Carbon emission flow calculation method and system of power system

Abstract

The invention discloses a carbon emission flow calculation method and system especially for an electric power system, and belongs to the technical field of carbon emission of the electric power system. The method comprises the steps of multi-modal data acquisition and fusion, dynamic graph construction, carbon flow distribution optimization based on combination of an adaptive genetic algorithm and reinforcement learning, carbon flow density analysis based on a graph neural network, carbon flow trend prediction based on a deep enhanced memory network, carbon emission flow feedback adjustment and carbon emission flow distribution. Through multi-level feedback control and path optimization, dynamic balance and adaptive distribution of the carbon emission flow of the electric power system are realized, the response efficiency of carbon flow distribution in the electric power system is improved, the stability and rationality of the carbon flow load under different conditions are guaranteed, and the adaptive capacity of dynamic change of the electric power system is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of carbon emission of power systems, and in particular to a method and system for calculating carbon emission flow of power systems. Background Art

[0002] The calculation of carbon emission flow in the current power system is of key significance for system optimization, achievement of environmental protection goals and policy compliance. The accurate calculation of carbon emission flow in the power system helps to track and regulate carbon emissions at each node in real time, thereby achieving effective management of carbon emissions. However, the existing power system carbon emission flow calculation method (such as the power flow method) (Chinese invention patent, publication number: CN118747575A, name: A method for calculating carbon emission flow in a power system) mainly relies on the power flow distribution matrix and basic power parameters such as nodes and branches of the power system. The voltage, angle, power, etc. of each node are calculated by the Newton-Raphson method to construct a carbon flow model, and the carbon flow direction is further derived based on the distribution of these physical parameters. The defect of this method is that it mainly focuses on the static distribution of carbon flow, and does not combine dynamic feedback mechanisms and intelligent optimization strategies. It has limited ability to optimize and improve the carbon emission flow in the system, and it is difficult to quickly respond and adjust the distribution of carbon emission flow when the carbon flow load fluctuates. Summary of the invention

[0003] In view of the many problems existing in the above-mentioned prior art, the present invention provides a method and system for calculating the carbon emission flow of the power system. The present invention realizes the dynamic allocation optimization and adaptive adjustment of the carbon emission flow of the power system through the fusion of multimodal data and dynamic graph construction technology, in conjunction with intelligent algorithms. Based on the carbon flow density, the present invention can realize real-time path optimization and load prediction, generate a carbon flow allocation scheme to dynamically balance the carbon emission load of the system, and improve the adaptability of the power system to dynamic changes.

[0004] A method for calculating carbon emission flow of a power system comprises the following steps:

[0005] Collecting multimodal data of the power system, performing feature fusion on the multimodal data, and generating a dynamic adjacency matrix to construct dynamic graph data of the power system;

[0006] Based on the dynamic graph data, an adaptive genetic algorithm and multi-agent reinforcement learning are used to optimize the distribution of carbon emission flows of power system nodes to generate coordinated distribution optimization data of the power system;

[0007] Based on the collaborative allocation optimization data, in combination with the regional feedback mechanism and the global feedback mechanism, the carbon emission flow intensity of the power system nodes is dynamically adjusted to generate feedback control data of the power system to ensure the balance of carbon emission flow within the power system;

[0008] Based on the feedback control data, a graph neural network is constructed to generate carbon emission flow density data between nodes of the power system, and the carbon emission flow transmission path of the power system is optimized by dynamic weight adjustment between nodes;

[0009] Based on the carbon emission flow density data, combined with multi-level trend prediction and adaptive fuzzy control, the carbon emission flow of the power system is feedback-regulated to generate the final carbon emission flow allocation data of the power system.

[0010] Preferably, the multimodal data includes real-time parameter data, environmental monitoring data and equipment status data of the power system, and the feature fusion extracts the spatiotemporal features of the multimodal data by constructing a convolutional network to generate dynamic graph data of the power system.

[0011] Preferably, the allocation optimization based on the dynamic graph data of the power system includes performing preliminary optimization allocation of the carbon emission flows of the power system nodes by an adaptive genetic algorithm, wherein the adaptive genetic algorithm evaluates the allocation scheme of each power system node based on a fitness function, wherein the fitness function is:

[0012]

[0013] Among them, F i Represents the fitness value of power system node i; N i represents the set of adjacent nodes connected to power system node i; W ij represents the weight of power system node i and adjacent node j; P j represents the carbon emission flow distribution of power system node j;

[0014] The fitness function is used to determine the optimal allocation plan, and the obtained allocation plan is used as the initial input of multi-agent reinforcement learning to further optimize the allocation strategy of carbon emission flows between nodes in the power system and generate collaborative allocation optimization data for the power system.

[0015] Preferably, the multi-agent reinforcement learning includes constructing an intelligent agent for each power system node, optimizing the carbon emission flow allocation strategy of each power system node through distributed training, each intelligent agent updates the strategy according to the status information of the power system node and the feedback from adjacent nodes, and the intelligent agents interact with each other through the dynamic adjacency matrix of the power system. After each round of training iteration, the local learning results are shared to achieve coordinated carbon emission flow allocation among the power system nodes.

[0016] Preferably, the dynamic adjustment based on the collaborative allocation optimization data of the power system includes combining the regional feedback mechanism and the global feedback mechanism to dynamically adjust the carbon emission flow intensity of the power system nodes to generate feedback control data of the power system and achieve dynamic balance of the carbon emission flow within the power system.

[0017] Preferably, the regional feedback mechanism is used to perform real-time regulation of the carbon emission flow of local nodes of the power system, and the global feedback mechanism is used to regulate the overall carbon emission flow of the power system to ensure the balance of carbon emission flow intensity at the regional and global layers of the power system.

[0018] Preferably, the carbon emission flow density calculation based on the feedback control data of the power system is performed through a graph neural network model, which takes the real-time status data of each node of the power system as input, performs convolution calculation on the features of each node through the adjacency matrix of the power system in each propagation layer, and generates the carbon emission flow density data of each power system node. The carbon emission flow density data D i Calculated by the following formula:

[0019]

[0020] Among them, D i represents the carbon emission flow density of power system node i; N i A represents the set of adjacent nodes connected to power system node i; ij represents the edge weight between power system node i and adjacent node j in the adjacency matrix of the power system; X j represents the characteristic data of power system node j; W represents the weight matrix of the graph convolution layer; σ represents the activation function;

[0021] The carbon emission flow density data is used to describe the distribution of carbon emission flows between power system nodes, and optimize the weight of the carbon emission flow transmission path according to the correlation and coordination relationship between power system nodes.

[0022] Preferably, the convolution calculation process in the graph neural network model includes aggregating and updating the features of adjacent nodes in the power system to generate the carbon emission flow density of each node in the power system, and adaptively adjusting the weights of the transmission paths between the nodes in the power system through the power system node influence propagation model, wherein the power system node influence is calculated by the transmission frequency and load weight of the power system node to construct a high-priority carbon emission flow transmission path.

[0023] Preferably, feedback regulation based on the carbon emission flow density data of the power system includes using a deep enhanced memory network to predict the short-term, medium-term and long-term trends of the carbon emission flow of the power system. The deep enhanced memory network generates multi-level prediction data of the carbon emission flow of the power system by analyzing the historical status data and current trend characteristics of the carbon emission flow of each node in the power system, and calculates the probability of the trend path of the multi-level prediction data in combination with Markov chain decision analysis. The carbon emission flow distribution of the power system is feedback regulated in real time through adaptive fuzzy control to achieve load adaptive balancing of the carbon emission flow of the power system.

[0024] A system for implementing the carbon emission flow calculation method of the power system, the system comprising:

[0025] A data acquisition module, used to collect multimodal data of the power system, wherein the multimodal data includes real-time parameter data, environmental monitoring data and equipment status data of the power system;

[0026] A feature fusion module, used for performing feature fusion on the multimodal data to generate a dynamic adjacency matrix to construct dynamic graph data of the power system;

[0027] A distribution optimization module is used to optimize the distribution of carbon emission flows of power system nodes based on the dynamic graph data by using an adaptive genetic algorithm and multi-agent reinforcement learning to generate collaborative distribution optimization data of the power system;

[0028] A feedback control module is used to dynamically adjust the carbon emission flow intensity of the power system nodes based on the collaborative allocation optimization data, combined with the regional feedback mechanism and the global feedback mechanism, and generate feedback control data of the power system to ensure the balance of carbon emission flow in the power system;

[0029] A carbon flow density calculation module is used to construct a graph neural network based on the feedback control data to generate carbon emission flow density data between nodes of the power system, and optimize the carbon emission flow transmission path of the power system by adjusting the dynamic weights between nodes;

[0030] A feedback regulation module is used to perform feedback regulation on the carbon emission flow of the power system based on the carbon emission flow density data, combined with multi-level trend prediction and adaptive fuzzy control, to generate the final carbon emission flow distribution data of the power system, so as to achieve dynamic load adaptive balance of the carbon emission flow of the power system.

[0031] Compared with the prior art, the advantages and beneficial effects of the present invention are:

[0032] The present invention realizes dynamic monitoring, intelligent optimization and adaptive feedback regulation of carbon emission flows in power systems through technical means such as adaptive genetic algorithms, multi-agent reinforcement learning, graph neural networks and deep enhanced memory networks, effectively making up for the shortcomings of existing technologies in dynamic carbon emission flow management. Through the combination of adaptive genetic algorithms and reinforcement learning, the present invention can collaboratively optimize carbon flow distribution between different nodes; using the feature aggregation calculation of graph neural networks, the scheme can intelligently analyze the carbon flow density between nodes; and the deep enhanced memory network provides multi-level prediction capabilities for carbon flow loads. The present invention not only improves the dynamic response speed of carbon flow distribution, but also realizes real-time dynamic balance of carbon emission flow loads, ensuring the scientificity and accuracy of carbon emission management and improving the adaptability of power systems to dynamic changes. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 It is a schematic diagram of the process of the present invention;

[0034] Figure 2 This is a schematic diagram of the structure of the intelligent optimization model of the present invention;

[0035] Figure 3 It is a schematic diagram of feedback mechanism adjustment and adaptive fuzzy control of the present invention;

[0036] Figure 4 It is a schematic diagram of the trend prediction model and decision analysis of the present invention;

[0037] Figure 5 It is a structural block diagram of the system of the present invention. DETAILED DESCRIPTION

[0038] Hereinafter, embodiments of the present disclosure will be described with reference to the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of the present disclosure. In the following detailed description, for ease of explanation, many specific details are set forth to provide a comprehensive understanding of the embodiments of the present disclosure. However, it is obvious that one or more embodiments may also be implemented without these specific details. In addition, in the following description, descriptions of known structures and technologies are omitted to avoid unnecessary confusion of the concepts of the present disclosure.

[0039] The terms used herein are only for describing specific embodiments and are not intended to limit the present disclosure. The terms "comprise", "include", etc. used herein indicate the existence of the features, steps, operations and / or components, but do not exclude the existence or addition of one or more other features, steps, operations or components.

[0040] All terms (including technical and scientific terms) used herein have the meanings commonly understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used herein should be interpreted as having a meaning consistent with the context of this specification, and should not be interpreted in an idealized or overly rigid manner.

[0041] like Figure 1 As shown, a method for calculating carbon emission flow of a power system includes the following steps:

[0042] Collecting multimodal data of the power system, performing feature fusion on the multimodal data, and generating a dynamic adjacency matrix to construct dynamic graph data of the power system;

[0043] Preferably, the multimodal data includes real-time parameter data, environmental monitoring data and equipment status data of the power system, and the feature fusion extracts the spatiotemporal features of the multimodal data by constructing a convolutional network to generate dynamic graph data of the power system.

[0044] In the present invention, multimodal data is collected and feature fusion is performed to generate a dynamic adjacency matrix to construct dynamic graph data of the power system, laying the foundation for the dynamic allocation and optimization of carbon emission flows.

[0045] First, multimodal data collection is an important step in realizing carbon emission flow modeling. To ensure the accurate distribution of carbon emission flow, the system needs to collect data in multiple dimensions, including real-time parameter data of the power system, environmental monitoring data and equipment status data. Among them, real-time parameter data mainly reflects the power load, power generation and transmission efficiency of each node in the power system, which can directly characterize the operating status and load conditions of the power system and affect the distribution of carbon emissions; environmental monitoring data includes information such as temperature, humidity and wind speed. These climate factors significantly affect the dynamic flow and transmission efficiency of carbon emissions, and are important factors in building an accurate carbon emission flow model; equipment status data records the current operating status, loss rate and maintenance status of power generation equipment and transmission equipment, which directly determines the occurrence and distribution of carbon flow, and plays a vital role in the balance of carbon emission flow and transmission stability of the entire system.

[0046] After completing the multimodal data collection, the system enters the feature fusion stage. Feature fusion is to extract and uniformly represent multi-dimensional and multi-source data features to support dynamic modeling of carbon emission flows. The present invention uses a convolutional neural network (CNN) to achieve this feature fusion. Convolutional neural networks can capture the spatiotemporal relationship in multimodal data, and extract key features from different modal data by sliding the convolution kernel in the spatial and temporal dimensions. Specifically, the feature fusion process can be expressed as the following calculation expression:

[0047]

[0048] Among them, H t represents the fusion feature extracted at time t; W i is the weight matrix of the convolution kernel; X i represents the data input of different modes; f is the activation function, and n is the total number of nodes; the use of the activation function introduces nonlinear characteristics, which enables the model to better capture complex spatiotemporal features. In the above process, the convolution kernel extracts the spatiotemporal features of each modal data through layer-by-layer convolution and fuses them into a unified spatiotemporal feature H t This result can effectively characterize the comprehensive state of the system at time t and provide key input data for the subsequent construction of the dynamic adjacency matrix.

[0049] After completing feature fusion, the system generates a dynamic adjacency matrix based on the extracted features. The generation of the dynamic adjacency matrix depends on the fusion features output by the convolutional network, combined with the relationship between nodes and parameters such as carbon emission weights, and updates the flow path of the carbon emission flow according to real-time changing data. Dynamic adjacency matrix A t The calculation formula is as follows:

[0050]

[0051] Among them, A t represents the dynamic adjacency matrix at time t; α ij is the association weight between node i and node j; H i and Represent the fusion features of node i and node j respectively; is the transposition of the feature of node j. This formula realizes the dynamic update of the system carbon emission flow relationship, so that the carbon emission flow path between nodes can timely reflect the current state of the system, and ensure the efficient transmission of carbon flow within the system by adjusting the association weight.

[0052] Through the feature fusion of multimodal data and the generation of dynamic adjacency matrix, the system can model the carbon emission flow of the power system in time and space and capture the real-time carbon flow relationship between nodes. Feature fusion enhances the system's ability to process multimodal data, making the model of the carbon emission flow of the power system more temporal and spatially accurate, while the generation of dynamic adjacency matrix enables the system to dynamically adjust the carbon flow distribution path between nodes according to real-time data, improving the response speed to changes in the distribution of carbon emission flow. In practical applications, for example, when a key node causes an increase in carbon emission flow due to excessive load, the convolutional neural network extracts the features of multimodal data such as the load state and temperature of the node, and the convolution kernel captures the characteristics of high load and temperature increase in space and time. Subsequently, the dynamic adjacency matrix automatically adjusts the association weights with other nodes according to the high load state of the node to reduce the concentration of carbon flow, thereby achieving adaptive balance of the system's carbon emission flow and avoiding the impact of excessive load on system stability. In summary, by collecting multimodal data of the power system, feature fusion, and dynamic adjacency matrix generation, the dynamic graph data of carbon emission flow was constructed, and the dynamic adjustment of the carbon flow distribution path was realized, providing a data basis and path optimization capability for the adaptive optimization of carbon flow in the power system.

[0053] Based on the dynamic graph data, an adaptive genetic algorithm and multi-agent reinforcement learning are used to optimize the distribution of carbon emission flows of power system nodes to generate coordinated distribution optimization data of the power system;

[0054] In the present invention, through the combination of adaptive genetic algorithm based on dynamic graph data and multi-agent reinforcement learning, the carbon emission flow of power system nodes can be efficiently allocated and optimized, and collaborative allocation optimization data can be generated. The design principle of this step is to preliminarily optimize the carbon flow allocation strategy through genetic algorithm based on the real-time dynamically constructed graph data, and then use multi-agent reinforcement learning (Multi-Agent Reinforcement Learning, MARL) to achieve collaborative regulation among multiple nodes to dynamically adapt to the carbon emission demand and load conditions of each node.

[0055] First, the adaptive genetic algorithm is mainly used for the preliminary allocation of carbon emission flows at each node in the power system. The core of the genetic algorithm is to evaluate the allocation strategy through the individual fitness function and continuously optimize the allocation of carbon flow by using genetic operations (selection, crossover and mutation).

[0056] In the genetic algorithm, through multiple iterations, the fitness function is used to evaluate the adaptability of each candidate allocation strategy, select the strategy with high fitness to retain, and generate a new generation of allocation schemes through crossover and mutation operations. In this way, the initially optimized carbon emission flow allocation scheme gradually tends to be balanced across the entire system.

[0057] like Figure 2As shown, after the initial allocation is completed, the multi-agent reinforcement learning mechanism intervenes to achieve collaborative optimization of carbon flow allocation. The basic principle of multi-agent reinforcement learning is to build an agent for each node. The agent continuously updates its carbon flow allocation strategy through interaction with surrounding nodes (i.e., adjacent nodes) under the reinforcement learning framework to achieve collaborative regulation of multiple nodes. In the present invention, the agent of each node selects the optimal allocation strategy through comprehensive information of the environment state, adjacent node state, and its own carbon flow allocation state, and updates the strategy according to the feedback of adjacent nodes in each iteration. Specifically, each agent calculates the profit function R according to the environment and node state. i , and continuously optimize the allocation strategy under the guidance of reinforcement learning algorithms such as Q-learning or policy gradient. i It can be defined as follows:

[0058] R i =α×U i -β×C i

[0059] Among them, R i represents the benefit value of node i; U i represents the carbon flow allocation efficiency of the node; C i is the carbon flow load cost of the node; α and β are trade-off parameters; used to adjust the weight between allocation efficiency and cost. The agent adjusts the allocation scheme by maximizing its benefit function to ensure the dynamic balance of carbon flow between nodes.

[0060] Through the preliminary allocation of the adaptive genetic algorithm and the collaborative optimization of multi-agent reinforcement learning, the final carbon emission flow allocation scheme achieves dynamic balance within the system. Specifically, the adaptive genetic algorithm provides a global preliminary balance, while the multi-agent reinforcement learning makes fine adjustments based on the dynamic load and state of each node. In principle, this combination gives full play to the global optimization ability of the genetic algorithm and the local adaptability of reinforcement learning, so that the carbon flow allocation not only meets the global balance requirements of the system, but also has the ability to adapt to the dynamic changes of nodes.

[0061] In practical applications, for example, when a node in the power grid experiences a sudden increase in load, the initially allocated carbon flow scheme may cause the load on neighboring nodes to be too high, thereby breaking the local balance. Through multi-agent reinforcement learning, the agents of adjacent nodes automatically adjust the allocation strategy according to the benefit function, reducing the carbon flow delivery to the nodes with increased load, thereby dispersing the load and maintaining the balance of the system's carbon emission flow. Ultimately, the scheme dynamically and collaboratively allocates carbon flows to ensure load balance and adaptive allocation of carbon emission flows during power system operation, thereby achieving efficient and stable operation of the power system's carbon flow.

[0062] Preferably, the allocation optimization based on the dynamic graph data of the power system includes performing preliminary optimization allocation of the carbon emission flows of the power system nodes by an adaptive genetic algorithm, wherein the adaptive genetic algorithm evaluates the allocation scheme of each power system node based on a fitness function, wherein the fitness function is:

[0063]

[0064] Among them, F i Represents the fitness value of power system node i; N i represents the set of adjacent nodes connected to power system node i; W ij represents the weight of power system node i and adjacent node j; P j represents the carbon emission flow distribution of power system node j;

[0065] The fitness function is used to determine the optimal allocation plan, and the obtained allocation plan is used as the initial input of multi-agent reinforcement learning to further optimize the allocation strategy of carbon emission flows between nodes in the power system and generate collaborative allocation optimization data for the power system.

[0066] In the present invention, the key to allocation optimization is to preliminarily optimize the carbon emission flow of power system nodes through adaptive genetic algorithms. In this scheme, the adaptive genetic algorithm is mainly used to globally optimize the carbon flow of each node in the power system based on dynamic graph data, thereby forming an optimal initial allocation scheme and providing efficient initial conditions for subsequent multi-agent reinforcement learning. The adaptive genetic algorithm is a global search algorithm based on biological evolution. Through iterative calculation, it continuously optimizes the carbon flow distribution of each node in the power system so as to achieve an overall balance of carbon flow in the whole system.

[0067] In the present invention, the allocation optimization process of the adaptive genetic algorithm revolves around a core - the fitness function F i Expand. Fitness function F i It is used to evaluate the effectiveness of the allocation scheme of each node and reflect the contribution of the node to the overall carbon emission flow balance of the system under a specific allocation strategy. The calculation formula of the fitness function is shown above.

[0068] In specific applications, the fitness function calculates the fitness value of the node by accumulating the carbon flow allocation of each adjacent node and its weight on the current node (that is, the relative influence between nodes). The value generated by the fitness function represents the degree of balance or rationality of the node under the current allocation strategy. The evolution of each generation of the algorithm is based on the value of the fitness function to carry out survival of the fittest for different allocation schemes: allocation schemes with higher fitness are retained first, and the next generation of allocation schemes are generated through genetic operations such as crossover and mutation. With multiple generations of iterations, the adaptive genetic algorithm gradually converges to a globally optimal initial carbon flow allocation scheme, achieving a preliminary balance of carbon flow.

[0069] The effect of this preliminary optimization is to achieve a relatively reasonable initial distribution of carbon emission flows throughout the system, thereby reducing the situation where local nodes in the system are overloaded or carbon emission flows are unbalanced. The balance optimization provided by the adaptive genetic algorithm at the global level lays the foundation for subsequent multi-agent reinforcement learning, allowing reinforcement learning to start from better initial conditions and focus on fine-tuning and local optimization of each node to achieve the final dynamic balance in the power system.

[0070] Preferably, assuming that some nodes in a power system need to share more carbon emission flows due to their geographical location or equipment characteristics, the initial state may cause the load to be concentrated near these nodes. Under the action of the fitness function of the adaptive genetic algorithm, the fitness values ​​of these nodes may be low, so that in the multi-generation evolution of the genetic algorithm, by redistributing weights, the carbon flow distribution spreads to the low-load area to avoid the overload problem of a single node. At this time, the optimal allocation plan obtained by the adaptive genetic algorithm is the initial input of the multi-agent reinforcement learning. The reinforcement learning mechanism can be further refined according to this optimized plan to achieve the coordinated allocation and dynamic adjustment of carbon emission flows between nodes.

[0071] In summary, the adaptive genetic algorithm generates a globally balanced initial allocation plan for carbon emission flows by evaluating the fitness function of the power system nodes. This optimization process ensures the global balance of the initial allocation plan through the gradual improvement of the fitness function, provides a good foundation for the carbon flow management of the entire power system, makes the subsequent collaborative optimization of multi-agent reinforcement learning more targeted and efficient, and finally generates the collaborative allocation optimization data of the power system.

[0072] Preferably, the multi-agent reinforcement learning includes constructing an intelligent agent for each power system node, optimizing the carbon emission flow allocation strategy of each power system node through distributed training, each intelligent agent updates the strategy according to the status information of the power system node and the feedback from adjacent nodes, and the intelligent agents interact with each other through the dynamic adjacency matrix of the power system. After each round of training iteration, the local learning results are shared to achieve coordinated carbon emission flow allocation among the power system nodes.

[0073] In the present invention, multi-agent reinforcement learning (MARL) is used to collaboratively optimize the carbon emission flow allocation strategy of each node in the system. Multi-agent reinforcement learning gradually optimizes the carbon flow allocation strategy by constructing an intelligent agent for each power system node based on a distributed reinforcement learning training mechanism, thereby achieving a dynamic balance between nodes as a whole. The core of this solution is to utilize the local adaptive characteristics of reinforcement learning, so that each intelligent agent can autonomously adjust its allocation strategy and interact with other intelligent agents based on its own state information and feedback from adjacent nodes.

[0074] In the present invention, the intelligent agent of each power system node continuously updates the allocation strategy based on the state information of the power system node and the feedback of the adjacent nodes. This strategy update process is based on the profit function design in reinforcement learning, and the intelligent agent selects the optimal carbon flow allocation strategy by maximizing the profit function.

[0075] Revenue function R i is defined as follows:

[0076] R i =α×U i -β×C i

[0077] Among them, R i represents the revenue value of power system node i; U i represents the carbon flow allocation efficiency of node i, that is, the load rate of unit carbon flow on the node; C i represents the carbon flow load cost of node i; α and β are trade-off parameters, which respectively control the weights of carbon flow allocation efficiency and load cost in the benefit calculation.

[0078] The intelligent agent adjusts its carbon emission flow allocation strategy according to the result of the profit function in each iteration. The profit function R i The design ensures that when nodes choose allocation strategies, they not only pursue the efficiency of carbon flow allocation, but also need to consider the constraints of node load costs, so that the transmission of carbon flow will not be concentrated on a single node, thereby effectively avoiding system imbalance.

[0079] In each training iteration, intelligent agents interact with each other through a dynamic adjacency matrix. t Update the interrelationships between nodes in real time, and determine the interaction weights between nodes based on node status and carbon flow allocation strategy. The interaction formula is as follows:

[0080] A t (i,j)=γ×S ij ×(R i +Rj )

[0081] Among them, A t (i, j) represents the interaction weight between node i and node j at time t; γ is the interaction influence factor; S ij is the similarity parameter between node i and node j, reflecting the degree of cooperation between the two nodes in carbon flow allocation; R i and R j Represent the profit values ​​of node i and node j respectively.

[0082] Each intelligent agent shares local learning results with the intelligent agents of adjacent nodes under the guidance of the dynamic adjacency matrix. The shared content includes the current carbon flow allocation strategy and revenue situation, so that the adjacent nodes can optimize their allocation strategy based on the revenue results of other nodes. After multiple rounds of iterations, each intelligent agent gradually achieves a balanced distribution of carbon emission flows in the whole system through revenue maximization and local sharing.

[0083] In a specific embodiment, for example, in a high-load area of ​​the power grid, the intelligent agent of a node generates a higher load cost C due to the high load. i , resulting in its revenue function R i At this time, the intelligent agent reduces the input carbon flow by adjusting the allocation strategy, and shares the adjustment information with the adjacent node intelligent agent through the dynamic adjacency matrix. The adjacent node intelligent agent perceives its load adjustment needs through revenue sharing, and actively assists in sharing part of the carbon flow load. After multiple iterations, the allocation strategy is gradually optimized, and the carbon flow of high-load nodes gradually diffuses to the low-load area, achieving a dynamic balance of the system carbon flow.

[0084] Through the collaborative optimization of carbon emission flow distribution achieved by multi-agent reinforcement learning, the system has the ability to adaptively adjust under dynamic load environments, ensuring that the carbon flow distribution within the power system is optimized between high efficiency and balance, thereby meeting the dynamic regulation needs of carbon emission flows and improving the system's operating stability and the overall efficiency of carbon flow distribution.

[0085] like Figure 3 As shown, based on the collaborative allocation optimization data, combined with the regional feedback mechanism and the global feedback mechanism, the carbon emission flow intensity of the power system node is dynamically adjusted to generate feedback control data of the power system to ensure the balance of carbon emission flow in the power system;

[0086] In the present invention, by dynamically adjusting the collaborative allocation optimization data, combined with the regional feedback mechanism and the global feedback mechanism, the balanced regulation of the intensity of the carbon emission flow of nodes in the power system is achieved. The principle of this step is to use a multi-level feedback regulation method: the local node carbon flow is fine-tuned through the regional feedback mechanism to achieve the balance of the local carbon flow, and the distribution of the overall carbon emission flow is macro-regulated through the global feedback mechanism. The feedback control data finally generated is based on these dynamic adjustment processes to ensure the dynamic balance of the system carbon emission flow, thereby achieving efficient and stable operation of the power system under a dynamic load environment.

[0087] First, the role of the regional feedback mechanism is to respond to the distribution status of carbon emission flows between local nodes in real time. It adjusts the carbon emission flow intensity of each node by monitoring the load level and carbon flow distribution efficiency of each node and its adjacent nodes. Assume that the carbon emission flow intensity of each node i is I i , the intensity is dynamically adjusted according to the distribution difference between adjacent nodes. Specifically, the flow intensity adjustment amount of node i is calculated as follows:

[0088]

[0089] Among them, ΔI i represents the adjustment amount of the carbon emission flow intensity of node i; λ is the adjustment coefficient, which is used to control the amplitude of the adjustment amount; N i is the set of adjacent nodes of node i; F j and F i They represent the carbon flow allocation efficiency of adjacent node j and node i respectively, reflecting the load status and carbon flow utilization of the node.

[0090] The regional feedback mechanism fine-tunes the carbon flow intensity of the node through the above formula. When the load of node i is significantly higher than that of the surrounding nodes, its carbon emission flow intensity I i It will automatically decrease to avoid the local carbon flow being too concentrated on the high-load node. At the same time, the carbon flow intensity of the adjacent nodes will increase accordingly to share the carbon flow pressure of the high-load node, thereby achieving a dynamic balance of carbon flow in the local range.

[0091] Different from the regional feedback mechanism, the global feedback mechanism adjusts the carbon emission flow intensity of each node in the whole system to ensure the balance of the overall load. The global feedback mechanism dynamically adjusts the carbon flow by calculating the deviation of each node from the average load of the system based on the load distribution of all nodes in the system. The adjustment formula of global feedback is as follows:

[0092]

[0093] in, represents the carbon emission flow intensity of node i after global feedback regulation; γ1 is the global regulation factor, which controls the regulation amplitude; M represents the set of all nodes in the system; F k is the load status of node k, reflecting the carbon flow allocation efficiency and load status of the current node; |M| is the total number of node sets, representing the number of nodes in the whole system.

[0094] Through the above global feedback adjustment formula, the carbon emission flow intensity in the whole system can be balanced according to the overall load condition. For nodes with load lower than the system average, the feedback mechanism will increase their carbon flow intensity. This can improve the load rate of the node, enable it to play a greater role in carbon flow distribution, and then disperse the pressure of other high-load nodes to achieve load balancing at the system level.

[0095] The combination of regional feedback mechanism and global feedback mechanism constructs the multi-level feedback control method of the present invention. The regional feedback mechanism quickly responds to local carbon flow imbalances and achieves carbon flow balance within the region through local adjustments between adjacent nodes; the global feedback mechanism adjusts the carbon flow intensity of each node based on the overall load condition of the system to ensure the balance of the system's carbon flow distribution at the macro level. The generated feedback control data records the carbon emission flow intensity, local load, and system balance adjustment of each node, providing a reference for the monitoring and optimization of the carbon flow status of subsequent nodes.

[0096] Taking the high-load area as an example, when the carbon emission flow of nodes in a certain area is concentrated, resulting in increased local load pressure, the regional feedback mechanism will first fine-tune the carbon flow intensity of the nodes in the area, and achieve regional balance by reducing the carbon flow of high-load nodes and increasing the intensity of low-load nodes. If the overall load of the area still exceeds the system average, the global feedback mechanism will reduce the carbon flow intensity of nodes in other areas within the system, introduce more carbon flow to the area, further disperse the load pressure, and achieve a dynamic balance of carbon emission flow in the entire system. Ultimately, the coordinated regulation of regional feedback and global feedback ensures the stable operation of the power system under complex load environments, making the distribution of carbon emission flows efficient and flexible.

[0097] Preferably, the dynamic adjustment based on the collaborative allocation optimization data of the power system includes combining the regional feedback mechanism and the global feedback mechanism to dynamically adjust the carbon emission flow intensity of the power system nodes to generate feedback control data of the power system and achieve dynamic balance of the carbon emission flow within the power system.

[0098] In the process of calculating the carbon emission flow of the power system of the present invention, in order to achieve the dynamic balance of the carbon emission flow in the system, the dynamic adjustment of the coordinated allocation optimization data is combined with the regional feedback mechanism and the global feedback mechanism to achieve the intelligent regulation of the carbon emission flow intensity of the power system node. This method coordinates from the local and overall levels of the system to ensure the balance of the global load of the system while meeting the needs of local nodes, thereby generating feedback control data to provide support for the carbon flow balance of the system.

[0099] The principle of the regional feedback mechanism is to adjust the carbon emission flow intensity of each node in real time based on the load status of adjacent nodes. This mechanism can allocate the intensity of carbon flow to a single node. i Fine-tuning is performed to prevent local overload or carbon flow concentration, and to ensure the balance of carbon flow distribution in the local area. The specific adjustment process can be expressed as the following formula:

[0100]

[0101] Among them, ΔI i represents the regulation amount of the carbon emission flow intensity of node i; α is the local regulation coefficient, which determines the sensitivity of the regulation amount; N i is the set of adjacent nodes of node i; F j and F i They represent the carbon flow allocation efficiency of adjacent node j and node i respectively, reflecting the load status and carbon flow utilization of the node.

[0102] Through this formula, the carbon emission flow intensity of node i will be automatically adjusted under the influence of adjacent nodes. When the load of node i is significantly higher than that of its adjacent nodes, ΔI i The value is negative, resulting in I i The traffic intensity of adjacent nodes with lighter loads will be increased accordingly, so that the carbon emission flow in the local area is evenly distributed among the nodes. This kind of regulation based on adjacency can quickly respond to local load changes and ensure the load balance of local nodes.

[0103] On the other hand, the global feedback mechanism is to macro-regulate the carbon emission flow intensity of each node based on the overall load of the system. This mechanism adjusts the carbon flow intensity of each node by comparing the load of each node with the average level of the overall load of the system to achieve global balance of the system. The global feedback adjustment formula is:

[0104]

[0105] in, represents the carbon emission flow intensity of node i after global feedback regulation; γ1 is the global regulation factor, which controls the regulation amplitude; M represents the set of all nodes in the system; F kis the load status of node k, reflecting the carbon flow allocation efficiency and load status of the current node; |M| is the total number of node sets, representing the number of nodes in the whole system.

[0106] The global feedback mechanism adjusts the carbon flow intensity of the node according to the load of the whole system. When the load of node i is lower than the average level of the system, the adjustment amount in the feedback formula is positive, that is, Increase, thereby increasing the carbon flow intensity of node i, so as to share the burden of other high-load nodes, and vice versa. Through the global feedback mechanism, the system can effectively control the load level of each node and ensure the uniform distribution of carbon flow in the global range of the system.

[0107] Through the coordinated regulation of the regional feedback mechanism and the global feedback mechanism, a two-layer dynamic balance of the local and global carbon emission flows of the system is achieved. The regional feedback mechanism quickly responds to the load changes of adjacent nodes, so that the nodes can flexibly respond to the carbon flow allocation needs in the local area; the global feedback mechanism balances the carbon flow intensity of each node at the macro level to prevent uneven load on the entire system. The feedback control data finally generated contains the real-time carbon flow intensity adjustment information of each node for dynamic carbon flow allocation and load monitoring of the system. Through this multi-level feedback control, the present invention ensures the carbon flow allocation efficiency and system stability of the power system under different load conditions, especially under peak load or abnormal operating conditions, and achieves efficient carbon flow management and balance.

[0108] Preferably, the regional feedback mechanism is used to perform real-time regulation of the carbon emission flow of local nodes of the power system, and the global feedback mechanism is used to regulate the overall carbon emission flow of the power system to ensure the balance of carbon emission flow intensity at the regional and global layers of the power system.

[0109] In the present invention, the regional feedback mechanism and the global feedback mechanism work together to achieve a dynamic balance of carbon emission flows in the system, thereby ensuring that the power system can achieve reasonable carbon flow distribution and load balancing at both the local node and the overall system level. The core of this step is to introduce a hierarchical regulation strategy into carbon flow control, adjust the carbon emission flow intensity of each node through real-time regional feedback and systematic global feedback, and finally generate feedback control data to ensure the global balance of carbon emission flows in the power system.

[0110] The principle of the regional feedback mechanism is to fine-tune the carbon emission flow intensity of the node through the mutual influence between adjacent nodes at the local level of the power system in order to dynamically adapt to the fluctuation of local load. This mechanism adjusts the carbon flow intensity I of each node under the current load state. i Make real-time adjustments.

[0111] When the load of node i is significantly higher than that of adjacent nodes, the adjustment amount ΔI iWhen the value is negative, the carbon flow intensity of the node decreases, thus effectively alleviating the situation of local overload. At the same time, the adjacent nodes with lower load will automatically increase the carbon flow intensity to share the carbon flow demand of the high-load nodes and achieve balanced distribution of local carbon flow. This mechanism can quickly respond to load changes between adjacent nodes and ensure the dynamic balance of carbon flow distribution in the local area.

[0112] The global feedback mechanism regulates the intensity of carbon emission flow from the overall system level to ensure load balance between regions. The global feedback mechanism is based on the principle of load balancing across the entire system and globally adjusts the carbon flow intensity of each node.

[0113] When the load of a node is significantly lower than the average level of the system, feedback regulation will increase the carbon emission flow intensity of the node, thereby effectively balancing the load of the entire system. For nodes with loads significantly higher than the average level, the system will appropriately reduce their carbon flow intensity to ensure that the system load is evenly distributed. The regulatory effect of the global feedback mechanism is reflected in the macroscopic balance of the global carbon flow distribution of the system. By adjusting the load of each node, the carbon flow transmission of the system tends to be uniform, avoiding the situation where a single node or area is overloaded.

[0114] The synergy of the regional feedback mechanism and the global feedback mechanism constructs a multi-level dynamic adjustment scheme, which ensures the balance of carbon flow distribution in the power system both locally and globally. The regional feedback mechanism achieves load balancing in the local area by responding locally to adjacent nodes, so that each node can quickly adapt to local changes and adjust in real time; the global feedback mechanism starts from the overall load of the system to ensure the balance of the global load. The feedback control data generated by the two mechanisms contains the real-time carbon flow adjustment information of each node, providing data support for system monitoring and dynamic adjustment, so that the carbon emission flow can effectively adapt to changes under complex load environments.

[0115] For example, when the local carbon flow intensity of a node in a certain area of ​​the power system increases due to increased load demand, the regional feedback mechanism will first make local adjustments to direct part of the carbon flow intensity to adjacent low-load nodes to achieve regional balance. If the overall load of the region is higher than the system average, the global feedback mechanism will impose adjustments on low-load nodes in other regions to increase their carbon flow intensity to share the local load, thereby ensuring the balance of carbon flow intensity throughout the system. This two-level adjustment method ensures that the system can maintain dynamic balance and efficient distribution of carbon flow under different load conditions, thereby optimizing the overall carbon flow management and load distribution effects of the power system.

[0116] Based on the feedback control data, a graph neural network is constructed to generate carbon emission flow density data between nodes of the power system, and the carbon emission flow transmission path of the power system is optimized by dynamic weight adjustment between nodes;

[0117] In the present invention, a graph neural network is constructed based on feedback control data to generate carbon emission flow density data between nodes in the power system, and to optimize the carbon emission flow transmission path of the power system through dynamic weight adjustment between nodes. The core of this step is to utilize the topological modeling capability of the graph neural network to represent the carbon flow relationship of nodes in the power system as a graph structure, and optimize the distribution and transmission path of the carbon flow through the training and dynamic weight adjustment of the graph neural network to ensure the efficient transmission of the carbon emission flow within the system. Specifically, a graph neural network (GNN) is a neural network model suitable for graph structure data. In the present invention, the nodes of the power system and their carbon flow relationships are represented as a dynamic graph structure, in which each node represents a key node in the power system (such as a power plant, substation, etc.), and the edge represents the transmission path of the carbon flow between the nodes. The input feature vector of each node is provided by the feedback control data, including the carbon flow intensity, load status and feedback information of the adjacent nodes of the node. Assume that the input feature vector of node i is x i , then the node features realize the modeling of carbon flow density through the aggregation function and update function of the graph neural network. The multi-layer information transmission formula of the graph neural network is:

[0118]

[0119] in, represents the feature representation of node i at layer t+1, σ is the activation function, usually ReLU or LeakyReLU, W (t) is the weight matrix of the tth layer, represents the feature representation of node j at layer t, b (t) is the bias term, and N(i) represents the set of adjacent nodes of node i. This formula describes the information transfer process in the graph neural network, that is, the feature representation of node i is updated by the feature aggregation of its adjacent nodes, reflecting the carbon flow distribution relationship of node i in the graph structure. Through multi-layer information transmission, the carbon emission flow density data of each node is finally obtained. This data represents the distribution of the carbon flow intensity of each node in the transmission path, which helps to determine the importance and flow concentration of each node in the system in the carbon flow transmission. The carbon flow density calculation formula is:

[0120]

[0121] Among them, D i represents the carbon flow density of node i, α ij is the dynamic weight between node i and node j, reflecting the priority of carbon flow allocation between the two nodes, h jis the characteristic representation of node j. Through the carbon flow density data, the system can clearly understand the carbon flow load of each node to achieve load balancing in the transmission path optimization. Next, by adjusting the dynamic weights between nodes, the system further optimizes the transmission path of the carbon emission flow. The dynamic weight adjustment between nodes is based on the carbon flow density and feedback control data of the nodes to ensure that the transmission of carbon flow within the system is more efficient. The dynamic weight update formula is:

[0122]

[0123] in, represents the dynamic weights of nodes i and j at time t+1, represents the dynamic weights of nodes i and j at time t, γ2 is the weight update coefficient, which is used to control the adjustment amplitude, D i and D j Represent the carbon flow density of node i and node j respectively. Through this adjustment formula, when the carbon flow density of node i is significantly higher than that of node j, the weight adjustment will increase by α ij , thereby reducing the inflow pressure of carbon to high-density nodes and improving the load balancing of the system.

[0124] Through the graph neural network constructed based on feedback control data, the system can dynamically generate carbon emission flow density data for each node, and optimize the carbon flow transmission path through dynamic weight adjustment between nodes, thereby achieving the following significant results: first, the generated carbon flow density data can reflect the carbon flow load of each node in real time, so that the system can quickly identify carbon flow intensive areas and provide data support for the optimization of the transmission path; secondly, the dynamic weight adjustment between nodes enables nodes with higher carbon flow concentration to appropriately reduce the input load, while improving the flow transmission of low-density nodes, thereby achieving carbon flow load balancing within the power system; finally, the optimized carbon flow path distribution helps to reduce the overload risk of local nodes within the system, improve the stability and efficiency of the power system in carbon flow transmission, and ensure the smooth operation of the carbon flow management system in a high-load environment.

[0125] Preferably, the carbon emission flow density calculation based on the feedback control data of the power system is performed through a graph neural network model, which takes the real-time status data of each node of the power system as input, performs convolution calculation on the features of each node through the adjacency matrix of the power system in each propagation layer, and generates the carbon emission flow density data of each power system node. The carbon emission flow density data D i Calculated by the following formula:

[0126]

[0127] Among them, D i represents the carbon emission flow density of power system node i; N iA represents the set of adjacent nodes connected to power system node i; ij represents the edge weight between power system node i and adjacent node j in the adjacency matrix of the power system; X j represents the characteristic data of power system node j; W represents the weight matrix of the graph convolution layer; σ represents the activation function;

[0128] The carbon emission flow density data is used to describe the distribution of carbon emission flows between power system nodes, and optimize the weight of the carbon emission flow transmission path according to the correlation and coordination relationship between power system nodes.

[0129] In the present invention, the calculation of carbon emission flow density data is based on the Graph Neural Network (GNN) model, through which the real-time calculation and optimization of the carbon flow distribution of each node in the system are realized. Specifically, the graph neural network model takes the real-time status data of each node in the power system as input, and performs convolution calculation on the characteristics of each node through the adjacency matrix in each propagation layer, and finally generates the carbon emission flow density data of each node. This density data is used to reflect the distribution of carbon emission flows between nodes in the power system, and at the same time optimizes the weight of the carbon flow transmission path according to the correlation and synergy between nodes, so as to achieve efficient and balanced transmission of carbon flow in the system.

[0130] In the process of each layer of propagation, the graph neural network realizes the transmission of information between nodes through the adjacency matrix A. The carbon emission flow density D of each node i It is a weighted aggregation of the features of adjacent nodes. The weight matrix W is used to control the influence of different features on the current node, and the activation function σ is applied to enhance the nonlinear expression ability of the model. This process enables nodes to not only obtain information about their direct adjacent nodes, but also understand the status of more distant nodes at a deeper level through multi-layer convolution stacking, thereby establishing a carbon flow synergy relationship between remote nodes in the system.

[0131] The carbon emission flow density data D generated by the graph neural network model i It reflects the degree of carbon flow concentration at each node in the power system and provides key data support for optimizing the carbon flow transmission path. Specifically, nodes with high carbon flow density indicate high carbon flow concentration, while low-density nodes reflect the dispersion of carbon flow. In order to further optimize the weight of the transmission path, the system dynamically adjusts the edge weight A between nodes according to the carbon flow density of each node and the correlation degree of adjacent nodes. ij , ensuring the efficiency and balance of carbon flow transmission. The edge weight adjustment formula is as follows:

[0132]

[0133] in, is the edge weight between node i and node j after update, represents the weight between node i and node j at time t; γ3 is the adjustment coefficient, which controls the weight adjustment amplitude; I i and I j Represent the influence of nodes i and j respectively.

[0134] This dynamic weight adjustment mechanism ensures that high-density nodes can reduce input carbon flow to avoid overload, and guide carbon flow to low-density nodes to achieve a balanced distribution of carbon flow in the system. Through this optimization process, the graph neural network model can continuously and dynamically update the carbon emission flow density data and continuously optimize the carbon flow transmission path during system operation.

[0135] Through the multi-layer propagation of the graph neural network model and the dynamic adjustment of the adjacency matrix, the carbon emission flow density in the power system can be calculated and distributed in real time. This process enables the system to quickly identify areas where carbon flow is concentrated under different node load conditions, and timely optimize the carbon flow transmission path to reduce local overload. At the same time, the dynamic weight adjustment mechanism ensures the stable operation of the system under peak load or complex working conditions, so that the carbon flow transmission path can achieve the best state in terms of efficiency and load balancing, thereby improving the carbon emission management efficiency and stability of the entire power system.

[0136] Preferably, the convolution calculation process in the graph neural network model includes aggregating and updating the features of adjacent nodes in the power system to generate the carbon emission flow density of each node in the power system, and adaptively adjusting the weights of the transmission paths between the nodes in the power system through the power system node influence propagation model, wherein the power system node influence is calculated by the transmission frequency and load weight of the power system node to construct a high-priority carbon emission flow transmission path.

[0137] In the present invention, the convolution calculation process of the graph neural network model is used to aggregate and update the characteristics of adjacent nodes in the power system to generate carbon emission flow density data for each node. This convolution process not only takes into account the direct characteristics of each node, but also integrates the status of its adjacent nodes through the information transfer model to form a distribution analysis of the carbon emission flow of the power system. In addition, the weight of the transmission path is adaptively adjusted through the power system node influence propagation model to ensure that the carbon emission flow can be preferentially transmitted through high-influence nodes, thereby constructing a high-priority carbon flow transmission path. The node influence is calculated through the transmission frequency and load weight of the node, which specifically affects the weight update strategy of the graph neural network.

[0138] Specifically, during the convolution calculation process of the graph neural network, the carbon emission flow density D of each node iIt is the weighted aggregation result of the characteristics of its adjacent nodes. This process integrates the status of adjacent nodes through the aggregation function, so that each node is affected by its adjacent nodes while considering its own characteristics, thus more comprehensively reflecting the distribution of carbon emission flows in the system.

[0139] Through this convolution calculation formula, the system can effectively aggregate the local neighborhood characteristics of each node to accurately characterize its carbon emission flow density. At the same time, the node influence propagation model further optimizes the weight adjustment in the transmission path. The calculation of node influence is based on two core elements: transmission frequency and load weight. i It can be expressed as:

[0140] I i =f i × i

[0141] Among them, I i is the influence of node i; f i represents the transmission frequency of node i, that is, the activity of the node in the carbon flow transmission in the system; w i is the load weight of node i, which indicates the node's ability to withstand carbon flow load.

[0142] In the influence propagation model, the transmission frequency f i The load weight w is calculated by recording the number or frequency of the node participating in the carbon flow transmission. i The influence determines the priority of the node in the carbon flow path, so nodes with high influence will undertake more carbon flow transmission tasks, and the system will assign higher path weights to these nodes.

[0143] Based on the node influence, the system adaptively adjusts the edge weights in the adjacency matrix to ensure that carbon flow preferentially passes through paths with high influence. The edge weight adjustment formula is as follows:

[0144]

[0145] in, is the edge weight between node i and node j after update, represents the weight between node i and node j at time t; γ3 is the adjustment coefficient, which controls the weight adjustment amplitude; I i and I j Represent the influence of nodes i and j respectively.

[0146] This formula ensures that when the influence of node i is greater than that of node j, the edge weight A ijIncrease, making the carbon flow more inclined to transfer from node i to adjacent node j. At the overall system level, this weight adjustment process gradually optimizes the carbon flow transmission path, allowing the carbon flow to preferentially pass through high-priority nodes, thereby improving the system's carbon flow transmission efficiency and load balancing.

[0147] The convolution calculation process of the graph neural network realizes the accurate modeling of the carbon flow density of the power system, and the node influence propagation model further optimizes the carbon flow transmission path. Through the dynamic calculation of node influence and adaptive weight adjustment, the system can prioritize the construction of high-influence paths, so that carbon flows flow to the most suitable nodes, avoid carbon flow backlogs in low-influence paths, and achieve system-wide carbon flow load balancing. In practical applications, this solution can effectively improve the transmission efficiency of carbon emission flows within the power system, ensuring that the system can maintain efficient carbon flow management and path optimization when the load changes or the carbon flow is unevenly distributed.

[0148] Based on the carbon emission flow density data, combined with multi-level trend prediction and adaptive fuzzy control, the carbon emission flow of the power system is feedback-regulated to generate the final carbon emission flow allocation data of the power system.

[0149] like Figure 4 As shown, the feedback regulation based on the carbon emission flow density data of the power system includes using a deep enhanced memory network to predict the short-term, medium-term and long-term trends of the carbon emission flow of the power system. The deep enhanced memory network generates multi-level prediction data of the carbon emission flow of the power system by analyzing the historical status data and current trend characteristics of the carbon emission flow of each node in the power system, and calculates the probability of the trend path of the multi-level prediction data in combination with Markov chain decision analysis. The carbon emission flow distribution of the power system is regulated in real time by adaptive fuzzy control to achieve load adaptive balancing of the carbon emission flow of the power system.

[0150] In the present invention, by combining carbon emission flow density data, multi-level trend prediction and adaptive fuzzy control, feedback regulation of carbon emission flow is realized to generate the final carbon emission flow distribution data of the power system, ensuring the adaptive load balance of the system at different time scales. Specifically, the process includes three core steps: using Deep Enhanced Memory Network (DEMN) for short-term, medium-term and long-term trend prediction, combining Markov chain decision analysis to determine the probability distribution of trend paths, and real-time feedback regulation of carbon emission flow distribution based on adaptive fuzzy control.

[0151] First, the deep enhanced memory network is used in the present invention to predict the short-term, medium-term and long-term trends of carbon emission flows of each node in the power system. The model performs multi-level analysis on the historical state data and current trend characteristics of the carbon emission flow of each node, and the generated multi-level prediction data covers the carbon flow distribution changes of the nodes in different periods. Let the historical carbon flow feature vector of node i be The DEMN model can be used to predict the short-term (e.g. minute level), medium-term (e.g. hour level) and long-term (e.g. day level) carbon flow trends of the node. for:

[0152]

[0153] in, represents the short-term carbon flow trend prediction of node i; represents the medium-term carbon flow trend; Represents the long-term carbon flow trend.

[0154] Through DEMN's multi-layer memory units (such as LSTM or GRU), the system can simultaneously capture the short-term fluctuations, medium-term trends, and long-term cyclical characteristics of node carbon flow changes. DEMN's deep memory of historical carbon flow data and comprehensive analysis of current trends enable the system to have full-cycle prediction capabilities and provide reliable trend data support for subsequent adjustments.

[0155] Next, the system applies Markov chain decision analysis to calculate the probability distribution of trend paths based on the multi-level prediction data generated by DEMN. This analysis process quantifies the transition probability of carbon flow trend paths by establishing a state transition matrix, helping the system to weigh different possible trend paths. For example, if the carbon flow trend state transition probability of node i is P ij , then the state transfer matrix P is expressed as:

[0156]

[0157] Among them, P ij Represents the probability of a node transferring from state i to state j. Through this matrix, the system can derive the probability distribution of different trend paths and provide a quantitative basis for the path selection of carbon flow allocation. Based on Markov chain analysis, the system can identify the carbon flow allocation mode with a higher probability in the trend path in advance, so as to select the optimal path in carbon flow regulation and ensure the smooth operation of the system.

[0158] Finally, based on the multi-level prediction data and path probability distribution, the system uses adaptive fuzzy control to achieve feedback regulation of carbon emission flow. Adaptive fuzzy control inputs the prediction data and the current load status of the node into the fuzzy logic system to generate real-time regulation instructions for carbon flow distribution. Assume that the current load of node i is Li , the average load of the system is L avg , then the rules of adaptive fuzzy control are as follows:

[0159] If L i Much higher than L avg , the system reduces the intensity of carbon flow allocated to node i;

[0160] If L i Close to L avg , keep the current allocation;

[0161] If L i Lower than L avg , increasing the carbon flow intensity allocated to node i.

[0162] The regulation output of fuzzy control adjusts the carbon flow load of each node in real time to ensure the balance of carbon flow distribution at each node in the system. By combining the trend prediction data generated by DEMN with the probability path of the Markov chain, the feedback regulation of adaptive fuzzy control continuously corrects the carbon flow distribution in actual operation, giving the system the adaptive ability to cope with load fluctuations and trend changes.

[0163] This feedback regulation process enables the system to have global judgment and adaptive control capabilities for short-term, medium-term and long-term carbon flow trends. The multi-level prediction data provided by the deep enhanced memory network enables the system to achieve accurate trend prediction in carbon flow load fluctuations; the trend path probability distribution of Markov chain analysis helps the system identify the optimal allocation path in complex trend paths; adaptive fuzzy control dynamically allocates carbon flow loads based on real-time feedback regulation to ensure the balance and stability of carbon flow in the power system under different load conditions. This method realizes adaptive allocation and load balancing of carbon flow in multi-level feedback regulation, thereby improving the carbon flow management efficiency and responsiveness of the system.

[0164] A system for implementing the carbon emission flow calculation method of the power system, the system comprising:

[0165] The data acquisition module is used to collect multimodal data of the power system, which includes real-time parameter data, environmental monitoring data and equipment status data of the power system; the data acquisition module is responsible for collecting multimodal data of the power system, mainly including real-time parameter data, environmental monitoring data and equipment status data. In specific applications, the data comes from the sensor network and monitoring equipment distributed in the power system. These devices can collect the power status, external meteorological conditions and equipment operation status of the nodes in real time to form multi-dimensional data input. The collected data is passed to the feature fusion module after preprocessing.

[0166] The feature fusion module is used to perform feature fusion on the multimodal data and generate a dynamic adjacency matrix to construct dynamic graph data of the power system. The feature fusion module extracts spatiotemporal features from the multimodal data through deep learning methods such as convolutional neural networks, and generates a dynamic adjacency matrix to construct dynamic graph data of the power system. This dynamic graph data represents the dynamic relationship between nodes in the power system.

[0167] The allocation optimization module is used to optimize the allocation of carbon emission flows of power system nodes based on the dynamic graph data using adaptive genetic algorithms and multi-agent reinforcement learning to generate collaborative allocation optimization data of the power system; the allocation optimization module uses adaptive genetic algorithms and multi-agent reinforcement learning (MARL) to optimize the allocation of carbon emission flows of power system nodes based on dynamic graph data. The adaptive genetic algorithm first performs a preliminary optimization of the carbon flow allocation, finds a preliminary allocation plan by evaluating the fitness of each node, and then passes the plan as input to the multi-agent reinforcement learning module. Multi-agent reinforcement learning further optimizes the carbon flow allocation plan through collaboration and strategy updates between agents to ensure that the carbon emission flow is efficiently allocated among different nodes in the system.

[0168] The feedback control module is used to dynamically adjust the carbon emission flow intensity of the power system nodes based on the collaborative allocation optimization data, combined with the regional feedback mechanism and the global feedback mechanism, and generate feedback control data for the power system to ensure the balance of carbon emission flow within the power system; the feedback control module is responsible for generating feedback control data based on the allocation optimization results to dynamically adjust the carbon flow intensity of the power system nodes to achieve the balance of the system carbon flow. The feedback control module includes a regional feedback mechanism and a global feedback mechanism: the regional feedback mechanism is used to fine-tune the local nodes of the system, while the global feedback mechanism is responsible for comprehensive adjustments to the carbon flow of the entire system to ensure the balance of the carbon flow distribution of the entire system. The generated feedback control data will guide the next step of path optimization.

[0169] The carbon flow density calculation module is used to construct a graph neural network based on the feedback control data to generate carbon emission flow density data between power system nodes, and optimize the carbon emission flow transmission path of the power system through dynamic weight adjustment between nodes; in the carbon flow density calculation module, the graph neural network (GNN) constructed based on the feedback control data generates carbon emission flow density data between nodes, and optimizes the transmission path through dynamic weight adjustment. The convolution calculation of the graph neural network can aggregate the characteristics of adjacent nodes, making the carbon flow density information of each node more comprehensive, and the optimized weight distribution ensures the stability and efficiency of the transmission path.

[0170] The feedback regulation module is used to perform feedback regulation on the carbon emission flow of the power system based on the carbon emission flow density data, combined with multi-level trend prediction and adaptive fuzzy control, to generate the final carbon emission flow allocation data of the power system, so as to achieve dynamic load adaptive balance of the carbon emission flow of the power system. The feedback regulation module uses the carbon emission flow density data, combined with trend prediction and adaptive fuzzy control to adjust the system carbon flow in real time. The deep enhanced memory network (DEMN) predicts the short-term, medium-term and long-term trends of the carbon flow in this module, and combines the decision analysis of the Markov chain to provide probabilistic support for path selection for carbon flow allocation. Finally, the adaptive fuzzy control adjusts the carbon flow allocation in real time according to the trend prediction and the current load status, and finally generates the carbon flow allocation data of the power system to ensure the adaptive balance of the system under the dynamic load environment.

[0171] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems or computer program products. Therefore, the present application may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware.

[0172] The above are only embodiments of the present application and are not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included in the scope of the claims of the present application.

Claims

1. A method for calculating carbon emission flow of a power system, characterized in that: The following steps are involved: Collecting multimodal data of the power system, performing feature fusion on the multimodal data, and generating a dynamic adjacency matrix to construct dynamic graph data of the power system; Based on the dynamic graph data, an adaptive genetic algorithm and multi-agent reinforcement learning are used to optimize the distribution of carbon emission flows of power system nodes to generate coordinated distribution optimization data of the power system; Based on the collaborative allocation optimization data, in combination with the regional feedback mechanism and the global feedback mechanism, the carbon emission flow intensity of the power system nodes is dynamically adjusted to generate feedback control data of the power system to ensure the balance of carbon emission flow within the power system; Based on the feedback control data, a graph neural network is constructed to generate carbon emission flow density data between nodes of the power system, and the carbon emission flow transmission path of the power system is optimized by dynamic weight adjustment between nodes; Based on the carbon emission flow density data, combined with multi-level trend prediction and adaptive fuzzy control, the carbon emission flow of the power system is feedback-regulated to generate the final carbon emission flow allocation data of the power system.

2. The method for calculating carbon emission flow of a power system according to claim 1, characterized in that: The multimodal data includes real-time parameter data, environmental monitoring data and equipment status data of the power system. The feature fusion extracts the spatiotemporal features of the multimodal data by constructing a convolutional network to generate dynamic graph data of the power system.

3. The method for calculating carbon emission flow of a power system according to claim 1, characterized in that: The allocation optimization based on the dynamic graph data of the power system includes performing preliminary optimization allocation of the carbon emission flow of the power system nodes by using an adaptive genetic algorithm, wherein the adaptive genetic algorithm evaluates the allocation scheme of each power system node based on a fitness function, wherein the fitness function is: Among them, F i Represents the fitness value of power system node i; N i represents the set of adjacent nodes connected to power system node i; W ij represents the weight of power system node i and adjacent node j; P j represents the carbon emission flow distribution of power system node j; The fitness function is used to determine the optimal allocation plan, and the obtained allocation plan is used as the initial input of multi-agent reinforcement learning to further optimize the allocation strategy of carbon emission flows between nodes in the power system and generate collaborative allocation optimization data for the power system.

4. The method for calculating carbon emission flow of a power system according to claim 1, characterized in that: The multi-agent reinforcement learning includes building an intelligent agent for each power system node, optimizing the carbon emission flow allocation strategy of each power system node through distributed training, each intelligent agent updates the strategy according to the state information of the power system node and the feedback of adjacent nodes, and the intelligent agents interact with each other through the dynamic adjacency matrix of the power system. After each round of training iteration, the local learning results are shared to achieve coordinated carbon emission flow allocation among the power system nodes.

5. The method for calculating carbon emission flow of a power system according to claim 1, characterized in that: The dynamic adjustment based on the collaborative allocation optimization data of the power system includes combining the regional feedback mechanism and the global feedback mechanism to dynamically adjust the carbon emission flow intensity of the power system nodes to generate feedback control data of the power system and achieve dynamic balance of the carbon emission flow within the power system.

6. The method for calculating carbon emission flow of a power system according to claim 1, characterized in that: The regional feedback mechanism is used to perform real-time regulation of the carbon emission flow of local nodes in the power system, and the global feedback mechanism is used to regulate the overall carbon emission flow of the power system to ensure the balance of carbon emission flow intensity at the regional and global levels of the power system.

7. The method for calculating carbon emission flow of a power system according to claim 1, characterized in that: The carbon emission flow density calculation based on the feedback control data of the power system is performed through a graph neural network model. The graph neural network model takes the real-time status data of each node in the power system as input, and performs convolution calculation on the features of each node through the adjacency matrix of the power system in each propagation layer to generate the carbon emission flow density data of each power system node. The carbon emission flow density data D i Calculated by the following formula: Among them, D i represents the carbon emission flow density of power system node i; N i A represents the set of adjacent nodes connected to power system node i; ij represents the edge weight between power system node i and adjacent node j in the adjacency matrix of the power system; X j represents the characteristic data of power system node j; W represents the weight matrix of the graph convolution layer; σ represents the activation function; The carbon emission flow density data is used to describe the distribution of carbon emission flows between power system nodes, and optimize the weight of the carbon emission flow transmission path according to the correlation and coordination relationship between power system nodes.

8. The method for calculating carbon emission flow of a power system according to claim 7, characterized in that: The convolution calculation process in the graph neural network model includes aggregating and updating the features of adjacent nodes in the power system to generate the carbon emission flow density of each node in the power system, and adaptively adjusting the weights of the transmission paths between the nodes in the power system through the power system node influence propagation model, where the power system node influence is calculated by the transmission frequency and load weight of the power system nodes to construct a high-priority carbon emission flow transmission path.

9. The method for calculating carbon emission flow of a power system according to claim 1, characterized in that: Feedback regulation based on the carbon emission flow density data of the power system includes using a deep enhanced memory network to predict the short-term, medium-term and long-term trends of the carbon emission flow of the power system. The deep enhanced memory network generates multi-level prediction data of the carbon emission flow of the power system by analyzing the historical status data and current trend characteristics of the carbon emission flow of each node in the power system, and calculates the probability of the trend path of the multi-level prediction data in combination with Markov chain decision analysis. Through adaptive fuzzy control, real-time feedback regulation of the carbon emission flow distribution of the power system is performed to achieve load adaptive balancing of the carbon emission flow of the power system.

10. A system for implementing the carbon emission flow calculation method of the power system according to any one of claims 1 to 9, characterized in that: The system comprises: A data acquisition module, used to collect multimodal data of the power system, wherein the multimodal data includes real-time parameter data, environmental monitoring data and equipment status data of the power system; A feature fusion module, used for performing feature fusion on the multimodal data to generate a dynamic adjacency matrix to construct dynamic graph data of the power system; A distribution optimization module is used to optimize the distribution of carbon emission flows of power system nodes based on the dynamic graph data by using an adaptive genetic algorithm and multi-agent reinforcement learning to generate collaborative distribution optimization data of the power system; A feedback control module is used to dynamically adjust the carbon emission flow intensity of the power system nodes based on the collaborative allocation optimization data, combined with the regional feedback mechanism and the global feedback mechanism, and generate feedback control data of the power system to ensure the balance of carbon emission flow in the power system; A carbon flow density calculation module is used to construct a graph neural network based on the feedback control data to generate carbon emission flow density data between nodes of the power system, and optimize the carbon emission flow transmission path of the power system by adjusting the dynamic weights between nodes; A feedback regulation module is used to perform feedback regulation on the carbon emission flow of the power system based on the carbon emission flow density data, combined with multi-level trend prediction and adaptive fuzzy control, to generate the final carbon emission flow distribution data of the power system, so as to achieve dynamic load adaptive balance of the carbon emission flow of the power system.

Images