A multi-microgrid hierarchical distributed collaborative scheduling method and system

Abstract

Embodiments of the present application disclose a kind of multi-microgrid hierarchical distributed collaborative scheduling method and system.In the method, carbon flow calculation function is constructed to obtain load node carbon potential, power generation node carbon emission intensity and other state data, and hierarchical model of device layer, microgrid layer and global layer is built.Device layer generates local low-carbon scheduling instruction based on reinforcement learning rolling optimization;Microgrid layer aggregates model gradient through federal learning, calculates carbon quota profit and loss and optimizes power mutual aid and quota allocation;Global layer stores carbon data and circulation records by means of distributed account, obtains external carbon emission sensitivity information through oracle, and generates internal optimization weight parameter through distributed voting.Based on the parameter, power market joint bidding strategy and internal low-carbon scheduling strategy are generated through double-layer collaborative optimization model, to minimize system total carbon emission, and improve renewable energy consumption efficiency and grid robustness, adaptability.

Description

Technical Field

[0001] This invention relates to the field of power system dispatching and optimization technology, and in particular to a hierarchical distributed collaborative dispatching method and system for multiple microgrids. Background Technology

[0002] With the acceleration of the energy transition, multi-microgrid systems are widely used due to their ability to effectively integrate distributed renewable energy sources. However, existing dispatching methods still face many challenges in achieving deeply decarbonized operation:

[0003] Insufficient precision in carbon emission quantification: Traditional methods lack accurate calculation of carbon potential at the grid node level, making it impossible to achieve precise source tracing and optimization of carbon flows.

[0004] Data silos and privacy and security issues: Each microgrid operates independently, the data is heterogeneous and involves commercial privacy, making it difficult to achieve centralized collaborative optimization.

[0005] Coordination across multiple time scales is challenging: Existing methods struggle to unify second-level real-time control with day-ahead scheduling planning, failing to balance system dynamic response with long-term low-carbon goals.

[0006] External carbon market signals are difficult to internalize: external signals such as carbon prices cannot be effectively transformed into internal technical parameters that guide system operation, and there is a lack of means to deeply integrate market mechanisms with physical scheduling.

[0007] Therefore, how to achieve refined and real-time carbon flow scheduling, enhance the robustness and adaptability of power grid operation, and avoid data silos and privacy and security issues are problems that need to be solved by those skilled in the art. Summary of the Invention

[0008] This invention provides a hierarchical distributed collaborative scheduling method and system for multi-microgrids. By constructing a carbon flow calculation function and a hierarchical model, and combining technologies such as reinforcement learning, federated learning, blockchain, and two-layer optimization, the system can minimize total carbon emissions and efficiently absorb renewable energy.

[0009] The first aspect of this invention provides a hierarchical distributed cooperative scheduling method for multiple microgrids, comprising: A carbon flow calculation function is constructed to quantify system carbon emissions in order to obtain system carbon flow state data, which includes the carbon potential of load nodes, the carbon emission intensity of power generation nodes, and the carbon potential changes of energy storage nodes. Construct a hierarchical model that includes a device layer, a microgrid layer, and a global layer; At the device layer, rolling optimization is performed based on reinforcement learning algorithms to generate low-carbon scheduling instructions for local devices. The optimization objective of reinforcement learning includes minimizing local carbon emission data calculated based on system carbon flow state data. At the microgrid layer, the model gradients of multiple microgrids are aggregated based on the federated learning algorithm, and the carbon quota surplus and deficit of each microgrid are calculated based on the system carbon flow state data, so as to collaboratively optimize the power mutual assistance and carbon quota allocation among microgrids and generate carbon quota flow records. At the global layer, distributed ledger technology is used to store local carbon emission data and carbon quota transfer records on the blockchain; carbon emission sensitivity information is obtained from the external carbon market through oracles; and based on the carbon emission sensitivity information, internal carbon emission optimization weight parameters are generated through a distributed voting mechanism. Based on the internal carbon emission optimization weight parameters, a two-layer collaborative optimization model is adopted to generate a joint bidding strategy for microgrid groups in the electricity market and a low-carbon dispatch strategy within them. The core objective of the two-layer collaborative optimization model is to minimize the total carbon emissions of the system determined based on the system carbon flow state data.

[0010] Optionally, the carbon flow calculation function includes: A carbon potential calculation function used to characterize the accumulation of carbon emissions generated at power grid generation nodes at power grid load nodes after being distributed by power flow; A power generation node carbon emission intensity function used to quantify carbon emissions per unit of electricity generated; A function describing the change in carbon potential at an energy storage node, used to describe the effect of energy storage charging and discharging on carbon potential absorption or release.

[0011] Optionally, at the device layer, rolling optimization is performed based on reinforcement learning algorithms to generate low-carbon scheduling instructions for the local device, including: Obtain the current system status, which includes real-time carbon intensity, energy storage system state of charge, local load forecast, and renewable energy output forecast provided by system carbon flow status data; The system state is input into a pre-trained reinforcement learning model to obtain the optimal action strategy, which includes the target output of the local generator set, the target charging and discharging power of the energy storage, and the target adjustment amount of the controllable load. The optimal action strategy is converted into control commands and sent to the corresponding local devices for execution, so as to minimize local carbon emissions while maintaining local power balance. In the next scheduling cycle, the above process is repeated based on the new system state to achieve real-time rolling generation and optimization of low-carbon scheduling instructions.

[0012] Optionally, at the microgrid layer, model gradients from multiple microgrids are aggregated based on a federated learning algorithm, and the carbon allowance surplus or deficit of each microgrid is calculated based on system carbon flow state data to collaboratively optimize power sharing and carbon allowance allocation among microgrids, and to generate carbon allowance transfer records, including: Based on the system carbon flow status data, the difference between the actual carbon emissions of each microgrid and the preset carbon quota benchmark value during the dispatch cycle is calculated to determine the carbon quota accounting result. The carbon quota accounting result includes the carbon quota surplus or deficit status of each microgrid. Each microgrid trains a scheduling model based on local data and uploads the model gradient to the regional scheduler. The regional scheduler uses a federated averaging algorithm to perform a weighted average of the received model gradients from each microgrid and aggregates them to generate a global scheduling model. Based on the carbon quota accounting results and the global scheduling model, the Monte Carlo hierarchical optimization method is adopted. Multiple candidate schemes for power mutual assistance and carbon quota allocation are generated by random sampling. The impact of each candidate scheme on the total carbon emissions of the system is simulated and evaluated. The candidate scheme that minimizes the total carbon emissions of the system is selected as the collaborative optimization scheme. The collaborative optimization scheme includes the optimal power mutual assistance strategy and carbon quota allocation strategy. According to the collaborative optimization scheme, corresponding carbon quota transfer records are generated. The carbon quota transfer records shall include at least the identifiers of the two parties involved in the carbon quota transfer, the amount of carbon quota transferred, the corresponding power mutual assistance amount, and timestamp information.

[0013] Optionally, distributed ledger technology can be used to store local carbon emission data and carbon quota transfer records on the blockchain, including: Standardize the format of local carbon emission data and carbon quota transfer records; The evidence storage smart contract is invoked through the application programming interface (API) and deployed in the distributed ledger network. The standardized local carbon emission data and carbon quota transfer records are used as the transaction content to generate a transaction to be uploaded to the blockchain and digitally signed by the data submitter. The signed transaction is broadcast to nodes in the distributed ledger network, and the validity and consistency of the transaction are verified by a consensus algorithm. Validated transactions are packaged into new blocks and added to the distributed ledger to form immutable on-chain evidence.

[0014] Optionally, based on carbon emission sensitivity information, internal carbon emission optimization weight parameters are generated through a distributed voting mechanism, including: Each microgrid node autonomously generates a carbon allowance transfer proposal containing desired internal carbon emission optimization weight parameters based on its own carbon emission surplus or deficit status, operating costs, and carbon emission sensitivity information, and broadcasts the carbon allowance transfer proposal to the distributed ledger network. Each microgrid node evaluates all received carbon quota transfer proposals and conducts multiple rounds of voting based on a Byzantine fault-tolerant consensus algorithm until a consensus is reached. Based on the consensus-reaching voting results, the final internal carbon emission optimization weight parameters applicable to all microgrid nodes are calculated using a weighted average or majority clearing algorithm. The internal carbon emission optimization weight parameters are written into the distributed ledger through a smart contract for final confirmation and then sent to the two-layer collaborative optimization model as input parameters.

[0015] Optionally, based on the internal carbon emission optimization weight parameters, a two-layer collaborative optimization model is adopted to generate a joint bidding strategy for the microgrid group in the electricity market and its internal low-carbon dispatch strategy, including: With the goal of minimizing the total operating cost of the microgrid cluster and the carbon emission technology penalty term weighted by the internal carbon emission optimization weight parameters, an upper-level optimization model is constructed and solved to generate a joint bidding strategy for the microgrid cluster in the electricity market, and the market clearing price and the shadow price of carbon emission constraints are output as dual variables. The lower-level optimization model is constructed and solved with the goal of minimizing the operating costs of each microgrid and the local carbon emissions driven by the internal carbon emission optimization weight parameters. This generates the power generation, energy storage and load dispatch strategies within the microgrid group and feeds back the marginal cost of each dispatch decision to the upper-level model. Repeat the solution and variable transfer process of the upper-level model and the lower-level model until the change in the joint bidding strategy and the internal scheduling strategy is less than the preset threshold, and obtain the converged joint bidding strategy and the converged internal low-carbon scheduling strategy. The converged joint bidding strategy is output to the electricity market, and the converged internal low-carbon dispatch strategy is distributed to each microgrid for execution.

[0016] A second aspect of the present invention provides a hierarchical distributed cooperative scheduling system for multiple microgrids, comprising: The carbon flow calculation module is used to construct a carbon flow calculation function for quantifying system carbon emissions in order to obtain system carbon flow status data, which includes the carbon potential of load nodes, the carbon emission intensity of power generation nodes, and the carbon potential changes of energy storage nodes. The hierarchical model architecture module is used to build a hierarchical model that includes a device layer, a microgrid layer, and a global layer. The device layer is used for rolling optimization based on reinforcement learning algorithms to generate low-carbon scheduling instructions for local devices. The optimization objective of reinforcement learning includes minimizing local carbon emission data calculated based on system carbon flow state data. The microgrid layer is used to aggregate the model gradients of multiple microgrids based on federated learning algorithms, and to calculate the carbon quota surplus and deficit of each microgrid based on the system carbon flow state data, so as to collaboratively optimize the power mutual assistance and carbon quota allocation among microgrids and generate carbon quota flow records. The global layer is used to store local carbon emission data and carbon quota transfer records on-chain using distributed ledger technology; obtain carbon emission sensitivity information from external carbon markets through oracles; and generate internal carbon emission optimization weight parameters based on the carbon emission sensitivity information through a distributed voting mechanism. The two-layer collaborative optimization module is used to generate joint bidding strategies for microgrid groups in the electricity market and their internal low-carbon dispatch strategies based on internal carbon emission optimization weight parameters and a two-layer collaborative optimization model. The core objective of the two-layer collaborative optimization model is to minimize the total carbon emissions of the system determined based on system carbon flow state data.

[0017] A third aspect of the present invention provides a multi-microgrid hierarchical distributed cooperative scheduling device, comprising: One or more processors; A memory on which one or more programs are stored; When the one or more programs are executed by the one or more processors, the one or more processors implement the multi-microgrid hierarchical distributed cooperative scheduling method as described in any of the preceding claims.

[0018] A fourth aspect of the present invention provides a computer storage medium for storing a program, which, when executed, is used to implement the method for hierarchical distributed cooperative scheduling of multiple microgrids as described in any of the preceding claims.

[0019] By constructing a hierarchical model, the precision and real-time control capabilities of carbon flow scheduling were significantly improved. A hybrid intelligent optimization algorithm combining federated learning and reinforcement learning was introduced, enhancing the autonomous decision-making capabilities of each microgrid node and the computational scalability of the system while ensuring data privacy, enabling more efficient response to low-carbon scheduling commands. Distributed ledger technology was introduced to achieve secure and transparent storage and quota transfer of carbon data and carbon footprint. Oracle technology was integrated to achieve decentralized confirmation and effective arbitration of external carbon emission sensitivity information in the distributed ledger network. Internal carbon emission optimization weight parameters were formed based on a distributed voting mechanism, thereby transforming external market signals into driving functions. Finally, a two-layer collaborative optimization model was used to achieve low-carbon allocation of power resources, significantly reducing total system carbon emissions, improving renewable energy consumption and energy utilization efficiency, and enhancing grid robustness and adaptability. Attached Figure Description

[0020] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 A flowchart illustrating a hierarchical distributed collaborative scheduling method for multiple microgrids provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of the structure of a multi-microgrid hierarchical distributed cooperative scheduling system provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of the structure of a multi-microgrid hierarchical distributed collaborative scheduling device provided in an embodiment of the present invention. Detailed Implementation

[0022] This invention provides a hierarchical distributed collaborative scheduling method and system for multi-microgrids. By constructing a carbon flow calculation function and a hierarchical model, and combining technologies such as reinforcement learning, federated learning, blockchain, and two-layer optimization, the system can minimize total carbon emissions and efficiently absorb renewable energy.

[0023] To facilitate understanding, the application scenarios of the embodiments of the present invention will be introduced first.

[0024] With the rapid development of distributed power sources and microgrids, multi-microgrid systems are playing an increasingly important role in improving energy efficiency and promoting the consumption of renewable energy. However, the dispatching of large-scale multi-microgrid systems faces many challenges, especially in achieving deep decarbonization: Insufficient Refined Management and Quantification of Carbon Emissions: Existing power grid systems lack sophisticated methods for assessing carbon emissions and quantifying emissions at the node level. Accurately calculating and optimizing the carbon potential and total carbon emissions of power grid load nodes, generation nodes, and energy storage nodes is a pressing technical problem that needs to be solved.

[0025] Data distribution heterogeneity and privacy protection requirements: Each microgrid has independent operating data, which involves trade secrets and privacy, making it difficult to upload centrally for unified scheduling, resulting in the problem of data silos.

[0026] Communication delay and bandwidth limitations: Large-scale microgrid systems suffer from high communication overhead and latency, which seriously affects real-time scheduling performance.

[0027] System heterogeneity and computational complexity: Different microgrids have large differences in equipment configuration and load characteristics. Traditional centralized optimization algorithms have high computational complexity and poor real-time performance when processing large-scale non-independent and identically distributed data.

[0028] The lack of multi-scale carbon flow coordination optimization: Existing scheduling methods mostly focus on a single time scale or physical scale, and lack the ability to unify the physical scale (equipment-microgrid-microgrid group) and the time scale (second-minute-intraday-pre-day) into a hierarchical model. This makes it difficult to effectively solve the alternating fast and slow dynamics, thus limiting the precise control of grid carbon flow.

[0029] Insufficient Decentralized Carbon Data Management and Technology Transformation of External Market Signals: Although multi-level carbon trading mechanisms have been applied in microgrids, they largely rely on centralized platforms, resulting in insufficient data transparency, low trading efficiency, and susceptibility to single points of failure. In a distributed microgrid federated learning environment, there is a lack of means to enable autonomous participation of microgrid nodes, protect privacy, and effectively transform external dynamic carbon emission market signals into technical parameters for optimizing total carbon emissions within the system.

[0030] While existing hierarchical distributed scheduling technologies, such as the Analytical Objective Concatenation (ATC) method, have been widely applied to solve the hierarchical distributed scheduling problem in multi-microgrid systems, achieving parallel optimization of each system layer, and federated learning technology has also made breakthroughs in power systems, demonstrating good performance in tasks such as energy forecasting and electricity theft detection, the edge-cloud federated partitioned learning framework has solved the computational limitations of resource-constrained devices such as smart meters, reducing communication overhead by 60-80% compared to traditional methods.2 However, current technologies have not yet been able to combine refined carbon emission assessment (such as carbon potential), federated learning, multi-scale hierarchical models with decentralized carbon data management and mechanisms for converting external market signals into core technology optimization parameters to achieve efficient, privacy-preserving, and carbon reduction-driven flexible scheduling. Therefore, an innovative technological approach is urgently needed to address these issues.

[0031] See Figure 1 This figure is a flowchart illustrating a hierarchical distributed collaborative scheduling method for multi-microgrids provided in an embodiment of the present invention. The hierarchical distributed collaborative scheduling method for multi-microgrids provided in this embodiment of the present invention can be implemented, for example, through the following steps S101-S106.

[0032] S101: Construct a carbon flow calculation function to quantify system carbon emissions in order to obtain system carbon flow state data.

[0033] In this embodiment of the invention, the system carbon flow state data includes the carbon potential of load nodes, the carbon emission intensity of power generation nodes, and the carbon potential change of energy storage nodes. The carbon flow calculation functions include: a carbon potential calculation function characterizing the accumulation of carbon emissions generated by power generation nodes at power grid load nodes after power flow distribution; a power generation node carbon emission intensity function quantifying carbon emissions per unit of power generation; and an energy storage node carbon potential change function describing the impact of energy storage charging and discharging on carbon potential absorption or release.

[0034] Specifically, this invention introduces innovative indicators to achieve in-depth optimization of carbon emissions, including the following refined carbon emission indicators: The carbon potential calculation function of the grid load node is used to characterize the carbon emissions generated by the grid generation node and their accumulation at the grid load node after power flow distribution, so as to realize the node-level quantification and visualization of carbon emissions. The carbon emission intensity function of the power grid generation node quantifies the carbon emissions generated per unit of power generation, guiding the priority dispatch of low-carbon units; The carbon potential change function of energy storage nodes accurately describes the impact of energy storage charging and discharging on carbon potential absorption or release, serving as an important variable for carbon flow optimization.

[0035] These metrics enable the algorithm to directly minimize "equivalent carbon overhead" or "total carbon emissions," transforming carbon reduction targets from external constraints into internal optimization objectives and improving the carbon efficiency of scheduling.

[0036] In this embodiment of the invention, carbon emissions are taken as one of the core technical objectives of dynamic optimization. A multi-objective function is used to optimize "system operating performance-carbon emissions", that is, to minimize the total system operating technical overhead and total carbon emissions. The comprehensive optimization objective function F is expressed as follows: ; in: To comprehensively optimize the objective function; The total operating technical overhead includes the physical cost of power generation, the physical cost of purchasing electricity, the carbon emission optimization guidance function value, and the operation and maintenance technical overhead. The total carbon emissions of the system are the core optimized technical indicator of this invention, and their calculation involves nodal carbon potential and carbon emission intensity. and To dynamically optimize the weighting coefficients and balance the system's operating technology costs and carbon emission reduction targets, their values ​​can be dynamically adjusted by the internal dynamic carbon emission optimization weighting estimation model based on external carbon market signals.

[0037] Total operating technical overhead Include: ; in, microgrid At any moment The physical cost of operating local power generation equipment; microgrid At any moment The physical cost of purchasing electricity from the main grid; microgrid At any moment Operation and maintenance technology costs; Indicates the first The physical cost of a microgrid; Indicates the system at time 10:00 The carbon emission optimization guiding function value; This represents a collection of microgrids. The function value is driven by internal guiding parameters, designed to technically guide the system to reduce carbon emissions. Its specific form can be seen in the ladder model described below, but its essence is a technical penalty / incentive term rather than a direct commercial transaction cost.

[0038] Total carbon emissions The calculation formula is as follows: ; in microgrid At any moment Carbon emissions generated by local power generation equipment (such as diesel engines, gas turbines, etc.); Indicates the system at time 10:00 Carbon emissions from purchasing electricity from the main grid (need to be calculated in conjunction with the real-time carbon intensity of the main grid). Indicates time Carbon emission reductions resulting from reduced load due to demand response (as a reduction item to optimize total emissions).

[0039] Carbon emission optimization guidance function The following is an example using a tiered model (technical penalties / incentives):

[0040] in, , These are internal guidance parameters (or penalty coefficients) generated by the system based on external carbon emission sensitivity signals, typically... To form a tiered constraint; Indicates time The system's real-time total carbon emissions; Indicates the system's carbon emission baseline (or exemption quota). This represents the first carbon emission technology threshold (exceeding this value will trigger a higher level of optimization weight).

[0041] Constraints include, but are not limited to: Power balance constraints: power balance within each microgrid and between microgrid groups.

[0042] Equipment operating constraints: generator output, energy storage system state of charge (SOC), power charge and discharge limits, reserve capacity, etc.

[0043] Power grid security constraints include node voltage, branch power flow, and line capacity.

[0044] Demand-side response constraints: demand-side response volume, response time, user comfort, etc.

[0045] Total carbon emission constraints: Real-time carbon emissions shall not exceed the preset technical limit, or be optimized under the guidance of the above internal parameters.

[0046] Communication constraints: Communication overhead limitations in federated learning.

[0047] S102: Construct a hierarchical model that includes a device layer, a microgrid layer, and a global layer.

[0048] In this embodiment of the invention, a hierarchical multi-scale design is adopted to unify the physical scale (device-microgrid-microgrid group) and the temporal scale (second-minute-intraday-pre-day) into a hierarchical model, constructing a hierarchical model that includes a device layer, a microgrid layer and a global layer.

[0049] Specifically, this invention adopts a three-layer distributed intelligent scheduling architecture, including a device layer (L0), a microgrid layer (L1), and a global layer (L2). Simultaneously, an edge-cloud collaborative computing framework is constructed to achieve efficient and secure scheduling.

[0050] Device Layer (L0): Responsible for local device data acquisition, status monitoring, local power balancing, and power constraints. It achieves rapid response through reinforcement learning-based rolling optimization and estimates local greenhouse gas emissions and node carbon potential in real time, providing refined carbon flow data input to the upper layers.

[0051] Microgrid Layer (L1): Responsible for power sharing, shared energy storage, and matching of local carbon quotas among multiple microgrids. Through federated averaging (FedAvg) combined with Monte Carlo hierarchical optimization, the gradients of each node's model are weighted to the "regional scheduler," eliminating the need to upload raw data, effectively protecting privacy, and achieving regional-level carbon flow optimization and collaboration.

[0052] Global Layer (L2): Serves as the interface between multiple microgrid clusters and the external electricity market and carbon emission management platform. Its core functions include distributed ledger-based carbon data management and carbon quota ledger, oracle-driven acquisition of external carbon emission sensitivity information, and internal carbon emission optimization weight estimation under a distributed physical scale (device-microgrid-microgrid cluster) and time-domain scale (second-minute-intraday-day-ahead) voting mechanism. This enables intraday-day-ahead joint bidding and autonomous management of carbon emission optimization parameters, transforming external market signals into driving functions and technical variables for internal optimization of total carbon emissions.

[0053] Edge-cloud collaborative computing framework: Edge computing nodes are deployed locally in each microgrid, responsible for real-time data processing and local optimization, while the cloud decision center is responsible for global coordination and long-term planning. Low-latency data transmission is achieved through 5G communication technology.

[0054] S103: The device layer performs rolling optimization based on reinforcement learning algorithms to generate low-carbon scheduling instructions for local devices.

[0055] In this embodiment of the invention, the optimization objective of reinforcement learning includes minimizing local carbon emissions calculated based on system carbon flow state data. The device layer is used to acquire the current system state, which includes real-time carbon intensity, energy storage system state of charge, local load forecast, and renewable energy output forecast provided by the system carbon flow state data. The system state is input into a pre-trained reinforcement learning model to obtain the optimal action strategy, which includes the target output of local generator sets, the target charging and discharging power of energy storage, and the target adjustment amount of controllable load. The optimal action strategy is converted into control commands and issued to the corresponding local devices for execution, thereby minimizing local carbon emissions while maintaining local power balance. In the next scheduling cycle, the above process is repeated based on the new system state to achieve real-time rolling generation and optimization of low-carbon scheduling commands.

[0056] Specifically, the device layer employs reinforcement learning-based rolling optimization to rapidly maintain local power balance and power constraints, and to estimate local greenhouse gas emissions in real time. The state space of this reinforcement learning-based rolling optimization includes the current output of locally controllable equipment, the state of charge (SOC) of the energy storage system, local load forecasting, renewable energy output forecasting, real-time carbon intensity, and real-time electricity prices. The action space includes adjustments to local generator output, energy storage charging and discharging power, and controllable load adjustments. A reward function is designed to minimize local carbon emissions and operating costs, while considering power balance and device constraints. The reward function directly reflects the reduction in carbon emissions and is expressed as follows: ; in, The state represents the time interval. The system's environmental status includes real-time carbon intensity, energy storage state of charge (SOC), local load forecast, and renewable energy output forecast. For actions, it represents the agent's actions at time t. The dispatching strategies adopted include generator output, energy storage charging and discharging power, and controllable load adjustment. Let be the reward function, representing the feedback signal of reinforcement learning. Since the goal is to maximize the reward while minimizing the cost and carbon emissions, all terms in the formula are negative. This is the economic weighting coefficient, representing the factor used to adjust for local operating technology costs. Its importance in the overall reward; This is the carpet weighting factor, representing the factor used to adjust local carbon emissions. The degree of importance in the total reward is the core parameter that this invention emphasizes for "low-carbon scheduling"; The balance penalty coefficient represents the coefficient used to adjust the degree of penalty imposed on the system by power imbalance; The power imbalance quantity represents the time. The difference between local supply and demand, with the squared term in the formula representing a severe penalty for large imbalances; For constraint index, representing the first... Technical constraints (such as voltage over-limit, line overload, and ramp rate limits); As a constraint penalty factor, corresponding to the first Penalty weight for class constraint violations (usually set to a large value to ensure that safety constraints are enforced); The number of violations indicates the number of violations. The degree of violation of a class constraint (such as the number of volts exceeded by the voltage).

[0057] Learning algorithms: Deep Q-Network (DQN), A3C (Asynchronous Advantage Actor-Critic), or PPO (Proximal Policy Optimization) RL algorithms can be used for training to enable the agent to learn the optimal low-carbon scheduling strategy under different states.

[0058] Rolling optimization: In each scheduling cycle (e.g., seconds or minutes), the RL agent makes decisions based on the current system state and re-evaluates and adjusts the strategy in the next cycle to ensure rapid response and optimization for real-time carbon emissions.

[0059] S104: The microgrid layer aggregates the model gradients of multiple microgrids based on the federated learning algorithm, and calculates the carbon quota surplus and deficit of each microgrid based on the system carbon flow state data, so as to collaboratively optimize the power mutual assistance and carbon quota allocation among microgrids and generate carbon quota flow records.

[0060] In this embodiment of the invention, based on system carbon flow status data, the difference between the actual carbon emissions of each microgrid during the scheduling cycle and the preset carbon quota benchmark value is calculated to determine the carbon quota accounting result. The carbon quota accounting result includes the carbon quota surplus or deficit status of each microgrid. Each microgrid trains a scheduling model based on local data and uploads the model gradient to the regional scheduler. The regional scheduler uses a federated averaging algorithm to perform a weighted average of the received model gradients of each microgrid and aggregates them to generate a global scheduling model. Based on the carbon quota accounting result and the global scheduling model, a Monte Carlo hierarchical optimization method is used to generate multiple candidate schemes for power mutual assistance and carbon quota allocation through random sampling. The impact of each candidate scheme on the total carbon emissions of the system is simulated and evaluated, and the candidate scheme that minimizes the total carbon emissions of the system is selected as the collaborative optimization scheme. The collaborative optimization scheme includes the optimal power mutual assistance strategy and carbon quota allocation strategy. According to the collaborative optimization scheme, a corresponding carbon quota transfer record is generated. The carbon quota transfer record includes at least the identifiers of both parties involved in the carbon quota transfer, the amount of carbon quota transfer, the corresponding amount of power mutual assistance, and timestamp information.

[0061] Specifically, the microgrid layer adopts FedAvg combined with Monte Carlo hierarchical optimization to achieve power sharing among multiple microgrids, shared energy storage, and matching of local carbon quotas, while protecting the data privacy of each microgrid and achieving regional-level coordinated optimization of carbon emissions. The specific steps include the following: Local model training: Each microgrid (client) independently trains its local scheduling model (e.g., L0 layer reinforcement learning model or prediction model) on its local dataset to obtain the gradient of the model parameters. wi.

[0062] Local model gradient upload: Each microgrid uploads encrypted or differential privacy noise-added local model gradients. Upload to the regional scheduler (server).

[0063] Federated aggregation: The regional dispatcher performs a weighted average of the gradients received from each microgrid to form a global model. .

[0064] ; in, For microgrids Data sample size, This represents the total sample size for all microgrids.

[0065] Model distribution: The regional dispatcher distributes the aggregated global model parameters to each microgrid to update its local model.

[0066] Monte Carlo hierarchical optimization: Based on the global model update, the regional scheduler uses the Monte Carlo method to perform hierarchical optimization of power exchange and carbon quota transfer among microgrids. Through random sampling and simulation, the impact of different power exchange and carbon quota allocation schemes on the overall total carbon emissions is evaluated, and the optimal scheme is selected. This process does not require the upload of raw data; it only performs collaborative decision-making based on model parameters.

[0067] The federated learning scheme in this embodiment of the invention focuses on solving the problems of data privacy and communication overhead in industrial microgrids.

[0068] Privacy Protection: Federated learning only exchanges model gradients, not raw production data (such as equipment efficiency and energy costs), effectively preventing information leakage. Combined with differential privacy or secure aggregation technologies, data privacy is further enhanced.

[0069] Communication overhead: Using gradient compression or knowledge distillation federated algorithms (“Teacher-Student” structure) can significantly reduce the amount of data uploaded and reduce the communication burden.

[0070] Heterogeneity handling: Personalized federated learning, such as P-FedAvg or model classifier gradient clustering, is used to address the heterogeneity of different microgrid data distributions and improve model adaptability.

[0071] S105: The global layer is used to store local carbon emission data and carbon quota transfer records on-chain using distributed ledger technology; obtain carbon emission sensitivity information from the external carbon market through oracles; and generate internal carbon emission optimization weight parameters based on the carbon emission sensitivity information through a distributed voting mechanism.

[0072] In this embodiment of the invention, local carbon emission data and carbon quota transfer records are processed in a standardized format; a notarization smart contract deployed in a distributed ledger network is called through an application programming interface; the standardized local carbon emission data and carbon quota transfer records are used as transaction content to generate a transaction to be uploaded to the blockchain, which is then digitally signed by the data submitter; the signed transaction is broadcast to nodes in the distributed ledger network, and the validity and consistency of the transaction are verified through a consensus algorithm; the verified and valid transaction is packaged into a new block and added to the distributed ledger to form an immutable on-chain notarization.

[0073] Each microgrid node autonomously generates a carbon allowance transfer proposal containing desired internal carbon emission optimization weight parameters based on its own carbon emission surplus or deficit status, operating costs, and carbon emission sensitivity information, and broadcasts the carbon allowance transfer proposal to the distributed ledger network. Each microgrid node evaluates all received carbon allowance transfer proposals and conducts multiple rounds of voting based on a Byzantine fault-tolerant consensus algorithm until a consensus is reached. Based on the voting results of the consensus, the final internal carbon emission optimization weight parameters applicable to all microgrid nodes are calculated through a weighted average or majority clearing algorithm. The internal carbon emission optimization weight parameters are finally confirmed by writing them into the distributed ledger through a smart contract and are then distributed as input parameters to the two-layer collaborative optimization model.

[0074] Specifically, the global layer integrates distributed ledger carbon data management, oracles, distributed voting pricing mechanisms, and a two-layer collaborative optimization model. It transforms external carbon emission-related market signals into internal technical parameters that drive the optimization of the system's total carbon emissions, and coordinates with the electricity market to ultimately minimize the system's total carbon emissions. The specific implementations of distributed ledger carbon data management, oracles, and distributed voting pricing mechanisms are as follows: Distributed Ledger Carbon Data Management and On-Chain Evidence Storage: Carbon emission data estimated and aggregated in real time at the device layer (L0) and microgrid layer (L1) of each microgrid is standardized and then connected to the distributed ledger network via API. Whenever a critical change in carbon emissions or a carbon quota transfer occurs, the corresponding carbon data and transfer records are written into the distributed ledger as part of the transaction information via smart contracts, forming a permanent, traceable, and tamper-proof on-chain carbon asset evidence. This process ensures the authenticity, transparency, immutability, and security of carbon data, providing a reliable basis for subsequent optimization of total carbon emissions. The carbon flow calculation method based on carbon storage ratio is applicable to various energy storage devices such as power storage and hydrogen storage, providing a foundation for carbon flow calculation.

[0075] Oracle-Driven Acquisition and Arbitration of External Carbon Emission Sensitivity Information: To obtain real-time, authoritative carbon emission sensitivity information (rather than direct trading prices) from external carbon trading markets, this invention introduces a decentralized oracle mechanism. A group of independent oracle nodes (which may consist of reputable third parties, energy data providers, or representatives of microgrids) are responsible for acquiring carbon emission sensitivity data from multiple external carbon trading markets. Since carbon trading prices (and the market sensitivity behind them) are dynamic and depend on the dynamic supply and demand of the network, the oracle mechanism is the core technical means for the decentralized confirmation and effective arbitration of dynamic carbon emission sensitivity information in the distributed ledger network.

[0076] Oracle nodes clean and reach consensus on the acquired sensitivity data through data aggregation and verification algorithms (such as median aggregation, weighted averaging, and outlier removal), and then upload the finally confirmed external reference carbon emission sensitivity parameters to the blockchain via encrypted transactions. This mechanism effectively solves the oracle problem that distributed ledgers cannot directly access external data and ensures the accuracy and resistance to manipulation of on-chain sensitivity data.

[0077] Dynamic Carbon Emission Optimization Weight Estimation and Strategy Generation under Distributed Voting Mechanism: Based on obtaining external reference carbon emission sensitivity parameters, this invention designs an internal carbon emission optimization weight estimation model based on a distributed voting mechanism. The aim is to transform external market signals into a technological driving force for internal optimization of total carbon emissions through technical means. Specifically, it includes the following steps: Bidding proposal stage: Each microgrid independently generates and broadcasts its carbon quota transfer proposal (including the quantity and expected internal carbon emission optimization weight parameters) based on its own carbon emission surplus / deficit, internal operating costs and external reference carbon emission sensitivity parameters provided by the oracle.

[0078] Voting Consensus Phase: All microgrid nodes (or authorized representative nodes) participating in carbon allowance transfer evaluate the received bidding proposals and cast their votes in favor or against. To improve robustness, a Byzantine Fault-Tolerant (BFT-like) consensus algorithm (such as a variant of Practical Byzantine Fault-Tolerant (PBFT) or Tendermint consensus) is adopted. This algorithm can guarantee the eventual consistency of the internal optimization weight parameters of the carbon allowances even in the presence of some malicious nodes or communication failures. Each node selects its voting strategy based on its own carbon emission minimization objective and its prediction of the behavior of other nodes.

[0079] Dynamic Internal Optimization Weight Estimation Model: Based on voting results, the smart contract calculates the final internal carbon emission optimization weight parameters using a weighted average or majority-clearing algorithm. These parameters consider carbon emission supply and demand, the carbon reduction potential of each microgrid (such as the amount of staked tokens, historical reputation, etc.), and external reference carbon emission sensitivity parameters. This internal optimization weight parameter directly affects the coefficients in the overall objective function, technically guiding the system to optimize carbon emissions rather than engaging in economic transactions.

[0080] Smart contract execution: Once the parameters are agreed upon, the smart contract automatically executes the transfer of carbon allowances and the corresponding internal points settlement, and records the transfer on the blockchain for evidence storage.

[0081] In this embodiment of the invention, the external market signal conversion parameters include external carbon emission sensitivity parameters and internal carbon emission optimization weight parameters. The external carbon emission sensitivity parameters reflect the sensitivity of the external market to changes in carbon emissions and are acquired and arbitrated by an oracle. The internal carbon emission optimization weight parameters are formed by a distributed voting mechanism and serve as input parameters or weighting coefficients of the objective function in a two-layer collaborative optimization model. They directly drive the upper-layer optimization to adjust market bidding strategies and, through the lower-layer optimization to adjust physical scheduling decisions, enable the system to minimize carbon emissions at the operational level.

[0082] S106: Based on the internal carbon emission optimization weight parameters, a two-layer collaborative optimization model is used to generate the joint bidding strategy of microgrid groups in the electricity market and their internal low-carbon dispatch strategy.

[0083] In this embodiment of the invention, the two-layer collaborative optimization model takes minimizing the total carbon emissions of the system determined based on system carbon flow state data as its core objective. To minimize the total operating cost of the microgrid cluster and the carbon emission technology penalty term weighted by internal carbon emission optimization weight parameters, an upper-layer optimization model is constructed and solved to generate a joint bidding strategy for the microgrid cluster in the electricity market. The market clearing price and the shadow price of carbon emission constraints are output as dual variables. Receiving the dual variables from the upper-layer model, a lower-layer optimization model is constructed and solved to minimize the operating cost of each microgrid and the local carbon emissions driven by internal carbon emission optimization weight parameters. This generates generation, energy storage, and load dispatch strategies within the microgrid cluster, and the marginal cost of each dispatch decision is fed back to the upper-layer model. The solution and variable transfer process of the upper and lower-layer models is repeated until the changes in the joint bidding strategy and the internal dispatch strategy are less than a preset threshold, resulting in a converged joint bidding strategy and a converged internal low-carbon dispatch strategy. The converged joint bidding strategy is output to the electricity market, and the converged internal low-carbon dispatch strategy is distributed to each microgrid for execution.

[0084] Specifically, this invention proposes an innovative two-layer collaborative optimization model, aiming to deeply couple the carbon emission optimization target, characterized by internal guiding parameters, with the electricity market price signal through technical means, thereby driving the generation of low-carbon dispatch strategies at the technical level. In intraday-day joint bidding, this model comprehensively considers real-time electricity prices, the internal carbon emission optimization guiding function value, and demand-side response capabilities to form the optimal bidding strategy and refined low-carbon dispatch instructions. The two-layer collaborative optimization model includes upper-layer optimization and lower-layer optimization, as detailed below: The upper-level optimization realizes the generation of market bidding strategies. Under the premise of meeting the overall power demand and total carbon emission constraints of the microgrid cluster, it optimizes the bidding strategy of the microgrid cluster in the power market, while minimizing the technical penalties related to carbon emissions or maximizing the incentives for carbon emission reduction technology contributions.

[0085] Decision variables: the amount of electricity purchased and sold by the microgrid group to the electricity market, the bidding curve, the turnover of carbon allowances, and the carbon emission optimization weight parameters.

[0086] Constraints include the total power generation capacity within the microgrid cluster, total load demand, total internal carbon quota (managed by distributed ledger), and coupling constraints with the lower-level optimization solution.

[0087] An iterative optimization algorithm based on dual decomposition is employed. The upper-level model passes the electricity market clearing price and internal carbon emission optimization weight parameters as dual variables to the lower-level optimization. After the lower-level optimization is completed, the marginal operating cost and carbon emission impact of the optimal dispatch strategy are fed back to the upper level for iterative convergence. This iterative mechanism ensures close technical consistency between the market bidding strategy and the actual physical dispatch, rather than simply transmitting price signals.

[0088] The lower-level optimization realizes the generation of low-carbon dispatch strategies within the microgrid group. Under the premise of satisfying various physical constraints within the microgrid (such as power balance, equipment output limits, voltage flow, energy storage state of charge, etc.), it minimizes the internal operating technology overhead and the total carbon emissions driven by the internal optimization weight parameters.

[0089] Decision variables: power output of generator units within each microgrid (including new energy and traditional units), energy storage charging and discharging power, demand-side response load adjustment, and interconnection power between microgrids.

[0090] Constraints include power balance of each microgrid, upper and lower limits of generator output, energy storage charge and discharge rates and state of charge constraints, line power flow and node voltage constraints, as well as technical incentives / penalties caused by market clearing prices and internal carbon emission optimization weight parameters transmitted from the upper layer.

[0091] A fusion algorithm combining personalized federated reinforcement learning and mixed-integer nonlinear programming (MINLP) is employed. First, the personalized federated reinforcement learning models at layers L0 and L1 provide efficient preliminary scheduling suggestions and parameters based on the microgrid's characteristics and local real-time data. Then, these suggestions are used as initial solutions or constraints for MINLP. By dynamically adjusting the penalty function or weights, internal carbon emission optimization weight parameters are effectively integrated into the optimization objective. For example, when the internal carbon emission optimization weight parameters are high (indicating high carbon emission priority), MINLP prioritizes the scheduling of low-carbon power sources through technological optimization methods (such as adjusting generator unit combinations, increasing energy storage charging and discharging, and guiding demand-side response), and incentivizes demand-side response to reduce electricity purchases and corresponding carbon emissions. This fusion algorithm ensures that the technological optimization objective for carbon emissions is achieved to the greatest extent possible while satisfying physical constraints. In one implementation of this invention, a dynamic carbon intensity assessment mechanism is employed, which updates node carbon intensity in real time based on power system flow calculations and uses machine learning to predict the changing trend of carbon emission factors. A carbon-electricity-power ternary coupling model, through a decentralized carbon data management mechanism, enables real-time linkage between carbon emissions and power flow, as well as external market sensitivity signals, thus enhancing the technical adaptability and optimization capabilities of low-carbon dispatch. A multi-energy coordinated dispatch strategy integrates various energy forms such as electricity, heat, cooling, and hydrogen energy to establish a unified energy flow model. A multi-objective particle swarm optimization algorithm is used to achieve a coordinated balance between energy efficiency, carbon emission minimization, and system reliability.

[0092] The technical solutions provided by this invention can significantly reduce the total carbon emissions and node carbon potential of the power grid system. Through refined carbon potential calculation and multi-objective optimization, the overall carbon emissions of power grid load nodes, generation nodes, and energy storage nodes are minimized, reducing carbon emissions by 50-80%. The invention also improves the absorption of renewable energy: by optimizing scheduling strategies and rationally arranging the output of new energy units, the utilization rate of renewable energy is increased by 20-35%. Furthermore, it optimizes power resource allocation and improves energy efficiency: by coordinating the relationship between power generation, energy storage, and power consumption, peak shaving and valley filling are achieved, improving energy efficiency by 15-25%. Finally, it enhances the robustness and adaptability of power grid operation: distributed intelligent algorithms and decentralized mechanisms improve the system's autonomous decision-making and fault tolerance capabilities in complex and dynamic environments. Finally, it ensures data privacy and security: federated learning and distributed ledger technology ensure the privacy of sensitive operational data and the immutability of carbon data notarization.

[0093] This invention discloses a hierarchical distributed collaborative scheduling method for multiple microgrids. This method creatively couples refined modeling of grid carbon flow, demand-side response, and multi-microgrid collaborative scheduling within the same algorithmic framework, aiming to minimize the total carbon emissions of the power grid system and the carbon potential of its nodes. By constructing a hierarchical model that unifies physical scale (device-microgrid-microgrid cluster) and temporal scale (second-minute-intraday-pre-day), it effectively solves the problem that traditional single-scale models cannot simultaneously account for alternating fast and slow dynamics, improving the refinement and real-time performance of carbon flow scheduling, thereby enhancing the system's precise control over carbon emissions. This invention introduces a hybrid intelligent optimization algorithm of "federated learning + reinforcement learning," which, while ensuring the privacy of data in each microgrid (only exchanging model gradients, without uploading original data), enhances the autonomous decision-making ability of each microgrid node and the computational scalability of the system, enabling it to respond more efficiently to low-carbon scheduling commands. Furthermore, this invention innovatively introduces distributed ledger technology to achieve secure, transparent management and on-chain notarization of grid carbon data and its physical carbon footprint, and supports the transfer of carbon quotas between microgrid nodes. Given the dynamic nature of external carbon market signals, this invention integrates oracle technology to achieve decentralized confirmation and effective arbitration of external carbon emission sensitivity information in a distributed ledger network. Based on this, the invention designs a dynamic carbon emission optimization weight estimation model based on a distributed voting mechanism (similar to Byzantine fault tolerance) to assist microgrid nodes in forming consensus carbon emission priority parameters in carbon quota trading, thereby transforming external market signals into a driving function for internal optimization of total carbon emissions. This method takes carbon emissions as the core technical objective of dynamic optimization and, through an innovative two-layer collaborative optimization model and a specific personalized federated reinforcement learning and mixed-integer nonlinear programming (MINLP) fusion algorithm, optimizes the allocation of power resources at different stages while satisfying physical constraints, significantly reducing the overall system carbon emissions, improving renewable energy absorption and energy utilization efficiency, and enhancing the robustness and adaptability of grid operation.

[0094] Based on the methods provided in the above embodiments, this invention also provides a multi-microgrid hierarchical distributed collaborative scheduling system, which is described below with reference to the accompanying drawings.

[0095] See Figure 2 The figure is a schematic diagram of the structure of a multi-microgrid hierarchical distributed collaborative scheduling system provided in an embodiment of the present invention.

[0096] The multi-microgrid hierarchical distributed collaborative scheduling system 200 provided in this embodiment of the invention includes: a carbon flow calculation module 201, a hierarchical model architecture module 202, and a two-layer collaborative optimization module 203.

[0097] The carbon flow calculation module 201 is used to construct a carbon flow calculation function for quantifying system carbon emissions in order to obtain system carbon flow status data, which includes the carbon potential of load nodes, the carbon emission intensity of power generation nodes, and the carbon potential changes of energy storage nodes. The hierarchical model architecture module 202 is used to construct a hierarchical model that includes a device layer, a microgrid layer, and a global layer. The device layer is used for rolling optimization based on reinforcement learning algorithms to generate low-carbon scheduling instructions for local devices. The optimization objective of reinforcement learning includes minimizing local carbon emission data calculated based on system carbon flow state data. The microgrid layer is used to aggregate the model gradients of multiple microgrids based on federated learning algorithms, and to calculate the carbon quota surplus and deficit of each microgrid based on the system carbon flow state data, so as to collaboratively optimize the power mutual assistance and virtual carbon quota allocation among microgrids, and generate carbon quota flow records. The global layer is used to store local carbon emission data and carbon quota transfer records on-chain using distributed ledger technology; obtain carbon emission sensitivity information from external carbon markets through oracles; and generate internal carbon emission optimization weight parameters based on the carbon emission sensitivity information through a distributed consensus mechanism. The two-layer collaborative optimization module 203 is used to generate a joint bidding strategy for the microgrid group in the electricity market and its internal low-carbon dispatch strategy based on the internal carbon emission optimization weight parameters and a two-layer collaborative optimization model. The core objective of the two-layer collaborative optimization model is to minimize the total carbon emissions of the system determined based on the system carbon flow state data.

[0098] In one possible implementation, the carbon flow calculation function includes: A carbon potential calculation function used to characterize the accumulation of carbon emissions generated at power grid generation nodes at power grid load nodes after being distributed by power flow; A power generation node carbon emission intensity function used to quantify carbon emissions per unit of electricity generated; A function describing the change in carbon potential at an energy storage node, used to describe the effect of energy storage charging and discharging on carbon potential absorption or release.

[0099] In one possible implementation, the hierarchical model architecture module 202 is specifically used for: Obtain the current system status, which includes real-time carbon intensity, energy storage system state of charge, local load forecast, and renewable energy output forecast provided by system carbon flow status data; The system state is input into a pre-trained reinforcement learning model to obtain the optimal action strategy, which includes the target output of the local generator set, the target charging and discharging power of the energy storage, and the target adjustment amount of the controllable load. The optimal action strategy is converted into control commands and sent to the corresponding local devices for execution, so as to minimize local carbon emissions while maintaining local power balance. In the next scheduling cycle, the above process is repeated based on the new system state to achieve real-time rolling generation and optimization of low-carbon scheduling instructions.

[0100] In one possible implementation, the hierarchical model architecture module 202 has features for: Based on the system carbon flow status data, the difference between the actual carbon emissions of each microgrid and the preset carbon quota benchmark value during the dispatch cycle is calculated to determine the carbon quota accounting result. The carbon quota accounting result includes the carbon quota surplus or deficit status of each microgrid. Each microgrid trains a scheduling model based on local data and uploads the model gradient to the regional scheduler. The regional scheduler uses a federated averaging algorithm to perform a weighted average of the received model gradients from each microgrid and aggregates them to generate a global scheduling model. Based on the carbon quota accounting results and the global scheduling model, the Monte Carlo hierarchical optimization method is adopted. Multiple candidate schemes for power mutual assistance and virtual carbon quota allocation are generated by random sampling. The impact of each candidate scheme on the total carbon emissions of the system is simulated and evaluated. The candidate scheme that minimizes the total carbon emissions of the system is selected as the collaborative optimization scheme. The collaborative optimization scheme includes the optimal power mutual assistance strategy and carbon quota allocation strategy. According to the collaborative optimization scheme, a corresponding virtual carbon quota transfer record is generated. The virtual carbon quota transfer record shall include at least the identifiers of the two parties involved in the carbon quota transfer, the amount of carbon quota transferred, the corresponding power mutual assistance amount, and timestamp information.

[0101] In one possible implementation, the hierarchical model architecture module 202 has features for: Standardize the format of local carbon emission data and carbon quota transfer records; The evidence storage smart contract is invoked through the application programming interface (API) and deployed in the distributed ledger network. The standardized local carbon emission data and carbon quota transfer records are used as the transaction content to generate a transaction to be uploaded to the blockchain and digitally signed by the data submitter. The signed transaction is broadcast to nodes in the distributed ledger network, and the validity and consistency of the transaction are verified by a consensus algorithm. Validated transactions are packaged into new blocks and added to the distributed ledger to form immutable on-chain evidence.

[0102] In one possible implementation, the hierarchical model architecture module 202 has features for: Each microgrid node autonomously generates a carbon allowance transfer proposal containing desired internal carbon emission optimization weight parameters based on its own carbon emission surplus or deficit status, operating costs, and carbon emission sensitivity information, and broadcasts the carbon allowance transfer proposal to the distributed ledger network. Each microgrid node evaluates all received carbon quota transfer proposals and conducts multiple rounds of voting based on a Byzantine fault-tolerant consensus algorithm until a consensus is reached. Based on the consensus-reaching voting results, the final internal carbon emission optimization weight parameters applicable to all microgrid nodes are calculated using a weighted average or majority clearing algorithm. The internal carbon emission optimization weight parameters are written into the distributed ledger through a smart contract for final confirmation and then sent to the two-layer collaborative optimization model as input parameters.

[0103] In one possible implementation, the two-layer collaborative optimization module 203 has the following features: With the goal of minimizing the total operating cost of the microgrid cluster and the carbon emission technology penalty term weighted by the internal carbon emission optimization weight parameters, an upper-level optimization model is constructed and solved to generate a joint bidding strategy for the microgrid cluster in the electricity market, and the market clearing price and the shadow price of carbon emission constraints are output as dual variables. The lower-level optimization model is constructed and solved with the goal of minimizing the operating costs of each microgrid and the local carbon emissions driven by the internal carbon emission optimization weight parameters. This generates the power generation, energy storage and load dispatch strategies within the microgrid group and feeds back the marginal cost of each dispatch decision to the upper-level model. Repeat the solution and variable transfer process of the upper-level model and the lower-level model until the change in the joint bidding strategy and the internal scheduling strategy is less than the preset threshold, and obtain the converged joint bidding strategy and the converged internal low-carbon scheduling strategy. The converged joint bidding strategy is output to the electricity market, and the converged internal low-carbon dispatch strategy is distributed to each microgrid for execution.

[0104] Since the system 200 is a system corresponding to the multi-microgrid hierarchical distributed cooperative scheduling method provided in the above method embodiments, the specific implementation of each unit of the system 200 is based on the same concept as in the above method embodiments. Therefore, for the specific implementation of each unit of the system 200, please refer to the description of the multi-microgrid hierarchical distributed cooperative scheduling method in the above method embodiments, and it will not be repeated here.

[0105] This invention also provides a multi-microgrid hierarchical distributed collaborative scheduling device, the device comprising: a processor and a memory; The memory is used to store instructions; The processor is used to execute the instructions in the memory to perform the multi-microgrid hierarchical distributed cooperative scheduling method mentioned in the above embodiments.

[0106] It should be noted that the hardware structure of the multi-microgrid hierarchical distributed collaborative scheduling device provided in the embodiments of the present invention can be as follows: Figure 3 The structure shown, Figure 3 This is a schematic diagram of the structure of a device provided in an embodiment of the present invention.

[0107] Please see Figure 3 As shown, device 300 includes: a processor 310, a communication interface 320, and a memory 330. The number of processors 310 in device 300 can be one or more. Figure 3 Taking a processor as an example, in this embodiment of the invention, the processor 310, communication interface 320, and memory 330 can be connected via a bus system or other means. Figure 3 Taking the connection between China and Israel via bus system 340 as an example.

[0108] Processor 310 may be a central processing unit (CPU), a network processor (NP), or a combination of a CPU and an NP. Processor 310 may further include hardware chips. These hardware chips may be application-specific integrated circuits (ASICs), programmable logic devices (PLDs), or combinations thereof. The PLD may be a complex programmable logic device (CPLD), a field-programmable gate array (FPGA), generic array logic (GAL), or any combination thereof.

[0109] The memory 330 may include volatile memory, such as random-access memory (RAM); the memory 330 may also include non-volatile memory, such as flash memory, hard disk drive (HDD) or solid-state drive (SSD); the memory 330 may also include a combination of the above types of memory.

[0110] Optionally, the memory 330 stores an operating system and programs, executable modules, or data structures, or subsets thereof, or extended sets thereof. The programs may include various operation instructions for implementing various operations. The operating system may include various system programs for implementing various basic services and handling hardware-based tasks. The processor 310 can read the programs in the memory 330 to implement the multi-microgrid hierarchical distributed cooperative scheduling method provided in this embodiment of the invention.

[0111] The bus system 340 can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. The bus system 340 can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 3 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0112] This invention also provides a computer-readable storage medium including instructions that, when executed on a computer, cause the computer to perform the multi-microgrid hierarchical distributed cooperative scheduling method mentioned in the above embodiments.

[0113] This invention also provides a computer program product containing instructions that, when run on a computer, causes the computer to execute the multi-microgrid hierarchical distributed cooperative scheduling method mentioned in the above embodiments.

[0114] Although the invention has been specifically shown and described in conjunction with preferred embodiments, those skilled in the art should understand that various changes in form and detail may be made to the invention without departing from the spirit and scope of the invention as defined in the appended claims, all of which shall be within the scope of protection of the invention.

Claims

1. A hierarchical distributed cooperative scheduling method for multiple microgrids, characterized in that, include: A carbon flow calculation function is constructed to quantify system carbon emissions in order to obtain system carbon flow state data, which includes the carbon potential of load nodes, the carbon emission intensity of power generation nodes, and the carbon potential changes of energy storage nodes. Construct a hierarchical model that includes a device layer, a microgrid layer, and a global layer; At the device layer, rolling optimization is performed based on reinforcement learning algorithms to generate low-carbon scheduling instructions for local devices. The optimization objective of the reinforcement learning includes minimizing the local carbon emission data calculated based on the system carbon flow state data. At the microgrid layer, the model gradients of multiple microgrids are aggregated based on the federated learning algorithm, and the carbon quota surplus or deficit of each microgrid is calculated based on the carbon flow state data of the system, so as to collaboratively optimize the power mutual assistance and carbon quota allocation among microgrids and generate carbon quota flow records. At the global layer, distributed ledger technology is used to store the local carbon emission data and the carbon quota transfer records on the blockchain; carbon emission sensitivity information is obtained from the external carbon market through oracles; and based on the carbon emission sensitivity information, internal carbon emission optimization weight parameters are generated through a distributed voting mechanism. Based on the internal carbon emission optimization weight parameters, a two-layer collaborative optimization model is adopted to generate a joint bidding strategy for the microgrid group in the electricity market and its internal low-carbon dispatch strategy. The core objective of the two-layer collaborative optimization model is to minimize the total carbon emissions of the system determined based on the system carbon flow state data.

2. The method according to claim 1, characterized in that, The carbon flow calculation function includes: A carbon potential calculation function used to characterize the accumulation of carbon emissions generated at power grid generation nodes at power grid load nodes after being distributed by power flow; A power generation node carbon emission intensity function used to quantify carbon emissions per unit of electricity generated; A function describing the change in carbon potential at an energy storage node, used to describe the effect of energy storage charging and discharging on carbon potential absorption or release.

3. The method according to claim 1, characterized in that, At the device layer, rolling optimization is performed based on a reinforcement learning algorithm to generate low-carbon scheduling instructions for the local device, including: Obtain the current system status, which includes real-time carbon intensity, energy storage system state of charge, local load forecast, and renewable energy output forecast provided by the system carbon flow status data; The system state is input into a pre-trained reinforcement learning model to obtain the optimal action strategy, which includes the target output of the local generator set, the target charging and discharging power of the energy storage, and the target adjustment amount of the controllable load. The optimal action strategy is converted into control commands and sent to the corresponding local devices for execution, so as to minimize local carbon emissions while maintaining local power balance. In the next scheduling cycle, the above process is repeated based on the new system state to realize the real-time rolling generation and optimization of the low-carbon scheduling instructions.

4. The method according to claim 1, characterized in that, At the microgrid layer, model gradients from multiple microgrids are aggregated based on a federated learning algorithm, and the carbon allowance surplus or deficit of each microgrid is calculated based on the system's carbon flow state data. This is to collaboratively optimize power sharing and carbon allowance allocation among microgrids and generate carbon allowance transfer records, including: Based on the carbon flow status data of the system, the difference between the actual carbon emissions of each microgrid during the scheduling cycle and the preset carbon quota benchmark value is calculated to determine the carbon quota accounting result. The carbon quota accounting result includes the carbon quota surplus or deficit status of each microgrid. Each microgrid trains a scheduling model based on local data and uploads the model gradient to the regional scheduler. The regional scheduler uses a federated averaging algorithm to perform a weighted average of the received model gradients of each microgrid and aggregates them to generate a global scheduling model. Based on the carbon quota calculation results and the global scheduling model, the Monte Carlo hierarchical optimization method is adopted to generate multiple candidate schemes for power mutual assistance and carbon quota allocation through random sampling. The impact of each candidate scheme on the total carbon emissions of the system is simulated and evaluated, and the candidate scheme that minimizes the total carbon emissions of the system is selected as the collaborative optimization scheme. The collaborative optimization scheme includes the optimal power mutual assistance strategy and carbon quota allocation strategy. According to the collaborative optimization scheme, a corresponding carbon quota transfer record is generated. The carbon quota transfer record includes at least the identifiers of both parties involved in the carbon quota transfer, the amount of carbon quota transferred, the corresponding power mutual assistance amount, and timestamp information.

5. The method according to claim 1, characterized in that, The method of using distributed ledger technology to perform on-chain notarization of the local carbon emission data and the carbon quota transfer records includes: The local carbon emission data and the carbon quota transfer records are processed in a standardized format. The evidence storage smart contract is invoked through the application programming interface (API) and deployed in the distributed ledger network. The local carbon emission data processed in a standardized format and the carbon quota transfer record are used as transaction content to generate a transaction to be uploaded to the blockchain and digitally signed by the data submitter. The signed transaction is broadcast to nodes in the distributed ledger network, and the validity and consistency of the transaction are verified by a consensus algorithm. Validated transactions are packaged into new blocks and added to the distributed ledger to form immutable on-chain evidence.

6. The method according to claim 1, characterized in that, The step of generating internal carbon emission optimization weight parameters based on the carbon emission sensitivity information through a distributed voting mechanism includes: Each microgrid node autonomously generates a carbon allowance transfer proposal containing desired internal carbon emission optimization weight parameters based on its own carbon emission surplus or deficit status, operating costs, and carbon emission sensitivity information, and broadcasts the carbon allowance transfer proposal to the distributed ledger network. Each microgrid node evaluates all received carbon quota transfer proposals and conducts multiple rounds of voting based on a Byzantine fault-tolerant consensus algorithm until a consensus is reached. Based on the consensus-reaching voting results, the final internal carbon emission optimization weight parameters applicable to all microgrid nodes are calculated using a weighted average or majority clearing algorithm. The internal carbon emission optimization weight parameters are written into the distributed ledger via a smart contract for final confirmation and then sent as input parameters to the two-layer collaborative optimization model.

7. The method according to claim 1, characterized in that, Based on the internal carbon emission optimization weight parameters, a two-layer collaborative optimization model is used to generate a joint bidding strategy for the microgrid group in the electricity market and its internal low-carbon dispatch strategy, including: With the goal of minimizing the total operating cost of the microgrid cluster and the carbon emission technology penalty term weighted by the internal carbon emission optimization weight parameters, an upper-level optimization model is constructed and solved to generate a joint bidding strategy for the microgrid cluster in the electricity market, and the market clearing price and the shadow price of carbon emission constraints are output as dual variables. The system receives dual variables from the upper-level model and aims to minimize the operating costs of each microgrid and the local carbon emissions driven by the internal carbon emission optimization weight parameters. It then constructs and solves the lower-level optimization model to generate power generation, energy storage, and load dispatch strategies within the microgrid group and feeds back the marginal costs of each dispatch decision to the upper-level model. Repeat the solution and variable transfer process of the upper-level model and the lower-level model until the change in the joint bidding strategy and the internal scheduling strategy is less than the preset threshold, and obtain the converged joint bidding strategy and the converged internal low-carbon scheduling strategy. The converged joint bidding strategy is output to the electricity market, and the converged internal low-carbon dispatch strategy is distributed to each microgrid for execution.

8. A hierarchical distributed collaborative scheduling system for multiple microgrids, characterized in that, include: The carbon flow calculation module is used to construct a carbon flow calculation function for quantifying system carbon emissions in order to obtain system carbon flow status data, which includes the carbon potential of load nodes, the carbon emission intensity of power generation nodes, and the carbon potential changes of energy storage nodes. The hierarchical model architecture module is used to build a hierarchical model that includes a device layer, a microgrid layer, and a global layer. The device layer is used to perform rolling optimization based on reinforcement learning algorithms to generate low-carbon scheduling instructions for local devices. The optimization objective of the reinforcement learning includes minimizing the local carbon emission data calculated based on the system carbon flow state data. The microgrid layer is used to aggregate the model gradients of multiple microgrids based on the federated learning algorithm, and to calculate the carbon quota surplus and deficit of each microgrid based on the carbon flow state data of the system, so as to collaboratively optimize the power mutual assistance and carbon quota allocation among microgrids and generate carbon quota flow records. The global layer is used to utilize distributed ledger technology to perform on-chain notarization of the local carbon emission data and the carbon quota transfer records; obtain carbon emission sensitivity information from the external carbon market through oracles; and generate internal carbon emission optimization weight parameters based on the carbon emission sensitivity information through a distributed voting mechanism. The two-layer collaborative optimization module is used to generate a joint bidding strategy for the microgrid group in the electricity market and its internal low-carbon dispatch strategy based on the internal carbon emission optimization weight parameters and a two-layer collaborative optimization model. The two-layer collaborative optimization model takes minimizing the total carbon emissions of the system determined based on the system carbon flow state data as its core objective.

9. A multi-microgrid hierarchical distributed collaborative scheduling device, characterized in that, The device includes: a processor and a memory; The memory is used to store instructions; The processor is configured to execute the instructions in the memory to perform the method according to any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, Including instructions that, when run on a computer, cause the computer to perform the method described in any one of claims 1-7 above.

Images