Low-voltage transformer area power distribution network topology identification and verification method, system and device based on low perceptibility and medium
By using data augmentation and adaptive density peak clustering algorithms, the problem of missing topology information in the topology identification of low-voltage distribution networks was solved, enabling accurate identification of distribution areas and customer phases, and improving the accuracy of topology identification and the ability to adapt to new energy sources in low-voltage distribution networks.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-11-24
- Publication Date
- 2026-03-10
AI Technical Summary
Due to poor planning and low level of informatization in the early stage, the topology of the low-voltage distribution network is inaccurate. After the new energy is connected, the topology information is missing, and the relationship between households and transformers and between households and phases cannot be accurately identified.
By acquiring observable node data, preprocessing it, and then inputting it into the TS-WCGAN-GP model for data augmentation, the DTW and Euclidean distance are calculated. An improved adaptive density peak clustering algorithm is used to determine the number of transformer substations and cluster centers. The DTW distance is then combined with the DTW distance to identify transformer substations and households.
In low-perception scenarios, it automatically identifies the topology of the transformer area, improves the accuracy of topology identification, reduces reliance on measurement devices, and enhances adaptability to high-penetration scenarios of new energy sources.
Smart Images
Figure CN121642991A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the technical field of power distribution network topology identification, and in particular to a low-voltage transformer area power distribution network topology identification and checking method, system, device and medium based on low perception. BACKGROUND
[0002] As a link between the power transmission network and the end user, the low-voltage power distribution network contains a large amount of data and information. How to accurately use such information for power distribution network planning, operation and control is of great significance to the stable and safe operation of the low-voltage power distribution network. Under this background, the correct topology structure of the low-voltage power distribution network is an important basis for ensuring the safe operation and normal maintenance of the power distribution network.
[0003] However, the low-voltage transformer area power distribution network currently has the characteristics of low perception due to poor early planning, low informatization level and other problems. The actual topology is not completely consistent with the designed topology, the low-voltage topology is not accurate enough, and the information interaction is limited to only some observable nodes. This leads to the lack of topology information and low data quality of the power distribution network, and the lack of network measurement points further exacerbates the difficulty of power distribution network topology identification. How to identify and check the topology of the low-voltage power distribution network based on limited data to mine the operation rules of the key nodes of the transformer area topology, analyze the voltage sequence similarity, and divide the specific connection relationship between the transformer, branch nodes, lines and users is a key problem that needs to be solved. SUMMARY
[0004] In view of the above existing problems, the present application is proposed.
[0005] Therefore, the present application provides a low-voltage transformer area power distribution network topology identification and checking method and system based on low perception to solve the problem that the data quality is poor, the topology information is missing after the access of new energy, and the house-transformer relationship and house-related relationship cannot be accurately identified under the current low perception.
[0006] To solve the above technical problems, the present application provides the following technical solutions: In a first aspect, the present application provides a low-voltage transformer area power distribution network topology identification and checking method based on low perception, comprising: obtaining the historical data sequence of the observable node data in the power distribution network and performing data preprocessing; inputting the preprocessed data into a TS-WCGAN-GP model to perform data enhancement and generate an enhanced voltage time series data set; based on the obtained enhanced voltage time series data set, calculating the DTW distance and Euclidean distance of each voltage sequence to obtain a comprehensive distance index; Based on the comprehensive distance index, the number of transformer areas and the corresponding voltage sequences of the clustering centers are automatically determined by the improved adaptive density peak clustering algorithm, which is used for transformer area attribution and house phase identification; The DTW distance between the preprocessed historical data sequence and the clustering center obtained by adaptive clustering is calculated, the data is clustered, and the transformer area attribution of the voltage sequence is obtained. By comparing the DTW distance of each voltage sequence and the low-voltage side three-phase voltage of the transformer area, the house phase identification in the transformer area is realized.
[0007] As a preferred scheme of the low-voltage transformer area power distribution network topology identification and checking method based on low perception degree, the preprocessed data is input into the TS-WCGAN-GP model, data enhancement is performed, and an enhanced voltage time series data set is generated, including: The preprocessed data is input into the space-time attention mechanism model, the time feature weight of each historical voltage sequence and the space feature weight of each sampling point are calculated, and the space-time feature weight is fused and generated; The space-time feature weight is input into the conditional generative adversarial network model, and the Wasserstein distance is used as the measurement standard to measure the distribution difference between the generated sample data and the real sample data; The gradient penalty is introduced to strengthen the Lipschitz constraint of the discriminant network in the adversarial network model, so that the loss function of the discriminant network satisfies the Lipschitz continuity, and an improved optimization objective function is obtained; By alternately optimizing the generator and the discriminator until Nash equilibrium is reached, the enhanced voltage time series data set is generated.
[0008] As a preferred scheme of the low-voltage transformer area power distribution network topology identification and checking method based on low perception degree, based on the comprehensive distance index, the number of transformer areas and the corresponding voltage sequences of the clustering centers are automatically determined by the improved adaptive density peak clustering algorithm, including: Based on the comprehensive distance, the local density and the relative distance of each voltage sequence are calculated; Select the points that meet the conditions of local density greater than the first threshold value and relative distance greater than the second threshold value as the initial clustering center candidate set.
[0009] As a preferred scheme of the low-voltage transformer area power distribution network topology identification and checking method based on low perception degree, based on the comprehensive distance index, the number of transformer areas and the corresponding voltage sequences of the clustering centers are automatically determined by the improved adaptive density peak clustering algorithm, and further including: Run the basic density peak clustering, calculate the initial clustering center set, and obtain the number of transformer areas represented by the number of clusters; Cluster the data based on the initial cluster centers and calculate the cluster boundary density and intersection density. Iterative merging is performed based on cluster boundary density and intersection density to output the final number of transformer areas, the corresponding cluster centers, and the corresponding voltage sequence.
[0010] The beneficial effects of this preferred technical solution are: in low-perception scenarios where the number of transformer substations is unknown, there is very little observable data, and the enhanced dataset is large in scale, the actual number of existing transformer substations can be determined automatically without any manual intervention. The adaptive merging mechanism can effectively filter out pseudo-centers and avoid over-clustering or under-clustering.
[0011] As a preferred embodiment of the low-voltage distribution network topology identification and verification method based on low-sensitivity described in this invention, the method includes: calculating the DTW distance between the preprocessed historical data sequences and the cluster centers obtained by adaptive clustering, performing data clustering, and obtaining the distribution area affiliation of the voltage sequences, including: Confirm the preprocessed historical data sequence and the number of distribution transformers, K. One voltage sequence DATA is randomly selected from the historical data sequence as the first initial cluster point; Calculate the DTW distance between the voltage sequences of all users and the cluster centers obtained by adaptive clustering. The probability of each voltage sequence being selected as the next location is denoted as... The voltage sequence with the highest probability value is selected as the next aggregation point; Repeat the calculation until K cluster points are selected; Cluster all voltage sequences with the DTW values of each cluster point, and assign the voltage sequence with the smallest DTW value with each cluster point to the corresponding cluster point's area, thus dividing all voltage sequences into K sets. Voltage sequences contained in the same set belong to the same user voltage sequence of the same transformer area.
[0012] As a preferred embodiment of the low-voltage distribution network topology identification and verification method based on low-sensitivity described in this invention, the method involves: achieving phase identification within the distribution area by comparing the DTW distance between each voltage sequence and the three-phase voltage on the low-voltage side of the distribution transformer in the respective distribution area, including: Calculate the DTW distance between the voltage sequence of each user's meter and the three-phase voltage sequence on the low-voltage side of the distribution transformer within the transformer area; By comparison, the user with the smallest distance from the voltage sequence DTW of phase A on the low-voltage side of the distribution transformer is classified as phase A; Repeat the comparison until all users in phases B and C have been identified.
[0013] As a preferred embodiment of the low-voltage distribution network topology identification and verification method based on low-sensitivity described in this invention, the method involves: calculating the DTW distance and Euclidean distance of each voltage sequence based on the obtained enhanced voltage time-series dataset to obtain a comprehensive distance index, including: The DTW distance and Euclidean distance of each normalized voltage sequence are added together to obtain the comprehensive distance index.
[0014] Secondly, the present invention provides a low-voltage distribution network topology identification and verification system based on low-sensitivity, comprising: The preprocessing module is used to acquire historical data sequences of observable nodes in the distribution network and perform data preprocessing. The data augmentation module is used to input the preprocessed data into the TS-WCGAN-GP model for data augmentation and to generate an enhanced voltage time series dataset. The distance calculation module is used to calculate the DTW distance and Euclidean distance of each voltage sequence based on the obtained enhanced voltage time series dataset, and obtain a comprehensive distance index. The first clustering module is used to automatically determine the number of transformer substations and the voltage sequence corresponding to the K cluster centers based on the comprehensive distance index and an improved adaptive density peak clustering algorithm, for substation affiliation and household phase identification. The second clustering module is used to calculate the DTW distance between the preprocessed historical data sequences and the cluster centers obtained by adaptive clustering, and to perform data clustering to obtain the substation affiliation of the voltage sequences. The comparison and identification module is used to identify the phase of each household within the distribution area by comparing the DTW distance between each voltage sequence and the three-phase voltage on the low-voltage side of the distribution transformer in the area to which it belongs.
[0015] Thirdly, the present invention provides a computer device, comprising: Memory and processor; The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions. When the computer-executable instructions are executed by the processor, they implement the steps of a method for identifying and verifying the topology of a low-voltage distribution network based on low-sensitivity.
[0016] Fourthly, the present invention provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the steps of the method for identifying and verifying the topology of a low-voltage distribution network based on low-sensitivity.
[0017] Compared with existing technologies, the beneficial effects of this invention are as follows: By correcting and normalizing limited measured voltage data through preprocessing, and combining it with the conditional WGAN-GP mechanism of spatiotemporal attention, high-quality data augmentation is achieved, effectively expanding the dataset size while fully preserving the original spatiotemporal correlation and fluctuation characteristics. Based on this, a comprehensive DTW and Euclidean distance index is constructed, and adaptive density peak clustering is used to automatically determine the actual number of transformer substations and their operating curves, thus solving the problem of unknown substation numbers. Furthermore, DTW similarity grading is used to accurately identify user-substation affiliation and the phase of households within a substation. Overall, automatic identification and real-time verification of the entire substation topology can be achieved by relying only on a small number of observable node voltages, improving topology accuracy, reducing dependence on measurement devices, and enhancing adaptability to high-penetration renewable energy scenarios. Attached Figure Description
[0018] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0019] Figure 1 This is a schematic diagram of the overall process of a low-voltage distribution network topology identification and verification method based on low-sensitivity, according to an embodiment of the present invention.
[0020] Figure 2 This is a schematic diagram of the spatiotemporal attention mechanism model structure in a low-voltage distribution network topology identification and verification method based on low-perception, as described in an embodiment of the present invention.
[0021] Figure 3 This is a schematic diagram of the generated model G structure in the TS-WCGAN-GP model of a low-voltage distribution network topology identification and verification method based on low-sensitivity, as described in an embodiment of the present invention.
[0022] Figure 4 This is a schematic diagram of the discriminant model D in the TS-WCGAN-GP model of a low-voltage distribution network topology identification and verification method based on low-sensitivity, as described in an embodiment of the present invention.
[0023] Figure 5 This is an exemplary topology diagram of a distribution network topology identification and verification method based on low-sensitivity in a certain embodiment of the present invention.
[0024] Figure 6 This is a schematic diagram of the heat map distribution of exemplary real data in a method for identifying and verifying the topology of a low-voltage distribution network based on low-sensitivity, as described in an embodiment of the present invention.
[0025] Figure 7 This is an exemplary data augmentation heatmap distribution diagram of a low-voltage distribution network topology identification and verification method based on low-sensitivity, as described in an embodiment of the present invention.
[0026] Figure 8 This is an exemplary real data fluctuation density diagram in a method for identifying and verifying the topology of a low-voltage distribution network based on low-sensitivity, as described in an embodiment of the present invention.
[0027] Figure 9 This is an exemplary data fluctuation density diagram illustrating a low-voltage distribution network topology identification and verification method based on low-sensitivity, as described in an embodiment of the present invention. Detailed Implementation
[0028] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.
[0029] Example 1, referring to Figure 1 As one embodiment of the present invention, a method for topology identification and verification of low-voltage distribution networks based on low-sensitivity is provided, comprising: S100: Acquire historical data sequences of observable nodes within the distribution network and perform data preprocessing; S200: Input the preprocessed data into the TS-WCGAN-GP model for data augmentation and generate an enhanced voltage time series dataset; S300: Based on the obtained enhanced voltage time series dataset, calculate the DTW distance and Euclidean distance of each voltage sequence to obtain a comprehensive distance index; S400: Based on the comprehensive distance index, the improved adaptive density peak clustering algorithm automatically determines the number of transformer substations and the voltage sequence corresponding to the cluster centers, which is used for transformer substation affiliation and household phase identification. S500: Calculate the DTW distance between the preprocessed historical data sequence and the cluster center obtained by adaptive clustering, perform data clustering, and obtain the substation affiliation of the voltage sequence; S600: By comparing the DTW distance between each voltage sequence and the three-phase voltage on the low-voltage side of the distribution transformer in its respective area, the phase identification of households within the distribution area is realized.
[0030] It should be noted that in steps S100-S600 above, the spatiotemporal feature weights of each node are calculated by considering the time-space attention mechanism, and the spatiotemporal feature sequences are fused and input into a Conditional Generative Adversarial Network (CGAN) for data augmentation. Wasserstein distance is used as a metric to measure the distribution difference between the generated sample data and the real sample data. Gradient penalty (GP) is added to improve the stability of the data augmentation algorithm. Through the TS-WCGAN-GP algorithm, the limited data of the low-voltage distribution network is augmented to form a large-scale dataset with the same characteristics. Based on the acquired key node dataset, the DTW distance and Euclidean distance (ED) of each voltage sequence are calculated to obtain a comprehensive distance judgment index. The density peak clustering method is used to calculate the number of transformer areas to which the user belongs and the voltage operation pattern. Furthermore, the DTW distance between the node data in the initial dataset and the cluster center is calculated to cluster different users within the same transformer area. This method enables accurate identification and real-time verification of transformer substation topology based on limited topology data, further improving the accuracy and speed of low-voltage distribution network topology identification.
[0031] Example 2, refer to Figures 1-4 As an embodiment of the present invention, based on the above embodiment, a method for identifying and verifying the topology of a low-voltage distribution network based on low-sensitivity is provided.
[0032] In this embodiment of the application, in step S100, historical data sequences of observable node data within the distribution network are obtained and data preprocessing is performed; Specifically, the data preprocessing for historical data of observable nodes includes: outlier handling and missing value imputation; after processing, the processed data is normalized in order to better evaluate the experimental results. Abnormal values include: exceeding the full load voltage value and garbled characters; among them, the handling method for exceeding the full load value and garbled characters is to use the instantaneous voltage at the two sampling times before and after to perform mean-variance normalization correction on the exceeding the full load value and garbled characters. The missing value is filled by using the instantaneous voltage at the two sampling times before and after.
[0033] The normalization method is shown in the following formula: In the formula, Category of customer-side load, For time scale, .
[0034] In one alternative implementation, in addition to outlier handling and missing value imputation, a sliding window mechanism is further introduced. At the same time, voltage sequences with different time granularities are resampled and generated. Then, the mean-variance normalization is performed independently for each granularity to obtain a multi-scale normalized voltage time series dataset.
[0035] In another alternative implementation, the preprocessing step may also consider using the multi-physical quantity sequences as multi-channel input sequences by vertically splicing or horizontally parallelizing them, and performing the same outlier processing, missing value imputation and normalization processing to form a multi-channel clean time series dataset.
[0036] In this embodiment of the application, step S200 involves inputting the preprocessed data into the TS-WCGAN-GP model for data augmentation to generate an enhanced voltage time series dataset, including the following steps A1-A4: A1: Input the preprocessed data into the spatiotemporal attention mechanism model, calculate the temporal feature weights of each historical voltage sequence and the spatial feature weights of each sampling point, and fuse them to generate spatiotemporal feature weights. It should be noted that, referring to Figure 2 In the spatiotemporal attention mechanism model (TS): For the set of attention heads, it can be derived from express; Each attention head contains a value object, and their collection constitutes the input value object. ; Each attention head has its own key-value object, and their collection constitutes the input key-value object. ; Each attention head contains a query object, and their set constitutes the input query object. .
[0037] The calculation process for the multi-head attention layer is shown in the following formula: ; ; ; In the formula, , , and Here are the weight matrices for each layer; This is a concatenation function; For the self-attention solution function; The activation function is used. The difference between temporal attention and spatial attention networks lies in the input. The quantity and value of latent features: The input latent feature quantity of the temporal attention network is the data of the key node voltage spatial dimension, and the input latent feature value is the data of the node voltage temporal dimension.
[0038] A2: Input the spatiotemporal feature weights into the conditional generative adversarial network model, and use the Wasserstein distance as a metric to measure the distribution difference between the generated sample data and the real sample data; It should be noted that neither the generator network G nor the discriminator network D in a GAN are expressed using explicit functions, and the choice of network structure offers considerable freedom and flexibility. Assuming the input real data is... The probability distribution that it satisfies is denoted as... The input noise sequence is defined as follows: This noise sequence typically originates from a simple existing distribution. The samples obtained are from uniform and Gaussian distributions.
[0039] Specifically, the input to the generator network G is a noise sequence Z, and the output is G(z), with the following distribution: The input to the discriminant network D is the original real data X and the generated data G(z), and the output is... and The parameters of the discriminant network D are defined as follows: The loss function is The parameters of the generating network G are: The loss function is The specific details are shown in the formula: In the formula, for Expectations; for The expectation. The expression aims to fix the parameters of the discriminant network. At that time, the probability that the discriminant network determines the data G(z) generated by the generator network to be true should be maximized, so that it can... Minimize as much as possible. The training objective of the discriminant network is to distinguish... and When the parameters of the generator network are fixed Update the discrimination network parameters The greater the difference in the discrimination, the stronger the discrimination network's judgment ability. When the discrimination network can distinguish the input samples well, its loss function LD is smaller.
[0040] The model training process constitutes a dynamic game, denoted as... In this game-theoretic process, as one network's loss function is optimized, the other network also updates its parameters. The generator network G and the adversarial network D are trained alternately, with one network's parameters fixed and the other's updated. This iterative process continues until a dynamic equilibrium, known as Nash equilibrium, is reached. The final training converges to obtain a pair of network parameters. These are two extreme points existing in the high-dimensional data space. The parameter update process of this generative adversarial network is an optimization problem of extreme values, and the objective function of the optimization is shown in the equation.
[0041] Conditional Generative Adversarial Networks (CGANs) generate highly realistic sample data through adversarial games between a generator network (G) and a discriminator network (D). CGANs, by introducing conditional variables (c) into the generator network (D) and the discriminator network (G) to constrain the features of the samples generated by the model, can effectively improve the model's ability to represent additional data features and increase the realism of the sample data generated by the model.
[0042] In the CGAN model, the prior input noise p(z) and conditional information The joint components form the joint hidden layer representation. The objective function of CGAN is an adversarial process of two minimaxes with conditional probabilities, as shown in the following equation.
[0043] In the formula, E(*) is the expected value of the distribution function.
[0044] CGAN uses JS divergence or KL divergence to measure the difference between two distributions. and When considering the distance between the generator and discriminator networks, it's difficult to properly manage their training levels during training, which can easily lead to deteriorating model performance. Furthermore, the training effectiveness and potential model collapse cannot be assessed through the losses of the generator and discriminator networks. To address these issues, this step employs the Wasserstein distance as a metric to measure the distributional difference between generated and real sample data, theoretically resolving the instability problem during CGAN training.
[0045] A3: Introduce gradient penalty to strengthen the Lipschitz constraint of the discriminant network in the adversarial network model, so that the loss function of the discriminant network satisfies Lipschitz continuity, and obtain the improved optimization objective function; It should be noted that, in order to satisfy the Lipschitz continuity condition, certain constraints need to be imposed on the discriminant network during training. (Function) The Lipschitz continuity refers to the existence of a constant under constraints. For two points within the domain of the function and The following relationship must be satisfied: When a function satisfies the K-Lipschitz continuity condition, for the function In other words, the magnitude of the function's local variation can be limited. To strengthen the Lipschitz continuity condition, weight clipping can be used to limit all parameters W to a certain range [-c, c]. If the weights of the discriminant network exceed this range during training, they are adjusted. At this point, the derivative with respect to the input sample x... It will not exceed a certain range. Therefore, there will always exist some unknown constant K such that... The local variation amplitude is limited to a certain range.
[0046] Given that WGAN employs a weight clipping strategy, the network weights tend to concentrate at two clipping boundaries during training. This can lead to slow training or even failure to converge. For example, setting the clipping boundaries too large or too small can cause gradient explosion or vanishing gradient problems. Therefore, the weight clipping strategy may make model optimization difficult, resulting in training instability in the WGAN model. Based on optimal transport theory, when the model is trained to its optimal state, the discriminant network G in the generation distribution... and data distribution The sampling points have a unit gradient norm.
[0047] Therefore, based on the output and input of the discriminant network, a gradient penalty (GP) can be added, as shown in the following equation: In the formula, For real samples With generated samples Random sampling on the connection line.
[0048] Considering that strengthening gradient norm constraints everywhere is difficult, this approach is an effective strategy for improving model training. After adding the gradient term GP, the loss function... It can be re-expressed using an equation: Specifically, by strengthening the Lipschitz constraint with a gradient penalty term (GP), the loss function of the discriminant network satisfies Lipschitz continuity. The distance metric function of the improved CWGAN-GP is shown in the following equation: In the formula, The constant representing the Wasserstein distance function; This indicates that the discriminant network function satisfies the constant. The Lipschitz continuity condition.
[0049] Specifically, the optimization objective function of the final improved CGAN-GP is shown in the equation.
[0050] A4: Generate an enhanced voltage time series dataset by alternately optimizing the generator and discriminator until Nash equilibrium is reached.
[0051] Furthermore, the network structure of the aforementioned TS-WCGAN-GP algorithm includes a generator network G and a discriminator network D, with the input data of the generator network G being condition variables. The input data of the adversarial network D consists of a randomly sampled random noise sequence z, where the condition variable c is the real-time acquired node voltage data; the input data of the adversarial network D is the condition variable. Generate sample data and real sample data The network structures for the generative model G and the discriminative model D are designed, and their specific structures are as follows: Figures 3-4 As shown.
[0052] from Figure 3 As can be seen, the generative model G contains a three-layer network structure: a feature layer, an intermediate layer, and an output layer. The feature layer consists of a temporal attention network and a spatial attention network, and its structure is defined as a TS module. The intermediate layer consists of four stacked layers: a 2D convolutional network Conv2D, a 2D deconvolutional network ConvTranspose2D, BN, and ReLU6. The output layer consists of an average pooling layer AvgPool2d, a linear layer Linear, BN, and a sigmoid activation function.
[0053] from Figure 4 As can be seen, the discriminant model D contains three network layers: a feature layer, an intermediate layer, and an output layer. The feature layer structure is the same as that of the discriminant model D, consisting of a temporal attention network and a spatial attention network. The intermediate layer is composed of four stacked layers: 2D convolutional layers (Conv2D), batch normalization (BN), and the ReLU6 activation function. The output layer consists of a linear layer (Linear), BN, and the sigmoid activation function.
[0054] In this embodiment of the application, step S300 calculates the DTW distance and Euclidean distance of each voltage sequence based on the obtained enhanced voltage time series dataset to obtain a comprehensive distance index, including: The DTW distance and Euclidean distance of each normalized voltage sequence are added together to obtain the comprehensive distance index.
[0055] Specifically, the DTW distance and Euclidean distance (ED) of each voltage sequence are calculated to obtain a comprehensive distance judgment index. The calculation formula is shown below: In the formula, d represents the dimension of each input element. for In the distance matrix, the first Distance elements corresponding to each coordinate; This represents the number of input voltage sequences.
[0056] Will , Normalize the results and then add them together to get the result. The calculation formula is as follows: In the formula, Distance The mean; Distance The variance.
[0057] In an optional implementation, the comprehensive distance index can also be obtained by weighted fusion combined with adaptive weights. For example, the weight coefficients are adaptively adjusted according to the actual characteristics of the transformer area. It can be automatically adjusted based on empirical values or the standard deviation of daily voltage fluctuations: the greater the fluctuation, The larger.
[0058] In another optional implementation, if multi-scale preprocessing is used in S100, the comprehensive distance index can be calculated at different time scales for DTW and Euclidean distance, and then weighted and merged.
[0059] S400: Based on the comprehensive distance index, the improved adaptive density peak clustering algorithm automatically determines the number of transformer substations and the voltage sequence corresponding to the cluster centers, which is used for transformer substation affiliation and household phase identification. Before explaining the S400, it should be noted that one... The voltage sequence curve of dimension 1 is a voltage sequence curve containing 1 dimension 2. The data points of the data, the curve dimension refers to the number of sampling periods of the curve. For example, a curve with a sampling interval of 1 hour has 24 sampling periods, that is, its dimension is 24; thus, two variables are introduced: local density and relative distance.
[0060] In this embodiment of the application, step S400 automatically determines the number of transformer areas and the voltage sequence corresponding to the corresponding cluster centers based on the comprehensive distance index using an improved adaptive density peak clustering algorithm, including the following steps B1-B2: B1: Calculate the local density and relative distance of each voltage sequence based on the comprehensive distance; Specifically, in B1, the first Local density (truncated kernel) of voltage sequence curves The definition is as follows: In the formula, Indicates the first The data point and the A comprehensive distance index for each data point The cutoff distance, This is a judgment function.
[0061] For relatively large datasets, the Gaussian kernel is used to calculate the local density, and its definition is as follows: Cut-off distance The definition is as follows: In the formula, The total number of data points in the dataset. This is a percentage parameter, typically ranging from 1 to 10. This indicates that the dataset is sorted in ascending order based on the combined distances between all data points. This indicates rounding to the nearest integer, meaning that after sorting all composite distances in ascending order, the integer part is taken as the nearest integer. The distance value is used as the cutoff distance.
[0062] Specifically, in B1, the first relative distance of data points The definition is as follows: In the formula, Indicates other data points up to the th The maximum distance between data points Indicates other data points up to the th The minimum distance between data points Indicates the first The maximum local density of the nth data point. When the nth data point... When the local density of a data point is maximized, it is likely to be a cluster center, and its relative distance to other data points is equal to the distance from the i-th data point to the j-th data point. The maximum distance between the nth data points; when the nth data point When the local density of a data point is not at its maximum, its relative distance is equal to the distance from the data point to all data points with local densities greater than the distance from the data point.
[0063] B2: Select points that simultaneously satisfy the condition that the local density is greater than the first threshold and the relative distance is greater than the second threshold as the initial cluster center candidate set.
[0064] Specifically, the method for determining the initial cluster centers of B2 is as follows: In the formula, It is the set of curve data points where the local density is greater than the average local density. It is a set of curve data points whose relative distance is greater than the average relative distance. These are data points on the curve. It is its corresponding local density. It is its corresponding relative distance. It is the average of the local density of all data points. It is the average of the relative distances of all data points. It is the initial cluster center set.
[0065] It should be noted that, in order to avoid noise points with low local density and large relative distance, and boundary points with high local density and small relative distance being selected as cluster centers, two thresholds are defined by B2 to improve the accuracy of the initial cluster centers.
[0066] In this embodiment of the application, step S400 automatically determines the number of transformer areas and the voltage sequence corresponding to the corresponding cluster centers based on the comprehensive distance index using an improved adaptive density peak clustering algorithm, and also includes the following steps B3-B5: B3: Run basic density peak clustering, calculate the initial cluster center set, and obtain the number of substations represented by the number of clusters; B4: Cluster the data based on the initial cluster centers and calculate the cluster boundary density and intersection density; B5: Iterative merging based on cluster boundary density and intersection density, outputting the final number of substations, corresponding cluster centers, and corresponding voltage sequences.
[0067] Specifically, in step B5, the cluster center numbers can be sorted in ascending order. It is determined whether the intersection density is greater than the boundary density. If so, the process returns to step B4 and updates the cluster centers. It is also determined whether the total number of points has been reached. If not, the boundary density and intersection density are modified in ascending order of density, and the process returns to step B5. Finally, the core point set and cluster set are output.
[0068] It should be noted that the DTW distance between two voltage sequences is inversely proportional to their similarity; the smaller the DTW distance, the higher the similarity. Therefore, by calculating the DTW distance between each sequence and performing data clustering, the identification of transformer substations and individual households can be achieved.
[0069] Specifically, traditional clustering methods like K-means require pre-setting or estimating K, i.e., the number of transformer substations. However, in practical applications, this cannot provide a reliable K value. Therefore, by using B3 to quickly obtain a large number of initial centers, and then using the iterative merging mechanism of B4-B5, multiple candidate centers belonging to the same transformer substation are automatically merged. The final output K value is completely consistent with the actual number of transformer substations on site, completely eliminating the reliance on manual intervention. The voltage sequence corresponding to each retained cluster center is the curve that most closely resembles the typical operating pattern within that transformer substation, providing a high signal-to-noise ratio comparison template for subsequent comparisons, improving the accuracy of transformer substation attribution identification and the stability of household phase identification.
[0070] In this embodiment of the application, step S500 calculates the DTW distance between the preprocessed historical data sequences and the cluster centers obtained by adaptive clustering, performs data clustering, and obtains the substation affiliation of the voltage sequences, including the following steps C1-C6: C1: Confirm the preprocessed historical data sequence and the number of distribution transformers, K; C2: Randomly select one voltage sequence DATA from the historical data sequence as the first initial cluster point; C3: Calculate the DTW distance between the voltage sequences of all users and the cluster centers obtained from adaptive clustering. The probability of each voltage sequence being selected as the next location is denoted as... The voltage sequence with the highest probability value is selected as the next aggregation point; C4: Repeat the calculation until K cluster points are selected; C5: Cluster all voltage sequences with the DTW values of each cluster point, and assign the voltage sequence with the smallest DTW value with each cluster point to the corresponding cluster point's area, thus dividing all voltage sequences into K sets. C6: Voltage sequences contained in the same set belong to the same transformer area.
[0071] It should be noted that within the same distribution area, the differences between voltage sequences of different phases are greater than those of voltage sequences of the same phase; furthermore, this step directly reuses the S400 adaptive cluster center, without requiring any distribution transformer quantity or household transformer relationship files provided by the marketing system, thus avoiding systematic bias caused by file errors; using the nearest center DTW distance principle, all user divisions can be completed without iterating through the entire distribution area once.
[0072] Therefore, by further comparing the DTW distance between each voltage sequence and the three-phase voltage on the low-voltage side of the distribution transformer in its respective area, the voltage phase of the user's line can be identified. The specific implementation steps are as follows: In this embodiment of the application, step S600 achieves phase identification within the distribution area by comparing the DTW distance between each voltage sequence and the three-phase voltage on the low-voltage side of the distribution transformer of its respective area, including the following steps D1-D3: D1: Calculate the DTW distance between the voltage sequence of each user's meter and the three-phase voltage sequence on the low-voltage side of the distribution transformer within the transformer area; D2: By comparison, the user with the smallest distance from the voltage sequence DTW of phase A on the low-voltage side of the distribution transformer is classified as phase A; D3: Repeat the comparison until all users in phases B and C have been identified.
[0073] It should be noted that the S600 only needs to use the existing three-phase voltmeters on the low-voltage side of the distribution transformer and the single-phase voltage data of ordinary smart meters to achieve phase identification of households throughout the entire distribution area, without the need for hardware modification.
[0074] Overall, this invention, based on limited data information, can accurately expand the data information while maintaining the original spatiotemporal characteristics of the data through data augmentation, and accurately simulate the operating characteristics such as voltage of key nodes. The clustering algorithm combined with the comprehensive distance index can adaptively calculate the cluster center based on the input data information when the number of distribution transformers is unknown, thereby obtaining the number of transformer areas and the characteristics of distribution transformer data information in the transformer area. At the same time, by using DTW distance to calculate the similarity of node voltage, missing values and garbled values in the dataset can be ignored. Combined with the proposed clustering algorithm, topology identification has a higher recognition accuracy.
[0075] Example 3, referring to Figures 5-9 According to Tables 1 and 2, this embodiment provides an application example of a low-voltage distribution network topology identification and verification method based on low-sensitivity, in order to verify the feasibility and beneficial effects of the present invention.
[0076] S1: Seven distribution transformer areas in a certain region were selected for verification, totaling 120 user data points. The distribution transformer topology is as follows: Figure 5 As shown.
[0077] S2: Perform data augmentation using limited user data, and verify the accuracy of the data augmentation method. The indicators include two aspects: volatility and spatiotemporal correlation, to analyze the augmented dataset.
[0078] (1) Volatility characteristics index The ramp-up rate is used to evaluate the short-term fluctuations in the output of distributed photovoltaic clusters, as defined in the following equation: In the formula, , They are respectively Time and Voltage observations / predictions at any given time; This represents the maximum voltage value.
[0079] (2) Spatiotemporal correlation indicators The Pearson correlation coefficient is used to measure the characterization of spatial correlation between different transformer substations. The specific formula is shown in the figure below: In the formula, Pearson correlation coefficient; and These represent the mean values of voltage sequences X and Y for different transformer substations, respectively.
[0080] Heatmaps of real data and heatmaps of augmented data Figure 6 and Figure 7 As shown, the spatiotemporal distribution characteristics of the data augmented using the TS-WCGAN-GP algorithm are very close to those of real data, possessing the spatiotemporal characteristics of real data.
[0081] The fluctuation values of real data and the fluctuation distribution of data augmentation are as follows: Figures 8-9 As shown, the data fluctuations after data augmentation are quite close to the real data.
[0082] S3: The proposed clustering algorithm was used to identify user areas and households within those areas. The results are shown in Tables 1 and 2.
[0083] Table 1: Transit Area Identification Results
[0084] Table 2: Household Identification Results
[0085] It can be seen that the transformer substation topology identification method proposed in this invention can effectively identify the topology of low-voltage distribution substations, and the adaptive clustering center algorithm can effectively determine the number of distribution transformers and voltage patterns in the corresponding substations.
[0086] Example 4 illustrates a schematic scheme for a low-voltage distribution network topology identification and verification method based on low-sensitivity. It should be noted that the technical solution of this low-voltage distribution network topology identification and verification system based on low-sensitivity is based on the same concept as the aforementioned low-voltage distribution network topology identification and verification method. Details not described in detail in this example of the low-voltage distribution network topology identification and verification system based on low-sensitivity can be found in the description of the aforementioned low-voltage distribution network topology identification and verification method.
[0087] This embodiment also provides another low-voltage distribution network topology identification and verification system based on low-sensitivity, including: The preprocessing module is used to acquire historical data sequences of observable nodes in the distribution network and perform data preprocessing. The data augmentation module is used to input the preprocessed data into the TS-WCGAN-GP model for data augmentation and to generate an enhanced voltage time series dataset. The distance calculation module is used to calculate the DTW distance and Euclidean distance of each voltage sequence based on the obtained enhanced voltage time series dataset, and obtain a comprehensive distance index. The first clustering module is used to automatically determine the number of transformer substations and the voltage sequence corresponding to the K cluster centers based on the comprehensive distance index and an improved adaptive density peak clustering algorithm, for substation affiliation and household phase identification. The second clustering module is used to calculate the DTW distance between the preprocessed historical data sequences and the cluster centers obtained by adaptive clustering, and to perform data clustering to obtain the substation affiliation of the voltage sequences. The comparison and identification module is used to identify the phase of each household within the distribution area by comparing the DTW distance between each voltage sequence and the three-phase voltage on the low-voltage side of the distribution transformer in the area to which it belongs.
[0088] This embodiment also provides a computer device applicable to a low-voltage distribution network topology identification and verification method based on low-sensitivity, comprising: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the low-voltage distribution network topology identification and verification method based on low-sensitivity proposed in the above embodiment.
[0089] This embodiment also provides a storage medium storing a computer program that, when executed by a processor, implements a method for identifying and verifying the topology of a low-voltage distribution network based on low-sensitivity, as proposed in the above embodiments.
[0090] The storage medium proposed in this embodiment and the method for implementing a low-voltage distribution network topology identification and verification method based on low-sensitivity proposed in the above embodiments belong to the same inventive concept. Technical details not described in detail in this embodiment can be found in the above embodiments, and this embodiment has the same beneficial effects as the above embodiments.
[0091] From the above description of the implementation methods, those skilled in the art will clearly understand that the present invention can be implemented using software and necessary general-purpose hardware. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as a computer floppy disk, read-only memory (ROM), random access memory (RAM), flash memory, hard disk, or optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods of the various embodiments of the present invention.
[0092] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not 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, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.
Claims
1. A low-perception-based low-voltage distribution network topology identification and verification method for a distribution area, characterized in that, The method comprises the following steps: Obtain the historical data sequence of the observable node data in the power distribution network and perform data preprocessing; Input the preprocessed data into the TS-WCGAN-GP model to perform data enhancement and generate an enhanced voltage time series data set; Based on the obtained enhanced voltage time series data set, calculate the DTW distance and Euclidean distance of each voltage sequence to obtain a comprehensive distance index; Based on the comprehensive distance index, automatically determine the number of transformer areas and the voltage sequences corresponding to the respective clustering centers by using an improved adaptive density peak clustering algorithm, and use the algorithm for transformer area attribution and phase identification; Calculate the DTW distance between the preprocessed historical data sequence and the clustering center obtained by adaptive clustering, perform data clustering, and obtain the transformer area attribution of the voltage sequence; By comparing the DTW distance between each voltage sequence and the low-voltage side three-phase voltage of the distribution transformer in the corresponding transformer area, the phase identification in the transformer area is realized.
2. The low-awareness-based low-voltage distribution network topology identification and checking method according to claim 1, characterized in that, The preprocessed data is input into the TS-WCGAN-GP model to perform data enhancement and generate an enhanced voltage time series data set, which comprises: Input the preprocessed data into the spatio-temporal attention mechanism model to calculate the time feature weight of each historical voltage sequence and the spatial feature weight at each sampling point, and fuse to generate a spatio-temporal feature weight; Input the spatio-temporal feature weight into the conditional generative adversarial network model, and use the Wasserstein distance as the measurement standard to measure the distribution difference between the generated sample data and the real sample data; Introduce gradient penalty to strengthen the Lipschitz constraint of the discriminator network in the adversarial network model, so that the loss function of the discriminator network satisfies the Lipschitz continuity, and obtain an improved optimization objective function; By alternately optimizing the generator and the discriminator until the Nash equilibrium is reached, an enhanced voltage time series data set is generated.
3. The low-awareness-based low-voltage distribution network topology identification and checking method according to claim 2, characterized in that, Based on the comprehensive distance index, the number of transformer areas and the voltage sequences corresponding to the respective clustering centers are automatically determined by using an improved adaptive density peak clustering algorithm, which comprises: Based on the comprehensive distance, the local density and the relative distance of each voltage sequence are calculated; Select the points that simultaneously satisfy the local density greater than the first threshold value and the relative distance greater than the second threshold value as the initial clustering center candidate set.
4. The low-awareness-based low-voltage distribution network topology identification and checking method according to claim 3, characterized in that, Based on the comprehensive distance index, the number of transformer areas and the voltage sequences corresponding to the respective clustering centers are automatically determined by using an improved adaptive density peak clustering algorithm, which further comprises: Run the basic density peak clustering to calculate the initial clustering center set and obtain the number of transformer areas represented by the clustering number; According to the initial clustering center, perform data clustering to calculate the clustering boundary density and the intersection density; Based on the clustering boundary density and the intersection density, perform iterative merging to output the final number of transformer areas, the corresponding clustering centers and the corresponding voltage sequences.
5. The low-awareness-based low-voltage distribution network topology identification and checking method according to claim 4, characterized in that, The DTW distance between the preprocessed historical data sequence and the clustering center obtained by adaptive clustering is calculated, the data is clustered, and the transformer area attribution of the voltage sequence is obtained, which comprises: Confirm the preprocessed historical data sequence and the number of distribution transformers K; Randomly select one voltage sequence DATA from the historical data sequence as the first initial clustering point; The DTW distance between the voltage sequence of all users and the clustering center obtained by adaptive clustering is calculated respectively, and the probability of each voltage sequence being selected as the next cluster center is recorded as The voltage sequence with the maximum probability value is selected as the next cluster center. Repeat the calculation until K clustering points are selected; All voltage sequences are clustered with DTW values between each cluster point, and the voltage sequence with the minimum DTW value of each cluster point is assigned to the corresponding cluster area where the cluster point is located, and all voltage sequences are divided into K sets; The voltage sequences contained in the same set belong to the user voltage sequences of the same transformer area.
6. The low-awareness-based low-voltage distribution network topology identification and checking method according to claim 5, characterized in that, By comparing the DTW distance of each voltage sequence and the low-voltage side three-phase voltage of the corresponding transformer area, the house phase recognition in the transformer area is realized, including: The DTW distance of each user meter voltage sequence and the low-voltage side three-phase voltage sequence in the transformer area is calculated respectively; By comparison, the user with the minimum DTW distance of the low-voltage side A-phase voltage sequence is assigned to A-phase; Repeat the comparison until all users of B-phase and C-phase are identified.
7. The low-awareness-based low-voltage distribution network topology identification and checking method according to claim 6, characterized in that, Based on the obtained enhanced voltage time series data set, the DTW distance and Euclidean distance of each voltage sequence are calculated to obtain a comprehensive distance index, including: The DTW distance and Euclidean distance of each normalized voltage sequence are added to obtain a comprehensive distance index.
8. A low-voltage distribution network topology identification and verification system based on low-awareness, applying the method of any one of claims 1-7, characterized in that, It includes: The preprocessing module is used to obtain the historical data sequence of the observable node data in the power distribution network and perform data preprocessing; The data enhancement module is used to input the preprocessed data into the TS-WCGAN-GP model to perform data enhancement and generate an enhanced voltage time series data set; The distance calculation module is used to calculate the DTW distance and Euclidean distance of each voltage sequence based on the obtained enhanced voltage time series data set to obtain a comprehensive distance index; The first clustering module is used to automatically determine the number of transformer areas and the voltage sequences corresponding to the K clustering centers based on the comprehensive distance index through an improved adaptive density peak clustering algorithm, which is used for transformer area attribution and house phase recognition; The second clustering module is used to calculate the DTW distance between the preprocessed historical data sequence and the clustering center obtained by adaptive clustering, and perform data clustering to obtain the transformer area attribution of the voltage sequence; The comparison and recognition module is used to realize the house phase recognition in the transformer area by comparing the DTW distance of each voltage sequence and the low-voltage side three-phase voltage of the corresponding transformer area.
9. A computer device, comprising: It includes: Memory and processor; The memory is used to store computer executable instructions, and the processor is used to execute the computer executable instructions, which realize the steps of the low-sensing low-voltage transformer area power distribution network topology identification and verification method according to any one of claims 1 to 7.
10. A computer-readable storage medium, characterized in that, It stores computer executable instructions, which realize the steps of the low-sensing low-voltage transformer area power distribution network topology identification and verification method according to any one of claims 1 to 7.