A power system carbon index joint evaluation method and device based on edge perception graph neural network

Abstract

The application relates to a power system carbon index joint evaluation method and device based on an edge perception graph neural network, and belongs to the field of power systems, aiming to solve the problems of low evaluation efficiency, weak coordination between node and line carbon indexes, low integrated output degree and poor consistency of multi-index parallel output of existing carbon index evaluation. A running condition sample set is constructed, five types of carbon index labels are generated through power flow calculation and carbon emission flow analysis; the power system is abstracted into a graph structure, node and edge features are extracted to construct a training data set; an edge perception graph neural network is constructed to fuse the node and edge features; carbon indexes are output through double-channel decoding; the model is trained using a joint loss function to realize rapid carbon index evaluation. The application realizes integrated joint evaluation of node and line carbon indexes, improves evaluation efficiency and output consistency, solves the problems of poor adaptability and slow response of traditional evaluation, and can support high-frequency near-real-time calls.

Description

Technical Field

[0001] This invention belongs to the field of power systems, specifically relating to a method and apparatus for joint evaluation of carbon indicators in power systems based on edge-aware graph neural networks. Background Technology

[0002] Currently, the main method for assessing carbon indicators in power systems is to calculate operating conditions sequentially. This involves obtaining node operating states, line transmission parameters, and corresponding carbon attribute indicators through power flow calculations and carbon emission flow analysis, which are then used for low-carbon operation of the power grid and carbon responsibility accounting. This type of assessment requires repeated numerical solutions and indicator calculations for different operating conditions, and the overall process relies on traversing operating conditions and performing step-by-step calculations.

[0003] Traditional carbon index assessment methods suffer from low overall execution efficiency and poor adaptability to large-scale systems and diverse operating conditions. As the number of operating conditions increases and the system scale expands, the overall processing time tends to rise, and the response speed is low when simultaneously outputting multiple types of carbon indices, making it difficult to support high-frequency, near real-time assessment requirements.

[0004] Existing technologies have the following drawbacks: existing assessment methods have weak collaborative characterization of node and line carbon indicators, low degree of integrated output, and poor consistency in parallel output of multiple indicators. These are the shortcomings of existing technologies.

[0005] In view of this, it is very necessary to provide a method and apparatus for joint evaluation of carbon indicators of power systems based on edge-aware graph neural networks to solve the above-mentioned defects in the prior art. Summary of the Invention

[0006] To address the technical problems of existing assessment methods, such as weak collaborative representation of carbon indicators at nodes and lines, low degree of integrated output, and poor consistency of parallel output of multiple indicators, this invention provides a joint assessment method and device for carbon indicators of power systems based on edge-aware graph neural networks, in order to solve the above-mentioned technical problems.

[0007] In a first aspect, the present invention provides a joint evaluation method for carbon indicators of power systems based on edge-aware graph neural networks, comprising: Step S1: Construct a basic operating condition sample of the power system. Based on the basic operating condition sample, apply disturbances to the load side and the generation side to construct an operating condition sample set. Perform AC power flow calculation and carbon emission flow analysis on the operating condition sample set to obtain sample labels. The sample labels include node carbon index labels and line carbon index labels. Select the basic network model of the target power system, and construct a sample of basic operating conditions of the power system based on the basic network model; Based on the basic operating condition samples, a set of operating condition samples is constructed by applying perturbations to the load side and the generation side. Specifically, applying a perturbation to the load side of the basic operating conditions yields the s-th operating condition sample. The active power and reactive power of node i are expressed as follows: ; ; in, , Let be the active power and reactive power of the load at node i under the s-th operating condition sample, respectively. , These represent the active and reactive loads of node i under the baseline operating conditions. Let be the load disturbance coefficient of node i in the s-th operating condition sample; The load disturbance coefficient of node i in the s-th operating condition sample satisfies: ; in, This is the lower limit of the load disturbance range. This is the upper limit of the load disturbance range; A disturbance is applied to the load side of the basic operating condition to obtain the s-th operating condition sample. The active power and reactive power of node i are expressed as follows: ; ; in, , Let be the active power and reactive power generated by node i under the s-th operating condition sample, respectively. , These represent the active power load and reactive power load of node i under the baseline operating conditions, respectively. Let be the power generation disturbance coefficient of node i in the s-th operating condition sample; In the s-th operating condition sample, the generation disturbance coefficient of node i satisfies: ; in, This is the lower limit of the power generation disturbance range. This represents the upper limit of the power generation disturbance range; AC power flow calculations are performed on the operating condition samples after applying disturbances to the load side and the generation side based on the basic operating condition samples. The operating condition sample set is represented as follows: ,in, This is the Mth operating condition sample, where M is the total number of operating condition samples. , For the m-th operating condition sample, Let be the disturbance coefficient of the i-th load node under the m-th operating condition sample. Let g be the disturbance coefficient of the g-th power generation node under the m-th operating condition sample; The node operating status includes node voltage amplitude, node voltage phase angle, node active power injection, and node reactive power injection. The power at both ends of the line includes the power at the transmitting end and the power at the receiving end. Line active power loss is the difference between the power at the transmitting end of the line and the power at the receiving end of the line. The five categories of results—node carbon intensity, node carbon emissions, line carbon inflow, line carbon loss, and line carbon outflow—are represented as follows: ; in, For node carbon strength, For node carbon emissions, For carbon inflow into the line, For line carbon loss, Carbon flows out of the circuit.

[0008] The generation of sample labels specifically includes: Under operating condition sample s, node carbon intensity is calculated for each node to generate node carbon intensity labels for operating condition sample s. The node carbon intensity of node i under operating condition sample s is represented as: ; in, Let i be the nodal carbon intensity of node i under the s-th operating condition sample. Let be the carbon flow corresponding to the local power generation of node i under the s-th operating condition sample. Let i be the set of upstream nodes of node i. Let be the amount of carbon flowing from node j into node i after transmission through the line in the s-th operating condition sample. Let be the active power generated by node i. Let be the active power at the line receiving end of the s-th operating condition sample; By calculating the node carbon emissions for each node, the node carbon emissions under operating condition sample s are generated, where the node carbon emissions of node i under the s-th operating condition sample are represented as: ; in, Let be the node carbon emissions of node i in the s-th operating condition sample. Let i be the nodal carbon intensity of node i under the s-th operating condition sample. Let be the load active power of node i under the s-th operating condition sample; By calculating the carbon flow entering the line at each node, the carbon inflow into the line under operating condition sample s is generated. In the s-th operating condition sample, the carbon inflow into the line from node i is calculated. carbon flow Represented as: ; in, Let i be the nodal carbon intensity of node i under the s-th operating condition sample. For the s-th operating condition sample, the line The transmitting power; Line carbon loss For the s-th operating condition sample, the line The difference between carbon inflow and carbon outflow from the line is expressed as: ;in, For the s-th operating condition sample, the line carbon outflow is the carbon flow rate output by the receiving end after transmission. Node carbon intensity, node carbon emissions, line carbon inflow, line carbon loss, and line carbon outflow are represented as sample labels corresponding to operating condition sample s. , is represented as: ; in, This is the node carbon intensity label vector under the operating condition sample s. This is the node carbon emission label vector under operating condition sample s. This is the carbon inflow label vector of the line under operating condition sample s. This is the line carbon loss label vector under operating condition sample s. This is the line carbon outflow label vector under the operating condition sample s.

[0009] In this step, disturbances are applied to the load side and the power generation side, and AC power flow calculations are completed. At the same time, five types of carbon index labels are calculated according to the carbon attribute transmission law, the operating condition samples and label generation logic are standardized, and then monitoring data is obtained to ensure the stability of the data.

[0010] Step S2: Represent the power system as a graph structure, extract node features and edge features from the sample set to form graph input samples, and combine them with sample labels to obtain the training dataset; The graphical structure of a power system is represented as follows: ; Where V represents the set of nodes, corresponding to the power system bus, and is represented as: ,in, Indicates the first 1 node Let E represent the number of bus nodes in the power system; E represents the set of edges, corresponding to the power system lines, expressed as: ,in, Indicates the first One line, This refers to the number of power system lines. For the s-th operating condition sample, the corresponding graphical input sample is represented as: ,in, Let be the node feature matrix of the s-th operating condition sample. The edges are connected. Let be the edge feature matrix of the s-th operating condition sample; The node feature matrix of the s-th operating condition sample is represented as: ,in, Let i be the feature vector of node i in the s-th operating condition sample, and be expressed as: ; in, The active power of node load at node i under operating condition sample s. The reactive power of node i under operating condition sample s is the node load. The active power generated by node i under operating condition sample s. The reactive power generated by node i under operating condition sample s. The node voltage amplitude of node i under operating condition sample s. The node voltage phase angle of node i under operating condition sample s. To inject net active power into node i under operating condition sample s. To inject net reactive power into node i under operating condition sample s. Encode the node type; The edge feature matrix of the s-th operating condition sample is represented as: ;in, Let be the number of directed edges. For any directed edge under the s-th operating condition sample The constructed edge feature vector is represented as: ; in, These represent the line parameters, including line impedance, line reactance, and line susceptance. Let these represent the active power at the transmitting end and the receiving end of the line, respectively, for the s-th operating condition sample. This represents the line active power loss for the s-th operating condition sample. Let represent the voltage difference and phase angle difference between the two ends of the line for the s-th operating condition sample, respectively.

[0011] Input sample based on the graph of the s-th operating condition sample , with sample label The training dataset is obtained by combining the data, and is represented as follows: ; By combining all operational condition samples, an offline training dataset is constructed, represented as: .

[0012] In this step, the power system is abstracted into a graph structure and the multidimensional features of nodes and edges are fully extracted. The graph input samples are paired with labels to construct an offline training dataset that adapts to the characteristics of the power grid topology and can provide standardized and unified input information for the model.

[0013] Step S3: Construct an edge-aware graph neural network and fuse node features and edge features through edge-aware message passing; The construction of an edge-aware graph neural network is represented as follows: ; in, The parameter is Graph neural network model, Let be the node feature matrix of the s-th operating condition sample. The edges are connected. Let be the edge feature matrix of the s-th operating condition sample. This represents the predicted output of the graph neural network model under operating condition sample s; The node feature matrix is ​​processed by node encoding to obtain the node hidden representation matrix of the s-th operating condition sample. , is represented as: ; in, Represents the node encoding function; Edge encoding is used to process the edge feature matrix to obtain the edge hidden representation matrix of the s-th operating condition sample. , is represented as: ; in, Represents the edge encoding function; Edge-aware message-passing fusion is used to process the node hidden representation matrix and the edge hidden representation matrix. The update process of node i in the s-th running condition sample is as follows: ; in, Describes the set of neighboring nodes of node i. This represents the message constructor, used for edge-aware message passing fusion. Represents aggregate functions, Represents a non-linear activation function. Let i be the final hidden representation of the l-th node i in the s-th operating condition sample. Let be the final hidden representation of the l-th node j of the s-th operating condition sample. Let be the edge hidden representation matrix for the s-th operating condition sample. Let be the weight matrix of the l-th layer.

[0014] In this step, node and edge features are encoded separately, edge-aware message passing is performed, and topology and line feature information are fused, which can improve the model's ability to learn the coupling relationship of the power grid.

[0015] Step S4: Perform node channel decoding and line channel decoding on the fused node features and edge features respectively, and output node carbon index and line carbon index; Node carbon indicators include node carbon intensity and node carbon emissions; line carbon indicators include line carbon inflow, line carbon loss, and line carbon outflow. The node carbon intensity prediction for the s-th operating condition sample is obtained through node channel decoding. , is represented as: ; in, This is the node hidden representation matrix after edge-aware message passing fusion. Represents the nodal carbon strength readout function; The node carbon emission prediction for the s-th operating condition sample is obtained through node channel decoding. , is represented as: ; in, For node carbon emission readout function; The predicted carbon inflow of the line is obtained by decoding the line channel for the s-th operating condition sample. Line carbon loss prediction Carbon outflow prediction of the line , is represented as: ; in, For line-side readout function, For the line fusion representation of the s-th operating condition sample, Let be the final hidden representation matrix of the starting node of the s-th operating condition sample. Let be the final hidden representation matrix of the termination node of the s-th operating condition sample. Let be the edge hidden representation matrix for the s-th operating condition sample. Let be the original edge feature matrix of the s-th operating condition sample; In this step, dual-channel decoding of nodes and lines is used to output the corresponding carbon indicators, and the prediction of multiple types of indicators is completed simultaneously, which can improve the completeness and correlation of indicator output.

[0016] Step S5: Train the edge-aware graph neural network using a joint loss function to perform online rapid evaluation of carbon indicators.

[0017] The joint loss function includes the node-side loss function and the line-side loss function; The node-side loss function is expressed as: ; in, Let be the true value of the carbon index for node i in the s-th operating condition sample. Let be the predicted value of node i for the s-th operating condition sample. The number of nodes; The line-side loss function is expressed as: ; in, This represents the actual line-side value under the s-th operating condition sample. This is the predicted value for the line side under the s-th operating condition sample. For L1 loss function, Let be the mean squared error, and α be the weighting coefficient of the mean squared error term.

[0018] In this step, a joint loss function that includes both node-side and line-side loss functions is used to train the model, constraining the error level of the prediction results. This can improve the accuracy of the model output and ensure the consistency and reliability of the parallel output of multiple indicators.

[0019] Secondly, the technical solution of the present invention also provides a joint evaluation device for carbon indicators of power systems based on edge-aware graph neural networks, including a sample construction module, a graph modeling module, a network construction module, an indicator decoding module, and a training and evaluation module; The sample construction module is used to construct a basic operating condition sample of the power system. Based on the basic operating condition sample, disturbances are applied to the load side and the generation side to construct an operating condition sample set. AC power flow calculation and carbon emission flow analysis are performed on the operating condition sample set to obtain sample labels. The sample labels include node carbon index labels and line carbon indexes. The graph modeling module is used to represent the power system as a graph structure. It extracts node features and edge features from the sample set to form graph input samples, which are then combined with sample labels to obtain the training dataset. The network building module is used to build edge-aware graph neural networks, fusing node features and edge features through edge-aware message passing; The network construction module also includes node coding unit, edge coding unit, and edge-aware message passing unit; Node encoding unit, used to map node features to node hidden representations; Edge coding unit is used to map edge features to edge hidden representations; The edge-aware message passing unit is used to update the hidden representation of a node by combining the node's neighborhood relationship with the edge's hidden representation, and to fuse node features and edge features.

[0020] The index decoding module performs node channel decoding and line channel decoding on the fused node features and edge features respectively, and outputs node carbon index and line carbon index. The training and evaluation module uses a joint loss function to train a side-aware graph neural network and performs online rapid evaluation of carbon indicators.

[0021] The beneficial effects of this invention are as follows: This invention provides a method and apparatus for joint evaluation of carbon indicators in power systems based on a side-aware graph neural network. Through a complete process of operating condition construction and label generation, graph structure modeling, side-aware feature fusion, dual-channel decoding, and joint loss training, it achieves integrated joint evaluation of carbon indicators for power system nodes and lines, improving the collaborative representation effect of carbon indicators and optimizing the ability to characterize carbon attributes throughout the entire process and the consistency level of multiple indicator outputs. Samples are constructed and five types of carbon indicator labels are generated through load and generation-side disturbances, providing the necessary full-condition monitoring data foundation for evaluation and improving data integrity. Graph structure modeling and feature extraction are used to construct a training dataset, adapting to power system characteristics and standardizing the model input format. Feature fusion is achieved through a side-aware graph neural network, mining grid coupling relationships and improving the model's ability to learn and utilize power system information. Dual-channel decoding output from nodes and lines reflects the distribution of carbon indicators, improving the correlation of parallel outputs of multiple indicators. Training is completed through a joint loss function and used for rapid online evaluation, improving the stability of the evaluation process and output accuracy, and adapting to practical application scenarios.

[0022] Furthermore, the design principle of this invention is reliable, the structure is simple, and it has a very wide range of application prospects. Attached Figure Description

[0023] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0024] Figure 1 This is a flowchart of a joint evaluation method for carbon indicators of power systems based on edge-aware graph neural networks provided by the present invention.

[0025] Figure 2 This is a schematic diagram of a power system carbon index joint evaluation device based on edge-aware graph neural network provided by the present invention. Detailed Implementation

[0026] To enable those skilled in the art to better understand the technical solutions of this invention, the technical solutions of the embodiments of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this invention, and not all embodiments. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of this invention.

[0027] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used herein in the description of the invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention.

[0028] Example 1: like Figure 1 As shown, this embodiment of the invention provides a joint evaluation method for carbon indicators of power systems based on edge-aware graph neural networks, including the following steps: Step S1: Construct a basic operating condition sample of the power system. Based on the basic operating condition sample, apply disturbances to the load side and the generation side to construct an operating condition sample set. Perform AC power flow calculation and carbon emission flow analysis on the operating condition sample set to obtain sample labels. The sample labels include node carbon index labels and line carbon index labels. Step S2: Represent the power system as a graph structure, extract node features and edge features from the sample set to form graph input samples, and combine them with sample labels to obtain the training dataset; Step S3: Construct an edge-aware graph neural network and fuse node features and edge features through edge-aware message passing; Step S4: Perform node channel decoding and line channel decoding on the fused node features and edge features respectively, and output node carbon index and line carbon index; Step S5: Train the edge-aware graph neural network using a joint loss function to perform online rapid evaluation of carbon indicators.

[0029] In step S1, the core task is to construct a power system operating condition sample set to cover different operating states, obtain complete node and line electrical operating information through AC power flow calculation, and generate five types of sample labels according to the physical transmission law of carbon emission flow: node carbon intensity, node carbon emission, line carbon inflow, line carbon loss, and line carbon outflow.

[0030] A basic network model of the target power system is selected. The basic network model includes bus topology information, line parameter information, generator operation information and load distribution information. Based on the basic network model, a sample of basic operating conditions of the power system can be constructed.

[0031] Based on the basic operating conditions, continuous perturbations within a preset range are applied to both the load side and the generation side, enabling the model to learn the variation patterns of carbon indicators across the entire operating range. The perturbation applied to the load side of the basic operating conditions yields the s-th operating condition sample, where the active power and reactive power of node i are represented as follows: ; ; in, , Let be the active power and reactive power of the load at node i under the s-th operating condition sample, respectively. , These represent the active and reactive loads of node i under the baseline operating conditions. Let be the load disturbance coefficient of node i in the s-th operating condition sample; The load disturbance coefficient of node i in the s-th operating condition sample satisfies: ; in, This is the lower limit of the load disturbance range. This is the upper limit of the load disturbance range; A disturbance is applied to the load side of the basic operating condition to obtain the s-th operating condition sample, where the active power and reactive power of node i are expressed as follows: ; ; in, , Let be the active power and reactive power generated by node i under the s-th operating condition sample, respectively. , These represent the active power load and reactive power load of node i under the baseline operating conditions, respectively. Let be the power generation disturbance coefficient of node i in the s-th operating condition sample; In the s-th operating condition sample, the generation disturbance coefficient of node i satisfies: ; in, This is the lower limit of the power generation disturbance range. This represents the upper limit of the power generation disturbance range; In some embodiments, the load disturbance range and the generation disturbance range may be the same range, for example: or The above ranges correspond to normal fluctuation scenarios and relatively strong fluctuation scenarios, respectively.

[0032] After completing the disturbance settings, AC power flow calculations are performed on the operating condition samples. For the s-th operating condition sample, the active power balance and reactive power balance of node i are expressed as follows: ; in, This refers to the number of bus nodes in the power system. For the s-th operating condition sample, the line Active power loss on For the s-th operating condition sample, the line The reactive power loss on the line is represented by E, which represents the set of edges and corresponds to the power system lines.

[0033] Therefore, the constructed operating condition sample set is represented as: ,in, This is the Mth operating condition sample, where M is the total number of operating condition samples. The node operating status includes node voltage amplitude, node voltage phase angle, node active power injection, and node reactive power injection. The power at both ends of the line includes the power at the transmitting end and the power at the receiving end. Under the s-th operating condition sample, the line active power loss Power at the transmitting end of the line With line receiving power The difference is expressed as: ; The five categories of results—node carbon intensity, node carbon emissions, line carbon inflow, line carbon loss, and line carbon outflow—are represented as follows: ; in, For node carbon strength, For node carbon emissions, For carbon inflow into the line, For line carbon loss, Carbon flows out of the circuit.

[0034] The generation of the five types of sample labels specifically includes: Under operating condition sample s, node carbon intensity is calculated for each node to generate node carbon intensity labels for operating condition sample s. The node carbon intensity of node i under operating condition sample s is represented as: ; in, Let i be the nodal carbon intensity of node i under the s-th operating condition sample. Let be the carbon flow corresponding to the local power generation of node i under the s-th operating condition sample. Let i be the set of upstream nodes of node i. Let be the amount of carbon flowing from node j into node i after transmission through the line in the s-th operating condition sample. Let be the active power generated by node i. Let be the active power at the line receiving end of the s-th operating condition sample; By calculating the node carbon emissions for each node, the node carbon emissions under operating condition sample s are generated, where the node carbon emissions of node i under the s-th operating condition sample are represented as: ; in, Let be the node carbon emissions of node i in the s-th operating condition sample. Let i be the nodal carbon intensity of node i under the s-th operating condition sample. Let be the load active power of node i under the s-th operating condition sample; By calculating the carbon flow entering the line at each node, the carbon inflow into the line under operating condition sample s is generated. In the s-th operating condition sample, the carbon inflow into the line from node i is calculated. carbon flow Represented as: ; in, Let i be the nodal carbon intensity of node i under the s-th operating condition sample. For the s-th operating condition sample, the line The transmitting power; Line carbon loss For the s-th operating condition sample, the line The difference between carbon inflow and carbon outflow from the line is expressed as: ;in, For the s-th operating condition sample, the line carbon outflow is the carbon flow rate output by the receiving end after transmission. The relationship between carbon inflow, carbon outflow, and carbon loss in the transmission lines is as follows: ; The above five categories of indicators are organized uniformly to form sample labels corresponding to a single operating condition. The sample labels include nodal carbon intensity, nodal carbon emissions, line carbon inflow, line carbon loss, and line carbon outflow, which are represented as the sample labels corresponding to operating condition sample s. , represented as: ; in, This is the node carbon intensity label vector under the operating condition sample s. This is the node carbon emission label vector under operating condition sample s. This is the carbon inflow label vector of the line under operating condition sample s. This is the line carbon loss label vector under operating condition sample s. This is the line carbon outflow label vector under the operating condition sample s.

[0035] Thus far, step S1 has constructed a monitoring data set with electrical and carbon attribute information through two-sided perturbation construction, AC power flow calculation, and sample tag generation, ensuring that the tags have physical consistency and cover the entire operating range.

[0036] In step S2, the core task is to abstract the power system into a graph structure that conforms to the input format of a graph neural network, extract node features and edge features that can reflect the operating state and physical characteristics, form standardized graph input samples, and pair the graph input samples with the sample labels obtained in step S1 to construct an offline training dataset that can be directly used for model training.

[0037] The graph structure of a power system, where buses correspond to nodes and transmission lines to edges, can be represented in a way that fully preserves the topological connections and electrical coupling characteristics, as follows: ; Where V represents the set of nodes, corresponding to the power system bus, and is represented as: ,in, Indicates the first 1 node Let E represent the number of bus nodes in the power system; E represents the set of edges, corresponding to the power system lines, expressed as: ,in, Indicates the first One line, This refers to the number of power system lines. For the s-th operating condition sample, the corresponding graphical input sample is represented as: ,in, Let be the node feature matrix of the s-th operating condition sample. This represents the edge connection relationship. Let be the edge feature matrix of the s-th operating condition sample; Node features and edge features can be extracted to form graph input samples. Node features are used to characterize the node's operating status and attribute information, covering electrical quantities and type attributes. The node feature vector includes node load active power, node load reactive power, node generator active power, node generator reactive power, node voltage amplitude, node voltage phase angle, node net injected active power, node net injected reactive power, and node type code. The node feature matrix of the s-th operating condition sample is specifically represented as follows: ,in, Let the node feature vector of the s-th operating condition sample be represented as: ; in, The active power of node load at node i under operating condition sample s. The reactive power of node i under operating condition sample s is the node load. The active power generated by node i under operating condition sample s. The reactive power generated by node i under operating condition sample s. The node voltage amplitude of node i under operating condition sample s. The node voltage phase angle of node i under operating condition sample s. To inject net active power into node i under operating condition sample s. To inject net reactive power into node i under operating condition sample s. Node type encoding is used to distinguish different types such as generator nodes, load nodes, and tie nodes. The net active power injected into a node reflects the node's power balance. The net active power injected into node i under operating condition sample s is written as: ; The net reactive power injected into node i under operating condition sample s is written as: ; Edge features are used to characterize the static parameters and dynamic operating status of the line, including both physical parameters and real-time operational quantities. The edge feature matrix is ​​constructed by stacking the feature vectors of all directed edges in sequence. The edge feature vectors include line parameters, line transmitting power, line receiving power, line active power loss, voltage difference between the two ends of the line, and phase angle difference between the two ends of the line. The edge feature matrix of the s-th operating condition sample is represented as: ;in, Let be the number of directed edges. For any directed edge under the s-th operating condition sample The constructed edge feature vector is represented as: ; in, These represent the line parameters, including line impedance, line reactance, and line susceptance. Let these represent the active power at the transmitting end and the receiving end of the line, respectively, for the s-th operating condition sample. This represents the line active power loss for the s-th operating condition sample. Let represent the voltage difference and phase angle difference between the two ends of the line for the s-th operating condition sample, respectively.

[0038] The voltage difference between the two ends of the line for the s-th operating condition sample is written as: ; The phase angle difference between the two ends of the line in the s-th operating condition sample is written as: ; An offline training dataset is constructed by combining the graph input samples with the sample labels from step S1. Graph input samples corresponding to a single working condition are paired with sample labels to form supervised learning sample pairs. This enables the model to learn the mapping relationship between the graph input and the carbon index. All sample pairs combined constitute a complete offline training dataset, which can be represented as: .

[0039] Thus far, step S2 completes the standardized mapping of the power system to a graph structure, constructs a graph input sample containing complete electrical features, operational features, and topological information, and completes the pairing and organization of graph samples and labels to form a standardized and complete offline training dataset, providing standardized input for the construction and training of edge-aware graph neural networks.

[0040] In step S3, the core task is to construct a side-aware graph neural network adapted to the assessment of carbon indicators in the power system. Through node encoding, edge encoding and side-aware message passing mechanism, node features, edge features and topological connection relationships are fully integrated to extract a deep graph representation that can reflect the coupling relationship between topology, parameters and state, providing a feature basis for subsequent joint prediction of multiple carbon indicators.

[0041] Construct a side-aware graph neural network. This network takes graph input samples as input and five carbon indicators as output. The overall mapping relationship is written as: ; in, The parameter is Graph neural network model, Let be the node feature matrix of the s-th operating condition sample. This represents the edge connection relationship. Let be the edge feature matrix of the s-th operating condition sample. This represents the predicted output of the graph neural network model under operating condition sample s; The overall mapping relationship can be further expressed as: ; This is the predicted node carbon intensity result for the s-th operating condition sample. This is the node carbon emission prediction result for the s-th operating condition sample. These are the predicted results of carbon inflow, carbon loss, and carbon outflow of the line under the s-th operating condition sample, respectively.

[0042] The node feature matrix is ​​processed by node encoding to obtain the node hidden representation matrix of the s-th operating condition sample. , represented as: ; in, Represents the node encoding function. , To hide the dimension of the representation; Edge encoding is used to process the edge feature matrix to obtain the edge hidden representation matrix of the s-th operating condition sample. , represented as: ; in, Represents the edge encoding function. .

[0043] By employing edge-aware message-passing fusion to process the node hidden representation matrix and the edge hidden representation matrix, the hidden representation matrix of all nodes in the (l+1)th layer of the s-th running condition sample can be written as: ; in, The hidden representation matrix for all nodes in the l-th layer of the s-th running condition sample. For the s-th operating condition sample, the l-th layer message passing function is used. The update process for node i in the s-th operating condition sample is as follows: ; in, Describes the set of neighboring nodes of node i. This represents the message constructor, used for edge-aware message passing fusion. Represents aggregate functions, This represents a non-linear activation function. Let i be the final hidden representation of the l-th node i in the s-th operating condition sample. Let be the final hidden representation of the l-th node j of the s-th operating condition sample. Let be the edge hidden representation matrix for the s-th operating condition sample. Here is the weight matrix for the l-th layer; In another embodiment, after each layer's message passing, layer normalization, non-linear activation, and residual connections are performed sequentially to prevent gradient vanishing in deep networks and improve training stability and deep feature representation capabilities. This update process can be written as: ; ; in, This indicates a normalization operation. This indicates the feedforward mapping module.

[0044] After L layers of edge-aware message passing, the final node hidden representation matrix, which integrates all information of topology, nodes, and edges, is obtained for the s-th operating condition sample. .

[0045] At this point, step S3 completes the overall construction and feature fusion calculation of the edge-aware graph neural network. Through node encoding, edge encoding, and edge-aware message passing, the original electrical features are transformed into a graph representation with strong expressive power.

[0046] In step S4, the core task is to build node channel decoding structure and line channel decoding structure based on the fusion features obtained in step S3, so as to realize the synchronous and joint output of node-side carbon index and line-side carbon index, ensure that the two types of indexes are predicted under the same framework, and improve the integrity and consistency of output data.

[0047] The fused node features and edge features are decoded using both node-level and line-level channels, respectively. Node-level decoding targets node-level carbon indicators, directly using the final node hidden representation for prediction, thus aligning with the local characteristics of node indicators. Line-level decoding targets line-level carbon indicators, requiring the fusion of node hidden representations and edge hidden representations at both ends of the line with the original edge features to form a fused representation specific to the line, aligning with the correlation characteristics of line indicators. This dual-channel structure balances the differences between node and line indicators, enabling joint output.

[0048] Node carbon indicators include node carbon intensity and node carbon emissions; line carbon indicators include line carbon inflow, line carbon loss, and line carbon outflow. The node carbon intensity prediction for the s-th operating condition sample is obtained through node channel decoding. , is represented as: ; in, This is the node hidden representation matrix after edge-aware message passing fusion. Represents the nodal carbon strength readout function; The node carbon emission prediction for the s-th operating condition sample is obtained through node channel decoding. , is represented as: ; in, For node carbon emission readout function; The predicted carbon inflow of the line is obtained by decoding the line channel for the s-th operating condition sample. Line carbon loss prediction Carbon outflow prediction of the line , is represented as: ; in, For line-side readout function, For the line fusion representation of the s-th operating condition sample, Let be the final hidden representation matrix of the starting node of the s-th operating condition sample. Let be the final hidden representation matrix of the termination node of the s-th operating condition sample. Let be the edge hidden representation matrix for the s-th operating condition sample. Let be the original edge feature matrix of the s-th operating condition sample.

[0049] The system outputs node carbon indicators and line carbon indicators. Node channels and line channels are calculated in the same forward propagation process, and five types of results are output simultaneously: node carbon intensity, node carbon emissions, line carbon inflow, line carbon loss, and line carbon outflow. This achieves integrated output of carbon indicators, covering the distribution characteristics of carbon attributes throughout the entire process of node injection, line transmission, line loss, and node absorption.

[0050] Thus far, step S4 uses a dual-decoding channel structure to accurately predict carbon indicators on both the node and line sides. The node channel relies on node hidden representation to ensure full utilization of local features, while the line channel uses multi-information fusion to ensure a complete characterization of the transmission process, thereby achieving joint output of five types of carbon indicators, strengthening the physical correlation between indicators, and improving the collaborative characterization effect.

[0051] In step S5, the core task is to construct a joint loss function that includes node-side supervision loss, line-side supervision loss, and physical consistency constraints, complete the full training of the edge-aware graph neural network, and perform online rapid evaluation of carbon indicators for new operating conditions after training.

[0052] A joint loss function is used to train the edge-aware graph neural network. This joint loss function considers both node-side and line-side prediction errors and enforces physical conservation relationships for the three line-side indicators, embedding engineering physics rules into the model learning process to improve the engineering usability and reliability of the output. The joint loss function is a weighted sum of the node-side and line-side loss functions, without adding any additional meaningless constraint terms.

[0053] The node-side loss function uses mean squared error loss to quantify the deviation between the predicted value and the true label of the node carbon index. The node-side loss function is expressed as follows: ; in, Let be the true value of the carbon index for node i in the s-th operating condition sample. Let be the predicted value of node i for the s-th operating condition sample. The number of nodes; The actual value on the line side under the s-th operating condition sample is represented as: ; The predicted value for the line side under the s-th operating condition sample is expressed as: ; The line-side loss function is expressed as: ; in, For L1 loss function, The mean square error is α, which is the weighting coefficient of the mean square error term. The line-side loss implicitly includes physical consistency constraints, so that the carbon inflow, loss and outflow of the line satisfy the conservation relationship.

[0054] The joint loss function is expressed as: ; In another embodiment, the training process uses the AdamW optimizer to iteratively update the parameters, combines gradient clipping to control the gradient magnitude, and adaptively adjusts the learning rate based on the validation set error until the model converges and stabilizes. The training process is completed offline and does not occupy online evaluation time.

[0055] For a new operating condition to be evaluated, a graph input sample is constructed according to step S2. This graph input sample is then fed into the trained edge-aware graph neural network to perform a forward inference, which can simultaneously output the prediction results of five types of carbon indicators. This eliminates the need to repeatedly perform tidal flow calculations and carbon flow calculations, thus reducing the evaluation time. The single-condition inference time is much shorter than the traditional per-condition calculation method, which can meet the needs of near real-time online monitoring and large-scale scene simulation engineering.

[0056] Thus far, step S5 completes model training through a joint loss function, embeds physical constraints into the learning process, ensures the accuracy of the output results is consistent with physical conditions, and achieves rapid online evaluation of carbon indicators through lightweight forward inference, thereby improving the evaluation efficiency in multi-condition and large-scale scenarios.

[0057] Example 2: like Figure 2 As shown, this embodiment also provides a joint evaluation device for carbon indicators of power systems based on edge-aware graph neural networks, including a sample construction module 1, a graph modeling module 2, a network construction module 3, an indicator decoding module 4, and a training and evaluation module 5. Sample construction module 1 is used to construct a basic operating condition sample of the power system. Based on the basic operating condition sample, disturbances are applied to the load side and the generation side to construct an operating condition sample set. AC power flow calculation and carbon emission flow analysis are performed on the operating condition sample set to obtain sample labels. The sample labels include node carbon index labels and line carbon indexes. Graph modeling module 2 is used to represent the power system as a graph structure, extract node features and edge features from the sample set to form graph input samples, and combine them with sample labels to obtain the training dataset; Network building module 3 is used to build edge-aware graph neural networks, fusing node features and edge features through edge-aware message passing; Among them, network construction module 3 also includes node coding unit, edge coding unit, and edge-aware message passing unit; Node encoding unit, used to map node features to node hidden representations; Edge coding unit is used to map edge features to edge hidden representations; The edge-aware message passing unit is used to update the hidden representation of a node by combining the node's neighborhood relationship with the edge's hidden representation, and to fuse node features and edge features.

[0058] The index decoding module 4 performs node channel decoding and line channel decoding on the fused node features and edge features respectively, and outputs node carbon index and line carbon index. Training and evaluation module 5 uses a joint loss function to train a side-aware graph neural network and performs online rapid evaluation of carbon indicators.

[0059] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. The methods disclosed in the embodiments are described simply because they correspond to the systems disclosed in the embodiments; relevant details can be found in the method section.

[0060] Those skilled in the art will further recognize that the 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 invention.

[0061] In the embodiments provided by this invention, it should be understood that the disclosed systems, methods, and approaches can be implemented in other ways. For example, the system embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between systems or units may be electrical, mechanical, or other forms.

[0062] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0063] In addition, the functional modules in the various embodiments of the present invention 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.

[0064] Similarly, in the various embodiments of the present invention, each processing unit can be integrated into a functional module, or each processing unit can exist physically, or two or more processing units can be integrated into a functional module.

[0065] The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein can be implemented directly by hardware, a software module executed by a processor, or a combination of both. The software module can be located in random access memory (RAM), main memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, removable disk, CD-ROM, or any other form of storage medium known in the art.

[0066] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0067] The above-disclosed embodiments are merely preferred embodiments of the present invention, but the present invention is not limited thereto. Any non-creative variations that can be conceived by those skilled in the art, as well as any improvements and modifications made without departing from the principles of the present invention, should fall within the protection scope of the present invention.

Claims

1. A joint evaluation method for carbon indicators of power systems based on edge-aware graph neural networks, characterized in that, Includes the following steps: Step S1: Construct a basic operating condition sample of the power system. Based on the basic operating condition sample, apply disturbances to the load side and the generation side to construct an operating condition sample set. Perform AC power flow calculation and carbon emission flow analysis on the operating condition sample set to obtain sample labels. The sample labels include node carbon index labels and line carbon indexes. The node carbon indexes include node carbon intensity and node carbon emissions. The line carbon indexes include line carbon inflow, line carbon loss, and line carbon outflow. Step S2: Represent the power system as a graph structure, extract node features and edge features from the sample set to form graph input samples, and combine them with sample labels to obtain the training dataset; Step S3: Construct an edge-aware graph neural network and fuse node features and edge features through edge-aware message passing; Step S4: Perform node channel decoding and line channel decoding on the fused node features and edge features respectively, and output node carbon index and line carbon index; Step S5: Train the edge-aware graph neural network using a joint loss function to perform online rapid evaluation of carbon indicators; In step S3, the edge-aware graph neural network is constructed as follows: ; in, The parameter is Graph neural network model, Let be the node feature matrix of the s-th operating condition sample. The edges are connected. Let be the edge feature matrix of the s-th operating condition sample. This represents the predicted output of the graph neural network model under operating condition sample s; The node feature matrix is ​​processed by node encoding to obtain the node hidden representation matrix of the s-th operating condition sample. , is represented as: ; in, Represents the node encoding function; Edge encoding is used to process the edge feature matrix to obtain the edge hidden representation matrix of the s-th operating condition sample. , is represented as: ; in, Represents the edge encoding function; Edge-aware message-passing fusion is used to process the node hidden representation matrix and the edge hidden representation matrix. The update process of node i in the s-th running condition sample is as follows: ; in, Describes the set of neighboring nodes of node i. This represents the message constructor, used for edge-aware message passing fusion. Represents aggregate functions, Represents a non-linear activation function. Let i be the final hidden representation of the l-th node i in the s-th operating condition sample. Let be the final hidden representation of the l-th node j of the s-th operating condition sample. Let be the edge hidden representation matrix for the s-th operating condition sample. Let be the weight matrix of the l-th layer.

2. The method for joint evaluation of carbon indicators in power systems based on edge-aware graph neural networks according to claim 1, characterized in that, Step S1 involves constructing a basic operating condition sample of the power system. Specifically, this involves applying disturbances to the load side and generation side based on this basic operating condition sample to construct an operating condition sample set. The operating condition sample set is represented as follows: ,in, This is the Mth operating condition sample, where M is the total number of operating condition samples. Performing AC power flow calculations and carbon emission flow analysis on the operating condition sample set specifically includes: Perform AC power flow calculations on the sample set of operating conditions; Based on the node operating status, power at both ends of the line, and active power loss of the line under the carbon emission flow analysis operating condition sample s, sample labels are generated including node carbon intensity, node carbon emission, line carbon inflow, line carbon loss, and line carbon outflow. , is represented as: ; in, This is the node carbon intensity label vector for the operating condition sample s. This is the node carbon emission label vector under operating condition sample s. This is the carbon inflow label vector of the line under operating condition sample s. This is the line carbon loss label vector under operating condition sample s. This is the line carbon outflow label vector under the operating condition sample s.

3. The method for joint evaluation of carbon indicators in power systems based on edge-aware graph neural networks according to claim 2, characterized in that, Under operating condition sample s, the nodal carbon intensity of each node is calculated to generate the nodal carbon intensity under operating condition sample s. The nodal carbon intensity of node i under operating condition sample s is expressed as: ; in, Let i be the nodal carbon intensity of node i under the s-th operating condition sample. Let be the carbon flow corresponding to the local power generation of node i under the s-th operating condition sample. Let i be the set of upstream nodes of node i. Let be the amount of carbon flowing from node j into node i after transmission through the line in the s-th operating condition sample. Let be the active power generated by node i. Let be the active power at the line receiving end of the s-th operating condition sample; By calculating the node carbon emissions for each node, the node carbon emissions under operating condition sample s are generated, where the node carbon emissions of node i under the s-th operating condition sample are represented as: ; in, Let be the node carbon emissions of node i in the s-th operating condition sample. Let i be the nodal carbon intensity of node i under the s-th operating condition sample. Let be the load active power of node i under the s-th operating condition sample; By calculating the carbon flow entering the line at each node, the carbon inflow into the line under operating condition sample s is generated. In the s-th operating condition sample, the carbon inflow into the line from node i is calculated. carbon flow Represented as: ; in, Let i be the nodal carbon intensity of node i under the s-th operating condition sample. For the s-th operating condition sample, the line The transmitting power; Line carbon loss For the s-th operating condition sample, the line The difference between carbon inflow and carbon outflow from the line is expressed as: ;in, The line carbon outflow is the carbon flow rate output by the receiving end after transmission in the s-th operating condition sample.

4. The method for joint evaluation of carbon indicators in power systems based on edge-aware graph neural networks according to claim 1, characterized in that, In step S2, the target power system is abstracted into a graph structure, where the set of nodes corresponds to the power system bus and the set of edges corresponds to the power system line. For the s-th operating condition sample, the corresponding graphical input sample is represented as: ,in, Let be the node feature matrix of the s-th operating condition sample. The edges are connected. Let be the edge feature matrix of the s-th operating condition sample; The node feature matrix is ​​represented as follows: , Let i be the feature vector of node i in the s-th operating condition sample, and be expressed as: ; in, The active power of node load at node i under operating condition sample s. The reactive power of node i under operating condition sample s is the node load. The active power generated by node i under operating condition sample s. The reactive power generated by node i under operating condition sample s. The node voltage amplitude of node i under operating condition sample s. The node voltage phase angle of node i under operating condition sample s. To inject net active power into node i under operating condition sample s. To inject net reactive power into node i under operating condition sample s. Encode the node type; The edge feature matrix of the s-th operating condition sample is represented as: ; in, Let be the number of directed edges. For any directed edge under the s-th operating condition sample The constructed edge feature vector is represented as: ; in, These represent the line parameters, including line impedance, line reactance, and line susceptance. Let these represent the active power at the transmitting end and the receiving end of the line, respectively, for the s-th operating condition sample. This represents the line active power loss for the s-th operating condition sample. Let represent the voltage difference and phase angle difference between the two ends of the line for the s-th operating condition sample, respectively.

5. The method for joint evaluation of carbon indicators in power systems based on edge-aware graph neural networks according to claim 4, characterized in that, Input sample based on the graph of the s-th operating condition sample , with sample label The training dataset is obtained by combining the data, and is represented as follows: ； Combining all operating condition samples Construct an offline training dataset, represented as: .

6. The method for joint evaluation of carbon indicators in power systems based on edge-aware graph neural networks according to claim 5, characterized in that, In step S4, the node carbon indicators include node carbon intensity and node carbon emissions; the line carbon indicators include line carbon inflow, line carbon loss, and line carbon outflow. The node carbon intensity prediction for the s-th operating condition sample is obtained through node channel decoding. , is represented as: ; in, This is the node hidden representation matrix after edge-aware message passing fusion. Represents the nodal carbon strength readout function; The node carbon emission prediction for the s-th operating condition sample is obtained by decoding the node channel. , is represented as: ; in, For node carbon emission readout function; The predicted carbon inflow of the line is obtained by decoding the line channel for the s-th operating condition sample. Line carbon loss prediction Carbon outflow prediction of the line , is represented as: ; in, For line-side readout function, For the line fusion representation of the s-th operating condition sample, Let be the final hidden representation matrix of the starting node of the s-th operating condition sample. Let be the final hidden representation matrix of the termination node of the s-th operating condition sample. Let be the edge hidden representation matrix for the s-th operating condition sample. Let be the original edge feature matrix of the s-th operating condition sample.

7. The method for joint evaluation of carbon indicators in power systems based on edge-aware graph neural networks according to claim 6, characterized in that, In step S5, the joint loss function includes the node-side loss function and the line-side loss function; The node-side loss function is expressed as: ; in, Let be the true value of the carbon index for node i in the s-th operating condition sample. Let be the predicted value of node i for the s-th operating condition sample. The number of nodes; The line-side loss function is expressed as: ; in, This represents the actual line-side value under the s-th operating condition sample. This is the predicted value for the line side under the s-th operating condition sample. For L1 loss function, Let be the mean squared error, and α be the weighting coefficient of the mean squared error term.

8. A joint evaluation device for carbon indicators of power systems based on edge-aware graph neural networks, characterized in that, It includes a sample construction module, a graph modeling module, a network construction module, an indicator decoding module, and a training and evaluation module; The sample construction module is used to construct basic operating condition samples of the power system. Based on the basic operating condition samples, disturbances are applied to the load side and the generation side to construct an operating condition sample set. AC power flow calculation and carbon emission flow analysis are performed on the operating condition sample set to obtain sample labels. The sample labels include node carbon index labels and line carbon indexes. The node carbon indexes include node carbon intensity and node carbon emissions. The line carbon indexes include line carbon inflow, line carbon loss, and line carbon outflow. The graph modeling module is used to represent the power system as a graph structure. It extracts node features and edge features from the sample set to form graph input samples, which are then combined with sample labels to obtain the training dataset. The network building module is used to build edge-aware graph neural networks, fusing node features and edge features through edge-aware message passing; The index decoding module performs node channel decoding and line channel decoding on the fused node features and edge features respectively, and outputs node carbon index and line carbon index. The training and evaluation module uses a joint loss function to train the edge-aware graph neural network and performs online rapid evaluation of carbon indicators. The construction of the edge-aware graph neural network is represented as follows: ; in, The parameter is Graph neural network model, Let be the node feature matrix of the s-th operating condition sample. The edges are connected. Let be the edge feature matrix of the s-th operating condition sample. This represents the predicted output of the graph neural network model under operating condition sample s; The node feature matrix is ​​processed by node encoding to obtain the node hidden representation matrix of the s-th operating condition sample. , is represented as: ; in, Represents the node encoding function; Edge encoding is used to process the edge feature matrix to obtain the edge hidden representation matrix of the s-th operating condition sample. , is represented as: ; in, Represents the edge encoding function; Edge-aware message-passing fusion is used to process the node hidden representation matrix and the edge hidden representation matrix. The update process of node i in the s-th running condition sample is as follows: ; in, Describes the set of neighboring nodes of node i. This represents the message constructor, used for edge-aware message passing fusion. Represents aggregate functions, Represents a non-linear activation function. Let i be the final hidden representation of the l-th node i in the s-th operating condition sample. Let be the final hidden representation of the l-th node j of the s-th operating condition sample. Let be the edge hidden representation matrix for the s-th operating condition sample. Let be the weight matrix of the l-th layer.

9. A joint evaluation device for carbon indicators of power systems based on edge-aware graph neural networks according to claim 8, characterized in that, The network construction module includes node coding units, edge coding units, and edge-aware message passing units; Node encoding unit, used to map node features to node hidden representations; Edge coding unit is used to map edge features to edge hidden representations; The edge-aware message passing unit is used to update the hidden representation of a node by combining the node's neighborhood relationship with the edge's hidden representation, and to fuse node features and edge features.

Images