Dynamic Evaluation Method, Device, Electronic Device and Storage Medium of Electrocarbon Factor

Abstract

The present invention discloses a method, device, electronic device and storage medium for dynamically evaluating the electro-carbon factor, which are used to solve the problems of limited real-time performance, low accuracy, lack of flexibility and poor adaptability in the existing related technologies. The method includes: obtaining node information data and power time-varying data of a power system; constructing a node feature matrix and a node graph structure of the power system according to the node information data, and performing local optimization of the graph structure with combined node embedding update on the node feature matrix and the node graph structure to obtain node embedding representations; performing time-series optimized feature extraction on the power time-varying data to obtain time-series features; performing linear feature fusion according to the node embedding representations and the time-series features to obtain node feature representations; calculating the user-side electro-carbon factor of the power system based on the node feature representations. Thus, through a relatively simple processing flow, fast and accurate dynamic evaluation of the user-side electro-carbon factor is realized.

Description

Technical Field

[0001] The present invention relates to the technical field of carbon emission assessment of power systems, and particularly to a method, device, electronic device and storage medium for dynamically assessing an electricity-carbon factor. Background Art

[0002] The power industry is one of the important pillars of the social economy and provides key power for all walks of life. With the increasing global attention to sustainable development and environment-friendly energy, the power industry is also in a state of continuous evolution. Therefore, it is necessary to continuously introduce new technologies and methods to meet the needs of contemporary society. The environmental impact of power systems is a key focus, especially issues related to carbon emissions. Carbon emissions are widely regarded as one of the main driving factors of climate change. Therefore, it has become particularly urgent to assess and reduce carbon emissions globally.

[0003] In this context, the power industry has begun to focus on the electricity-carbon factor (carbon dioxide emissions per unit of electric energy) and regard it as a key performance indicator. In actual situations, the electricity-carbon factor is not only an important basis for evaluating the environmental friendliness of power systems, but also directly affects the adoption decisions of users, enterprises and governments regarding clean and sustainable energy.

[0004] Currently, a relatively in-depth solution for the assessment of carbon emissions in power systems is the combination of carbon emission analysis and power system power flow calculation. Further research mostly makes improvements or optimizations based on the carbon emission flow theory, such as the recursive algorithm for carbon emission flow in power systems.

[0005] From the current research, in traditional power systems, the assessment of the electricity-carbon factor usually relies on statistical models and rule-based assessment methods. Statistical models are based on historical data and statistical algorithms, and estimate the electricity-carbon factor by analyzing and modeling the existing data. The rule-based assessment method relies on predefined rules and parameters for assessment. Although these methods provide some convenience for electricity-carbon assessment on the user side to a certain extent, they have the following series of limitations:

[0006] 1. Limited real-time performance: Traditional electricity-carbon factor assessment methods often rely on historical data and statistical models, resulting in significant deficiencies in real-time performance. The dynamic changes in power systems require more timely electricity-carbon factor data to support real-time decision-making. Traditional methods are difficult to provide immediate assessment results in a rapidly changing environment and cannot meet the user's demand for real-time data.

[0007] 2. Lack of flexibility and adaptability: Rule-based user-side assessment methods may lack flexibility in complex power system environments. Predefined rules and parameters often cannot comprehensively cover various changes and special situations in power systems, restricting the adaptability of the assessment method, especially under different energy mixes and operating conditions. Due to the diverse structures and operating conditions of power systems, it is difficult for predefined rules and parameters to cover all possible scenarios, thus affecting the accuracy and adaptability of the assessment.

[0008] 3. Limited assessment accuracy: The assessment accuracy of traditional statistical models may be limited in some cases. Since these models usually rely on the statistical analysis of historical data, their response to the introduction of new energy or other system changes may not be sensitive enough, resulting in the estimation of the electricity-carbon factor possibly deviating from the actual situation.

[0009] 4. Complex structure and data processing: Traditional methods may involve complex mathematical models and data processing procedures, making the entire assessment process relatively cumbersome. This complexity not only increases the cost of the assessment but also reduces the assessment efficiency, especially when dealing with large-scale data.

[0010] 5. Lack of personalized user experience: Traditional methods often fail to provide personalized electricity-carbon factor assessments, unable to fully consider users' specific needs and preferences, thus limiting users' in-depth understanding of the impact of electricity on the environment and hindering personalized decision-making in energy consumption.

[0011] 6. Poor adaptability to emerging energy: Since traditional methods were not fully designed considering the complexity of emerging energy, in power systems with a relatively high proportion of renewable energy, their assessments may exhibit relatively low efficiency and inaccuracy. Summary of the Invention

[0012] The present invention provides a dynamic assessment method, device, electronic device, and storage medium for electricity-carbon factors, which are used to solve or partially solve the technical problems of limited real-time performance, low accuracy, lack of flexibility, and poor adaptability in existing related assessment means due to complex node structures and data processing processes.

[0013] A dynamic assessment method for electricity-carbon factors provided by the present invention, the method includes:

[0014] Obtain node information data and power time-varying data of the power system;

[0015] According to the node information data, construct a node feature matrix and a node graph structure of the power system, and perform local optimization of the graph structure with combined node embedding update on the node feature matrix and the node graph structure to obtain node embedding representations;

[0016] Perform time-series optimization feature extraction on the time-varying power data to obtain time-series features;

[0017] Perform linear feature fusion based on the node embedding representation and the time-series features to obtain a node feature representation;

[0018] Calculate the user-side electricity-carbon factor of the power system based on the node feature representation.

[0019] Optionally, the node information data includes the node information of all nodes in the power system and the connection relationships between each pair of nodes; constructing the node feature matrix and the node graph structure of the power system according to the node information data includes:

[0020] Construct the node feature matrix of the power system according to the node information of all nodes;

[0021] Construct an adjacency matrix according to the connection relationships between each pair of nodes, and the adjacency matrix is used to represent the node graph structure of the power system.

[0022] Optionally, the graph structure local optimization of combining node embedding update for the node feature matrix and the node graph structure to obtain a node embedding representation includes:

[0023] Based on the node embedding dynamic update mechanism, update the node feature matrix and the adjacency matrix respectively to obtain an updated node feature matrix and an updated adjacency matrix;

[0024] During the dynamic update process of the node feature matrix and the adjacency matrix, after each update, perform local neighbor sampling on each node respectively to obtain a sampling neighbor set, and based on the sampling neighbor set, perform local feature extraction on the updated node feature matrix and the updated adjacency matrix respectively to obtain a local node feature matrix and a local adjacency matrix;

[0025] Input the local node feature matrix and the local adjacency matrix into a pre-trained graph neural network for multi-layer convolution to obtain a node embedding representation.

[0026] Optionally, the node embedding dynamic update mechanism includes a node feature update mechanism and an adjacency matrix maintenance mechanism, and based on the node embedding dynamic update mechanism, updating the node feature matrix and the adjacency matrix respectively includes:

[0027] Based on the adjacency matrix maintenance mechanism, update the adjacency matrix, and the update process is as follows:

[0028]

[0029] Based on the node feature update mechanism, the node feature matrix is updated, and the update process is as follows:

[0030]

[0031] Among them, represents the weight of the adjacency relationship from node i to node j in the power system at time t; represents the degree of node i at time t, that is, the number of edges connected to node i; represents the degree of node j at time t, that is, the number of edges connected to node j; represents the weight of the adjacency relationship from node i to node j in the power system at time t - 1; represents the embedding representation of node i at time t; represents the activation function; W represents the learning weight parameter; represents the embedding representation of node i at time t - 1; represents the embedding representation of node j at time t - 1.

[0032] Optionally, the power time-varying data includes time series data and node-related time-varying information; the obtaining of time series features by performing temporal optimization feature extraction on the power time-varying data includes:

[0033] Perform data preprocessing on the time series data and the node-related time-varying information respectively, and integrate all the data obtained after data preprocessing to obtain spatio-temporal correlation data;

[0034] Analyze the spatio-temporal correlation data, and based on the analysis results, adjust the time series processing length and the number of hidden units of the pre-trained multi-layer LSTM structure to dynamically adjust and optimize the multi-layer LSTM structure;

[0035] Perform temporal feature extraction on the spatio-temporal correlation data based on the optimized multi-layer LSTM structure to obtain time series features.

[0036] Optionally, the obtaining of the node feature representation by performing linear feature fusion according to the node embedding representation and the time series features includes:

[0037] Based on the linear combination method of the weight matrix, perform linear feature fusion on the node embedding representation and the time series features to obtain the node feature representation. The linear feature fusion process is as follows:

[0038]

[0039] Among them, Z i represents the node feature representation of node i; ReLU represents the rectified linear unit activation function; Wfusion and b fusion respectively represent the weight and bias parameters of the feature fusion layer; [H i , h T represents concatenating the node embedding representation H i and the LSTM hidden state h at the last moment T for feature concatenation.

[0040] Optionally, calculating the user-side electricity-carbon factor of the power system based on the node feature representation includes:

[0041] Inputting the node feature representation into a pre-trained electricity-carbon factor evaluation model for carbon emission evaluation calculation to obtain the user-side electricity-carbon factor of the power system;

[0042] wherein, the electricity-carbon factor evaluation model adopts a regression model structure of a feedforward neural network with multiple hidden layers.

[0043] The present invention also provides an electricity-carbon factor dynamic evaluation device, including:

[0044] A power system data acquisition module for acquiring node information data and power time-varying data of the power system;

[0045] A node embedding representation generation module for constructing a node feature matrix and a node graph structure of the power system according to the node information data, and performing local optimization of the graph structure with combined node embedding update on the node feature matrix and the node graph structure to obtain a node embedding representation;

[0046] A time-series optimization feature extraction module for performing time-series optimization feature extraction on the power time-varying data to obtain time-series features;

[0047] A linear feature fusion module for performing linear feature fusion according to the node embedding representation and the time-series features to obtain a node feature representation;

[0048] A user-side electricity-carbon factor calculation module for calculating the user-side electricity-carbon factor of the power system based on the node feature representation.

[0049] The present invention also provides an electronic device, the device includes a processor and a memory:

[0050] The memory is used for storing program codes and transmitting the program codes to the processor;

[0051] The processor is used for executing the electricity-carbon factor dynamic evaluation method as described in any one of the above according to the instructions in the program codes.

[0052] The present invention also provides a computer-readable storage medium for storing program code for executing the electric carbon factor dynamic evaluation method described in any one of the above.

[0053] As can be seen from the above technical solutions, the present invention has the following advantages:

[0054] A user-side electric carbon factor dynamic evaluation method is provided. First, node information data and power time-varying data of the power system are obtained; then, according to the node information data, a node feature matrix and a node graph structure of the power system are constructed, and local optimization of the graph structure with combined node embedding update is performed on the node feature matrix and the node graph structure to obtain node embedding representations; time-series optimization feature extraction is performed on the power time-varying data to obtain time-series features; then, linear feature fusion is performed according to the node embedding representations and the time-series features to obtain node feature representations; finally, based on the node feature representations, the user-side electric carbon factor of the power system is calculated. Thus, by using a relatively simplified feature representation method of the graph structure and node features, and combining further time-series optimization and linear feature fusion processing, a relatively simple electric carbon factor evaluation processing flow is provided, which can realize fast and accurate dynamic evaluation of the user-side electric carbon factor. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the following drawings are only some embodiments of the present invention, and those of ordinary skill in the art can obtain other drawings without creative efforts based on these drawings.

[0056] Figure 1 It is a step flowchart of an electric carbon factor dynamic evaluation method;

[0057] Figure 2 It is a schematic diagram of the overall process of an electric carbon factor dynamic evaluation method;

[0058] Figure 3 It is a structural block diagram of an electric carbon factor dynamic evaluation device. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0059] The embodiments of the present invention provide an electric carbon factor dynamic evaluation method, device, electronic device, and storage medium, which are used to solve or partially solve the technical problems of limited real-time performance, low accuracy, lack of flexibility, and poor adaptability caused by the complex node structure and data processing process in the existing related evaluation means.

[0060] In order to make the object, features, and advantages of the present invention more obvious and understandable, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the embodiments described below are only a part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts belong to the scope of protection of the present invention.

[0061] As an example, in the power industry, the electro-carbon factor is not only an important basis for evaluating the environmental friendliness of the power system but also directly affects the adoption decisions of users, enterprises, and governments regarding clean and sustainable energy. Currently, the relatively in-depth solution for evaluating the carbon emissions of the power system is the combination of carbon emissions analysis and power system power flow calculation. Further research mostly makes improvements or optimizations based on the carbon emission flow theory, such as the recursive algorithm for the carbon emission flow of the power system.

[0062] From the current research, in traditional power systems, the evaluation of the electro-carbon factor usually relies on statistical models and rule-based evaluation methods. The statistical model is based on historical data and statistical algorithms, and estimates the electro-carbon factor by analyzing and modeling the existing data. The rule-based evaluation method, on the other hand, relies on predefined rules and parameters for evaluation. Although these methods provide some convenience for the electro-carbon evaluation on the user side to a certain extent, there are technical problems such as limited real-time performance, low accuracy, complex data processing process, lack of flexibility, and poor adaptability. These problems in the current technology have given rise to an urgent need for a more advanced, accurate, and real-time evaluation method for the electro-carbon factor on the user side.

[0063] With the continuous development of deep learning technology, especially the application of models such as graph neural networks and recurrent neural networks, the present invention believes that the limitations of traditional methods can be solved through innovative methods to improve the accuracy and efficiency of electro-carbon factor evaluation.

[0064] Therefore, one of the core inventive points of the embodiments of the present invention lies in: aiming at the deficiencies existing in the prior art and overcoming the limitations of traditional methods. First, by combining a graph neural network and a recurrent neural network, a deep learning model of the graph convolutional neural network (GCN) in the graph neural network and the long short-term memory network (LSTM) in the recurrent neural network is introduced, and the graph data and time series data in the power system are ingeniously combined to construct a powerful GCN+LSTM model. This combination enables the model to simultaneously capture the relationships between nodes in the power system (GCN) and the spatio-temporal correlation of time series data (LSTM), thereby realizing a more comprehensive and accurate evaluation of the user-side electricity-carbon factor. Second, in the process of processing the graph structure, a local neighbor sampling strategy is introduced. Through an efficient graph structure representation, the computational efficiency and accuracy of node embedding in the power system are improved, enabling the model to better adapt to the huge power system and solve the complex structure and data processing problems faced in traditional methods. At the same time, the combined GCN+LSTM model can not only perform deep learning on the overall power system but also realize the learning and adaptation to the personalized needs of users. By capturing node relationships through GCN and processing time series data through LSTM, the model can better understand the consumption patterns and preferences of users, thereby providing a more personalized and intelligent user-side electricity-carbon factor evaluation experience. Further, another important innovation point of the present invention lies in the strong adaptability of the model, especially in processing emerging energy data. Through the flexible learning of the graph structure by GCN and the sensitive processing of time series data by LSTM, the model can better adapt to the emerging energy introduced into the power system, improve the accuracy and real-time performance of carbon emission evaluation for these energies, and this adaptability is not possessed by traditional methods, thus enabling the present invention to perform more excellently in the face of the challenges of power system structure changes and new energy penetration. In summary, the introduction of the deep learning model will provide a more flexible and dynamic solution for electricity-carbon factor evaluation, provide comprehensive power environment information for users, and promote the power system towards a cleaner and more sustainable future. It is expected to provide a more intelligent, transparent, and environmentally friendly solution for the power system, meeting the growing demand of users for clean energy.

[0065] Referring to Figure 1 , a flowchart of the steps of a method for dynamically evaluating an electricity-carbon factor provided by an embodiment of the present invention is shown, which may specifically include the following steps:

[0066] Step 101, obtain node information data and power time-varying data of the power system;

[0067] In a specific implementation, in order to construct a graph structure related to nodes in the power system, embodiments of the present invention first collect power system-related data including node information, connection relationships, etc., collectively referred to as node information data. Among them, each node can be represented as an element in the power system (such as a power device, a power station, etc.), and the connection relationship is used to represent the physical or logical connection between nodes.

[0068] Meanwhile, embodiments of the present invention also make preparations for collecting time series-related data (collectively referred to as power time-varying data) that will be used later. Specifically, time series data including information such as power consumption and power generation, as well as other time-varying information related to each node (i.e., node-related time-varying information) are mainly collected. Among them, the other time-varying information referred to in embodiments of the present invention generally refers to other information that may change over time and is related to each node in addition to power consumption and power generation. These information include but are not limited to: temperature data - the temperature change in the area where each node is located may affect the operation of the power system, especially the energy consumption of air conditioners and heating systems; humidity data - the change in humidity may affect the generation and consumption of energy, especially for power generation equipment related to some renewable energy sources (such as wind energy, solar energy).

[0069] Step 102, according to the node information data, construct the node feature matrix and the node graph structure of the power system, and perform local optimization of the graph structure with combined node embedding update on the node feature matrix and the node graph structure to obtain node embedding representations;

[0070] In embodiments of the present invention, a graph convolutional neural network GCN (simply referred to as a graph neural network in the present invention) is first introduced to process the node embedding of the power system.

[0071] Combined with the foregoing content, it can be known that the node information data can at least include the node information of all nodes in the power system and the connection relationships between each pair of nodes. Then, in a specific implementation, according to the node information data, constructing the node feature matrix and the node graph structure of the power system can be as follows: First, the node feature matrix of the power system can be constructed according to the node information of all nodes; Second, an adjacency matrix A can be constructed according to the connection relationships between each pair of nodes, where the adjacency matrix A can be used to represent the node graph structure of the power system.

[0072] Further, the calculation expression of the adjacency matrix A is as follows:

[0073]

[0074] Among them, i and j represent grid nodes.

[0075] In the embodiments of the present invention, the introduction of the graph neural network GCN is to overcome the complex node correlations in the power system. Based on the construction method of the graph structure of the node feature matrix and the adjacency matrix, in the subsequent processing, through the convolutional operations of multiple layers of GCN, not only can the information of the node itself be captured in the node embedding, but also the relationship between the node and its neighbor nodes can be successfully obtained. And the learning of this relationship enables the model to more accurately simulate the mutual influence between nodes in the power system.

[0076] Specifically, in the technical solution of the present invention, a dynamic update mechanism of node embedding is adopted to ensure that the model can reflect the structural changes of the power system in a timely manner. Among them, this dynamic update mechanism mainly includes the maintenance of the adjacency matrix and the real-time update of node features to meet the dynamic requirements of the power system.

[0077] In a specific implementation, the process of obtaining the node embedding representation by performing local optimization of the graph structure that combines the node feature matrix and the node graph structure for node embedding update can be achieved by executing the following sub-steps S1021 to S1023:

[0078] Step S1021: Based on the dynamic update mechanism of node embedding, update the node feature matrix and the adjacency matrix respectively to obtain the updated node feature matrix and the updated adjacency matrix;

[0079] As can be seen from the foregoing content, the dynamic update mechanism of node embedding can include a node feature update mechanism and an adjacency matrix maintenance mechanism. Then, the dynamic update in step S1021 can be divided into a dynamic update process for the adjacency matrix and an update process for the node feature matrix in combination with the update of the adjacency matrix.

[0080] Among them, based on the adjacency matrix maintenance mechanism, the process of updating the adjacency matrix is as follows:

[0081]

[0082] Based on the node feature update mechanism, the process of updating the node feature matrix is as follows:

[0083]

[0084] Among them, represents the weight of the adjacency relationship from node i to node j in the power system at time step t. This weight usually represents the association strength between two nodes in the context of the graph neural network GCN, and can be understood as the connection degree between node i and node j; t represents the time step. In the dynamic graph neural network, the update of the adjacency matrix A may change over time to reflect the evolution of the power system structure at different time points. Based on this evolution characteristic, The \(t\) in it can be used to reflect the state of the adjacency matrix at time \(t\); denotes the degree of node \(i\) at time \(t\), that is, the number of edges connected to node \(i\). The degree information is used to normalize the adjacency matrix \(A\) to ensure that the weight of each node takes into account its connection degree relative to other nodes; denotes the degree of node \(j\) at time \(t\), that is, the number of edges connected to node \(j\); denotes the weight of the adjacency relationship from node \(i\) to node \(j\) in the power system at time \(t - 1\); denotes the embedding representation of node \(i\) at time \(t\); denotes the activation function; \(W\) denotes the learning weight parameter; denotes the embedding representation of node \(i\) at time \(t - 1\); denotes the embedding representation of node \(j\) at time \(t - 1\).

[0085] Step S1022: During the dynamic update process of the node feature matrix and the adjacency matrix, after each update, local neighbor sampling is performed on each node respectively to obtain a sampled neighbor set. Then, based on the sampled neighbor set, local feature extraction is performed on the updated node feature matrix and the updated adjacency matrix respectively to obtain a local node feature matrix and a local adjacency matrix;

[0086] For the dynamic update process of the matrix, to cope with the huge number of nodes in the power system, a local neighbor sampling strategy is adopted in the embodiments of the present invention. For each node \(i\), its local neighbors are selected for sampling to obtain a sampled neighbor set \(N\) i . Then, based on the sampled neighbor set \(N\) obtained after sampling i , local feature extraction is performed on the updated node feature matrix and the updated adjacency matrix respectively to obtain a local node feature matrix and a local adjacency matrix. This local optimization strategy can effectively reduce the computational complexity while retaining key information.

[0087] Among them, the specific sampling method of the sampled neighbors can be random sampling or a sampling strategy based on node importance to ensure that the sampled neighbors can represent the overall graph structure and effectively reduce the computational overhead.

[0088] The above local optimization method realizes local optimization for the matrix after each update. In practical applications, regarding the update frequency or update method, those skilled in the art can set it according to the actual situation. For example, after multiple updates are completed, local sampling optimization is performed based on the matrix after the last update to obtain the corresponding locally optimized matrix. The difference is that in the former optimization method, except for the first update, the data used before each update is the data obtained from the local optimization based on the updated matrix in the previous time (that is, local optimization is performed on the matrix for each update); in the latter optimization method, the data used before each update is the data obtained after the previous update, and local optimization is only performed on the matrix after the last update (that is, only one local optimization is performed for the entire update process). By using the former optimization method, through continuous capture and update of local key information, compared with the latter method, it has stronger real-time dynamics and key information capture ability, and reduces the amount of calculation data for each update, making it more suitable for scenarios with a large amount of data and high requirements for real-time performance.

[0089] Step S1023: Input the local node feature matrix and the local adjacency matrix into the pre-trained graph neural network for multi-layer convolution to obtain the node embedding representation.

[0090] Then, the adjacency matrix A and the node feature matrix X after local feature extraction can be input into the graph neural network GCN model. Through multi-layer convolution operations, the embedding representation H of each node can be learned. i . After multi-layer convolution, these embedding vectors capture the semantic information of each node in the power system graph structure and can better reflect the relationship between nodes. The construction and training related processes of the graph neural network GCN model have been publicly disclosed in the existing related technologies. Those skilled in the art can implement it with reference to the relevant methods and will not be elaborated here.

[0091] The specific calculation formula of the GCN layer is as follows:

[0092]

[0093] Among them, is the embedding representation of node i at the l-th layer; is the adjacency matrix A plus the self-loop matrix; is the degree matrix of A; is the weight matrix of the l-th layer; σ is the activation function.

[0094] Through this step, the embedding representation H of each node in the power system can be obtained. i , thus preparing for the synergy in the subsequent steps.

[0095] Step 103: Perform time-series optimization feature extraction on the power time-varying data to obtain time-series features;

[0096] Meanwhile, to better process the time-series data in the power system, the embodiment of the present invention also introduces the long short-term memory network (LSTM) in the recurrent neural network to process the time-series data of the power system and the node-related time-varying information.

[0097] Specifically, the process of performing time-series optimization feature extraction on the power time-varying data to obtain time-series features can be implemented by executing the following sub-steps S1031 to S1033:

[0098] Step S1031: Perform data preprocessing on the time-series data and the node-related time-varying information respectively, and integrate all the data obtained after data preprocessing to obtain spatio-temporal correlation data;

[0099] The time-series data and the node-related time-varying information are used to model the spatio-temporal correlation in the power system. In this step, appropriate standardization and normalization processing are performed on the time-series data and the node-related time-varying information to ensure the comparability of the data and the robustness of the model.

[0100] Step S1032: Analyze the spatio-temporal correlation data, and based on the analysis results, adjust the time-series processing length and the number of hidden units of the pre-trained multi-layer LSTM structure to dynamically adjust and optimize the multi-layer LSTM structure;

[0101] Considering the complexity of the power time-varying data, a multi-layer LSTM structure is pre-constructed in the embodiment of the present invention and trained based on historical time-varying data. Among them, the embodiment of the present invention also optimizes the LSTM structure in many aspects, including the adoption of appropriate activation functions, the introduction of attention mechanisms, etc. This series of adjustments makes the model more flexible and can better adapt to the characteristics of different power systems. In the design of the model, through the training and verification process of the multi-layer LSTM structure, fine-tuning of parameters such as the sequence length and the number of hidden units of the model is performed to ensure that the model can effectively capture time-series features at different time scales as much as possible. The construction and training-related processes of the LSTM have been publicly disclosed in the existing related technologies, and those skilled in the art can implement them with reference to the relevant methods, and will not be elaborated here.

[0102] Perform convolution operations on the subsequent collected actual time-varying data using the trained multi-layer LSTM structure. Each layer of LSTM can capture different abstract features in the data, thereby improving the model's ability to model the time-series changes in the power system.

[0103] Based on the pre-construction and training of a multi-layer LSTM structure, in this step, by analyzing spatio-temporal correlation data (mainly time series data of the power system), and based on the analysis results, an appropriate sequence length T can be selected so that the model can effectively capture temporal features at different time scales. At the same time, adjust the number of hidden units H in the LSTM to balance the expressive power of the model and the computational efficiency.

[0104] For example, analyze relevant information such as the data volume and data complexity of the spatio-temporal correlation data, and judge whether the best processing effect can be obtained when processing with the current model parameters based on the pre-trained multi-layer LSTM structure. For example, judge whether there is a large difference (the difference threshold can be set according to the actual situation) between the data to be processed currently and the data (information such as data volume and data complexity) used for training the model previously. If the difference between the two is small, or small enough to be negligible, then the current model parameters can be not adjusted, and directly perform multi-layer convolution operations on the spatio-temporal correlation data, which can save the time difference of parameter adjustment and has stronger real-time performance; if the difference between the two is large, then the length of the time series processing and the number of hidden units can be adjusted accordingly according to the size of the difference to meet the current data processing requirements.

[0105] Step S1033: Extract temporal features from the spatio-temporal correlation data based on the optimized multi-layer LSTM structure to obtain time series features.

[0106] Specifically, the calculation formula of the LSTM layer is as follows:

[0107]

[0108]

[0109]

[0110]

[0111]

[0112]

[0113] Among them, respectively represent the values of the forget gate, input gate, memory unit, cell state, and output gate; σ is the activation function; ⊙ represents element-wise multiplication; W f , W i , W c , W o respectively represent the weight matrices of the forget gate, input gate, memory unit, and output gate; b f , b i , bc , b o respectively represent the bias parameters of the forget gate, input gate, memory cell, and output gate; x_t represents the input sequence; h t-1 represents the hidden state output of the LSTM cell at time t-1; h t represents the hidden state output of the LSTM cell at time t.

[0114] It should be noted that multiple steps in the embodiments of the present invention involve the use of activation functions. Using the same parameter symbol represents the same activation function (for example, the sigmoid activation function is used in all cases). Setting the same activation function aims to maintain consistency and ensure that the model uses the same activation function for non-linear transformations at different stages. And the sharing of parameters can help the model better learn the overall features.

[0115] Through step 103, more detailed modeling of the time series data in the power system can be carried out, thereby improving the prediction performance of the model.

[0116] Step 104, perform linear feature fusion according to the node embedding representation and the time series features to obtain a node feature representation;

[0117] One of the improvement points in the embodiments of the present invention is that the graph neural network GCN and the long short-term memory network LSTM are organically combined to form a GCN+LSTM model to jointly process the graph data and time series data in the power system.

[0118] The key point of the GCN+LSTM model lies in the synergy between the two. The present invention has conducted in-depth research on this synergy relationship to understand how these two different components complement each other, so as to achieve a more comprehensive and efficient evaluation of the user-side electricity-carbon factor.

[0119] Among them, in the process of feature fusion, the embodiments of the present invention introduce a non-linear activation function, such as the rectified linear unit (ReLU), to better capture the non-linear relationship in the feature fusion process. Performing linear correction on the non-linear relationship can not only improve the calculation efficiency but also enable the model to better adapt to the complex relationship of the electricity-carbon factor in the power system.

[0120] Then in the specific implementation, performing linear feature fusion according to the node embedding representation and the time series features to obtain a node feature representation can be: based on the linear combination method of the weight matrix, using a weight matrix W fusion , for the node embedding representation H i and the time series feature h t to perform linear feature fusion (that is, perform a linear combination of the two) to obtain a node feature representation. The linear feature fusion process is as follows:

[0121]

[0122] Among them, Z i represents the node feature representation of node i; ReLU represents the rectified linear unit activation function; W fusion and b fusion respectively represent the weight and bias parameters of the feature fusion layer; ReLU, W fusion and b fusion These parameters are used to linearly combine the features from the graph neural network GCN and the long short-term memory network LSTM to generate the final feature representation. Specifically, W fusion is a learned weight matrix used to weightedly fuse the two parts of features, and b fusion is the bias parameter of the feature fusion layer used to adjust the overall translation; [H i , h T represents concatenating the node embedding representation H i and the LSTM hidden state h T at the last moment.

[0123] Step 105, based on the node feature representation, calculate the user-side electricity-carbon factor of the power system.

[0124] Finally, by inputting the fused features into a final calculation module, the real-time and accurate evaluation of the user-side electricity-carbon factor is achieved.

[0125] Specifically, based on the node feature representation, calculating the user-side electricity-carbon factor of the power system can be: inputting the node feature representation into a pre-trained electricity-carbon factor evaluation model for carbon emission evaluation calculation to obtain the user-side electricity-carbon factor of the power system; among them, the electricity-carbon factor evaluation model adopts the regression model structure of a feedforward neural network with multiple hidden layers.

[0126] Combined with the foregoing content, it can be seen that the input received by the final calculation module is the node feature representation output by the feature fusion layer. These features include the information from the graph neural network GCN and the long short-term memory network LSTM, and are more representative node representations formed after fusion.

[0127] In the technical solution of the present invention, the final calculation module is designed as a deep learning model (electric carbon factor evaluation model) that combines advanced algorithms and real-time requirements. Specifically, it is a neural network regression model that uses a carbon emission flow analysis method based on the principle of proportional sharing in deep learning as the calculation method, and is a regression model structure of a feedforward neural network (FNN) with multiple hidden layers. During the construction and training of the model, a large number of trainable parameters are included in the hidden layer of the electric carbon factor evaluation model, and it learns through the training dataset to capture the non-linear relationship between node features and electric carbon factors. That is to say, this model is mainly trained to learn the complex mapping relationship between input features and electric carbon factors, and its main task is to receive the node feature representations from the feature fusion layer, and then through these node feature representations, to evaluate the user-side electric carbon factor in real time and accurately.

[0128] In addition, the model can be adjusted according to actual needs to make the output of the model meet the specific requirements of the power system. For example, according to the actual needs of the power system, adjust the calculation accuracy of the model. Some application scenarios may have high-precision requirements for electric carbon factors, while some scenarios may be more concerned about the calculation speed. At this time, the depth, width or other hyperparameters of the model can be adjusted to balance the calculation speed and accuracy.

[0129] For another example, different power system applications may have different requirements for real-time performance. For some scenarios, it may be required that the model can achieve real-time output within milliseconds, while in other scenarios, a slightly delayed output may be acceptable without much impact. Therefore, the structure of the model can be adjusted, a more lightweight model can be adopted, or hardware acceleration can be used to meet different real-time requirements.

[0130] In some special cases, special power system parameters (such as frequency, voltage, etc.) can also be considered. When designing the model, these parameters are taken into account to better adapt to different types of power systems.

[0131] In addition, in some application scenarios, more attention may be paid to the electric carbon factors of specific nodes or regions in the power system. At this time, the relevant weights of the model can be adjusted, and specially designed node representations can be introduced to pay more attention to these specific regions.

[0132] In the embodiment of the present invention, through the combination of GCN+LSTM, the problems of poor real-time performance and lack of flexibility in traditional methods are solved, and a more accurate and efficient user-side electric carbon factor evaluation effect is achieved, which can better meet the user's needs.

[0133] On this basis, combined with personalized electro-carbon factor calculation methods and analysis and optimization technologies, a complete user-side dynamic electro-carbon factor evaluation system can be built.

[0134] Among them, the personalized electro-carbon factor calculation methods mainly include:

[0135] User behavior modeling: By monitoring the behavior of users interacting with the power system, machine learning models are used to model the preferences and needs of users.

[0136] Personalized visualization interface: Based on the user behavior model, a personalized visualization interface is designed and implemented. The visualization interface can display content such as metrics, charts, and real-time data that the user is interested in, so as to provide electro-carbon factor information that conforms to the user's habits and needs.

[0137] User feedback and adjustment: A user feedback mechanism is introduced, allowing users to provide feedback on the electro-carbon factor calculation. This feedback can be used to adjust the model to better meet the personalized calculation needs of users.

[0138] User configuration options: User-configurable options are provided, allowing users to customize parameters, weights, or other factors included in the calculation model to ensure that the calculation experience better meets individual needs.

[0139] The analysis and optimization technologies mainly include:

[0140] Data analysis: Data analysis technologies are used to deeply analyze the operation data of the power system, including time series analysis, trend analysis, anomaly detection, etc., to understand the operation status of the power system.

[0141] Optimization algorithms: Optimization algorithms, such as genetic algorithms, particle swarm algorithms, etc., are applied to optimize the parameters of the power system, including objectives such as optimizing energy distribution and reducing carbon emissions.

[0142] Energy consumption optimization: Energy consumption optimization technologies, such as dynamic electricity price response, load management, etc., are applied to reduce the overall power consumption of the system.

[0143] In the embodiments of the present invention, a user-side electro-carbon factor dynamic evaluation method combining the graph neural network GCN and the recurrent neural network LSTM is proposed. Considering the calculation of the electro-carbon factor, the combined evaluation system not only provides a more flexible and dynamic solution for the electro-carbon factor evaluation, provides comprehensive power environment information for users, but also can provide personalized electro-carbon factor calculation and evaluation services for users, and helps the power system achieve a more efficient and low-carbon operation state, promoting the power system towards a cleaner and more sustainable future, and is expected to provide a more intelligent, transparent and environmentally friendly solution for the power system to meet the growing demand of users for clean energy.

[0144] For better illustration, refer to Figure 2 , which shows a schematic diagram of the overall process of a dynamic evaluation method for electro-carbon factors provided by an embodiment of the present invention. It should be noted that this embodiment only briefly describes the general process of dynamic evaluation of electro-carbon factors. The specific implementation process of each step can be understood by referring to the relevant content in the foregoing embodiments, and will not be elaborated here. It can be understood that the present invention is not limited thereto.

[0145] A. Collect power system information, that is, the original data of unprocessed node information data and power time-varying data;

[0146] B. Perform corresponding data processing (data preprocessing means such as removing duplicate data, deleting error data, filling in missing data, etc.) on the original data of node information data and power time-varying data respectively;

[0147] C. According to the preprocessed node information data, construct the node graph structure of the power system, and input the node graph structure into the graph neural network GCN, and perform processing by combining local neighbor sampling and graph structure representation optimization means to obtain node embedding representation;

[0148] D. Input the power time-varying data (mainly power consumption time series data) into the optimized multi-layer LSTM for feature extraction to obtain time series features;

[0149] E. Perform linear feature fusion according to the node embedding representation and the time series features to obtain node feature representation;

[0150] F. Based on the node feature representation, combined with a comprehensive understanding of the power system, realize real-time electro-carbon factor calculation, that is, input the node feature representation into the pre-trained electro-carbon factor evaluation model for carbon emission evaluation calculation to obtain the user-side electro-carbon factor of the power system;

[0151] G. Combine personalized electro-carbon factor calculation experience and analysis and optimization technologies to build a complete user-side dynamic electro-carbon factor evaluation system. On the one hand, it can output a more accurate user-side electro-carbon factor to the user, and on the other hand, it can feedback the analysis and optimization results or user personalized demand settings to the relevant processing models, so as to provide a more personalized evaluation service.

[0152] By adopting the technical solution provided by the embodiment of the present invention, the following remarkable effects can be brought:

[0153] 1. Transcend real-time limitations: Through local neighbor sampling and efficient graph structure representation, GCN achieves real-time and efficient learning of the node relationships in the power system. The introduction of LSTM significantly enhances the model's real-time modeling ability for time series data in the power system through reasonable selection of the sequence length. This enables the model to achieve more immediate evaluation of the electro-carbon factor in a dynamically changing power system, helps solve the problem of poor real-time performance in traditional methods, and provides users with more timely electro-carbon factor evaluations.

[0154] 2. Improve flexibility and adaptability: The embedding representation of GCN captures the complex relationships between the nodes in the power system, improving the flexibility and adaptability of the model. Compared with rule-based methods, GCN can better adapt to different power system structures and operating conditions. At the same time, through the application of multi-layer LSTM, the learning of abstract features of time series data in the power system is strengthened, and the temporal dependencies can be better captured, thus better adapting to changes at different time scales in the power system. In summary, the present invention improves the adaptability of the model and alleviates the problem of lack of flexibility in traditional methods.

[0155] 3. Solve the problem of limited evaluation accuracy: Through in-depth learning of the graph structure, GCN improves the accuracy of power system node representation. Compared with traditional statistical models, GCN can better adapt to emerging energy sources and complex changes in the power system. LSTM improves the computational accuracy for time series data through adjustment of the multi-layer structure and the number of hidden units. This helps overcome the problem of limited evaluation accuracy in traditional methods and makes the evaluation of the electro-carbon factor more reliable and accurate.

[0156] 4. Solve complex structure and data processing: By adopting the combination of GCN + LSTM, the present invention can optimize the structure and data processing flow for electro-carbon factor evaluation. The local neighbor sampling and graph structure representation of GCN improve the efficiency of data processing and reduce the complexity. At the same time, the selection of the sequence length and adjustment of the number of hidden units in LSTM make the processing of time series data more refined and efficient. The feature of this combination is that it can reduce the overall complexity while maintaining the model performance.

[0157] 5. Enrich the personalized user experience: The combined architecture of GCN + LSTM has stronger flexibility, enabling the model to better adapt to different user needs and provide a more personalized electro-carbon factor evaluation experience. The ability of GCN lies in capturing the relationships between nodes in the power system, while LSTM can capture individual differences in time series data. By integrating the two, the present invention can more comprehensively understand the user's consumption patterns and preferences and provide personalized electro-carbon factor evaluation results that better meet the actual needs of users.

[0158] 6. Improve adaptability to emerging energy sources: The combined model of GCN + LSTM has stronger adaptability, especially in dealing with emerging energy data. GCN can flexibly learn the characteristics of new energy nodes in the graph structure, while LSTM can better capture the time series changes of new energy. This comprehensive adaptability enables the model to better handle the complexity brought by the introduction of emerging energy sources in the power system, improving the accuracy and real-time performance of the electro-carbon assessment of these new energy sources.

[0159] Referring to Figure 3 , a structural block diagram of an electro-carbon factor dynamic assessment device provided by an embodiment of the present invention is shown, which may specifically include:

[0160] A power system data acquisition module 301, configured to acquire node information data and power time-varying data of the power system;

[0161] A node embedding representation generation module 302, configured to construct a node feature matrix and a node graph structure of the power system according to the node information data, and perform local optimization of the graph structure with combined node embedding update on the node feature matrix and the node graph structure to obtain a node embedding representation;

[0162] A time series optimization feature extraction module 303, configured to perform time series optimization feature extraction on the power time-varying data to obtain time series features;

[0163] A linear feature fusion module 304, configured to perform linear feature fusion according to the node embedding representation and the time series features to obtain a node feature representation;

[0164] A user-side electro-carbon factor calculation module 305, configured to calculate the user-side electro-carbon factor of the power system based on the node feature representation.

[0165] In an optional embodiment, the node information data includes the node information of all nodes in the power system and the connection relationship between any two nodes; the node embedding representation generation module 302 includes:

[0166] A node feature matrix construction module, configured to construct a node feature matrix of the power system according to the node information of all nodes;

[0167] An adjacency matrix construction module, configured to construct an adjacency matrix according to the connection relationship between any two nodes, and the adjacency matrix is used to represent the node graph structure of the power system.

[0168] In an optional embodiment, the node embedding representation generation module 302 further includes:

[0169] The node embedding dynamic update module is used to update the node feature matrix and the adjacency matrix respectively based on the node embedding dynamic update mechanism, and obtain the updated node feature matrix and the updated adjacency matrix;

[0170] The local neighbor sampling optimization module is used to, during the dynamic update process of the node feature matrix and the adjacency matrix, perform local neighbor sampling on each node respectively every time an update is completed, obtain the sampled neighbor set, and perform local feature extraction on the updated node feature matrix and the updated adjacency matrix respectively according to the sampled neighbor set, so as to obtain the local node feature matrix and the local adjacency matrix;

[0171] The multi-layer convolution module is used to input the local node feature matrix and the local adjacency matrix into a pre-trained graph neural network for multi-layer convolution to obtain the node embedding representation.

[0172] In an alternative embodiment, the node embedding dynamic update mechanism includes a node feature update mechanism and an adjacency matrix maintenance mechanism, and the node embedding dynamic update module includes:

[0173] The adjacency matrix update module is used to update the adjacency matrix based on the adjacency matrix maintenance mechanism, and the update process is as follows:

[0174]

[0175] The node feature update module is used to update the node feature matrix based on the node feature update mechanism, and the update process is as follows:

[0176]

[0177] where represents the weight of the adjacency relationship from node i to node j in the power system at time t; represents the degree of node i at time t, that is, the number of edges connected to node i; represents the degree of node j at time t, that is, the number of edges connected to node j; represents the weight of the adjacency relationship from node i to node j in the power system at time t - 1; represents the embedding representation of node i at time t; represents the activation function; W represents the learning weight parameter; represents the embedding representation of node i at time t - 1; represents the embedding representation of node j at time t - 1.

[0178] In an alternative embodiment, the time-varying power data includes time series data and node-related time-varying information; the timing optimization feature extraction module 303 includes:

[0179] A data preprocessing module, configured to perform data preprocessing on the time series data and the node-related time-varying information respectively, and integrate all the data obtained after data preprocessing to obtain spatio-temporal correlation data;

[0180] A multi-layer LSTM structure optimization module, configured to analyze the spatio-temporal correlation data, and based on the analysis result, adjust the time series processing length and the number of hidden units of a pre-trained multi-layer LSTM structure to dynamically adjust and optimize the multi-layer LSTM structure;

[0181] A timing feature extraction module, configured to perform timing feature extraction on the spatio-temporal correlation data based on the optimized multi-layer LSTM structure to obtain time series features.

[0182] In an alternative embodiment, the linear feature fusion module 304 is specifically configured to:

[0183] Based on the linear combination method of the weight matrix, perform linear feature fusion on the node embedding representation and the time series features to obtain a node feature representation. The linear feature fusion process is as follows:

[0184]

[0185] where Z i represents the node feature representation of node i; ReLU represents the rectified linear unit activation function; W fusion and b fusion represent the weight and bias parameters of the feature fusion layer respectively; [H i , h T represents concatenating the node embedding representation H i and the LSTM hidden state h T at the last time step for feature splicing.

[0186] In an alternative embodiment, the user-side electricity-carbon factor calculation module 305 is specifically configured to:

[0187] Input the node feature representation into a pre-trained electricity-carbon factor evaluation model for carbon emission evaluation calculation to obtain the user-side electricity-carbon factor of the power system;

[0188] where the electricity-carbon factor evaluation model adopts a regression model structure of a feedforward neural network with multiple hidden layers.

[0189] For the apparatus embodiments, since they are basically similar to the method embodiments, they are described relatively simply. For the relevant parts, refer to the corresponding descriptions in the foregoing method embodiments.

[0190] An embodiment of the present invention further provides an electronic device, which includes a processor and a memory:

[0191] The memory is used to store program codes and transmit the program codes to the processor;

[0192] The processor is used to execute the electro-carbon factor dynamic evaluation method according to any embodiment of the present invention according to the instructions in the program codes.

[0193] An embodiment of the present invention further provides a computer-readable storage medium, which is used to store program codes, and the program codes are used to execute the electro-carbon factor dynamic evaluation method according to any embodiment of the present invention.

[0194] Those skilled in the art can clearly understand that for the convenience and conciseness of description, the specific working processes of the above-described systems, apparatuses, and units can refer to the corresponding processes in the foregoing method embodiments, and will not be elaborated herein.

[0195] In several embodiments provided by the present invention, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative. For example, the division of the units is only a logical function division, and there can be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed couplings or direct couplings or communication connections to each other can be through some interfaces, and the indirect couplings or communication connections of the apparatuses or units can be in electrical, mechanical, or other forms.

[0196] The units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they can be located in one place, or they can be distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0197] In addition, in each embodiment of the present invention, the functional units can be integrated into one processing unit, or each unit can exist physically alone, or two or more units can be integrated into one unit. The above-mentioned integrated units can be implemented in the form of hardware or in the form of software functional units.

[0198] When the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as USB flash drives, mobile hard disks, read-only memories (ROMs), random access memories (RAMs), magnetic disks, or optical discs.

[0199] As described above, the above embodiments are only used to illustrate the technical solution of the present invention and are not intended to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements for some of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A method for dynamic evaluation of electric carbon factor, characterized in that: include: Obtain node information data and power time-varying data of the power system; According to the node information data, a node feature matrix and a node graph structure of the power system are constructed, and the node feature matrix and the node graph structure are locally optimized in combination with node embedding update to obtain a node embedding representation; Performing time series optimization feature extraction on the power time-varying data to obtain time series features; Performing linear feature fusion according to the node embedding representation and the time series features to obtain a node feature representation; Based on the node characteristic representation, calculating the user-side electric carbon factor of the power system; The node graph structure is represented by an adjacency matrix; The node feature matrix and the node graph structure are locally optimized in combination with node embedding update to obtain a node embedding representation, including: Based on the node embedding dynamic update mechanism, the node feature matrix and the adjacency matrix are updated respectively to obtain an updated node feature matrix and an updated adjacency matrix; In the dynamic update process of the node feature matrix and the adjacency matrix, each time an update is completed, local neighbor sampling is performed on each node to obtain a sampled neighbor set, and local features are extracted from the updated node feature matrix and the updated adjacency matrix according to the sampled neighbor set to obtain a local node feature matrix and a local adjacency matrix; Inputting the local node feature matrix and the local adjacency matrix into a pre-trained graph neural network for multi-layer convolution to obtain a node embedding representation; The node embedding dynamic update mechanism includes a node feature update mechanism and an adjacency matrix maintenance mechanism. The node feature matrix and the adjacency matrix are updated based on the node embedding dynamic update mechanism, respectively, including: Based on the adjacency matrix maintenance mechanism, the adjacency matrix is ​​updated, and the updating process is as follows: Based on the node feature update mechanism, the node feature matrix is ​​updated, and the update process is as follows: in, represents the weight of the adjacency relationship from node i to node j in the power system at time t; represents the degree of node i at time t, that is, the number of edges connected to node i; represents the degree of node j at time t, that is, the number of edges connected to node j; represents the weight of the adjacency relationship from node i to node j in the power system at time t-1; represents the embedding representation of node i at time t; σ represents the activation function; W represents the learning weight parameter; represents the embedding representation of node i at time t-1; represents the embedding representation of node j at time t-1.

2. The method for dynamic evaluation of electric carbon factor according to claim 1, characterized in that: The node information data includes node information of all nodes in the power system and connection relationships between each node; constructing a node feature matrix and a node graph structure of the power system according to the node information data includes: Constructing a node characteristic matrix of the power system according to the node information of all the nodes; An adjacency matrix is ​​constructed according to the connection relationship between each pair of nodes, and the adjacency matrix is ​​used to represent the node graph structure of the power system.

3. The method for dynamic evaluation of electric carbon factor according to claim 1, characterized in that: The power time-varying data includes time series data and node-related time-varying information; the time series optimization feature extraction of the power time-varying data to obtain the time series feature includes: Preprocessing the time series data and the node-related time-varying information respectively, and integrating all the data obtained after the data preprocessing to obtain spatiotemporal correlation data; Analyze the spatiotemporal correlation data, and based on the analysis results, adjust the time series processing length and the number of hidden units of the pre-trained multi-layer LSTM structure to dynamically adjust and optimize the multi-layer LSTM structure; Based on the optimized multi-layer LSTM structure, time series features are extracted from the spatiotemporal correlation data to obtain time series features.

4. The method for dynamic evaluation of electric carbon factor according to claim 3, characterized in that: The performing linear feature fusion according to the node embedding representation and the time series feature to obtain the node feature representation includes: Based on the linear combination method of the weight matrix, linear feature fusion is performed on the node embedding representation and the time series feature to obtain the node feature representation. The linear feature fusion process is as follows: WITH i =ReLU(W fusion ·[H i ,h T ]+b fusion ) Among them, Z i represents the node feature representation of node i; ReLU represents the rectified linear unit activation function; W fusion and b fusion Respectively represent the weight and bias parameters of the feature fusion layer; [H i ,h T ] means embedding the node into H i and the LSTM hidden state h at the last moment T Perform feature stitching.

5. The method for dynamic evaluation of electric carbon factor according to any one of claims 1 to 4, characterized in that: The calculating the user-side electric carbon factor of the power system based on the node characteristic representation includes: Inputting the node feature representation into a pre-trained electric carbon factor assessment model to perform carbon emission assessment calculation to obtain the user-side electric carbon factor of the power system; Among them, the electric carbon factor evaluation model adopts a regression model structure of a feedforward neural network with multiple hidden layers.

6. A device for dynamic evaluation of electric carbon factor, characterized in that: include: The power system data acquisition module is used to obtain node information data and power time-varying data of the power system; A node embedding representation generation module is used to construct a node feature matrix and a node graph structure of the power system according to the node information data, and perform local optimization of the graph structure combined with node embedding update on the node feature matrix and the node graph structure to obtain a node embedding representation; A time series optimization feature extraction module is used to extract time series optimization features from the power time-varying data to obtain time series features; A linear feature fusion module, used to perform linear feature fusion according to the node embedding representation and the time series feature to obtain a node feature representation; A user-side electric carbon factor calculation module, used to calculate the user-side electric carbon factor of the power system based on the node feature representation; The node graph structure is represented by an adjacency matrix; The node embedding representation generation module comprises: A node embedding dynamic update module, used to update the node feature matrix and the adjacency matrix respectively based on the node embedding dynamic update mechanism to obtain an updated node feature matrix and an updated adjacency matrix; A local neighbor sampling optimization module is used to perform local neighbor sampling on each node to obtain a sampled neighbor set each time an update is completed during the dynamic update process of the node feature matrix and the adjacency matrix, and to extract local features of the updated node feature matrix and the updated adjacency matrix based on the sampled neighbor set to obtain a local node feature matrix and a local adjacency matrix; A multi-layer convolution module, used to input the local node feature matrix and the local adjacency matrix into a pre-trained graph neural network for multi-layer convolution to obtain a node embedding representation; The node embedding dynamic update mechanism includes a node feature update mechanism and an adjacency matrix maintenance mechanism, and the node embedding dynamic update module includes: The adjacency matrix update module is used to update the adjacency matrix based on the adjacency matrix maintenance mechanism. The update process is as follows: The node feature update module is used to update the node feature matrix based on the node feature update mechanism. The update process is as follows: in, represents the weight of the adjacency relationship from node i to node j in the power system at time t; represents the degree of node i at time t, that is, the number of edges connected to node i; represents the degree of node j at time t, that is, the number of edges connected to node j; represents the weight of the adjacency relationship from node i to node j in the power system at time t-1; represents the embedding representation of node i at time t; σ represents the activation function; W represents the learning weight parameter; represents the embedding representation of node i at time t-1; represents the embedding representation of node j at time t-1.

7. An electronic device, characterized in that: The device comprises a processor and a memory: The memory is used to store program code and transmit the program code to the processor; The processor is used to execute the electric carbon factor dynamic evaluation method described in any one of claims 1-5 according to the instructions in the program code.

8. A computer-readable storage medium, characterized in that: The computer-readable storage medium is used to store program code, and the program code is used to execute the dynamic evaluation method of electric carbon factor according to any one of claims 1-5.

Images