Power station remote simulation method and device, electronic equipment and storage medium

Abstract

The invention provides a power station remote simulation method and device, electronic equipment and a storage medium, and belongs to the technical field of power station simulation, and the method comprises the steps: obtaining multi-source data of the equipment according to a locally deployed edge node of a power station; determining graph structure data according to the multi-source data, inputting the graph structure data into a pre-trained graph neural network, and outputting a coding vector; determining an operation scene of the power station according to the coding vector, and determining a configuration parameter corresponding to the operation scene according to the operation scene; constructing a double-layer digital twinborn model according to the physical entity of the power station and each device of the power station; and obtaining current operation parameters respectively corresponding to each device, and determining an operation state and a device health state respectively corresponding to each device of the power station according to the double-layer digital twin model, the configuration parameters and the operation parameters. According to the power station remote simulation method and device, the electronic equipment and the storage medium provided by the invention, more accurate and more comprehensive remote simulation of the power station can be realized.

Description

Technical Field

[0001] This application belongs to the field of power plant simulation technology, and more specifically, relates to a remote simulation method and device for power plants, electronic equipment, and storage medium. Background Technology

[0002] Remote simulation technology plays a crucial role in the field of industrial intelligence, serving as a key supporting means for achieving remote equipment monitoring, fault prediction, and optimized operation. For power plants, remote simulation enables a comprehensive and in-depth simulation and analysis of the plant's operating status without interfering with its normal operation, thus providing a scientific basis for the safe, efficient, and stable operation of the power plant.

[0003] In existing technologies, remote simulation of power plants often involves directly collecting various data from the power plant equipment and then inputting this data into traditional neural network models or simple mathematical models to simulate the equipment's operating status. However, this approach struggles to fully uncover the inherent relationships between data points when dealing with complex and multi-source power plant data. This leads to an inability to accurately determine the operating scenario, which in turn affects the accuracy of subsequent assessments of equipment operating status and health. Summary of the Invention

[0004] The purpose of this application is to provide a method, device, electronic equipment, and storage medium for remote simulation of power plants, so as to accurately determine the operating scenario and improve the accuracy of the assessment of the operating status and health status of equipment from both macroscopic and microscopic levels, thereby achieving more accurate and comprehensive remote simulation of power plants.

[0005] In a first aspect, this embodiment provides a method for remote simulation of a power plant, comprising: The power station acquires multi-source data of the equipment based on the edge nodes deployed locally. The power station has multiple edge nodes deployed locally, and each edge node represents one device. The multi-source data of the device includes the multi-source data corresponding to each device, including device vibration spectrum data, discharge signals, and environmental data. The graph structure data is determined based on multi-source data. The graph structure data is then input into a pre-trained graph neural network, which outputs an encoding vector. The encoding vector is used to reflect the relevant characteristics of the operation of each piece of equipment in the power plant. The operating scenario of the power plant is determined based on the encoding vector, and the configuration parameters corresponding to the operating scenario are determined based on the operating scenario. A two-layer digital twin model is constructed based on the physical entity of the power station and its various equipment. The two-layer digital twin model includes a macroscopic layer and a microscopic layer. Obtain the current operating parameters of each device, and determine the operating status and health status of each device in the power plant based on the two-layer digital twin model, configuration parameters, and current operating parameters of each device.

[0006] A second aspect of this embodiment provides a power plant remote simulation device, comprising: The data acquisition module is used to acquire multi-source data of the equipment based on the edge nodes deployed locally in the power plant. There are multiple edge nodes deployed locally in the power plant, and each edge node is a device. The multi-source data of the equipment includes the multi-source data corresponding to each device, including equipment vibration spectrum data, discharge signals and environmental data. The encoding vector determination module is used to determine graph structure data based on multi-source data. It inputs the graph structure data into a pre-trained graph neural network and outputs an encoding vector. The encoding vector is used to reflect the relevant characteristics of the operation of each piece of equipment in the power plant. The configuration parameter determination module is used to determine the operating scenario of the power plant based on the encoding vector, and to determine the configuration parameters corresponding to the operating scenario based on the operating scenario. The two-layer digital twin model determination module is used to construct a two-layer digital twin model based on the physical entity of the power station and its various equipment. The two-layer digital twin model includes a macroscopic layer and a microscopic layer. The simulation evaluation module is used to obtain the current operating parameters of each device. Based on the two-layer digital twin model, configuration parameters and the current operating parameters of each device, the module determines the operating status and health status of each device in the power plant.

[0007] In a third aspect of this embodiment, an electronic device is provided, including a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor executes the computer program to implement the steps of the above-described power plant remote simulation method.

[0008] In a fourth aspect of this embodiment, a computer-readable storage medium is provided, which stores a computer program that, when executed by a processor, implements the steps of the above-described power plant remote simulation method.

[0009] The beneficial effects of the power plant remote simulation method and device, electronic equipment, and readable storage medium provided in this embodiment are as follows: This embodiment acquires multi-source data from the device through edge nodes deployed locally at the power plant, then determines graph structure data based on the multi-source data, and inputs the graph structure data into a pre-trained graph neural network model. After the graph neural network model processes the input graph structure data, it fully explores the inherent relationships between each node and each edge in the graph structure data, thereby accurately judging the operating scenario of the power plant; This embodiment constructs a two-layer digital twin model containing macro and micro layers, thereby simulating the power plant from macro to micro levels with high precision, and more accurately reflecting the actual operating conditions of the power plant. Attached Figure Description

[0010] To more clearly illustrate the technical solutions in this embodiment, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0011] Figure 1 A flowchart illustrating a remote power plant simulation method provided in an embodiment of this application; Figure 2 A structural block diagram of a power plant remote simulation device provided in an embodiment of this application; Figure 3 This is a schematic block diagram of an electronic device provided in an embodiment of this application. Detailed Implementation

[0012] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments. However, those skilled in the art will understand that the present application may be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of the present application with unnecessary detail.

[0013] It is understood that in this embodiment, data related to user information is involved. When this embodiment is applied to specific products or technologies, user permission or consent is required, and the collection, use and processing of related data must comply with relevant laws, regulations and standards.

[0014] It should be noted that the terms "first," "second," etc., used in the specification, claims, and drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such use of data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in sequences other than those illustrated or described herein.

[0015] To make the objectives, technical solutions, and advantages of this application clearer, the following description will be provided in conjunction with the accompanying drawings and specific embodiments.

[0016] Please refer to Figure 1 , Figure 1 This is a flowchart illustrating a power plant remote simulation method according to an embodiment of this application. The method may include: S101: Obtain multi-source data from the device based on the edge nodes deployed locally at the power station.

[0017] In this embodiment, the power station locally deploys multiple edge nodes, each edge node being a device. The multi-source data for each device includes multi-source data corresponding to each device, including device vibration spectrum data, discharge signals, and environmental data. The device vibration spectrum data is obtained by converting time-domain vibration signals collected during device operation due to vibrations caused by component movement, friction, or imbalance into frequency-domain signals. The discharge signals can be electrical, optical, or thermal signals generated by high-voltage electrical equipment during partial discharge. Environmental data can include temperature, humidity, and air pressure. This embodiment deploys edge nodes on various devices within the power station, or at a predetermined distance from each device, to reduce signal attenuation and avoid areas with strong electromagnetic interference, such as near high-voltage busbars.

[0018] In this embodiment, the edge nodes can be deployed according to the physical areas of the power plant, such as the plant buildings, generating units, and equipment clusters. Specifically, one edge node is deployed for each piece of equipment in each physical area of ​​the power plant. For example, for the generating unit area of ​​the power plant, one edge node is deployed for generator unit 1 in that area, and another edge node is deployed for generator unit 2 in that area. The edge node corresponding to each generator unit can collect the vibration spectrum data, discharge signals, and environmental data of the corresponding generator unit.

[0019] In this embodiment, the deployment of edge nodes can also be based on the relationships between various devices within the power plant. Specifically, for functionally related groups of devices within the power plant, edge nodes are bound to core devices for deployment. For example, if a transformer within the power plant is considered as a functional unit, an edge node is deployed for that transformer to collect its vibration spectrum data, discharge signals, and environmental data; if a transmission line within the power plant is considered as another functional unit, an edge node is deployed for that functional unit.

[0020] In this embodiment, the multi-source data acquisition method is as follows: Within a preset acquisition period, vibration spectrum data of each device is acquired using an accelerometer, and high-frequency current signals generated by partial discharge within each device are detected using a high-frequency current sensor. The preset acquisition period is once every 2 minutes. The preset acquisition period for environmental data is once every 5 minutes.

[0021] S102: Determine the graph structure data based on the multi-source data, input the graph structure data into the pre-trained graph neural network, and output the encoding vector.

[0022] In this embodiment, step S102, "determining graph structure data based on multi-source data," specifically includes: Determine the corresponding feature vectors for each of the multiple data sources; Each piece of equipment in the power plant is defined as a node in a graph structure, and the physical or functional connections between the nodes are defined as edges in the graph structure. The weights corresponding to each edge of the graph structure are obtained based on the self-attention mechanism; The adjacency matrix is ​​determined based on the edges and corresponding weights of the graph structure. The adjacency matrix is ​​used to represent the connection relationships and connection attributes between nodes. Graph structure data is determined based on the eigenvectors and adjacency matrices corresponding to each of the multiple data sources.

[0023] In one possible embodiment, the specific method for determining the corresponding feature vectors based on multi-source data is as follows: The multi-source data is preprocessed by cleaning and outlier handling of the equipment vibration spectrum data, discharge signal data, and environmental data. For the processed equipment vibration spectrum data, feature frequencies are extracted using Fourier transform, converting the high-dimensional spectrum into a fixed-dimensional vibration spectrum vector. In this embodiment, the vibration spectrum vector can be a 128-dimensional vector. For the processed discharge signal data, statistical characteristics (e.g., pulse amplitude, number of discharges, and average discharge amount) and waveform characteristics (e.g., pulse rise time and fall time) are obtained. These characteristics are then integrated into a fixed-dimensional discharge signal vector. In this embodiment, the discharge signal vector can be a 64-dimensional vector. For the processed environmental data, the temperature, humidity, and air pressure values ​​are standardized to the [0, 1] interval, forming a low-dimensional 3D environmental vector.

[0024] In this embodiment, after obtaining the feature vectors corresponding to each multi-source data, the three types of data are associated and aligned based on a combination of timestamps and device IDs. This ensures that the vibration spectrum vector and discharge signal vector of the same device at the same time point can be mapped to the same device; and that the environmental vector of the same monitoring point at the same time point can be mapped to an environmental node and form an association mapping with the devices covered by that monitoring point, avoiding environmental data without corresponding devices or device data without corresponding environments. The feature vectors corresponding to the aligned multi-source data are then concatenated to obtain the concatenated feature vector of that node.

[0025] In this embodiment, each piece of equipment within the power plant is defined as a node in a graph structure. For example, generators, transformers, circuit breakers, and transmission lines can each be defined as a node in the graph structure. Based on the physical wiring diagrams of each piece of equipment in the power plant and the functional relationships between them, the physical connection edges and functional dependency edges between the equipment are determined. Physical connection edges refer to edges where there is a direct physical pipeline or line connection between the equipment, such as the pipeline connection between the feedwater pump and the economizer. Functional dependency edges refer to edges where there is no direct physical connection between the equipment, but there is a functional dependency or influence relationship, such as the excitation system and the generator (the output of the excitation system affects the voltage of the generator).

[0026] In this embodiment, each edge can be represented as: u → v ,in, u As the source node, vThe target node is defined as follows: The input vector for the self-attention mechanism is determined for each edge. This input vector includes the state of the source node, the state of the target node, and the connection type. For example, for an edge consisting of a transformer and a circuit breaker in a power plant, since electricity is transmitted from the transformer to the circuit breaker, the source node of this edge can be the transformer, and the target node can be the circuit breaker.

[0027] In this embodiment, the learnable parameters of the self-attention mechanism are the query matrix and the key matrix. The input vector is mapped to the query vector and key vector respectively using these learnable parameters. In this embodiment, the query vector and key vector can be obtained through mapping the input vector. Based on the query vector and key vector, the first attention score for that edge is determined according to the first attention score calculation formula. The first attention score calculation formula is as follows: ,in, s The first attention score, q For query vector, k For key vectors, T For transpose operation, The dimension of the key vector is determined based on the actual task requirements. Different task requirements can adjust the dimension of the key vector according to the actual scenario.

[0028] The attention scores corresponding to all edges from the same source node are normalized. In this embodiment, the softmax function can be used to normalize the attention scores to obtain the weights of the corresponding edges. The sum of the weights of all edges from the same source node is 1. The specific method is as follows: If a source node u have m Edges, determining the target nodes respectively The corresponding attention scores are respectively For each attention score, take the exponent to determine the exponential score; sum the exponential scores of all edges of the source node to obtain the denominator normalized by the softmax function; for each edge, divide its exponential score by the sum of the exponential scores of all edges to obtain the weight of that edge. The edge weight is calculated as follows: ,in, Let be the weight of the edge formed by the source node and the i-th target node. Let be the exponential score of the edge formed by the source node and the i-th target node. It is the sum of the exponentialized fractions of all edges.

[0029] This embodiment amplifies the differences in attention scores between different edges by rounding the attention scores to integers, thereby more sensitively capturing the differences in the importance of the association between the source node and different target nodes. This embodiment normalizes according to the softmax function to determine the weights corresponding to each edge, which can intuitively understand the relative importance of each edge, improve the performance of the model in monitoring the status of equipment in the power plant, and provide a basis for the maintenance and optimization of the power plant.

[0030] This embodiment determines the adjacency matrix based on the edges and corresponding weights of the graph structure. This adjacency matrix represents the connection relationships and connection attributes between nodes, where the connection attribute is the edge weight of the edge formed by two nodes. In this embodiment, the power station has N equipment nodes, and the adjacency matrix A is an N×N matrix. The source node is used as the row of the adjacency matrix, and the target node is used as the column. The elements A in this adjacency matrix... ij This represents the edge weight of the edge formed by the i-th source node and the j-th target node. In this embodiment, the elements of the adjacency matrix are determined by the edge weights. If there is an edge (including physical or functional edges) between node i and node j, then element A... ij The element value is the weight of the edge; if there is no physical edge or functional edge between node i and node j, then element A... ij The element value is 0.

[0031] This embodiment can sort the acquired spliced ​​feature vectors of each node according to a preset order to determine the feature matrix X of all nodes in the graph structure, where, ,in, R For the set of real numbers, n The total number of nodes. d The feature dimensions for each node.

[0032] For example, consider a power plant with three devices, each with a dimension of 3. Without prior sorting, the devices can be in any order. The feature vector of the second device is [0.4, 0.5, 0.6], the first device's feature vector is [0.1, 0.2, 0.3], and the third device's feature vector is [0.7, 0.8, 0.9]. If the default order is sorted by device ID (first device, second device, third device), then the sorted feature vectors are: first device [0.1, 0.2, 0.3], second device [0.4, 0.5, 0.6], and third device [0.7, 0.8, 0.9]. Stacking these sorted feature vectors yields the feature matrix of all nodes in the graph structure. The feature matrix has a total of 3 nodes, and each node has 3 dimensions.

[0033] The adjacency matrix is ​​normalized. In this embodiment, symmetric normalization is used to normalize the adjacency matrix. The symmetric normalization calculation formula is as follows: ,in, The normalized adjacency matrix, D It is a diagonal matrix. ,in, The element in the i-th row and i-th column of the diagonal matrix (the element on the diagonal) is used to characterize the connectivity of node i in the graph. This is to sum all elements in the i-th row of the adjacency matrix.

[0034] Graph convolution is performed on the feature matrices and adjacency matrices of all nodes in the graph structure to obtain the feature representations of all nodes in the graph structure. The graph convolution operation can be: ,in, Let W be the feature representation of all nodes in the graph structure obtained after the graph convolution operation, and let W be the learnable weight matrix.

[0035] For example, if a power plant has three nodes, and the feature vector dimension of each node is two, then the feature matrix... Adjacency matrix This adjacency matrix indicates that node 1 is connected to nodes 2 and 3, node 2 is connected only to node 1, and node 3 is connected only to node 1. (Diagonal matrix) The diagonal matrix indicates that node 1 has a degree of 2, and nodes 2 and 3 both have a degree of 1.

[0036] The normalized adjacency matrix is ​​determined based on the diagonal matrix, the adjacency matrix, and the symmetric normalization formula. The normalized adjacency matrix is ​​as follows: In this embodiment, the learnable weight matrix is ​​set as follows: .

[0037] The feature representations of all nodes in the graph structure are determined based on the normalized adjacency matrix, the learnable weight matrix, and the node feature matrix. These feature representations can be the feature parts of the nodes after information transmission. The graph structure data G is then determined based on the feature representations of each node, edge, and all nodes in the graph.

[0038] The acquired graph structure data G of the current power plant is input into a pre-trained graph neural network model. The graph structure data G starts from the input layer of the graph neural network model and passes through each layer. The node features of the graph structure data are updated based on the execution results of each layer. The propagation stops when the forward propagation reaches a pre-specified intermediate layer. The feature matrix H of the nodes is extracted from this pre-specified intermediate layer. The dimension of feature matrix H is N× ,in, This represents the feature dimension output by the intermediate layer. Encoding vectors are extracted based on this feature matrix H, where each row of the feature matrix H corresponds to the encoding vector of the corresponding node.

[0039] In this embodiment, the training method for the graph neural network can be: Acquire historical operating data of the power plant, including operating data under normal operating conditions and operating data under fault conditions. Based on this historical operating data, determine multiple sets of graph structure data G, each set of graph structure data G corresponding to one time slice (e.g., determine one set of graph structure data every 5 minutes), and label it with the corresponding encoding vector. Divide the graph structure data G into training set and validation set according to a preset ratio (the preset ratio can be 7:3).

[0040] The parameters of the graph neural network model are initialized by inputting the graph structure data G from the training set into the model. The aggregation result of node features is calculated through forward propagation; the model parameters are updated through backpropagation until the loss function converges or the preset number of training iterations is reached. Using the graph structure data G from the training set, the accuracy of the model in tasks such as state recognition and fault detection is evaluated using a validation set to ensure that the model can effectively learn the interaction patterns of power plant equipment.

[0041] Before obtaining the weights corresponding to each edge of the graph structure based on the self-attention mechanism, this embodiment also includes: For each edge, determine the initial attention coefficient for that edge; The first node of the edge is taken as the current node, and the second node of the edge is taken as the neighboring node. Based on the initial attention coefficient of the edge, the information of the neighboring nodes is fused into the current node to obtain the updated feature representation of the first node. The second node of the edge is taken as the current node, and the first node of the edge is taken as the neighboring node. Based on the initial attention coefficient of the edge, the information of the neighboring nodes is fused into the current node to obtain the updated feature representation of the second node. Based on the updated first node feature representation and the updated second node feature representation, determine the edge feature corresponding to the edge; The weights corresponding to each edge of the graph structure are obtained based on the self-attention mechanism, including: The attention coefficients of each edge are determined based on the self-attention mechanism and edge characteristics; The attention coefficients of each edge are normalized to obtain the weights of each edge.

[0042] In this embodiment, an initial attention coefficient is assigned to each edge in the graph structure based on experience or preset rules, or the initial attention coefficient is determined based on a prior matrix. This initial attention coefficient is used to initially measure the importance of the edge in device interaction.

[0043] Taking the first node of the edge as the current node and the second node of the edge as the neighboring node, we obtain the feature vectors of the current node and the neighboring node respectively, where the first node is the source node and the second node is the target node. Based on the initial attention coefficients, we then apply these features to the neighboring nodes (…). The feature vectors of are weighted and then compared with the current node ( The feature vectors of the first node are fused to obtain the updated feature representation of the first node. The fusion formula can be: ,in, This is the updated feature representation of the first node. As the source node, Let be the target node, and s be the first attention score.

[0044] Taking the second node of the edge as the current node and the first node of the edge as the neighboring node, we obtain the feature vectors of the current node and the neighboring node respectively, where the second node is the source node and the first node is the target node. Based on the initial attention coefficients, we then apply these features to the neighboring nodes (…). The feature vectors of are weighted and then compared with the current node ( The feature vectors of the second node are fused to obtain the updated feature representation of the second node. The fusion formula can be: ,in, This is the updated feature representation of the second node.

[0045] The updated features of the first node are concatenated with the updated features of the second node to obtain the concatenated features, which are then used as edge features.

[0046] Based on the edge features obtained above, the edge features are processed by the softmax function to obtain the attention coefficient of the edge. In this embodiment, the edge features serve as an intermediate carrier for calculating the attention coefficient, integrating the original features of the source node and the target node, which helps to determine the attention coefficient of the edge more accurately.

[0047] The calculated attention coefficients of each edge are normalized and converted into corresponding weight values. The influence of each edge on the relationship between devices is accurately measured based on the weight of each edge.

[0048] This embodiment first determines the potential characteristics of the connections between devices through steps such as initial attention coefficients and updating node features. Then, it calculates attention coefficients using a self-attention mechanism and normalizes these coefficients to obtain weights, accurately reflecting the importance of physical or functional connections between devices. This provides a more reliable structural foundation for subsequent graph neural network processing of multi-source data and extraction of encoding vectors. Through the above-mentioned edge weight calculation process, this embodiment enables the graph neural network to better capture the complex interaction relationships between power plant devices, improve the model's ability to represent the power plant's operating status, and thus enhance the accuracy and reliability of subsequent operation scenario recognition, equipment status assessment, and other processes.

[0049] Before determining the initial attention coefficient for each edge, this embodiment also includes: For each edge, determine the parameter data corresponding to the physical constraint type based on the physical constraint type of the device corresponding to that edge; Quantize the parameter data into the physical constraint coefficients of that edge; The prior matrix is ​​determined based on the physical constraint coefficients of each edge; For each edge, determine the initial attention coefficient for that edge, including: The initial attention weight of the edge is determined based on the prior matrix and the graph attention mechanism; The initial attention coefficient of the edge is determined based on the constraint coefficient and the initial attention weight.

[0050] In this embodiment, the prior matrix is ​​used to constrain and guide the calculation of the initial attention weights, thereby providing an initial weight allocation reference for the initial attention coefficients.

[0051] This embodiment addresses each edge representing a connection between power station equipment in the graph structure. Based on the power station's physical structure (e.g., equipment A1 and equipment A2), it determines the physical constraint types of equipment A1 and equipment A2 connected to that edge, and obtains these constraints. For example, constraint types can be voltage limits, current limits, and power limits. Based on the determined constraint types, it retrieves the corresponding parameter data from the equipment's operating data. If the constraint type is a voltage limit, it obtains the voltage range and rated voltage during normal operation of the equipment.

[0052] This embodiment quantizes parameter data according to a preset quantization method to determine the physical constraint coefficient of each edge. In this embodiment, the physical parameters of the edges can be voltage, current, and power, etc., and a standard value is set for each physical parameter, such as rated voltage, rated current, and rated power. The actual parameter value of each edge at the current moment can be obtained by reading the corresponding actual parameter value in real time from deployed sensors. This embodiment can quantize each physical parameter according to a linear quantization method or a nonlinear quantization method to obtain the quantized parameters corresponding to each physical parameter. For each edge, all the quantized parameters of that edge are integrated into the physical constraint coefficient of that edge. For example, for the voltage parameter, if the rated voltage is... U 0, actual operating voltage is U The quantization coefficients of the voltage parameters are determined using a linear method. The formula for calculating the quantization coefficients of the voltage parameters is as follows: ,in, k 1 represents the quantization coefficient for the voltage parameter. The quantization coefficient for the current parameter can be determined in the same way. k 2, and the quantization coefficient of power parameters. k 3. In this embodiment, the physical constraint coefficient of the edge can be determined using a weighted average method. The weighted average formula is as follows: ,in, p This represents the physical constraint coefficient for that edge. These are the weights corresponding to the quantization coefficients of the voltage parameter, the current parameter, and the power parameter, respectively. These weights can be determined based on the actual physical parameters' influence on that side.

[0053] The physical constraint coefficient of this side is determined based on the quantization coefficient of the voltage parameters and the voltage fluctuation.

[0054] The physical constraint coefficient corresponding to each edge is determined according to the preset quantization method.

[0055] In this embodiment, the dimension of the prior matrix is ​​related to the number of nodes in the graph structure. This embodiment has N device nodes, and the prior matrix is ​​an N×N matrix. The element P in this prior matrix... ij The element P is the physical constraint coefficient of the edge formed by nodes i and j. ji The element values ​​are the physical constraint coefficients of the edge formed by nodes j and i. The edge formed by nodes i and j has the same physical constraint coefficient as the edge formed by nodes j and i.

[0056] The physical constraint coefficients corresponding to each edge in the graph structure are filled into the corresponding positions of the prior matrix to determine the prior matrix.

[0057] For each edge, the feature vector of node i on that edge is... hi and the eigenvectors of node j h j Perform a linear transformation to obtain the transformed feature vector. , ,in, Let be the feature vector after the transformation of node i. Let be the transformed feature vector of node j, and W be the learnable weight matrix.

[0058] In this embodiment, the attention score of an edge can also be determined according to the second attention score calculation formula. The specific method for determining the attention score of an edge according to the second attention score calculation formula is as follows: Based on the transformed feature vectors of node i and node j, the prior matrix, and the formula for calculating the second attention coefficient, the attention score of the edge formed by nodes i and j is calculated. The formula for calculating the second attention score is as follows: ,in, The attention score for that edge. Let be the element values ​​in the i-th row and j-th column of the prior matrix. For activation function, l is the learnable attention vector, and || is the vector concatenation.

[0059] In this embodiment, the attention scores of all edges connected to node i are softmax normalized to obtain the initial attention weight of that edge relative to node i. The initial attention coefficient of the edge is obtained by multiplying the physical constraint coefficient by the initial attention coefficient.

[0060] S103: Determine the operating scenario of the power plant based on the encoding vector, and determine the configuration parameters corresponding to the operating scenario based on the operating scenario.

[0061] In this embodiment, the encoding vector is used to reflect the operating characteristics of each piece of equipment in the power plant. The operating scenarios in this embodiment include stable operation scenario, full load operation scenario and equipment maintenance scenario. The operating scenario of the power plant is determined according to the encoding vector.

[0062] This embodiment determines the features of the encoding vector that can reflect key information such as the status and interaction relationships of power plant equipment based on the encoding vector output by the graph neural network; compares the features of the encoding vector with the feature patterns of various predefined power plant operation scenarios (e.g., normal operating conditions, different load conditions, or fault warning scenarios) to determine the operating scenario corresponding to the current encoding vector features; based on the obtained operating scenario, the configuration parameters that should be adopted by various equipment in the power plant under the scenario are searched in the scenario mapping library. For example, the configuration parameters can be the operating power, pressure, temperature, etc. of the equipment.

[0063] In this embodiment, the operating scenarios include stable operation scenarios, full-load operation scenarios, and equipment maintenance scenarios. Step S103 specifically includes: Key features are determined based on the encoding vector, and these key features are used to reflect the operating status of the power plant. Based on the cosine similarity method, the similarity between key features and various operating scenarios in the scenario mapping library is determined. The scenario mapping library is used to reflect the relationship between different operating scenarios of the power plant and key features. The operating scenario corresponding to the encoding vector is determined based on similarity. Based on the scenario mapping library, the configuration parameters for the operating scenario are determined. The scenario mapping library is also used to reflect the relationship between different operating scenarios of the power plant and the configuration parameters.

[0064] In this embodiment, key feature vectors reflecting the operating status of the power plant are extracted from the encoded vectors output by the graph neural network. These feature vectors include typical parameter features of the equipment (typical parameter features may include features such as temperature, pressure, and flow rate) and features of interactions between equipment. Based on the extracted key feature vectors and the feature vectors corresponding to each operating scenario in the scenario mapping library, the cosine similarity between the key feature vectors and the feature vectors corresponding to the operating scenarios is determined. This scenario mapping library is used to reflect the correlation between different operating scenarios of the power plant and key features.

[0065] In this embodiment, the formula for calculating cosine similarity is: ,in, This represents the cosine similarity value between the key feature vector and the feature vector corresponding to the running scenario. As the key feature vector, The feature vector corresponding to the running scenario. The cosine similarity value is the angle between two feature vectors. In this embodiment, the closer the cosine similarity value is to 1, the more similar the two feature vectors are.

[0066] The operating scenario corresponding to the largest cosine similarity value is selected and used as the current operating scenario of the power plant corresponding to the encoding vector. Based on the current operating scenario of the power plant and the scenario mapping library, which also reflects the relationship between different operating scenarios of the power plant and configuration parameters, the configuration parameters associated with the current operating scenario of the power plant are searched in the scenario mapping library, and these configuration parameters are used as the parameter settings that should be adopted by various equipment of the power plant under this operating scenario.

[0067] S104: Construct a two-layer digital twin model based on the physical entity of the power station and its various equipment.

[0068] In this embodiment, the two-layer digital twin model includes a macroscopic layer and a microscopic layer. The macroscopic layer digital twin model is determined as follows: the power plant is abstracted into several macroscopic units according to their functions, a topology diagram of each device in the power plant is drawn, and the input-output logic of the macroscopic layer digital twin model is determined. For example, in a thermal power plant scenario, the power plant is abstracted into a fuel supply unit, a boiler unit, a turbine unit, a generator unit, and a condensate return unit according to their functions. The logical relationship between the input and output of each unit is determined as fuel system → boiler → turbine → generator. Global parameters are defined for each macroscopic unit. These global parameters may include fuel flow rate, fuel calorific value, boiler steam production, turbine steam intake, turbine output power, and generator efficiency. Connection equations between units are established based on physical conservation laws. For example, boiler steam production = turbine steam intake + pipeline losses; generator output power = turbine output power × generator efficiency.

[0069] Historical operating condition data is acquired, which represents typical operating conditions, including full-load and half-load data. For example, under full-load conditions, rated fuel quantity, rated main steam pressure, temperature, and rated power generation can be obtained. Under half-load conditions, 50% of the rated fuel quantity, the corresponding main steam parameters, and power generation can be obtained. In this embodiment, the full-load data (rated fuel quantity, rated main steam pressure, and temperature) is input into the macroscopic-level digital twin model. The deviation between the power generation output by the model and the actual rated power generation is compared. The model parameters are adjusted based on this deviation until the deviation between the output power generation and the actual rated power generation is ≤5%.

[0070] The micro-level digital twin model focuses on the internal physical processes, local states, and potential faults of the equipment. The micro-level digital twin model is determined by disassembling and modeling each unit in the macro-level. This embodiment takes a boiler unit as an example, disassembling it into boiler sub-components such as the furnace, water-cooled walls, superheater, and economizer. Fluid dynamics models can be established for each sub-component to simulate the combustion process and temperature distribution within the furnace; finite element thermal stress models can be established to simulate the stress distribution of the water-cooled wall tubes at high temperatures and predict potential boiler faults. In the macro-level, the boiler merely converts the input fuel to output steam.

[0071] The parameters of the macroscopic layer are used as boundary conditions for the equipment in the microscopic layer. This boundary condition can be that the macroscopic boiler outlet pressure equals the microscopic steam turbine inlet pressure, thus ensuring data communication between the microscopic and macroscopic layers. Historical parameter data of the equipment is acquired, including historical operating data and historical fault data. The historical parameter data of the boiler is used as input to the microscopic digital twin model. The parameters output by the model are compared with the actual parameters. Based on the comparison results, the component mechanism parameters (e.g., thermal conductivity coefficient, friction coefficient) are adjusted to improve the simulation accuracy of the equipment.

[0072] Establish a data interaction interface between the macro-level digital twin model and the micro-level digital twin model, and establish parameter transfer from the macro-level to the micro-level, as well as parameter transfer from the micro-level to the macro-level.

[0073] The transfer of parameters from the macroscopic to the microscopic level can be achieved by using simulation results from the microscopic level to provide boundary conditions for the operation of the microscopic level. For example, the total steam production and total heat absorption of the boiler determined at the macroscopic level can be transferred to the microscopic level as the overall objective and constraint for the microscopic level simulation.

[0074] Parameter transfer from the microscopic to the macroscopic level can be achieved by using simulation results from the microscopic level to provide correction conditions for key parameters of the macroscopic model and to set safety boundary conditions. For example, if microscopic simulation reveals a risk of overheating (local state) in a certain area of ​​the water-cooled wall under the current load, this risk information will be fed back to the macroscopic level. The boiler efficiency parameters or maximum allowable steam production parameters (physical constraints) in the macroscopic digital twin model will then be dynamically adjusted based on this risk information, thereby avoiding unsafe operating conditions in subsequent simulations.

[0075] S105: Obtain the current operating parameters of each device, and determine the operating status and health status of each device in the power plant based on the two-layer digital twin model, configuration parameters and the current operating parameters of each device.

[0076] In this embodiment, based on the two-layer digital twin model, configuration parameters, and the current operating parameters of each device, the operating status and health status of each device in the power plant are determined, including: The control parameters corresponding to each piece of equipment in the power plant are determined based on the macroscopic layer and configuration parameters of the two-layer digital twin model. Based on the micro-level of the two-layer digital twin model, the control parameters, and the current operating parameters of each device, the operating status and health status of each device in the power plant are determined.

[0077] In this embodiment, taking the steam turbine unit as an example, the determined power plant operation scenario configuration parameters are input into the macro layer of the two-layer digital twin model to determine the macro operating conditions corresponding to the configuration parameters (e.g., full-load power generation condition, peak load condition, etc.), thereby generating macro-level target parameters (e.g., total boiler steam production, total steam turbine output power, and total generator transmission power, etc.).

[0078] The macro-level target parameters are broken down into individual sub-equipment parameters (e.g., the total steam production of the boiler is broken down into the total fuel supply to each burner, and the total output power of the turbine is broken down into the power allocation values ​​of the high-pressure cylinder and the low-pressure cylinder, etc.), forming sub-level control targets. Combining the pre-defined parameter relationships between the macro and micro levels, the sub-level control targets are transformed into the control parameters of the corresponding equipment, ensuring that the control parameters of the equipment meet the macro-level operating condition requirements.

[0079] The obtained control and operating parameters are input into a micro-level digital twin model to simulate and generate the equipment's operating baseline parameters under these control parameters (e.g., steam drum water level range, bearing temperature threshold, vibration frequency range, etc. under normal operating conditions). Based on these operating baseline parameters and the equipment's real-time operating parameters, the operating status and health status of the power plant equipment are determined. For example, if the equipment's real-time operating parameters are within the operating baseline range, the equipment is determined to be in normal operating condition; if the equipment's real-time operating parameters exceed the operating baseline range but do not reach the preset fault association threshold, it is determined to be in a sub-healthy operating state; if the equipment's real-time operating parameters exceed the operating baseline range and reach the preset fault association threshold, it is determined to be in an abnormal operating state. In this embodiment, the equipment's real-time operating parameters may include the equipment's operating power, operating speed (e.g., turbine speed), operating voltage (e.g., input and output voltage of substation equipment), and operating pressure (e.g., fluid pressure in pipelines).

[0080] In this embodiment, the multi-source data and the real-time operating data of the equipment are interconnected and complementary. The multi-source data in this embodiment can describe the relevant characteristics of the equipment's operation from different perspectives (vibration, discharge, and environmental influences), while the equipment's operating parameters are the specific numerical manifestations of the actual operating state of the equipment under the influence of multi-source data. For example, the equipment vibration spectrum data can reflect the vibration situation during equipment operation. By analyzing the vibration spectrum, it can be determined whether the equipment's operating speed is stable. Discharge signal data can detect whether there is partial discharge in the equipment (e.g., high-voltage electrical equipment). If partial discharge exists, it may affect the insulation performance of the equipment, which may adversely affect the operating voltage, operating current, and other parameters of the equipment, or even lead to equipment failure. If the temperature data in the environmental data is too high, it may cause difficulty in heat dissipation of the equipment, causing the operating temperature of the equipment to rise, which in turn affects the operating power, operating speed, and other parameters of the equipment.

[0081] As can be seen from the above, this embodiment processes multi-source data by combining graph neural networks with a self-attention mechanism. By obtaining the weights corresponding to each edge and constructing an adjacency matrix through the self-attention mechanism, it can accurately capture the complex physical and functional connections between devices, making the output encoded vector more accurately reflect the overall state of the power plant. This embodiment identifies operating scenarios based on encoded vectors and cosine similarity, enabling more efficient and accurate matching of scenarios such as stable operation, full load, and equipment maintenance, thus improving the accuracy and efficiency of scenario identification. This embodiment also utilizes a two-layer digital twin model. At the macro level, it combines configuration parameters to determine the control parameters of the equipment, achieving macro-level control of the overall power plant operation. At the micro level, it focuses on equipment details, combining control parameters and equipment operating parameters to accurately assess the operating and health status of the equipment. This embodiment, through the two-layer digital twin model, ensures both a global understanding of the overall operating status of the power plant and in-depth, precise monitoring and diagnosis at the equipment level, achieving comprehensive coverage from the whole to the part. This improves the accuracy and comprehensiveness of state assessment in remote power plant simulation, providing strong support for remote monitoring, fault warning, and optimized operation of the power plant, and enhancing the intelligence and refinement of power plant operation management.

[0082] In one embodiment of this application, after determining the operating status and health status of each device in the power plant based on the two-layer digital twin model, configuration parameters, and the current operating parameters of each device, the method further includes: The graph structure data and encoding vectors are used to update the pre-trained graph neural network to obtain the updated graph neural network.

[0083] In this embodiment, the mean squared error loss function can be used to measure the difference between the encoded vector output by the graph neural network and the expected result. If the expected encoded vector is... The actual output encoded vector is The difference between the encoded vector and the expected result is determined based on the mean squared error loss function. The mean squared error loss function is: ,in, L The difference between the encoded vector and the expected result. n 1 represents the dimension of the encoded vector. For the first in the expected vector One element, The first in the actual output vector Each element.

[0084] The acquired graph structure data is input into a pre-trained graph neural network to obtain the encoding vector. Based on the mean squared error loss function, the gradient of the difference value with respect to each parameter of the graph neural network (e.g., weight matrix, bias, etc.) is calculated. The parameters of the graph neural network are updated based on this gradient and a pre-defined optimization algorithm to obtain the updated graph neural network.

[0085] This embodiment can use an updated graph neural network to process multi-source data, thereby determining an updated encoding vector to more accurately reflect the subsequent operation of the power plant.

[0086] In one embodiment of this application, feature vectors corresponding to each of the multi-source data of the device are determined, and the corresponding feature vectors are concatenated to obtain the concatenated feature vector corresponding to the device, including: Based on the multi-source data of the device, the multi-source feature vectors corresponding to each multi-source data are determined. The multi-source feature vectors include vibration feature vectors, discharge feature vectors, and environmental feature vectors. The multi-source feature vector is input into a pre-trained fully connected neural network model, which outputs the dynamic weights corresponding to each of the multi-source feature vectors. The weighted multi-source feature vectors are determined based on the multi-source feature vectors and their corresponding dynamic weights. The weighted multi-source feature vectors are concatenated to obtain the concatenated feature vector corresponding to the device.

[0087] In this embodiment, the current multi-source data of the device is acquired, and the corresponding feature vectors are determined based on the multi-source data. For the device vibration spectrum data, the vibration feature vector can be determined based on the time domain characteristics (e.g., root mean square) and frequency characteristics (e.g., fault characteristic frequency amplitude) of the data. For the discharge signal data, the discharge feature vector can be determined based on the discharge statistical characteristics (e.g., average discharge amount) of the data. For the environmental data, the environmental feature vector can be determined based on data such as temperature and humidity.

[0088] The obtained multi-source feature vectors are input into a pre-trained fully connected neural network model, which outputs the dynamic weights corresponding to each multi-source feature vector. The multi-source feature vectors are weighted according to their respective dynamic weights to obtain the weighted multi-source feature vectors. The weighted multi-source feature vectors are then concatenated to obtain the concatenated feature vector.

[0089] The training process for the pre-trained fully connected neural network model is as follows: Acquire multiple sets of historical, multi-source data from the equipment under different operating conditions; For each set of historical multi-source data, obtain the corresponding multi-source feature vector, which includes vibration feature vector, discharge feature vector, and environmental feature vector. Divide all multi-source feature vectors according to a preset ratio (such as 7:3) to obtain a training set and a validation set. Use the training set to train the model, calculate the dynamic weights of the model output through forward propagation, calculate the loss value according to the loss function, and then update the model parameters (weights and biases, etc.) through the backpropagation algorithm. Use the validation set to monitor the model's performance and prevent overfitting.

[0090] This embodiment overcomes the limitations of a single data dimension by concatenating multi-source feature vectors, comprehensively depicting the equipment status from multiple perspectives such as mechanical vibration, electrical discharge, and environmental influences. This allows the corresponding feature vectors to more accurately represent the equipment's operating condition. Furthermore, this embodiment outputs dynamic weights through a pre-trained fully connected neural network model, adaptively adjusting the weights of each feature vector based on the real-time performance of multi-source data during actual equipment operation. By weighting and concatenating the multi-source feature vectors, this embodiment effectively highlights features that are more discriminative of the equipment status, reduces interference from irrelevant or secondary features, and improves the quality of the concatenated feature vector.

[0091] Based on the same inventive concept, this embodiment also provides a power plant remote simulation device for implementing the power plant remote simulation method described above. The solution provided by this device is similar to the solution described in the above method; therefore, the specific limitations in one or more power plant remote simulation device embodiments provided below can be found in the limitations of the power plant remote simulation method described above, and will not be repeated here.

[0092] This embodiment provides a remote power plant simulation device, such as... Figure 2 As shown, the remote simulation device 20 for the power station includes: a data acquisition module 21, an encoding vector determination module 22, a configuration parameter determination module 23, a two-layer digital twin model determination module 24, and a simulation evaluation module 25.

[0093] In one embodiment of this application, the data acquisition module 21 is used to acquire multi-source data of the device based on the edge nodes deployed locally in the power plant. The power plant has multiple edge nodes deployed locally, and each edge node is a device. The multi-source data of the device includes the multi-source data corresponding to each device. The multi-source data includes device vibration spectrum data, discharge signal and environmental data. The encoding vector determination module 22 is used to determine graph structure data based on multi-source data, input the graph structure data into a pre-trained graph neural network, and output encoding vectors. The encoding vectors are used to reflect the relevant characteristic information of the operation of each piece of equipment in the power plant. The configuration parameter determination module 23 is used to determine the operating scenario of the power plant based on the encoding vector, and to determine the configuration parameters corresponding to the operating scenario based on the operating scenario. The two-layer digital twin model determination module 24 is used to construct a two-layer digital twin model based on the physical entity of the power station and each piece of equipment in the power station. The two-layer digital twin model includes a macroscopic layer and a microscopic layer. The simulation evaluation module 25 is used to obtain the current operating parameters of each device and determine the operating status and health status of each device in the power plant based on the two-layer digital twin model, configuration parameters and the current operating parameters of each device.

[0094] In one embodiment of this application, when determining graph structure data based on multi-source data, the encoding vector determination module 22 is specifically used for: Determine the corresponding feature vectors for each of the multiple data sources; Each piece of equipment in the power plant is defined as a node in a graph structure, and the physical or functional connections between the nodes are defined as edges in the graph structure. The weights corresponding to each edge of the graph structure are obtained based on the self-attention mechanism; The adjacency matrix is ​​determined based on the edges and corresponding weights of the graph structure. The adjacency matrix is ​​used to represent the connection relationships and connection attributes between nodes. Graph structure data is determined based on the eigenvectors and adjacency matrices corresponding to each of the multiple data sources.

[0095] In one embodiment of this application, before obtaining the weights corresponding to each edge of the graph structure according to the self-attention mechanism, the encoding vector determination module 22 is specifically used for: For each edge, determine the initial attention coefficient for that edge; The first node of the edge is taken as the current node, and the second node of the edge is taken as the neighboring node. Based on the initial attention coefficient of the edge, the information of the neighboring nodes is fused into the current node to obtain the updated feature representation of the first node. The second node of the edge is taken as the current node, and the first node of the edge is taken as the neighboring node. Based on the initial attention coefficient of the edge, the information of the neighboring nodes is fused into the current node to obtain the updated feature representation of the second node. Based on the updated first node feature representation and the updated second node feature representation, determine the edge feature corresponding to the edge; The weights corresponding to the edges of each graph neural network are obtained based on the self-attention mechanism, including: The attention coefficients of each edge are determined based on the self-attention mechanism and edge characteristics; The attention coefficients of each edge are normalized to obtain the weights of each edge.

[0096] In one embodiment of this application, before determining the initial attention coefficient for each edge, the encoding vector determination module 22 is specifically used for: For each edge, determine the parameter data corresponding to the physical constraint type based on the physical constraint type of the device corresponding to that edge; Quantize the parameter data into the physical constraint coefficients of that edge; The prior matrix is ​​determined based on the physical constraint coefficients of each edge; For each edge, determine the initial attention coefficient for that edge, including: The initial attention weight of the edge is determined based on the prior matrix and the graph attention mechanism; The initial attention coefficient of the edge is determined based on the constraint coefficient and the initial attention weight.

[0097] In one embodiment of this application, the operating scenarios include stable operation scenarios, full-load operation scenarios, and equipment maintenance scenarios. When determining the operating scenario of the power plant based on the encoding vector, and determining the configuration parameters corresponding to the operating scenario based on the operating scenario, the configuration parameter determination module 23 is specifically used for: Key features are determined based on the encoding vector, and these key features are used to reflect the operating status of the power plant. Based on the cosine similarity method, the similarity between key features and various operating scenarios in the scenario mapping library is determined. The scenario mapping library is used to reflect the relationship between different operating scenarios of the power plant and key features. The operating scenario corresponding to the encoding vector is determined based on similarity. Based on the scenario mapping library, the configuration parameters for the operating scenario are determined. The scenario mapping library is also used to reflect the relationship between different operating scenarios of the power plant and the configuration parameters.

[0098] In one embodiment of this application, when determining the operating status and health status of each device in the power plant based on the two-layer digital twin model, configuration parameters, and the current operating parameters of each device, the simulation evaluation module 25 is specifically used for: The control parameters corresponding to each piece of equipment in the power plant are determined based on the macroscopic layer and configuration parameters of the two-layer digital twin model. Based on the micro-level of the two-layer digital twin model, the control parameters, and the current operating parameters of each device, the operating status and health status of each device in the power plant are determined.

[0099] In one embodiment of this application, after determining the operating status and health status of each device in the power plant based on the two-layer digital twin model, configuration parameters, and the current operating parameters of each device, the simulation evaluation module 25 is specifically used for: The graph structure data and encoding vectors are used to update the pre-trained graph neural network to obtain the updated graph neural network.

[0100] See Figure 3 , Figure 3 This is a schematic block diagram of an electronic device provided according to an embodiment of this application. Figure 3 The electronic device 300 in this embodiment may include one or more processors 301, one or more input devices 302, one or more output devices 303, and one or more memories 304. The processors 301, input devices 302, output devices 303, and memories 304 communicate with each other via a communication bus 305. The memories 304 store computer programs, including program instructions. The processors 301 execute the program instructions stored in the memories 304. Specifically, the processors 301 are configured to invoke the program instructions to perform the functions of each module / unit in the above-described device embodiments, for example... Figure 2 The functions of the data acquisition module 21, the encoding vector determination module 22, the configuration parameter determination module 23, the two-layer digital twin model determination module 24, and the simulation evaluation module 25 are shown.

[0101] It should be understood that, in this embodiment, the processor 301 may be a Central Processing Unit (CPU), but it can also be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor.

[0102] Input device 302 may include a touchpad, a fingerprint sensor (for collecting the user's fingerprint information and fingerprint orientation information), a microphone, etc., and output device 303 may include a display (LCD, etc.), a speaker, etc.

[0103] The memory 304 may include read-only memory and random access memory, and provides instructions and data to the processor 301. A portion of the memory 304 may also include non-volatile random access memory. For example, the memory 304 may also store graph structure data, encoding vectors, configuration parameters corresponding to power plant operating scenarios, and equipment health status information.

[0104] In specific implementation, the processor 301, input device 302, and output device 303 described in this embodiment can execute the implementation method described in the power plant remote simulation method provided in this embodiment, or they can execute the implementation method of the electronic device described in this embodiment, which will not be repeated here.

[0105] In another embodiment of this application, a computer-readable storage medium is provided. This computer-readable storage medium stores a computer program, which includes program instructions. When executed by a processor, the program instructions implement all or part of the processes in the methods described above. Alternatively, the computer program can instruct related hardware to complete the process. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include any entity or device capable of carrying computer program code, a recording medium, a USB flash drive, a portable hard drive, a magnetic disk, an optical disk, a computer memory, a read-only memory (ROM), a random access memory (RAM), an electrical carrier signal, a telecommunication signal, and a software distribution medium, etc.

[0106] The computer-readable storage medium can be an internal storage unit of the electronic device in any of the foregoing embodiments, such as a hard disk or memory of the electronic device. The computer-readable storage medium can also be an external storage device of the electronic device, such as a plug-in hard disk, smart media card (SMC), secure digital card (SD), flash card, etc., provided on the electronic device. Furthermore, the computer-readable storage medium can include both internal and external storage units of the electronic device. The computer-readable storage medium is used to store computer programs and other programs and data required by the electronic device. The computer-readable storage medium can also be used to temporarily store data that has been output or will be output.

[0107] Those skilled in the art will recognize that the modules / units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this application.

[0108] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the electronic devices and units described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0109] In the several embodiments provided in this application, it should be understood that the disclosed electronic devices and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For instance, the division of modules / units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple modules, units, or components may be combined or integrated into another system, or some features may be ignored or not executed. In addition, the mutual coupling or direct coupling or communication connection shown or discussed may be indirect coupling or communication connection through some interfaces or modules / units, or it may be an electrical, mechanical, or other form of connection.

[0110] The modules / units described as separate components may or may not be physically separate. Similarly, the components shown as modules / units may or may not be physical modules / units; they may be located in one place or distributed across multiple network modules / units. Some or all of the modules / units can be selected to achieve the purpose of this embodiment, depending on actual needs.

[0111] Furthermore, the functional modules in the various embodiments of this application can be integrated into one processing unit, or each module can exist physically separately, or two or more modules can be integrated into one unit. The integrated modules / units described above can be implemented in hardware or in the form of software functional modules / units.

[0112] The above are merely specific embodiments of this application, but the scope of protection of this application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this application, and these modifications or substitutions should all be covered within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for remote simulation of a power plant, characterized in that, include: The power plant acquires multi-source data of the equipment based on the edge nodes deployed locally. The power plant has multiple edge nodes deployed locally, and each edge node represents one device. The multi-source data of the device includes the multi-source data corresponding to each device, including device vibration spectrum data, discharge signals, and environmental data. Graph structure data is determined based on the multi-source data, and the graph structure data is input into a pre-trained graph neural network to output an encoding vector. The encoding vector is used to reflect the relevant characteristics of the operation of each piece of equipment in the power plant. The operating scenario of the power plant is determined based on the encoding vector, and the configuration parameters corresponding to the operating scenario are determined based on the operating scenario. A two-layer digital twin model is constructed based on the physical entity of the power station and its various equipment. The two-layer digital twin model includes a macroscopic layer and a microscopic layer. Obtain the current operating parameters of each device, and determine the operating status and health status of each device in the power station based on the two-layer digital twin model, the configuration parameters, and the current operating parameters of each device.

2. The power plant remote simulation method as described in claim 1, characterized in that, The step of determining graph structure data based on the multi-source data includes: For each device, the corresponding feature vector is determined based on the multi-source data of that device, and the corresponding feature vectors are concatenated to obtain the concatenated feature vector corresponding to that device. Each piece of equipment in the power plant is defined as a node in a graph structure, and the physical or functional connections between the nodes are defined as edges in the graph structure. The weights corresponding to each edge of the graph structure are obtained based on the self-attention mechanism; The adjacency matrix is ​​determined based on the edges and corresponding weights of the graph structure. The adjacency matrix is ​​used to represent the connection relationships and connection attributes between nodes. The graph structure data is determined based on the splicing feature vectors corresponding to each device and the adjacency matrix.

3. The power plant remote simulation method as described in claim 2, characterized in that, Before obtaining the weights corresponding to each edge of the graph structure based on the self-attention mechanism, the method further includes: For each edge, determine the initial attention coefficient for that edge; The first node of the edge is taken as the current node, and the second node of the edge is taken as the neighboring node. Based on the initial attention coefficient of the edge, the information of the neighboring nodes is fused into the current node to obtain the updated feature representation of the first node. The second node of the edge is taken as the current node, and the first node of the edge is taken as the neighboring node. Based on the initial attention coefficient of the edge, the information of the neighboring nodes is fused into the current node to obtain the updated feature representation of the second node. Based on the updated first node feature representation and the updated second node feature representation, determine the edge feature corresponding to the edge; The step of obtaining the weights corresponding to each edge of the graph structure based on the self-attention mechanism includes: Based on the self-attention mechanism and the characteristics of each edge, the attention coefficients corresponding to each edge are determined respectively. Normalize the attention coefficients corresponding to each edge to obtain the weights of each edge.

4. The power plant remote simulation method as described in claim 3, characterized in that, Before determining the initial attention coefficient for each edge, the process also includes: For each edge, based on the physical constraint type of the device corresponding to the edge, the parameter data corresponding to the physical constraint type is determined, and the parameter data is quantized into the physical constraint coefficient of the edge. The physical constraint type represents the physical constraint type between the two devices corresponding to the edge. The prior matrix is ​​determined based on the physical constraint coefficients corresponding to each edge; The process of determining the initial attention coefficient for each edge includes: The initial attention coefficient of an edge is determined based on its constraint coefficient and initial attention weight. The initial attention weights corresponding to each edge are determined based on the prior matrix and the graph attention mechanism.

5. The power plant remote simulation method as described in claim 1, characterized in that, The operating scenarios include stable operation scenarios, full-load operation scenarios, and equipment maintenance scenarios. The process of determining the power plant's operating scenario based on the encoding vector, and determining the corresponding configuration parameters based on the operating scenario, includes: Key features are determined based on the encoding vector, and these key features are used to reflect the operating status of the power station; The similarity between the key features and each operating scenario in the scenario mapping library is determined based on the cosine similarity method. The scenario mapping library is used to reflect the association between different operating scenarios of the power plant and the key features. The running scenario corresponding to the encoding vector is determined based on the similarity. Based on the scenario mapping library, the configuration parameters of the operating scenario are determined. The scenario mapping library is also used to reflect the relationship between different operating scenarios of the power plant and the configuration parameters.

6. The power plant remote simulation method as described in claim 1, characterized in that, The step of determining the operating status and health status of each device in the power station based on the two-layer digital twin model, the configuration parameters, and the current operating parameters of each device includes: The control parameters corresponding to each piece of equipment in the power plant are determined based on the macroscopic layer of the two-layer digital twin model and the configuration parameters. Based on the micro-level of the two-layer digital twin model, the control parameters, and the current operating parameters of each device, the operating status and health status of each device in the power plant are determined.

7. The power plant remote simulation method as described in claim 6, characterized in that, After determining the operating status and health status of each device in the power station based on the two-layer digital twin model, the configuration parameters, and the current operating parameters of each device, the method further includes: The graph structure data and the encoding vector are used to update the pre-trained graph neural network to obtain the updated graph neural network.

8. A remote simulation device for a power plant, characterized in that, include: The data acquisition module is used to acquire multi-source data of the equipment based on the edge nodes deployed locally in the power plant. There are multiple edge nodes deployed locally in the power plant, and each edge node is a device. The multi-source data of the device includes the multi-source data corresponding to each device. The multi-source data includes the device vibration spectrum data, discharge signal and environmental data. The encoding vector determination module is used to determine graph structure data based on the multi-source data, input the graph structure data into a pre-trained graph neural network, and output an encoding vector. The encoding vector is used to reflect the relevant characteristic information of the operation of each piece of equipment in the power plant. The configuration parameter determination module is used to determine the operating scenario of the power plant based on the encoding vector, and to determine the configuration parameters corresponding to the operating scenario based on the operating scenario. A two-layer digital twin model determination module is used to construct a two-layer digital twin model based on the physical entity of the power station and each piece of equipment in the power station. The two-layer digital twin model includes a macroscopic layer and a microscopic layer. The simulation evaluation module is used to obtain the current operating parameters of each device, and determine the operating status and health status of each device in the power plant based on the two-layer digital twin model, the configuration parameters and the current operating parameters of each device.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method as described in any one of claims 1 to 7.

Images