An online prediction method and device for node voltage of a time-varying topological DC distribution network

Abstract

The present invention discloses an online prediction method and device for node voltages in a time-varying topology DC distribution network. Based on the historical operation data of the DC distribution network, a simulation model is built in combination with the actual grid topology, and a time-series training dataset including voltages, powers, and line states is generated through dynamic topology simulation. Based on the training dataset, a dynamic spatio-temporal graph neural network that integrates topological dynamic changes is constructed, and the time-series module is combined to learn the voltage time-series evolution law. The dynamic spatio-temporal graph neural network is used to collect node power data and topological information in real time, dynamically update the network topology input, and output future short-time domain voltage prediction values. For the scenario of grid topology changes, the graph connection parameters associated with the changed nodes are updated to achieve rapid fine-tuning of the model and stable prediction. The present invention can dynamically input network topology information under the condition of topology change in the DC distribution network, and achieve rapid and stable prediction of node voltages in the DC distribution network under time-varying topology.

Description

Technical Field

[0001] The present invention relates to the field of DC distribution network node voltage prediction, and particularly to an online prediction method and device for the node voltage of a time-varying topology DC distribution network. Background Art

[0002] In a DC distribution network, voltage is the most important indicator to measure the system stability. However, due to the small inertia and strong voltage sensitivity of the DC distribution network system, and at the same time, the topology of the DC distribution network changes frequently caused by distributed power sources and load switching, resulting in strong time-varying and non-linear fluctuation characteristics of the node voltage. It is difficult to ensure the stability and security of the system operation. Therefore, it is urgent to improve the short-term prediction ability of the DC distribution network voltage situation to ensure the safe and reliable operation of the DC distribution network. Traditional prediction methods based on fixed network models are mostly pure data-driven methods, which predict by learning the development trend of historical voltage data. It is difficult to utilize the spatial correlation characteristics between different node voltages, and most of them are based on fixed network models and cannot effectively handle the scenarios of topology changes caused by load switching and other reasons in the power grid. Therefore, how to construct an online voltage prediction model that can comprehensively utilize spatio-temporal coupling information and effectively handle time-varying topology situations has become an important technical bottleneck in the construction of high-reliability DC distribution systems.

[0003] The Dynamic Graph Neural Network (DGNN) combines the Graph Convolutional Layer (GCN) and the Transformer time series processing module, and can effectively model the dynamic characteristics of a time-varying topology distribution network, and has significant advantages in power system time series tasks such as voltage prediction. The graph convolutional layer aggregates local topology information through the adjacency matrix to capture the spatial correlation between nodes. When dealing with time-varying topology scenarios, it can adaptively adjust the adjacency weights according to the dynamic topology changes and optimize the feature propagation path in real time. By stacking multiple layers of graph convolutional structures, the model can not only capture long-range dependence relationships such as the voltage coupling effect of long-distance buses in the power network, but also deeply integrate the topological dynamic characteristics to realize multi-scale modeling and spatio-temporal collaborative characterization of the node association strength in a complex power grid environment. The Transformer time series processing module analyzes the global time series dependence relationship of the voltage sequence through the self-attention mechanism, adaptively assigns the attention weights of different time steps according to the dynamic time series pattern, and accurately captures the cross-cycle impact of key events on voltage evolution. The dynamic graph neural network combines the graph convolutional layer and the Transformer to achieve spatio-temporal two-dimensional modeling, efficiently captures the spatial correlation characteristics of the graph structure, and accurately captures the dynamic evolution pattern of each node voltage in the long-term time series, and realizes the efficient modeling of complex spatio-temporal associations and online generalization prediction of node voltages for a time-varying topology distribution network. Summary of the Invention

[0004] In view of the above problems, the present invention proposes an online prediction method and device for node voltages in a time-varying topology DC distribution network. The method first builds a simulation model based on the historical operation data of the DC distribution network and the actual grid topology, and generates a training data set including voltages, powers, and line states through simulation for subsequent neural network training. A dynamic graph neural network combining a dynamic graph convolutional layer and a Transformer time series processing module is constructed. The dynamic graph convolutional layer is trained to capture the real-time correlations between nodes to obtain high-dimensional node features, and the Transformer time series processing module is further used to process the high-dimensional node features to learn the time series evolution rules of the voltages of each node. At the same time, physical constraints in the DC distribution network are embedded in the loss function during the training process to ensure that the final prediction output conforms to the actual grid rules. When the model is applied online, by inputting the real-time measurement data of nodes and the current topology information, the predicted values of the voltages of each node in the future for a period of time can be quickly generated. When the topology changes, the model can quickly fine-tune by training a few new parameters, accept the real-time input under the current topology, and quickly generate accurate predicted values of node voltages.

[0005] To achieve the above object, the present invention provides an online prediction method for node voltages in a time-varying topology DC distribution network, including the following steps:

[0006] Step 1: Build a simulation model based on the historical operation data and the actual topology of the DC distribution network, and generate a time series training data set including voltages, powers, and line states through dynamic topology simulation;

[0007] Step 2: Compose the required training samples according to the training data obtained in Step 1, train a dynamic spatio-temporal graph neural network that integrates topological dynamic changes, capture the real-time correlations between nodes through the dynamic graph convolutional layer to obtain high-dimensional node features, process them to further learn the time series evolution rules of voltages and perform node voltage deduction and prediction, and embed the physical constraints in the DC distribution network in the loss function to train a complete dynamic spatio-temporal graph neural network;

[0008] Step 3: According to the dynamic spatio-temporal graph neural network obtained in Step 2, input the real-time collected node power data and network topology information, and the dynamic spatio-temporal graph neural network outputs the predicted values of the future voltages of each node;

[0009] Step 4: According to the dynamic spatio-temporal graph neural network obtained in Step 2, when the grid topology changes, fine-tune the dynamic graph neural network by updating the graph connection parameters associated with the changed nodes to ensure that the model can be quickly adjusted to continue prediction.

[0010] Furthermore, a simulation model is built based on the real distribution network topology. By using the historical operation data of each node and the network topology information of the DC distribution network, the power flow calculation in the AC distribution network is simplified to a linear equation in the DC distribution network to calculate the voltage amplitude of each node in the DC distribution network. Further combined with the line on / off state, training samples are formed. The specific method is as follows:

[0011]

[0012] Among them, represents the set of nodes adjacent to node ; node is the node adjacent to node in the DC distribution network; is the voltage of node at time ; is the voltage of node at time ; is the resistance between node and node ; is the power injection of node at time ; is the load power of node at time ; is the on / off state of node and node at time ;

[0013] According to the calculated node voltages of each node at different times, combined with the power data 、 of each node and the corresponding line on / off state to form the subsequent training dataset of the dynamic spatio-temporal graph neural network.

[0014] Furthermore, a dynamic spatio-temporal graph neural network is constructed. For the time-series training dataset obtained in Step 1, graph convolution operations are performed on the data of each node in the order of time steps;

[0015] After the training data of the training set undergoes graph convolution operations, high-dimensional node feature sequences of each node are obtained. The high-dimensional feature sequence of each node contains the node information 、 、 of each node and the network topology information .

[0016] Furthermore, in the time series processing module of the dynamic spatio-temporal graph neural network, a Transformer is used to further process the high-dimensional node feature sequence, learn the law of voltage time series evolution, and perform node voltage deduction and prediction for the nodes in the distribution network.

[0017] Furthermore, in the loss calculation of the model training of the dynamic spatio-temporal graph neural network, the loss function adds physical constraints on the basis of the error term to make the model output more in line with the actual physical laws. The specific operations are as follows:

[0018]

[0019] Among them, is the final loss function after combining the prediction error term and physical constraints; is the prediction error term; is the Kirchhoff's current law constraint term; is the model predicted voltage value; is the true voltage value; is the node at the power injection power at the moment; is the node at the load power at the moment; is the weight of the physical constraint term in the loss function; is the dynamic conductance matrix.

[0020] Furthermore, for the online application of the trained dynamic spatio-temporal graph neural network, the real-time topology information of the DC distribution network and the power data of each node are used as the input of the dynamic spatio-temporal graph neural network, and the dynamic spatio-temporal graph neural network outputs the predicted values of the voltage amplitudes of each node in the DC distribution network for a period of time in the future.

[0021] Furthermore, for the case of grid topology change, the structural parameters of the original neural network time series processing module are frozen, the dynamic graph generation parameters and the weights of the graph convolution layer are adjusted, and only the partial parameters of the graph convolution module are trained with the short-period real-time data under the new topology to achieve rapid fine-tuning of the prediction model. Combining with the real-time updated input data, the accurate prediction of the voltages of each node under the change of the distribution network topology is further realized.

[0022] In a second aspect, the present invention also provides an online prediction device for the node voltages of a time-varying topology DC distribution network, including a memory and one or more processors. Executable code is stored in the memory, and when the processor executes the executable code, the online prediction method for the node voltages of a time-varying topology DC distribution network as described above is implemented.

[0023] Thirdly, the present invention also provides a computer-readable storage medium, on which a program is stored. When the program is executed by a processor, the on-line prediction method for node voltage of a time-varying topology DC distribution network as described above is realized.

[0024] Fourthly, the present invention also provides a computer program product, including a computer program. When the computer program is executed by a processor, the on-line prediction method for node voltage of a time-varying topology DC distribution network as described above is realized.

[0025] Advantages of the present invention:

[0026] (1) The dynamic spatio-temporal graph neural network of the present invention is trained and completed in the off-line stage. When used on-line, by inputting the real-time measurement data of the distribution network nodes and the network topology data, only one network model can be used to quickly predict the voltages of multiple nodes in the distribution network simultaneously, which can provide efficient guidance for the node voltage prediction and system early warning in actual engineering of DC distribution networks.

[0027] (2) The spatio-temporal graph neural network proposed by the present invention accepts the input of dynamic topology information, can autonomously identify topology changes according to the input, and realizes the adaptive fine-tuning of the model within dozens of seconds through the adjustment of a small number of partial structure parameters. The proposed model has high generalization ability and can provide guidance for the voltage prediction of DC distribution network systems with complex operation situations and frequent topology changes. Description of the drawings

[0028] Figure 1 It is a flowchart of an on-line prediction method for node voltage of a time-varying topology DC distribution network proposed by the present invention.

[0029] Figure 2 It is a topology graph of a DC distribution network used for verification of the present invention.

[0030] Figure 3 It is a schematic structural diagram of the dynamic spatio-temporal graph neural network proposed by the present invention.

[0031] Figure 4 It is a flowchart of a method for fine-tuning the model for topology changes proposed by the present invention.

[0032] Figure 5 It is the voltage prediction effect of the dynamic spatio-temporal graph neural network proposed by the present invention at the charging pile node.

[0033] Figure 6 It is the voltage prediction effect of the dynamic spatio-temporal graph neural network proposed by the present invention at the photovoltaic node.

[0034] Figure 7 It is the voltage prediction effect of the dynamic spatio-temporal graph neural network proposed by the present invention when the topology of the distribution network changes.

[0035] Figure 8 This is the structural diagram of an on - line prediction device for node voltages of a time - varying topology DC distribution network provided by the present invention. Detailed implementation manners

[0036] The present invention provides an on - line prediction method for node voltages of a time - varying topology DC distribution network. It can dynamically adjust input data and perform rapid model fine - tuning to further achieve voltage prediction when the topology of the DC distribution network changes due to load switching and other reasons, and can provide guidance for DC distribution network voltage prediction in actual engineering.

[0037] In order to make the objectives, technical solutions and advantages of the present invention clearer and more understandable, the following further elaborates on the specific implementation manners of the present invention in conjunction with the attached drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0038] As shown in the attached Figure 1 drawings, the present invention provides an on - line prediction method for node voltages of a time - varying topology DC distribution network, including the following steps:

[0039] Step 1: As shown in the attached Figure 2 drawings, based on the actual topology of the distribution network, build a simulation model. Using the historical power data of each node and the network topology information of the DC distribution network, simplify the power flow calculation in the traditional AC distribution network into a linear equation in the DC distribution network, calculate the voltage amplitudes of each node in the DC distribution network, and further combine the line on - off states to form training samples. The specific method is as follows:

[0040]

[0041] Among them, represents the set of nodes adjacent to node ; node is the node adjacent to node in the DC distribution network; is the voltage of node at time ; is the voltage of node at time ; is the resistance between node and node ; is the power injected by the power supply of node at time ; is the load power of node at time ; is the current between node and node At the on - off state at a moment, indicates that the line is disconnected, indicates that the line is closed.

[0042] According to the above method, the node voltages at different moments of each node are calculated , combined with the power data of each node 、 and the corresponding line on - off states to form the subsequent dynamic spatio - temporal graph neural network training dataset , where is the number of nodes in the DC distribution network.

[0043] Step 2: As shown in the appendix Figure 3 , construct a dynamic convolutional graph neural network that combines a graph convolutional module and a Transformer time - series processing module. Before starting the model training, first optimize the design of the loss function. Based on the traditional error term of the loss function, add physical constraints to make the model output more in line with the actual power grid operation rules and ensure the reliability of the voltage prediction results. The specific implementation operations are as follows:

[0044]

[0045] Among them, is the final loss function after combining the traditional prediction error term and physical constraints; is the prediction error term; is the Kirchhoff's current law constraint term; is the voltage value predicted by the model; is the true voltage value; is the node at the power injection at a moment; is the node at the load power at a moment; is the weight of the physical constraint term in the loss function; is the dynamic conductance matrix, and the calculation method is as follows:

[0046]

[0047] After preparing the training data and designing the loss function, train the graph convolutional layer part in the model. For the dynamic spatio - temporal graph neural network training dataset obtained through Step 1, perform graph convolutional operations on the data of each node in the order of time steps. The specific operations are as follows:

[0048]

[0049] Among them, is the graph convolution operation function; is the high-dimensional node feature generated after the graph convolution layer operation; is the adjacency matrix; it is dynamically generated by the line on / off state and the line resistance parameter The calculation formula is as follows:

[0050]

[0051] After the training data of the training set undergoes graph convolution operation, the high-dimensional node feature sequences of each node are obtained , T is the time step, that is, the sequence length of the node features in the time dimension. The high-dimensional feature sequence of each node contains the node information of each node , , and the network topology information .

[0052] After obtaining the high-dimensional feature sequence of each node by using the dynamic graph convolution operation, the Transformer time series processing module is used to further process the high-dimensional node feature sequence to learn the voltage time series evolution law and perform node voltage deduction and prediction. The specific operations are as follows:

[0053]

[0054] Among them, is the multi-step voltage prediction result of the Transformer for nodes in the distribution network; is the feature of the last time step obtained after the Transformer performs multiple operations such as positional encoding, attention weighting, and forward propagation on the high-dimensional features of the input nodes; among them in represents extracting the feature of the last time step obtained after the input data undergoes multiple operations; is the transpose of the weight matrix; is the bias term;

[0055] Step 3: Apply the trained dynamic spatio-temporal graph neural network online. Use the real-time topology information of the DC distribution network and the power data of each node as the input of the dynamic spatio-temporal graph neural network. The dynamic spatio-temporal graph neural network outputs the predicted values of the voltage amplitudes of each node in the DC distribution network for a period of time in the future;

[0056] Step 4: Attachment Figure 4 It is a specific operation process for the situation of grid topology change. When the grid topology changes, some structural parameters of the original neural network are frozen, and only some parameters of the graph convolution module are trained with short-period real-time data under the new topology to achieve rapid fine-tuning of the prediction model. Combining with the real-time updated input data, it further realizes the accurate prediction of the voltage of each node under the change of the distribution network topology.

[0057] To reflect the actual effect of the present invention, an embodiment is described by referring to the actual topology of a DC distribution network project in Guangxi Zhuang Autonomous Region and appropriately adjusting and simplifying it. This embodiment builds a distribution network model as shown in Figure 2 the MATLAB / Simulink platform. The referenced DC distribution network is a two-terminal distribution type DC distribution network with a DC bus voltage of 750V. This DC distribution network model has a total of 11 nodes, among which nodes 2, 5, and 6 are charging pile access nodes, nodes 3, 8, 10, and 11 are PV access nodes, and nodes 7 and 11 are defaulted to be disconnected and can be changed to a connected state. Each node has historical power data every 15 minutes / each time. Through model simulation calculation, historical voltage data is obtained. The historical operation data of this area for 360 days is collected, with a collection interval of 15 minutes, a total of 34,560 moments. Data such as voltage and power are used as samples for subsequent neural network training.

[0058] Select the measurement data of each node and the distribution network topology information to form the subsequent dynamic spatio-temporal graph neural network training dataset , where the first 29,560 groups of data are used as training samples, and the last 5,000 groups of data are used as test samples. Further, a dynamic spatio-temporal graph neural network model is constructed. In the constructed dynamic spatio-temporal graph neural network, it includes a dynamic graph generation layer and a graph convolution layer. The number of graph convolution layers is 3, and a residual connection is added to each layer to solve the problems of gradient vanishing and explosion. A Relu activation function layer is added after each graph convolution layer to enhance the non-linear modeling ability of the model. After the graph convolution structure is the Transformer time series processing module. In the Transformer module, the number of encoder layers is 3, the number of attention heads is 8, and layer normalization and residual connection are added to improve the stability of the model. Further, in the model loss function add a physical constraint term based on the KCL condition of the DC distribution network to ensure that the model prediction results conform to the actual grid rules.

[0059] The dynamic spatio-temporal graph neural network model is trained in a regression manner. The network input is the power injection of each node in the DC distribution network in the DC distribution network , the load power of each node in the DC distribution network and the DC distribution network topology information , the output is the voltage prediction values of each node in the next period of time , set the initial learning rate to 0.0001, use the Adam optimization algorithm for training, and the initial number of training rounds is 30. When the power grid topology changes, the topology information input to the model will be adjusted accordingly. At the same time, the model is fine-tuned. At this time, the learned parameters of the Transformer time series processing module will be frozen, and the dynamic graph generation parameters and layer weights of the graph convolutional layer in the graph convolutional module will be fine-tuned. After the model adjustment is completed, the node voltage prediction continues.

[0060] In terms of node voltage prediction, attach Figure 5 shows the short-term voltage prediction effect (predicting 20 time steps backward) of the dynamic spatio-temporal graph neural network at node 5 (charging pile node). Attach Figure 6 shows the short-term voltage prediction effect (predicting 20 time steps backward) of the dynamic spatio-temporal graph neural network at node 11 (photovoltaic node). From the results of the entire test set, the average absolute error of the voltage prediction values of the dynamic spatio-temporal graph neural network for each node is 0.762V, and the maximum absolute error is 1.036V. It can be seen that the dynamic spatio-temporal graph neural network can well learn the trend of voltage changes at each node of the DC distribution network and can accurately predict the voltage amplitude of the nodes. To verify the voltage prediction effect of the proposed method under the condition of topology change, some data before and after topology change are appropriately selected as data input on the time scale. Attach Figure 7 shows the voltage prediction effect of the dynamic spatio-temporal graph neural network at node 11 under the condition of topology change. The topology changes at time 10, and nodes 7 and 11 are connected. The model receives the data under the new topology and quickly fine-tunes. Under the new topology condition, the average absolute error of the voltage prediction values of the dynamic spatio-temporal graph neural network for each node is 0.821V, and the maximum absolute error is 1.729V. It can be seen that the proposed method can quickly fine-tune the model and stably and accurately predict the node voltage for the scenario of topology change.

[0061] Corresponding to the foregoing embodiment of an online prediction method for node voltages of a time-varying topology DC distribution network, the present invention also provides an embodiment of an online prediction device for node voltages of a time-varying topology DC distribution network.

[0062] See Figure 8 , an online prediction device for node voltages of a time-varying topology DC distribution network provided by an embodiment of the present invention includes a memory and one or more processors. An executable code is stored in the memory. When the processor executes the executable code, it is used to implement an online prediction method for node voltages of a time-varying topology DC distribution network in the foregoing embodiment.

[0063] An embodiment of the on-line prediction device for node voltage of a time-varying topology DC distribution network provided by the present invention can be applied to any device with data processing capabilities, and such a device with data processing capabilities can be a device or apparatus such as a computer. The device embodiment can be implemented by software, or by hardware or a combination of software and hardware. Taking software implementation as an example, as a logically meaningful device, it is formed by the processor of any device with data processing capabilities reading the corresponding computer program instructions in the non-volatile memory into the memory for operation. From the hardware level, as Figure 8 shown, it is a hardware structure diagram of any device with data processing capabilities where the on-line prediction device for node voltage of a time-varying topology DC distribution network provided by the present invention is located. In addition to Figure 8 the processor, memory, network interface, and non-volatile memory shown, usually according to the actual functions of any device with data processing capabilities where the device in the embodiment is located, other hardware may also be included, which will not be elaborated here.

[0064] The implementation processes of the functions and roles of each unit in the above device are specifically described in detail in the implementation processes of the corresponding steps in the above method, which will not be elaborated here.

[0065] For the device embodiment, since it basically corresponds to the method embodiment, the relevant parts can be referred to the partial description of the method embodiment. The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place, or may be distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of the present invention. Those of ordinary skill in the art can understand and implement it without creative efforts.

[0066] The embodiment of the present invention also provides a computer-readable storage medium, on which a program is stored. When the program is executed by a processor, it implements the on-line prediction method for node voltage of a time-varying topology DC distribution network in the above embodiment.

[0067] The computer-readable storage medium may be an internal storage unit of any device with data processing capabilities described in any of the foregoing embodiments, such as a hard disk or memory. The computer-readable storage medium may also be an external storage device of any device with data processing capabilities, such as a plug-in hard disk, a Smart Media Card (SMC), an SD card, a Flash Card, etc. equipped on the device. Further, the computer-readable storage medium may also include both an internal storage unit and an external storage device of any device with data processing capabilities. The computer-readable storage medium is used to store the computer program and other programs and data required by any device with data processing capabilities, and may also be used to temporarily store data that has been output or is to be output.

[0068] The present invention also provides a computer program product, including a computer program, which when executed by a processor, implements the online prediction method for node voltage of a time-varying topological DC distribution network described above.

[0069] The above are only the preferred embodiments of the present invention, and the protection scope of the present invention is not limited to the above embodiments. All technical solutions falling within the concept of the present invention belong to the protection scope of the present invention. It should be noted that for those of ordinary skill in the art in this technical field, several improvements and refinements made without departing from the principle of the present invention should be regarded as within the protection scope of the present invention.

Claims

1. An online prediction method for node voltage in a time-varying topology DC distribution network, characterized in that, It includes the following steps: Step 1: Build a simulation model based on the historical operation data and the real topological structure of the DC distribution network. Generate a time-series training data set including voltage, power, and line status through dynamic topology simulation. Build a simulation model based on the real distribution network topology. Using the historical operation data of each node and the network topology information of the DC distribution network, simplify the power flow calculation in the AC distribution network into a linear equation in the DC distribution network, calculate the voltage amplitude of each node in the DC distribution network, and further combine the line on / off status to form training samples. The specific method is as follows: ； Among them, represents the set of nodes adjacent to node is the node adjacent to node in the DC distribution network; is the voltage of node at time is the voltage of node at time ; is the resistance between node and node ; is the power injection of node at time ; is the load power of node at time ; is the on / off state of node and node at time ; Calculate the node voltages of each node at different times , combined with the power data of each node , and the corresponding line on-off status to form a training dataset for the subsequent dynamic spatio-temporal graph neural network; Step 2: According to the training data obtained in Step 1, form the required training samples, and train a dynamic spatio-temporal graph neural network that integrates topological dynamic changes. Capture the real-time association between nodes through the dynamic graph convolutional layer to obtain high-dimensional node features. After processing, further learn the voltage time-series evolution law and perform node voltage deduction and prediction, and embed the physical constraints in the DC distribution network in the loss function to train a complete dynamic spatio-temporal graph neural network. In the loss calculation of the dynamic spatio-temporal graph neural network during model training, the loss function adds physical constraints on the basis of the error term to make the model output more in line with the actual physical laws. The specific operation is as follows: ； Among them, is the final loss function after combining the prediction error term and physical constraints; is the prediction error term; is the Kirchhoff's current law constraint term; is the voltage value predicted by the model; is the true voltage value; is the node at the power injection of the power source at time; is the node at the load power at time; is the weight of the physical constraint term in the loss function; is the dynamic conductance matrix; Step 3: According to the dynamic spatio-temporal graph neural network obtained in Step 2, input the node power data and network topology information collected in real time, and the dynamic spatio-temporal graph neural network outputs the predicted values of the future voltages of each node. Step 4: According to the dynamic spatio-temporal graph neural network obtained in Step 2, when the power grid topology changes, fine-tune the model of the dynamic graph neural network by updating the graph connection parameters associated with the changed nodes to ensure that the model can be quickly adjusted to continue prediction.

2. The on-line prediction method for node voltage of a time-varying topology DC distribution network according to claim 1, characterized in that Build a dynamic spatio-temporal graph neural network. For the time-series training data set obtained in Step 1, perform graph convolution operations on the data of each node in the order of time steps. After the training data of the training set undergoes graph convolution operations, high-dimensional node feature sequences for each node are obtained, and the node information of each node is included in the high-dimensional feature sequence of each node , 、 and network topology information 。 3. An online prediction method for node voltage of a time-varying topology DC distribution network according to claim 2, characterized in that, In the time series processing module of the dynamic spatio-temporal graph neural network, the Transformer is used to further process the high-dimensional node feature sequence, learn the law of voltage time series evolution, and predict the node voltage deduction of nodes in the distribution network.

4. An online prediction method for node voltage of a time-varying topology DC distribution network according to claim 1, characterized in that, Perform online application on the trained dynamic spatio-temporal graph neural network. Take the real-time topological information of the DC distribution network and the power data of each node as the input of the dynamic spatio-temporal graph neural network, and the dynamic spatio-temporal graph neural network outputs the predicted values of the voltage amplitudes of each node in the DC distribution network in the future for a period of time.

5. An online prediction method for node voltage of a time-varying topology DC distribution network according to claim 1, characterized in that, For the case of power grid topology change, freeze the structural parameters of the original neural network time-series processing module, adjust the dynamic graph generation parameters and the weights of the graph convolutional layer, and only train some parameters of the graph convolutional module with short-period real-time data under the new topology to achieve fast fine-tuning of the prediction model. Combine the real-time updated input data to further achieve accurate prediction of the voltages of each node under the change of the power grid topology.

6. An on-line prediction device for node voltage of a time-varying topology DC distribution network, comprising a memory and one or more processors, wherein executable code is stored in the memory, characterized in that, When the processor executes the executable code, it implements an online prediction method for the node voltage of a time-varying topology DC distribution network as described in any one of claims 1-5.

7. A computer-readable storage medium having a program stored thereon, characterized in that, When the program is executed by the processor, it implements an online prediction method for the node voltage of a time-varying topology DC distribution network as described in any one of claims 1-5.

8. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements an online prediction method for the node voltage of a time-varying topology DC distribution network as described in any one of claims 1-5.

Images