotemporal coupling feature matrix. : Where core is the core tensor; The projection matrix is ​​the spatial dimension. This is the projection matrix along the time dimension; It is a tensor-matrix multiplication operation, performed only along the first, second, and third dimensions of the tensor; It should be noted that by fusing microscopic and macroscopic feature vectors and performing spatiotemporal dimension expansion and tensor dimensionality reduction, high-dimensional features are compressed into a computable coupling matrix, which can provide input for subsequent federated learning models. Through the concatenation of microscopic and macroscopic features and tensor decomposition, the data dimensionality is reduced while preserving spatiotemporal correlation information, thus addressing the curse of dimensionality in high-resolution situation fields. The constructed spatiotemporal coupled feature matrix... It simultaneously includes details of local perturbations and global trend features, providing multi-scale, coupled feature representations for refined situational assessment within the array.

[0035] In this embodiment of the invention, the above steps can break through the limitations of traditional big data analysis, which focuses on nodes but neglects the whole domain and macroscopic aspects, and thus can provide multi-scale, high-resolution feature basis for early fault warning and performance optimization within new energy power generation arrays.

[0036] The internal health perception and analysis module is used to input the spatiotemporal coupled feature matrix into the federated learning model for distributed training, output the real-time health assessment results of each unit within the array, and display the differences in health status among different regions of the new energy power generation array. Specific steps include: Based on the physical topology of new energy power generation arrays, such as photovoltaic module strings and wind turbine clusters, an improved K-means algorithm is used to divide the array into K edge computing nodes. The central server receives local gradients and local validation set accuracy uploaded by each edge computing node, and calculates the node contribution weights: ;in, For edge computing nodes The number of samples; This is the weighted sum of the health assessment accuracy and path prediction accuracy on the local validation set; m is the summation variable; where the weight of the health assessment accuracy is 0.7 and the weight of the path prediction accuracy is 0.3; the processing of local gradient and local validation set accuracy is a conventional technique, and the specific implementation steps are not described here. The central server aggregates gradients based on contribution weights, using the following expression: ;in, Upload local gradients to different edge computing nodes; Update global model parameters using aggregated gradients: , To update the coefficients, the specific values ​​can be customized according to the actual application scenario, and the updated global model will be applied. Distribute the data to each edge computing node and iteratively train it until the loss function converges; Each edge computing node loads the global model For real-time spatiotemporal coupling characteristics Perform inference and output the unit health status. , Use the Sigmoid activation function; This is the weight matrix of the fully connected layer; For the extracted deep abstract features; For health assessment, head bias is used. Generate a health heatmap based on the output cell health data for each grid cell. The color depth visually displays the differences in health status of different areas of the new energy power generation array, with the color depth indicating implementation based on preset rules.

[0037] It should be noted that by modularly processing and combining the obtained health heatmap, the health assessment is visualized from local units to the entire array, helping maintenance personnel to quickly locate abnormal areas, such as low-health grid clusters, which can effectively solve the limitation of traditional monitoring that only covers key nodes.

[0038] In this embodiment of the invention, the above steps can break through the limitations of traditional centralized model data silos and privacy leaks, and realize big data analysis with full coverage, unit-level fine detail, and spatiotemporal coupling within the new energy power generation array, providing key technical support for early fault warning and operation and maintenance optimization.

[0039] like Figure 2 As shown, a big data analysis method for the renewable energy consumption capacity of a distribution area includes: Deploy a distributed fiber optic sensor array to collect real-time operating parameters and environmental micro-disturbance data of each unit inside the new energy power generation array, and generate a high-density situation field matrix. The high-density situation field matrix is ​​reconstructed by a gridded method based on an adaptive spatiotemporal interpolation algorithm to obtain a refined situation field covering the entire array domain. By extracting the micro-perturbation features and macro-trend features of the refined situation field through multi-scale feature decomposition, a spatiotemporal coupled feature matrix is ​​constructed. The spatiotemporal coupling feature matrix is ​​input into the federated learning model for distributed training, and the real-time health assessment results of each unit within the array are output, and the differences in health status of each region of the new energy power generation array are displayed.

[0040] In the several embodiments provided by this invention, it should be understood that the disclosed system can be implemented in other ways. For example, the embodiments of the invention described above are merely illustrative; for example, the division of modules is only a logical functional division, and there may be other division methods in actual implementation.

[0041] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical modules; they may be located in one place or distributed across multiple network modules. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0042] Furthermore, the functional modules in the various embodiments of the present invention can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module. The integrated module can be implemented in hardware or in the form of hardware plus software functional modules.

[0043] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the essential characteristics of the present invention.

[0044] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A big data analysis system for the renewable energy absorption capacity of a transformer substation, characterized in that, include: The multi-source data acquisition and processing module is used to deploy a distributed fiber optic sensor array to collect real-time operating parameters and environmental micro-disturbance data of each unit inside the new energy power generation array, and generate a high-density situation field matrix. The dataset gridding reconstruction module is used to perform gridding reconstruction of the high-density situation field matrix based on the adaptive spatiotemporal interpolation algorithm to obtain a refined situation field covering the entire array domain. The spatiotemporal coupling processing module is used to extract the micro-perturbation features and macro-trend features of the refined situation field through multi-scale feature decomposition, and to construct the spatiotemporal coupling feature matrix. The internal health perception and analysis module is used to input the spatiotemporal coupling feature matrix into the federated learning model for distributed training, output the real-time health assessment results of each unit inside the array, and display the differences in health status of each region of the new energy power generation array.

2. The big data analysis system for the renewable energy absorption capacity of a transformer substation according to claim 1, characterized in that, The dataset gridding reconstruction module includes: Spatiotemporal weights are calculated by jointly calculating and normalizing the obtained spatial and temporal weights to obtain joint weights. The joint weights are then used to interpolate grid node values ​​to obtain the situation values ​​of the target grid nodes.

3. The big data analysis system for the renewable energy absorption capacity of a transformer substation according to claim 2, characterized in that, Arrange the situation values ​​of all target grid nodes according to spatial coordinates to form a refined situation field of three-dimensional tensor.

4. The big data analysis system for the renewable energy absorption capacity of a transformer substation according to claim 1, characterized in that, The refined situation field is preprocessed by spatiotemporal standardization and outputs a standardized and denoised spatiotemporal matrix. When performing multi-scale feature decomposition, VMD decomposition is performed on the time series of each spatial grid of the spatiotemporal matrix.

5. The big data analysis system for the renewable energy absorption capacity of a transformer substation according to claim 4, characterized in that, Calculate the center frequency of each modal component. If the center frequency is greater than the frequency threshold, mark the corresponding component as the micro-perturbation mode set and mark the remaining components as the macro-trend mode set.

6. The big data analysis system for the renewable energy absorption capacity of a transformer substation according to claim 5, characterized in that, Microscopic perturbation features and macroscopic trend features are extracted separately. The microscopic and macroscopic feature vectors extracted from each spatial grid are concatenated according to their dimensions to obtain the spatiotemporal feature vector. The spatiotemporal feature vectors of all spatial grids are arranged into a four-dimensional tensor according to spatial coordinates and time window, and then reduced to a three-dimensional matrix through tensor decomposition to obtain the spatiotemporal coupling feature matrix.

7. The big data analysis system for the renewable energy absorption capacity of a transformer substation according to claim 1, characterized in that, Based on the physical topology of the new energy power generation array, an improved K-means algorithm is used to divide it into K edge computing nodes; The central server receives the local gradients and local validation set accuracy uploaded by each edge computing node, and calculates the node contribution weights.

8. The big data analysis system for the renewable energy absorption capacity of a transformer substation according to claim 7, characterized in that, The central server aggregates gradients weighted by contribution, uses the aggregated gradients to update the global model parameters, and distributes the updated global model to each edge computing node for iterative training until the loss function converges.

9. A big data analysis system for the renewable energy absorption capacity of a transformer substation according to claim 8, characterized in that, Each edge computing node loads the global model, infers the real-time spatiotemporal coupling characteristics, outputs the unit health status, and generates a health status heatmap by grid unit. The differences in health status of each region of the new energy power generation array are intuitively displayed by the color depth.

10. A big data analysis method for the renewable energy absorption capacity of a transformer substation, applied to a big data analysis system for renewable energy absorption capacity of a transformer substation as described in any one of claims 1-9, characterized in that, include: Deploy a distributed fiber optic sensor array to collect real-time operating parameters and environmental micro-disturbance data of each unit inside the new energy power generation array, and generate a high-density situation field matrix. The high-density situation field matrix is ​​reconstructed by a gridded method based on an adaptive spatiotemporal interpolation algorithm to obtain a refined situation field covering the entire array domain. By extracting the micro-perturbation features and macro-trend features of the refined situation field through multi-scale feature decomposition, a spatiotemporal coupled feature matrix is ​​constructed. The spatiotemporal coupling feature matrix is ​​input into the federated learning model for distributed training, and the real-time health assessment results of each unit within the array are output, and the differences in health status of each region of the new energy power generation array are displayed.

Images