Fragmented reactive power response evaluation method and system for power system based on multimodality and graph convolution

Abstract

The present invention discloses a method and system for evaluating the fragmented reactive power response of an electric power system based on multimodality and graph convolution. The method comprises: collecting relevant characteristic quantity factors affecting the reactive power of the electric power system; classifying the data corresponding to the relevant characteristic quantity factors according to the time of their collection; constructing a reactive power data convergence; performing multimodal processing on all nodes, determining the direction between adjacent nodes, and thus obtaining an energy flow diagram of reactive information in reactive power; using a graph convolutional neural network to obtain the fusion of the energy flow diagram of reactive information in reactive power and the data energy flow at each node; the present invention fully analyzes the temporal sequence and equipment spatial distribution characteristics of power generation and power consumption equipment in each part of the new electric power system, and uses a multimodal graph convolutional network and a spatial attention mechanism to establish a multi-time scale evaluation model, through which reactive power can be more effectively monitored, predicted and adjusted, and optimal management of the power grid can be achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of power distribution networks, and in particular to a method and system for evaluating fragmented reactive power response of a power system based on multi-modality and graph convolution. Background Art

[0002] In the current power system, with the large-scale access of renewable energy, traditional power grids are facing unprecedented challenges. Especially in distributed and discrete new energy bases, such as wind power and photovoltaic power, the grid structure is usually weak, far away from the load center, and lacks sufficient synchronous power support. These factors together lead to low interference immunity and weak support of the power grid, which significantly reduces the voltage stability margin and dynamic voltage support capability. Therefore, it is particularly important to accurately evaluate and optimize the management of reactive power.

[0003] Effective management of reactive power is of great significance for improving the voltage stability of the power grid, optimizing the power quality and improving the operating efficiency of the system. However, in a fragmented power grid environment, the uneven distribution and difficulty of reactive power regulation bring additional complexity. Constructing a comprehensive evaluation index system to evaluate the potential and regulation effect of fragmented reactive power in the power system has become a key means to improve the performance and reliability of the power grid.

[0004] The comprehensive evaluation index system constructed includes the following aspects:

[0005] 1. Reactive power regulation capability: Evaluate the ability of different power sources to regulate reactive power under varying load conditions.

[0006] 2. Voltage stability: measures the stability and resilience of grid voltage under different operating conditions.

[0007] 3. Dynamic response capability: Analyze the system's response speed and effectiveness to load changes and fault conditions.

[0008] 4. Economic feasibility: Evaluate the economic feasibility of reactive power regulation measures from a cost-effectiveness perspective, including operating costs and maintenance costs.

[0009] 5. Environmental impact: Evaluate the impact of reactive power regulation on the environment, especially when using fossil fuel power sources.

[0010] Combining these indicators can more comprehensively understand and optimize the potential of fragmented reactive power in the power system, thereby improving the efficiency and reliability of the entire system. However, in practice, these indicators come from different devices, data types using different modes, and data from different software versions. Therefore, a comprehensive evaluation of multimodal data is required, which not only helps stabilize the operation of the power grid, but also promotes a higher proportion of renewable energy access and sustainable development of the power system. The existing technology does not yet have a method for efficient and accurate processing of the above multimodal data. Summary of the invention

[0011] Purpose of the invention: To solve the problem of unusable fragmented reactive power between various plant areas in the existing new power system. The present invention fully analyzes the temporal sequence and equipment spatial distribution characteristics of the power generation and power consumption equipment in each part of the new power system, and uses a multi-modal graph convolutional network and spatial attention mechanism to establish a multi-time scale evaluation model.

[0012] Technical solution: According to a first aspect of the present invention, a method for evaluating fragmented reactive power response of a power system based on multimodality and graph convolution is provided, the method comprising:

[0013] S1 collects relevant characteristic quantity factors that affect the reactive power of the power system, including: bus voltage, line flow diagram, voltage margin, active margin, reactive margin, transformer load rate, node voltage offset, system overload degree, system load rate and node voltage change rate index. The above characteristic quantity factors are used to characterize the state of fragmented energy in the power system;

[0014] S2 classifies the data corresponding to the relevant feature quantity factors according to the time of their collection to obtain a corresponding data set, wherein the data set includes multiple modal forms, wherein the modal forms include: text, image, value, pulse or waveform;

[0015] S3 constructs reactive power data confluence, firstly, the power data acquisition equipment, the substation relay equipment and the server receiving equipment in the power system are all regarded as nodes of a power data confluence, and the data set is used as the input node of the data confluence, mapped to the corresponding power data acquisition equipment, the substation relay equipment or the server receiving equipment, to form multiple nodes;

[0016] S4 performs multimodal processing on all nodes to determine the direction between adjacent nodes, thereby obtaining an energy flow diagram of reactive information in reactive power; a graph convolutional neural network is used to obtain the fusion of the energy flow diagram of reactive information in reactive power and the data energy flow at each node;

[0017] After S5 obtains the corresponding graph of the data energy flow, the amount of fragmented reactive power available for each node can be obtained, and the amount of fragmented reactive power available for each node can be evaluated.

[0018] Further, it includes:

[0019] In step S4, multi-modal processing is performed on all nodes to determine the direction between adjacent nodes, thereby obtaining an energy flow diagram of reactive power information in reactive power, including:

[0020] Step 1: According to the functional classification of the server in the power system, it includes: power data acquisition device A, substation area relay device B, and server receiving device C;

[0021] Step 2: Pack the data set at a certain moment or during a certain period collected into corresponding data packets, perform spatial positioning on the data packets according to the location of the corresponding server, and number them and mark their location information;

[0022] Step 3: Input the corresponding data packets at the same moment or during a certain period into the power data acquisition device A respectively, and transmit them in the substation area relay device B and the server receiving device C, thereby forming multiple nodes of the energy flow diagram;

[0023] Step 4: Align the different modal data from asynchronous distributions according to the time sequence to obtain the edge vectors corresponding to the energy flow diagram, and then construct an energy flow diagram with directions.

[0024] Further, it includes:

[0025] In step 4, align the different modal data from asynchronous distributions according to the time sequence to obtain the edge vectors corresponding to the energy flow diagram, specifically including:

[0026] If only considering the transmission process of the data packets at the same moment, then determine the direction between the adjacent nodes corresponding to the same data packet, and use the multi-modal processing method to calculate the data value of the data packet in the corresponding server of the node, denoted as X , and calculate the data value of the data packet in the corresponding server of the adjacent node in the same way, denoted as Y. If X < Y, then the direction is from this node to the adjacent node; otherwise, if X Y, then the direction is from the adjacent node to this node;

[0027] Otherwise, if considering the transmission process of the data packets over a period of time, then it is also necessary to calculate the direction between the adjacent nodes corresponding to the data packets at different times. The steps adopted are:

[0028] First, determine whether the servers corresponding to the two nodes are the same. If they are the same, directly compare the time of the two data packets, set the direction according to the time sequence, and the corresponding data packet with the earlier time points to the corresponding data packet with the later time. Otherwise, if the two nodes correspond to different servers, first determine the time and calculate the data value of the data packet after being processed by the corresponding server using the multimodal method. The direction is determined only when the time sequence and data value size mentioned above are met. If only one of them is met or neither is met, the corresponding direction is not set.

[0029] Further, including:

[0030] The multimodal processing method comprises the following steps:

[0031] The multimodal dataset and node data transmitted in the node are converted into feature space using linear transformation, which is expressed as: in, is the transformation coefficient, which is determined by the server type, node status, multi-modal type and actual project status of the data. a Represents the node data value corresponding to the power data acquisition device A, b is the node data value corresponding to relay device B, c The server receives the node data value corresponding to device C, and They are the data sets before and after linear transformation respectively;

[0032] Calculate the influence of the surrounding nodes on the modal data of the central node, expressed as: in, is the side information, indicating and The transmission process and the degree of mutual influence, leaky ReLU() is the selected activation function, which is a variant of the ReLU() activation function, which allows , The value is negative, and a non-zero gradient is added to the negative value. Leaky ReLU() helps to keep the gradient flow of the network, so that more neurons remain active;

[0033] Normalize the data to improve the generalization of the algorithm. exp represents the natural logarithm e, For edge information, the normalization process is as follows: Further, including:

[0034] In step S4, a graph convolutional neural network is used to obtain the fusion of the energy flow diagram of reactive information in reactive power and the data energy flow at each node, including:

[0035] Influencing information Perform convolution encapsulation and represent the influence of the node on the next edge at the node, encapsulating it as , which is calculated based on the surrounding nodes combined with attention, that is: Where, N is the total number of nodes, N i For the i nodes; So far, we have obtained the data node Z obtained through convolution encapsulation, and the edge information By obtaining these two pieces of information, we can get the energy flow diagram of reactive information in reactive power evaluation.

[0036] Further, including:

[0037] The method further includes:

[0038] In multimodal fusion processing, if there are too many nodes linked to the same node, it will cause data bloat and congestion, and it will be impossible to focus on the content containing more information. Graph convolution is needed to remove the edges whose mean is less than a certain threshold.

[0039] That is, in calculation After that, if If it is less than a certain threshold, the corresponding package Z data can be directly deleted to remove the invalid edges;

[0040] According to the vector value of each node and the heat map of the correlation between each node, the node fusion status and the data energy flow at the data node can be known.

[0041] Further, including:

[0042] In step S5, the amount of fragmented reactive power available for use is evaluated, including: after obtaining the energy flow, the amount of fragmented reactive power available for use at each node can be obtained, but the result needs to be evaluated to measure the accuracy of the model, and the root mean square error RMSE indicator is used to measure the accuracy of the model reactive power evaluation, and its expression is as follows: in, represents the fusion result, Indicates the actual situation of the node. Indicates the number of nodes for reactive power evaluation. The smaller the RMSE value, the more accurate the fusion result.

[0043] On the other hand, the present invention also provides a fragmented reactive power response evaluation system for a power system based on multimodality and graph convolution, the system comprising:

[0044] The acquisition module is used to collect relevant characteristic quantity factors that affect the reactive power of the power system, including: bus voltage, line flow diagram, voltage margin, active margin, reactive margin, transformer load rate, node voltage offset, system overload degree, system load rate and node voltage change rate indicators. The above characteristic quantity factors are used to characterize the state of fragmented energy in the power system;

[0045] A data set construction module is used to classify the data corresponding to the relevant feature quantity factors according to the time of their collection to obtain a corresponding data set, wherein the data set includes multiple modal forms, wherein the modal forms include: text, image, value, pulse or waveform;

[0046] The node construction module is used to construct reactive power data confluence. First, the power data acquisition equipment, the substation relay equipment and the server receiving equipment in the power system are all regarded as nodes of a power data confluence, and the data set is used as an input node of the data confluence, mapped to the corresponding power data acquisition equipment, the substation relay equipment or the server receiving equipment to form multiple nodes.

[0047] The direction determination module is used to perform multimodal processing on all nodes and determine the direction between adjacent nodes, so as to obtain the energy flow diagram of reactive information in reactive power; a graph convolutional neural network is used to obtain the fusion of the energy flow diagram of reactive information in reactive power and the data energy flow at each node;

[0048] The evaluation module is used to obtain the amount of fragmented reactive power available for each node according to the corresponding graph of the data energy flow situation, and to evaluate the amount of fragmented reactive power available for each node.

[0049] Further, including:

[0050] In the direction determination module, multi-modal processing is performed on all nodes to determine the direction between adjacent nodes, thereby obtaining an energy flow diagram of reactive information in reactive power, including:

[0051] Step 1: Classify the server according to its functions in the power system, including: power data acquisition equipment A, area relay equipment B and server receiving equipment C;

[0052] Step 2: Pack the collected data sets at a certain time or time period into corresponding data packets, spatially locate the data packets according to the location of the corresponding server, and number and mark their location information;

[0053] Step 3: Input the corresponding data packets at the same moment or a certain time period into the power data acquisition device A, and transmit them in the substation relay device B and the server receiving device C, thereby forming multiple nodes of the energy flow diagram;

[0054] Step 4: Align different modality data from asynchronous distribution according to the time sequence to obtain the edge vectors corresponding to the energy flow diagram, and then construct an energy flow diagram with directions.

[0055] Further, it includes:

[0056] In the said Step 4, aligning different modality data from asynchronous distribution according to the time sequence to obtain the edge vectors corresponding to the energy flow diagram specifically includes:

[0057] If only considering the transmission process of data packets at the same moment, then determine the direction between adjacent nodes corresponding to the same data packet, and use the multi-modal processing method to calculate the data value of the data packet in the server corresponding to the node, denoted as X , and calculate the data value of the data packet in the server corresponding to the adjacent node in the same way, denoted as Y. If X < Y, then the direction is from this node to the adjacent node; otherwise, if X ≥ Y, then the direction is from the adjacent node to this node;

[0058] Otherwise, if considering the transmission process of data packets within a period of time, then it is also necessary to calculate the direction between adjacent nodes corresponding to data packets at different times. The steps adopted are as follows:

[0059] First, determine whether the servers corresponding to the two nodes are the same. If they are the same, then directly compare the times of the two data packets and set the direction according to the time sequence, with the data packet with an earlier time pointing to the data packet with a later time. Otherwise, if the servers corresponding to the two nodes are different, then first determine the time and calculate the data values of the data packets processed by the multi-modal method in the corresponding servers. The direction is determined only when both the above-mentioned time sequence and data value size are satisfied. If only one of them is satisfied or neither is satisfied, then the corresponding direction is not set.

[0060] Beneficial effects: Compared with the prior art, the present invention has the following advantages:

[0061] The present invention maps the multimodal data set to the corresponding server, and numbers the corresponding data packets. After the mapping is completed, the server is classified, and the data packets are transmitted to the corresponding server as input, and then the corresponding nodes are first obtained, and then from point to line, where the point is the server mapping, the content inside is the data set, the position is the server position, and the node type is related to the server type. On this basis, the data from different modal forms of asynchronous distribution are aligned according to the time sequence, and the data node graph is constructed, and then the nodes and the directional energy flow graph are obtained. After the above graph is convolved, the energy flow graph of reactive information in reactive power evaluation is obtained. Through the construction of the energy flow graph, the transmission process of the data flow in reality is mapped to the algorithm as completely as possible, so that after calculation, the flow process of the reactive source is accurately controlled, and the precise control of reactive power in a large range is completed. Therefore, the present invention fully analyzes the timing and spatial distribution characteristics of the power generation and power-consuming equipment in each part of the new power system, and adopts a multi-modal graph convolutional network and a spatial attention mechanism to establish a multi-time scale evaluation model. Through this model, reactive power can be more effectively monitored, predicted and adjusted to achieve optimal management of the power grid. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] Figure 1 A schematic diagram of a novel power system fragmented reactive power response evaluation method of a multimodal graph convolutional network according to the present invention;

[0063] Figure 2 This is a flow chart of fragmented reactive power evaluation based on a multimodal data fusion algorithm according to an embodiment of the present invention;

[0064] Figure 3 A thermal result diagram of the correlation of reactive power related features based on the multimodal data fusion algorithm according to an embodiment of the present invention;

[0065] Figure 4 The fragmented reactive power described in the embodiment of the present invention can be applied to the reactive power evaluation result diagram;

[0066] Figure 5 A schematic diagram of the positional relationship of multimodal data in a server according to an embodiment of the present invention;

[0067] Figure 6 This is a data preprocessing flow chart according to an embodiment of the present invention. DETAILED DESCRIPTION

[0068] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention. Embodiment 1

[0069] like Figure 1 As shown, the present invention provides a novel power system fragmentation reactive power response evaluation method based on a multimodal graph convolutional network, and the steps are as follows:

[0070] Step 1: Data collection. First, collect relevant characteristic factors that affect the reactive power of the new power system, including bus voltage, line flow diagram, voltage margin, active power margin, reactive power margin, transformer load rate, node voltage offset, system overload degree, system load rate, and node voltage change rate indicators, so as to characterize the state of fragmented energy in the power system as much as possible. Among them, bus voltage, line flow diagram, voltage margin, active power margin, reactive power margin, and transformer load rate directly affect the utilization path and available energy of reactive power in the system, while node voltage offset, system overload degree, system load rate, and node voltage change rate will affect the stability of the system and thus affect the utilization efficiency of reactive power. In addition, collect regional load-related data, covering historical load data and user electricity consumption behavior, so as to interact with reactive power utilization later.

[0071] Step 2: data preprocessing.

[0072] According to the data information collected in the previous step, first, the bus voltage, line flow diagram, voltage margin, active power margin, reactive power margin, transformer load rate, node voltage offset, system overload degree, system load rate, and node voltage change rate indicators are classified according to the modal form in which they are collected, and the data sets are formed according to the modal form, where the modal forms include: text, image, value, pulse and waveform.

[0073] For example, in an embodiment of the present invention, data corresponding to all modal forms of bus voltage indicators within a period of time are formed into a data set, or data corresponding to all modal forms of line power flow diagram indicators within a period of time are formed into a data set.

[0074] Secondly, after the data set is constructed, data packaging and transmission are carried out. Each new power system data collection point, relay point and server node is regarded as a data confluence node, and the data set collected in step 1 is used as the data of the input node to construct the reactive power data flow.

[0075] The process of building reactive power data stream specifically includes:

[0076] Taking the transmitted data packet as the content of the node and the transmission flow direction as the edge, we determine the data set to be packaged and transmitted between servers. We need to locate the server node in space and mark its location information. Specifically, since the data input is the content of the node and the node itself is the mapping of the server in the graph, when constructing the node in the graph, we also need to mark the spatial position of the node to draw the image. This position corresponds to the actual position of the server. That is, the data collection points, relay nodes and server nodes correspond to the positions in multiple servers.

[0077] like Figure 5 As shown in the figure, based on the location information, the power communication data and server information can be constructed as a node in the graph convolutional network. The data flow information to be constructed is represented by Π. According to the different ways of processing the data set, the related equipment can be divided into three categories. In the following, this article will use A, B, and C to represent different types of node data.

[0078] Among them, ABC is classified according to the role played by its node in the data transmission process. (1) There are multiple power data collection devices A, substation relay devices B, and server receiving devices C. The corresponding data packets are transmitted to the power data collection devices A, substation relay devices B, and server receiving devices C to form multiple nodes.

[0079] Therefore, the purpose of the embodiment of the present invention is to map the actual data flow into a graph. The mapping method divides the data set according to the purpose of the device, such as mapping the collected data set to the corresponding server, and numbering the corresponding data packets. After the mapping is completed, the computer can calculate according to the graph. The drawing method is: from point to line, the point is the server mapping, the content inside is the data set, the location is the server location, and the node type is related to the server type, that is, the three types mentioned above.

[0080] Finally, the data from different modal forms of asynchronous distribution are aligned according to the time sequence, and a data node graph is constructed. The asynchronous distribution here refers to the corresponding values ​​of data at different times or time periods. The purpose of alignment is to prevent the time difference caused by the transmission process, but it is actually the same data packet, from being processed as different data packets by constructing directed edges.

[0081] like Figure 6 As shown, the line segments connecting nodes have directions. In this embodiment, how to determine whether server A flows to server B or server B flows to server A depends on the change pattern of the internal data of their nodes. If it increases gradually, it is positive, otherwise it decreases.

[0082] In actual data processing work, it is first mapped into an image on a computer. After mapping, the computer calculates and analyzes the image. This part of the embodiment describes how to map, using points and edges to draw a picture so that the computer can understand it.

[0083] Specifically, since the attention operation used in the fusion stage is non-aligned, the edge construction needs to be directed. The direction of the directed vector is determined by the data change. For example, if the modal data in a relay node B increases over time, the vector direction is a positive vector, and its change slope is set as the vector direction. The vector sum of all modal data in the same node is defined as the node vector, where the vector direction is determined according to the data value calculated by the corresponding data packet in different servers after the linear transformation of multimodal processing. Specifically, if the data value calculated by the data packet on server A is x , the data value calculated on server B is y ,and x < y , then the direction of the vector is AB.

[0084] The above method can be used to determine the direction between adjacent nodes for the same data packet and different data packets, so as to obtain all energy flow node vectors within a period of time. Specifically: if only the data set collected at a certain moment is considered, it is only necessary to calculate the direction between two adjacent nodes of the same data packet. If the data value of the data packet calculated by the server in multimodal processing is less than the server data value corresponding to the adjacent node, the vector direction points to the server corresponding to the adjacent node;

[0085] If processing is performed on data collected over a period of time, it is also necessary to calculate the direction between two adjacent nodes of different data packets, that is, first determine whether the servers corresponding to the two nodes are the same. If they are the same, directly compare the time of the two data packets, and set the direction according to the time sequence. The corresponding data packet with the earlier time points to the corresponding data packet with the later time. Otherwise, if the two nodes correspond to different servers, first determine the time and calculate the data value of the data packet after the corresponding server modal processing. The direction is determined only when the time sequence and data value size mentioned above are met. If only one of them is met or neither is met, the corresponding direction is not set.

[0086] If the node vector of node A is positive, and the data flowing out of A is input into the next node B, then an edge vector graph from A to B is constructed.

[0087] This embodiment is exemplified as follows, and the following flow directions represent A to B, A to C, and B to C. (2) Perform the above processing on the data set in the modal form corresponding to each indicator to obtain the energy flow node vector, thereby constructing the edge vector graph of the energy flow.

[0088] Since these indicators in the embodiment of the present invention come from different devices, data types using different modes, and data of different software versions (referred to as multimodal data), it is necessary to perform a comprehensive evaluation on the multimodal data.

[0089] Therefore, in the selected method, a multimodal graph convolutional neural network (GCNN) is a deep learning model used to process graph structure information with multiple types of data (multimodal data). This model combines the graph structure processing capability of the GCNN and the multimodal data processing capability to improve the understanding and prediction accuracy of complex data.

[0090] In the fragmented reactive power evaluation network, each node contains not only traditional single-form features (such as text or numerical data), but also different forms of data such as images, sounds, and videos. These different types of data are integrated together and processed through a graph convolutional network to more comprehensively capture the intrinsic connections and complex patterns between data. Its method can match the needs of fragmented reactive power evaluation. It can well solve the problem of a surge in parameters and orders caused by the increase in the number of data types and complex connections in the evaluation and the increase in the number of hidden layer nodes.

[0091] Firstly, multimodal processing is performed on the features of data stream nodes for reactive power collection of all new power systems. The present invention uses linear transformation to convert the multimodal data sets transmitted in the nodes into feature space. (3) (4) Among them, Represents the fragmented reactive power evaluation data flow, for each node , and the surrounding nodes are Formula 3 is a linear transformation. is the transformation coefficient, which is determined by the server type, node status, multi-modal type and actual project status of the data. , Represent the data values ​​of the data set before and after the linear transformation respectively.

[0092] After the linear transformation, the multimodal features can be used in the form of numerical values ​​to participate in the unified calculation. type identifier as well as temporal identifier Two identifiers are used to calculate the influence of surrounding nodes on the modal data of the central node, such as Figure 2 shown. (5) Among them, is the side information, indicating and The transmission process and the degree of mutual influence. leaky ReLU (LeakyRectified Linear Unit) is type identifier as well as temporal identifier The activation function selected by the recognizer is a variant of the ReLU (Rectified Linear Unit) activation function. type identifier as well as temporal identifier Identifier transformation can allow , By taking a negative value and adding a non-zero gradient to negative values, Leaky ReLU helps keep the gradient of the network flowing, allowing more neurons to remain active. When the stable process and the transmission process are j Node and i The degree of mutual influence between nodes.

[0093] In impact analysis, due to differences in data types and collection devices, the data needs to be normalized to improve the generalization of the algorithm. exp Represents the natural logarithm e , For edge information, the normalization process is as follows: (6) According to the above multimodal processing method, the normalized data value of each data packet is obtained and the corresponding node influence is obtained.

[0094] Step 3: Model construction

[0095] After the multimodal processing is completed, the graph convolution model calculation is performed.

[0096] The influence information is convoluted and encapsulated, and the influence of the node on the next edge is represented at the node, which is encapsulated as , which is calculated based on the surrounding nodes combined with attention, that is: (7) Where N is the total number of nodes, N i For the i At this point, we can get the data node Z obtained through convolution encapsulation and the edge information By obtaining these two pieces of information, we can obtain the energy flow diagram of reactive information in reactive power evaluation.

[0097] In multimodal fusion processing, if there are too many nodes linked to the same node, it will cause data bloat and congestion, and it will be impossible to focus on the content containing more information. Graph convolution is needed to remove edges whose mean is less than a certain threshold.

[0098] In calculation After that, if If it is less than a certain threshold, the corresponding package Z data can be directly deleted to remove the invalid edges.

[0099] According to the vector value of each node and the heat map of the correlation between each node, the node fusion status and the data energy flow at the data node can be known.

[0100] Step 4: reactive power evaluation. After obtaining the energy flow, the amount of fragmented reactive power available for each node can be obtained. However, the results need to be evaluated to measure the accuracy of the model. The present invention uses the root mean square error RMSE indicator to measure the accuracy of the model reactive power evaluation. Its expression is as follows: (10) Among them, represents the fusion result, Indicates the actual situation of the node. Indicates the number of nodes for reactive power evaluation. The smaller the RMSE value, the more accurate the fusion result. two

[0102] The present invention also provides a power system fragmentation reactive power response evaluation system based on multi-modality and graph convolution, the system comprising:

[0103] The acquisition module is used to collect relevant characteristic quantity factors that affect the reactive power of the power system, including: bus voltage, line flow diagram, voltage margin, active margin, reactive margin, transformer load rate, node voltage offset, system overload degree, system load rate and node voltage change rate indicators. The above characteristic quantity factors are used to characterize the state of fragmented energy in the power system;

[0104] A data set construction module is used to classify the data corresponding to the relevant feature quantity factors according to the time of their collection to obtain a corresponding data set, wherein the data set includes multiple modal forms, wherein the modal forms include: text, image, value, pulse or waveform;

[0105] The node construction module is used to construct reactive power data confluence. First, the data acquisition equipment, relay equipment and server receiving equipment in the power system are all regarded as nodes of a power data confluence, and the data set is used as an input node of the data confluence, mapped to the corresponding power data acquisition equipment, area relay equipment or server receiving equipment to form multiple nodes;

[0106] The direction determination module is used to perform multimodal processing on all nodes and determine the direction between adjacent nodes, so as to obtain the energy flow diagram of reactive information in reactive power; a graph convolutional neural network is used to obtain the fusion of the energy flow diagram of reactive information in reactive power and the data energy flow at each node;

[0107] An evaluation module, configured to obtain the amount of fragmented reactive power available at each node according to the corresponding graph of the data energy flow situation, and evaluate the amount of fragmented reactive power available at each node.

[0108] Further, it includes:

[0109] In the direction determination module, multi-modal processing is performed on all nodes to determine the direction between adjacent nodes, so as to obtain the energy flow graph of the reactive power information, including:

[0110] Step 1: Classify according to the functions of the server in the power system, including: power data acquisition device A, substation area relay device B, and server receiving device C;

[0111] Step 2: Package the data set collected at a certain moment or during a certain period into corresponding data packets, perform spatial positioning on the data packets according to the location of the corresponding server, and number and label their location information;

[0112] Step 3: Input the corresponding data packets at the same moment or during a certain period into the power data acquisition device A respectively, and transmit them in the substation area relay device B and the server receiving device C, thereby forming multiple nodes of the energy flow graph;

[0113] Step 4: Align the different modal data from asynchronous distributions according to the time sequence to obtain the edge vectors corresponding to the energy flow graph, and then construct an energy flow graph with directions.

[0114] Further, it includes:

[0115] In step 4, aligning the different modal data from asynchronous distributions according to the time sequence to obtain the edge vectors corresponding to the energy flow graph specifically includes:

[0116] If only considering the transmission process of the data packets at the same moment, then determine the direction between the adjacent nodes corresponding to the same data packet, and use the multi-modal processing method to calculate the data value of the data packet in the corresponding server of the node, denoted as X , and calculate the data value of the data packet in the corresponding server of the adjacent node in the same way, denoted as Y. If X < Y, then the direction is from this node to the adjacent node; otherwise, if X ≥ Y, then the direction is from the adjacent node to this node;

[0117] Otherwise, if considering the transmission process of the data packets over a period of time, then it is also necessary to calculate the direction between the adjacent nodes corresponding to the data packets at different times. The steps adopted are:

[0118] First, determine whether the servers corresponding to the two nodes are the same. If they are the same, directly compare the time of the two data packets, set the direction according to the time sequence, and the corresponding data packet with the earlier time points to the corresponding data packet with the later time. Otherwise, if the two nodes correspond to different servers, first determine the time and calculate the data value of the data packet after being processed by the corresponding server using the multimodal method. The direction is determined only when the time sequence and data value size mentioned above are met. If only one of them is met or neither is met, the corresponding direction is not set.

[0119] Other technical features of the power system fragmented reactive power response evaluation system based on multimodality and graph convolution described in the present invention are similar to the corresponding methods and will not be repeated here.

[0120] Figure 3 It is the "Thermal result diagram of correlation of reactive power-related features based on multimodal data fusion algorithm", which shows the reactive power distribution of each node in the reactive power evaluation system and the connection between the nodes. Through this figure, you can intuitively see the fragmented application and concentration points of reactive power in the distribution network, which is crucial for the application of fragmented reactive power. The data used in the design of this invention is derived from the actual operating data provided by State Grid Corporation of China in the first quarter of 2023. The heat map uses a gradient color from light to dark, where light colors represent areas with lower reactive power and dark colors indicate peak reactive power areas. From Figure 3 It can be clearly seen that the distribution of reactive power is not uniform, but shows obvious aggregation in some areas. This distribution characteristic suggests that there may be potential problems with voltage stability in the region. In addition, some isolated dark high-value areas can be seen in the figure, which may be related to the improper application of reactive power by local power equipment. The analysis results of the heat map show that optimizing the distribution and adjustment of reactive power is the key to improving the stability and efficiency of the power grid. Based on this, it is recommended that grid operators add reactive compensation equipment such as capacitors and phase shifters in areas with high reactive power concentration to improve voltage quality and reduce losses. In addition, such heat maps can also be used as a basis for implementing dynamic electricity price policies and planning power grid upgrades.

[0121] Figure 4 Titled "Evaluation diagram of reactive power available at each node in the distribution system", it shows in detail the reactive power applicability evaluation of each node in the distribution network. This diagram is used to evaluate the performance and potential of each node in reactive power management and is a key analysis tool for optimizing power system performance. The data used in the design of this invention is derived from the actual operating data of four typical days in the first quarter of 2023 provided by State Grid Corporation of China, and the peak period of reactive power abnormality in the first quarter was selected. Figure 4It can be observed that the scores of the central area and important load nodes of the system are higher, indicating that the reactive power management effect in these areas is better, and the stability and reliability are higher. The lowest score of reactive power evaluation is 10 points, which appears at nodes 3 and 16, and node 3 has low scores on two typical days, so there may be reactive power problems. Through the comprehensive evaluation of the reactive power of each node in the new power system, key improvement areas can be identified and targeted improvement measures can be proposed. For example, adding reactive compensation devices, optimizing voltage regulation strategies, or upgrading aging equipment. In addition, the evaluation chart can also serve as an important reference for the design of new power systems or power grid transformation to achieve higher energy efficiency and power grid security.

[0122] It will be appreciated by those skilled in the art that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented 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.

[0123] Although the preferred embodiments of the present invention have been described, those skilled in the art may make other changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention.

[0124] Obviously, those skilled in the art can make various changes and modifications to the embodiments of the present invention without departing from the spirit and scope of the embodiments of the present invention. Thus, if these modifications and variations of the embodiments of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A fragmented reactive power response evaluation method for power systems based on multimodality and graph convolution, characterized in that: The method includes: S1 collects relevant characteristic quantity factors that affect the reactive power of the power system, including: bus voltage, line flow diagram, voltage margin, active margin, reactive margin, transformer load rate, node voltage offset, system overload degree, system load rate and node voltage change rate index. The above characteristic quantity factors are used to characterize the state of fragmented energy in the power system; S2 classifies the data corresponding to the relevant feature quantity factors according to the time of their collection to obtain a corresponding data set, wherein the data set includes multiple modal forms, wherein the modal forms include: text, image, value, pulse or waveform; S3 constructs reactive power data confluence, firstly, the power data acquisition equipment, the substation relay equipment and the server receiving equipment in the power system are all regarded as nodes of a power data confluence, and the data set is used as the input node of the data confluence, mapped to the corresponding power data acquisition equipment, the substation relay equipment or the server receiving equipment, to form multiple nodes; S4 performs multimodal processing on all nodes to determine the direction between adjacent nodes, thereby obtaining an energy flow diagram of reactive information in reactive power; a graph convolutional neural network is used to obtain the fusion of the energy flow diagram of reactive information in reactive power and the data energy flow at each node; After obtaining the corresponding graph of the data energy flow, S5 can obtain the amount of fragmented reactive power available for each node and evaluate the amount of fragmented reactive power available for each node; In step S4, multi-modal processing is performed on all nodes to determine the direction between adjacent nodes, thereby obtaining an energy flow diagram of reactive information in reactive power, including: Step 1: Classify the server according to its functions in the power system, including: power data acquisition equipment A, area relay equipment B and server receiving equipment C; Step 2: Pack the collected data sets at a certain time or time period into corresponding data packets, spatially locate the data packets according to the location of the corresponding server, and number the data packets and mark their location information; Step 3: Input the corresponding data packets at the same moment or a certain time period into the power data acquisition device A, and transmit them in the substation relay device B and the server receiving device C, thereby forming multiple nodes of the energy flow diagram; Step 4: Align different modal data from asynchronous distribution according to the time sequence to obtain the edge vector corresponding to the energy flow graph, and then construct a directional energy flow graph; Among them, the multimodal processing includes the following steps: The multimodal dataset and node data transmitted in the node are converted into feature space using linear transformation, which is expressed as: in, is the transformation coefficient, which is determined by the server type, node status, multi-modal type and actual project status of the data. a Represents the node data value corresponding to the power data acquisition device A, b is the node data value corresponding to the relay device B in the substation area, c The server receives the node data value corresponding to device C, and They are the data sets before and after linear transformation respectively; Calculate the influence of the surrounding nodes on the modal data of the central node, expressed as: in, is the edge information, indicating the node and The degree of mutual influence during the transmission process, leaky ReLU is the selected activation function, which is a variant of the ReLU activation function, which allows , The value is negative, and a non-zero gradient is added for negative values; When the stable process and the transmission process are j Node and i The degree of mutual influence between nodes; Normalize the data. exp Represents the natural logarithm e , is the side information, N i For the i nodes, so the normalization process is as follows: In step S4, a graph convolutional neural network is used to obtain the fusion of the energy flow diagram of reactive information in reactive power and the data energy flow at each node, including: The edge information is convoluted and encapsulated, and the influence of the node on the next edge is represented at the node, and finally encapsulated as , It is calculated based on the surrounding nodes combined with attention, that is: Where, N is the total number of nodes, N i For the i nodes; So far, we have obtained the data node Z obtained through convolution encapsulation, and the edge information By obtaining these two pieces of information, we can get the energy flow diagram of reactive information in reactive power evaluation.

2. According to claim 1, the method for evaluating the fragmented reactive power response of a power system based on multimodality and graph convolution is characterized in that: In the fourth step, different modal data from asynchronous distribution are aligned according to the time sequence to obtain the edge vector corresponding to the energy flow graph, which specifically includes: If only considering the transmission process of data packets at the same time, then determine the direction between adjacent nodes corresponding to the same data packet, and use the multimodal processing method to calculate the data value of the data packet in the server corresponding to the node, denoted as X , and calculate the data value of the data packet in the server corresponding to the adjacent node in the same way, denoted as Y. If X < Y, the direction is from this node to the adjacent node; otherwise, if X ≥ Y, the direction is from the adjacent node to this node; otherwise, if considering the data packet transmission process over a period of time, it is also necessary to calculate the direction between adjacent nodes corresponding to data packets at different times. The steps are as follows: First, determine whether the servers corresponding to the two nodes are the same. If they are the same, directly compare the time of the two data packets, set the direction according to the time sequence, and the corresponding data packet with the earlier time points to the corresponding data packet with the later time. Otherwise, if the two nodes correspond to different servers, first determine the time and calculate the data value of the data packet after being processed by the corresponding server using the multimodal method. The direction is determined only when the time sequence and data value size mentioned above are met. If only one of them is met or neither is met, the corresponding direction is not set.

3. The method for evaluating fragmented reactive power response of a power system based on multimodality and graph convolution according to claim 1, characterized in that: The method further includes: In multimodal fusion processing, if there are too many nodes linked to the same node, it will cause data bloat and congestion, and it will be impossible to focus on the content containing more information. Graph convolution is needed to remove the edges whose mean is less than a certain threshold. That is, in calculation After that, if If it is less than a certain threshold, the corresponding package Z data can be directly deleted to remove the invalid edges; Based on the vector value of each node and the heat map of the correlation between each node, the node fusion status and the data energy flow at the data node can be known.

4. The method for evaluating fragmented reactive power response of a power system based on multimodality and graph convolution according to claim 1, characterized in that: In step S5, the amount of fragmented reactive power available for use is evaluated, including: after obtaining the energy flow, the amount of fragmented reactive power available for use at each node can be obtained, but the result needs to be evaluated to measure the accuracy of the model, and the root mean square error RMSE indicator is used to measure the accuracy of the model reactive power evaluation, and its expression is as follows: in, represents the fusion result, Indicates the actual situation of the node. Indicates the number of nodes for reactive power evaluation. The smaller the RMSE value, the more accurate the fusion result.

5. A fragmented reactive power response evaluation system for power systems based on multimodality and graph convolution, characterized in that: The system includes: The acquisition module is used to collect relevant characteristic quantity factors that affect the reactive power of the power system, including: bus voltage, line flow diagram, voltage margin, active margin, reactive margin, transformer load rate, node voltage offset, system overload degree, system load rate and node voltage change rate indicators. The above characteristic quantity factors are used to characterize the state of fragmented energy in the power system; A data set construction module is used to classify the data corresponding to the relevant feature quantity factors according to the time of their collection to obtain a corresponding data set, wherein the data set includes multiple modal forms, wherein the modal forms include: text, image, value, pulse or waveform; The node construction module is used to construct reactive power data confluence. First, the power data acquisition equipment, the substation relay equipment and the server receiving equipment in the power system are all regarded as nodes of a power data confluence, and the data set is used as an input node of the data confluence, mapped to the corresponding power data acquisition equipment, the substation relay equipment or the server receiving equipment to form multiple nodes. The direction determination module is used to perform multimodal processing on all nodes and determine the direction between adjacent nodes, so as to obtain the energy flow diagram of reactive information in reactive power; a graph convolutional neural network is used to obtain the fusion of the energy flow diagram of reactive information in reactive power and the data energy flow at each node; An evaluation module, used for obtaining the amount of fragmented reactive power available for each node according to a corresponding graph of the data energy flow, and evaluating the amount of fragmented reactive power available for each node; In the direction determination module, multi-modal processing is performed on all nodes to determine the direction between adjacent nodes, thereby obtaining an energy flow diagram of reactive information in reactive power, including: Step 1: Classify the server according to its functions in the power system, including: power data acquisition equipment A, area relay equipment B and server receiving equipment C; Step 2: Pack the collected data sets at a certain time or time period into corresponding data packets, spatially locate the data packets according to the location of the corresponding server, and number the data packets and mark their location information; Step 3: Input the corresponding data packets at the same moment or a certain time period into the power data acquisition device A, and transmit them in the substation relay device B and the server receiving device C, thereby forming multiple nodes of the energy flow diagram; Step 4: Align different modal data from asynchronous distribution according to the time sequence to obtain the edge vector corresponding to the energy flow graph, and then construct a directional energy flow graph; Among them, the multimodal processing includes the following steps: The multimodal dataset and node data transmitted in the node are converted into feature space using linear transformation, which is expressed as: in, is the transformation coefficient, which is determined by the server type, node status, multi-modal type and actual project status of the data. a Represents the node data value corresponding to the power data acquisition device A, b is the node data value corresponding to the relay device B in the substation area, c The server receives the node data value corresponding to device C, and They are the data sets before and after linear transformation respectively; Calculate the influence of the surrounding nodes on the modal data of the central node, expressed as: in, is the edge information, indicating the node and The transmission process and the degree of mutual influence, leaky ReLU is the selected activation function, which is a variant of the ReLU activation function, which allows , The value is negative, and a non-zero gradient is added for negative values; When the stable process and the transmission process are j Node and i The degree of mutual influence between nodes; Normalize the data. exp Represents the natural logarithm e , is the side information, N i For the i nodes, so the normalization process is as follows: The use of graph convolutional neural network to obtain the fusion of the energy flow diagram of reactive information in reactive power and the data energy flow at each node includes: The edge information is convoluted and encapsulated, and the influence of the node on the next edge is represented at the node, and finally encapsulated as , It is calculated based on the surrounding nodes combined with attention, that is: Where, N is the total number of nodes, N i For the i nodes; At this point, we get the data node Z obtained through convolution encapsulation and the edge information By obtaining these two pieces of information, we can get the energy flow diagram of reactive information in reactive power evaluation.

6. The power system fragmentation reactive power response evaluation system based on multimodality and graph convolution according to claim 5, characterized in that: In the fourth step, different modal data from asynchronous distribution are aligned according to the time sequence to obtain the edge vector corresponding to the energy flow graph, which specifically includes: If only considering the transmission process of data packets at the same moment, then determine the direction between adjacent nodes corresponding to the same data packet, and use the multimodal processing method to calculate the data value of the data packet in the server corresponding to the node, denoted as X , and calculate the data value of the data packet in the server corresponding to the adjacent node in the same way, denoted as Y. If X < Y, the direction is from this node to the adjacent node; otherwise, if X ≥ Y, the direction is from the adjacent node to this node; Otherwise, if the data packet transmission process within a period of time is considered, the directions between adjacent nodes corresponding to data packets at different times need to be calculated. The steps used are: First, determine whether the servers corresponding to the two nodes are the same. If they are the same, directly compare the time of the two data packets, set the direction according to the time sequence, and the corresponding data packet with the earlier time points to the corresponding data packet with the later time. Otherwise, if the two nodes correspond to different servers, first determine the time and calculate the data value of the data packet after being processed by the corresponding server using the multimodal method. The direction is determined only when the time sequence and data value size mentioned above are met. If only one of them is met or neither is met, the corresponding direction is not set.

Images