Regulation and control method, device and equipment of three-phase power distribution network and medium

Abstract

The invention discloses a three-phase power distribution network regulation and control method, device and equipment and a medium, and belongs to the field of power distribution network regulation and control, and the method comprises the steps: constructing a physical information graph of a three-phase power distribution network, and extracting spatial-temporal characteristics from the physical information graph; determining the equipment information of the three-phase power distribution network according to the spatial-temporal characteristics: the equipment switch state of the three-phase power distribution network and the position information of a voltage regulator; constructing an overall loss function of the model by using the equipment information, and training a space-time diagram convolutional network based on the overall loss function; and the prediction processing of the space-time diagram convolutional network is simplified into an initial problem, a preset solver is called to solve the initial problem to obtain a regulation and control scheme, and the three-phase power distribution network is regulated and controlled according to the regulation and control scheme. According to the method, the physical information graph is constructed, the spatial-temporal characteristics are extracted from the physical information graph, prediction is carried out according to the spatial-temporal characteristics, the actual state of the power distribution network can be fit, the regulation and control scheme meeting the requirements of the power distribution network is generated, the regulation and control precision is improved, and regulation and control errors are reduced.

Description

Technical Field

[0001] This invention relates to the technical field of power distribution network control, and more particularly to a control method, device, equipment and medium for a three-phase power distribution network. Background Technology

[0002] Dynamic Network Reconfiguration (DNR) has become an effective method for reducing network losses in distribution systems, regulating bus voltage, and improving system resilience. With the increasing integration of distributed energy resources (DERs) into the distribution network (e.g., the increasing penetration of wind and solar photovoltaic power), the active power loss of the system will be reduced proportionally. Simultaneously, due to the increasing penetration of DERs, voltage regulation in unbalanced distribution systems is becoming increasingly complex and critical, and voltage fluctuations are strongly influenced by the topology of the distribution system.

[0003] Therefore, it is necessary to restructure the distribution network to improve its stability. One common approach is to construct an optimal network refactoring problem for the unbalanced distribution system, including renewable energy sources (DERs) and voltage regulators (VRs), considering the timing of distributed energy access, distribution, and transmission. This optimal network refactoring problem is defined as a mixed-integer semidefinite programming problem with slack variables. By solving the optimal network refactoring problem, a refactoring scheme can be obtained, which can then regulate the unbalanced distribution system to reduce voltage fluctuations.

[0004] However, the commonly used methods have the following technical problems: Since unbalanced power distribution systems operate for long periods of time, the optimal network reconstruction problem constructed for a specific time is difficult to adapt to the actual working conditions of the power distribution system, the solution results deviate from the actual situation of the power distribution system, and the control accuracy is low. Summary of the Invention

[0005] This invention provides a method, device, equipment, and medium for regulating a three-phase power distribution network, which can solve the technical problems of deviation between the regulation scheme solved by the prior art and the actual situation, and low regulation accuracy.

[0006] A first aspect of this invention provides a control method for a three-phase distribution network, the method comprising: Construct a physical information map of a three-phase distribution network and extract spatiotemporal features from the physical information map; The equipment information of the three-phase power distribution network is determined based on the spatiotemporal characteristics, wherein the equipment information includes: the switching status of the equipment in the three-phase power distribution network and the location information of the voltage regulator; The overall loss function of the model is constructed using the device information, and a spatiotemporal graph convolutional network is trained based on the overall loss function; The prediction process of the spatiotemporal graph convolutional network is simplified into an initial problem, and a preset solver is called to solve the initial problem to obtain a control scheme. The three-phase power distribution network is then controlled according to the control scheme.

[0007] This invention constructs a physical information map and extracts its spatiotemporal features. Based on these features, predictions can be made that align with the actual state of the power distribution network, thereby generating control schemes that meet the needs of the power distribution network, thus improving the accuracy of control and reducing control errors. Furthermore, during the prediction process, the model's predictions are transformed into problems for solving, which simplifies the processing flow and reduces computational complexity, thereby improving processing efficiency.

[0008] In conjunction with the first aspect, in one implementation, constructing a physical information map of a three-phase distribution network and extracting spatiotemporal features from the physical information map includes: Obtain the node set, the edge set, and the line segment information of the three-phase distribution network, respectively, wherein the line segment information is the adjacency matrix of the line segment connectivity; A physical information graph is constructed using the node set, the edge set of the line, and the line segment information; The physical information graph is subjected to node feature definition and vector adjustment processing to obtain an adjusted information graph; Spatial features are extracted from the adjustment information map using a preset Chebyshev convolutional layer, and temporal features are extracted from the adjustment information map using a preset causal convolutional layer; The spatiotemporal features are obtained by combining the temporal and spatial features using a preset spatiotemporal convolutional block.

[0009] In conjunction with the first aspect, in one implementation, the step of defining node features and adjusting vectors in the physical information graph to obtain an adjusted information graph includes: The physical information graph is defined by defining node features to obtain a defined information graph, wherein the node features include: the vector of the maximum active power generation of the three-phase distributed energy, the three-phase active load demand, and the three-phase reactive load demand; The defined information graph is subjected to value replacement processing to obtain an adjusted information graph. The value replacement processing involves setting the vector of maximum active power generation, three-phase active load demand and three-phase reactive load demand of non-distributed energy or non-load-bearing nodes as zero vectors, and using preset values ​​to represent non-three-phase buses.

[0010] In conjunction with the first aspect, in one implementation, determining the equipment information of the three-phase distribution network based on the spatiotemporal characteristics includes: A preset link classifier is invoked to perform a Hadamard product operation on the spatiotemporal features to obtain the edge embedding of each device in the three-phase power distribution network. The device scalar is obtained by summing the edge embeddings of the switch on the feature dimension, and the device switch state is obtained by using the sigmoid function based on the device scalar. The edge embedding of the voltage regulator is preprocessed using regular expressions to obtain processed embeddings, and the probability distribution of the tap positions of the voltage regulator is calculated using the processed embeddings to obtain the position information of the voltage regulator.

[0011] In conjunction with the first aspect, in one implementation, the step of constructing an overall loss function for the model using the device information and training a spatiotemporal graph convolutional network based on the overall loss function includes: A first loss function and a second loss function are constructed using the device's on / off state, and a third loss function is constructed using the location information; The overall loss function is obtained by combining the first loss function, the second loss function, and the third loss function. A prediction network for predicting the tap position and switching state of a voltage regulator is constructed based on the overall loss function, and the prediction network is trained to obtain a spatiotemporal graph convolutional network.

[0012] In conjunction with the first aspect, in one implementation, the prediction processing of the spatiotemporal graph convolutional network is simplified into an initial problem, and a preset solver is invoked to solve the initial problem to obtain a control scheme. The control scheme is then used to regulate the three-phase power distribution network, including: Based on the binary decision variable fixing algorithm of the spatiotemporal graph convolutional network, the prediction process of the spatiotemporal graph convolutional network is simplified into an initial problem; The initial problem is solved by calling a preset solver to obtain a control scheme. The preset solver is a MID solver constructed by a preset objective function. The equipment in the three-phase power distribution network is regulated and controlled according to the aforementioned control scheme.

[0013] In conjunction with the first aspect, in one implementation, the method for determining the constraints of the preset objective function includes: The output deviation, voltage deviation, number of switching actions, and voltage regulator adjustment amount of the three-phase distribution network are obtained, wherein the output deviation is to minimize the output deviation of the three-phase distribution network under distributed energy. An objective function is constructed using the output deviation, the voltage deviation, the number of switching actions, and the voltage regulator adjustment amount, and the constraints of the objective function are determined. The constraints include: power flow constraints, voltage constraints, voltage regulator planning constraints, network radiation constraints, distributed energy output constraints, branch capacity constraints, and switching operation constraints.

[0014] A second aspect of the present invention provides a control device for a three-phase distribution network, the device comprising: An extraction module is used to construct a physical information map of a three-phase power distribution network and extract spatiotemporal features from the physical information map; The determination module is used to determine the equipment information of the three-phase power distribution network based on the spatiotemporal characteristics, wherein the equipment information includes: the switching status of the equipment in the three-phase power distribution network and the location information of the voltage regulator; The training module is used to construct the overall loss function of the model using the device information, and to train the spatiotemporal graph convolutional network based on the overall loss function; The control module is used to simplify the prediction processing of the spatiotemporal graph convolutional network into an initial problem, and call a preset solver to solve the initial problem to obtain a control scheme, and then control the three-phase power distribution network according to the control scheme.

[0015] Compared to existing technologies, the present invention provides a method, apparatus, equipment, and medium for regulating a three-phase distribution network, which offers the following advantages: The present invention can construct a physical information map of the three-phase distribution network and extract spatiotemporal features from the physical information map; determine the equipment information of the three-phase distribution network based on the spatiotemporal features; construct the overall loss function of the model using the equipment information, and train a spatiotemporal graph convolutional network based on the overall loss function; simplify the prediction processing of the spatiotemporal graph convolutional network into an initial problem, and call a preset solver to solve the initial problem to obtain a regulation scheme; and regulate the three-phase distribution network according to the regulation scheme. By constructing a physical information map and extracting spatiotemporal features from it, and making predictions based on these features, the present invention can closely match the actual state of the distribution network, thereby generating a regulation scheme that meets the needs of the distribution network, thus improving the accuracy of regulation and reducing errors; furthermore, by converting the model's prediction into a problem to be solved during prediction, the processing flow and computational complexity can be simplified, thereby improving processing efficiency. Attached Figure Description

[0016] Figure 1 This is a schematic flowchart of a control method for a three-phase power distribution network provided in an embodiment of the present invention; Figure 2 This is an operation flowchart of a three-phase power distribution network control method provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of the structure of a control device for a three-phase power distribution network provided in an embodiment of the present invention. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] Dynamic Network Reconfiguration (DNR) has become an effective method for reducing network losses in distribution systems, regulating bus voltage, and improving system resilience. With the increasing integration of distributed energy resources (DERs) into the distribution network (e.g., the increasing penetration of wind and solar photovoltaic power), the active power loss of the system will be reduced proportionally. Simultaneously, due to the increasing penetration of DERs, voltage regulation in unbalanced distribution systems is becoming increasingly complex and critical, and voltage fluctuations are strongly influenced by the topology of the distribution system.

[0019] Therefore, it is necessary to restructure the distribution network to improve its stability. One common approach is to construct an optimal network refactoring problem for the unbalanced distribution system, including renewable energy sources (DERs) and voltage regulators (VRs), considering the timing of distributed energy access, distribution, and transmission. This optimal network refactoring problem is defined as a mixed-integer semidefinite programming problem with slack variables. By solving the optimal network refactoring problem, a refactoring scheme can be obtained, which can then regulate the unbalanced distribution system to reduce voltage fluctuations.

[0020] However, the commonly used methods have the following technical problems: Since unbalanced power distribution systems operate for long periods of time, the optimal network reconstruction problem constructed for a specific time is difficult to adapt to the actual working conditions of the power distribution system, the solution results deviate from the actual situation of the power distribution system, and the control accuracy is low.

[0021] To address the aforementioned issues, the following detailed embodiments will be used to describe and explain a three-phase power distribution network control method, apparatus, equipment, and medium provided in this application.

[0022] To address the technical problem of discrepancies between existing control schemes and actual conditions, resulting in low control accuracy, this paper refers to... Figure 1 The diagram shows a flowchart of a control method for a three-phase power distribution network according to an embodiment of the present invention.

[0023] As an example, the control method for the three-phase distribution network may include: S11. Construct a physical information map of a three-phase power distribution network and extract spatiotemporal features from the physical information map.

[0024] In one embodiment, a physical information diagram of a three-phase distribution network or distribution system can be constructed first. The physical information diagram can determine the equipment connection structure of the three-phase distribution network, and then different equipment can be adjusted according to its equipment connection structure to fit the actual situation of the three-phase distribution network, so that the three-phase distribution network can operate in the optimal state and improve the accuracy of control.

[0025] To improve the efficiency of regulation, spatiotemporal features can be extracted from the physical information map. By extracting spatiotemporal features, spatial and temporal load, DER and voltage information can be effectively captured, thereby making accurate predictions of the positions of switches and voltage regulator taps. Then, by using these spatiotemporal features to solve the problem, the optimal regulation scheme can be determined.

[0026] In one embodiment, to construct a corresponding physical information map based on the connection structure of each device in a three-phase power distribution network, and then extract corresponding spatiotemporal features from it, the following sub-steps may be included as an example: S111. Obtain the node set, the edge set, and the line segment information of the three-phase distribution network, respectively, wherein the line segment information is the adjacency matrix of the line segment connectivity.

[0027] S112. Construct a physical information graph using the node set, the edge set of the line, and the line segment information.

[0028] S113. Perform node feature definition and vector adjustment processing on the physical information graph to obtain the adjusted information graph.

[0029] S114. Extract spatial features from the adjusted information map using a preset Chebyshev convolutional layer, and extract temporal features from the adjusted information map using a preset causal convolutional layer.

[0030] S115. Call a preset spatiotemporal convolutional block to combine the temporal features and the spatial features to obtain spatiotemporal features.

[0031] In one embodiment, the node set, the edge set, and the line segment information of the three-phase distribution network can be obtained respectively, wherein the line segment information is an adjacency matrix of line segment connectivity.

[0032] Specifically, the above information can be presented as follows: V represents the set of nodes in the power distribution system, |V| = N, where N is the total number of nodes in the power distribution system. This is used to explain the concept of a physical information graph and to map it to the structure of the power distribution system. E represents the set of edges of the lines in the power distribution system. A∈R N × N It is an adjacency matrix representing the connectivity of line segments. If line segment (i, j) is connected, then Aij = 1, otherwise Aij = 0.

[0033] Next, the above information can be used to construct a physical information diagram. Specifically, the physical information diagram can be as follows: G = (V, E, A); The above formula can be used to represent a three-phase unbalanced distribution network as a diagram.

[0034] To identify the spatiotemporal features of the physical information graph, we can first define the node features of the physical information graph, and then perform vector adjustment processing on the vectors of the physical information graph to make the physical information graph meet the requirements of feature extraction, thus obtaining the adjusted information graph.

[0035] Next, spatial features can be extracted from the adjusted information map using a pre-defined Chebyshev convolutional layer.

[0036] This invention employs Chebyshev convolutional layers to extract spatial features of graphs because Chebyshev convolutional layers achieve local information aggregation by allowing the model to focus on a finite number of neighboring nodes, making them suitable for graph-structured data. For each time interval t, the node embedding is calculated using the following convolution formula.

[0037] Specifically, the spatial characteristics of the infographic over time period t can be adjusted as shown in the following formula: ; In the above formula, x` t It represents spatial features. Θ is the convolution kernel. * G is the spatial convolution operator, which represents a signal x. t The product with a spatial kernel Θ. This can be achieved by setting different fields K. s , will K s Information from neighboring nodes within the range is propagated to the central node. Let λ be the scaled Laplace matrix. max for L The largest eigenvalue. It is a k-th order Chebyshev polynomial. , , .

[0038] The Laplace matrix of the normalized adjusted infographic can be defined as follows: L = I - D -1 / 2 AD -1 / 2 ; In the above formula, I It is the identity matrix. D Let be the angle matrix. If there are two branches connected to node i, then D ii = 2, D ij =0, i = j A is the adjacency matrix. Since the calculation results present different forms depending on the structure of different network structures, this embodiment can use... L Abstraction is used instead.

[0039] Simultaneously, causal convolutional layers can be used to extract temporal features from the adjustment infographic. Specifically, causal convolutional layers with a kernel of width Kt can be used to extract temporal features from the adjustment infographic, preserving the temporal causal relationships of the input sequence. The causal convolutional layers are followed by gated linear units (GLUs) to model nonlinearities.

[0040] Specifically, the time characteristics of the infographic can be adjusted over time period t as shown in the following formula: ; Where, x` i It is the temporal characteristic of the physical information graph at node i; x i This represents a time signal at node i in the signal graph at time t; W1 and W2 ∈ R. kt × ci × co Γ is an arbitrary convolution kernel, kt is the size of the temporal convolution kernel, i.e., how many time steps the convolution spans on the time axis; ci and co are the number of input channels and output channels of convolution kernels W1 and W2, respectively; * is the convolution operation; σ() is the Sigmoid function. This invention introduces *T as the temporal convolution operator, and Γ is the temporal kernel.

[0041] Finally, a spatiotemporal convolutional block can be constructed to jointly process and adjust the spatial and temporal features of the infographic. A "sandwich" structure can be used, placing the spatial layer between two temporal layers. The output of the computation block x' can be expressed as follows: ; Where x` represents the spatiotemporal characteristics. , These are the upper and lower temporal kernels, respectively; since x is a three-dimensional variable, the time kernel... , The spatial kernel is applied to each node of x, while the spatial kernel is applied to each time step of x, where x refers to the characteristic signal of the information graph at node i at time t.

[0042] In an optional embodiment, to capture the various characteristics of an unbalanced three-phase distribution network and to make the physical information map more uniform, the step of defining node features and adjusting vectors in the physical information map to obtain an adjusted information map, as an example, may include the following sub-steps: S1131. Define the node features of the physical information graph to obtain the defined information graph, wherein the node features include: the vector of the maximum active power generation of the three-phase distributed energy, the three-phase active load demand, and the three-phase reactive load demand.

[0043] S1132. Perform value replacement processing on the defined information graph to obtain an adjusted information graph. The value replacement processing involves setting the vector of maximum active power generation, three-phase active load demand and three-phase reactive load demand of non-distributed energy or non-load-bearing nodes as zero vectors, and using preset values ​​to represent non-three-phase buses.

[0044] To capture the various characteristics of an unbalanced three-phase distribution network, node characteristics can be defined in the physical information diagram. These defined node characteristics can be recorded... , x∈R N × 11 × T .

[0045] in It is the vector of the maximum active power generation of the three-phase DER on bus i at time t. and Let T represent the three-phase active and reactive load demands at node i at time t, respectively, where T is the time step length. This represents the upper limit of the voltage at node i. This is the lower limit of the voltage at node i.

[0046] To ensure a uniform physical information graph, for nodes without DERs and loads, and , Set as the zero vector, for buses that are not three-phase, use the value 0 to represent the missing phase.

[0047] S12. Determine the equipment information of the three-phase power distribution network based on the spatiotemporal characteristics, wherein the equipment information includes: the switching status of the equipment in the three-phase power distribution network and the location information of the voltage regulator.

[0048] In one embodiment, equipment information of a three-phase power distribution network can be determined based on spatiotemporal characteristics, wherein the equipment information includes: the switching status of equipment in the three-phase power distribution network and the location information of voltage regulators.

[0049] The device's on / off status can be information related to whether it is in an open or closed state. The voltage regulator's position information can be the probability distribution of the voltage regulator's tap positions.

[0050] In an optional embodiment, determining the equipment information of the three-phase distribution network based on the spatiotemporal characteristics may include the following sub-steps: S121. Call the preset link classifier to perform Hadamard product operation on the spatiotemporal features to obtain the edge embedding of each device in the three-phase power distribution network.

[0051] S122. The device scalar is obtained by summing the edge embeddings of the switch on the feature dimension, and the device switch state is obtained by using the sigmoid function based on the device scalar.

[0052] S123. The edge embedding of the voltage regulator is preprocessed using regular expressions to obtain processed embedding, and the probability distribution of the tap position of the voltage regulator is calculated using the processed embedding to obtain the position information of the voltage regulator.

[0053] In one embodiment, after obtaining the spatiotemporal features, a preset link classifier can be used to predict the connectivity of the switch and the tap position of the voltage based on the spatiotemporal features to obtain device information.

[0054] In one operating mode, spatiotemporal features can be input into a preset link classifier, which generates edge embeddings using node embeddings.

[0055] Specifically, the spatiotemporal features of endpoint nodes i and j at time t are obtained. , Then, the pre-defined linker classifier can perform Hadamard multiplication on the spatiotemporal features, making the result x... i ′⊙x j ′ is used as the input for the edge; then it passes through K. f Each fully connected layer obtains edge embedding. l ij .

[0056] For each time step, the edge embedding for each device can be obtained as follows: ; In the above formula: W k K is the weight matrix of the fully connected layer k; f This is the last fully connected layer; For the switch accuracy threshold, This is the voltage regulator difference threshold.

[0057] Take the D obtained in the previous step fThe edges of the dimensional device are embedded in the feature dimension and summed to obtain a scalar. Then, using the sigmoid function, the probability that the switching device is closed at time t can be obtained, thus determining the device's switching state. In one embodiment, the device's switching state can be calculated as follows: ; Among them, D f It is the last fully connected layer K f The output dimension of D can be used to determine the output dimension of D. f Set to the maximum number of taps for all voltage regulators.

[0058] In the above operating method, l ij,d,t It's about the embedded features of the switches. These features are summed together and then processed by the Sigmoid function. This gives us a number between 0 and 1, representing the probability that the switch will close at time t. The closer the number is to 1, the higher the probability that the switch will close; conversely, the lower the number is, the lower the probability that the switch will close.

[0059] For voltage regulators, regular expressions can be used to encode the natural sequence between discrete actions of tap position settings.

[0060] Specifically, regular expressions can be used to preprocess the edge embedding of the voltage regulator to obtain the processed embedding, where m ≤ M for each tap position. ij The preprocessed embedding can be obtained as shown in the following formula: ; in M is the sequential encoding of the tap position m of the voltage regulator (i, j) on line ij at time t. ij It is the total number of tap positions of the voltage regulator between node i and node j.

[0061] Next, the position information of the voltage regulator can be obtained by processing the probability distribution of the tap positions embedded in the calculation. The position information of the voltage regulator can be calculated as follows: ; The above formula can be expressed as a softmax function, yielding the probability corresponding to each tap position m, representing the likelihood that the regulator is at that tap position at time t. The probabilities of all positions are summed to 1.

[0062] It should be noted that for each node i, the node embedding can also be calculated using Causal convolution. By changing the value of Kt, the input sequence x can be explored. i Kt nearest neighbors of an inner element.

[0063] S13. Construct the overall loss function of the model using the device information, and train the spatiotemporal graph convolutional network based on the overall loss function.

[0064] In one embodiment, the overall loss function of the spatiotemporal graph convolutional network can be constructed using device information, and the spatiotemporal graph convolutional network can be trained based on the overall loss function.

[0065] Subsequently, a spatiotemporal graph convolutional network can be invoked to predict the switch state and the tap position of the voltage regulator based on the edge embeddings obtained in the above steps. This allows subsequent operations to construct solutions based on the prediction results and select the optimal solution.

[0066] In one embodiment, the step of constructing the overall loss function of the model using the device information and training the spatiotemporal graph convolutional network based on the overall loss function may include the following sub-steps: S131. Construct a first loss function and a second loss function using the device's on / off state, and construct a third loss function using the location information.

[0067] S132. Combine the first loss function, the second loss function, and the third loss function to obtain the overall loss function.

[0068] S133. Construct a prediction network for predicting the tap position and switching state of the voltage regulator based on the overall loss function, and train the prediction network to obtain a spatiotemporal graph convolutional network.

[0069] In one embodiment, the overall loss function of the spatiotemporal graph convolutional network can consist of three parts. Specifically, a first loss function and a second loss function can be constructed first using the device's on / off state, and then a third loss function can be constructed using the location information.

[0070] In one operating mode, the first loss function can be the logistic loss function of the switch, which aims to improve the accuracy of predicting the switch state. Specifically, the first loss function can be as follows: ; The second loss function incorporates the radial constraint of the physical network, ensuring that the distribution network topology remains radial. Specifically, the second loss function can be expressed as follows: ; Where, N s This represents the total number of substation nodes.

[0071] The third loss function corresponds to the cross-entropy loss of the voltage regulator and is used to improve the prediction accuracy of the tap position. Specifically, the third loss function can be expressed as follows: ; Where z ij,m,t It is a 0-1 variable used in SOS1. If the tap position of the voltage regulator is at position m at time t, it is set to 1; otherwise, it is 0.

[0072] Next, the first loss function, the second loss function, and the third loss function can be combined to obtain the overall loss function. Specifically, the overall loss function can be expressed as follows: ; Where Ξ is the set of all convolutional kernels. W is the set of weight matrices associated with the fully connected layer.

[0073] In one embodiment, after training the spatiotemporal graph convolutional network, the network can be evaluated to determine its prediction accuracy. Specifically, two evaluation metrics related to the prediction performance of the switch and voltage regulator tap variables can be established. These two metrics include prediction accuracy (Acc) and prediction bias (Dcc). For all time steps t within the set T, the prediction accuracy (Acc) can be defined as follows: ; The prediction bias (Dcc) can be defined as follows: ; in, , These are the switching variables and voltage regulator tap variables to be predicted, respectively. and The nth graph G n The switching variables and voltage regulator tap variables to be predicted; N g It represents the number of graphs in the dataset; It is the predicted state of the switch; It is the predicted tap position of the voltage regulator.

[0074] In one embodiment, the training operation of the spatiotemporal graph convolutional network can first construct a training set D from the dataset. train With the validation set D val Initialize the graph adjacency matrix A, and normalize the graph Laplacian matrix L.

[0075] The spatiotemporal features of the training set sequence are extracted using the above formula. Next, the weights Wk of the fully connected layers are initialized, and Model is defined as: Convolutional network + Kf fully connected layers + sigmoid function. Then, the number of training epochs E is determined. The training set D is used. train Perform the training operation. Forward propagation calculations can be performed during training. Simultaneously, the loss Loss can be calculated based on the formulas in steps S121-S123 and the overall loss function, and then the weights W can be updated using gradient descent backpropagation. k Repeat the training for round E until the operation is completed.

[0076] In one embodiment, constraints can be added to the training instance data to selectively modify variables based on comparisons of the values ​​and thresholds of two evaluation metrics.

[0077] For example, calculation The prediction accuracy Acc, if training instance data F Add constraints, for example .

[0078] For example, calculation The prediction bias Dcc, if training instance data F Add constraints, for example .

[0079] In the above formula, G represents the physical information diagram of the power distribution system. The threshold for switch accuracy. The voltage regulator difference threshold. Let be the variable of the voltage regulator tap on line ij at time t.

[0080] S14. Simplify the prediction processing of the spatiotemporal graph convolutional network into an initial problem, and call a preset solver to solve the initial problem to obtain a control scheme. Then, control the three-phase power distribution network according to the control scheme.

[0081] In one embodiment, after training the spatiotemporal graph convolutional network, the spatiotemporal graph convolutional network can be used to predict the relevant prediction parameters of the switch and voltage regulator taps, and then the prediction results can be used as a control scheme for control processing.

[0082] To improve processing efficiency, the prediction process of the spatiotemporal graph convolutional network can be simplified into an initial problem. A pre-defined solver can then be used to solve this initial problem, yielding the optimal control scheme. Finally, the control scheme can be applied to regulate the equipment in the three-phase power distribution network, ensuring that the network operates at its best.

[0083] In an optional embodiment, simplifying the prediction process of the spatiotemporal graph convolutional network into an initial problem, and calling a preset solver to solve the initial problem to obtain a control scheme, and then controlling the three-phase distribution network according to the control scheme, may include the following sub-steps: S141. Based on the binary decision variable fixing algorithm of the spatiotemporal graph convolutional network, the prediction process of the spatiotemporal graph convolutional network is simplified into an initial problem.

[0084] S142. Call the preset solver to solve the initial problem and obtain the control scheme. The preset solver is a MID solver constructed by a preset objective function.

[0085] S143. The equipment of the three-phase power distribution network is regulated and processed according to the regulation scheme.

[0086] By fixing some decision variables predicted by the proposed physical information graph convolutional model, the original optimization problem is simplified into a sub-MIP problem that is easier to solve.

[0087] In one embodiment, the prediction process of a spatiotemporal graph convolutional network (SPCNN) can be simplified into an initial problem using a binary decision variable fixing algorithm. A preset solver (MID) is then invoked to solve the initial problem, yielding a control scheme. The preset solver is a MID solver constructed using a preset objective function. Finally, the control scheme is applied to the equipment in the three-phase power distribution network.

[0088] By simplifying and solving the problem, computational complexity can be reduced. In one approach, the linker classifier can be integrated into the spatiotemporal graph convolutional network of physical information, forming an ensemble network (STGCN). This method achieves efficient solution generation by fixing specific variables in instances and using a MIP solver to solve the simplified problem, thereby quickly generating a control scheme for convenient processing based on the control scheme.

[0089] In one embodiment, the control scheme may include switch states, tap positions of voltage regulators, and DER output. Finally, the equipment in the three-phase distribution network can be controlled according to the content of the control scheme.

[0090] In an optional embodiment, the method for determining the constraints of the preset objective function may include the following sub-steps: S21. Obtain the output deviation, voltage deviation, number of switching actions, and voltage regulator adjustment amount of the three-phase distribution network, wherein the output deviation is to minimize the output deviation of the three-phase distribution network under distributed energy.

[0091] S22. Construct an objective function using the output deviation, the voltage deviation, the number of switching actions, and the voltage regulator adjustment amount, and determine the constraints of the objective function.

[0092] The constraints include: power flow constraints, voltage constraints, voltage regulator planning constraints, network radiation constraints, distributed energy output constraints, branch capacity constraints, and switching operation constraints.

[0093] In one embodiment, the output deviation, voltage deviation, number of switching operations, and voltage regulator adjustment of the three-phase distribution network can be obtained separately. The output deviation is to minimize the output deviation of the three-phase distribution network under distributed energy resources. An objective function can be constructed using the above parameters. This objective function is then solved, and network reconfiguration and voltage regulation can be coordinated to achieve economic and environmental benefits.

[0094] In one embodiment, the objective function can be expressed as follows (1): (1); in, (2); In the above formula: , The active and reactive power output of DER at node i with phase φ at time t is represented. To maximize the active power output of DER; , This represents the square magnitude of the voltage at phase φ at time node i, and its deviation from the square magnitude of the reference voltage. It is a 0-1 variable, which takes the value of 1 when the switch between nodes ij is closed at time t, and 0 otherwise. It is a 0-1 variable, which takes the value of 1 when node j is the parent node of node i at time t, and 0 otherwise; This is an integer variable representing the tap position of the voltage regulator between nodes ij at time t; It is a 0-1 variable, representing the tap position state of the voltage regulator between nodes ij at time t; The cost coefficient for distributed energy generation limitation; Cost factor for voltage deviation; This is the cost factor for switching operations; Cost factor for adjusting taps in voltage regulators; This is a 0-1 variable, representing the operational state of the switch between nodes ij at time t. A value of 1 indicates that the switch is closed, otherwise it is not closed. This represents the change in the position of the voltage regulator taps between nodes ij compared to the previous time t; , , These are collections of all lines, switches, and voltage regulators; A collection of distributed energy resources; The set of all nodes; Let i be the set of phases of node i.

[0095] After constructing the objective function, the constraints of the user-defined objective function can be obtained. These constraints include: power flow constraints, voltage constraints, voltage regulator planning constraints, network radiation constraints, distributed energy output constraints, branch capacity constraints, and switching operation constraints.

[0096] For power flow constraints, it is assumed that the voltages at each node are approximately balanced. Neglecting higher-order terms, the active and reactive power balance equations for node i are obtained (3) - (6), which can be shown in the following equations: (3); (4); (5); (6); In the above formula, A collection of substations; For node i at time t, in phase φ The active power; For node i at time t, in phase φ reactive power; and These represent the active and reactive power of the substation at time t, respectively. , Let g be the active and reactive power of DER at time t; , Let i be the phase from node i to j at time t. φ The trend of meritorious and ineffective actions.

[0097] Similarly, for each line (i,j)∈ By ignoring higher-order terms and adding / subtracting a large positive number, the voltage amplitude constraint is obtained, as shown in equations (7) and (8) below. For, ∈Φ ij For t∈T, we have: (7); (8); Where M is a sufficiently large positive number; It is a composite index; if the phase of line ij is... φ The value is 1 if it is set to 1, otherwise it is 0. Let be the set of phases of line ij; , The resistance and reactance matrices are after equivalent transformation; , Let represent the active and reactive power flow from node i to j at time t.

[0098] For voltage constraints, in addition to the voltage amplitude constraints in equations (12) and (13), voltage imbalance is also a major concern in the power distribution system. Phase imbalance constraints are established using equations (14) and (15). Finally, the deviation between the square of the voltage amplitude and the reference value is calculated using equations (16) and (17).

[0099] (12); (13); (14); (15); (16); (17); In the above formula, Let be the average square of the three-phase voltage at node i at time t; This is the maximum voltage imbalance limit for three-phase voltage. , These are the upper and lower limits of the square of the voltage at node i, respectively; For a set of three-phase nodes, For time t, φ On phase i, the magnitude of the square voltage amplitude deviates from the reference value.

[0100] Regarding the voltage regulator planning constraints, in order to obtain the linear expression of the voltage regulator, it is assumed that there is a three-phase, double-grounded voltage regulator between bus i and bus j, and it is equivalently modeled as an ideal transformer with series impedance, where the bus is a virtual node connected between bus i and bus j on the secondary side of the ideal transformer. For each phase, the ideal transformer can be expressed as equation (18), where η ij This refers to the tap ratio.

[0101] (18); in, For virtual node i s phase voltage amplitude, Phase of node i The voltage amplitude; , These are virtual nodes i s Phase with actual node i Injected active power; , These are virtual nodes i s Phase with actual node i Injected reactive power.

[0102] Assume τ ij It is an integer variable representing the position of the tap; the tap is the same as the header. , 0≤τ ij ≤M ij M ij This represents the total number of tap positions of the voltage regulators between nodes ij. It is the lower bound of the tap changer ratio. This is the change in the transformer ratio during each tap adjustment. Therefore: (19); Since equation (19) contains a nonlinear term, which is the product of a continuous variable and a square integer variable, it cannot be handled by most MIP solvers. Therefore, a special ordered set (SOS1) is applied to solve this problem. Formally, for a set of related nonnegative weights {w1, w2, ..., w...}, ... Mij} 0-1 variables { z1, z2, ..., z Mij The bilinear form of the voltage regulator is given as follows: (20); (twenty one); (twenty two); In the above formula, This refers to the voltage of phase φ on the virtual bus. Let i be the voltage of phase φ at node i. s This is a virtual bus used to define the voltage regulator (i, j).

[0103] Regarding the network's radial constraint, to enhance the system's radial and connectivity properties, it is ensured that all nodes except substation nodes have exactly one parent node. The specific constraint conditions are shown in the following equation: (twenty three); (twenty four); (25); (26); In the above formula, N(i) is the set of neighboring buses of bus i.

[0104] The output constraints of distributed energy resources can be expressed as follows: (27); (28); In the above formula, For node i, DER in phase capacity, Let gi be the reactive power of phase φ of generator at time t. Let be the upper limit of the active power of generator gi in phase φ at time t. Let gi be the active power of phase φ of generator at time t.

[0105] Regarding line capacity constraints, the following formula can be used: (29); In the above formula, This represents the upper limit of line ij's capacity.

[0106] Regarding the operating constraints of switches and voltage regulators, in order to reduce excessive operation of switches and voltage regulators and extend their lifespan, a maximum operating time constraint is introduced for each switch and voltage regulator, as shown in the following formula: (30); (31); (32); (33); (34); (35); In the above formula, This is a 0-1 variable, representing the operational state of the switch between nodes ij at time t. A value of 1 indicates that the switch is closed, otherwise it is not closed. This represents the maximum number of switch operations within the time period T. This represents the change in the position of the voltage regulator taps between nodes ij compared to the previous time t; This represents the maximum number of tap operations within the time period T.

[0107] Reference Figure 2 The diagram shows an operation flowchart of a control method for a three-phase power distribution network provided by an embodiment of the present invention.

[0108] Specifically, the operation of the control method for a three-phase distribution network may include the following steps: Step 1, Data Processing: Distribution System Physical Information Map G, Node Timing Characteristics x i,t And perform standardized Laplace calculations.

[0109] The second step involves constructing a spatiotemporal graph of physical priors within time period T using the formulas in steps S111-115, and inputting it into a spatiotemporal graph convolutional network to achieve spatiotemporal feature extraction.

[0110] The third step involves using a link classifier to embed the switch and voltage regulator devices into an edge, and using the edge embedding to represent the switch state and voltage regulator tap position, as detailed in the formulas of steps S121-123.

[0111] The fourth step is to define the loss function (refer to the formulas in steps S131-133 for details), and train the established physical information neural network model on the training set to obtain two types of probability outputs: switch closure probability. The probability of tap position of voltage regulator .

[0112] The fifth step is to use the validation set to evaluate the prediction performance and calculate the prediction accuracy and prediction bias of the state results of each component at time t.

[0113] The sixth step involves using a binary decision variable fixed simplification algorithm based on a physical information graph convolution model to progressively simplify the original prediction problem, and then using a solver to calculate the simplified initial problem.

[0114] Step 7, Output switch status Voltage regulator tap position And the DER output, and adjust it according to the output control scheme.

[0115] In this embodiment, the present invention provides a method for regulating a three-phase distribution network. Its advantages are as follows: the present invention can construct a physical information map of the three-phase distribution network and extract spatiotemporal features from the physical information map; determine the equipment information of the three-phase distribution network based on the spatiotemporal features; construct the overall loss function of the model using the equipment information, and train a spatiotemporal graph convolutional network based on the overall loss function; simplify the prediction processing of the spatiotemporal graph convolutional network into an initial problem, and call a preset solver to solve the initial problem to obtain a regulation scheme; and regulate the three-phase distribution network according to the regulation scheme. By constructing a physical information map and extracting spatiotemporal features from it, and making predictions based on these features, the present invention can closely match the actual state of the distribution network to generate a regulation scheme that meets the needs of the distribution network, thereby improving the accuracy of regulation and reducing the error of regulation; furthermore, during prediction, converting the model's prediction into a problem for solution simplifies the processing flow and computational complexity, thereby improving processing efficiency.

[0116] This invention also provides a control device for a three-phase power distribution network, see [link to relevant documentation]. Figure 3 The diagram shows a schematic diagram of the structure of a control device for a three-phase power distribution network according to an embodiment of the present invention.

[0117] As an example, the control device for the three-phase distribution network may include: Extraction module 201 is used to construct a physical information map of a three-phase power distribution network and extract spatiotemporal features from the physical information map; The determining module 202 is used to determine the equipment information of the three-phase power distribution network based on the spatiotemporal characteristics, wherein the equipment information includes: the switching status of the equipment in the three-phase power distribution network and the location information of the voltage regulator; Training module 203 is used to construct the overall loss function of the model using the device information, and to train the spatiotemporal graph convolutional network based on the overall loss function; The control module 204 is used to simplify the prediction processing of the spatiotemporal graph convolutional network into an initial problem, and call a preset solver to solve the initial problem to obtain a control scheme, and control the three-phase power distribution network according to the control scheme.

[0118] Optionally, the construction of the physical information map of the three-phase distribution network and the extraction of spatiotemporal features from the physical information map include: Obtain the node set, the edge set, and the line segment information of the three-phase distribution network, respectively, wherein the line segment information is the adjacency matrix of the line segment connectivity; A physical information graph is constructed using the node set, the edge set of the line, and the line segment information; The physical information graph is subjected to node feature definition and vector adjustment processing to obtain an adjusted information graph; Spatial features are extracted from the adjustment information map using a preset Chebyshev convolutional layer, and temporal features are extracted from the adjustment information map using a preset causal convolutional layer; The spatiotemporal features are obtained by combining the temporal and spatial features using a preset spatiotemporal convolutional block.

[0119] Optionally, the step of defining node features and adjusting vectors in the physical information graph to obtain an adjusted information graph includes: The physical information graph is defined by defining node features to obtain a defined information graph, wherein the node features include: the vector of the maximum active power generation of the three-phase distributed energy, the three-phase active load demand, and the three-phase reactive load demand; The defined information graph is subjected to value replacement processing to obtain an adjusted information graph. The value replacement processing involves setting the vector of maximum active power generation, three-phase active load demand and three-phase reactive load demand of non-distributed energy or non-load-bearing nodes as zero vectors, and using preset values ​​to represent non-three-phase buses.

[0120] Optionally, determining the equipment information of the three-phase distribution network based on the spatiotemporal characteristics includes: A preset link classifier is invoked to perform a Hadamard product operation on the spatiotemporal features to obtain the edge embedding of each device in the three-phase power distribution network. The device scalar is obtained by summing the edge embeddings of the switch on the feature dimension, and the device switch state is obtained by using the sigmoid function based on the device scalar. The edge embedding of the voltage regulator is preprocessed using regular expressions to obtain processed embeddings, and the probability distribution of the tap positions of the voltage regulator is calculated using the processed embeddings to obtain the position information of the voltage regulator.

[0121] Optionally, the step of constructing the overall loss function of the model using the device information and training the spatiotemporal graph convolutional network based on the overall loss function includes: A first loss function and a second loss function are constructed using the device's on / off state, and a third loss function is constructed using the location information; The overall loss function is obtained by combining the first loss function, the second loss function, and the third loss function. A prediction network for predicting the tap position and switching state of a voltage regulator is constructed based on the overall loss function, and the prediction network is trained to obtain a spatiotemporal graph convolutional network.

[0122] Optionally, the step of simplifying the prediction process of the spatiotemporal graph convolutional network into an initial problem, and calling a preset solver to solve the initial problem to obtain a control scheme, and then controlling the three-phase distribution network according to the control scheme, includes: Based on the binary decision variable fixing algorithm of the spatiotemporal graph convolutional network, the prediction process of the spatiotemporal graph convolutional network is simplified into an initial problem; The initial problem is solved by calling a preset solver to obtain a control scheme. The preset solver is a MID solver constructed by a preset objective function. The equipment in the three-phase power distribution network is regulated and controlled according to the aforementioned control scheme.

[0123] Optionally, the method for determining the constraints of the preset objective function includes: The output deviation, voltage deviation, number of switching actions, and voltage regulator adjustment amount of the three-phase distribution network are obtained, wherein the output deviation is to minimize the output deviation of the three-phase distribution network under distributed energy. An objective function is constructed using the output deviation, the voltage deviation, the number of switching actions, and the voltage regulator adjustment amount, and the constraints of the objective function are determined. The constraints include: power flow constraints, voltage constraints, voltage regulator planning constraints, network radiation constraints, distributed energy output constraints, branch capacity constraints, and switching operation constraints.

[0124] Those skilled in the art will understand that, for ease of description and brevity, the specific working process of the device described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0125] Furthermore, this application also provides an electronic device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the three-phase power distribution network control method as described in the above embodiments.

[0126] Furthermore, embodiments of this application also provide a computer-readable storage medium storing a computer-executable program, the computer-executable program being used to cause a computer to perform the three-phase power distribution network control method as described in the above embodiments.

[0127] In the description of the embodiments of the present invention, it should be noted that the terms "above," "below," etc., indicating the orientation or positional relationship are based on the orientation or positional relationship shown in the accompanying drawings, and are only for the convenience of describing the embodiments of 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, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the present invention. When an element such as a layer, region, or substrate is referred to as being "above" or "on top of" another element, it may be directly on the other element, or there may be an intermediate element. Conversely, when an element is referred to as being "directly on" or "above" another element, there is no intermediate element. It should also be understood that when an element is referred to as being "below" or "under" another element, it may be directly below or under the other element, or there may be an intermediate element. Conversely, when an element is referred to as being "directly below" or "under" another element, there is no intermediate element. Unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to fixed connections, detachable connections, or integral connections; they can refer to mechanical connections or electrical connections; they can refer to direct connections or indirect connections through an intermediate medium; and they can refer to the internal connection between two components. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0128] Those skilled in the art will understand that embodiments of this application may also include computer program products. Therefore, this application may take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application may take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0129] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), devices, and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0130] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0131] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0132] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the technical principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A control method for a three-phase power distribution network, characterized in that, The method includes: Construct a physical information map of a three-phase distribution network and extract spatiotemporal features from the physical information map; The equipment information of the three-phase power distribution network is determined based on the spatiotemporal characteristics, wherein the equipment information includes: the switching status of the equipment in the three-phase power distribution network and the location information of the voltage regulator; The overall loss function of the model is constructed using the device information, and a spatiotemporal graph convolutional network is trained based on the overall loss function; The prediction process of the spatiotemporal graph convolutional network is simplified into an initial problem, and a preset solver is called to solve the initial problem to obtain a control scheme. The three-phase power distribution network is then controlled according to the control scheme.

2. The control method for a three-phase distribution network according to claim 1, characterized in that, The construction of the physical information map of the three-phase distribution network and the extraction of spatiotemporal features from the physical information map include: Obtain the node set, the edge set, and the line segment information of the three-phase distribution network, respectively, wherein the line segment information is the adjacency matrix of the line segment connectivity; A physical information graph is constructed using the node set, the edge set of the line, and the line segment information; The physical information graph is subjected to node feature definition and vector adjustment processing to obtain an adjusted information graph; Spatial features are extracted from the adjustment information map using a preset Chebyshev convolutional layer, and temporal features are extracted from the adjustment information map using a preset causal convolutional layer; The spatiotemporal features are obtained by combining the temporal and spatial features using a preset spatiotemporal convolutional block.

3. The control method for a three-phase distribution network according to claim 2, characterized in that, The process of defining node features and adjusting vectors in the physical information graph to obtain an adjusted information graph includes: The physical information graph is defined by defining node features to obtain a defined information graph, wherein the node features include: the vector of the maximum active power generation of the three-phase distributed energy, the three-phase active load demand, and the three-phase reactive load demand; The defined information graph is subjected to value replacement processing to obtain an adjusted information graph. The value replacement processing involves setting the vector of maximum active power generation, three-phase active load demand and three-phase reactive load demand of non-distributed energy or non-load-bearing nodes as zero vectors, and using preset values ​​to represent non-three-phase buses.

4. The control method for a three-phase distribution network according to claim 1, characterized in that, The process of determining the equipment information of the three-phase distribution network based on the spatiotemporal characteristics includes: A preset link classifier is invoked to perform a Hadamard product operation on the spatiotemporal features to obtain the edge embedding of each device in the three-phase power distribution network. The device scalar is obtained by summing the edge embeddings of the switch on the feature dimension, and the device switch state is obtained by using the sigmoid function based on the device scalar. The edge embedding of the voltage regulator is preprocessed using regular expressions to obtain processed embeddings, and the probability distribution of the tap positions of the voltage regulator is calculated using the processed embeddings to obtain the position information of the voltage regulator.

5. The control method for a three-phase distribution network according to claim 1, characterized in that, The step of constructing the overall loss function of the model using the device information and training the spatiotemporal graph convolutional network based on the overall loss function includes: A first loss function and a second loss function are constructed using the device's on / off state, and a third loss function is constructed using the location information; The overall loss function is obtained by combining the first loss function, the second loss function, and the third loss function. A prediction network for predicting the tap position and switching state of a voltage regulator is constructed based on the overall loss function, and the prediction network is trained to obtain a spatiotemporal graph convolutional network.

6. The control method for a three-phase distribution network according to any one of claims 1-5, characterized in that, The process of simplifying the prediction processing of the spatiotemporal graph convolutional network into an initial problem, and calling a preset solver to solve the initial problem to obtain a control scheme, and then performing control processing on the three-phase distribution network according to the control scheme, includes: Based on the binary decision variable fixing algorithm of the spatiotemporal graph convolutional network, the prediction process of the spatiotemporal graph convolutional network is simplified into an initial problem; The initial problem is solved by calling a preset solver to obtain a control scheme. The preset solver is a MID solver constructed by a preset objective function. The equipment in the three-phase power distribution network is regulated and controlled according to the aforementioned control scheme.

7. The control method for a three-phase distribution network according to claim 6, characterized in that, The method for determining the constraints of the preset objective function includes: The output deviation, voltage deviation, number of switching actions, and voltage regulator adjustment amount of the three-phase distribution network are obtained, wherein the output deviation is to minimize the output deviation of the three-phase distribution network under distributed energy. An objective function is constructed using the output deviation, the voltage deviation, the number of switching actions, and the voltage regulator adjustment amount, and the constraints of the objective function are determined. The constraints include: power flow constraints, voltage constraints, voltage regulator planning constraints, network radiation constraints, distributed energy output constraints, branch capacity constraints, and switching operation constraints.

8. A control device for a three-phase power distribution network, characterized in that, The device includes: An extraction module is used to construct a physical information map of a three-phase power distribution network and extract spatiotemporal features from the physical information map; The determination module is used to determine the equipment information of the three-phase power distribution network based on the spatiotemporal characteristics, wherein the equipment information includes: the switching status of the equipment in the three-phase power distribution network and the location information of the voltage regulator; The training module is used to construct the overall loss function of the model using the device information, and to train the spatiotemporal graph convolutional network based on the overall loss function; The control module is used to simplify the prediction processing of the spatiotemporal graph convolutional network into an initial problem, and call a preset solver to solve the initial problem to obtain a control scheme, and then control the three-phase power distribution network according to the control scheme.

9. An electronic device, comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, when the processor executes the computer program, it implements the control method for a three-phase power distribution network as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer-executable program for causing a computer to perform the control method for a three-phase power distribution network as described in any one of claims 1-7.

Images