Power grid carbon footprint dynamic management method based on digital twinning enhancement

Abstract

The invention discloses a power grid carbon footprint dynamic management method based on digital twin strengthening. The method comprises the steps that a power grid digital twinborn model of key physical assets in a power grid is constructed, and a full-life-cycle carbon footprint accounting engine is developed to dynamically calculate the carbon emission of the power grid in each stage from raw material exploitation to retirement recovery; designing and training an intensified training agent, performing deep learning through interaction with the environment in a simulation environment provided by a power grid digital twin model and a full-life-cycle carbon footprint accounting engine, and continuously optimizing a decision strategy; and performing deep fusion and collaboration on the power grid digital twin model, the full-life-cycle carbon footprint accounting engine and the enhanced practice agent, so that the enhanced practice agent outputs an optimal low-carbon decision strategy according to a power grid real-time state provided by the power grid digital twin model. According to the invention, through deep fusion of digital twinning and reinforcement learning technologies, the intelligent level and emission reduction actual effect of power grid carbon footprint management are significantly improved.

Description

Technical Field

[0001] This invention relates to the field of dynamic management technology of power grid carbon footprint, specifically to a dynamic management method for power grid carbon footprint based on digital twin enhancement. Background Technology

[0002] Currently, the severe challenges posed by global climate change are accelerating the energy and power industry's transition to low-carbon practices. As the core carrier of the energy system, the power grid's life-cycle carbon footprint management has become a crucial link in achieving the "dual carbon" goals. Traditionally, power grid carbon footprint accounting has primarily relied on inventory methods or life cycle assessments (LCA) based on historical data and static models. While these methods have laid the foundation for carbon emission assessment, they generally suffer from the following problems: 1. Lack of static accounting and real-time perception: Existing carbon footprint accounting relies heavily on historical data and static models, which cannot reflect the changes in carbon emissions of the power grid under complex operating conditions in real time and dynamically, and lacks the ability to accurately and realistically map the carbon emissions of physical entities in real time.

[0003] 2. Lack of adaptability in optimization strategies: Traditional carbon emission reduction optimization methods are mainly based on preset rules or experience models, which are difficult to effectively cope with dynamic and complex environments such as power grid load fluctuations and new energy grid connection, and cannot achieve adaptive and intelligent decision-making for carbon emissions.

[0004] 3. Insufficient coordination in full life cycle management: Existing technologies lack systematic integration for the carbon footprint management of power grids from planning and design to decommissioning and recycling. Carbon emission optimization at each stage is carried out independently, making it difficult to achieve global optimization and coordinated emission reduction. Summary of the Invention

[0005] The purpose of this invention is to address the shortcomings of existing technologies by providing a dynamic management method for the carbon footprint of power grids based on digital twin enhancement.

[0006] To achieve the above objectives, this invention provides a dynamic management method for the carbon footprint of power grids based on digital twin enhancement, comprising: A digital twin model of key physical assets in the power grid is constructed. The digital twin model integrates multi-source heterogeneous data to achieve real-time mapping and two-way synchronization between the physical world and the digital space. Based on the aforementioned power grid digital twin model, a full life cycle carbon footprint accounting engine was developed to obtain real-time updates of equipment operating status, energy consumption data, and material status from the digital twin, and to dynamically calculate the carbon emissions of the power grid at each stage from raw material mining to decommissioning and recycling. The reinforcement training agent is designed and trained. The reinforcement training agent performs deep learning by interacting with the environment in the simulation environment provided by the power grid digital twin model and the life cycle carbon footprint accounting engine, and continuously optimizes its decision-making strategy. The power grid digital twin model, the life-cycle carbon footprint accounting engine, and the reinforcement training agent are deeply integrated and coordinated so that the reinforcement training agent can output the optimal low-carbon decision-making strategy based on the real-time power grid status provided by the power grid digital twin model.

[0007] Furthermore, the multi-source heterogeneous data is collected from the acquisition and monitoring control system, sensors, geographic information system, equipment ledger, operation and maintenance records, and video surveillance based on the data acquisition module.

[0008] Furthermore, the multi-source heterogeneous data is transmitted in real time via OPC UA and MQTT protocols, and preprocessed and aligned using a data fusion algorithm to achieve centimeter-level spatial positioning and millisecond-level time synchronization between the physical world and the power grid digital twin model.

[0009] Furthermore, each type of power grid digital twin model is linked to its full life-cycle carbon emission factor database, and the carbon emissions of the equipment unit are calculated in real time: ; in, To calculate the carbon emissions of the equipment unit, For the i-th activity data, Let be the emission factor corresponding to the i-th activity data.

[0010] Furthermore, the life-cycle carbon footprint accounting engine includes: The data interface layer is used to subscribe in real time to equipment operating status, energy consumption data, material status and environmental parameters from the power grid digital twin model via API or message queue; The accounting logic layer is used to dynamically calculate the carbon emissions of the power grid at each stage from raw material extraction to decommissioning and recycling. The data storage layer is used to store the calculated carbon emissions; The presentation layer provides a visualization interface for carbon footprint, including overall carbon emission trends, the proportion of carbon emissions at each stage, and the carbon emission contribution of key equipment, enabling real-time monitoring and tracing of carbon emissions.

[0011] Furthermore, the reinforcement training agent includes: The state space is used to reflect comprehensive information about the operation of the power grid in real time. Action space, used to provide a set of executable discrete or continuous decisions; The reward function is used to guide the reinforcement training agent to learn a strategy that minimizes carbon emissions over its entire life cycle.

[0012] Furthermore, the reward function is expressed as: ; in, To reinforce the actions performed by the training agent at time t, The increase in carbon emissions per unit time. For operating costs, Penalties for violating operational constraints , , These are the weighting coefficients.

[0013] Beneficial Effects: This invention significantly improves the intelligence level and emission reduction effectiveness of power grid carbon footprint management by deeply integrating digital twin and reinforcement learning technologies. By constructing a high-fidelity digital twin model of the power grid's physical entities, it achieves real-time, dynamic, and refined perception and high-precision calculation of carbon emissions throughout the power grid's entire lifecycle, completely changing the traditional static, lagging, and coarse-grained calculation mode, resulting in a significant improvement in the timeliness and accuracy of carbon emission data. Simultaneously, the introduction of reinforcement learning as an intelligent decision engine endows the system with adaptive optimization capabilities, enabling it to autonomously learn and generate optimal scheduling, operation and maintenance, and investment strategies based on the complex dynamic changes in power grid operation, achieving continuous dynamic optimization and significant reduction of carbon emissions throughout the entire lifecycle. Furthermore, this invention realizes a closed-loop collaborative optimization of the entire carbon footprint management process from planning and design to decommissioning and recycling, effectively avoiding the overall suboptimal problem that may result from local optimization, and significantly improving the overall efficiency and resilience of green power grid operation. This invention has strong theoretical innovation and engineering practical value, providing powerful technical support for achieving low-carbon operation of the power grid and the national "dual-carbon" goals. Attached Figure Description

[0014] Figure 1 This is a flowchart illustrating the dynamic management method for power grid carbon footprint based on digital twin enhancement, according to an embodiment of the present invention. Figure 2 This is a schematic diagram of the full life cycle carbon footprint accounting engine according to an embodiment of the present invention. Detailed Implementation

[0015] The present invention will be further illustrated below with reference to the accompanying drawings and specific embodiments. These embodiments are implemented based on the technical solutions of the present invention, and it should be understood that these embodiments are only used to illustrate the present invention and are not intended to limit the scope of the present invention.

[0016] like Figure 1 As shown, this embodiment of the invention provides a method for dynamic management of power grid carbon footprint based on digital twin enhancement, including: A high-fidelity digital twin model of key physical assets in the power grid is constructed. This model integrates multi-source heterogeneous data to achieve real-time mapping and bidirectional synchronization between the physical world and the digital space. These key physical assets include generating units, transmission lines, transformers, distribution equipment, energy storage facilities, and maintenance vehicles. The multi-source heterogeneous data is collected from Supervisory Control and Data Acquisition (SCADA) systems, sensors, geographic information systems, equipment ledgers, maintenance records, and video surveillance via a data acquisition module. This data is transmitted in real-time via OPC UA and MQTT protocols, and preprocessed and aligned using data fusion algorithms to achieve centimeter-level spatial positioning and millisecond-level time synchronization between the physical world and the power grid digital twin model. The power grid digital twin model not only reflects the geometry and operating status of equipment but also embeds its carbon emission characteristics, laying the foundation for subsequent real-time carbon footprint perception. Specifically, each type of power grid digital twin model is associated with its full life-cycle carbon emission factor database, and the carbon emissions of the equipment unit are calculated in real-time. ; in, To calculate the carbon emissions of the equipment unit, For the i-th activity data, Let be the emission factor corresponding to the i-th activity data.

[0017] A lifecycle carbon footprint accounting engine was developed based on a power grid digital twin model to acquire real-time updates of equipment operating status, energy consumption data, and material status from the digital twin, and dynamically calculate the carbon emissions of the power grid at each stage from raw material extraction to decommissioning and recycling. Through continuous data flow and feedback from the digital twin, online, visualized, and refined accounting of the power grid's carbon footprint is achieved, overcoming the limitations of traditional static accounting and providing highly timely carbon emission data. For details, see [link to details]. Figure 2 The full lifecycle carbon footprint accounting engine includes: data interface layer 1, accounting logic layer 2, data storage layer 3, and presentation layer 4.

[0018] The data interface layer 1 is used to subscribe in real time to equipment operating status, energy consumption data, material status, and environmental parameters from the power grid digital twin model via API or message queue. Energy consumption data includes transformer power loss and generator fuel consumption, while material status includes the type and quantity of materials replaced during maintenance. Environmental parameters include the impact of temperature on transmission losses. The accounting logic layer 2 dynamically calculates the carbon emissions of the power grid at each stage from raw material extraction to decommissioning and recycling. Specifically, the accounting logic layer integrates the Life Cycle Assessment (LCA) method based on the ISO 14040 / 14044 standard and extends it to suit the characteristics of the power grid, dynamically calculating carbon emissions across the five stages of "raw material extraction - production and manufacturing - transportation and installation - operation and maintenance - decommissioning and recycling". The data storage layer 3 stores the calculated carbon emissions. The presentation layer 4 provides an online, multi-dimensional, and highly timely carbon footprint visualization interface, including overall carbon emission trends, the proportion of carbon emissions at each stage, and the carbon emission contribution of key equipment, enabling real-time monitoring and traceability of carbon emissions.

[0019] A reinforcement learning agent is designed and trained to continuously optimize its decision-making strategies through deep learning in a simulation environment provided by a power grid digital twin model and a life-cycle carbon footprint accounting engine, by interacting with the environment. Specifically, the reinforcement learning agent includes a state space, an action space, and a reward function. The state space reflects comprehensive information about power grid operation in real time, including power grid operation data and carbon emission data. The action space provides a set of executable discrete or continuous decisions, including scheduling, operation and maintenance, and investment strategies. The reward function guides the reinforcement learning agent to learn strategies that minimize life-cycle carbon emissions, and can be specifically expressed as: ; in, To reinforce the actions performed by the training agent at time t, The increase in carbon emissions per unit time. For operating costs, Penalties for violating operational constraints , , These are the weighting coefficients.

[0020] The power grid digital twin model, the life cycle carbon footprint accounting engine, and the reinforcement training agent are deeply integrated and coordinated to enable the reinforcement training agent to output the optimal low-carbon decision-making strategy based on the real-time power grid status provided by the power grid digital twin model.

[0021] For example: This invention was deployed and validated in practice on a large hub substation complex and its associated transmission and distribution network under the jurisdiction of a provincial power grid company. This substation complex (covering multiple voltage levels of 500kV and 220kV, equipped with advanced equipment such as intelligent inspection robots and online monitoring sensors, and connected to some renewable energy power plants in the region) served as the implementation object for engineering verification. The specific construction scheme included: First, on the high-voltage side, by integrating real-time substation operation data, equipment status information, and surrounding environmental parameters, a digital twin model of the power grid for each substation and transmission line was constructed to achieve real-time perception and mapping of carbon emissions. On the medium-voltage side, a digital twin platform was used to simulate the impact of power grid topology optimization and power flow control on carbon emissions under different load scenarios, and a digital twin of maintenance vehicles was integrated to track their travel paths and energy consumption carbon emissions. On the low-voltage side, the focus was on intelligent scheduling of the distribution network and optimization of distributed energy access. Through reinforcement learning to optimize the control system, dynamic prediction, real-time calculation, and adaptive optimization of the power grid's carbon footprint throughout its entire lifecycle were achieved, comprehensively verifying the system's emission reduction efficiency and intelligent management level under complex power grid operating conditions. Specifically as follows: 1. Construction and Data Integration of Digital Twin Model of Power Grid Physical Entity The initial focus was on the hub substations and related transmission and distribution networks under the jurisdiction of the aforementioned provincial power grid companies. Specifically, high-precision geometric and topological digital models were constructed based on their 3D point cloud data, CAD drawings, and BIM models for key physical assets such as 500kV / 220kV main transformers, GIS circuit breakers, busbars, overhead transmission lines, intelligent inspection robots, and regional maintenance vehicles. Simultaneously, data acquisition modules were deployed to integrate real-time heterogeneous data streams from multiple sources, including SCADA systems (such as transformer oil temperature, load current, active / reactive power), various intelligent sensors (such as ambient temperature and humidity, partial discharge, and fiber optic temperature measurement), geographic information systems (GIS, providing line routes and environmental altitude), equipment ledgers (equipment models, rated parameters, and material lists), maintenance records (maintenance plans, fault logs, and energy consumption data), and video surveillance. This data was transmitted in real-time via protocols such as OPC UA and MQTT, and preprocessed and aligned using data fusion algorithms (such as Kalman filtering and extended Kalman filtering) to achieve centimeter-level spatial positioning and millisecond-level time synchronization between the physical world and the power grid digital twin model. To embed carbon emission characteristics, each type of power grid digital twin model is linked to its entire lifecycle carbon emission factor database. For example, the transformer model is linked to the carbon footprint of its manufacturing materials and the carbon emission factor of its operating losses, while the maintenance vehicle model is linked to the carbon emission factor of its fuel consumption rate and mileage, thereby calculating the carbon emissions of the equipment unit. This integration approach ensures that the digital twin model of the power grid can not only accurately reflect the physical state, but also perceive its carbon emission characteristics in real time and dynamically, laying a solid foundation for subsequent refined carbon footprint accounting.

[0022] 2. Development of a real-time sensing and refined accounting engine for carbon footprint throughout the entire life cycle The provincial power grid digital twin model constructed above serves as the core data source. This engine adopts a microservice architecture, comprising a data interface layer, an accounting logic layer, and a data storage and presentation layer. It can dynamically calculate carbon emissions across five stages: raw material extraction, production and manufacturing, transportation and installation, operation and maintenance, and decommissioning and recycling. For the operation stage, for example, the real-time line loss carbon emissions of transmission lines can be dynamically calculated using the following formula: ; in, To calculate the real-time line loss carbon emissions of the transmission line, For real-time line loss power, To calculate the time step, This represents the carbon emission factor for the power grid. The accounting engine can further incorporate an activity-data-driven dynamic emission factor adjustment mechanism, such as adjusting the grid emission factor based on actual fuel composition or power supply structure. This accounting engine, integrating digital twins and dynamic LCA, significantly improves the real-time performance and granularity of carbon footprint data, providing a decision-making basis for the low-carbon operation of the power grid.

[0023] 3. Strengthen the design and training of intelligent agents. The reinforcement learning agent is designed as a deep reinforcement learning model. Its training environment is jointly provided by a power grid digital twin model and a life-cycle carbon footprint accounting engine, forming a high-fidelity simulation and reward feedback closed loop. The life-cycle carbon footprint management of the power grid is modeled as a Markov decision process (MDP), with its key elements defined as follows: 1) State Space (S): including but not limited to the output of each generator set (MW), power flow of transmission lines (kA), substation load rate (%), equipment health status (%), location and task status of maintenance vehicles, and real-time carbon emissions (tCO2e) and carbon emission trends at each stage provided by the carbon footprint accounting engine.

[0024] 2) Action Space (A): This encompasses multiple aspects of the entire power grid lifecycle. Specifically, it includes: power source combination optimization on the generation side (such as adjusting the output ratio of thermal power and wind power), power flow optimization on the transmission and distribution side (such as changing the status of line switches), equipment maintenance strategies (such as determining maintenance timing and priority), operation and maintenance path planning (such as selecting the optimal route), ancillary service invocation (such as energy storage charging and discharging), and recommendations for investment or decommissioning of specific equipment.

[0025] 3) Reward Function (R): This function guides the reinforcement learning agent to learn a strategy that minimizes carbon emissions over its entire lifecycle. It evaluates the impact of agent actions on carbon emissions and operating costs in real time through a digital twin environment and provides feedback. The agent is trained using the Proximal Policy Optimization (PPO) algorithm, learning the optimal policy network through continuous interaction with the digital twin environment. During training, the agent explores and optimizes its decisions through millions of simulation iterations provided by the digital twin, enabling the final policy to adapt to complex grid conditions and achieve adaptive minimization of carbon emissions. Its core elements are shown in Table 1. Table 1: Examples of Definitions of Core Elements of Reinforcement Training Agents ; 4. Collaborative optimization of operation control using digital twins and reinforcement learning The power grid digital twin model, the life-cycle carbon footprint accounting engine, and the reinforcement learning agent are deeply integrated to form a closed-loop system of "real-time perception - intelligent decision-making - adaptive control". Based on the real-time power grid operating status (such as load, power output, and equipment health) obtained from the power grid digital twin model and the real-time carbon emission data (including total emissions and emission composition at each stage) provided by the carbon footprint accounting engine, the reinforcement learning agent outputs a series of low-carbon optimization decisions within each decision cycle, according to the optimal strategy it has learned. These decisions are transformed into specific control commands or operational suggestions through standardized digital interfaces (such as IEC61850 and Modbus TCP), and are directly or indirectly applied to the actual operation of the power grid. For example, on the generation side, the reinforcement learning agent can adjust the real-time output allocation between new energy sources and conventional units to minimize carbon emissions while meeting load and stability requirements; on the transmission and distribution side, the reinforcement learning agent can optimize power flow paths to reduce network losses; and for operation and maintenance, it recommends the lowest-carbon inspection routes and preventative maintenance timing. The entire process follows a feedback control mechanism: after the reinforcement learning agent performs an action, the state of the power grid's digital twin model is updated, the carbon footprint accounting engine recalculates, and a new environmental state and reward are fed back to the reinforcement learning agent, prompting its strategy to continuously adapt and optimize. For example, the power generation allocation decision can be represented as: ; in, Let be the total power supplied by the power grid at time t. Let be the total power generation at time t. for, Let be the total load on the power grid at time t. The goal of the reinforcement learning agent is to select... The combination of these factors results in total carbon emissions from power generation. Minimize emissions while simultaneously satisfying various operational constraints. This collaborative mechanism ensures that the power grid operates efficiently and intelligently under complex dynamic conditions, consistently striving to minimize carbon emissions throughout its entire lifecycle. Specific examples of collaborative optimization decisions and power grid control / emission reduction can be found in Table 2. Table 2: Examples of Collaborative Optimization Decision-Making and Power Grid Control / Emission Reduction ; 5. System integration simulation verification and carbon emission reduction benefit assessment System integration simulation verification and carbon emission reduction benefit assessment are key steps in ensuring the effectiveness and advancement of the "Dynamic Management Method for Power Grid Carbon Footprint Based on Digital Twin Reinforcement". A high-precision power grid simulation platform (such as PSCAD / EMTDC, DIgSILENT PowerFactory combined with MATLAB / Simulink) is used to fully integrate the power grid digital twin model, carbon footprint accounting engine, and reinforcement learning optimization agent. This platform can simulate power grid operation scenarios covering various complex conditions, including typical load curves, randomness of renewable energy output, equipment failures, extreme weather events, and cyberattacks, thereby conducting comprehensive performance verification of the system under controlled conditions. Evaluation indicators include: carbon emission reduction rate, decision response time, carbon footprint accounting accuracy, power grid operation economy, system stability, and power grid resilience indicators. By comparing the performance of the system of this invention with traditional carbon footprint management schemes based on experience or fixed rules under the same simulation scenario, its beneficial effects are quantitatively analyzed. For example, the life-cycle carbon emission reduction rate: ; in, The calculated carbon emission efficiency. This represents the total carbon emissions under traditional methods. This represents the total carbon emissions under the proposed solution. Economic benefits are assessed through factors such as operating cost savings and ancillary service revenue growth. Specific evaluation indicators and comparisons with expected benefits are shown in Table 3. ; It can be seen that this invention has significant achievements and innovative value in reducing carbon emissions, improving the green operation efficiency of the power grid, and promoting the green transformation of the power grid.

[0026] The above description is merely a preferred embodiment of the present invention. It should be noted that for those skilled in the art, other parts not specifically described are existing technology or common knowledge. Several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A dynamic management method for the carbon footprint of power grids based on digital twin enhancement, characterized in that, include: A digital twin model of key physical assets in the power grid is constructed. The digital twin model of the power grid achieves real-time mapping and two-way synchronization between the physical world and the digital space by integrating multi-source heterogeneous data. Based on the aforementioned power grid digital twin model, a full life cycle carbon footprint accounting engine was developed to obtain real-time updates of equipment operating status, energy consumption data, and material status from the digital twin, and to dynamically calculate the carbon emissions of the power grid at each stage from raw material mining to decommissioning and recycling. The reinforcement training agent is designed and trained. The reinforcement training agent performs deep learning by interacting with the environment in the simulation environment provided by the power grid digital twin model and the life cycle carbon footprint accounting engine, and continuously optimizes its decision-making strategy. The power grid digital twin model, the life-cycle carbon footprint accounting engine, and the reinforcement training agent are deeply integrated and coordinated so that the reinforcement training agent can output the optimal low-carbon decision-making strategy based on the real-time power grid status provided by the power grid digital twin model.

2. The method for dynamic management of power grid carbon footprint based on digital twin enhancement according to claim 1, characterized in that, The multi-source heterogeneous data is collected from the acquisition and monitoring control system, sensors, geographic information system, equipment ledger, operation and maintenance records and video surveillance by the data acquisition module.

3. The method for dynamic management of power grid carbon footprint based on digital twin enhancement according to claim 1, characterized in that, The multi-source heterogeneous data is transmitted in real time via OPC UA and MQTT protocols, and preprocessed and aligned using a data fusion algorithm to achieve centimeter-level spatial positioning and millisecond-level time synchronization between the physical world and the digital twin model of the power grid.

4. The method for dynamic management of power grid carbon footprint based on digital twin enhancement according to claim 1, characterized in that, Each type of power grid digital twin model is linked to its full life-cycle carbon emission factor database and calculates the carbon emissions of equipment units in real time: ； in, To calculate the carbon emissions of the equipment unit, For the i-th activity data, Let be the emission factor corresponding to the i-th activity data.

5. The method for dynamic management of power grid carbon footprint based on digital twin enhancement according to claim 1, characterized in that, The lifecycle carbon footprint calculation engine includes: The data interface layer is used to subscribe in real time to equipment operating status, energy consumption data, material status and environmental parameters from the power grid digital twin model via API or message queue; The accounting logic layer is used to dynamically calculate the carbon emissions of the power grid at each stage from raw material extraction to decommissioning and recycling. The data storage layer is used to store the calculated carbon emissions; The presentation layer provides a visualization interface for carbon footprint, including overall carbon emission trends, the proportion of carbon emissions at each stage, and the carbon emission contribution of key equipment, enabling real-time monitoring and tracing of carbon emissions.

6. The method for dynamic management of power grid carbon footprint based on digital twin enhancement according to claim 1, characterized in that, The reinforcement training agent includes: The state space is used to reflect comprehensive information about the operation of the power grid in real time. Action space, used to provide a set of executable discrete or continuous decisions; The reward function is used to guide the reinforcement training agent to learn a strategy that minimizes carbon emissions over its entire life cycle.

7. The method for dynamic management of power grid carbon footprint based on digital twin enhancement according to claim 6, characterized in that, The reward function is expressed as follows: ； in, To reinforce the actions performed by the training agent at time t, The increase in carbon emissions per unit time. For operating costs, Penalties for violating operational constraints , , These are the weighting coefficients.

Images