Industrial user power indirect carbon emission monitoring method and system based on graphsage-mrf

Abstract

The application discloses a kind of industrial user power indirect carbon emission monitoring method and system based on GraphSAGE-MRF, the method first obtains the electrical information of power system node, carbon emission information and topological structure information, pre-processes to obtain node electricity carbon coupling information and adjacency matrix;Subsequently, a GraphSAGE-MRF model is built, node feature learning and neighbor information aggregation are carried out in the GraphSAGE module therein, and graph-level electricity carbon coupling features are output;Then, the spatial dependence relationship between the characteristics of the node itself and the adjacent nodes is modeled through the MRF module, and the node electricity carbon factor is output;Finally, the power indirect carbon emission is calculated by combining the industrial user electricity consumption.The application combines the feature extraction capability of graph neural network and the spatial modeling advantage of Markov random field, and still maintains stable accuracy under the dynamic change scene of power grid topology, meets the monitoring requirements of industrial users, and provides reliable support for low-carbon operation of power system.

Description

Technical Field

[0001] This invention belongs to the field of online carbon emission monitoring technology for power systems, and particularly relates to a method and system for monitoring indirect carbon emissions from industrial users' electricity based on GraphSAGE-MRF. Background Technology

[0002] Existing monitoring methods are mostly based on statistical regression or simple neural networks, which fail to fully consider the complex topological dependencies and electrical-carbon coupling relationships between power system nodes, resulting in limited monitoring accuracy, especially insufficient adaptability when the power grid structure changes dynamically.

[0003] Traditional methods have the following main drawbacks:

[0004] 1. It failed to effectively model the complex topology of the power system and ignored the spatial dependencies between nodes;

[0005] 2. Lack of in-depth exploration of the electro-carbon coupling characteristics, resulting in limited feature extraction capabilities;

[0006] 3. Poor robustness in scenarios with dynamic changes in power grid topology;

[0007] 4. The monitoring accuracy is difficult to meet the ±5% accuracy requirements of industrial users.

[0008] Therefore, there is an urgent need for a high-precision carbon emission monitoring method that can integrate topological structures, extract deep features, and introduce physical constraints. Summary of the Invention

[0009] In a first aspect, the present invention provides a method for monitoring indirect carbon emissions from industrial user electricity based on GraphSAGE-MRF, comprising:

[0010] Obtain electrical information, carbon emission information, and topology information of power system nodes; construct node electrical-carbon coupling information based on the electrical information and carbon emission information; and construct an adjacency matrix based on the topology information.

[0011] A GraphSAGE-MRF model is constructed, and the node electro-carbon coupling information and the adjacency matrix are input into the GraphSAGE module of the GraphSAGE-MRF model. Node feature learning and neighbor node information aggregation are performed through multi-layer graph convolution operations, and graph-level electro-carbon coupling features are output.

[0012] The graph-level electrocarbon coupling features are input into the MRF module of the GraphSAGE-MRF model. The node’s own electrocarbon characteristics are modeled by a univariate potential function, and the spatial dependence between adjacent nodes is modeled by a binary potential function. The indirect electrocarbon factor of the node where the industrial user is located is output.

[0013] The power indirect carbon emission amount is calculated based on the indirect electric carbon factor of the node where the industrial user is located and the power consumption of the industrial user.

[0014] In a second aspect, the present application provides a GraphSAGE-MRF-based industrial user power indirect carbon emission monitoring system, comprising:

[0015] An acquisition module is configured to acquire electrical information, carbon emission information and topological structure information of a power system node, construct node electric-carbon coupling information based on the electrical information and the carbon emission information, and construct an adjacency matrix based on the topological structure information;

[0016] A first output module is configured to construct a GraphSAGE-MRF model, input the node electric-carbon coupling information and the adjacency matrix into a GraphSAGE module in the GraphSAGE-MRF model, perform node feature learning and neighbor node information aggregation through multi-layer graph convolution operation, and output graph-level electric-carbon coupling features;

[0017] A second output module is configured to input the graph-level electric-carbon coupling features into an MRF module in the GraphSAGE-MRF model, model the electric-carbon characteristics of the node itself through a unary potential function, model the spatial dependency relationship between adjacent nodes through a binary potential function, and output an indirect electric carbon factor of the node where the industrial user is located.

[0018] A calculation module is configured to calculate the power indirect carbon emission amount based on the indirect electric carbon factor of the node where the industrial user is located and the power consumption of the industrial user.

[0019] In a third aspect, an electronic device is provided, comprising at least one processor and a memory connected to the at least one processor in communication, wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the steps of the GraphSAGE-MRF-based industrial user power indirect carbon emission monitoring method of any embodiment of the present application.

[0020] In a fourth aspect, the present application further provides a computer-readable storage medium having a computer program stored thereon, wherein the program instructions are executed by a processor to enable the processor to perform the steps of the GraphSAGE-MRF-based industrial user power indirect carbon emission monitoring method of any embodiment of the present application.

[0021] The GraphSAGE-MRF-based industrial user power indirect carbon emission monitoring method and system provided by the application can deeply extract node self electric carbon features and aggregate neighbor node information through multi-order graph convolution operation of the GraphSAGE module, accurately capture the correlation influence between different nodes, and combine the potential function modeling of the MRF module to describe the internal carbon emission characteristics of a single node through a unary potential function, constrain the spatial dependence relationship of adjacent nodes through a binary potential function, ensure that the output node electric carbon factor conforms to the physical law of the power system, fundamentally solve the problem of insufficient accuracy caused by insufficient modeling of node correlation in the traditional method, and realize fine measurement of industrial user power indirect carbon emission. BRIEF DESCRIPTION OF DRAWINGS

[0022] In order to more clearly illustrate the technical solutions of the embodiments of the application, the drawings needed in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are some embodiments of the application, and other drawings can also be obtained by those skilled in the art without creative labor.

[0023] Figure 1 A flowchart of a GraphSAGE-MRF-based industrial user power indirect carbon emission monitoring method provided by an embodiment of the application is provided.

[0024] Figure 2 A network structure diagram of the GraphSAGE module provided by an embodiment of the application is provided.

[0025] Figure 3 A whole block diagram of the GraphSAGE-MRF model provided by an embodiment of the application is provided.

[0026] Figure 4 A loss function curve of the GraphSAGE-MRF model provided by an embodiment of the application is provided.

[0027] Figure 5 A visualization diagram of the average absolute error of each node provided by an embodiment of the application is provided.

[0028] Figure 6 An error comparison diagram under different network topology change scenarios provided by an embodiment of the application is provided.

[0029] Figure 7 A structural block diagram of a GraphSAGE-MRF-based industrial user power indirect carbon emission monitoring system provided by an embodiment of the application is provided.

[0030] Figure 8 A structural schematic diagram of an electronic device provided by an embodiment of the application is provided. DETAILED DESCRIPTION

[0031] In order to make the purposes, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some but not all of the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by a person of ordinary skill in the art without creative work fall within the protection scope of the present application.

[0032] Referring to Figure 1 , a flowchart of an industrial user power indirect carbon emission monitoring method based on GraphSAGE-MRF is shown.

[0033] As Figure 1 shown, the industrial user power indirect carbon emission monitoring method based on GraphSAGE-MRF specifically comprises the following steps:

[0034] Step S101, electrical information, carbon emission information and topological structure information of a power system node are acquired, node electricity-carbon coupling information is constructed based on the electrical information and the carbon emission information, and an adjacency matrix is constructed based on the topological structure information.

[0035] In this step, the electrical information is characterized by node injection power, wherein:

[0036] If the node i is a unit node, the injection power is the active power of the generator unit .

[0037] If the node i is a load node, the injection power is the negative value of the active power of the load .

[0038] If the node i is a mixed node, the injection power is .

[0039] If the node i is an intermediate node, the injection power is 0.

[0040] It should be noted that the carbon emission information includes a first carbon emission intensity of a generator unit node and a second carbon emission intensity of a non-generator unit node.

[0041] The expression of the first carbon emission intensity is:

[0042] ,

[0043] In the formula, is the first carbon emission intensity, is the direct carbon emission amount of the source-side node generator unit measured in real time at the time period t, is the power generation power of the source-side node generator unit at the time period t, a generator set operation time of a source side node;

[0044] The expression of the second carbon emission intensity is:

[0045] ,

[0046] In the formula, is the second carbon emission intensity, is a load average carbon flow rate, is a load average active power, is a total number of generator set nodes, is a total number of load nodes, is a generator set active power, is a load active power.

[0047] Step S102, constructing a GraphSAGE-MRF model, and inputting the node electric carbon coupling information and the adjacency matrix into a GraphSAGE module in the GraphSAGE-MRF model, performing node feature learning and neighbor node information aggregation through multi-layer graph convolution operation, and outputting to obtain a graph-level electric carbon coupling feature.

[0048] In this step, the GraphSAGE module includes three cascaded SAGEConv graph convolution layers, and the node feature learning and neighbor node information are fused through a hierarchical message passing mechanism, wherein for node i in the first The representation of the SAGEConv convolution layer in the first

[0049] ,

[0050] In the formula, is the neighborhood aggregation feature of node is the feature representation of node in the first layer, is the neighborhood set of node is the neighborhood set of node is the normalized coefficient between node and node is the learned weight matrix of the first layer feature aggregation, is an activation function.

[0051] ​​It should be noted that the GraphSAGE module also includes a global pooling layer, wherein after being processed by three cascaded SAGEConv graph convolution layers, the node features are aggregated based on a global pooling mechanism of the global pooling layer to form a graph-level electric-carbon coupling feature representation, and an expression of the global pooling layer is:

[0052] ,

[0053] In the formula, is a graph-level feature representation output by the global pooling layer, is a pooling operation function, is a feature representation of a node in the i-th layer, is a feature representation of a node in the i-th layer, is a feature representation of a node in the i-th layer, is a set of nodes in the graph.

[0054] Further, the GraphSAGE-MRF model is trained by a multi-objective loss function, and a calculation formula of the loss function is:

[0055] ,

[0056] In the formula, is a mean square error loss, is an MRF regularization loss.

[0057] In step S103, the graph-level electric-carbon coupling feature is input into an MRF module in the GraphSAGE-MRF model, a unary potential function is used to model the electric-carbon characteristics of a node itself, a binary potential function is used to model the spatial dependence relationship between adjacent nodes, and an indirect electric-carbon factor of a node where an industrial user is located is output.

[0058] In this step, an expression of the unary potential function is:

[0059] ,

[0060] In the formula, is a unary potential function, representing the intrinsic characteristics of a single node, is a result of full connection layer mapping, representing a feature representation of each node, is a node bias parameter, used to adjust the basic feature of each node;

[0061] An expression of the binary potential function is:

[0062] ,

[0063] In the formula, is a binary potential function, representing the spatial dependence relationship and interaction between nodes, ​is an interaction weight matrix, used to represent the interaction strength between nodes, is an element-level point multiplication operation, used to ensure that there is an interaction between adjacent nodes, is a power system adjacency matrix.

[0064] In step S104, the indirect power carbon emission quantity is calculated based on the indirect power carbon factor of the node where the industrial user is located and the power consumption of the industrial user.

[0065] In this step, the indirect power carbon factor and the power consumption of the industrial user are multiplied to calculate the indirect power carbon emission quantity.

[0066] In summary, the method of the present application can deeply extract the node's own electric carbon features and aggregate neighbor node information through the multi-order graph convolution operation of the GraphSAGE module, accurately capture the correlation between different nodes, and combine the potential function modeling of the MRF module to describe the intrinsic carbon emission characteristics of a single node through a unary potential function and constrain the spatial dependence relationship of adjacent nodes through a binary potential function, so as to ensure that the output node carbon factor conforms to the physical law of the power system, fundamentally solve the problem of insufficient accuracy caused by insufficient modeling of node correlation in traditional methods, and realize the fine measurement of the indirect carbon emission of industrial users.

[0067] In one specific embodiment, node features are extracted by the GraphSAGE module, spatial consistency is enhanced by the MRF module, and finally end-to-end electric carbon factor mapping is realized. This embodiment takes the IEEE-118 node system as an example to illustrate the implementation process of the method described in the present application.

[0068] The electrical information, carbon emission information and topological structure information of the power system nodes are obtained, and the node electric carbon coupling information and the adjacency matrix are preprocessed.

[0069] The initial electrical information (source-side generator active power, load active power) and initial carbon emission information (unit direct carbon emission, power generation data) of each node in the IEEE-118 node system are collected. Among them, the generator active power is obtained in real time through the source-side monitoring system (SCADA), the load active power is indirectly calculated after being measured synchronously through the load-side μPMU device, the unit direct carbon emission is measured by the power plant, and the power generation data is derived from the power grid dispatching system.

[0070] The basic data of the IEEE-118 node system is collected, including 118 bus nodes, 54 power generation nodes and 175 transmission line models. A dual-voltage hierarchical structure of 138kV and 345kV is constructed. The system nodes are divided into three categories according to their functions:

[0071] Generator nodes: 9 nodes containing only generators, such as large-capacity thermal power generator nodes 10, 26, etc., and combined cycle gas turbine generator nodes 25, etc.

[0072] Load nodes: 64 nodes containing only loads;

[0073] Mixed nodes: 45 nodes containing both generators and loads.

[0074] According to the type of generator, the carbon emission intensity of the 54 generator nodes is classified and set:

[0075] Large-capacity thermal power generators: distributed in 8 nodes such as nodes 10, 26, etc., with a carbon emission intensity of 0.875 ;

[0076] Large-capacity combined cycle gas turbine generators: distributed in 6 nodes such as nodes 25, 49, etc., with a carbon emission intensity of 0.65 ;

[0077] Medium-capacity gas turbine generators: distributed in 35 nodes such as nodes 1, 4, etc., with a carbon emission intensity of 0.45 ;

[0078] Zero-carbon renewable energy generators: distributed in 5 nodes such as nodes 8, 27, etc., with a carbon emission intensity of 0 kg .

[0079] For the 64 non-generator nodes (load nodes or mixed nodes), their carbon emission information is represented by the second carbon emission intensity.

[0080] The input data set includes node injection power, carbon emission parameters, and node electric carbon factor true values calculated based on carbon emission flow theory. To obtain sufficient samples, Monte Carlo simulation is used to collect power flow data to build the data set: considering the uncertainty of distributed generator output, the generator output is set to randomly fluctuate in the range of the minimum operating threshold to the rated capacity, and the load level is dynamically adjusted according to the standard benchmark value of 0.5-1.5 times. After convergence verification and screening of effective samples, the system defaults the fixed carbon emission intensity parameter, and finally constructs 5000 groups of effective samples, which are divided into training set (4000 groups), validation set (500 groups), and test set (500 groups) according to the ratio of 8:1:1. Two groups of sample input data of nodes 1 to 5 are shown in Table 1.

[0081] Table 1 Input data example

[0082] ,

[0083] Using the GraphSAGE module, node feature learning and neighbor node information aggregation are realized.

[0084] The GraphSAGE module is responsible for extracting deep features of nodes and aggregating neighbor information, and the network structure mainly comprises three SAGEConv layers, forming an end-to-end feature learning process. Figure 2

[0085] After completing the three-layer GraphSAGE convolution operation, a global pooling mechanism based on a full connection layer is introduced, and a high-dimensional node feature matrix is mapped to a target prediction space through a nonlinear transformation;

[0086] The MRF module is used for processing the spatial dependence relationship between nodes in the power system, and capturing the mutual influence between nodes.

[0087] The MRF module is responsible for modeling the spatial dependence relationship between nodes, and the rationality of the node electricity-carbon factor is constrained by a potential function;

[0088] The GraphSAGE module and the MRF module are cooperatively optimized to build a GraphSAGE-MRF model, so as to realize end-to-end mapping optimization from node electricity-carbon coupling information and topological structure information to a node electricity-carbon factor.

[0089] The GraphSAGE-MRF model proposed in the application couples the GraphSAGE module and the MRF module in depth, and constructs an industrial user power indirect carbon emission monitoring method suitable for a complex power grid.

[0090] As shown in Figure 3 The GraphSAGE-MRF model constructed by the application can realize end-to-end accurate mapping of node electricity-carbon coupling information and power system topological structure information to a node electricity-carbon factor. The model design adopts a cascade structure, the front end is composed of three-layer GraphSAGE graph convolution networks, cooperates with a ReLU activation function and a Dropout regularization mechanism, realizes deep feature extraction, the middle part is dimensionally mapped through a full connection layer, and the rear end MRF layer is based on a learning parameter to calculate a unary potential and a binary potential function, and finally generates a node electricity-carbon factor prediction value. The optimization strategy fuses an MSE loss and an MRF regularization term, the latter explicitly constrains the difference between carbon emission factors of adjacent nodes in the network topology, and ensures that the prediction result meets the spatial continuity characteristics of the physical network. The structure not only fully utilizes the advantages of GraphSAGE in feature learning, but also introduces prior physical constraints through the MRF component, forming an electricity-carbon coupling prediction framework with theoretical explanation and practical value.

[0091] ​Based on the IEEE-118 node system topology characteristics and the electric carbon information characteristics, the initial parameter range is set. Then the control variable method is used to adjust each parameter one by one, and the MAE and MAPE indicators of the GraphSAGE-MRF model on the validation set are evaluated, and finally the current parameter combination is selected. The parameter setting is shown in Table 2.

[0092] Table 2 GraphSAGE-MRF model parameter setting

[0093] ,

[0094] From Table 2, it can be seen that the hidden layer dimension of the GraphSAGE module is set to 256, which can ensure the expression ability of the model and avoid overfitting; the number of convolution layers is set to 3, which can effectively capture the node multi-hop neighborhood information; the input dimension of the fully connected layer is 256 and the output dimension is 118, which ensures the effectiveness of feature dimension reduction; the Dropout ratio of 0.2 helps to improve the generalization ability of the model. In the MRF module, the number of nodes is fixed at 4 to match the key topology characteristics of the IEEE-118 system; the interaction weight matrix [118, 118] and the bias vector

[118] accurately model the spatial dependence between nodes; the regularization coefficient is set to 0.1, which balances the spatial correlation constraint between nodes and the independent prediction ability. The overall parameter setting aims to optimize the performance of the model in predicting the electric carbon factor of the IEEE-118 node system while maintaining its computational efficiency.

[0095] The initial learning rate of the model is 0.0005, and L2 regularization (weight decay coefficient 5e-5) is used to suppress overfitting; the optimizer is Adam, combined with gradient clipping (max_norm=1.0) to prevent gradient explosion and ensure training stability. The loss function is composed of mean square error (MSE) and MRF regularization term, which can effectively capture the spatial correlation between nodes.

[0096] As shown in Figure 4 , the convergence trend of the loss function during the model training process is shown, where the error curves of the training set and the validation set show obvious change rules with the increase of the training rounds. It can be observed that the model training process is smooth and there is no obvious shock, and the convergence trend of the validation set loss is consistent with that of the training set, indicating that the model has good generalization ability and does not appear overfitting phenomenon.

[0097] The electric carbon factor of each node of the power system is output and the indirect carbon emissions of industrial users are calculated.

[0098] The GraphSAGE-MRF model effectively integrates the feature extraction ability of graph neural networks and the spatial modeling advantage of Markov random fields, solving the spatial dependence problem in the prediction of carbon emission factors between nodes of the power system, and finally outputting the node electric carbon factor , the indirect carbon emission of power of an industrial user at a certain node can be calculated by combining the power consumption data of the industrial user at the node with the node carbon factor .

[0099] The effect of the application can be obtained through the following simulation experiment, specifically as follows:

[0100] Effect analysis of the power indirect carbon monitoring method:

[0101] To evaluate the effect of the proposed indirect carbon monitoring method, two evaluation indexes, MAE and MAPE, are used for evaluation. The mean absolute error (MAE) quantifies the absolute deviation between the indirect carbon emission monitoring value and the true value, and the mean absolute percentage error (MAPE) quantifies the relative error, and the calculation formula is as follows:

[0102] ,

[0103] ,

[0104] In the formula, , the number of nodes in each sample and the number of samples contained in the test set are 118 and 500, respectively; is the indirect carbon emission of power obtained based on the proposed power indirect carbon monitoring method; is the true carbon emission calculated based on the carbon emission flow theory; is a very small positive number, which is used to avoid the case that the denominator is meaningless when the indirect carbon emission of power is 0, and the value is 10 -6 .

[0105] The experimental results show that the proposed indirect carbon monitoring method of industrial users based on GraphSAGE-MRF in the application achieves significant performance on the test set: the MAE is only 0.0178 , and the MAPE is only 3.99%. From the perspective of absolute error, the MAE of 0.0178 means that for unit power consumption, the average deviation between the predicted value and the actual value of the indirect carbon emission of industrial users is less than 18 , which has reached the industrial monitoring accuracy; secondly, from the perspective of relative error, the MAPE of 3.99% indicates that the deviation ratio of the predicted value and the true value of the proposed method is very small, which is much lower than the 10% error threshold generally accepted by the industry.

[0106] To further study the error of the proposed power indirect carbon monitoring method in the application on 118 nodes, the average absolute error MAEy of node y on each sample in the test set is defined as follows:

[0107] ​ ,

[0108] like Figure 5 As shown, the mean absolute error of the 118 nodes on each sample of the test set is visualized. In the IEEE-118 node system, the mean absolute error of indirect carbon monitoring is relatively small for most nodes, with 100 nodes having a mean absolute error of less than 0.03. The accuracy requirements for indirect carbon monitoring by industrial users have been met. Specifically, the average absolute error of carbon monitoring for industrial users located at node 3 is only 0.01218. Specifically, node 111 exhibits the smallest mean absolute error in carbon monitoring, at only 0.0005. The errors of nodes 87, 108, 109, and 71 (arranged in ascending order of error) were also relatively small; while the indirect carbon monitoring average absolute error of node 69 was the largest, reaching 0.07034. Secondly, nodes 47, 49, 88, 79, and 30 (arranged in descending order of error) also showed significant errors. This spatial heterogeneity in error distribution reflects the differences in the difficulty of carbon monitoring among different nodes.

[0109] Robustness analysis of the present invention:

[0110] To verify the robustness of the proposed indirect carbon monitoring method for power systems, the topology of the IEEE-118 node system was modified. The IEEE-118 node system topology was modified in two main ways: first, by changing the local topology, i.e., modifying the connections of certain nodes, and analyzing the model's sensitivity to topology changes at different locations; second, by reducing the overall number of nodes, and analyzing the model's adaptability and accuracy under different scale power systems. Experiments were conducted in five scenarios:

[0111] Scenario 1: Without changing the topology of the IEEE-118 node system, as a baseline scenario;

[0112] Scenario 2: Connection change of high centrality nodes, that is, select nodes with a connectivity degree ≥ 5 (such as node 5, node 49, node 80, etc.), disconnect them from a neighboring node and connect them to another node to verify the performance stability of the model when the main power grid line is adjusted. In this scenario, the connection between node 49 and node 51 is disconnected and connected to node 52.

[0113] Scenario 3: Low centrality node connection change. Select nodes with a connectivity of ≤2 (such as node 10, node 73, node 87, etc.), disconnect them from a neighboring node and then connect them to another node to evaluate the model's adaptability to small-scale changes at the end of the power grid. In this scenario, disconnect node 87 from node 86 and connect it to node 85.

[0114] Scenario 4: High connectivity key node removal, that is, a key hub node (such as node 12, node 49, etc.) in the network is selected, and the node and all its connections are completely removed, to simulate the indirect carbon emission prediction ability when the main substation of the power grid fails or is maintained offline, in which scenario the connections of node 12 and node 2, 3, 7, 11, 14, 16, 117 are all removed;

[0115] Scenario 5: Low connectivity edge node removal, that is, a low connectivity node at the edge of the network (such as node 111, node 116, etc.) is selected, and the edge node is removed to verify the adaptability of the model to changes in the network edge node, to simulate the common scenario of edge users exiting the node system, in which scenario the connection between node 116 and node 68 is disconnected and node 116 is removed.

[0116] Only some node connections of the node system are modified, and the GraphSAGE-MRF model parameters and structure are not adjusted, and MAE and topological weighted mean absolute error (TWMAE) are used to evaluate the stability of the method

[0117] The method is evaluated for stability, TWMAE combines the topological characteristics of the node power system network, and the degree centrality of the node is used as a weight factor to emphasize the prediction accuracy of key nodes in the power network, and the calculation formula is as follows:

[0118] ,

[0119] In the formula, is the weight of node i, which is usually calculated based on the degree centrality of the node; is the degree (i.e., the number of connections) of node ; is the maximum node degree in the network, since the connection degree of node 49 is the largest, it is connected to 9 adjacent nodes, so the value is 9; is a weight decay parameter (default value is 0.7) for controlling the importance of highly central nodes.

[0120] Through the above five network topology change scenarios, the robustness of the indirect carbon monitoring method is verified, and the MAE and TWMAE of each scenario are obtained, as shown in Figure 6 The MAE of the baseline scenario is 0.0172 , and the TWMAE is 0.0203 After the topology changes, the model has performance differences, but always maintains low error, and the robustness of the edge and low centrality node topology change is significant. In the scene of removing the low connection degree edge node, the model MAE, TWMAE is close to the benchmark, which proves its good adaptability to the topology change of the power grid end; in the extreme scene of completely removing the key hub node 12, the model still has good prediction performance. This not only verifies the stability of the algorithm under severe topology changes, but also reveals the regulation mechanism of key nodes on carbon emission flow, providing a theoretical basis for power grid planning and carbon monitoring system optimization.

[0121] As can be seen from the above, the industrial user power indirect carbon emission monitoring method based on the GraphSAGE-MRF hybrid model is proposed, through the synergistic effect of the multi-order graph convolution of the GraphSAGE module and the potential function modeling of the MRF module, the accurate monitoring and analysis of the industrial user power indirect carbon emission under the complex power system are realized. The results show that: the GraphSAGE module can effectively extract the electric-carbon coupling characteristics of the power system nodes, and the MRF module constructed significantly optimizes the spatial consistency of the node electric-carbon factor; based on the hybrid model, the power indirect carbon emission parameters are detected, which has high precision in the multi-node complex carbon flow path scene, and it is verified that this model can effectively overcome the influence of the dynamic change of the power system topology, and has strong anti-interference performance. The algorithm proposed in the present application is easy to implement, has high real-time detection precision, and meets the requirements of fine monitoring of industrial user power indirect carbon emission, and provides reliable data support and decision basis for carbon measurement, emission reduction strategy making and low-carbon operation of the power system of industrial users.

[0122] Please refer to Figure 7 , which shows the structure block diagram of the industrial user power indirect carbon emission monitoring system based on GraphSAGE-MRF of the present application.

[0123] As Figure 7 shown, the industrial user power indirect carbon emission monitoring system 200 includes an acquisition module 210, a first output module 220, a second output module 230, and a calculation module 240.

[0124] The acquisition module 210 is configured to acquire electrical information, carbon emission information and topological structure information of a power system node, construct node electricity-carbon coupling information based on the electrical information and the carbon emission information, and construct an adjacency matrix based on the topological structure information; the first output module 220 is configured to construct a GraphSAGE-MRF model, input the node electricity-carbon coupling information and the adjacency matrix into a GraphSAGE module in the GraphSAGE-MRF model, perform node feature learning and neighbor node information aggregation through multi-layer graph convolution operation, and output graph-level electricity-carbon coupling features; the second output module 230 is configured to input the graph-level electricity-carbon coupling features into an MRF module in the GraphSAGE-MRF model, model node self electricity-carbon characteristics through a unary potential function, model spatial dependency relationships between adjacent nodes through a binary potential function, and output an indirect electricity-carbon factor of a node where an industrial user is located; and the calculation module 240 is configured to calculate a power indirect carbon emission amount based on the indirect electricity-carbon factor of the node where the industrial user is located and power consumption of the industrial user.

[0125] It should be understood that Figure 7 the modules described in the above Figure 1 correspond to the respective steps in the methods described with reference to Figure 7 the above description of the operations and features and the corresponding technical effects apply equally to the modules in

[0126] In some other embodiments, the present application also provides a computer readable storage medium having a computer program stored thereon, wherein the program instructions are executed by a processor to cause the processor to perform the GraphSAGE-MRF-based industrial user power indirect carbon emission monitoring method in any of the above method embodiments.

[0127] As an implementation form, the computer readable storage medium of the present application stores computer executable instructions, and the computer executable instructions are configured to:

[0128] acquire electrical information, carbon emission information and topological structure information of a power system node, construct node electricity-carbon coupling information based on the electrical information and the carbon emission information, and construct an adjacency matrix based on the topological structure information;

[0129] construct a GraphSAGE-MRF model, input the node electricity-carbon coupling information and the adjacency matrix into a GraphSAGE module in the GraphSAGE-MRF model, perform node feature learning and neighbor node information aggregation through multi-layer graph convolution operation, and output graph-level electricity-carbon coupling features;

[0130] The MRF module in the GraphSAGE-MRF model is input with the graph-level electrical carbon coupling feature, a unary potential function is used to model the electrical carbon characteristics of a node itself, a binary potential function is used to model the spatial dependence relationship between adjacent nodes, and an indirect electrical carbon factor of a node where the industrial user is located is output.

[0131] Based on the indirect electrical carbon factor of the node where the industrial user is located and the power consumption of the industrial user, the power indirect carbon emission is calculated.

[0132] The computer-readable storage medium can include a program storage area and a data storage area, wherein the program storage area can store an operating system and application programs required by at least one function; and the data storage area can store data created according to the use of the GraphSAGE-MRF-based industrial user power indirect carbon emission monitoring system, and the like. In addition, the computer-readable storage medium can include a high-speed random access memory, and can also include a memory such as at least one magnetic disk storage device, a flash memory device, or other non-volatile solid-state memory device. In some embodiments, the computer-readable storage medium can optionally include a memory that is remotely arranged relative to the processor, and these remote memories can be connected to the GraphSAGE-MRF-based industrial user power indirect carbon emission monitoring system through a network. Examples of the above network include but are not limited to the Internet, an intranet, a local area network, a mobile communication network, and a combination thereof.

[0133] Figure 8 is a structural schematic diagram of an electronic device provided by an embodiment of the present application, as Figure 8 indicated, the device includes a processor 310 and a memory 320. The electronic device can also include an input device 330 and an output device 340. The processor 310, the memory 320, the input device 330, and the output device 340 can be connected by a bus or other means, Figure 8 for example, by a bus connection. The memory 320 is the computer-readable storage medium described above. The processor 310 performs various functional applications and data processing of the server by running the non-volatile software programs, instructions, and modules stored in the memory 320, that is, implements the GraphSAGE-MRF-based industrial user power indirect carbon emission monitoring method of the above method embodiment. The input device 330 can receive input digital or character information, and generate key signal inputs related to user settings and functional control of the GraphSAGE-MRF-based industrial user power indirect carbon emission monitoring system. The output device 340 can include a display device such as a display screen.

[0134] The above electronic device can execute the method provided by the embodiments of the present application, and has the corresponding functional modules and beneficial effects of executing the method. Technical details not described in detail in the present embodiment can be referred to the method provided by the embodiments of the present application.

[0135] As an implementation form, the above electronic device is applied to a GraphSAGE-MRF-based industrial user power indirect carbon emission monitoring system, and is used for a client, comprising: at least one processor; and a memory connected with the at least one processor in communication; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to:

[0136] obtain electrical information, carbon emission information and topological structure information of a power system node, construct node electricity-carbon coupling information based on the electrical information and the carbon emission information, and construct an adjacency matrix based on the topological structure information;

[0137] construct a GraphSAGE-MRF model, input the node electricity-carbon coupling information and the adjacency matrix into a GraphSAGE module in the GraphSAGE-MRF model, perform node feature learning and neighbor node information aggregation through multi-layer graph convolution operation, and output to obtain graph-level electricity-carbon coupling features;

[0138] input the graph-level electricity-carbon coupling features into an MRF module in the GraphSAGE-MRF model, model the node's own electricity-carbon characteristics through a unary potential function, model the spatial dependency relationship between adjacent nodes through a binary potential function, and output to obtain an indirect electricity-carbon factor of a node where an industrial user is located;

[0139] based on the indirect electricity-carbon factor of the node where the industrial user is located and the electricity consumption of the industrial user, calculate to obtain a power indirect carbon emission amount.

[0140] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be realized by means of software and a necessary general hardware platform, and of course, it can also be realized by hardware. Based on such understanding, the above technical solutions essentially or say the part that contributes to the prior art can be embodied in the form of a software product, which can be stored in a computer readable storage medium such as ROM / RAM, magnetic disk, optical disk, etc., and includes a plurality of instructions to make a computer device (which can be a personal computer, a server, or a network device, etc.) execute the methods of each embodiment or some parts of the embodiment.

[0141] It should be pointed out finally that the above embodiments are only used to illustrate the technical solutions of the present application, but not to limit the same; and although the present application has been described in detail with reference to the foregoing embodiments, it should be appreciated by those skilled in the art that the technical solutions recorded in the foregoing embodiments can be modified, or some technical features thereof can be replaced equivalently; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for monitoring indirect carbon emissions from industrial user electricity based on GraphSAGE-MRF, characterized in that, include: The electrical information, carbon emission information, and topology information of power system nodes are obtained. Based on the electrical information and carbon emission information, node electrical-carbon coupling information is constructed, and based on the topology information, an adjacency matrix is ​​constructed. The carbon emission information includes the first carbon emission intensity of generator nodes and the second carbon emission intensity of non-generator nodes. The expression for the first carbon emission intensity is: ， In the formula, The highest carbon emission intensity, This represents the direct carbon emissions measured in real time by the generator unit at the source-side node during time period t. The power generation of the generator unit at the source node during time period t. This refers to the operating time of the generator set at the source-side node. The expression for the second carbon emission intensity is: ， In the formula, The second highest carbon emission intensity, The average carbon flow rate under load. The average active power of the load. This represents the total number of generator set nodes. This represents the total number of load nodes. The active power of the generator set. The active power of the load; A GraphSAGE-MRF model is constructed, and the node electro-carbon coupling information and the adjacency matrix are input into the GraphSAGE module of the GraphSAGE-MRF model. Node feature learning and neighbor node information aggregation are performed through multi-layer graph convolution operations, and graph-level electro-carbon coupling features are output. The graph-level electrocarbon coupling features are input into the MRF module of the GraphSAGE-MRF model. The node's own electrocarbon characteristics are modeled using a univariate potential function, and the spatial dependencies between adjacent nodes are modeled using a binary potential function. The indirect electrocarbon factor of the node where the industrial user is located is output. The expression for the univariate potential function is: ， In the formula, This is a univariate potential function, representing the intrinsic properties of a single node. This is the result after mapping the fully connected layer, representing the feature representation of each node. These are node bias parameters used to adjust the basic features of each node; The expression for the binary potential function is: ， In the formula, Let be a binary potential function, representing the spatial dependencies and interactions between nodes. This is the interaction weight matrix, used to represent the strength of the mutual influence between nodes. This is an element-wise dot product operation used to ensure that interactions only occur between adjacent nodes. For the power system adjacency matrix; The indirect carbon emissions from electricity are calculated based on the indirect carbon factor of the node where the industrial user is located and the electricity consumption of the industrial user.

2. The method for monitoring indirect carbon emissions from industrial user electricity based on GraphSAGE-MRF according to claim 1, characterized in that, The electrical information is characterized by node-injected power, wherein: If node i is a generator node, then the injected power is the active power of the generator set. ; If node i is a load node, then the injected power is the negative of the load's active power. ; If node i is a hybrid node, then the injected power is If node i is an intermediate node, then the injected power is 0.

3. The method for monitoring indirect carbon emissions from industrial user electricity based on GraphSAGE-MRF according to claim 1, characterized in that, The GraphSAGE module contains three cascaded SAGEConv graph convolutional layers, which fuse node feature learning with neighbor node information through a hierarchical message passing mechanism. Specifically, for node i at the th... The SAGEConv convolutional layer is represented as follows: ， In the formula, For nodes Neighborhood aggregation characteristics For nodes In the Layer feature representation, For nodes The neighborhood set, For nodes and nodes The normalization coefficients between them For the first Learnable weight matrix for layer feature aggregation This is the activation function.

4. The method for monitoring indirect carbon emissions from industrial user electricity based on GraphSAGE-MRF according to claim 3, characterized in that, The GraphSAGE module further includes a global pooling layer. After processing by three cascaded SAGEConv graph convolutional layers, node features are aggregated based on the global pooling mechanism of the global pooling layer to form a graph-level electro-carbon coupled feature representation. The expression for the global pooling layer is as follows: ， In the formula, This represents the graph-level feature representation output by the global pooling layer. This is the pooling operation function. In the first Layer nodes Feature representation, Let be the set of nodes in the graph.

5. The method for monitoring indirect carbon emissions from industrial user electricity based on GraphSAGE-MRF according to claim 1, characterized in that, in, The expression for the indirect electrocarbon factor is: ， In the formula, It is an indirect carbon factor.

6. A monitoring system for indirect carbon emissions from industrial users' electricity consumption based on GraphSAGE-MRF, characterized in that, include: The acquisition module is configured to acquire electrical information, carbon emission information and topology information of power system nodes, construct node electrical-carbon coupling information based on the electrical information and carbon emission information, and construct an adjacency matrix based on the topology information. The carbon emission information includes the first carbon emission intensity of generator nodes and the second carbon emission intensity of non-generator nodes. The expression for the first carbon emission intensity is: ， In the formula, The highest carbon emission intensity, This represents the direct carbon emissions measured in real time by the generator unit at the source-side node during time period t. The power generation of the generator unit at the source node during time period t. This refers to the operating time of the generator set at the source-side node. The expression for the second carbon emission intensity is: ， In the formula, The second highest carbon emission intensity, The average carbon flow rate under load. The average active power of the load. This represents the total number of generator set nodes. This represents the total number of load nodes. The active power of the generator set. The active power of the load; The first output module is configured to construct a GraphSAGE-MRF model and input the node electro-carbon coupling information and the adjacency matrix into the GraphSAGE module of the GraphSAGE-MRF model. The module performs node feature learning and neighbor node information aggregation through multi-layer graph convolution operations and outputs graph-level electro-carbon coupling features. The second output module is configured to input the graph-level electrocarbon coupling features into the MRF module of the GraphSAGE-MRF model, model the node's own electrocarbon characteristics through a univariate potential function, model the spatial dependency between adjacent nodes through a binary potential function, and output the indirect electrocarbon factor of the node where the industrial user is located. The expression of the univariate potential function is: ， In the formula, This is a univariate potential function, representing the intrinsic properties of a single node. This is the result after mapping the fully connected layer, representing the feature representation of each node. These are node bias parameters used to adjust the basic features of each node; The expression for the binary potential function is: ， In the formula, Let be a binary potential function, representing the spatial dependencies and interactions between nodes. This is the interaction weight matrix, used to represent the strength of the mutual influence between nodes. This is an element-wise dot product operation used to ensure that interactions only occur between adjacent nodes. For the power system adjacency matrix; The calculation module is configured to calculate the indirect carbon emissions of electricity based on the indirect carbon factor of the node where the industrial user is located and the electricity consumption of the industrial user.

7. An electronic device, characterized in that, include: At least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor to enable the at least one processor to perform the method according to any one of claims 1 to 5.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the method described in any one of claims 1 to 5.

Images