Neural network embedding method and device for power distribution network state estimation and medium

Abstract

The invention discloses a neural network embedding method and device for power distribution network state estimation, electronic equipment and a medium. The method comprises the following steps: acquiring an input sequence; time information and node type information are embedded into a vector through space-time priori information embedding, and a space-time embedded vector is obtained; sampling interested nodes according to the power flow direction of the optimal power flow, and fusing node information to obtain a node embedding vector; using a graph isomorphic neural network to capture a local part of the graph, and embedding the local part into the feature vector to obtain a structure embedding vector; and fusing the input sequence and the three vectors, inputting into the graph space-time prediction network, and outputting a prediction result. By introducing the space-time prior information, the graph node embedding based on the optimal power flow direction and the graph structure embedding of the graph isomorphic neural network, the modeling of the power distribution network characteristics is realized, the defects of the prior art on the proprietary modeling of the power distribution network are made up, and the accuracy of the state estimation of the power distribution network is improved.

Description

Technical Field

[0001] The present invention relates to a power distribution network state estimation technology, and in particular to a neural network embedding method, device, electronic equipment and medium for power distribution network state estimation. Background Art

[0002] With the large-scale grid connection of new distributed power sources and electric vehicle charging stations, the distribution network faces highly fluctuating charging loads, which may lead to network instability and imbalance, thus seriously affecting its safe operation. In this context, real-time monitoring and state estimation are crucial to the operation of the distribution network. However, the spatiotemporal heterogeneity of the distribution network poses great challenges to real-time and accurate state estimation. In the spatial dimension, the node states and characteristics of different regions vary significantly; in the temporal dimension, the power load fluctuates greatly in different time periods and seasons. In the temporal dimension of state estimation, traditional methods usually assume that the network state changes slowly and the measurement noise and time series distribution are ideal. Under this assumption, methods such as least squares and Kalman filtering are often used for state estimation. However, the actual noise is random and may fluctuate violently in a short period of time, which makes it difficult for traditional methods to cope with it. In the spatial dimension of state estimation, traditional methods rely on redundant measurement data to correct and estimate the system state through multi-point measurement data. However, due to the lack of real-time synchronized phasor measurement equipment in the distribution network, the estimation of the overall network state becomes more difficult.

[0003] In order to solve these problems, data-driven neural network technology uses historical data of the distribution network to model and realize dynamic state estimation of the distribution network. This method can make up for the lack of real-time data and improve the accuracy and reliability of state estimation. Among them, recurrent neural network (RNN) and its variants can capture the temporal characteristics of long time series, but ignore the spatial information between nodes. Graph neural network (GNN) and its variants can capture the relationship and graph structure information between nodes and model complex topological structures. Combined with RNN or temporal convolutional network (TCN), it can achieve accurate prediction of spatiotemporal information. The general spatiotemporal graph neural network model can estimate the state of the distribution network, but there are still the following problems: (1) Lack of embedding of prior spatiotemporal knowledge. Ignoring time information: Existing models often ignore the embedding of time periods and dates, resulting in power consumption fluctuations in the distribution network, and different power consumption distributions in different seasons. For example, there will be a peak in daily power consumption during rest time, and power consumption in southern China will increase in summer. Ignoring spatial information: Existing models generally treat each node as homogeneous, ignoring the differences between nodes. For example, distributed power sources and loads have different properties. (2) Lack of modeling of the impact of physical properties between nodes. There are complex physical interactions between nodes in the distribution network. Factors such as the direction of power flow are difficult to accurately capture and model. (3) Lack of embedding of structural information. The structural information of the distribution network should be fully considered during modeling, including the adjacency relationship between nodes and the overall topological structure. The topological structure can reflect the regional information and geographical distribution of the distribution network, and will also affect the analysis of the overall status. When modeling, it is necessary to design a reasonable embedding algorithm to embed this information into the data characteristics of the distribution system. Summary of the invention

[0004] In order to solve at least one of the technical problems existing in the prior art to a certain extent, the object of the present invention is to provide a neural network embedding method, device, electronic device and medium for power distribution network state estimation.

[0005] The first technical solution adopted by the present invention is:

[0006] A neural network embedding method for power distribution network state estimation comprises the following steps:

[0007] Acquire an input sequence, wherein the input sequence includes node information (such as voltage, active power, reactive power, etc.), time information, node type, and connection information between nodes in the power distribution network;

[0008] By embedding the spatiotemporal prior information, the time information and node type information are embedded into the vector to obtain the spatiotemporal embedding vector;

[0009] The nodes of interest are sampled according to the power flow direction of the optimal power flow, and the node information is fused to obtain the node embedding vector;

[0010] Use graph isomorphism neural network to capture the local part of the graph, embed it into the feature vector, and obtain the structural embedding vector;

[0011] The input sequence, spatiotemporal embedding vector, node embedding vector and structure embedding vector are fused and input into the graph spatiotemporal prediction network to output the prediction result.

[0012] Furthermore, the graph spatiotemporal prediction network can be implemented using an existing graph neural network model.

[0013] Furthermore, the spatiotemporal prior information embedding includes three steps: time embedding, space embedding and spatiotemporal combination:

[0014] For the input sequence, after preprocessing, we get a node and a piece of data in the mid-day time period t d , day of the week t w , mid-year date t y and node type c;

[0015] Time embedding: Consider the time period embedding, weekday date embedding, and yearday date embedding of data collection to construct three embedding matrices, and use the time as the index to obtain the corresponding embedding vector;

[0016] Time period embedding: Assume N samples are taken every day d Get the time series and randomly initialize an embedding matrix Where D is the embedding dimension;

[0017] Weekday embedding: the number of days in a week is N w , randomly initialize an embedding matrix

[0018] Year-end date embedding: number of days per year is N y , randomly initialize an embedding matrix

[0019] Spatial embedding: Consider the node types in the distribution network, construct an embedding matrix, and use the node type and node number as indexes to obtain the corresponding embedding vector; a total of N c Types of nodes, randomly initialize a category embedding matrix

[0020] Time-space combination: According to the information of each node, query the four embedding matrices to obtain four embedding vectors with dimension D:

[0021] e d =E d [t d ]

[0022] e w =E w [t w ]

[0023] e y =E y [t y ]

[0024] e c =E c [c]

[0025] The four vectors are connected and passed through a multilayer perceptron (MLP) to obtain the spatiotemporal embedding vector.

[0026] Furthermore, the node types include: 1) load types: residential load nodes, commercial load nodes, electric vehicle charging stations, distributed energy storage nodes; 2) distributed power supply nodes: photovoltaic nodes, wind power nodes.

[0027] Furthermore, sampling nodes of interest according to the power flow direction of the optimal power flow and fusing node information to obtain node embedding vectors includes:

[0028] Calculate the optimal power flow of the distribution network; the optimal power flow means that the distribution network achieves the optimal performance index under the condition that all operating constraints are met;

[0029] Construct a distribution network graph: The distribution network is represented as a graph structure, where nodes represent nodes in the distribution network, edges are weighted edges, and the weights are the power changes of the corresponding lines;

[0030] Set the node walk algorithm: the hyperparameter is the number of node walks n w , return parameter P, control parameter q; set sampling probability, select nodes with power transfer according to weights to sample n w times, get a node list L;

[0031] Use word vector embedding model for training and embedding: treat the node list L as a sequence, each node as a word, and the node sequence generated by the entire walk process as a sentence. Use the word vector embedding model to learn the node vector representation; finally, obtain the high-dimensional vector representation of each node;

[0032] Construct graph node embedding matrix E based on high-dimensional vector representation N ∈R N×D, where N is the total number of nodes in the graph and D is the embedding dimension; using the graph node number as the index, the node embedding vector e is obtained in the embedding matrix N :

[0033] e N =E N [i].

[0034] Furthermore, the calculating of the optimal power flow of the power distribution network includes:

[0035] Determine the objective function:

[0036]

[0037] In the formula, f i (x) is the objective function; x is the set of decision variables, including the node voltage V, voltage phase angle θ, active power P and reactive power Q; N is the total number of nodes;

[0038] Identify the constraints:

[0039]

[0040] Where P Gi is the active power generation of the ith node, P Di is the active power load of the ith node; Q Gi is the reactive power generation of the ith node, Q Di is the reactive power load of the ith node; V i is the voltage of the ith node, V j is the voltage of the jth node; θ i is the voltage phase angle of the ith node, θ j is the voltage phase angle of the jth node; G ij is the real part of the line admittance between node i and node j, B ij is the imaginary part of the line admittance between nodes j and j, is the minimum value of active power generation of the ith node, is the maximum value of active power generation of the ith node, is the minimum value of reactive power generation of the ith node, is the maximum value of reactive power generation of the ith node, is the minimum allowable value of the voltage amplitude of the ith node, is the maximum allowable value of the voltage amplitude of the ith node;

[0041] The parameters of the optimal power flow are solved to obtain the inflow and outflow power of each node under the optimal power flow conditions, thereby obtaining the power flow direction.

[0042] Further, the setting of the sampling probability includes:

[0043] Let V be the node sampled last time, X be the current sampling node, and set the probability of sampling the last sampled node V to When the node connection weight w xi The sampling probability is weighted by w xi , when the node connection weight w xi When it is negative, the sampling probability is

[0044] Furthermore, the use of a graph isomorphism neural network to capture a part of the graph and embed it into a feature vector to obtain a structural embedding vector includes:

[0045] Construct a power distribution network graph, wherein the nodes of the graph are the nodes of the power distribution network;

[0046] Preprocess the distribution network graph to generate node feature vectors;

[0047] The distribution network graph is embedded using a graph isomorphic neural network to obtain a structural embedding vector.

[0048] Furthermore, the graph isomorphism neural network works as follows:

[0049] The graph isomorphic neural network includes multiple GIN layers, each of which is responsible for aggregating neighborhood information of nodes;

[0050] In each layer of GIN, the aggregate features of the nodes are calculated. The specific formula is:

[0051]

[0052] In the formula, represents the feature vector of node v at layer k, represents the set of neighbor nodes of node v, MLP represents multi-layer perceptron, ∈ represents a learnable irrational number parameter;

[0053] The output feature vectors of the multi-layer GIN layer are processed by the fully connected layer and the activation function to obtain the final structural embedding vector.

[0054] The second technical solution adopted by the present invention is:

[0055] An electronic device comprises a processor and a memory, wherein the memory stores at least one instruction, at least one program, a code set or an instruction set, and the at least one instruction, the at least one program, the code set or the instruction set is loaded and executed by the processor to implement a neural network embedding method for power distribution network state estimation as described above.

[0056] The third technical solution adopted by the present invention is:

[0057] A computer-readable storage medium, wherein at least one instruction, at least one program, code set or instruction set is stored in the storage medium, wherein the at least one instruction, the at least one program, the code set or instruction set is loaded and executed by a processor to implement a neural network embedding method for power distribution network state estimation as described above.

[0058] The fourth technical solution adopted by the present invention is:

[0059] A computer program product or a computer program includes computer instructions stored in a computer-readable storage medium. A processor of a computer device can read the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computer device executes the above method.

[0060] The beneficial effects of the present invention are as follows: the present invention realizes the modeling of distribution network characteristics by introducing spatiotemporal prior information, graph node embedding based on optimal power flow direction, and graph structure embedding of graph isomorphic neural network, takes into account the proprietary properties of the distribution network, makes up for the shortcomings of the prior art in proprietary modeling of distribution networks, and effectively improves the accuracy of distribution network state estimation. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the embodiments of the present invention or the drawings of related technical solutions in the prior art are introduced below. It should be understood that the drawings introduced below are only for the convenience of clearly describing some embodiments of the technical solutions of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative work.

[0062] Figure 1 is an overall flow chart of the distribution network state estimation model in an embodiment of the present invention;

[0063] Figure 2 is an overall flow chart of time-space embedding in an embodiment of the present invention;

[0064] Figure 3 is an improved node wandering probability graph in an embodiment of the present invention;

[0065] Figure 4 Schematic diagram of training a sampled sequence by a word vector embedding model in an embodiment of the present invention;

[0066] Figure 5 It is a flowchart of the steps of a neural network embedding method for power distribution network state estimation in an embodiment of the present invention. DETAILED DESCRIPTION

[0067] The embodiments of the present invention are described in detail below, and examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and are not to be construed as limitations of the present invention. For the step numbers in the following embodiments, they are only provided for the convenience of explanation, and the order between the steps is not limited in any way, and the execution order of each step in the embodiment can be adaptively adjusted according to the understanding of those skilled in the art.

[0068] In the description of the present invention, it should be understood that descriptions involving orientations, such as up, down, front, back, left, right, etc., and orientations or positional relationships indicated are based on the orientations or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be understood as a limitation on the present invention.

[0069] In the description of the present invention, "several" means one or more, "more" means more than two, "greater than", "less than", "exceed" etc. are understood as not including the number itself, and "above", "below", "within" etc. are understood as including the number itself. If there is a description of "first" or "second", it is only used for the purpose of distinguishing the technical features, and cannot be understood as indicating or implying the relative importance or implicitly indicating the number of the indicated technical features or implicitly indicating the order of the indicated technical features.

[0070] In the description of the present invention, unless otherwise clearly defined, terms such as setting, installing, connecting, etc. should be understood in a broad sense, and technicians in the relevant technical field can reasonably determine the specific meanings of the above terms in the present invention based on the specific content of the technical solution.

[0071] In order to incorporate more information into the graph neural network model of distribution network state estimation, the present invention proposes an information embedding scheme for the distribution network state estimation task. The scheme can effectively embed spatiotemporal prior information, power flow direction and graph topology information. Specifically, it includes the following three main methods: (1) Spatiotemporal prior information embedding: embedding time information (time of day, daily cycle) and spatial information (such as node type) into the feature vector. (2) Graph node embedding based on optimal power flow direction: using the optimal power flow direction to guide the node embedding process to ensure that the power transmission path and direction information are fully utilized. (3) Graph structure embedding based on graph isomorphism neural network: using a graph isomorphism neural network (GraphIsomorphism Network, GIN) to capture the local part of the graph and embed it into the feature vector.

[0072] The present invention fully considers the proprietary nature of the distribution network, and realizes the modeling of the distribution network characteristics by introducing spatiotemporal prior information, graph node embedding based on the optimal power flow direction, and graph structure embedding of graph isomorphic neural networks. First, the spatiotemporal information and node type information unique to the distribution network are added, so that the model can more accurately reflect the actual situation of the distribution network. Secondly, it can adapt to the existing graph neural network model, and the newly added features can be directly used for training and inference without making major modifications to the original model architecture. In addition, the solution of the present invention is optimized in terms of computational cost. Through carefully designed embedding strategies and efficient graph isomorphic neural network algorithms, only a limited amount of computation is added, which will not impose an excessive burden on the computing power requirements of the existing system.

[0073] Example 1

[0074] like Figure 1 and Figure 5 As shown, this embodiment provides a neural network embedding method for power distribution network state estimation, comprising the following steps:

[0075] S1. Obtain an input sequence, where the input sequence includes node information, time information, node type, and connection information between nodes in the power distribution network;

[0076] S2, embed the time information and node type information into the vector through spatiotemporal prior information embedding to obtain the spatiotemporal embedding vector;

[0077] S3, sampling nodes of interest according to the power flow direction of the optimal power flow, and fusing node information to obtain node embedding vectors;

[0078] S4. Use graph isomorphism neural network to capture the local part of the graph, embed it into the feature vector, and obtain the structural embedding vector;

[0079] S5: Fusion the input sequence, spatiotemporal embedding vector, node embedding vector, and structure embedding vector, and input them into the graph spatiotemporal prediction network to output the prediction result.

[0080] Figure 1 The overall process of the embedding method is described. The input data (i.e., input sequence) includes node information in the distribution network (such as voltage, active power, reactive power, etc.), time information, node type, and connection information between nodes. Spatiotemporal prior information embedding embeds time and node type information into a vector of a specific dimension. Graph node embedding samples the nodes of interest according to the power flow direction of the optimal power flow, and fuses the node information and embeds it into a vector. Graph structure embedding uses a graph isomorphic network to extract graph structure information, ensuring that subsequent models can distinguish isomorphic distribution network graph structures. The three modules obtain the embedded vector and add it to the input sequence as the input of the graph spatiotemporal prediction network.

[0081] The above method is explained in detail below with reference to the accompanying drawings and specific embodiments.

[0082] (1) Embedding spatiotemporal prior information

[0083] Figure 2 It represents the overall process of the spatiotemporal embedding method, which mainly includes two parts: time embedding and space embedding.

[0084] For the input data, after preprocessing, we can get the mid-day time period t for one node and one piece of data. d , day of the week t w , mid-year date t y and node type c.

[0085] Time embedding: Consider the time period embedding, weekday date embedding, and yearly date embedding of data collection to construct three embedding matrices, and use the time as the index to obtain the corresponding embedding vector.

[0086] Time period embedding: Assume N samples are taken every day d Get the time series and randomly initialize an embedding matrix Where D is the embedding dimension.

[0087] Weekday embedding: There are seven days in a week, namely N w is 7, randomly initialize an embedding matrix

[0088] Year date embedding: Each year has 366 or 365 days, set N y 366 or 365, randomly initialize an embedding matrix

[0089] Spatial embedding: Consider the node types in the distribution network: (1) load types: residential load nodes, commercial load nodes, electric vehicle charging stations, distributed energy storage nodes, etc., (2) distributed power generation nodes: photovoltaic nodes, wind power nodes, etc. Construct an embedding matrix and use the node type and node number as indexes to obtain the corresponding embedding vector.

[0090] Total N c Types of nodes, randomly initialize a category embedding matrix

[0091] Time-space combination: According to the information of each node, query the above four embedding matrices to obtain four embedding vectors with dimension D

[0092] e d =E d [t d ] (1)

[0093] ew =E w [t w ] (2)

[0094] e y =E y [t y ] (3)

[0095] e c =E c [c] (4)

[0096] Connect the four vectors and pass them through a multilayer perceptron (MLP) to get the spatiotemporal embedding vector:

[0097] e concat =[e d ;e w ;e y ;e c ]∈R 4D (5)

[0098] e st =MLP(e concat )∈R D (6)

[0099] (2) Graph node embedding based on optimal power flow direction

[0100] The main algorithm improves the walking strategy of the graph node embedding algorithm so that the embedded nodes can reflect the scope of their influence and being influenced. The main steps of the algorithm include:

[0101] Step 1. Calculate the optimal power flow of the distribution network: The optimal power flow means that the distribution network achieves the optimal performance indicator under the condition of satisfying all operating constraints.

[0102] Objective function (minimizing a certain cost or loss):

[0103]

[0104] In the formula, f i (x) is the objective function; x is the set of decision variables, including the node voltage V, voltage phase angle θ, active power P and reactive power Q; N is the total number of nodes; c i is the weight of the ith node.

[0105] The following constraints are included:

[0106]

[0107] The constraints include power balance constraints, generator output constraints, node voltage constraints, etc.

[0108] Use Newton's method, interior point method or intelligent optimization algorithm to solve the parameters of the optimal power flow, obtain the inflow and outflow power of each node under the optimal power flow conditions, and thus obtain the power flow direction.

[0109] Step 2. Construct a distribution network graph: The distribution network is represented as a graph structure, where nodes represent nodes in the distribution network, edges are weighted edges, and the weights are the power changes of the corresponding lines.

[0110] Step 3. Set the node walk algorithm: the hyperparameter is the number of node walks n w , return parameter P, control parameter q. Set the sampling probability, such as Figure 3 , V node is the node sampled last time, X is the current sampling node, and the probability of sampling the last sampled node V is set to When the node connection weight w xi The sampling probability is weighted by w xi , when the node connection weight w xi When it is negative, the sampling probability is

[0111]

[0112] Sample nodes with power transfer according to weight selection n w times, and obtain a node list L.

[0113] Step 4. Use the word embedding model for training and embedding: treat the node list L as a sequence, each node as a word, and the node sequence generated by the entire walk process as a sentence. Use the Skip-gram method of the word embedding model to learn the node vector representation. The specific process is as follows Figure 4 :Each node and its node pair in the context window are used as training samples. The size of the context window can be adjusted according to the specific application. Through the neural network model, the node vector is trained so that the adjacent nodes are closer in the vector space, thereby reflecting the structural and functional relationship between the nodes. Finally, a high-dimensional vector representation of each node is obtained.

[0114] Step 5. Construct graph node embedding matrix E N ∈R N×D , where N is the total number of nodes in the graph and D is the embedding dimension. Using the graph node number as the index, the embedding vector e is obtained in the embedding matrix N :

[0115] e N =E N [i] (14)

[0116] Parameter explanation: adjust the number of node walks nw The range of node wandering can be controlled, thereby controlling the node's receptive field. Adjusting the return parameter p can control the tendency to sample the nodes just visited, and the wandering parameter q can control the tendency to sample in the direction of power inflow or power outflow.

[0117] (3) Graph Structure Embedding Based on Graph Isomorphic Neural Network

[0118] The purpose of graph structure embedding is to add the structural features of the distribution network to the features. Graph isomorphic neural network has the function of distinguishing isomorphic graphs, so we use it to extract the structural features of the distribution network. Graph structure embedding is achieved through the following steps:

[0119] Step 1. Construct a distribution network graph, where the nodes are nodes of the distribution network.

[0120] Step 2: Preprocess the distribution network diagram to generate node feature vectors.

[0121] Step 3. Use graph isomorphism neural network to embed the distribution network graph.

[0122] As an optional implementation, step 3 includes the following steps:

[0123] 1) Define the graph isomorphic neural network model structure, including multiple GIN layers, each of which is responsible for aggregating the neighborhood information of the node.

[0124] 2) In each layer of GIN, calculate the aggregate features of the nodes. The specific formula is:

[0125]

[0126] in represents the feature vector of node v at layer k, represents the set of neighbor nodes of node v, and MLP represents multi-layer perceptron.

[0127] 3) The output feature vectors of the multi-layer GIN layer are processed by the fully connected layer and activation function to obtain the final structural embedding vector.

[0128] In summary, the existing distribution network state estimation methods based on graph neural networks mainly rely on the powerful feature extraction capabilities of neural networks and large-scale high-quality distribution network data, which are more accurate than traditional state estimation methods. However, these models all use general graph structures for modeling, ignore the unique properties of distribution networks, and lack proprietary modeling of distribution networks. The embedding algorithm used in the existing models is only a general graph spatiotemporal embedding, which cannot effectively reflect the characteristics of the distribution network. The present invention designs a method for embedding the properties of the distribution network into the distribution network graph, which includes the embedding of spatiotemporal prior information, the embedding of graph nodes based on the optimal power flow direction, and the embedding of graph structures based on graph isomorphic neural networks, providing additional data feature support for the existing models. Specifically, a time encoding module is designed to embed week and date information, and the node type information is embedded in combination with the space encoding module, and these category information is learned through the embedding layer; a graph node embedding algorithm walking strategy based on the optimal power flow direction is adopted to embed each node, making full use of the information of the power transmission direction in the distribution network; and the local graph structure is embedded using a graph isomorphic network to better capture the structural information of the distribution network. Through these improvements, the solution of the present invention can improve the accuracy of distribution network state estimation and make up for the shortcomings of the prior art in proprietary modeling of distribution networks.

[0129] Example 2

[0130] An embodiment of the present invention further provides an electronic device, the electronic device comprising a processor and a memory, the memory storing at least one instruction, at least one program, a code set or an instruction set, the at least one instruction, the at least one program, the code set or the instruction set being loaded and executed by the processor to implement the following Figure 1 or Figure 5 A neural network embedding method for distribution network state estimation is shown.

[0131] It is understood that the memory may include a random access memory (RAM) or a read-only memory (ROM). Optionally, the memory includes a non-transitory computer-readable storage medium. The memory may be used to store instructions, programs, codes, code sets, or instruction sets. The memory may include a program storage area and a data storage area, wherein the program storage area may store instructions for implementing an operating system, instructions for at least one function, instructions for implementing the above-mentioned various method embodiments, etc.; the data storage area may store data created according to the use of the server, etc.

[0132] The processor may include one or more processing cores. The processor uses various interfaces and lines to connect the various parts of the entire server, and executes various functions of the server and processes data by running or executing instructions, programs, code sets or instruction sets stored in the memory, and calling data stored in the memory. Optionally, the processor can be implemented in at least one hardware form of digital signal processing (DSP), field programmable gate array (FPGA), and programmable logic array (PLA). The processor can integrate one or a combination of a central processing unit (CPU) and a modem. Among them, the CPU mainly processes the operating system and application programs; the modem is used to process wireless communications. It can be understood that the above-mentioned modem may not be integrated into the processor, but implemented separately through a chip.

[0133] Since the electronic device is an electronic device corresponding to a neural network embedding method for power distribution network state estimation in an embodiment of the present invention, and the principle of solving the problem by the electronic device is similar to that of the method, the implementation of the electronic device can refer to the implementation process of the above-mentioned method embodiment, and the repeated parts will not be repeated.

[0134] Example 3

[0135] The embodiment of the present invention further provides a computer-readable storage medium, wherein the storage medium stores at least one instruction, at least one program, code set or instruction set, and the at least one instruction, the at least one program, the code set or instruction set is loaded and executed by a processor to implement the following Figure 1 or Figure 5 A neural network embedding method for distribution network state estimation is shown.

[0136] Those skilled in the art can understand that all or part of the steps in the various methods of the above embodiments can be completed by instructing related hardware through a program, and the program can be stored in a computer-readable storage medium, and the storage medium includes a read-only memory (ROM), a random access memory (RAM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), a one-time programmable read-only memory (OTPROM), an electronically erasable rewritable read-only memory (EEPROM), a compact disc (CD-ROM) or other optical disc storage, magnetic disk storage, magnetic tape storage, or any other computer-readable medium that can be used to carry or store data.

[0137] Since the storage medium is a storage medium corresponding to a neural network embedding method for power distribution network state estimation in an embodiment of the present invention, and the principle of solving the problem by the storage medium is similar to that of the method, the implementation of the storage medium can refer to the implementation process of the above-mentioned method embodiment, and the repeated parts will not be repeated.

[0138] Example 4

[0139] In some possible implementations, various aspects of the method of the embodiment of the present invention may also be implemented in the form of a program product, which includes a program code. When the program product is run on a computer device, the program code is used to enable the computer device to execute the steps of a neural network embedding method for power distribution network state estimation according to various exemplary embodiments of the present application described above in this specification. Among them, the executable computer program code or "code" for executing various embodiments can be written in a high-level programming language such as C, C++, C#, Smalltalk, Java, JavaScript, Visual Basic, structured query language (e.g., Transact-SQL), Perl, or in various other programming languages.

[0140] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, a plurality of steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0141] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification and the features of the different embodiments or examples, without contradiction.

[0142] The above embodiments are only for illustrating the technical concept and features of the present invention, and their purpose is to enable ordinary technicians in the field to understand the content of the present invention and implement it accordingly, and they cannot be used to limit the protection scope of the present invention. Any equivalent changes or modifications made based on the essence of the content of the present invention should be included in the protection scope of the present invention.

Claims

1. A neural network embedding method for power distribution network state estimation, characterized in that: The following steps are involved: Acquire an input sequence, wherein the input sequence includes node information, time information, node type, and connection information between nodes in the power distribution network; By embedding the spatiotemporal prior information, the time information and node type information are embedded into the vector to obtain the spatiotemporal embedding vector; the nodes of interest are sampled according to the power flow direction of the optimal power flow, and the node information is fused to obtain the node embedding vector; Use graph isomorphism neural network to capture the local part of the graph, embed it into the feature vector, and obtain the structural embedding vector; The input sequence, spatiotemporal embedding vector, node embedding vector and structure embedding vector are fused and input into the graph spatiotemporal prediction network to output the prediction result.

2. A neural network embedding method for power distribution network state estimation according to claim 1, characterized in that: The spatiotemporal prior information embedding includes three steps: time embedding, space embedding and spatiotemporal combination: For the input sequence, after preprocessing, we get a node and a piece of data in the mid-day time period t d , day of the week t w , mid-year date t y and node type c; Time embedding: Consider the time period embedding, weekday date embedding, and yearday date embedding of data collection to construct three embedding matrices, and use the time as the index to obtain the corresponding embedding vector; Time period embedding: Assume N samples are taken every day d Get the time series and randomly initialize an embedding matrix Where D is the embedding dimension; Weekday embedding: the number of days in a week is N w , randomly initialize an embedding matrix Year-end date embedding: number of days per year is N y , randomly initialize an embedding matrix Spatial embedding: Consider the node types in the distribution network, construct an embedding matrix, and use the node type and node number as indexes to obtain the corresponding embedding vector; a total of N c Types of nodes, randomly initialize a category embedding matrix Time-space combination: According to the information of each node, query the four embedding matrices to obtain four embedding vectors with dimension D: And d =And d [t d ] And w =And w [t w ] And y =And y [t y ] And c =And c [c] The four vectors are concatenated and passed through a multi-layer perceptron to obtain the spatiotemporal embedding vector.

3. A neural network embedding method for power distribution network state estimation according to claim 2, characterized in that: The node types include: 1) load types: residential load nodes, commercial load nodes, electric vehicle charging stations, distributed energy storage nodes; 2) distributed power supply nodes: photovoltaic nodes, wind power nodes.

4. A neural network embedding method for power distribution network state estimation according to claim 1, characterized in that: The sampling of nodes of interest according to the power flow direction of the optimal power flow and fusing the node information to obtain the node embedding vector include: Calculate the optimal power flow of distribution networks; Construct a distribution network graph: The distribution network is represented as a graph structure, where nodes represent nodes in the distribution network, edges are weighted edges, and the weights are the power changes of the corresponding lines; Set the node walk algorithm: the hyperparameter is the number of node walks n w , return parameter P, control parameter q; set sampling probability, select nodes with power transfer according to weights to sample n w times, get a node list L; Use word vector embedding model for training and embedding: treat the node list L as a sequence, each node as a word, and the node sequence generated by the entire walk process as a sentence. Use the word vector embedding model to learn the node vector representation; finally, obtain the high-dimensional vector representation of each node; Construct graph node embedding matrix E based on high-dimensional vector representation N ∈R N×D , where N is the total number of nodes in the graph and D is the embedding dimension; using the graph node number as the index, the node embedding vector e is obtained in the embedding matrix N : yes N =E N [i].

5. A neural network embedding method for power distribution network state estimation according to claim 4, characterized in that: The method of calculating the optimal power flow of the power distribution network comprises: Determine the objective function: In the formula, f i (x) is the objective function; x is the set of decision variables, including the node voltage V, voltage phase angle θ, active power P and reactive power Q; N is the total number of nodes; Identify the constraints: Where P Gi is the active power generation of the ith node, P Di is the active power load of the ith node; Q Gi is the reactive power generation of the ith node, Q Di is the reactive power load of the ith node; V i is the voltage of the ith node, V j is the voltage of the jth node; θ i is the voltage phase angle of the ith node, θ j is the voltage phase angle of the ith node; G ij is the real part of the line admittance between node i and node j, B ij is the imaginary part of the line admittance between node i and node j, is the minimum value of active power generation of the ith node, is the maximum value of active power generation of the ith node, is the minimum value of reactive power generation of the ith node, is the maximum value of reactive power generation of the ith node, is the minimum allowable value of the voltage amplitude of the ith node, is the maximum allowable value of the voltage amplitude of the ith node; The parameters of the optimal power flow are solved to obtain the inflow and outflow power of each node under the optimal power flow conditions, thereby obtaining the power flow direction.

6. A neural network embedding method for power distribution network state estimation according to claim 4, characterized in that: The setting of the sampling probability comprises: Let V be the node sampled last time, X be the current sampling node, and set the probability of sampling the last sampled node V to When the node connection weight w xi The sampling probability is weighted by w xi , when the node connection weight w xi When it is negative, the sampling probability is 7. A neural network embedding method for power distribution network state estimation according to claim 1, characterized in that: The method of using a graph isomorphism neural network to capture a part of the graph and embed it into a feature vector to obtain a structural embedding vector includes: Construct a power distribution network graph, wherein the nodes of the graph are the nodes of the power distribution network; Preprocess the distribution network graph to generate node feature vectors; The distribution network graph is embedded using a graph isomorphic neural network to obtain a structural embedding vector.

8. A neural network embedding method for power distribution network state estimation according to claim 7, characterized in that: The graph isomorphism neural network works as follows: The graph isomorphic neural network includes multiple GIN layers, each of which is responsible for aggregating neighborhood information of nodes; In each layer of GIN, the aggregate features of the nodes are calculated. The specific formula is: In the formula, represents the feature vector of node v at layer k, represents the set of neighbor nodes of node v, MLP represents multi-layer perceptron, ∈ represents a learnable irrational number parameter; The output feature vectors of the multi-layer GIN layer are processed by the fully connected layer and the activation function to obtain the final structural embedding vector.

9. An electronic device, characterized in that: The electronic device includes a processor and a memory, wherein the memory stores at least one instruction, at least one program, a code set or an instruction set, and the at least one instruction, the at least one program, the code set or the instruction set is loaded and executed by the processor to implement the method described in any one of claims 1 to 8.

10. A computer-readable storage medium, characterized in that: The storage medium stores at least one instruction, at least one program, a code set or an instruction set, and the at least one instruction, the at least one program, the code set or the instruction set is loaded and executed by the processor to implement the method according to any one of claims 1 to 8.

Images