Partition regulation and control method and device for receiving end circuit of power distribution network

Abstract

The invention discloses a partition regulation and control method and device for a receiving end circuit of a power distribution network, and relates to the field of power grid control, and the method comprises the steps: obtaining the real-time operation data of each node of the receiving end circuit of the power distribution network and the electrical association data between the nodes; calculating an electrical distance between the nodes according to the electrical association data; performing time sequence processing on the real-time operation data to obtain time sequence data; performing fast Fourier transform compression on the time sequence data of each node, and decomposing the time sequence data into a trend component and a seasonal component; inputting the trend component of each node, the seasonal component of each node and the electrical distance between the nodes into a feature extraction model to be converted into an embedded vector, and fusing to obtain a high-dimensional feature vector; and according to the high-dimensional feature vectors of all the nodes, determining a target partitioning strategy so as to perform partitioning regulation and control on a receiving end circuit. By implementing the invention, the problem that the control action lags behind the dynamic change of the power grid in the existing threshold triggering type scheduling can be solved, and the operation stability of the power grid is improved.

Description

Technical Field

[0001] This invention relates to the field of power grid control, and in particular to a method and apparatus for zoned control of the receiving-end circuit of a distribution network. Background Technology

[0002] Under the "dual carbon" background, the distribution network is transforming into an interactive network integrating power sources, grids, loads and energy storage. Urban distribution networks, especially high-load receiving-end circuits, have been connected to a large number of distributed new energy sources and flexible loads with strong randomness, resulting in frequent voltage fluctuations and over-limit occurrences, increased operating costs, widening power peak-valley differences, and threats to system stability.

[0003] Faced with the aforementioned complex operating conditions, existing threshold-triggered scheduling often only performs zone adjustments after voltage exceedances occur, resulting in control actions lagging significantly behind the dynamic changes, and in severe cases, even threatening the safe operation of the power system. Summary of the Invention

[0004] This invention provides a method and apparatus for zoned control of the receiving-end circuit of a distribution network, which can solve the problem that the control action lags behind the dynamic changes of the power grid in the existing threshold-triggered scheduling, and improve the stability of power grid operation.

[0005] An embodiment of the present invention provides a method for zoned control of the receiving-end circuit of a distribution network, comprising: Acquire real-time operating data of each node in the receiving-end circuit of the distribution network and electrical correlation data between nodes; electrical correlation data includes: line impedance or physical length; Calculate the electrical distance between nodes based on the electrical correlation data between nodes; Perform time-series processing on real-time running data to obtain corresponding time-series data; For each node, the time series data is compressed by Fast Fourier Transform to obtain compressed data, and the compressed data is decomposed into trend components and seasonal components according to the preset decomposition frequency. The trend component, seasonal component, and electrical distance between each node are input into a pre-defined feature extraction model, which transforms the trend component, seasonal component, and electrical distance into embedding vectors to obtain trend feature vectors, seasonal feature vectors, and electrical distance vectors. The trend feature vectors, seasonal feature vectors, and electrical distance vectors are then fused to obtain high-dimensional feature vectors. The target partitioning strategy is determined based on the high-dimensional feature vectors of all nodes; Based on the target zoning strategy, the receiving-end circuits of the distribution network are zoned and regulated.

[0006] Furthermore, after acquiring the real-time operating data of each node in the receiving-end circuit of the distribution network and the electrical correlation data between the nodes, it also includes: The real-time running data of each node is denoised, missing values ​​are filled in, and normalized to obtain the preprocessed real-time running data of each node.

[0007] Furthermore, the time-series data is compressed using a Fast Fourier Transform to obtain compressed data, including: Perform a Fast Fourier Transform on the time-series data to obtain the frequency domain data; Based on the preset cutoff frequency, the frequency domain data is filtered to obtain the filtered frequency domain data; The filtered frequency domain data is subjected to inverse fast Fourier transform to obtain compressed data.

[0008] Furthermore, the feature extraction model is determined in the following way: Acquire historical operating sequence data of each node in the receiving-end circuit of the distribution network and the historical electrical distance between the corresponding nodes; The historical runtime sequence data of each node is compressed using Fast Fourier Transform to obtain the corresponding historical compressed data. Based on the preset decomposition frequency, the historical compressed data is decomposed into trend-based historical components and seasonal historical components. The trend-based historical components of each node, the seasonal historical components of each node, and the historical electrical distance between nodes are input into the feature embedding reconstruction model to be trained, so that the feature embedding reconstruction model takes the trend-based historical components, seasonal historical components, and historical electrical distance as inputs and the trend-based reconstruction components, seasonal reconstruction components, and reconstruction electrical distance as outputs for iterative training. In each training process, the built-in encoding layer transforms the trend historical components, seasonal historical components, and historical electrical distance into embedding vectors, resulting in trend historical feature vectors, seasonal historical feature vectors, and historical electrical distance vectors. The trend historical feature vectors, seasonal historical feature vectors, and historical electrical distance vectors are then fused to obtain high-dimensional historical feature vectors. Calculate the trend similarity between different nodes based on the historical trend feature vector; calculate the seasonal similarity between different nodes based on the historical seasonal feature vector. The high-dimensional historical feature vector is decoded and reconstructed through the built-in decoding layer to obtain the trend reconstruction component, the seasonal reconstruction component, and the reconstruction electrical distance. The reconstruction error is calculated by combining the trend historical component, the seasonal historical component, and the historical electrical distance. Based on trend similarity, seasonal similarity, and reconstruction error, the joint loss is determined; the model parameters of the feature extraction model are adjusted according to the joint loss until the joint loss converges, resulting in a trained feature embedding reconstruction model. The feature extraction model is determined based on the model parameters of the encoding layer in the feature embedding reconstruction model.

[0009] Furthermore, based on the high-dimensional feature vectors of all nodes, the target partitioning strategy is determined, including: Clustering is performed on the high-dimensional feature vectors of all nodes to obtain several clustering partitions; The global comprehensive performance index at the current moment is determined based on each cluster partition and the electrical distance between nodes; Obtain the global comprehensive performance index of the previous moment; If the absolute difference between the global comprehensive performance index at the current moment and the global comprehensive performance index at the previous moment is greater than a preset dynamic threshold, partition optimization is performed on several clustering partitions to obtain the target partitioning strategy.

[0010] Furthermore, based on each cluster partition and the electrical distance between nodes, the global comprehensive performance index at the current moment is determined, including: Based on the electrical distance between nodes, determine the edge weights between nodes, the weighted degree of nodes, and the total edge weight of the entire network; The global modularity index is determined based on the clustering partitioning, edge weights between nodes, weighted degree of nodes, and total edge weight of the entire network; the clustering partitioning is indicated by an indicator function. To characterize, among which, Represents a node Cluster partition label, Represents a node The cluster partition label to which the node belongs, when the node and nodes When they belong to the same cluster partition, the indicator function Otherwise, the indicator function ; Obtain the adjustable power potential of each node in each cluster partition; The global voltage regulation capability index is determined based on the adjustable power potential of each node in each cluster partition. The global membership stability index is determined based on the proportion of edge weights within cluster partitions. The global modularity index, global voltage regulation capability index, and global membership stability index are weighted and summed to obtain the global comprehensive performance index at the current moment.

[0011] Furthermore, partition optimization is performed on several clustering partitions to obtain the target partitioning strategy, including: The particle positions are initialized based on the cluster centers in the cluster partition to obtain the initial particle positions. The particle velocity, inertia weight, individual learning factor and group learning factor are also initialized to obtain the initial particle velocity, initial inertia weight, initial individual learning factor and initial group learning factor of the corresponding particle. Repeat the partition optimization process to obtain the target partitioning strategy; The partition optimization process includes: Calculate the comprehensive objective function value based on the current particle position; If the overall objective function value converges, the target partitioning strategy is determined based on the global best position among the current particle positions. Otherwise, based on the current particle position, current particle velocity, current inertia weight, current individual learning factor, and current group learning factor, update the particle position and particle velocity to obtain the updated particle position and particle velocity; where the current particle position, current particle velocity, current inertia weight, current individual learning factor, and current group learning factor during the first execution of the partition optimization process are the initial particle position, initial particle velocity, initial inertia weight, initial individual learning factor, and initial group learning factor, respectively. The current inertia weight, current individual learning factor, and current group learning factor are updated non-linearly in decreasing order to obtain the updated inertia weight, updated individual learning factor, and updated group learning factor. The updated particle position, updated particle velocity, updated inertia weight, updated individual learning factor, and updated group learning factor are used as the current particle position, current particle velocity, current inertia weight, current individual learning factor, and current group learning factor for the next partition optimization process, respectively.

[0012] Furthermore, based on the target zoning strategy, the receiving-end circuits of the distribution network are subjected to zoning control, including: The receiving-end circuits of the distribution network are partitioned according to the target partitioning strategy; The voltage regulation capability of each zone is evaluated to obtain the corresponding voltage regulation margin; Based on the range corresponding to the voltage regulation margin, reactive power control is performed on each zone accordingly.

[0013] Based on the above method embodiments, the present invention provides corresponding device embodiments, including: a data acquisition module, an electrical distance calculation module, a timing processing module, a compression decomposition module, a feature extraction module, a partitioning strategy determination module, and a partitioning control module; The data acquisition module is used to acquire real-time operating data of each node in the receiving-end circuit of the distribution network and electrical correlation data between nodes; the electrical correlation data includes: line impedance or physical length; The electrical distance calculation module is used to calculate the electrical distance between nodes based on the electrical association data between nodes; The timing processing module is used to perform timing processing on real-time running data to obtain corresponding timing data. The compression and decomposition module is used to perform fast Fourier transform compression on time series data for each node to obtain compressed data, and decompose the compressed data into trend components and seasonal components according to the preset decomposition frequency. The feature extraction module is used to input the trend component, seasonal component, and electrical distance between each node into a preset feature extraction model, so that the feature extraction model can transform the trend component, seasonal component, and electrical distance into embedding vectors to obtain trend feature vector, seasonal feature vector, and electrical distance vector; and fuse the trend feature vector, seasonal feature vector, and electrical distance vector to obtain a high-dimensional feature vector. The partitioning strategy determination module is used to determine the target partitioning strategy based on the high-dimensional feature vectors of all nodes. The zone control module is used to perform zone control on the receiving-end circuits of the distribution network according to the target zone strategy.

[0014] Furthermore, the zoning control device for the receiving-end circuit of the distribution network also includes: a preprocessing module; The preprocessing module is used to denoise, fill in missing values, and normalize the real-time running data of each node to obtain the preprocessed real-time running data of each node.

[0015] Compared with the prior art, the beneficial effects of this embodiment are as follows: This invention first acquires real-time operational data and electrical correlation data between nodes of the receiving-end circuit of the distribution network. Based on the electrical correlation data, it calculates the electrical distance between nodes and quantifies the electrical coupling strength between nodes, providing a structural basis for zoning. The real-time operational data undergoes time-series processing to form continuous time-series data. For each node, the time-series data is compressed using a Fast Fourier Transform to reduce redundant information. The compressed data is decomposed into trend components and seasonal components, thereby extracting long-term trends and intraday / seasonal fluctuation features respectively. A feature extraction model transforms the trend components, seasonal components, and electrical distance into embedded vectors, encoding them into a unified vector space. The embedded vectors are fused to obtain high-dimensional feature vectors, achieving an accurate characterization of the dynamic coupling relationship between power grid nodes and providing a data foundation for the rationality of zoning. Finally, based on the high-dimensional feature vectors of all nodes, a target zoning strategy is determined, and then, based on the target zoning strategy, the receiving-end circuit of the distribution network is zoned and regulated.

[0016] In summary, this invention performs trend decomposition and seasonal decomposition on power grid operation data, extracts long-term trends and intraday and seasonal fluctuation features respectively, and establishes high-dimensional feature vectors by combining the electrical distance between nodes, providing a data foundation for the rationality of zoning. This solves the problem of control actions lagging behind the dynamic changes of the power grid in the existing threshold-triggered scheduling, and improves the stability of power grid operation. Attached Figure Description

[0017] Figure 1 This is a schematic flowchart of a method for zoned control of a distribution network receiving-end circuit according to an embodiment of the present invention; Figure 2 This is a schematic diagram of the structure of a zone control device for the receiving end circuit of a power distribution network provided in an embodiment of the present invention. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] In the description of this invention, it should be understood that the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated.

[0020] like Figure 1 As shown, in order to solve the problem that control actions lag behind the dynamic changes of the power grid in existing threshold-triggered scheduling, an embodiment of the present invention provides a method for zoned control of the receiving-end circuit of a distribution network, which includes at least the following steps: Step S1: Obtain real-time operating data of each node in the receiving-end circuit of the distribution network and electrical correlation data between nodes; electrical correlation data includes: line impedance or physical length; For step S1, for the receiving-end circuit of the distribution network, this embodiment corresponds to the urban ultra-high load receiving-end power grid. Real-time operating data of each node and electrical correlation data between nodes are collected from the monitoring equipment of the receiving-end power grid.

[0021] The aforementioned real-time operating data includes: grid terminal voltage, return current, active power, reactive power, load curve, distributed photovoltaic output, energy storage device output, adjustable power of flexible load, reactive power compensation equipment parameters, active voltage sensitivity data, and reactive voltage sensitivity data. The aforementioned electrical correlation data includes: line impedance or physical length; in addition, the electrical correlation data also includes: the weight of the connection edges between nodes.

[0022] In a preferred embodiment, after acquiring the real-time operating data of each node in the receiving-end circuit of the distribution network and the electrical correlation data between the nodes, the method further includes: The real-time running data of each node is denoised, missing values ​​are filled in, and normalized to obtain the preprocessed real-time running data of each node.

[0023] Specifically, in order to clearly identify the receiving-end circuits of the distribution network and facilitate accurate analysis and efficient management of the node operation status, each node of the receiving-end circuit is uniformly numbered. Based on the node number and its corresponding attribute information, the nodes are labeled. For the operation attributes of a single node, it can be labeled as the terminal voltage of node i, etc. For the electrical association attributes between nodes, it can be labeled as the line impedance of node ij, etc.

[0024] Next, the real-time operating data of each node is preprocessed. Since the monitoring data of the receiving-end power grid is susceptible to electromagnetic interference, temporary equipment fluctuations, and other factors, abnormal noise is introduced. These non-realistic fluctuations are filtered out using methods such as filtering to achieve noise reduction and ensure the authenticity of operating data such as voltage and power. For data gaps caused by equipment failures or communication delays during monitoring, reasonable methods such as interpolation of synchronous data from adjacent nodes and fitting of similar load data are used to fill the gaps, ensuring missing values ​​are filled and preventing data gaps from affecting subsequent analysis. Finally, considering the significant differences in units and numerical ranges among different operating data, these data are uniformly mapped to the same numerical range to achieve normalization, eliminating weight bias caused by differences in units. Ultimately, the preprocessed real-time operating data of each node is obtained.

[0025] It should be noted that the preprocessed real-time running data can be used as the basis for partitioning in the future.

[0026] Step S2: Calculate the electrical distance between nodes based on the electrical correlation data between nodes; For step S2, the electrical distance between nodes is calculated using the following formula based on the electrical association data between nodes: ; in, Represents a node With nodes The electrical distance between them, with a value ranging from [0,1]. Represents a node With nodes Electrical correlation data between them, such as line impedance or physical length, This represents the maximum value of the line impedance between all nodes, or the maximum value of the physical length between all nodes. Normalization eliminates the differences in the magnitude of different line parameters, enabling an objective measurement of the tightness of electrical coupling between nodes.

[0027] Step S3: Perform time-series processing on the real-time running data to obtain the corresponding time-series data; For step S3, in order to provide suitable input data for subsequent Fast Fourier Transform (FFT) frequency domain analysis and to accurately extract the frequency characteristics of each node in the distribution network receiving-end circuit, the data is sorted according to the chronological order of real-time operation data acquisition to construct a continuous time series and obtain the corresponding time series data. ,in, This indicates the time when the operating data of each node in the receiving-end circuit of the distribution network is collected.

[0028] Step S4: For each node, perform Fast Fourier Transform compression on the time series data to obtain compressed data, and decompose the compressed data into trend components and seasonal components according to the preset decomposition frequency. In a preferred embodiment, the time-series data is compressed using a Fast Fourier Transform to obtain compressed data, including: Perform a Fast Fourier Transform on the time-series data to obtain the frequency domain data; Based on the preset cutoff frequency, the frequency domain data is filtered to obtain the filtered frequency domain data; The filtered frequency domain data is subjected to inverse fast Fourier transform to obtain compressed data.

[0029] For step S4, firstly, for any time-series signal, its decomposition form is: ; in, Represents time series data, This represents the long-term trend, or trend component, and is used to reflect the long-term changing patterns of the data. This indicates a seasonal component, used to reflect the characteristic of data fluctuating over a fixed period. This represents the residual component, which indicates the irregular, random fluctuations in the data.

[0030] Next, for time-series data in the time domain The data is then converted to frequency domain data using a Fast Fourier Transform, as shown in the following formula: ; Based on the preset cutoff frequency, the frequency domain data is filtered using the following formula to obtain the filtered frequency domain data: ; Perform an inverse fast Fourier transform on the filtered frequency domain data to convert it back to the time domain, resulting in compressed data. ; in, This represents the frequency domain data after the time series data has undergone a Fast Fourier Transform. Represents the Fast Fourier Transform. This indicates the cutoff frequency, which, by limiting the frequency range, retains only low-frequency components to reflect intraday and seasonal characteristics. This represents the filtered frequency domain data. This represents the inverse fast Fourier transform. This indicates compressed data.

[0031] The FFT compression described above can filter out irregular residual components in time series data. While reducing redundant information, it retains key dynamic characteristics related to voltage control, namely trend components. and seasonal ingredients .

[0032] It should be noted that the trend component This refers to the long-term, slow-changing characteristics of time-series data, corresponding to low-frequency bands, while seasonal components... It refers to the characteristic of time-series data fluctuating with a fixed period (such as daily or weekly), corresponding to a mid-frequency band that matches the period, and the pre-set decomposition frequency. A frequency domain threshold is used to distinguish between these two types of frequencies, and the frequency is decomposed. Requires a frequency less than the cutoff frequency .

[0033] Based on decomposition frequency A second frequency filtering is performed on the filtered frequency domain data to... The low-frequency components are retained as the frequency domain data corresponding to the trend components. ,Will The mid-frequency components are retained as the frequency domain data corresponding to the seasonal components. Then, respectively for and Performing an inverse fast Fourier transform and converting back to the time domain yields the trend components. and seasonal ingredients .

[0034] Step S5: Input the trend component of each node, the seasonal component of each node, and the electrical distance between nodes into the preset feature extraction model, so that the feature extraction model can convert the trend component, seasonal component, and electrical distance into embedding vectors to obtain trend feature vector, seasonal feature vector, and electrical distance vector; fuse the trend feature vector, seasonal feature vector, and electrical distance vector to obtain a high-dimensional feature vector; For step S5, the trend component of each node is... Seasonal components of each node Electrical distance between nodes The input is fed into a pre-defined feature extraction model, which transforms the trend component, seasonal component, and electrical distance into embedding vectors, resulting in trend feature vectors, seasonal feature vectors, and electrical distance vectors. ; ; ; in, Indicates the first The trend components of each node Indicates the first The seasonal components of each node, Indicates the first The set of electrical distances between a node and all other nodes in the distribution network. , and Representing the trend feature vector, seasonal feature vector, and electrical distance vector. , and Both refer to the embedded mapping processing built into the feature extraction model, which is used to map different types of input data into low-dimensional embedding vectors. The subscript distinction is to adapt to the feature attributes of different input data. For example, trend components are the long-term time-series features of nodes, seasonal components are the periodic fluctuation features, and electrical distance is the correlation feature between nodes. The mapping details are processed in a targeted manner within a unified framework.

[0035] After obtaining the trend feature vector With seasonal feature vectors Then, the trend feature vector is processed through the fnccLearning dual-view embedding module within the feature extraction model. With seasonal feature vectors The trend-seasonal fused embedding vector is obtained by fusion: ; in, Indicates the first Trend-seasonal fusion embedding vectors of individual nodes This indicates a deep fusion process for trend feature vectors and seasonal feature vectors. During model training, it has learned how to extract key information from trend feature vectors and seasonal feature vectors, and to uncover the complex intrinsic relationships between them.

[0036] Building on this, to further integrate information regarding electrical distance, trend-seasonal fusion is embedded into the vector. With electrical distance vector By concatenating the vectors, we obtain a high-dimensional feature vector: ; in, Indicates the first High-dimensional feature vectors of each node This indicates a vector concatenation operation.

[0037] By fusing the extracted trend feature vector, seasonal feature vector, and electrical distance vector, the potential connections between different types of features are further explored. The resulting high-dimensional feature vector integrates key information from multiple aspects and can more comprehensively and accurately represent the characteristics of the data compared to a single feature vector.

[0038] Next, the training process of the feature extraction model will be explained in detail: In a preferred embodiment, the feature extraction model is determined in the following manner: Acquire historical operating sequence data of each node in the receiving-end circuit of the distribution network and the historical electrical distance between the corresponding nodes; The historical runtime sequence data of each node is compressed using Fast Fourier Transform to obtain the corresponding historical compressed data. Based on the preset decomposition frequency, the historical compressed data is decomposed into trend-based historical components and seasonal historical components. The trend-based historical components of each node, the seasonal historical components of each node, and the historical electrical distance between nodes are input into the feature embedding reconstruction model to be trained, so that the feature embedding reconstruction model takes the trend-based historical components, seasonal historical components, and historical electrical distance as inputs and the trend-based reconstruction components, seasonal reconstruction components, and reconstruction electrical distance as outputs for iterative training. In each training process, the built-in encoding layer transforms the trend historical components, seasonal historical components, and historical electrical distance into embedding vectors, resulting in trend historical feature vectors, seasonal historical feature vectors, and historical electrical distance vectors. The trend historical feature vectors, seasonal historical feature vectors, and historical electrical distance vectors are then fused to obtain high-dimensional historical feature vectors. Calculate the trend similarity between different nodes based on the historical trend feature vector; calculate the seasonal similarity between different nodes based on the historical seasonal feature vector. The high-dimensional historical feature vector is decoded and reconstructed through the built-in decoding layer to obtain the trend reconstruction component, the seasonal reconstruction component, and the reconstruction electrical distance. The reconstruction error is calculated by combining the trend historical component, the seasonal historical component, and the historical electrical distance. Based on trend similarity, seasonal similarity, and reconstruction error, the joint loss is determined; the model parameters of the feature extraction model are adjusted according to the joint loss until the joint loss converges, resulting in a trained feature embedding reconstruction model. The feature extraction model is determined based on the model parameters of the encoding layer in the feature embedding reconstruction model.

[0039] Specifically, the feature extraction model of this invention is based on the E2USD model and is adapted to the feature encoding module trained in the power distribution network scenario.

[0040] Before model training begins, historical runtime sequence data of each node in the receiving-end circuit of the distribution network and the historical electrical distance between the corresponding nodes are obtained. The historical runtime sequence data of each node is compressed by Fast Fourier Transform (FFT) to remove residual components and decompose to obtain trend historical components and seasonal historical components, which are aligned with the real-time data of actual applications.

[0041] The processed trend-based historical components, seasonal historical components, and historical electrical distances are input into the feature embedding reconstruction model to be trained. The model first uses an internal encoding layer to convert these three types of historical features of different dimensions into corresponding embedding vectors. Then, these vectors are fused into a high-dimensional historical feature vector, allowing the model to learn how to map the original features into low-dimensional vectors that retain core information. At the same time, the model calculates the trend similarity and seasonal similarity between different nodes based on the trend-based historical feature vector and the seasonal historical feature vector, thereby capturing the feature association patterns between nodes. ; ; in, Indicates the first The node and the first The trend similarity between nodes is used to measure the degree of similarity in the long-term operating trend characteristics of two nodes. Indicates the first The node and the first Seasonal similarity between nodes is used to measure the degree of similarity in the periodic fluctuation characteristics of the two nodes. and They represent the first The node and the first The historical trend feature vector of each node, and The respective numbers represent the first The node and the first Seasonal historical feature vectors of each node, This represents the dot product operation of vectors.

[0042] It should be noted that, , and , These are all node identification symbols. Different letters are used here to represent the scenarios corresponding to the feature vectors, and are used during training. , corresponding Equivalent vectors are node feature vectors generated based on historical data; while in application... , corresponding Equal vectors are node feature vectors generated based on real-time data.

[0043] Next, the feature embedding reconstruction model, through a decoding layer, reverse-constructs the high-dimensional historical feature vector into trend-based reconstruction components, seasonal reconstruction components, and reconstructed electrical distance. Then, it compares the reconstruction results with the original historical components to calculate the reconstruction error. Finally, it combines trend similarity, seasonal similarity, and reconstruction error to obtain the joint loss. ;in, Indicates joint loss, and The weighting coefficients for the trend similarity constraint and the seasonal similarity constraint are represented. Indicates the first Historical compressed data of each node Indicates the first The reconstructed compressed data of each node consists of trend-based and seasonal reconstructed components. Represents a set of node pairs All nodes in Summation, It is a set of node combinations during the training phase.

[0044] The joint loss ensures that the feature vectors can reconstruct the original information and that the feature relationships between nodes are reasonable. The feature embedding reconstruction model continuously adjusts its parameters based on the joint loss until the joint loss converges, resulting in a well-trained feature embedding reconstruction model.

[0045] At this point, the parameters of the encoding layer are extracted from the trained feature embedding reconstruction model, and the corresponding module is the final feature extraction model, enabling the feature extraction model to accurately extract the features of the receiving-end circuit data of the distribution network in practical applications.

[0046] By leveraging the encoder-decoder framework in E2USD, historical features of the distribution network are input into the model. The encoding layer converts these features into embedding vectors, and the decoding layer reconstructs the original features. Furthermore, considering the actual needs of the distribution network, constraints on inter-node trend and seasonal similarity are added to E2USD. The reconstruction error and similarity features are combined into a joint loss, ensuring that the embedding vectors learned by the model retain information from the original features while also reflecting the operational correlation patterns between nodes.

[0047] Step S6: Determine the target partitioning strategy based on the high-dimensional feature vectors of all nodes; In a preferred embodiment, the target partitioning strategy is determined based on the high-dimensional feature vectors of all nodes, including: Clustering is performed on the high-dimensional feature vectors of all nodes to obtain several clustering partitions; The global comprehensive performance index at the current moment is determined based on each cluster partition and the electrical distance between nodes; Obtain the global comprehensive performance index of the previous moment; If the absolute difference between the global comprehensive performance index at the current moment and the global comprehensive performance index at the previous moment is greater than a preset dynamic threshold, partition optimization is performed on several clustering partitions to obtain the target partitioning strategy.

[0048] For step S6, based on the high-dimensional feature vectors, preliminary clustering of nodes is performed. Using the high-dimensional feature vectors of each node as a basis, the L2 norm is used to calculate the value of any two nodes. , Based on the feature differences, construct the distance metric for clustering: ; in, Indicates the first step of preliminary clustering. The node and the first Feature distance metric between nodes Indicates the first High-dimensional feature vectors of each node Indicates the first A high-dimensional feature vector of each node.

[0049] A lightweight K-means clustering algorithm is used to initially group all nodes. The optimization objective of the algorithm is to minimize the sum of squared feature distances from each node to the center of its cluster, as shown in the following formula: ; in, Indicates the first An initial cluster of nodes, This is the preset initial number of clusters. It is the first Cluster center.

[0050] By optimizing this objective, the high-dimensional features of nodes within the same cluster can be made as similar as possible, ultimately resulting in several initial clustering partitions. .

[0051] After completing the initial clustering based on high-dimensional feature vectors, although the high-dimensional feature vectors have incorporated the features of electrical distance, the initial clustering focuses more on the similarity matching of feature dimensions. In order to make the distribution network partitioning take into account both the operating characteristics of nodes and the electrical connections at the physical level, the rationality of the initial partitioning is verified by combining the electrical distance between nodes. By correlating the partitioning results of the initial clustering with the electrical distance between nodes, the global comprehensive performance index at the current moment is quantitatively evaluated.

[0052] In a preferred embodiment, the global comprehensive performance index at the current moment is determined based on each cluster partition and the electrical distance between nodes, including: Based on the electrical distance between nodes, determine the edge weights between nodes, the weighted degree of nodes, and the total edge weight of the entire network; The global modularity index is determined based on the clustering partitioning, edge weights between nodes, weighted degree of nodes, and total edge weight of the entire network; the clustering partitioning is indicated by an indicator function. To characterize, among which, Represents a node Cluster partition label, Represents a node The cluster partition label to which the node belongs, when the node and nodes When they belong to the same cluster partition, the indicator function Otherwise, the indicator function ; Obtain the adjustable power potential of each node in each cluster partition; The global voltage regulation capability index is determined based on the adjustable power potential of each node in each cluster partition. The global membership stability index is determined based on the proportion of edge weights within cluster partitions. The global modularity index, global voltage regulation capability index, and global membership stability index are weighted and summed to obtain the global comprehensive performance index at the current moment.

[0053] Specifically, based on the electrical distance between nodes, the edge weights between nodes, the weighted degree of each node, and the total edge weight of the entire network are determined using the following formulas: ; ; ; in, Represents a node With nodes Edge weights between them This represents a scaling parameter used to control the rate of decay of the weights on opposite sides of the electrical distance. Represents a node The weighting degree, It represents the total edge weight of the entire network.

[0054] In a given partition Below, in order to measure a single node Define the node's contribution to the rationality of the current clustering partitioning. Module contribution : ; in, Represents a node The module degree contribution; the higher this value, the stronger the node's contribution. The more concentrated the electrically connected nodes are in the partition, the better. Indicates an indicator function, Represents a node Cluster partition label, Represents a node The cluster partition label to which the node belongs, when the node and nodes When they belong to the same cluster partition, the indicator function Otherwise, the indicator function .

[0055] The contributions of all nodes are aggregated to obtain the global modularity index. : ; in, The global modularity index reflects the degree of electrical correlation matching under the current partitioning scheme. The higher the global modularity index, the better the current clustering partitioning matches the actual electrical correlation logic of the distribution network.

[0056] Next, the adjustable power potential of each node in each cluster partition is obtained. It should be noted that adjustable power potential refers to the range within which a node in a distribution network can flexibly adjust its output or consumption power under the constraints of safe grid operation. The greater the adjustable power potential, the wider the power range that the node can adjust, and the stronger its ability to regulate voltage.

[0057] Specifically, The calculation formula is as follows: ; in, and Represents a node The upper and lower limits of the adjustable power.

[0058] For each cluster partition, the adjustable power potential of each node is normalized and calculated to obtain the node's... Voltage regulation potential score: ; in, Represents a node The voltage regulation potential score, Represents a node Adjustable power potential, and This represents the minimum and maximum adjustable power potential of all nodes within the current cluster partition.

[0059] To standardize the measurement of the overall voltage regulation level of each cluster partition, a partition regulation capability index is defined. : ; in, Indicates the first The partition regulation capability index of each cluster partition directly reflects the average voltage regulation capability of the cluster partition. Indicates the first The set of the number of nodes within a cluster partition. For the first The number of nodes within a cluster partition.

[0060] The regulation capability indices of all clustered partitions are weighted and integrated to obtain the global voltage regulation capability index. : ; in, This indicates the overall voltage regulation capability. Indicates the first The weight coefficients of each cluster partition.

[0061] The node membership score is determined using the following formula, based on the proportion of edge weights within the cluster partition: ; in, Represents a node The node membership score is used to measure the node's membership degree. Regarding the degree of membership to the current partition, the higher the node's membership score, the stronger the node's membership. The electrical connections are more concentrated within their respective partitions, and the stronger the binding with the partition.

[0062] To measure the overall cohesion of a single cluster partition, the mean membership degree at the partition level is calculated: ; in, Indicates the first Mean membership degree of each cluster partition. The higher the value, the better. The nodes within a cluster partition are more electrically connected to the nodes within that partition as a whole, and the stronger the structural cohesion of the partition.

[0063] To assess the structural stability of all zones in the entire distribution network, all Each cluster partition The average value is used to obtain the global membership stability index: ; in, This represents the global membership stability index, reflecting the average cohesion level of all partitions.

[0064] Based on the global modularity index, global voltage regulation capability index, and global membership stability index obtained above, a weighted sum is used to obtain the global comprehensive performance index at the current moment: ; in, Indicates the current time The overall comprehensive performance index, , and They represent the current time. The global modularity index, global voltage regulation capability index, and global membership stability index. and This represents the global weighting coefficient.

[0065] Retrieve the global comprehensive performance index from the previous moment from the power distribution network's operation data storage system. This serves as a benchmark for state comparison. It should be noted that the global comprehensive performance index from the previous moment... It is also obtained through the same calculation method.

[0066] Calculate the absolute difference between the global comprehensive performance index at the current moment and the global comprehensive performance index at the previous moment. ,Right now Pre-set dynamic thresholds, when When the value is less than or equal to the dynamic threshold, it indicates that the existing clustering partitioning strategy still adapts well to the current operational requirements, and there is no need to adjust the existing partitions. The current partitioning scheme should be maintained, and the changes in the global comprehensive performance indicators at subsequent time points should continue to be monitored. When the value exceeds the dynamic threshold, it indicates that the overall operation or regional status of the distribution network has changed beyond the tolerance range, such as increased operational fluctuations or decreased adaptability of existing regions. In this case, the current clustering region can no longer match the new operational requirements, and it is necessary to optimize several clustering regions to obtain the target region strategy.

[0067] In a preferred embodiment, partition optimization is performed on several clustering partitions to obtain a target partitioning strategy, including: The particle positions are initialized based on the cluster centers in the cluster partition to obtain the initial particle positions. The particle velocity, inertia weight, individual learning factor and group learning factor are also initialized to obtain the initial particle velocity, initial inertia weight, initial individual learning factor and initial group learning factor of the corresponding particle. Repeat the partition optimization process to obtain the target partitioning strategy; The partition optimization process includes: Calculate the comprehensive objective function value based on the current particle position; If the overall objective function value converges, the target partitioning strategy is determined based on the global best position among the current particle positions. Otherwise, based on the current particle position, current particle velocity, current inertia weight, current individual learning factor, and current group learning factor, update the particle position and particle velocity to obtain the updated particle position and particle velocity; where the current particle position, current particle velocity, current inertia weight, current individual learning factor, and current group learning factor during the first execution of the partition optimization process are the initial particle position, initial particle velocity, initial inertia weight, initial individual learning factor, and initial group learning factor, respectively. The current inertia weight, current individual learning factor, and current group learning factor are updated non-linearly in decreasing order to obtain the updated inertia weight, updated individual learning factor, and updated group learning factor. The updated particle position, updated particle velocity, updated inertia weight, updated individual learning factor, and updated group learning factor are used as the current particle position, current particle velocity, current inertia weight, current individual learning factor, and current group learning factor for the next partition optimization process, respectively.

[0068] Specifically, by improving the particle swarm optimization (PSO) algorithm to optimize cluster centers, the partitioning can both conform to the high-dimensional characteristics of nodes and meet the operational constraints such as power balance of the distribution network.

[0069] First, based on the current cluster partition centers, the initial parameters for the particle swarm optimization algorithm are set, where the particle position corresponds to a combination of cluster centers: ; in, Indicates the current time The position of the particle. Indicates the current time Next Cluster centers of each partition, Indicates the number of partitions. It represents the dimension of the high-dimensional feature space of the node.

[0070] The cluster centers of each partition are directly used as the initial particle positions, and the velocity vectors of the particles are randomly initialized to ensure the diversity of the initial search. The inertia weight is initially set to the maximum value, while the individual learning factor and the group learning factor are respectively set to their respective preset upper limits.

[0071] Next, the partition optimization process is iteratively executed to find better cluster center combinations, thereby improving the rationality and operational adaptability of the partitions. In each partition optimization process, the comprehensive objective function is calculated based on the cluster center combination corresponding to the current particle position. : ; in, Represents the overall objective function. Represents a node The overall performance indicators of the nodes Represents a node The high-dimensional feature vector, Represents a node The cluster center of the partition to which it belongs. This indicates the power imbalance within each zone, used to quantify the degree of power supply and demand imbalance within a zone. This represents the weighting coefficient of the power imbalance penalty term.

[0072] Power imbalance The specific calculation formula is as follows: ; in, Indicates partition The total active power output of all power sources within the system, including photovoltaic, energy storage, and distributed generation units; Indicates partition The sum of active power loads, This indicates the total amount of reactive power provided by reactive power support resources. Indicates partition reactive power demand, The weighting coefficient represents the balance between active and reactive power, and is used to adjust the relative influence of the two.

[0073] Node comprehensive performance indicators The specific calculation formula is as follows: ; in, This indicates the number of dimensions involved in the overall node performance calculation. In this embodiment, , Represents a node The membership score corresponds to the above. , Represents a node The voltage regulation potential score corresponds to the above. , , and This represents the weighting coefficient of each dimension indicator.

[0074] The comprehensive objective function is calculated. After determining the value of the comprehensive objective function, when the value converges, it indicates that the current particle swarm has found a relatively optimal combination of cluster centers. At this point, the globally optimal position is selected from the current particle positions. Subsequently, the high-dimensional feature vector of each node is... It is assigned to the cluster center closest to it. The corresponding partitions are used to determine the target partitioning strategy. When the overall objective function value fails to converge, it indicates that there is still room for optimization in the current cluster center combination, and a better solution needs to be searched. Based on the current particle position, current particle velocity, current inertia weight, current individual learning factor, and current group learning factor, the particle position and particle velocity are updated using the following formulas to obtain the updated particle position and particle velocity: ; ; in, Indicates the current particle velocity. This represents the updated particle velocity. Indicates the current inertia weight. and This represents the current individual learning factor and the current group learning factor. and This represents a uniformly random number between 0 and 1, introducing randomness into the particle search process and preventing the algorithm from getting trapped in local optima. Indicates the current particle position. This indicates the particle's current best historical position. This represents the global optimal position of the entire particle swarm. This indicates the updated particle position.

[0075] It should be noted that the current particle position, current particle velocity, current inertia weight, current individual learning factor, and current group learning factor during the first execution of the partition optimization process are the initial particle position, initial particle velocity, initial inertia weight, initial individual learning factor, and initial group learning factor, respectively.

[0076] The current inertia weights are updated non-linearly and incrementally: ; The current individual learning factor and the current group learning factor are updated in a non-linear decreasing manner: ; ; in, and These represent the fixed upper and lower limits of the inertia weight, respectively. and These represent the fixed upper and lower limits of an individual's learning factor, respectively. and These represent the fixed upper and lower limits of the group learning factor, respectively. Indicates the maximum number of iterations. This represents the coefficient that adjusts the deceleration rate.

[0077] As the number of iterations t increases, the inertia weight... from Nonlinear decrease to , , Each value decreases non-linearly from its maximum value to its minimum value.

[0078] The updated particle positions, velocities, inertia weights, individual learning factors, and group learning factors obtained in this iteration are directly used as the current data for the next partitioning optimization process. Through this iterative transmission of parameters, the partitioning optimization process forms a continuous search chain, continuously advancing towards the convergence of the comprehensive objective function, and gradually improving the rationality of clustering partitions and their adaptability to power distribution network operation.

[0079] Step S7: Based on the target zoning strategy, perform zoning control on the receiving-end circuits of the distribution network.

[0080] In a preferred embodiment, the distribution network receiving-end circuit is zoned and regulated according to a target zoning strategy, including: The receiving-end circuits of the distribution network are partitioned according to the target partitioning strategy; The voltage regulation capability of each zone is evaluated to obtain the corresponding voltage regulation margin; Based on the range corresponding to the voltage regulation margin, reactive power control is performed on each zone accordingly.

[0081] For step S7, based on the target zoning strategy, the receiving-end circuit of the distribution network is divided into corresponding zones, and the historical voltage over-limit frequency, over-limit duration, and adjustable power potential within each zone are statistically analyzed. The voltage regulation capability of each zone is evaluated using a group-level voltage regulation capability evaluation function to obtain the corresponding voltage regulation margin, thereby measuring the strength of the voltage regulation capability of the zone. ; in, Indicates the first Voltage regulation margin for each zone, Indicates the first The historical voltage of each partition exceeded the frequency limit. Indicates the first The duration of the out-of-limit period for each partition. Indicates the first The percentage of nodes in each partition Indicates the first The adjustable power potential of each partition's grouping , , and All of these represent weighting coefficients.

[0082] By using a pre-set low margin threshold and high margin threshold The voltage regulation margin and the low margin threshold are compared. and high margin threshold By comparing the data, we can determine the range of voltage regulation margin. In this case, the voltage regulation margin is in the range where the regulation capability is relatively insufficient; Under these circumstances, the voltage regulation margin is in the range of moderate regulation capability; Under these conditions, the voltage regulation margin is in the range where the regulation capability is relatively strong.

[0083] For zones with relatively insufficient voltage regulation margin and corresponding regulation capacity, distributed photovoltaic inverters will be prioritized to provide reactive power support. When this method is insufficient to meet the demand, the charging and discharging regulation of energy storage units and the switching of capacitor and reactor equipment will be added. If extreme conditions are encountered, flexible loads will be further guided to adjust. For zones with strong voltage regulation margin and corresponding regulation capacity, a mild reactive power regulation strategy will be adopted. Through such differentiated reactive power regulation, the overall voltage optimization of the distribution network and the power balance of the urban receiving end will be ultimately achieved.

[0084] Preferably, after completing reactive power regulation, the total amount of reactive power actually drawn from reactive power support equipment during this reactive power regulation cycle is calculated. The evaluation targets include the reactive power of photovoltaic inverters, the reactive power output of energy storage inverters, and the adjustable reactive power capacity of capacitor / reactor equipment. It also calculates the maximum adjustable reactive power that all the aforementioned reactive power support equipment can provide within the same adjustment cycle. The reactive power utilization rate can be calculated using the following formula. : ; Reactive resource utilization rate This serves as the basis for judging whether the current reactive power regulation is qualified. If the reactive power resource utilization rate... If the preset criteria are met, the reactive power adjustment is deemed satisfactory, and the current support strategy is maintained. Otherwise, the reactive power adjustment is deemed unsatisfactory, and the reactive power support strategy is updated. Specifically, when the utilization rate of a certain type of equipment is low, its participation is increased by raising its priority or expanding its adjustment range; when equipment is operating under high load or has insufficient adjustment margin, its maximum adjustable reactive power capacity is limited. The frequency of calls is controlled to ensure operational security.

[0085] It should be noted that the maximum adjustable power is not a fixed constant, but is dynamically calculated based on the equipment's operating conditions: the maximum reactive power of a photovoltaic inverter is determined by its rated capacity and real-time active power output; the maximum reactive power output of an energy storage device is determined by its inverter's rated capacity, current SOC, and operating mode; and the maximum adjustable capacity of a capacitor-reactor device is determined by its switchable range and current operating status.

[0086] like Figure 2 As shown, based on the above method embodiments, corresponding apparatus embodiments are provided; One embodiment of the present invention provides a zoning control device for the receiving-end circuit of a distribution network, comprising: a data acquisition module, an electrical distance calculation module, a timing processing module, a compression decomposition module, a feature extraction module, a zoning strategy determination module, and a zoning control module; The data acquisition module is used to acquire real-time operating data of each node in the receiving-end circuit of the distribution network and electrical correlation data between nodes; the electrical correlation data includes: line impedance or physical length; The electrical distance calculation module is used to calculate the electrical distance between nodes based on the electrical association data between nodes; The timing processing module is used to perform timing processing on real-time running data to obtain corresponding timing data. The compression and decomposition module is used to perform fast Fourier transform compression on time series data for each node to obtain compressed data, and decompose the compressed data into trend components and seasonal components according to the preset decomposition frequency. The feature extraction module is used to input the trend component, seasonal component, and electrical distance between each node into a preset feature extraction model, so that the feature extraction model can transform the trend component, seasonal component, and electrical distance into embedding vectors to obtain trend feature vector, seasonal feature vector, and electrical distance vector; and fuse the trend feature vector, seasonal feature vector, and electrical distance vector to obtain a high-dimensional feature vector. The partitioning strategy determination module is used to determine the target partitioning strategy based on the high-dimensional feature vectors of all nodes. The zone control module is used to perform zone control on the receiving-end circuits of the distribution network according to the target zone strategy.

[0087] In a preferred embodiment, the zoning control device for the receiving-end circuit of the distribution network further includes: a preprocessing module; The preprocessing module is used to denoise, fill in missing values, and normalize the real-time running data of each node to obtain the preprocessed real-time running data of each node.

[0088] It is understood that the above-described device embodiments correspond to the method embodiments of the present invention, and can implement the zonal control method for the receiving-end circuit of the distribution network provided by any of the above-described method embodiments of the present invention.

[0089] It should be noted that the device embodiments described above are merely illustrative, and some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Furthermore, in the accompanying drawings of the device embodiments provided by this invention, the connection relationships between modules indicate that they have communication connections, which can specifically be implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement this without any creative effort.

[0090] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A method for zoned control of the receiving-end circuit of a distribution network, characterized in that, include: Acquire real-time operating data of each node in the receiving-end circuit of the distribution network and electrical correlation data between nodes; The electrical correlation data includes: line impedance or physical length; Calculate the electrical distance between nodes based on the electrical correlation data between nodes; Perform time-series processing on real-time running data to obtain corresponding time-series data; For each node, the time series data is compressed by Fast Fourier Transform to obtain compressed data, and the compressed data is decomposed into trend components and seasonal components according to the preset decomposition frequency. The trend component, seasonal component, and electrical distance between each node are input into a preset feature extraction model, which converts the trend component, seasonal component, and electrical distance into embedding vectors to obtain trend feature vector, seasonal feature vector, and electrical distance vector. The trend feature vector, seasonal feature vector, and electrical distance vector are then fused to obtain a high-dimensional feature vector. The target partitioning strategy is determined based on the high-dimensional feature vectors of all nodes; Based on the target zoning strategy, the receiving-end circuits of the distribution network are zoned and regulated.

2. The method for zoned control of the receiving-end circuit of a distribution network according to claim 1, characterized in that, After acquiring the real-time operating data of each node in the receiving-end circuit of the distribution network and the electrical correlation data between the nodes, it also includes: The real-time running data of each node is denoised, missing values ​​are filled in, and normalized to obtain the preprocessed real-time running data of each node.

3. The method for zoned control of the receiving-end circuit of a distribution network according to claim 1, characterized in that, The time-series data is compressed using a Fast Fourier Transform, resulting in compressed data, including: Perform a Fast Fourier Transform on the time-series data to obtain the frequency domain data; Based on a preset cutoff frequency, the frequency domain data is frequency filtered to obtain filtered frequency domain data. The filtered frequency domain data is subjected to inverse fast Fourier transform to obtain compressed data.

4. The method for zoned control of the receiving-end circuit of a distribution network according to claim 1, characterized in that, The feature extraction model is determined in the following way: Acquire historical operating sequence data of each node in the receiving-end circuit of the distribution network and the historical electrical distance between the corresponding nodes; The historical runtime sequence data of each node is compressed using Fast Fourier Transform to obtain the corresponding historical compressed data. Based on the preset decomposition frequency, the historical compressed data is decomposed into trend-based historical components and seasonal historical components. The trend-based historical components of each node, the seasonal historical components of each node, and the historical electrical distance between nodes are input into the feature embedding reconstruction model to be trained, so that the feature embedding reconstruction model takes the trend-based historical components, seasonal historical components, and historical electrical distance as inputs and the trend-based reconstruction components, seasonal reconstruction components, and reconstruction electrical distance as outputs for iterative training. In each training process, the built-in encoding layer transforms the trend historical components, seasonal historical components, and historical electrical distance into embedding vectors, resulting in trend historical feature vectors, seasonal historical feature vectors, and historical electrical distance vectors. The trend historical feature vectors, seasonal historical feature vectors, and historical electrical distance vectors are then fused to obtain high-dimensional historical feature vectors. Calculate the trend similarity between different nodes based on the historical trend feature vector; calculate the seasonal similarity between different nodes based on the historical seasonal feature vector. The high-dimensional historical feature vector is decoded and reconstructed through the built-in decoding layer to obtain the trend reconstruction component, the seasonal reconstruction component, and the reconstruction electrical distance. The reconstruction error is calculated by combining the trend historical component, the seasonal historical component, and the historical electrical distance. Based on trend similarity, seasonal similarity, and reconstruction error, the joint loss is determined; the model parameters of the feature extraction model are adjusted according to the joint loss until the joint loss converges, resulting in a trained feature embedding reconstruction model. The feature extraction model is determined based on the model parameters of the encoding layer in the feature embedding reconstruction model.

5. The method for zoned control of the receiving-end circuit of a distribution network according to claim 1, characterized in that, Based on the high-dimensional feature vectors of all nodes, the target partitioning strategy is determined, including: Clustering is performed on the high-dimensional feature vectors of all nodes to obtain several clustering partitions; The global comprehensive performance index at the current moment is determined based on each cluster partition and the electrical distance between nodes; Obtain the global comprehensive performance index of the previous moment; If the absolute difference between the global comprehensive performance index at the current moment and the global comprehensive performance index at the previous moment is greater than a preset dynamic threshold, partition optimization is performed on several clustering partitions to obtain the target partitioning strategy.

6. The method for zoned control of the receiving-end circuit of a distribution network according to claim 5, characterized in that, Based on each cluster partition and the electrical distance between nodes, the global comprehensive performance index at the current moment is determined, including: Based on the electrical distance between nodes, determine the edge weights between nodes, the weighted degree of nodes, and the total edge weight of the entire network; The global modularity index is determined based on the clustering partitioning, edge weights between nodes, weighted degree of nodes, and total edge weight of the entire network; the clustering partitioning is indicated by an indicator function. To characterize, among which, Represents a node Cluster partition label, Represents a node The cluster partition label to which the node belongs, when the node and nodes When they belong to the same cluster partition, the indicator function Otherwise, the indicator function ; Obtain the adjustable power potential of each node in each cluster partition; The global voltage regulation capability index is determined based on the adjustable power potential of each node in each cluster partition. The global membership stability index is determined based on the proportion of edge weights within cluster partitions. The global modularity index, global voltage regulation capability index, and global membership stability index are weighted and summed to obtain the global comprehensive performance index at the current moment.

7. The method for zoned control of the receiving-end circuit of a distribution network according to claim 5, characterized in that, Partition optimization is performed on several cluster partitions to obtain the target partitioning strategy, including: The particle positions are initialized based on the cluster centers in the cluster partition to obtain the initial particle positions. The particle velocity, inertia weight, individual learning factor and group learning factor are also initialized to obtain the initial particle velocity, initial inertia weight, initial individual learning factor and initial group learning factor of the corresponding particle. Repeat the partition optimization process to obtain the target partitioning strategy; The partition optimization process includes: Calculate the comprehensive objective function value based on the current particle position; If the overall objective function value converges, the target partitioning strategy is determined based on the global best position among the current particle positions. Otherwise, based on the current particle position, current particle velocity, current inertia weight, current individual learning factor, and current group learning factor, update the particle position and particle velocity to obtain the updated particle position and particle velocity; where the current particle position, current particle velocity, current inertia weight, current individual learning factor, and current group learning factor during the first execution of the partition optimization process are the initial particle position, initial particle velocity, initial inertia weight, initial individual learning factor, and initial group learning factor, respectively. The current inertia weight, current individual learning factor, and current group learning factor are updated non-linearly in decreasing order to obtain the updated inertia weight, updated individual learning factor, and updated group learning factor. The updated particle position, updated particle velocity, updated inertia weight, updated individual learning factor, and updated group learning factor are used as the current particle position, current particle velocity, current inertia weight, current individual learning factor, and current group learning factor for the next partition optimization process, respectively.

8. The method for zoned control of the receiving-end circuit of a distribution network according to claim 1, characterized in that, Based on the target zoning strategy, zoned regulation is implemented for the receiving-end circuits of the distribution network, including: The receiving-end circuits of the distribution network are partitioned according to the target partitioning strategy; The voltage regulation capability of each zone is evaluated to obtain the corresponding voltage regulation margin; Based on the range corresponding to the voltage regulation margin, reactive power control is performed on each zone accordingly.

9. A zone control device for the receiving-end circuit of a distribution network, characterized in that, include: The system includes a data acquisition module, an electrical distance calculation module, a time series processing module, a compression and decomposition module, a feature extraction module, a partitioning strategy determination module, and a partitioning control module. The data acquisition module is used to acquire real-time operating data of each node in the receiving-end circuit of the distribution network and electrical correlation data between nodes; The electrical correlation data includes: line impedance or physical length; The electrical distance calculation module is used to calculate the electrical distance between nodes based on the electrical association data between nodes; The timing processing module is used to perform timing processing on real-time running data to obtain corresponding timing data; The compression and decomposition module is used to perform fast Fourier transform compression on the time series data for each node to obtain compressed data, and decompose the compressed data into trend components and seasonal components according to a preset decomposition frequency. The feature extraction module is used to input the trend component of each node, the seasonal component of each node, and the electrical distance between nodes into a preset feature extraction model, so that the feature extraction model converts the trend component, seasonal component, and electrical distance into embedding vectors to obtain trend feature vector, seasonal feature vector, and electrical distance vector; and fuses the trend feature vector, seasonal feature vector, and electrical distance vector to obtain a high-dimensional feature vector. The partitioning strategy determination module is used to determine the target partitioning strategy based on the high-dimensional feature vectors of all nodes. The zoning control module is used to perform zoning control on the receiving-end circuit of the distribution network according to the target zoning strategy.

10. The zone control device for the receiving-end circuit of a distribution network according to claim 9, characterized in that, Also includes: Preprocessing module; The preprocessing module is used to perform noise reduction, missing value completion, and normalization on the real-time running data of each node to obtain the preprocessed real-time running data of each node.

Images