Low-voltage area operation state evaluation method and terminal

Abstract

The application discloses a low-voltage transformer area operation state evaluation method and a terminal, and relates to the technical field of power supply, and in particular relates to a low-voltage transformer area operation state evaluation method and a terminal. The method comprises the following steps: acquiring power data samples of a low-voltage transformer area; constructing a data dimension reduction model based on a deep self-encoding network model and the power data samples; constructing a data fitting model based on a Gaussian mixture model and the power data samples; constructing a data evaluation model based on a deep neural network model and the power data samples; and evaluating the operation state of the low-voltage transformer area based on the data dimension reduction model, the data fitting model and the data evaluation model. The application can not only improve the depth and breadth of data processing, thereby improving the accuracy of overall evaluation, but also reduce the data processing amount of the model, thereby improving the evaluation efficiency.

Description

Technical Field

[0001] The present invention relates to the technical field of power systems, and in particular to a method and terminal for evaluating the operating status of a low-voltage substation. Background Art

[0002] In today's energy industry, with the rapid development and widespread application of smart grid technologies, the operational efficiency and reliability of low-voltage substations have become key factors in improving overall power system performance. The implementation of smart grids requires not only efficient management of electricity consumption but also accurate assessment of the real-time status of the power system. However, traditional power system assessment methods often rely on simplified mathematical models and manual empirical judgment, which are inaccurate and inadequate when dealing with complex system data and dynamic changes. Especially in low-voltage substations, the diversity and uncertainty of power loads make system status assessment more difficult, which directly affects the efficiency of power distribution and system stability. Summary of the Invention

[0003] The technical problem to be solved by the present invention is to provide a method and a terminal for evaluating the operating status of a low-voltage substation, so as to improve the accuracy and efficiency of the evaluation of the operating status of a low-voltage substation.

[0004] In order to solve the above technical problems, the technical solution adopted by the present invention is:

[0005] A method for evaluating the operating status of a low-voltage substation area, comprising:

[0006] Obtain power data samples from low-voltage substations;

[0007] Building a data dimensionality reduction model based on the deep autoencoder network model and the power data sample;

[0008] Building a data fitting model based on a Gaussian mixture model and the power data sample;

[0009] Building a data evaluation model based on a deep neural network model and the power data sample;

[0010] The operating status of the low-voltage substation is evaluated based on the data dimension reduction model, the data fitting model and the data evaluation model.

[0011] In order to solve the above technical problems, another technical solution adopted by the present invention is:

[0012] A low-voltage substation operation status assessment terminal includes a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the computer program, each step in the above-mentioned low-voltage substation operation status assessment method is implemented.

[0013] The beneficial effects of the present invention are: constructing a data dimensionality reduction model based on a deep autoencoding network model to achieve dimensionality reduction processing of the power data of the low-voltage substation, thereby better capturing the key information in the power data and effectively reducing data noise and redundancy, so as to reduce the complexity of data processing by subsequent models and improve data processing efficiency. At the same time, constructing a data fitting model based on a Gaussian mixture model to identify different clustering and distribution trends from the power data, thereby providing detailed information describing the probability characteristics of the data for the subsequent data evaluation model. In addition, constructing a data evaluation model based on a deep neural network model can accurately and efficiently extract the relevant features of the power data, thereby realizing an intelligent evaluation of the operating status of the low-voltage substation. The step-by-step processing based on the data dimensionality reduction model, the data fitting model and the data evaluation model can not only improve the depth and breadth of data processing, thereby improving the accuracy of the overall evaluation, but also reduce the amount of data processing of the model, thereby improving the evaluation efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 A flowchart of a method for evaluating the operating status of a low-voltage substation provided by an embodiment of the present invention;

[0015] Figure 2 A schematic diagram of the architecture of a deep spatiotemporal neural network provided by an embodiment of the present invention;

[0016] Figure 3 A topological diagram of a low-voltage area provided by an embodiment of the present invention;

[0017] Figure 4 A schematic diagram of the architecture of a deep neural network model provided by an embodiment of the present invention;

[0018] Figure 5 A schematic diagram of the structure of a low-voltage substation operation status assessment terminal provided by an embodiment of the present invention;

[0019] Description of labels:

[0020] 100. A low-voltage substation operation status assessment terminal; 101. Memory; 102. Processor. DETAILED DESCRIPTION

[0021] To illustrate the technical content, achieved objectives and effects of the present invention in detail, the following description is given in conjunction with the embodiments and accompanying drawings.

[0022] An embodiment of the present invention provides a method for evaluating the operating status of a low-voltage substation, including:

[0023] Obtain power data samples from low-voltage substations;

[0024] Building a data dimensionality reduction model based on the deep autoencoder network model and the power data sample;

[0025] Building a data fitting model based on a Gaussian mixture model and the power data sample;

[0026] Building a data evaluation model based on a deep neural network model and the power data sample;

[0027] The operating status of the low-voltage substation is evaluated based on the data dimension reduction model, the data fitting model and the data evaluation model.

[0028] From the above description, it can be seen that the beneficial effects of the present invention are: constructing a data dimensionality reduction model based on a deep autoencoding network model to realize dimensionality reduction processing of the power data of the low-voltage substation, so as to better capture the key information in the power data and effectively reduce data noise and redundancy, so as to reduce the complexity of data processing by subsequent models and improve data processing efficiency. At the same time, constructing a data fitting model based on a Gaussian mixture model to realize the identification of different clustering and distribution trends from the power data, thereby providing detailed information describing the probability characteristics of the data for the subsequent data evaluation model. In addition, constructing a data evaluation model based on a deep neural network model can accurately and efficiently extract the relevant features of the power data, thereby realizing an intelligent evaluation of the operating status of the low-voltage substation. The step-by-step processing based on the data dimensionality reduction model, the data fitting model and the data evaluation model can not only improve the depth and breadth of data processing, thereby improving the accuracy of the overall evaluation, but also reduce the amount of data processing of the model, thereby improving the evaluation efficiency.

[0029] Furthermore, constructing a data dimensionality reduction model based on the deep autoencoding network model and the power data sample includes:

[0030] Constructing a first training set based on the power data samples;

[0031] Initialize the weights and bias values ​​of each network structure layer in the deep autoencoder network model;

[0032] Inputting the first training set into the network structure layer for data processing to obtain dimension-reduced output data;

[0033] Calculating a dimensionality reduction model loss value based on the first training set and the dimensionality reduction output data;

[0034] If the dimensionality reduction model loss value does not meet the preset training value, the weight value and the bias value are optimized and updated, and the step of inputting the first training set into the network structure layer for data processing to obtain output data is returned to be executed;

[0035] If the loss value of the dimensionality reduction model meets the preset training value, the training is terminated, and the current weight value and bias value are retained to obtain the data dimensionality reduction model.

[0036] From the above description, it can be seen that the deep autoencoding network model realizes nonlinear transformation of power data through multi-layer network structure layers, and can extract more complex data features than traditional linear methods, thereby ensuring the processing accuracy of subsequent data fitting models and data evaluation models.

[0037] Furthermore, the network structure layer includes a long short-term memory layer, an encoder layer and a decoder layer;

[0038] The step of inputting the first training set into the network structure layer for data processing to obtain dimension-reduced output data includes:

[0039] Inputting the first training set into the long short-term memory layer to perform feature learning to obtain an original feature sequence;

[0040] Compressing the original feature sequence based on the encoder layer to obtain a low-dimensional feature sequence;

[0041] The low-dimensional feature sequence is converted based on the decoder layer to obtain reduced-dimensionality output data.

[0042] As can be seen from the above description, the deep autoencoder network model, based on a combination of long-short-term memory (LSTM) layers, encoder layers, and decoder layers, can effectively reduce data dimensionality while retaining key information. The LSTM layers process the time series characteristics of power data, while the encoder and decoder layers effectively extract the spatial sequence characteristics of power data. This allows for the learning and compression of the most critical features of power data. Compared to linear compression methods, this nonlinear compression method can retain more key features.

[0043] Furthermore, inputting the first training set into the long short-term memory layer for feature learning to obtain an original feature sequence includes:

[0044] The first training set is input into the long short-term memory layer for feature learning, and the output features of the long short-term memory layer are processed by a Dropout algorithm to obtain an original feature sequence.

[0045] From the above description, we can see that when training the long short-term memory layer, the Dropout algorithm randomly sets the output of some nodes in its network layer to zero, and does not update the relevant weight values ​​and bias values. This can effectively reduce the synergistic effect between different features, eliminate and weaken the joint adaptability between neuron nodes, and thus prevent model overfitting.

[0046] Furthermore, the optimizing and updating the weight value and the bias value includes:

[0047] The weight value and the bias value are optimized and updated based on the Adam algorithm.

[0048] Furthermore, constructing a data fitting model based on the Gaussian mixture model and the power data sample includes:

[0049] Performing dimensionality reduction processing on the power data sample using the data dimensionality reduction model to obtain dimensionality-reduced power data, and constructing a second training set based on the dimensionality-reduced power data;

[0050] Initialize the target parameters of each Gaussian component of the Gaussian mixture model;

[0051] The second training set is input into the Gaussian mixture model for calculation, and based on the calculation result, the target parameter in each Gaussian component is iteratively corrected by the maximum expectation method to obtain a data fitting model.

[0052] As can be seen from the above description, constructing a second training set based on the reduced-dimensional power data output by the data dimensionality reduction model enables the data fitting model to effectively capture the data characteristics output by the data dimensionality reduction model, improving the adaptability between the data dimensionality reduction model and the data fitting model. Furthermore, the Gaussian mixture model implements soft clustering, and its output data represents the probability of a particular data point belonging to each cluster. This can better identify different clustering characteristics and distribution trends in the data, thus providing detailed information describing the data's probabilistic characteristics for subsequent data evaluation models.

[0053] Furthermore, the deep neural network model includes a deep spatiotemporal neural network; the deep spatiotemporal neural network includes multiple time resolution layers, multi-branch convolution blocks and fully connected layers designed based on a time pyramid structure;

[0054] The constructing of a data evaluation model based on a deep neural network model and the power data sample includes:

[0055] Acquire key power indicators from the power data samples, and construct a third training set based on the key power indicators;

[0056] Initializing network parameters of the deep spatiotemporal neural network;

[0057] Inputting the third training set into different time resolution layers to perform feature extraction to obtain time feature data;

[0058] Inputting the temporal feature data into the multi-branch convolution block to extract spatial feature data;

[0059] fusing the temporal feature data and the spatial feature data based on the fully connected layer to obtain fused output data;

[0060] Calculate the evaluation model loss value based on the third training set and the fusion output data;

[0061] If the loss value of the evaluation model does not meet the preset training value, the network parameters are optimized and updated, and the step of inputting the third training set into different time resolution layers for feature extraction to obtain time feature data is returned to be executed;

[0062] If the loss value of the evaluation model meets the preset training value, the training is terminated and the current network parameters are retained to obtain the data evaluation model.

[0063] As can be seen from the above description, the multiple time resolution layers designed based on the time pyramid structure can better handle data changes across time scales, thereby effectively analyzing and capturing the data characteristics of power data at different time scales. The multi-branch convolutional block can process data in parallel at multiple spatial scales, thereby more comprehensively capturing the spatial correlations and characteristics of power data. The deep spatiotemporal neural network combined with the time pyramid and multi-branch convolutional blocks can effectively improve the model's adaptability and generalization capabilities to data, achieving a transition from coarse-grained to fine-grained, allowing each network layer in the model to capture different data characteristics, effectively enhancing the model's feature learning capabilities.

[0064] Furthermore, the deep neural network model also includes a graph convolutional neural network;

[0065] If the loss value of the evaluation model meets the preset training value, the training is terminated, and the current network parameters are retained to obtain the data evaluation model, which includes:

[0066] If the loss value of the evaluation model meets the preset training value, the training is terminated and the current network parameters are retained;

[0067] Based on the deep spatiotemporal neural network outputting fused output data under the current network parameters, the graph convolutional neural network is trained by the fused output data to obtain a data evaluation model.

[0068] From the above description, it can be seen that training the graph convolutional neural network through the fusion output data of the deep spatiotemporal neural network can improve the graph convolutional neural network's ability to learn the features of the overall structure of the low-voltage substation, thereby accurately evaluating each node.

[0069] Furthermore, the evaluating the operating status of the low-voltage substation based on the data dimension reduction model, the data fitting model, and the data evaluation model includes:

[0070] Obtaining current power data of the low-voltage area;

[0071] Inputting the power data into the data dimensionality reduction model to obtain dimensionality reduced data;

[0072] Inputting the dimension-reduced data into the data fitting model to obtain clustering features of the power data;

[0073] The dimension reduction data and the clustering features are input into the data evaluation model to obtain a comprehensive evaluation score of the operating status of the low-voltage substation.

[0074] From the above description, it can be seen that compared with the evaluation method that relies on a single model or simple indicators, the present invention conducts a comprehensive evaluation of the operating status of the low-voltage substation based on the data dimensionality reduction model, the data fitting model and the data evaluation model, so as to achieve a comprehensive evaluation of the status of the power system from multiple angles, effectively improving the accuracy of the evaluation results, thereby providing a richer and more detailed basis for the optimization and decision-making of the power system.

[0075] Another embodiment of the present invention provides a low-voltage substation operation status assessment terminal, comprising a memory, a processor, and a computer program stored on the memory and running on the processor. When the processor executes the computer program, it implements each step of the above-mentioned low-voltage substation operation status assessment method.

[0076] From the above description, it can be seen that the beneficial effects of the present invention are: constructing a data dimensionality reduction model based on a deep autoencoding network model to realize dimensionality reduction processing of the power data of the low-voltage substation, so as to better capture the key information in the power data and effectively reduce data noise and redundancy, so as to reduce the complexity of data processing by subsequent models and improve data processing efficiency. At the same time, constructing a data fitting model based on a Gaussian mixture model to realize the identification of different clustering and distribution trends from the power data, thereby providing detailed information describing the probability characteristics of the data for the subsequent data evaluation model. In addition, constructing a data evaluation model based on a deep neural network model can accurately and efficiently extract the relevant features of the power data, thereby realizing an intelligent evaluation of the operating status of the low-voltage substation. The step-by-step processing based on the data dimensionality reduction model, the data fitting model and the data evaluation model can not only improve the depth and breadth of data processing, thereby improving the accuracy of the overall evaluation, but also reduce the amount of data processing of the model, thereby improving the evaluation efficiency.

[0077] The embodiments of the present invention provide a low-voltage substation operating status assessment method and terminal, which can be applied to real-time or near-real-time power system performance monitoring and assessment scenarios, can improve the accuracy and efficiency of low-voltage substation operating status assessment, and provide strong support for power system operation and maintenance management and optimization decision-making. The following is an illustration of the method and terminal through specific embodiments:

[0078] Please refer to Figures 1 to 4 , embodiment 1 of the present invention is:

[0079] A method for evaluating the operating status of a low-voltage substation area, comprising:

[0080] S1. Obtain power data samples from low-voltage substations.

[0081] In some embodiments, the power data samples include power data related to low-voltage substations and power data related to transformers. The power data related to low-voltage substations includes load data, maximum photovoltaic power generation data, three-phase current or three-phase voltage at each node, transmitted and received power data, and load peak-to-valley difference data; the power data related to transformers includes rated capacity and actual load data.

[0082] S2. Constructing a data dimensionality reduction model based on the deep autoencoding network model and the power data sample.

[0083] Specifically, step S2 includes:

[0084] S21. Construct a first training set based on the power data samples.

[0085] In some embodiments, step S21 is specifically as follows: first, each power data sample is subjected to an outlier analysis, abnormal outlier data is eliminated, and default values ​​are filled; then, the load data, photovoltaic maximum power generation data, three-phase current, three-phase voltage and other data in the power data sample are normalized; then, all power data samples are arranged in order from early to late according to their collection time values; finally, the sorted power data samples are converted into corresponding feature matrices, and a first training set is constructed according to the feature matrix.

[0086] In some embodiments, filling the power data sample with a default value specifically includes: filling the default value of the power data sample at time t with the average value of the power data samples collected at time t-1 and time t+1.

[0087] S22. Initialize the weight and bias values ​​of each network structure layer in the deep autoencoder network model.

[0088] S23: Input the first training set into the network structure layer for data processing to obtain dimension-reduced output data.

[0089] The network structure layer includes a long short-term memory layer, an encoder layer and a decoder layer.

[0090] The step S23 includes:

[0091] S231. Input the first training set into the long short-term memory layer to perform feature learning to obtain an original feature sequence.

[0092] In some embodiments, the long short-term memory layer is specifically:

[0093]

[0094] Among them, f t Represents the activation value of the forget gate, i tRepresents the activation value of the input gate, C~ t represents the value of the candidate memory cell, o t represents the activation value of the output gate, C t Indicates the value of the memory cell at the current moment, h t represents the hidden state at the current moment, σ represents the Sigmoid function, tanh represents the hyperbolic tangent function, ω represents the weight value, b represents the bias value, and x t represents the input at time t, h t-1 represents the hidden state at the previous moment, C t-1 Indicates the value of the memory cell at the previous moment.

[0095] In an optional embodiment, the step S231 includes: inputting the first training set into the long short-term memory layer for feature learning, and processing the output features of the long short-term memory layer through a Dropout algorithm to obtain an original feature sequence.

[0096] S232. Compress the original feature sequence based on the encoder layer to obtain a low-dimensional feature sequence.

[0097] S233. Convert the low-dimensional feature sequence based on the decoder layer to obtain reduced-dimensionality output data.

[0098] In some embodiments, the network structure of the deep autoencoder network model further includes an input layer, a fully connected layer, and an output layer. The encoder layer and the decoder layer are composed of convolutional layers. Specifically, the network structure of the deep autoencoder network model is as follows: the input layer is sequentially connected to a long short-term memory layer, an encoder layer, a decoder layer, a fully connected layer, and an output layer. Specifically, the input layer is used to directly receive raw multi-dimensional power data samples to ensure that all important system measurement indicators in the power data samples can be learned by the model, allowing the data dimensionality reduction model to understand the full picture of the input data. The long short-term memory layer (LSTM) is used to process the time series characteristics of the power data samples. The LSTM layer can remember long-term dependencies, which is particularly critical for predicting power demand and load changes. Therefore, feature extraction based on the LSTM layer enables the data dimensionality reduction model to understand and predict time-dependent features of the power data samples, such as the diurnal cycle of load data. The encoder and decoder, composed of convolutional layers, are used to extract spatial features from the power data samples. By learning the local features of the data, the convolutional layer enables the data dimensionality reduction model to capture the spatial correlation between indicators such as voltage and current, helping the data dimensionality reduction model understand and predict the interactions between various components of the power system. The fully connected layer further compresses the data and reduces its dimensionality. It merges the output data from the previous network structure layer to form a global data set, effectively capturing the most critical information from the power data sample. This nonlinear compression method retains more useful information than linear methods. The output layer outputs the features of the reduced-dimensional data. The output layer reconstructs the input data, minimizing the difference between the input data of the data dimensionality reduction model and the reconstructed, reduced-dimensional output data. This enables the data dimensionality reduction model to learn compression methods that characterize the most critical features of the data.

[0099] S24. Calculate a dimensionality reduction model loss value based on the first training set and the dimensionality reduction output data.

[0100] In some embodiments, the loss function uses mean square error (MSE) to calculate the dimensionality reduction model loss value. Where n represents the amount of data in the first training set, x i represents the i-th power data sample in the first training set, Indicates that the ith power data sample corresponds to the reduced-dimensional output data.

[0101] S25. If the dimensionality reduction model loss value does not meet the preset training value, the weight value and the bias value are optimized and updated, and the process returns to execute step S23.

[0102] In an optional implementation, the weight value and the bias value are optimized and updated based on the Adam algorithm.

[0103] S26. If the loss value of the dimensionality reduction model meets the preset training value, the training is terminated, and the current weight value and bias value are retained to obtain the data dimensionality reduction model.

[0104] In some embodiments, when the loss value of the dimensionality reduction model is minimized, the training of the model is terminated, and the current weight value ω and bias value b are retained. A data dimensionality reduction model is obtained based on the current weight value ω and bias value b and the network structure layer of the deep autoencoder network model.

[0105] S3. Constructing a data fitting model based on the Gaussian mixture model and the power data sample.

[0106] Specifically, step S3 includes:

[0107] S31. Performing dimensionality reduction processing on the power data samples using the data dimensionality reduction model to obtain dimensionality-reduced power data, and constructing a second training set based on the dimensionality-reduced power data.

[0108] In some embodiments, each reduced-dimensional power data is subjected to outlier analysis, abnormal outlier data is removed and filled with default values, and the filled reduced-dimensional power data is arranged in order from early to late according to the collection time value to obtain a second training set.

[0109] S32. Initialize the target parameters of each Gaussian component of the Gaussian mixture model.

[0110] In some embodiments, the target parameters include a mean, a covariance matrix, and a mixing coefficient.

[0111] Each Gaussian component has a corresponding mean and covariance structure, making the Gaussian mixture model adaptable to various clustering forms. Adjusting the covariance parameter allows clusters to be elliptical or other shapes, closer to the natural form of real data distribution.

[0112] S33, inputting the second training set into the Gaussian mixture model for calculation, and iteratively correcting the target parameters in each Gaussian component by the maximum expectation method based on the calculation results to obtain a data fitting model.

[0113] In some embodiments, the maximum expectation method includes an E (Expectation) step and an M (Maximization) step. The E step is used to calculate the probability that each data sample belongs to each cluster (i.e., the weight of the soft cluster), and the M step is used to update the target parameters of the Gaussian mixture model to maximize the likelihood function of the data. This method of iteratively optimizing the target parameters can effectively find the optimal parameters. In the E step, the posterior probability of each data in the second training set under different Gaussian components is calculated based on the current target parameters. In the M step, the target parameters of each Gaussian component are updated to maximize the likelihood function of the data. The likelihood function is specifically:

[0114]

[0115] Where L(x|λ) represents the log-likelihood function, T represents the length of the sequence data input to the Gaussian mixture model, N represents the probability density function of the cth Gaussian component, μ c represents the mean vector, c represents the number of Gaussian components in the Gaussian mixture model, x (t) Represents the power data at time t.

[0116] It should be noted that the power data is clustered based on the data fitting model to analyze the discrimination and clustering trends between different power data. Specifically, the output of the data fitting model enables the subsequent data evaluation model to obtain the uncertainty of each power data point belonging to each cluster and the clustering structure of all power data.

[0117] In some embodiments, the reduced-dimensionality power data is input into a trained data fitting model to calculate its log-likelihood, where the log-likelihood reflects the degree of fit between the data and the model. The log-likelihood is converted into a comprehensive evaluation score between 0 and 1 using a preset mapping function. The comprehensive evaluation score is then used to optimize the third training set of the data evaluation model described below.

[0118] S4. Construct a data evaluation model based on the deep neural network model and the power data sample.

[0119] The deep neural network model includes a deep spatiotemporal neural network; the deep spatiotemporal neural network includes multiple time resolution layers, multi-branch convolution blocks and fully connected layers designed based on a time pyramid structure. Figure 2As shown in the figure, the deep spatiotemporal neural network includes four time resolution layers and four multi-branch convolution blocks. Each time resolution layer is connected to a corresponding multi-branch convolution block, and the four multi-branch convolution blocks are all connected to the fully connected layer. Among them, X1 is the original input data, representing the finest-grained power data, with a time resolution of 5 minutes, used to capture subtle changes in the power data of the low-voltage substation on a short time scale; X2 represents the data after preliminary downsampling or aggregation of the original input data X1, with a time resolution of 10 minutes; X3 represents the data after further downsampling of the data X2, with a time resolution of 25 minutes; X4 represents the highest-level data, used to capture the performance trends and periodic changes of the power data of the low-voltage substation on the longest time scale, with a time resolution of 50 minutes.

[0120] It should be noted that multiple temporal resolution layers are used to capture data features of varying lengths. Furthermore, multi-branch convolutional blocks are used to extract spatial features from these data features, ensuring that the data evaluation model accurately captures the dynamic changes in the power system across both spatial and temporal dimensions. Fully connected layers are used to further process and fuse temporal and spatial features.

[0121] Specifically, step S4 includes:

[0122] S41. Obtain key power indicators from the power data samples, and construct a third training set based on the key power indicators.

[0123] In some embodiments, the comprehensive evaluation score calculated based on the above-mentioned data fitting model is used to screen out data with higher fitting degrees from the power data samples, and a third training set is constructed based on these power data samples with higher fitting degrees, and then the weight of the data is adjusted, so that the data evaluation model pays more attention to areas with poor fitting degrees during training.

[0124] In some embodiments, key power indicators in the power data sample can be determined by combining the collected historical power data and real-time power data related to the low-voltage substation with expert experience.

[0125] S42: Initialize the network parameters of the deep spatiotemporal neural network.

[0126] S43: Input the third training set into different time resolution layers to perform feature extraction to obtain time feature data.

[0127] S44. Input the temporal feature data into the multi-branch convolution block to extract spatial feature data.

[0128] S45. Fusing the temporal feature data and the spatial feature data based on the fully connected layer to obtain fused output data.

[0129] S46. Calculate the evaluation model loss value based on the third training set and the fusion output data.

[0130] In some embodiments, the loss function uses mean square error (MSE) to calculate the loss value of the evaluation model.

[0131] S47. If the loss value of the evaluation model does not meet the preset training value, the network parameters are optimized and updated, and the process returns to execute step S43.

[0132] In some embodiments, the network parameters of the model are updated based on the Adam optimization algorithm and through back propagation.

[0133] S48. If the loss value of the evaluation model meets the preset training value, the training is terminated and the current network parameters are retained to obtain the data evaluation model.

[0134] In some embodiments, a validation dataset may be constructed based on key power indicators, and the performance of the trained data evaluation model may be evaluated based on the validation dataset to monitor the training accuracy and prevent the data evaluation model from overfitting.

[0135] In an optional embodiment, the deep neural network model also includes a graph convolutional neural network, and step S48 includes: if the loss value of the evaluation model meets the preset training value, then the training is terminated and the current network parameters are retained; based on the deep spatiotemporal neural network output, the fused output data under the current network parameters is output, and the graph convolutional neural network is trained by the fused output data to obtain a data evaluation model.

[0136] S5. Evaluate the operating status of the low-voltage substation based on the data dimension reduction model, the data fitting model and the data evaluation model.

[0137] Specifically, step S5 includes:

[0138] S51. Obtain current power data of the low-voltage substation.

[0139] In some embodiments, as Figure 3 As shown, the low-voltage area includes energy storage devices, household photovoltaics, load units (such as electric vehicles, etc.) and voltage source converters. Specifically, the load data of the current low-voltage area, the maximum photovoltaic power generation, the three-phase voltage of each node, the three-phase current of each branch, and the time data and environmental data related to the node to be evaluated are obtained. Among them, the time data is recorded in the format of year-month-day-hour. The status evaluation is performed based on the relevant data of the low-voltage area running for one hour. If the current time is 12:00 on March 1, 2024, the time when the operating status needs to be evaluated is 13:00 on March 1, 2024.

[0140] S52: Input the power data into the data dimensionality reduction model to obtain dimensionality-reduced data.

[0141] It should be noted that the data dimensionality reduction model is used to preprocess and reduce the dimensionality of the collected raw power data, thereby reducing the complexity of data processing for subsequent models and improving data processing efficiency. The data dimensionality reduction model converts high-dimensional data into lower-dimensional data through an encoding process, thereby capturing key information in the power data and reducing noise and redundancy. Simultaneously, the data is reconstructed through a decoding process, enabling the model to learn to retain the most critical features.

[0142] S53: Input the dimension-reduced data into the data fitting model to obtain clustering features of the power data.

[0143] It should be noted that the data fitting model is used for statistical analysis and feature extraction of power data. It identifies different clustering and distribution trends from the power data, providing detailed information describing the probabilistic characteristics of the data for subsequent data evaluation models. The data fitting model models the probability distribution of the dimensionality-reduced data.

[0144] S54. Input the dimension reduction data and the clustering features into the data evaluation model to obtain a comprehensive evaluation score of the operating status of the low-voltage substation.

[0145] In some embodiments, as Figure 4 As shown in Figure 1, the dimensionality reduction data and clustering features of each node are input into the data evaluation model. The data is processed by a multi-branch convolution module to extract features of different scales. Then, different LSTM (long short-term memory) layers are used to capture the time dependency in the data and extract high-level temporal features. The data processed by the LSTM layer is input into the linear regression layer to generate a score, and finally the score is output. Where (80) and (40) represent the number of neurons in different LSTM layers. Thus, the data evaluation model outputs the performance score of each node by learning the relationship between its feature information and key power indicators, and then determines the total evaluation score of the current low-voltage substation operation status.

[0146] It should be noted that the data evaluation model is used for comprehensive feature learning and performance evaluation of power data. Through its complex network structure layer, it integrates time series and spatial features, and accurately maps these features to key indicators of the power system to achieve accurate evaluation of the power status of the low-voltage substation.

[0147] Please refer to Figure 5 , the second embodiment of the present invention is:

[0148] A low-voltage substation operation status assessment terminal 100 includes a memory 101, a processor 102, and a computer program stored in the memory 101 and running on the processor 102. When the processor 102 executes the computer program, it implements each step of a low-voltage substation operation status assessment method described in Example 1.

[0149] In summary, the present invention provides a method and terminal for evaluating the operating status of a low-voltage substation, which constructs a data dimensionality reduction model based on a deep autoencoding network model to achieve dimensionality reduction processing of the power data of the low-voltage substation, which not only improves the ability to process large-scale power data, but also better captures the key information in the power data and effectively reduces data noise and redundancy, reduces the complexity of subsequent models for data processing, and improves data processing efficiency. At the same time, a data fitting model is constructed based on a Gaussian mixture model to identify different clustering and distribution trends from the power data, thereby providing detailed information describing the probability characteristics of the data for subsequent data evaluation models. In addition, the data evaluation model constructed based on the deep neural network model can accurately and efficiently extract the relevant features of the power data, thereby achieving intelligent evaluation of the operating status of the low-voltage substation. Compared with the evaluation method that relies on a single model or simple indicator, the present invention integrates a data dimensionality reduction model, a data fitting model, and a data evaluation model to perform power performance evaluation, which can not only improve the depth and breadth of data processing, thereby improving the accuracy of the overall evaluation, but also reduce the data processing volume of the model, thereby improving the evaluation efficiency.

[0150] The above descriptions are merely embodiments of the present invention and are not intended to limit the patent scope of the present invention. Any equivalent transformations made using the contents of the present invention's description and drawings, or directly or indirectly applied in related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A method for evaluating the operating status of a low-voltage substation, characterized in that: include: Obtain power data samples from low-voltage substations; Building a data dimensionality reduction model based on the deep autoencoder network model and the power data sample; Building a data fitting model based on a Gaussian mixture model and the power data sample; Building a data evaluation model based on a deep neural network model and the power data sample; Evaluate the operating status of the low-voltage substation based on the data dimension reduction model, the data fitting model, and the data evaluation model; The constructing of a data fitting model based on the Gaussian mixture model and the power data sample includes: Performing dimensionality reduction processing on the power data sample using the data dimensionality reduction model to obtain dimensionality-reduced power data, and constructing a second training set based on the dimensionality-reduced power data; Initialize the target parameters of each Gaussian component of the Gaussian mixture model; Inputting the second training set into the Gaussian mixture model for calculation, and iteratively correcting the target parameters in each Gaussian component by the maximum expectation method based on the calculation results to obtain a data fitting model; The deep neural network model includes a deep spatiotemporal neural network; the deep spatiotemporal neural network includes multiple time resolution layers, multi-branch convolution blocks and fully connected layers designed based on a time pyramid structure; The constructing of a data evaluation model based on a deep neural network model and the power data sample includes: Acquire key power indicators from the power data samples, and construct a third training set based on the key power indicators; Initializing network parameters of the deep spatiotemporal neural network; Inputting the third training set into different time resolution layers to perform feature extraction to obtain time feature data; Inputting the temporal feature data into the multi-branch convolution block to extract spatial feature data; fusing the temporal feature data and the spatial feature data based on the fully connected layer to obtain fused output data; Calculate the evaluation model loss value based on the third training set and the fusion output data; If the loss value of the evaluation model does not meet the preset training value, the network parameters are optimized and updated, and the step of inputting the third training set into different time resolution layers for feature extraction to obtain time feature data is returned to be executed; If the loss value of the evaluation model meets the preset training value, the training is terminated and the current network parameters are retained to obtain the data evaluation model.

2. A method for evaluating the operation status of a low-voltage substation according to claim 1, characterized in that: The constructing of a data dimensionality reduction model based on the deep autoencoding network model and the power data sample includes: Constructing a first training set based on the power data samples; Initialize the weights and bias values ​​of each network structure layer in the deep autoencoder network model; Inputting the first training set into the network structure layer for data processing to obtain dimension-reduced output data; Calculating a dimensionality reduction model loss value based on the first training set and the dimensionality reduction output data; If the dimensionality reduction model loss value does not meet the preset training value, the weight value and the bias value are optimized and updated, and the step of inputting the first training set into the network structure layer for data processing to obtain output data is returned to be executed; If the loss value of the dimensionality reduction model meets the preset training value, the training is terminated, and the current weight value and bias value are retained to obtain the data dimensionality reduction model.

3. A method for evaluating the operation status of a low-voltage substation according to claim 2, characterized in that: The network structure layer includes a long short-term memory layer, an encoder layer and a decoder layer; The step of inputting the first training set into the network structure layer for data processing to obtain dimension-reduced output data includes: Inputting the first training set into the long short-term memory layer to perform feature learning to obtain an original feature sequence; Compressing the original feature sequence based on the encoder layer to obtain a low-dimensional feature sequence; The low-dimensional feature sequence is converted based on the decoder layer to obtain reduced-dimensionality output data.

4. A method for evaluating the operation status of a low-voltage substation according to claim 3, characterized in that: Inputting the first training set into the long short-term memory layer to perform feature learning to obtain an original feature sequence includes: The first training set is input into the long short-term memory layer for feature learning, and the output features of the long short-term memory layer are processed by a Dropout algorithm to obtain an original feature sequence.

5. A method for evaluating the operation status of a low-voltage substation according to claim 2, characterized in that: The optimizing and updating of the weight value and the bias value includes: The weight value and the bias value are optimized and updated based on the Adam algorithm.

6. A method for evaluating the operation status of a low-voltage substation according to claim 1, characterized in that: The deep neural network model also includes a graph convolutional neural network; If the loss value of the evaluation model meets the preset training value, the training is terminated, and the current network parameters are retained to obtain the data evaluation model, which includes: If the loss value of the evaluation model meets the preset training value, the training is terminated and the current network parameters are retained; Based on the deep spatiotemporal neural network outputting fused output data under the current network parameters, the graph convolutional neural network is trained by the fused output data to obtain a data evaluation model.

7. A method for evaluating the operation status of a low-voltage substation according to claim 1, characterized in that: The evaluating the operating status of the low-voltage substation based on the data dimension reduction model, the data fitting model, and the data evaluation model includes: Obtaining current power data of the low-voltage area; Inputting the power data into the data dimensionality reduction model to obtain dimensionality reduced data; Inputting the dimension-reduced data into the data fitting model to obtain clustering features of the power data; The dimension reduction data and the clustering features are input into the data evaluation model to obtain a comprehensive evaluation score of the operating status of the low-voltage substation.

8. A low-voltage substation operation status assessment terminal, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that: When the processor executes the computer program, each step of the low-voltage substation operation status assessment method according to any one of claims 1 to 7 is implemented.

Images