Multi-element intelligent computing power platform

Abstract

The invention belongs to the technical field of computing power platforms, and discloses a multi-element intelligent computing power platform, which comprises a region division module, a computing power management module, a computing power management module, a power management module and a computing power management module, and is characterized in that the region division module divides a power network into a plurality of regions and sets computing power resources for each node; the data acquisition module acquires basic data and advanced data of the power equipment; the computing power scheduling module calculates a data importance index according to the basic data and determines an edge or center analysis scheme; the edge computing power module executes an edge analysis scheme and performs node analysis on data with high importance; the central computing power module executes the central analysis scheme and performs regional analysis on all data in the region; the state display module displays node states in real time. According to the system, data importance indexes are introduced, dynamic allocation of computing power resources, rapid processing of important data and deep analysis of common data are achieved, potential risks of the system are found in time while the computing power resources are reasonably utilized, and the efficiency and accuracy of power network analysis are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of computing power platforms, and more specifically, to a multi-intelligent computing power platform. Background Art

[0002] With the rapid development of the power system and the continuous improvement of the degree of intelligence, the scale and complexity of the power grid are increasing day by day. In order to ensure the safe and stable operation of the power system, it has become increasingly important to monitor and analyze the operating status of power equipment in real time. However, there are still many deficiencies in the existing power monitoring systems in achieving efficient and accurate status analysis.

[0003] Traditional power monitoring systems often adopt a unified data processing mode, which is difficult to meet the monitoring needs of power grids of different scales and complexities. In terms of the allocation of computing power resources, there is usually a problem of unreasonable computing power allocation in the existing power monitoring systems. Due to the lack of preliminary analysis and classification of data, the system often uniformly uses high-performance or low-performance computing power resources for processing. This leads to two main problems: on the one hand, for some simple data processing tasks, the system may call high-performance computing resources, resulting in waste of resources and increased energy consumption; on the other hand, when facing complex system-level problems, the system may only use limited computing resources, resulting in insufficient in-depth and comprehensive analysis and inability to detect potential system risks in a timely manner. Summary of the Invention

[0004] In order to overcome the problems of resource waste and inability to detect potential risks in a timely manner caused by the existing technology, the present invention proposes a multi-intelligent computing power platform to solve the above problems.

[0005] The present invention provides the following technical solutions:

[0006] A multi-intelligent computing power platform, comprising:

[0007] A regional division module, configured to divide the power grid into multiple regions according to geographical locations, obtain power equipment within the regions as nodes within the regions, and set corresponding computing power resources for each node;

[0008] A data acquisition module, configured to acquire power data of power equipment and label the nodes corresponding to the power data, where the power data includes basic data and advanced data;

[0009] A computing power scheduling module, configured to obtain a data importance index according to the basic data in the acquired power data; and determine an analysis scheme according to the importance index, where the analysis scheme includes an edge analysis scheme and a central analysis scheme;

[0010] An edge computing power module for executing an edge analysis solution, which includes calling corresponding computing resources according to the nodes corresponding to power data to perform node analysis on the power data to obtain corresponding node states;

[0011] A central computing power module for executing a central analysis solution, which includes obtaining all power data within a region and calling central computing power to perform regional analysis on all power data within the region to obtain all node states within the region;

[0012] A status display module for real-time displaying the node status of each node.

[0013] Preferably, the basic data includes voltage, current, power, and frequency, and the advanced data includes voltage amplitude, voltage phase angle, and active power.

[0014] Preferably, the obtaining of the data importance index according to the basic data in the collected power data includes:

[0015] Obtaining power data and calculating the importance index using the following formula:

[0016]

[0017] In the formula, L represents the stability index, V represents voltage, Vn represents the rated voltage, I represents current, In represents the rated current, P represents power, Pn represents the rated power, f represents frequency, fn represents the rated frequency, and ω1, ω2, ω3, and ω4 respectively represent the weight values of voltage, current, power, and frequency;

[0018] The determining of the analysis solution according to the importance index includes:

[0019] Presetting an importance threshold. When the importance index is less than or equal to the importance threshold, the determined analysis solution is an edge analysis solution; when the importance index is greater than the importance threshold, the determined analysis solution is a central analysis solution.

[0020] Preferably, the steps of the edge analysis solution include:

[0021] Receiving power data from the data acquisition module and confirming the node corresponding to the data; calling the edge computing resources corresponding to the node; preprocessing the power data, including denoising and standardization;

[0022] Calculating key indicators, which include voltage deviation rate, power deviation rate, and frequency deviation rate;

[0023] Among them, the voltage deviation rate, power deviation rate, and frequency deviation rate are all calculated by dividing the corresponding collected data by the rated value;

[0024] For each key metric, calculate the moving average over the past T hours, where T = 324, and calculate the percentage difference between the current value and the moving average; if the percentage difference exceeds a preset threshold, mark it as an abnormal trend. Set a safety threshold for each key metric;

[0025] If any key metric exceeds the safety threshold, determine that the node status is abnormal;

[0026] If there is an abnormal trend mark but it does not exceed the safety threshold, determine that the node status is a warning;

[0027] If all metrics do not exceed the safety threshold and there is no abnormal trend mark, determine that the node status is normal.

[0028] Preferably, the steps of the central analysis scheme include:

[0029] Obtain the physical connection relationship of power equipment in the area and construct a graph structure;

[0030] Obtain all power data in the area and map the power data into the graph structure;

[0031] Construct input features based on the graph structure after mapping the power data;

[0032] Input the constructed input features into a pre-trained node status recognition network to obtain the node status output by the node status recognition network.

[0033] Preferably, the steps of constructing the graph structure include:

[0034] Construct a graph structure G, where G=(V, E), where V is the set of nodes representing power equipment, and E is the set of edges representing transmission lines;

[0035] The steps of obtaining all power data in the area and mapping the power data into the graph structure include:

[0036] Create a node feature vector for each node in the node set V

[0037] where Vm represents the voltage amplitude, θ represents the voltage phase angle, and Ps represents the active power;

[0038] Create an edge feature vector for each edge in the edge set E

[0039] where v1 represents the current flowing into the node, v2 represents the current flowing out of the node, V_v1 represents the voltage of the node where the current flows in, V_v2 represents the voltage of the node where the current flows out, and Q represents the weight value of the edge;

[0040] The steps for obtaining the weight value of the edge include:

[0041] Obtain the parameters of the power transmission line corresponding to the edge in the graph structure, where the parameters include line length, rated capacity, and impedance value;

[0042] After normalizing the parameters of the power transmission line using the maximum-minimum normalization method, perform weighted summation to obtain the weight value of the edge;

[0043] Map the power data to the corresponding edge feature vectors and node feature vectors.

[0044] Preferably, the step of constructing the input features according to the graph structure after mapping the power data includes constructing a node feature matrix, where each row in the node feature matrix corresponds to a node feature vector, and constructing an edge feature matrix, where each row in the edge feature matrix corresponds to an edge feature vector.

[0045] Preferably, the pre-trained node status recognition network includes:

[0046] Construct a graph neural network including a graph convolutional layer, a fully connected layer, and an output layer;

[0047] Collect historical power data, mark the node status corresponding to the power data, obtain the graph structure corresponding to the historical power data, and obtain the input features according to the historical power data and the graph structure;

[0048] Use the mini-batch stochastic gradient descent method to optimize the graph neural network; adopt a learning rate decay strategy to gradually reduce the learning rate as the number of training rounds increases; use early stopping to prevent overfitting and stop training when the performance of the validation set no longer improves;

[0049] Calculate the score distributions of normal state nodes, warning state nodes, and abnormal state nodes on the validation set; set appropriate abnormal thresholds according to the distribution;

[0050] Use the graph neural network with the set abnormal threshold as the trained node status recognition network.

[0051] The present invention provides a multi-intelligent computing power platform, which has the following beneficial effects:

[0052] By introducing the calculation of the data importance index, the system can perform preliminary analysis and classification on the collected power data. This method avoids the problem of uniformly using high-performance or low-performance computing resources, and realizes the dynamic allocation of computing resources. For data with a higher importance index, the system adopts an edge analysis solution and uses local computing resources for rapid processing, avoiding the waste of overusing high-performance resources for simple tasks. For data with a lower importance index, the system calls the central analysis solution and makes full use of high-performance computing resources for in-depth analysis, ensuring that complex system-level problems receive sufficient attention. Thus, potential risks of the system can be detected in a timely manner on the premise of reasonably utilizing computing resources. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 It is a schematic diagram of the modules of a multi-intelligent computing power platform of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0054] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0055] Embodiment 1

[0056] Please refer to Figure 1 , in this embodiment, a multi-intelligent computing power platform includes:

[0057] A regional division module, configured to divide the power grid into multiple regions according to geographical locations, obtain the power equipment within the regions as nodes within the regions, and set corresponding computing resources for each node;

[0058] In this embodiment, the operation of the regional division module can be carried out according to the following steps: First, obtain the geographical distribution information of the power grid, including the position coordinates of each power equipment. Then, divide the power grid into multiple regions according to the preset geographical division criteria (such as administrative divisions). Next, for each divided region, identify and list all the power equipment within the region, and use these equipment as the nodes within the region. Finally, set corresponding levels of computing resources for each node. It should be noted that the computing resources include: hardware devices such as edge computing servers, industrial-grade computers, and embedded systems, as well as corresponding computing capabilities (such as the number of CPU cores, main frequency), memory capacity, storage space, etc.

[0059] A data acquisition module, configured to acquire the power data of the power equipment and label the nodes corresponding to the power data, where the power data includes basic data and advanced data;

[0060] The basic data includes voltage, current, power, and frequency, and the advanced data includes voltage amplitude, voltage phase angle, and active power.

[0061] In this embodiment, the operation of the data acquisition module can be carried out according to the following steps: First, install data acquisition devices, such as smart meters or sensors, on each power device. These devices are used to collect power data in real time. Then, set the data acquisition frequency, for example, collect data once a minute. Next, the acquisition devices start to collect power data according to the set frequency, including basic data (voltage, current, power, and frequency) and advanced data (voltage amplitude, voltage phase angle, and active power). The collected data will be marked with the unique identifier of the corresponding node so that the data source can be identified during subsequent processing.

[0062] The computing power scheduling module is used to obtain the data importance index based on the basic data in the collected power data; and determine the analysis scheme according to the importance index, where the analysis scheme includes an edge analysis scheme and a central analysis scheme.

[0063] The obtaining of the data importance index based on the basic data in the collected power data includes:

[0064] Obtain the power data and calculate the importance index using the following formula:

[0065]

[0066] In the formula, L represents the stability index, V represents voltage, Vn represents the rated voltage, I represents current, In represents the rated current, P represents power, Pn represents the rated power, f represents frequency, fn represents the rated frequency, and ω1, ω2, ω3, and ω4 respectively represent the weight values of voltage, current, power, and frequency.

[0067] The determining of the analysis scheme according to the importance index includes:

[0068] Preset an importance threshold. When the importance index is less than or equal to the importance threshold, determine the analysis scheme as the edge analysis scheme; when the importance index is greater than the importance threshold, determine the analysis scheme as the central analysis scheme.

[0069] In this embodiment, the operation of the computing power scheduling module can be carried out according to the following steps: First, obtain the basic data in the power data from the data acquisition module. Then, calculate the data importance index using a given formula, where the values of each parameter can be directly obtained from the collected data, and the weight values can be preset according to the actual situation. It should be noted that the principle of the formula design is that the higher the importance index, the closer the data is to the rated value of the device, indicating that the device is in a more stable operating state and the potential risk is lower. On the contrary, the lower the importance index, the more the data deviates from the rated value, and there may be potential risks. Next, set an importance threshold. For example, the threshold can be set to 0.8. Finally, compare the calculated importance index with the preset threshold. If the importance index is less than or equal to 0.8, determine to use the central analysis scheme for more detailed analysis; if the importance index is greater than 0.8, determine to use the edge analysis scheme, and the analysis can be completed through edge computing. This design can reasonably utilize computing power resources while ensuring the analysis quality and improve the overall efficiency.

[0070] The edge computing power module is used to execute the edge analysis scheme, and the edge analysis scheme includes calling the corresponding computing power resources according to the node corresponding to the power data to perform node analysis on the power data to obtain the corresponding node status;

[0071] The steps of the edge analysis scheme include:

[0072] Receive the power data from the data acquisition module and confirm the node corresponding to the data; call the edge computing resources corresponding to the node; preprocess the power data, including denoising and standardization;

[0073] Calculate the key indicators, and the key indicators include voltage deviation rate, power deviation rate, and frequency deviation rate;

[0074] Among them, the voltage deviation rate, power deviation rate, and frequency deviation rate are all calculated by dividing the corresponding collected data by the rated value;

[0075] For each key indicator, calculate the moving average value in the past T hours, T = 324, calculate the percentage difference between the current value and the moving average value; if the percentage difference exceeds the preset threshold, mark it as an abnormal trend. Set a safety threshold for each key indicator;

[0076] If any key indicator exceeds the safety threshold, determine that the node status is abnormal;

[0077] If there is an abnormal trend mark but does not exceed the safety threshold, determine that the node status is a warning;

[0078] If all indicators do not exceed the safety threshold and there is no abnormal trend mark, determine that the node status is normal.

[0079] In this embodiment, the operation of the edge analysis solution can be carried out according to the following steps:

[0080] First, the edge computing device receives power data from the data acquisition module and confirms the node corresponding to the data. Then, it calls the edge computing resources corresponding to this node, which are usually embedded devices or industrial computers with relatively low computing power. Next, simple preprocessing is performed on the power data, including removing obvious noise points and data standardization.

[0081] Subsequently, key indicators are calculated, including voltage deviation rate, power deviation rate, and frequency deviation rate. These indicators are obtained by dividing the collected data by the corresponding rated value. For each key indicator, the moving average of the past hour is calculated, and the percentage difference between the current value and the moving average is calculated. If the percentage difference exceeds a preset threshold (such as 5%), it is marked as an abnormal trend.

[0082] Finally, based on the preset safety threshold and abnormal trend markings, the node status is quickly judged. If any key indicator exceeds the safety threshold, it is determined as abnormal; if there is an abnormal trend marking but does not exceed the safety threshold, it is determined as a warning; if all indicators do not exceed the safety threshold and there is no abnormal trend marking, it is determined as normal. This method can quickly process data with relatively low risks and give preliminary judgment results in a timely manner to meet the requirements of real-time monitoring.

[0083] The central computing power module is used to execute the central analysis solution. The central analysis solution includes obtaining all power data within the region, calling the central computing power to perform regional analysis on all power data within the region, and obtaining the status of all nodes within the region;

[0084] The steps of the central analysis solution include:

[0085] Obtain the physical connection relationship of power equipment within the region and construct a graph structure;

[0086] Obtain all power data within the region and map the power data into the graph structure;

[0087] Construct input features according to the graph structure after mapping the power data;

[0088] Input the constructed input features into a pre-trained node status recognition network to obtain the node status output by the node status recognition network.

[0089] The steps of constructing the graph structure include:

[0090] Construct a graph structure G, where G = (V, E), where V is the set of nodes representing power equipment, and E is the set of edges representing transmission lines;

[0091] The steps of obtaining all power data within the acquisition area and mapping the power data to the graph structure include:

[0092] Create a node feature vector for each node in the node set V

[0093] where Vm represents the voltage amplitude, θ represents the voltage phase angle, and Ps represents the active power;

[0094] Create an edge feature vector for each edge in the edge set E

[0095] where v1 represents the node where the current flows in, v2 represents the node where the current flows out, V_v1 represents the voltage of the node where the current flows in, V_v2 represents the voltage of the node where the current flows out, and Q represents the weight value of the edge;

[0096] The steps of obtaining the weight value of the edge include:

[0097] Obtain the parameters of the transmission line corresponding to the edge in the graph structure, and the parameters include line length, rated capacity, and impedance value;

[0098] After normalizing the parameters of the transmission line using the maximum-minimum normalization method, perform weighted summation to obtain the weight value of the edge;

[0099] Map the power data to the corresponding edge feature vector and node feature vector.

[0100] The steps of constructing the input features according to the graph structure after mapping the power data include constructing a node feature matrix, where each row in the node feature matrix corresponds to a node feature vector, and constructing an edge feature matrix, where each row in the edge feature matrix corresponds to an edge feature vector.

[0101] The pre-trained node status recognition network includes:

[0102] Construct a graph neural network including a graph convolutional layer, a fully connected layer, and an output layer;

[0103] Collect historical power data, label the node status corresponding to the power data, obtain the graph structure corresponding to the historical power data, and obtain the input features according to the historical power data and the graph structure;

[0104] Use the mini-batch stochastic gradient descent method to optimize the graph neural network; adopt a learning rate decay strategy to gradually reduce the learning rate as the number of training rounds increases; use the early stopping method to prevent overfitting and stop training when the performance of the validation set no longer improves;

[0105] Calculate the score distributions of normal state nodes, warning state nodes, and abnormal state nodes on the validation set; set appropriate abnormal thresholds according to the distribution;

[0106] Use the graph neural network with the set abnormal threshold as the trained node state recognition network.

[0107] In this embodiment, the operation of the central computing power module to execute the central analysis scheme can be carried out according to the following steps:

[0108] First, obtain the physical connection relationships of all power devices in the area and construct a graph structure. This graph structure reflects the topological relationship of the power network, where nodes represent power devices and edges represent transmission lines.

[0109] Then, obtain all the power data in the area and map it into the graph structure. For each node, create a feature vector containing voltage amplitude, voltage phase angle, and active power. For each edge, create a feature vector containing current inflow and outflow information, voltage information, and weight value.

[0110] The calculation process of the weight value of the edge is more detailed: First, obtain the parameters of the transmission line, including line length, rated capacity, and impedance value. Then use the maximum-minimum normalization method to normalize these parameters so that parameters with different dimensions can be compared. Finally, perform weighted summation on the normalized parameters to obtain the weight value of the edge. This weight value reflects the importance and characteristics of the transmission line.

[0111] Next, construct input features according to the graph structure mapped with power data. Specifically, construct a node feature matrix, where each row corresponds to the feature vector of a node; at the same time, construct an edge feature matrix, and each row corresponds to the feature vector of an edge. These two matrices together constitute the input features describing the state of the entire power network.

[0112] Finally, input the constructed input features into the pre-trained node state recognition network. This network is a graph neural network that has been carefully designed and trained, including graph convolutional layers, fully connected layers, and output layers. It can make full use of the topological structure of the power network and node-edge features for deep learning, so as to identify the state of each node.

[0113] Utilize the powerful computing power of the central server to conduct a comprehensive and in-depth analysis of the power network in the entire area. Considering the complex mutual influences between devices, it can more accurately identify potential abnormal states, and is especially suitable for dealing with complex system-level problems and identifying potential systemic risks.

[0114] A state display module for real-time displaying the node state of each node.

[0115] In this embodiment, the status information of each node is obtained from the edge computing power module and the central computing power module. Then, this information is updated in real time on the display interface. Specific display methods may include: using the power grid topology map of the geographic information system, and representing the status of each node with icons of different colors on the map. For example, green indicates normal, yellow indicates warning, and red indicates abnormal; using a data dashboard to display the statistical information of the status of various types of nodes in the form of charts, such as a pie chart showing the proportions of normal, warning, and abnormal nodes, etc.

[0116] In several embodiments provided by the present invention, it should be understood that the disclosed systems, devices, and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of the units is only one way, and there may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point, the displayed or discussed coupling or direct coupling or communication connection between each other can be through some interfaces, and the indirect coupling or communication connection of the devices or units can be in an electrical, mechanical, or other form.

[0117] As mentioned above, the above is only the specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention can easily think of changes or substitutions, which should all be covered by the protection scope of the present invention.

[0118] Finally: The above is only the preferred embodiment of the present invention and is not used to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A multi-intelligence computing power platform, characterized in that, Including: A region division module, which is used to divide the power grid into multiple regions according to geographical locations, obtain power equipment within the regions as nodes within the regions, and set corresponding computing power resources for each node; A data acquisition module, which is used to collect power data of power equipment and label the nodes corresponding to the power data, and the power data includes basic data and advanced data; A computing power scheduling module, which is used to obtain a data importance index according to the basic data in the collected power data; Determine an analysis scheme according to the importance index, and the analysis scheme includes an edge analysis scheme and a central analysis scheme; An edge computing power module, which is used to execute the edge analysis scheme. The edge analysis scheme includes calling the corresponding computing power resources according to the nodes corresponding to the power data to perform node analysis on the power data to obtain the corresponding node status; A central computing power module, which is used to execute the central analysis scheme. The central analysis scheme includes obtaining all power data within the region and calling the central computing power to perform region analysis on all power data within the region to obtain the statuses of all nodes within the region; A status display module, which is used to display the node status of each node in real time.

2. The multi-intelligent computing power platform according to claim 1, wherein The basic data includes voltage, current, power, and frequency, and the advanced data includes voltage amplitude, voltage phase angle, and active power.

3. The multi-intelligent computing power platform according to claim 2, characterized in that, The obtaining the data importance index according to the basic data in the collected power data includes: Obtain the power data and calculate the importance index using the following formula: In the formula, L represents the stability index, V represents voltage, Vn represents the rated voltage, I represents current, In represents the rated current, P represents power, Pn represents the rated power, f represents frequency, fn represents the rated frequency, and ω1, ω2, ω3, and ω4 respectively represent the weight values of voltage, current, power, and frequency; The determining the analysis scheme according to the importance index includes: Preset an importance threshold. When the importance index is less than or equal to the importance threshold, determine the analysis scheme as the edge analysis scheme; when the importance index is greater than the importance threshold, determine the analysis scheme as the central analysis scheme.

4. The multi-intelligent computing power platform according to claim 3, wherein The described multi-intelligent computing power platform is characterized in that the steps of the edge analysis scheme include: Receive the power data from the data acquisition module and confirm the nodes corresponding to the data; call the edge computing resources corresponding to the nodes; perform preprocessing on the power data, including denoising and standardization; Calculate key indicators, and the key indicators include voltage deviation rate, power deviation rate, and frequency deviation rate; Among them, the voltage deviation rate, power deviation rate, and frequency deviation rate are all calculated by dividing the corresponding collected data by the rated value; For each key indicator, calculate the moving average value in the past T hours, T>324, calculate the difference percentage between the current value and the moving average value; if the difference percentage exceeds the preset threshold, mark it as an abnormal trend Set a safety threshold for each key indicator; If any key indicator exceeds the safety threshold, determine the node status as abnormal; If there is an abnormal trend mark but does not exceed the safety threshold, determine the node status as a warning; If all indicators do not exceed the safety threshold and there is no abnormal trend mark, determine the node status as normal.

5. A multi-intelligent computing power platform according to claim 1, characterized in that, The steps of the central analysis scheme include: Obtain the physical connection relationship of power equipment in the area and construct a graph structure; Obtain all power data in the area and map the power data into the graph structure; Construct input features according to the graph structure after mapping the power data; Input the constructed input features into a pre-trained node status recognition network to obtain the node status output by the node status recognition network.

6. The multi-intelligence computing power platform according to claim 5, characterized in that, The steps of constructing the graph structure include: Construct a graph structure G, where G = (V, E), where V is the set of nodes representing power equipment, and E is the set of edges representing transmission lines; The steps of obtaining all power data in the area and mapping the power data into the graph structure include: Create a node feature vector for each node in the node set V Among them, Vm represents the voltage amplitude, θ represents the voltage phase angle, and Ps represents the active power; Create an edge feature vector for each edge in the edge set E Among them, v1 represents the current flowing into the node, v2 represents the current flowing out of the node, V_v1 represents the voltage of the current flowing into the node, V_v2 represents the voltage of the current flowing out of the node, and Q represents the weight value of the edge; The steps of obtaining the weight value of the edge include: Obtain the parameters of the transmission line corresponding to the edge in the graph structure, and the parameters include line length, rated capacity, and impedance value; After normalizing the parameters of the transmission line using the maximum-minimum normalization method, perform weighted summation to obtain the weight value of the edge; Map the power data into corresponding edge feature vectors and node feature vectors.

7. The multi-intelligence computing power platform according to claim 6, characterized in that, The steps of constructing input features according to the graph structure after mapping the power data include constructing a node feature matrix, where each row in the node feature matrix corresponds to a node feature vector, and constructing an edge feature matrix, where each row in the edge feature matrix corresponds to an edge feature vector.

8. The multi-intelligent computing power platform according to claim 7, wherein, The pre-trained node status recognition network includes: Construct a graph neural network including a graph convolutional layer, a fully connected layer, and an output layer; Collect historical power data, mark the node status corresponding to the power data, obtain the graph structure corresponding to the historical power data, and obtain input features according to the historical power data and the graph structure; Use the mini-batch stochastic gradient descent method to optimize the graph neural network; adopt a learning rate decay strategy to gradually reduce the learning rate as the number of training rounds increases; use early stopping to prevent overfitting and stop training when the performance of the validation set no longer improves; Calculate the score distributions of normal state nodes, warning state nodes, and abnormal state nodes on the validation set; according to the distribution, set appropriate anomaly thresholds; Take the graph neural network with the set anomaly threshold as the trained node status recognition network.

Images