A power distribution network operation planning scheme intelligent generation method, system, device and medium

Abstract

The application discloses a power distribution network operation planning scheme intelligent generation method, system, equipment and medium, and belongs to the technical field of power distribution network planning, and comprises the following steps: performing feature extraction on multi-source operation data and multi-source standard data respectively through a multi-modal large model, obtaining operation data features and standard data features, and mapping the standard data features and the operation data features to a unified semantic space to obtain a multi-modal joint feature vector; performing space-time mining and analysis on the multi-modal joint feature vector to determine a space-time dependency relationship; constructing a typical problem graph based on the space-time dependency relationship, an operation abnormal mode and a preset power grid ontology knowledge base; inputting the typical problem graph into a conditional generative adversarial network to obtain a preliminary planning scheme set output by the conditional generative adversarial network, and generating an optimal operation planning scheme according to a preset constraint target. The application improves the effectiveness and practicality of the operation planning scheme.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power distribution network planning, and in particular to a power distribution network operation planning scheme intelligent generation method, system, device and medium. BACKGROUND

[0002] With the rapid development of artificial intelligence technology, multi-modal large models have increasingly deepened their application in the field of power distribution network planning and operation due to their ability to process multiple types of data, such as analyzing power grid operation data to assist decision-making through deep learning algorithms.

[0003] However, there are still significant technical bottlenecks in the intelligent generation of power distribution network operation planning schemes based on multi-modal large models. On the one hand, since power distribution network operation data covers multiple modalities such as text, charts, and images, and multi-modal data such as power grid guidelines, guidelines, and standards are frequently updated, existing technologies are difficult to effectively integrate these heterogeneous data, thereby failing to fully exploit the potential associations between data, resulting in insufficient accuracy of problem diagnosis and planning decisions. On the other hand, accurately extracting key information from massive multi-modal data and constructing a comprehensive and accurate power distribution network typical problem library is the basis for intelligent planning, but due to the complex and variable nature of power grid operation environment and large data noise interference, existing methods are difficult to effectively identify typical problem patterns, and are prone to problem omission or misjudgment, resulting in the automatically generated typical problem library failing to cover all potential problems, severely restricting the effectiveness of the operation planning scheme. SUMMARY

[0004] In view of the above existing problems, the present application is proposed.

[0005] Therefore, the present application provides a power distribution network operation planning scheme intelligent generation method, system, device and medium to solve the problem that existing technologies are difficult to effectively integrate multi-source heterogeneous data and accurately extract key information in power distribution network operation planning, resulting in inaccurate problem diagnosis and difficulty in identifying typical problem patterns, thereby limiting the effectiveness and comprehensiveness of the operation planning scheme.

[0006] To solve the above technical problems, the present application provides the following technical solutions:

[0007] In a first aspect, the present application provides a power distribution network operation planning scheme intelligent generation method, comprising:

[0008] obtaining multi-source operation data and multi-source standard data;

[0009] extracting features from the multi-source operation data and the multi-source standard data through a multi-modal large model to obtain operation data features and standard data features, and mapping the standard data features and the operation data features to a unified semantic space to obtain a multi-modal joint feature vector;

[0010] The multi-modal joint feature vector is subjected to spatio-temporal mining and analysis of data to determine a spatio-temporal dependency relationship;

[0011] Based on the spatio-temporal dependency relationship, the operation abnormal mode and the preset power grid ontology knowledge base, a typical problem atlas is constructed.

[0012] The typical problem atlas is input into a conditional generative adversarial network to obtain a preliminary planning scheme set output by the conditional generative adversarial network, and an optimal operation planning scheme is generated according to a preset constraint target.

[0013] As a preferred scheme of the power distribution network operation planning scheme intelligent generation method, the spatio-temporal dependency relationship is determined by performing spatio-temporal mining and analysis of data on the multi-modal joint feature vector, including:

[0014] The multi-modal joint feature vector is divided based on the distribution of the power distribution network region to obtain a plurality of regional sub-vectors.

[0015] The time series change rate is determined based on the regional sub-vectors, and the regional sub-vectors are filtered based on the time series change rate to obtain an abnormal time series segment.

[0016] For a region pair composed of a first sub-region and a second sub-region in the plurality of regional sub-vectors, a region interrelation matrix is constructed by traversing all region pairs according to the coincidence degree of the abnormal time series segment of the first sub-region and the time series change of the second sub-region in the region pair; the first sub-region and the second sub-region are any region in the plurality of regional sub-vectors, and the first sub-region and the second sub-region are different.

[0017] The time series dependency weight of the first sub-region to the second sub-region is determined based on the region interrelation matrix.

[0018] The spatio-temporal dependency relationship is determined based on the plurality of regional sub-vectors and the time series dependency weight.

[0019] As a preferred scheme of the power distribution network operation planning scheme intelligent generation method, the spatio-temporal dependency relationship is determined based on the plurality of regional sub-vectors and the time series dependency weight, including:

[0020] The spatial position and the time series dependency relationship of each region pair are taken as feature points, and the feature points are clustered to obtain a spatio-temporal dependency clustering cluster.

[0021] The dependency weight change amount is determined based on the spatio-temporal dependency clustering cluster and the time series dependency weight obtained in the next preset sliding time window, and the abnormal spatio-temporal dependency is determined based on the dependency weight change amount.

[0022] The region pairs in the abnormal spatio-temporal dependency are causally verified based on a preset power distribution network topology to obtain a causal verification result.

[0023] Fusing the plurality of regional sub-vectors based on the timing-dependent weight and the causal verification result, constructing a global spatio-temporal dependence model, and determining the global spatio-temporal dependence model as a spatio-temporal dependence relationship.

[0024] As a preferred scheme of the power distribution network operation planning scheme intelligent generation method, wherein: the standard data features and operation data features are mapped to a unified semantic space to obtain a multi-modal joint feature vector, comprising:

[0025] The standard data features and operation data features are respectively subjected to manifold learning dimension reduction to obtain standard data manifold features and operation data manifold features.

[0026] The standard data manifold features and operation data manifold features are respectively clustered to obtain standard data category clusters and operation data category clusters.

[0027] For a category cluster pair composed of any category cluster in the standard data category cluster and any category cluster in the operation data category cluster, a correspondence matrix is constructed based on the feature similarity between the category clusters in the category cluster pair.

[0028] Based on the cosine similarity between the category clusters in each group of the category cluster pair in the correspondence matrix, a preliminary fusion feature is determined.

[0029] The preliminary fusion feature is mapped to a unified semantic space to obtain a multi-modal joint feature vector.

[0030] As a preferred scheme of the power distribution network operation planning scheme intelligent generation method, wherein: the preliminary fusion feature is determined based on the cosine similarity between the category clusters in each group of the category cluster pair in the correspondence matrix, comprising:

[0031] The cosine similarity between the category clusters in each group of the category cluster pair in the correspondence matrix is processed to determine an adjacency matrix.

[0032] Based on the adjacency matrix, a degree matrix is determined, and based on the degree matrix, a Laplacian matrix is constructed.

[0033] The standard data category clusters and the operation data category clusters are sequentially spliced to obtain an initial fusion matrix.

[0034] The initial fusion matrix is optimized based on the Laplacian matrix to determine the preliminary fusion feature.

[0035] As a preferred scheme of the power distribution network operation planning scheme intelligent generation method, wherein: the preliminary fusion features are mapped into a unified semantic space to obtain a multi-modal joint feature vector, including:

[0036] Wavelet transform is performed on the preliminary fusion features to obtain time-frequency components of different scales;

[0037] Based on the energy distribution of each time-frequency component, the attention weight is determined, and the time-frequency components are weighted and reconstructed based on the attention weight to obtain the reconstructed fusion features;

[0038] Based on the reconstructed fusion features, a time-frequency correlation matrix is determined, and a semantic transformation matrix is constructed based on the time-frequency correlation matrix;

[0039] The reconstructed fusion features are mapped and aligned based on the semantic transformation matrix to obtain the aligned fusion features;

[0040] The aligned fusion features are subjected to nonlinear transformation and normalization processing to obtain a multi-modal joint feature vector.

[0041] As a preferred scheme of the power distribution network operation planning scheme intelligent generation method, wherein: based on the spatio-temporal dependency relationship, the operation abnormal mode and the preset power grid ontology knowledge base, a typical problem graph is constructed, including:

[0042] Based on the spatio-temporal dependency relationship, the operation abnormal mode and the preset power grid ontology knowledge base, a problem element set is determined;

[0043] The semantic network of the power grid ontology knowledge base is used to perform semantic analysis on the problem element set to obtain a semantic relationship analysis result;

[0044] Based on the semantic relationship analysis result, each problem element in the problem element set is classified into a pre-constructed multi-layer typical problem graph to obtain a preliminary problem graph, and the spatio-temporal dependency relationship is integrated into the preliminary problem graph to obtain a fusion problem graph;

[0045] The fusion problem graph is logically verified based on the power grid ontology knowledge base to obtain a verification result;

[0046] Based on the verification result, the fusion problem graph is screened and deleted to obtain a typical problem graph.

[0047] In a second aspect, the present application provides a power distribution network operation planning scheme intelligent generation system, comprising:

[0048] An acquisition module is configured to acquire multi-source operation data and multi-source standard data;

[0049] The feature fusion module is used to extract features from multi-source operational data and multi-source standard data through a multimodal large model, to obtain operational data features and standard data features, and to map the standard data features and operational data features to a unified semantic space to obtain a multimodal joint feature vector;

[0050] The analysis and mining module is used to perform spatiotemporal mining and analysis of multimodal joint feature vectors to determine spatiotemporal dependencies.

[0051] The graph construction module is used to construct typical problem graphs based on spatiotemporal dependencies, abnormal operation modes, and a preset power grid ontology knowledge base.

[0052] The scheme generation module is used to input typical problem graphs into the conditional generative adversarial network (GAN) to obtain a preliminary planning scheme set output by the GAN, and generate the optimal operation planning scheme according to the preset constraint objectives.

[0053] Thirdly, the present invention provides an electronic device, including a memory and a processor, wherein the memory stores a computer program, characterized in that the processor executes the computer program to implement the steps of the intelligent generation method for a power distribution network operation planning scheme.

[0054] Fourthly, the present invention provides a computer-readable storage medium having a computer program stored thereon, characterized in that the computer program, when executed by a processor, implements the steps of the intelligent generation method for a power distribution network operation planning scheme.

[0055] The beneficial effects of this invention are as follows: By utilizing a multimodal large model to extract features from operational multi-source data and standard multi-source data respectively, this invention breaks down the barriers between heterogeneous multi-source data. The extracted features are mapped to a unified semantic space, eliminating semantic differences between modalities, enabling the discovery of potential correlations between data, improving data fusion efficiency, and thus enhancing the accuracy of problem diagnosis and planning decisions in subsequent processes. Furthermore, by determining spatiotemporal dependencies, the inherent connections between distribution network operation data in time and space dimensions can be captured. Combined with the identification of abnormal operation patterns, the interference of complex power grid environments and data noise is reduced, while lowering the rate of missed and false positives in problem identification. This ensures that the automatically generated typical problem map can cover various potential problems. Finally, combined with generative adversarial networks and multi-objective optimization, a more accurate and reliable operation planning scheme can be generated, improving the effectiveness and practicality of the operation planning scheme. Attached Figure Description

[0056] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings needed to be used in the embodiments will be briefly introduced as follows. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0057] Figure 1 A basic flowchart of an intelligent generation method of a power distribution network operation planning scheme provided by an embodiment of the present application is shown. DETAILED DESCRIPTION

[0058] In order to make the above-mentioned objects, features and advantages of the present application more apparent and easy to understand, the specific embodiments of the present application will be described in detail below with reference to the drawings in the specification. Obviously, the described embodiments are only some embodiments of the present application, not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor should be within the scope of protection of the present application.

[0059] Embodiment 1, refer to Figure 1 For an embodiment of the present application, an intelligent generation method of a power distribution network operation planning scheme is provided, comprising:

[0060] S100: acquiring multi-source operation data and multi-source standard data;

[0061] In the embodiments of the present application, the scheme intelligent generation system collects multi-source operation data through intelligent electric meters, sensors, cameras and other devices in the actual operation of the power distribution network. The multi-source operation data includes voltage, current, power, temperature, device image and other data, as well as historical operation data such as device historical fault records and maintenance logs. At the same time, multi-source standard data outside the power distribution network can also be obtained through network crawlers, information reprints and other methods, including specification files, industry technical guidelines and standard data from State Grid and other data. Comprehensive multi-source data acquisition can reflect the operation state and specification requirements of the power distribution network from multiple dimensions. Among them, the operation data can capture real-time state, the standard data provides specification criteria, and the combination of the two can avoid the planning scheme from deviating from reality or violating industry standards.

[0062] S200: performing feature extraction on the multi-source operation data and the multi-source standard data through a multi-modal large model to obtain operation data features and standard data features, and mapping the standard data features and the operation data features to a unified semantic space to obtain a multi-modal joint feature vector;

[0063] In the embodiment of the present application, after the multi-source operation data and the multi-source standard data are acquired, the scheme intelligent generation system adopts a multi-modal large model (such as a BERT-based model) to extract features from the multi-source operation data and the multi-source standard data respectively due to different data structures and semantics of different modalities. For text form data in the multi-source operation data and the multi-source standard data, a natural language processing technology is adopted to convert the text form data into word vectors through word segmentation and embedding layer conversion, and then the semantic features are extracted through a multi-layer Transformer encoder. For image form data (such as device images), a convolutional neural network (CNN) is adopted to extract visual features such as edges and textures in the images. For numerical form data (such as voltage and current sequences), a long short-term memory network can be adopted to extract time sequence features, and finally the operation data features and the standard data features are obtained.

[0064] After the operation data features and the standard data features are determined, the scheme intelligent generation system maps the operation data features and the standard data features to a unified semantic space to obtain a multi-modal joint feature vector in order to realize fusion of different modal data, which is specifically described in steps S210-S250.

[0065] S210: Manifold learning dimension reduction is respectively performed on the standard data features and the operation data features to obtain standard data manifold features and operation data manifold features.

[0066] In the embodiment of the present application, the scheme intelligent generation system respectively performs manifold learning dimension reduction on the operation data features and the standard data features to map high-dimensional features to a low-dimensional manifold space.

[0067] Taking the operation data features as an example, let the operation data feature set be .

[0068] ;

[0069] For each data point in , find its k nearest neighbors , calculate the local linear reconstruction weight value of the data point and its neighbors , so that:

[0070] ;

[0071] wherein is an index variable.

[0072] Under the weight constraint, the low-dimensional embedding coordinates are solved to obtain the operation data manifold features .

[0073] Similarly, the same processing is performed on the standard data feature set to obtain the standard data manifold feature .

[0074] In the embodiment of the present application, to ensure that the running data and the standard data are in a unified low-dimensional semantic space for subsequent similarity calculation and feature fusion, the following is usually performed wherein, is the target dimension of manifold learning dimension reduction, representing the feature dimension of the low-dimensional embedding space (D). This dimension will be used as the unified feature space dimension for subsequent feature fusion, semantic mapping and spatiotemporal analysis, and remains unchanged throughout the entire processing flow.

[0075] S220: Cluster the standard data manifold feature and the running data manifold feature respectively to obtain the standard data category cluster and the running data category cluster.

[0076] In the embodiment of the present application, the scheme intelligent generation system clusters the running data manifold feature and the standard data manifold feature using a dynamic clustering algorithm (such as K-means clustering, DBSCAN, etc.).

[0077] In the embodiment of the present application, the number of running data clusters is set to K , and the number of standard data clusters is set to K .

[0078] Initialize K running data cluster centers and K standard data cluster centers .

[0079] Calculate the distance of each data point to the cluster center, and assign it to the nearest cluster center category, update the cluster center, and repeat the process until convergence to obtain the running data category cluster and the standard data category cluster

[0080] S230: For any pair of prototypes in the standard data category cluster and any state grouping in the running data category cluster, construct a correspondence matrix based on the feature similarity of the two categories in the pair.

[0081] In the embodiment of the present application, the scheme intelligent generation system calculates the similarity between category clusters using cosine similarity.

[0082] In this embodiment of the invention, the intelligent scheme generation system uses cosine similarity to calculate the similarity between category clusters. In K-means clustering, the cluster center of a category cluster is the mean vector of all data points within that cluster. Therefore, let the running data... Each cluster of runtime data categories The cluster centers are It is equivalent to the mean vector of data points within the cluster. ,Right now Similarly, standard data category clusters Cluster center Equivalent to .

[0083] Let the pairing be the first of the running data. Each cluster of runtime data categories and the first standard data Standard data category clusters Calculate the arithmetic mean vector of all data points within it. and Then the cosine similarity between them is expressed as:

[0084] ;

[0085] Traverse all Construct a correspondence matrix ,in , This represents the feature similarity between the i-th running data category cluster and the j-th standard data category cluster.

[0086] S240: Based on the cosine similarity between the category clusters in each pair of category clusters in the correspondence matrix, determine the preliminary fusion features.

[0087] In this embodiment of the invention, the intelligent solution generation system generates solutions based on a corresponding relationship matrix. The cosine similarity between each group of category clusters is used to fuse the features of the two category clusters to obtain preliminary fused features, as described in steps S241-S244.

[0088] S241: Perform cosine similarity processing on the category clusters in each pair of matching category clusters in the correspondence matrix to determine the adjacency matrix.

[0089] In this embodiment of the invention, the intelligent solution generation system generates solutions based on a corresponding relationship matrix. ,in For the number of data category clusters during operation, Construct an adjacency matrix of an undirected graph to represent the number of standard data category clusters. .

[0090] The rules for constructing the adjacency matrix are as follows:

[0091] For connections between nodes within a standard data category cluster, if the two nodes are not directly related... If there is a direct correlation, a value is assigned based on the actual degree of correlation; the same processing is applied to the running data category clusters. As for the connection between the standard data category cluster and the running data category cluster, , ,in, This represents the connection weight in the adjacency matrix from the i-th running data category cluster to the j-th standard data category cluster. This represents the connection weight in the adjacency matrix from the j-th standard data category cluster to the i-th running data category cluster.

[0092] This process transforms the correspondence matrix into an adjacency matrix, which more intuitively represents the strong associations between category clusters. The adjacency matrix clearly highlights meaningful associations between category clusters while filtering out weaker associations that might interfere with subsequent analysis.

[0093] S242: Determine the degree matrix based on the adjacency matrix, and construct the Laplace matrix based on the degree matrix.

[0094] In this embodiment of the invention, the intelligent solution generation system uses the adjacency matrix... Construct the degree matrix It is a diagonal matrix, degree matrix The diagonal elements The degree of node i is represented by the sum of the weights of the edges connected to node i.

[0095] ;

[0096] Then construct the Laplace matrix The topological structure information of the graph is reflected through the Laplace function.

[0097] S243: Sequentially concatenate the standard data category cluster and the running data category cluster to obtain the initial fusion matrix.

[0098] In this embodiment of the invention, the intelligent scheme generation system concatenates the mean feature vectors of the running data category cluster and the standard data category cluster vertically in sequence to form an initial fusion matrix. Each row corresponds to the central feature vector of a category cluster.

[0099] In this embodiment of the invention, the feature matrices of five policy text standard data category clusters, such as "electricity price policy" and "equipment standard", and the feature matrices of three load data operation data category clusters, such as "high load", "medium load" and "low load", are concatenated in sequence to obtain an initial fusion matrix. At this time, the matrix contains 8 rows (corresponding to 8 category clusters) and feature dimension columns after dimensionality reduction.

[0100] S244: The initial fusion matrix is ​​optimized based on the Laplace matrix to obtain preliminary fusion features.

[0101] In this embodiment of the invention, the intelligent solution generation system optimizes the initial fusion matrix using the Laplacian matrix obtained in step S242, that is, it uses a Laplacian regularization-based optimization objective function:

[0102] ;

[0103] in, This indicates the preliminary fusion characteristics to be solved. Represents the trace of a matrix. It is a regularization parameter used to balance the Laplacian regularization term and preserve the original data information term; This represents the Frobenius norm, which indicates the initial fusion feature. With the initial fusion matrix The degree of difference. This objective function is typically solved using optimization algorithms such as gradient descent to obtain preliminary fusion features. And in this process, through the Laplace regularization term This encourages similar category clusters (with strong correlations in the adjacency matrix) to be closer in the fused features, thereby achieving feature fusion optimization based on category cluster correlation.

[0104] This invention considers the similarity and global structure between standard data category clusters and operational data category clusters, achieving effective fusion of multimodal data features in a mathematical and structured manner. Compared to simple feature splicing or fusion methods, it can more accurately capture the inherent relationships between data, filter out interference information, optimize feature representation, and obtain preliminary fused features with higher quality and representativeness. This lays a solid foundation for subsequent spatiotemporal analysis, problem diagnosis, and solution generation in distribution network operation planning, improving the accuracy and reliability of the entire intelligent generation method.

[0105] S250: The initial fused features are mapped onto a unified semantic space to obtain a multimodal joint feature vector.

[0106] In this embodiment of the invention, the intelligent scheme generation system can map the obtained preliminary fusion features to a unified semantic space through nonlinear transformation to obtain a multimodal joint feature vector, as described in steps S251-S255.

[0107] This invention achieves the fusion of standard data features and operational data features to obtain a high-quality multimodal joint feature vector. It effectively solves the problems of heterogeneous multimodal data and large semantic differences, providing a comprehensive, accurate, and intrinsically linked data representation for distribution network operation planning, and improving the accuracy and efficiency of subsequent spatiotemporal mining, problem diagnosis, and planning scheme generation.

[0108] In this embodiment of the invention, steps S251-S255 are described as follows:

[0109] S251: Perform wavelet transform on the preliminary fusion features to obtain time-frequency components at different scales.

[0110] In this embodiment of the invention, the intelligent scheme generation system performs wavelet transform based on the determined preliminary fusion features, and adopts a method for analyzing the preliminary fusion features. Assuming its dimension is "Consider the sample size as a single signal," and perform discrete wavelet transform on each feature dimension. The formula for discrete wavelet transform is:

[0111] ;

[0112] in, It is the original signal (i.e., one-dimensional data of the preliminary fused features). Describe the wavelet basis functions. Indicated in scale Translation parameters The wavelet transform coefficients.

[0113] By selecting different scales This allows us to obtain time-frequency components at different resolutions, enabling us to capture the changes in the initial fusion features at different times and frequencies.

[0114] In this embodiment of the invention, taking a distribution network data scenario as an example, the initial fusion feature includes information fused from policy text and load data. Taking a certain dimension feature (such as a feature reflecting the correlation between load fluctuations and policies) as an example, a discrete wavelet transform is performed on it. By selecting three different scales, low-frequency overview components and high-frequency detail components are obtained. These components describe the time-frequency characteristics of the feature from different perspectives.

[0115] S252: Based on the energy distribution of each time-frequency component, determine the attention weights, and reconstruct the time-frequency components based on the attention weights to obtain the reconstructed fusion features.

[0116] In this embodiment of the invention, the intelligent scheme generation system calculates the energy of each time-frequency component based on the acquired time-frequency components, where the energy of each time-frequency component reflects its importance in the entire signal.

[0117] ;

[0118] in, Indicates the first The energy of each time-frequency component Indicates the first Wavelet transform coefficients of each time-frequency component under scale p and translation parameter l;

[0119] For the first The energy of each time-frequency component is normalized to obtain the attention weights. :

[0120] ;

[0121] in, This represents the total number of time-frequency components.

[0122] By weighting the time-frequency components based on attention weights, the reconstructed fused features are obtained. The calculation formula is:

[0123] ;

[0124] in, This represents the characteristic representation of the i-th time-frequency component.

[0125] That is, weighted summation is performed according to the importance of each time-frequency component, highlighting important time-frequency features and suppressing unimportant features.

[0126] S253: Based on the reconstructed fusion features, determine the time-frequency correlation matrix, and based on the time-frequency correlation matrix, construct the semantic transformation matrix.

[0127] In this embodiment of the invention, the intelligent solution generation system generates solutions based on the acquired reconstruction and fusion features. Calculate the Pearson correlation coefficient between the i-th time-frequency component and the j-th time-frequency component. :

[0128] ;

[0129] in, and They are the first The and the first The value of the time-frequency component on the k-th sample and It is the mean of the corresponding components.

[0130] Based on the time-frequency correlation matrix, a semantic transformation matrix is ​​constructed using methods such as singular value decomposition. Singular value decomposition (SVD) will transform the time-frequency correlation matrix. Decomposed into By selecting appropriate singular values ​​and corresponding singular vectors, a semantic transformation matrix is ​​constructed through a certain combination method. .

[0131] In this embodiment of the invention, for the reconstructed fusion features, the Pearson correlation coefficient between different time-frequency components is calculated to obtain the time-frequency correlation matrix. For example, it is found that the time-frequency component reflecting load fluctuations under high-temperature weather has a high correlation with the time-frequency component reflecting the impact of corresponding policies. Singular value decomposition is performed on the time-frequency correlation matrix, and the singular vectors corresponding to the first few larger singular values ​​are selected to construct a semantic transformation matrix.

[0132] S254: Based on the semantic transformation matrix, feature mapping and alignment are performed on the reconstructed fusion features to obtain aligned fusion features.

[0133] In this embodiment of the invention, the intelligent solution generation system employs a semantic transformation matrix. Reconstruction and fusion features Perform a linear transformation to obtain the alignment and fusion features. :

[0134] ;

[0135] In this process, the feature distribution and semantic meaning of different modalities are adjusted under the action of the semantic transformation matrix, so that they have better alignment and consistency in the new space.

[0136] S255: Perform nonlinear transformation and normalization on the aligned and fused features to obtain the multimodal joint feature vector.

[0137] Despite alignment and fusion features Alignment and structured representation of multimodal features have been achieved in a unified semantic space, but their essence is still the result of linear or quasi-linear transformation, which is difficult to fully characterize the nonlinear dynamic behavior (such as equipment saturation, sudden faults, multi-source coupled oscillations, etc.) that exists in the operation of the distribution network.

[0138] To enhance the model's ability to express complex nonlinear modes and improve the robustness and adaptability of subsequent scheme generation, the intelligent scheme generator further introduces a nonlinear transformation mechanism. Specifically, the system uses nonlinear activation functions (such as ReLU, Sigmoid, etc.) to... Element-wise nonlinear mapping is performed to generate a more discriminative nonlinear feature representation, the calculation formula of which is:

[0139] ;

[0140] in, For activation function, This represents the characteristics after a nonlinear transformation.

[0141] Perform normalization processing to map the eigenvalues ​​to The interval is used to obtain the final multimodal joint feature vector. .

[0142] This invention fully leverages the feature information of multimodal data in the time-frequency domain, utilizes an attention mechanism to highlight key features, achieves feature alignment and unified representation through semantic transformation, and obtains a high-quality multimodal joint feature vector after nonlinear transformation and normalization. It effectively integrates data from different modalities, enhances the semantic consistency and expressive power of features, and improves the accuracy and robustness of multimodal data processing tasks (such as image and text retrieval, multimodal classification, etc.).

[0143] S300: Perform spatiotemporal mining and analysis of multimodal joint feature vectors to determine spatiotemporal dependencies;

[0144] In this embodiment of the invention, after obtaining the multimodal joint feature vector, the intelligent scheme generation system performs spatiotemporal mining and analysis of the multimodal joint feature vector from two dimensions: time and space, in order to determine the spatiotemporal dependency relationship, as described in steps 310-350, so as to achieve a more accurate grasp of the operation law of the distribution network and provide a more reliable basis for operation planning.

[0145] In this embodiment of the invention, steps S310-S350 are described as follows:

[0146] S310: Based on the regional distribution of the power distribution network, the multimodal joint feature vector is divided to obtain multiple regional sub-vectors.

[0147] In this embodiment of the invention, since the distribution network is physically composed of multiple regions, each region contains different types of equipment (such as transformers, feeders, etc.), and their operating states and external influences differ. Therefore, the intelligent solution generation system divides the multimodal joint feature vector according to the actual regional distribution of the distribution network. Specifically, it is assumed that the multimodal joint feature vector... ,in, This is expressed as the sample size. Represented as a feature dimension, if the distribution network is divided into Each region ,from Extract the feature dimensions related to the device to obtain the region sub-vector, denoted as:

[0148] ;

[0149] in, For the region The number of associated feature dimensions.

[0150] In subsequent analyses, we will use the same approach. Indicates the first Region subvectors of each region.

[0151] In this embodiment of the invention, a city's power distribution network is divided into three regions: A, B, and C. The multimodal joint feature vector includes features fused from power distribution network operation data (such as load and voltage) and external standard data (such as weather and policy). Feature data related to equipment in region A (such as transformers and feeders in region A) are extracted to form a sub-vector for region A. Similarly, we can obtain and .

[0152] S320: Determine the time series change rate based on the regional sub-vectors, and filter the regional sub-vectors based on the time series change rate to obtain abnormal time series segments.

[0153] In this embodiment of the invention, the intelligent solution generation system first calculates the rate of change of each regional sub-vector over time to reflect the dynamic changes in the equipment's operating status over time. The rate of change can be calculated as the ratio of the difference between feature values ​​at adjacent time points to the time interval. Specifically, let the regional sub-vector... In time eigenvalues Then the rate of change over time Represented as:

[0154] ;

[0155] in, The time interval is defined as [time interval]. A threshold is set to filter out abnormal time segments. ,when Time markers are designated as outliers, and consecutive outliers constitute an abnormal time sequence.

[0156] In this embodiment of the invention, taking the urban power distribution network as an example, for the regional sub-vector Calculate its intervals every 15 minutes. The time-series change rate of the load characteristic value (minutes). It was found that between 2 PM and 3 PM on a certain day, the load change rate repeatedly exceeded the set threshold. If this time period is abnormal, then this time period is identified as an abnormal time series. The same method can be used to determine... and Abnormal timing segment.

[0157] S330: For a region pair consisting of the first and second sub-regions in multiple region sub-vectors, based on the overlap between the abnormal time series segment of the first sub-region and the time series change of the second sub-region in the region pair, traverse all region pairs and construct the inter-regional correlation matrix; the first and second sub-regions are any regions in multiple region sub-vectors, and the first and second sub-regions are not the same.

[0158] In this embodiment of the invention, the intelligent solution generation system measures the correlation between different region pairs by the degree of overlap between abnormal time segments and the set of time points showing significant changes in another region. Specifically, the correlation degree is obtained by calculating the proportion of abnormal time segments in the first sub-region within the significant change period in the second sub-region. Let the first sub-region be... The second sub-region is j, and the set of abnormal time segments in the first sub-region is j. The temporal variation data of the second sub-region within the corresponding time range are as follows: Calculate the degree of overlap:

[0159] ;

[0160] in, This indicates the degree of overlap between the abnormal time series of the first sub-region i and the data of the second sub-region j within the corresponding time range.

[0161] Traverse all distinct region pairs and construct the region association matrix. Among them, the inter-regional correlation matrix In the equation, the correlation strength between region i and region j is... Inter-regional correlation matrix diagonal elements .

[0162] In this embodiment of the invention, for the region sub-vector and , The abnormal time period was from 2 pm to 3 pm. (Statistics) Within this timeframe, identify periods of significant change (e.g., times when the rate of change exceeds a threshold), forming a set of time points showing significant change. Calculate the degree of overlap between these two sets. .

[0163] Similarly, calculate and , By quantifying the overlap between all region pairs, an inter-regional correlation matrix is ​​constructed. For example, the resulting inter-regional correlation matrix is:

[0164] .

[0165] S340: Determine the temporal dependency weight of the first sub-region on the second sub-region based on the inter-regional correlation matrix.

[0166] In this embodiment of the invention, the intelligent solution generation system further determines the temporal dependency weights between each pair of regions based on the overlap values ​​in the inter-regional correlation matrix. A normalization method can be used to convert the overlap values ​​into weights. Let the inter-regional correlation matrix be... The temporal dependency weight of the first sub-region on the second sub-region is then... Represented as:

[0167] ;

[0168] in, Representing the inter-regional correlation matrix The element in the i-th row and j-th column.

[0169] The larger the weight, the greater the impact of changes in the device status of the first sub-region on the second sub-region.

[0170] In this embodiment of the invention, for the above-mentioned inter-regional correlation matrix ,calculate right Temporal dependency weights Represented as:

[0171] ;

[0172] Similarly, calculate the temporal dependency weights of other region pairs to obtain the complete weight matrix.

[0173] S350: Determines spatiotemporal dependencies based on multiple regional subvectors and temporal dependency weights.

[0174] In this embodiment of the invention, the intelligent solution generation system combines multiple regional sub-vectors and temporal dependency weights to comprehensively analyze the dependency relationships of distribution network equipment states from both temporal and spatial dimensions. This is specifically described in S351-S354.

[0175] This invention enables precise, granular analysis of the spatiotemporal dependencies between equipment states in different areas of a power distribution network. Compared to traditional methods, it better adapts to the complex spatial-temporal interaction scenarios of power distribution networks. It can accurately pinpoint the linkage patterns of cross-regional equipment states, helping maintenance personnel to provide early warnings of cascading failure risks and prevent the escalation of faults.

[0176] In this embodiment of the invention, S351-S354 are described as follows:

[0177] S351: Use the spatial location and temporal dependency of each region pair as feature points, and cluster the feature points to obtain spatiotemporal dependency clusters.

[0178] In this embodiment of the invention, the intelligent solution generation system, based on previously obtained multiple regional sub-vectors and temporal dependency weights between regional pairs, recognizes that each regional pair corresponds to a certain spatial location (e.g., the relative geographical locations of region A and region B) and has a specific temporal dependency relationship (reflected by the temporal dependency weights). Therefore, the spatial location information (e.g., latitude and longitude coordinates) and temporal dependency weights of the regional pairs are combined to form a multi-dimensional feature point. Clustering algorithms (e.g., DBSCAN, K-Means) are then used to cluster these feature points. Taking K-Means clustering as an example, the number of clusters K is first determined, K cluster centers are randomly initialized, the distance from each feature point to the cluster center is calculated, and the feature point is assigned to the cluster containing the nearest cluster center. The cluster centers are then updated, and this process is iterated until the cluster centers no longer change significantly, ultimately resulting in K spatiotemporally dependent clusters.

[0179] In this embodiment of the invention, the previous example of three areas, A, B, and C, of ​​the urban power distribution network is continued. (Area pair) The spatial distance is 10 kilometers, and the temporal dependence weights are... , represented as feature points ; region pair The spatial distance is 15 kilometers, and the temporal dependence weights are... , represented as ; region pair The spatial distance is 8 kilometers, and the temporal dependence weights are... , represented as Using the K-Means clustering algorithm, with the number of clusters K set to 2, after iterative calculation, two spatiotemporally dependent clusters were obtained, one of which contains region pairs. and Another cluster contains region pairs .

[0180] S352: Based on the spatiotemporal dependency clusters and the temporal dependency weights obtained within the next preset sliding time window, determine the change in dependency weights, and based on the change in dependency weights, determine abnormal spatiotemporal dependencies.

[0181] In this embodiment of the invention, after obtaining the spatiotemporal dependency clusters, the intelligent solution generation system considers the time factor and sets a sliding time window (e.g., 1 hour, half a day, etc.). For each cluster, the set of temporal dependency weights within the current time window is obtained. and the set of time-dependent weights within the next sliding time window. And calculate the change in dependency weights for a given region. For example, its weight change Represented as:

[0182] ;

[0183] in, and These are respectively represented as the next window and the area within the current window. The time-dependent weights are then determined. Next, a threshold for the amount of change is set. ,when When this occurs, it indicates that the spatiotemporal dependency of the region is abnormal, and it is identified as an abnormal spatiotemporal dependency.

[0184] In this embodiment of the invention, taking the urban power distribution network as an example, it is assumed that one of the previously obtained spatiotemporal dependency clusters contains region pairs. and Within the current 1-hour time window, ; within the next 1-hour sliding time window, .but Set a threshold. ,because and Both are greater than Therefore, the region is and All spatiotemporal dependencies were identified as anomalous spatiotemporal dependencies.

[0185] S353: Based on the preset distribution network topology, perform causal verification on regional pairs in abnormal spatiotemporal dependencies to obtain causal verification results.

[0186] In this embodiment of the invention, since the distribution network has a specific topology that describes the connection relationships between various regions and devices, the intelligent solution generation system performs causal verification on the region pairs identified in the abnormal spatiotemporal dependencies, in conjunction with the preset distribution network topology. If there is no direct electrical connection between region i and region j in the topology, but they exhibit abnormal temporal dependency weight changes, then this dependency relationship may be spurious or subject to other interference factors. Therefore, by analyzing information such as connection paths and device connection relationships in the topology, it is determined whether there is a reasonable causal relationship between the region pairs in the abnormal spatiotemporal dependencies. If a reasonable causal path exists (such as being connected through transformers, feeders, etc., and conforming to the logic of power transmission and influence), then the causal relationship is considered valid and marked as a positive result; otherwise, it is marked as a negative result, thereby obtaining the causal verification result for each abnormal spatiotemporal dependency region pair.

[0187] In this embodiment of the invention, within the aforementioned determined anomalous spatiotemporal dependency region... and In the process of examining the distribution network topology, it was found that region A and region B are connected by a main feeder, and region B and region C are connected by a tie switch. The electrical connections between them can explain the anomalous changes in time-dependent weights. Therefore, for region A and region B... and The causal verification result is positive, indicating that this abnormal dependency has a real physical connection basis.

[0188] S354: Based on temporal dependency weights and causal verification results, multiple regional sub-vectors are fused to construct a global spatiotemporal dependency model, and the global spatiotemporal dependency model is determined as a spatiotemporal dependency relationship.

[0189] In this embodiment of the invention, the intelligent solution generation system comprehensively considers temporal dependency weights and causal verification results to fuse multiple region sub-vectors. For region pairs with positive causal verification results, the correlation between region sub-vectors is strengthened according to their corresponding temporal dependency weights; for region pairs with negative causal verification results, their correlation is weakened or ignored. Specifically, a global spatiotemporal dependency model can be constructed through weighted fusion. Let the region sub-vectors be... , region The time-dependent weights are The causal verification result is (in This indicates that a causal relationship is established. (This indicates the condition is not met). Therefore, the global spatiotemporal dependency model... It can be represented as:

[0190] ;

[0191] in, Represented as a matrix, the model comprehensively describes the spatiotemporal dependencies of the distribution network by integrating the characteristics of regional sub-vectors, the temporal dependency weights between regions, and causal relationships.

[0192] In this embodiment of the invention, for the region sub-vector Combined with the temporal dependency weights obtained earlier and causal verification results Assuming , , Substituting these values ​​into the above formula, the global spatiotemporal dependency model is calculated. This model provides a clear understanding of the spatial and temporal interdependencies among different areas of the distribution network, such as the regional... and region How can we influence each other's operational state through their dependency weights and physical connections, while excluding regional... and region Unreasonable dependencies between them.

[0193] The embodiments of the present invention can more accurately identify spatiotemporal dependency patterns with actual physical meaning, effectively filter false anomaly information, and the constructed global spatiotemporal dependency model comprehensively reflects the operating characteristics of the power distribution network.

[0194] S400: Construct a typical problem map based on spatiotemporal dependencies, abnormal operation modes, and a pre-set power grid ontology knowledge base;

[0195] In this embodiment of the invention, the abnormal operation mode is obtained by comparative learning based on historical multi-source operation data.

[0196] In this embodiment of the invention, the intelligent solution generation system pre-sets a power grid ontology knowledge base and anomaly operation modes. The power grid ontology knowledge base includes knowledge such as equipment types, fault types, and operating parameters. The anomaly operation modes are obtained through comparative learning of historical multi-source operating data. For example, a comparative learning algorithm is used to compare normal operating data and fault data to identify characteristic differences before a fault occurs, forming an anomaly operation mode. Then, combined with the spatiotemporal dependencies obtained in step S300, a typical problem graph is constructed, as described in steps S410-S450. Nodes represent problems, and edges represent the relationships between problems. This ensures that the constructed typical problem graph can intuitively display the problems and relationships in the distribution network operation, facilitating planners to quickly understand the nature and scope of the problems.

[0197] In this embodiment of the invention, taking the operation of a city power distribution network in high summer temperatures as an example, based on the spatiotemporal dependency relationship determined in step S300, it is found that a certain area shows a voltage drop trend during high-temperature periods. Combining the abnormal operation mode of "high temperature leads to load surge and voltage drop" obtained by historical data comparison and learning, and the knowledge of "voltage drop may cause equipment damage" in the power grid ontology knowledge base, a typical problem graph containing nodes and related edges such as "high temperature", "load surge", "voltage drop" and "equipment damage" is constructed.

[0198] In this embodiment of the invention, steps S410-S450 are described as follows:

[0199] S401: Based on spatiotemporal dependencies, abnormal operation modes, and a pre-defined power grid ontology knowledge base, determine the set of problem elements.

[0200] In this embodiment of the invention, the intelligent solution generation system extracts key elements related to distribution network problems from these three aspects: spatiotemporal dependencies, operational anomaly patterns, and a pre-defined power grid ontology knowledge base. The spatiotemporal dependencies reveal the mutual influence of equipment operating states in different regions and at different times within the distribution network; the operational anomaly patterns are characteristic patterns of equipment faults or abnormal operations obtained through comparative learning of historical data; and the power grid ontology knowledge base includes knowledge in areas such as equipment parameters, operating rules, and fault types.

[0201] In this embodiment of the invention, the following are extracted: "load transfer dependency between region A and region B during peak hours" is extracted from spatiotemporal dependencies; "overload fault mode with excessively high transformer oil temperature and abnormal short-circuit current" is extracted from abnormal operation modes; and knowledge such as "transformer rated capacity" and "line safe current threshold" is extracted from the power grid knowledge base. These key pieces of information are used as problem elements to form a problem element set, such as {load transfer dependency between region A and region B, transformer overload fault mode, transformer rated capacity, line safe current threshold}.

[0202] S420: Based on the semantic network of the power grid ontology knowledge base, semantic analysis is performed on the set of problem elements to obtain the semantic relationship analysis results.

[0203] In this embodiment of the invention, the power grid ontology knowledge base typically stores knowledge in the form of a semantic network. The semantic network consists of nodes (representing concepts, such as "transformer" and "short-circuit fault") and edges (representing relationships between concepts, such as "transformer - includes - winding" and "short-circuit fault - leads to - power outage"). Therefore, the intelligent solution generation system searches for the corresponding concept node in the semantic network for each element in the problem element set and analyzes its connection relationships with other nodes. For example, for the problem element "transformer overload fault mode," the semantic network finds the nodes "transformer" and "overload fault," and analysis reveals causal relationship edges between "overload fault" and nodes such as "insufficient transformer capacity" and "excessive load"; similarly, "load transfer dependence between region A and region B" has association relationships with nodes such as "regional power grid topology" and "load allocation rules." In this way, the semantic relationships between each problem element and other concepts are obtained, forming semantic relationship analysis results.

[0204] S430: Based on the semantic relationship analysis results, each problem element in the problem element set is classified into a pre-constructed multi-layer typical problem graph to obtain a preliminary problem graph. The spatiotemporal dependency relationship is then integrated into the preliminary problem graph to obtain a fused problem graph.

[0205] In this embodiment of the invention, the pre-constructed multi-layered typical problem graph within the intelligent solution generation system is a framework built based on common distribution network problem types and hierarchical structures, such as being divided into an equipment layer (transformer problems, line problems, etc.), a fault type layer (overload faults, short-circuit faults, etc.), and an impact consequence layer (power outages, voltage anomalies, etc.). After obtaining the semantic relationship analysis results, problem elements are categorized into nodes at the corresponding levels. For example, "transformer rated capacity" is categorized under the "transformer" node in the equipment layer; "transformer overload fault mode" is categorized under the "overload fault" node in the fault type layer. Simultaneously, edge connections are established with the "transformer" node in the equipment layer and other related nodes through semantic relationships to form a preliminary problem graph. Spatiotemporal dependencies are then integrated into the preliminary problem graph. For example, the spatiotemporal dependency "region A and region B have load transfer dependencies during peak hours" is represented by adding an edge between the nodes of region A and region B in the graph, and the load transfer dependency and time period information are labeled to obtain a fused problem graph. This results in a graph that not only contains problem elements and semantic relationships but also reflects spatiotemporal connections.

[0206] S440: Logically verify the fusion problem graph based on the power grid ontology knowledge base, and filter and delete the fusion problem graph based on the verification results to obtain a typical problem graph.

[0207] In this embodiment of the invention, the power grid ontology knowledge base contains the logical rules and professional knowledge of distribution network operation, which are used as the basis for verifying the fusion problem graph. Therefore, the intelligent solution generation system checks whether the relationships between nodes in the graph conform to actual logic. For example, it determines whether the association between "transformer overload fault" and "line safe current threshold" is reasonable. If relationships that do not conform to the power grid operation logic appear in the graph (such as incorrect causal relationships or unreasonable parameter associations), the corresponding nodes or edges are deleted. For example, if the fusion problem graph contains a relationship where "transformer overload fault" leads to "reduction of line safe current threshold," which does not conform to actual logic, then this relationship edge is deleted. By filtering and deleting unreasonable parts, the final typical problem graph is obtained.

[0208] This invention constructs a typical problem graph through a series of steps, including determining the set of problem elements, performing semantic analysis, classification and fusion, and logical verification. This allows for the organic integration of complex information such as spatiotemporal dependencies and abnormal operation modes in distribution network operation with knowledge from the power grid domain, forming a structured and logical knowledge graph. This graph comprehensively and accurately reflects the inherent logic and various relationships of typical distribution network problems, providing an intuitive and effective analytical tool for power grid fault diagnosis, operation optimization, and knowledge management. It helps improve the intelligence level of power grid operation and management, and enhances the power grid's ability to cope with various problems.

[0209] S500: Input the typical problem graph into the conditional generative adversarial network to obtain the preliminary planning scheme set output by the conditional generative adversarial network, and generate the optimal operation planning scheme according to the preset constraint objectives.

[0210] In this embodiment of the invention, the intelligent solution generation system inputs the acquired typical problem graph into the conditional generative adversarial network (cGAN). The generation system of the conditional generative adversarial network generates preliminary planning solutions based on the typical problem graph, and the discriminator judges whether the generated solutions are real and reasonable, and finally obtains a set of preliminary planning solutions that are judged to be reasonable.

[0211] In this embodiment of the invention, the preset constraints within the intelligent scheme generation system include cost constraints (i.e., how to reduce operating costs), reliability constraints (such as reducing power outage time), and safety constraints (how to avoid equipment overload). A multi-objective optimization algorithm (such as NSGA-II) can be employed to iteratively optimize the generated preliminary planning scheme set using cost, reliability, and safety constraints as conditions. By calculating the fitness values ​​of different operational planning schemes under different objectives, the scheme parameters are continuously adjusted until the optimal operational planning scheme that satisfies the constraints is selected.

[0212] In this embodiment of the invention, taking the operation of a city's power distribution network during high summer temperatures as an example, the typical problem map obtained in step S400 is input into cGAN to generate a preliminary planning scheme set, such as adjusting transformer taps and adding backup lines. Then, based on constraints such as cost constraints (reducing equipment investment costs), safety constraints (ensuring voltage is within the normal range), and reliability constraints, the NSGA-II algorithm is used to iteratively optimize the schemes, ultimately determining the optimal operation planning scheme for high-temperature periods by adjusting some transformer taps and rationally allocating loads.

[0213] Example 2 is an embodiment of the present invention. This embodiment differs from the first embodiment in that it provides an intelligent generation system for power distribution network operation planning schemes.

[0214] It should be noted that the technical solution of the intelligent generation system for distribution network operation planning scheme is based on the same concept as the technical solution of the intelligent generation method for distribution network operation planning scheme described above. For details not described in detail in the technical solution of the intelligent generation system for distribution network operation planning scheme in this embodiment, please refer to the description of the technical solution of the intelligent generation method for distribution network operation planning scheme described above.

[0215] This embodiment describes an intelligent generation system for power distribution network operation planning schemes, comprising:

[0216] The acquisition module is used to acquire multi-source runtime data and multi-source standard data.

[0217] The feature fusion module is used to extract features from multi-source operational data and multi-source standard data through a multimodal large model, to obtain operational data features and standard data features, and to map the standard data features and operational data features to a unified semantic space to obtain a multimodal joint feature vector;

[0218] The analysis and mining module is used to perform spatiotemporal mining and analysis of multimodal joint feature vectors to determine spatiotemporal dependencies.

[0219] The graph construction module is used to construct typical problem graphs based on spatiotemporal dependencies, abnormal operation modes, and a preset power grid ontology knowledge base.

[0220] The scheme generation module is used to input typical problem graphs into the conditional generative adversarial network (GAN) to obtain a preliminary planning scheme set output by the GAN, and generate the optimal operation planning scheme according to the preset constraint objectives.

[0221] This embodiment also provides an electronic device applicable to a method for intelligently generating power distribution network operation planning schemes, including:

[0222] The system includes a memory and a processor. The memory stores computer-executable instructions, and the processor executes these instructions to implement a method for intelligently generating power distribution network operation planning schemes, as proposed in the above embodiments.

[0223] This embodiment also provides a storage medium storing a computer program, which, when executed by a processor, implements a method for intelligently generating a power distribution network operation planning scheme as proposed in the above embodiments.

[0224] The storage medium proposed in this embodiment and the intelligent generation method for a power distribution network operation planning scheme 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.

[0225] Based on the above description of the implementation methods, those skilled in the art can clearly understand that the present invention can be implemented using software and necessary general-purpose hardware, and of course, it can also be implemented using hardware, but in many cases the former is a better implementation method. 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.

[0226] 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 method for intelligently generating power distribution network operation planning schemes, characterized in that, include: Acquire multi-source operational data and multi-source standard data; By using a multimodal large model, features are extracted from multi-source operational data and multi-source standard data respectively to obtain operational data features and standard data features. The standard data features and operational data features are then mapped to a unified semantic space to obtain a multimodal joint feature vector. Spatiotemporal mining and analysis of multimodal joint feature vectors are performed to determine spatiotemporal dependencies. Based on spatiotemporal dependencies, abnormal operation modes, and a pre-set power grid ontology knowledge base, a typical problem map is constructed. The typical problem graph is input into the conditional generative adversarial network to obtain the preliminary planning scheme set output by the conditional generative adversarial network, and the optimal operation planning scheme is generated according to the preset constraint objectives. The spatiotemporal mining and analysis of multimodal joint feature vectors to determine spatiotemporal dependencies includes: Based on the regional distribution of the power distribution network, the multimodal joint feature vector is divided to obtain multiple regional sub-vectors; The temporal change rate is determined based on the regional sub-vectors, and the regional sub-vectors are filtered based on the temporal change rate to obtain abnormal time series segments. For a region pair consisting of the first and second sub-regions in multiple region sub-vectors, based on the overlap between the abnormal time series segment of the first sub-region and the time series change of the second sub-region in the region pair, all region pairs are traversed to construct an inter-regional correlation matrix; the first and second sub-regions are any regions in multiple region sub-vectors, and the first and second sub-regions are not the same. Based on the inter-regional correlation matrix, determine the temporal dependency weight of the first sub-region on the second sub-region; Based on multiple regional sub-vectors and temporal dependency weights, the spatiotemporal dependency relationship is determined; The process of mapping standard data features and operational data features to a unified semantic space to obtain a multimodal joint feature vector includes: Manifold learning and dimensionality reduction are performed on the standard data features and the operational data features respectively to obtain the standard data manifold features and the operational data manifold features; Clustering is performed on the standard data manifold features and the operational data manifold features respectively to obtain standard data category clusters and operational data category clusters; For any category cluster in the standard data category cluster and any category cluster in the running data category cluster, a correspondence matrix is ​​constructed based on the feature similarity between the category clusters in the category cluster pair; Based on the cosine similarity between the category clusters in each pair of category clusters in the correspondence matrix, preliminary fusion features are determined. The initial fused features are mapped onto a unified semantic space to obtain a multimodal joint feature vector.

2. The intelligent generation method for power distribution network operation planning schemes as described in claim 1, characterized in that: The determination of spatiotemporal dependencies based on multiple regional sub-vectors and temporal dependency weights includes: The spatial location and temporal dependency of each region pair are used as feature points, and the feature points are clustered to obtain a spatiotemporal dependency cluster. Based on the spatiotemporal dependency clusters and the temporal dependency weights obtained within the next preset sliding time window, the change in dependency weights is determined, and based on the change in dependency weights, abnormal spatiotemporal dependencies are determined. Causal verification is performed on the regional pairs in the abnormal spatiotemporal dependency based on the preset distribution network topology to obtain the causal verification results; Based on the temporal dependency weights and causal verification results, the multiple regional sub-vectors are fused to construct a global spatiotemporal dependency model, and the global spatiotemporal dependency model is determined as a spatiotemporal dependency relationship.

3. The intelligent generation method for power distribution network operation planning schemes as described in claim 2, characterized in that: The preliminary fusion features are determined based on the cosine similarity between the category clusters in each pair of category clusters in the correspondence matrix, including: For each pair of matching category clusters in the correspondence matrix, cosine similarity processing is performed between the category clusters to determine the adjacency matrix; Based on the adjacency matrix, determine the degree matrix, and based on the degree matrix, construct the Laplace matrix; The standard data category cluster and the running data category cluster are sequentially concatenated to obtain an initial fusion matrix; The initial fusion matrix is ​​optimized based on the Laplacian matrix to determine preliminary fusion characteristics.

4. The intelligent generation method for power distribution network operation planning schemes as described in claim 3, characterized in that: The process of mapping the initial fused features onto a unified semantic space to obtain a multimodal joint feature vector includes: Wavelet transform is applied to the preliminary fusion features to obtain time-frequency components at different scales; Based on the energy distribution of each time-frequency component, an attention weight is determined, and the time-frequency components are reconstructed by weighting based on the attention weight to obtain the reconstructed fusion feature; Based on the reconstructed fusion features, the time-frequency correlation matrix is ​​determined, and based on the time-frequency correlation matrix, a semantic transformation matrix is ​​constructed. Based on the semantic transformation matrix, feature mapping and alignment are performed on the reconstructed fusion features to obtain aligned fusion features; The aligned and fused features are subjected to nonlinear transformation and normalization to obtain a multimodal joint feature vector.

5. The intelligent generation method for power distribution network operation planning schemes as described in claim 4, characterized in that: The typical problem graph is constructed based on spatiotemporal dependencies, abnormal operation modes, and a pre-set power grid ontology knowledge base, including: Based on the aforementioned spatiotemporal dependencies, combined with the abnormal operation mode and the preset power grid ontology knowledge base, the set of problem elements is determined; Semantic analysis of the problem element set is performed based on the semantic network of the power grid ontology knowledge base to obtain semantic relationship analysis results; Based on the semantic relationship analysis results, each problem element in the problem element set is classified into a pre-constructed multi-layer typical problem graph to obtain a preliminary problem graph. The spatiotemporal dependency is then incorporated into the preliminary problem graph to obtain a fused problem graph. Logical verification of the fusion problem graph is performed based on the power grid ontology knowledge base, and the verification results are obtained. Based on the verification results, the fusion problem map is filtered and deleted to obtain the typical problem map.

6. A distribution network operation planning scheme intelligent generation system, using the distribution network operation planning scheme intelligent generation method as described in any one of claims 1-5, characterized in that, include: The acquisition module is used to acquire multi-source runtime data and multi-source standard data. The feature fusion module is used to extract features from multi-source operational data and multi-source standard data through a multimodal large model, to obtain operational data features and standard data features, and to map the standard data features and operational data features to a unified semantic space to obtain a multimodal joint feature vector; The analysis and mining module is used to perform spatiotemporal mining and analysis of multimodal joint feature vectors to determine spatiotemporal dependencies. The graph construction module is used to construct typical problem graphs based on spatiotemporal dependencies, abnormal operation modes, and a preset power grid ontology knowledge base. The scheme generation module is used to input typical problem graphs into the conditional generative adversarial network (GAN) to obtain a preliminary planning scheme set output by the GAN, and generate the optimal operation planning scheme according to the preset constraint objectives.

7. An electronic device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the intelligent generation method for power distribution network operation planning scheme according to any one of claims 1 to 5.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the intelligent generation method for power distribution network operation planning scheme according to any one of claims 1 to 5.

Images