Multi-energy complementary zero-carbon campus ac-dc microgrid configuration method and system

Abstract

The application discloses a multi-energy complementary zero-carbon campus AC / DC microgrid configuration method and system, relates to the technical field of communication network energy management, and comprises a photovoltaic power generation unit, a wind power generation unit, an energy storage unit and a communication base station load, specifically comprises the following steps: collecting system state data of the microgrid, wherein the system state data comprises real-time wind and light power generation power, an energy storage unit state of charge, base station load power demand and real-time carbon intensity information; based on the system state data, a base station microgrid multi-time scale optimization scheduling model is constructed by fusing dynamic carbon perception; by establishing a carbon flow tracking model and embedding it as an endogenous variable and a constraint condition into the optimization scheduling model, the operation optimization logic of the microgrid is fundamentally changed, carbon emission is changed from a post-event statistical index into a pre-event control target and a process constraint, and synchronous optimization and minimization of energy flow and carbon flow are realized.

Description

Technical Field

[0001] This invention relates to the field of energy management technology for communication networks, and more specifically, to a method and system for configuring a zero-carbon AC / DC microgrid on campus with multiple complementary energy sources. Background Technology

[0002] With the rapid development of 5G / 6G communication technologies, the number of communication base stations and their energy consumption have increased dramatically, making their operating costs and carbon emissions increasingly prominent. Utilizing renewable energy sources such as photovoltaics and wind power to supply power to base stations and constructing off-grid or grid-connected microgrids has become an important technological path for the industry to reduce carbon emissions.

[0003] However, existing microgrid operation methods for base stations have the following significant drawbacks: First, most existing optimization methods focus solely on economic cost or power supply reliability as primary objectives, lacking direct and refined control over the key indicator of "carbon emissions." Carbon emissions are typically treated as post-hoc statistics rather than real-time control variables during operation, making it difficult for the system to achieve truly "low-carbon" or "zero-carbon" operation. Second, the intermittency of renewable energy and the sudden fluctuations in base station service loads pose significant challenges to the real-time power balance and stability control of microgrids. Traditional control methods based on fixed rules or simple model predictions have poor adaptability and limited optimization effects when facing complex and changing weather conditions and load scenarios. Finally, the fixed topology of existing microgrids makes it difficult to adapt to dynamic changes in base station loads (such as service migration and equipment expansion) or the need for fault isolation, resulting in insufficient system flexibility and resilience.

[0004] Therefore, there is an urgent need for a base station microgrid operation optimization scheme that can deeply embed carbon emissions into the core of operation control, has a high degree of self-adaptation and self-optimization capabilities, and can flexibly adjust the system structure. Summary of the Invention

[0005] To overcome the aforementioned deficiencies of existing technologies, this invention provides a method and system for configuring a zero-carbon AC / DC microgrid on campus with multiple complementary energy sources. Its core objective is to deeply integrate carbon emission flow as an endogenous constraint and optimization target into the full-scale scheduling and control of the microgrid, and to achieve synergistic optimization of the system under multiple objectives of economy, reliability and low carbon emissions through digital twin and advanced control technologies.

[0006] To achieve the above objectives, the present invention provides the following technical solution: A method for configuring a zero-carbon AC / DC microgrid for a campus with multiple complementary energy sources, including photovoltaic power generation units, wind power generation units, energy storage units, and communication base station loads, specifically includes the following steps: Collecting system status data of the microgrid, including real-time wind and solar power generation, energy storage unit state of charge, base station load power demand, and real-time carbon intensity information; Based on the system status data, constructing a multi-timescale optimized scheduling model for the base station microgrid integrating dynamic carbon sensing; The optimized scheduling model takes operational economy, power supply reliability, and carbon emission intensity as collaborative optimization objectives, and includes equipment operation constraints and carbon flow constraints; Inputting the system status data into a pre-trained digital twin agent model, outputting future multi-timescale wind and solar power generation and load demand prediction results; Based on the prediction results and the optimized scheduling model, solving for the optimized scheduling command sequences of the photovoltaic power generation units, wind power generation units, and energy storage units at the day-ahead, intraday, and real-time scales; Based on the optimized scheduling command sequences, using a coordinating controller to perform real-time control of each unit in the microgrid to achieve low-carbon and economical operation.

[0007] In a preferred embodiment, based on the system state data, a multi-timescale optimization scheduling model for a base station microgrid integrating dynamic carbon sensing is constructed. This optimization scheduling model takes operational economy, power supply reliability, and carbon emission intensity as collaborative optimization objectives, and includes equipment operation constraints and carbon flow constraints. Specifically, it establishes a carbon flow tracking model for the microgrid, quantifies the carbon emission flow within the microgrid and each power supply link based on the real-time carbon intensity information and power flow calculation, treats the carbon emission flow as an endogenous variable, and uses it together with system operating costs and load shortage rate to form a multi-objective optimization function. Based on the physical characteristics and operational limitations of the equipment in the microgrid, a set of equipment operation constraints is constructed, including energy storage charging and discharging status, power ramping, and bus voltage stability. Combined with the carbon flow tracking model, a carbon flow constraint set is constructed, which is used to control the total carbon emissions or carbon emission intensity of the microgrid operation within a preset threshold.

[0008] In a preferred embodiment, the system state data is input into a pre-trained digital twin proxy model, which outputs predictions of future wind and solar power generation and load demand at multiple time scales. Specifically, based on the historical operating data of the microgrid, high-precision weather forecast data, and base station service data, a system-level digital twin is constructed to perform a high-fidelity mapping of the physical microgrid. Massive scenario simulations are performed within the digital twin to generate a training sample set covering various typical and extreme weather conditions and load fluctuations. A deep reinforcement learning algorithm is used, with the optimization objective of the optimized scheduling model as the reward function, to train the proxy model, enabling it to quickly infer an approximately optimal scheduling strategy based on the input system state data.

[0009] In a preferred embodiment, based on the prediction results and the optimized scheduling model, the optimized scheduling instruction sequences for the photovoltaic power generation unit, wind power generation unit, and energy storage unit at the day-ahead, intraday, and real-time scales are obtained. Specifically: at the day-ahead scale, based on refined weather forecasts for the next 24-72 hours, with the goal of minimizing total daily operating costs and total carbon emissions, the basic charge-discharge plan for the energy storage unit and the day-ahead plan curves for each power generation unit are obtained; at the intraday rolling scale, based on updated ultra-short-term wind and solar forecasts and load forecasts, with the goal of minimizing adjustment costs and carbon emission fluctuations, the day-ahead plan curves are rolled over and corrected; at the real-time scale, based on the second- to minute-level optimization instructions output by the proxy model, the output of the energy storage converter and distributed power sources is dynamically adjusted to quickly smooth out power fluctuations and ensure that real-time carbon intensity meets constraints.

[0010] In a preferred embodiment, the method further includes: real-time monitoring of changes in the topology and load demand of the microgrid; when a significant change in the load of the communication base station is detected due to service migration or equipment start-up and shutdown, or when some lines of the microgrid need maintenance, dynamically reconstructing the network topology of the microgrid; after topology reconstruction, performing distributed collaborative control on multiple parallel-operating energy storage converters based on a consensus algorithm to achieve autonomous power distribution and synchronization, and maintain system stability.

[0011] In a preferred embodiment, the microgrid's topology and load demand changes are monitored in real time. When a significant change in the communication base station load is detected due to service migration or equipment start-up / shutdown, or when some lines of the microgrid require maintenance, the network topology of the microgrid is dynamically reconstructed. Specifically, based on the distribution and importance level of the communication base station load, the microgrid is logically divided into multiple power supply zones, including core load areas and flexible load areas. Through software-defined networking and solid-state switches, in response to load changes or faults, the network connection mode is automatically switched to achieve flexible networking and energy sharing between the power supply zones.

[0012] In a preferred embodiment, the optimized scheduling command sequence of the photovoltaic power generation unit, wind power generation unit, and energy storage unit at the day-ahead, intraday, and real-time scales is obtained by solving the solution, and the following steps are also performed: according to the power fluctuation spectrum characteristics in the prediction results, the high-frequency and low-energy power fluctuation components are allocated to the power-type electrochemical energy storage for rapid response; the low-frequency and high-energy energy transfer requirements are allocated to the energy-type energy storage.

[0013] A photovoltaic-wind power complementary microgrid operation optimization system for low-carbon base stations, the system comprising: The data acquisition module is used to collect system status data of the microgrid; The model building module is used to build and update the multi-time-scale optimization scheduling model of the base station microgrid that integrates dynamic carbon sensing. The prediction and decision-making module, which is embedded with the pre-trained digital twin agent model, is used to receive the system state data and make predictions and rapid optimization decisions to generate an optimized scheduling instruction sequence. The coordination and control module is used to execute the optimized scheduling instruction sequence and send control instructions to the photovoltaic inverters, wind power converters, energy storage converters and network switches in the microgrid; The topology management module is used to monitor the system status and dynamically manage the network topology of the microgrid. In a preferred embodiment, the system further includes an energy management platform deployed in the cloud, which is communicatively connected to multiple microgrid systems; The energy management platform is used to aggregate the regulation capabilities of multiple microgrids and, based on blockchain technology, to realize point-to-point green electricity trading and carbon emission reduction verification among multiple microgrids.

[0014] The technical effects and advantages of the multi-energy complementary zero-carbon campus AC / DC microgrid configuration method and system of this invention are as follows: By establishing a carbon flow tracking model and embedding it as an endogenous variable and constraint into the optimization scheduling model, the operation optimization logic of microgrids has been fundamentally changed. Carbon emissions have been transformed from a post-event statistical indicator into a pre-event control target and process constraint, realizing the synchronous optimization and minimization of energy flow and carbon flow. This effectively overcomes the technical contradiction of the difficulty in coordinating economy and low carbon in traditional methods.

[0015] A digital twin agent model based on deep reinforcement learning is introduced. By learning from massive amounts of complex scenarios in a virtual twin, this model achieves near-global optimal rapid decision-making capabilities, effectively addressing the strong uncertainties of wind and solar resources and loads. It solves the problems of slow online computation speed and high dependence on model accuracy in traditional optimization algorithms, realizing an upgrade from model-driven to data and model fusion-driven approaches.

[0016] By using software-defined networking and solid-state switches to achieve dynamic reconfiguration of the network topology, the microgrid can flexibly adjust its structure according to load changes and operating conditions, isolate faults, and optimize power flow distribution. Combined with distributed energy storage collaborative control based on consensus algorithms, the system is ensured to stabilize rapidly after topology changes, significantly improving its adaptability to internal disturbances and external uncertainties, as well as its power supply resilience.

[0017] A multi-timescale optimized scheduling framework was constructed, encompassing day-ahead planning, intraday rolling correction, and real-time second-level control, achieving seamless integration of long-term energy management and instantaneous power balance. Combined with spectrum-level management of hybrid energy storage, precise suppression of power fluctuations at different frequencies and energy levels was achieved, optimizing the operational lifespan and efficiency of energy storage devices while ensuring power quality. Attached Figure Description

[0018] Figure 1 This is a flowchart illustrating the multi-energy complementary zero-carbon campus AC / DC microgrid configuration method of the present invention.

[0019] Figure 2 This is a schematic diagram of the structure of the multi-energy complementary zero-carbon campus AC / DC microgrid configuration system of the present invention.

[0020] Figure 3 This is a schematic diagram of the multi-timescale optimization scheduling mechanism that integrates dynamic carbon sensing in the multi-energy complementary zero-carbon campus AC / DC microgrid configuration method of the present invention.

[0021] Figure 4 This is a schematic diagram of the dynamic reconfiguration of microgrid topology and coordinated control of energy storage in the multi-energy complementary zero-carbon campus AC / DC microgrid configuration method of the present invention. Detailed Implementation

[0022] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0023] Example 1, Figure 1 , Figure 2 and Figure 3This invention presents a method for configuring a zero-carbon AC / DC microgrid on a campus using multiple energy sources, including photovoltaic (PV) power generation units, wind power generation units, energy storage units, and communication base station loads. The method comprises the following steps: collecting system state data of the microgrid, including real-time wind and solar power generation, energy storage unit state of charge, base station load power demand, and real-time carbon intensity information; constructing a multi-timescale optimized scheduling model for the base station microgrid based on the system state data; the optimized scheduling model uses operational economy, power supply reliability, and carbon emission intensity as collaborative optimization objectives, and includes equipment operation constraints and carbon flow constraints; inputting the system state data into a pre-trained digital twin proxy model, outputting future multi-timescale predictions of wind and solar power generation and load demand; solving for optimized scheduling command sequences for the PV power generation units, wind power generation units, and energy storage units at the day-ahead, intraday, and real-time scales based on the prediction results and the optimized scheduling model; and using a coordinating controller to perform real-time control of each unit in the microgrid based on the optimized scheduling command sequences to achieve low-carbon and economical operation.

[0024] In this embodiment, a multi-timescale optimization scheduling model for a base station microgrid integrating dynamic carbon sensing is constructed based on system state data. The optimization scheduling model takes operational economy, power supply reliability, and carbon emission intensity as synergistic optimization objectives and includes equipment operation constraints and carbon flow constraints. Specifically, a carbon flow tracking model for the microgrid is established, and the carbon emission flow within the microgrid and each power supply link is quantified based on real-time carbon intensity information and power flow calculation. The carbon emission flow is treated as an endogenous variable and, together with system operating cost and load shortage rate, constitutes a multi-objective optimization function. Based on the physical characteristics and operational limitations of the equipment in the microgrid, a set of equipment operation constraints is constructed, including energy storage charging and discharging status, power ramping, and bus voltage stability. Combined with the carbon flow tracking model, a set of carbon flow constraints is constructed, which is used to control the total carbon emissions or carbon emission intensity of the microgrid operation within a preset threshold.

[0025] It is important to note that establishing a carbon flow tracing model for microgrids quantifies carbon emission flows at each stage. Traditional microgrid carbon emission calculations only count the total carbon emissions, failing to pinpoint the sources of carbon emissions (such as zero-carbon wind and solar power, equivalent carbon electricity from energy storage, and electricity purchased from the main grid) and the carbon flow distribution at each power supply stage. This results in low-carbon optimization being only statistically analyzed retrospectively, without process control. This step, through carbon flow tracing, achieves precise source tracking and full-chain quantification of carbon emissions, providing a data foundation for subsequent carbon constraint embedding.

[0026] Based on two core data sources—real-time carbon intensity information and power flow calculation—a full-link carbon flow quantification system is built, specifically divided into three levels: **Basic Data Reliance:** Real-time carbon intensity information primarily refers to the real-time carbon intensity of the external power grid (dynamically changing with the regional power generation structure; for example, a high proportion of thermal power results in high carbon intensity, while a high proportion of hydropower / new energy results in low carbon intensity). Simultaneously, the equivalent carbon intensity during energy storage charging needs to be calculated (different energy storage charging sources contribute different amounts of carbon emissions during discharge; for example, charging wind and solar power results in an equivalent carbon intensity of 0, while charging the main grid results in an equivalent carbon intensity equal to the grid carbon intensity during charging). **Power Flow Calculation:** Through microgrid power flow analysis, the power flow direction and magnitude of each node (PV / wind power nodes, energy storage nodes, load nodes, and main grid interaction nodes) are clarified, determining the energy transmission path within the microgrid. **Quantification Core Action:** A carbon intensity transfer matrix is ​​constructed at the node level, binding real-time carbon intensity with power flow, and calculating carbon emission flows link by link. For example: Wind and solar power nodes: carbon intensity is 0 (zero-carbon power source), and the carbon flow carried by its output power is 0; Energy storage discharge nodes: carbon flow is the equivalent carbon intensity accumulated during the energy storage charging stage multiplied by the discharge power, realizing full-cycle traceability of carbon emissions from charging to discharging; Large grid interaction nodes: carbon flow is the real-time carbon intensity of the grid multiplied by the interaction power (carbon flow is introduced when electricity is purchased and output when electricity is sold); Base station load nodes: carbon flow is the superposition of carbon flows from each power supply path (wind and solar, energy storage, and the large grid), ultimately quantifying the carbon emissions corresponding to the electricity consumed by the load. This transforms abstract carbon emissions into measurable and traceable carbon flows, breaking the limitation of only calculating the total amount, making the carbon emission contribution of each link in the microgrid visible, and providing a precise basis for subsequently incorporating carbon emissions into optimization targets.

[0027] A multi-objective optimization function incorporating carbon emission flows is constructed. Traditional optimization functions often prioritize minimizing operating costs or power shortage rates as a single or dual objective, which can easily lead to contradictions such as sacrificing low-carbon emissions for cost reduction or increasing costs and reducing reliability for low-carbon emissions. This step treats carbon emission flows as an endogenous variable (rather than an externally imposed indicator) and deeply integrates them with economic and reliability objectives to construct a synergistic optimization function that achieves a balance among the three.

[0028] The specific connotations and integration methods of the three objectives are as follows: Objective 1: Operational economy (core is cost control), covering the entire life cycle operation cost of the microgrid, including: wind and solar curtailment loss cost (opportunity cost caused by abandoning zero-carbon electricity), energy storage charging and discharging loss cost (energy loss discount during charging and discharging), interaction cost with the main grid (electricity purchase cost - electricity sales revenue), and equipment operation and maintenance cost (daily maintenance costs of photovoltaic / wind turbine / energy storage). Objective 2: Power supply reliability (core is ensuring base station power supply), using the load deficit rate (LPSP) as a quantitative indicator, which is the proportion of electricity that the base station load demand is not met to the total demand. Campus communication base stations have extremely high reliability requirements (usually above 99.99%), and optimization is needed to ensure that the load can still be stably powered when wind and solar power fluctuates or equipment fails. Objective 3: Carbon emission intensity (core is achieving the zero-carbon target), using the total carbon emissions calculated by the carbon flow tracking model or the carbon intensity per unit of electricity as an indicator, to ensure that carbon emissions do not exceed the standard during the optimization process.

[0029] Integration Approach: Instead of prioritizing a single objective, the relationship between the three objectives is coordinated through a weighted approach or hierarchical optimization. For example, the weight of each objective is determined by the entropy weight method (the weight of carbon emission intensity can be appropriately increased in conjunction with the needs of zero-carbon campus construction), transforming multiple objectives into a single objective function, ensuring that the optimization instructions simultaneously meet the requirements of cost control, stable power supply, and carbon emission compliance.

[0030] As an endogenous variable, carbon emission flows mean that they are not independent of economic and reliability objectives, but rather interconnected. For example, when the carbon intensity of the power grid is high, the optimization function will automatically prioritize the use of wind, solar, and energy storage discharge (zero-carbon / low-carbon power sources). Even if short-term operation and maintenance costs increase slightly, the overall goal can be optimized by reducing carbon emissions. Conversely, when the carbon intensity of the power grid is extremely low, the purchase of electricity can be moderately increased to reduce energy storage charging and discharging losses, thus balancing economic and low-carbon objectives.

[0031] Constructing a set of equipment operation constraints is crucial. Solving the optimization model requires basing it on the equipment's physical characteristics and operational limits to avoid generating theoretically feasible but practically unexecutable instructions (such as overcapacity charging of energy storage or sudden power surges in wind turbines exceeding the equipment's capacity). This set of constraints is the core of ensuring the engineering feasibility of the optimization instructions.

[0032] Specific constraints (in the context of a campus microgrid): Energy storage unit constraints: State of charge (SOC) constraint: The SOC of the energy storage must be within a preset range (e.g., 20%-80%) to avoid overcharging and over-discharging, which could damage the battery and extend the equipment's lifespan; Charging and discharging power constraint: The charging and discharging power cannot exceed the rated power of the equipment, and the charging and discharging state cannot switch instantaneously (e.g., from full-power charging to full-power discharging immediately), and the switching buffer time requirement must be met. Power ramping constraint: The output change rate of photovoltaic and wind turbines cannot exceed the maximum ramping rate (e.g., the output change of wind turbines per minute cannot exceed 10% of the rated power) to avoid sudden power changes causing voltage and frequency fluctuations in the microgrid, affecting the stable operation of the base station load. Bus voltage stability constraint: The AC and DC bus voltages of the microgrid must be maintained within the allowable deviation range of the rated voltage (e.g., AC 380V±5%, DC 48V±3%) to ensure the normal operation of base station communication equipment, energy storage converters, and other components, and to avoid equipment failure due to voltage exceeding limits. Other auxiliary constraints include photovoltaic panel operating temperature constraints (to avoid a sudden drop in efficiency at high temperatures) and wind turbine cut-in / cut-out wind speed constraints (no power generation below the cut-in wind speed, and shutdown protection above the cut-out wind speed), to adapt to the installation environment characteristics of wind and solar equipment on campus.

[0033] Constructing a carbon flow constraint set: This transforms the zero-carbon goal from an optimization trend into a rigid boundary, ensuring that the carbon emissions of the microgrid operation are always controlled within a preset range, avoiding exceeding the zero-carbon threshold in pursuit of economy or reliability. Constraint objects: Based on the campus zero-carbon construction plan, either total carbon emissions or carbon emission intensity can be selected as the constraint object: Total carbon emission constraint: Set a daily / monthly upper limit for the total carbon emissions of the microgrid (e.g., no more than 50 kg CO2 per day), suitable for scenarios with clear total carbon emission assessments; Carbon emission intensity constraint: Set an upper limit for carbon emissions per unit of electricity supplied (e.g., no more than 0.03 kg CO2 / kWh), suitable for scenarios pursuing a zero-carbon power supply ratio. Constraint implementation logic: Based on real-time carbon emission data calculated by the carbon flow tracking model, the constraint conditions are embedded into the optimization model. When carbon emissions approach the threshold, the model automatically adjusts the scheduling strategy (e.g., increasing wind and solar power output, reducing power purchases from the main grid, and prioritizing the release of zero-carbon energy from energy storage) to ensure that carbon emissions do not exceed the limit. For example: If the real-time carbon intensity exceeds the constraint threshold, the model will forcibly reduce the power purchased from the main grid, prioritizing the carbon constraint even if it incurs a small amount of energy storage operation and maintenance costs.

[0034] In this embodiment, system state data is input into a pre-trained digital twin proxy model, which outputs predictions of wind and solar power generation and load demand over multiple time scales. Specifically, a system-level digital twin is constructed based on historical microgrid operation data, high-precision weather forecast data, and base station service data to perform high-fidelity mapping of the physical microgrid. Massive scenario simulations are performed within the digital twin to generate a training sample set covering various typical and extreme weather conditions and load fluctuations. A deep reinforcement learning algorithm is used, with the optimization objective of the scheduling model as the reward function, to train the proxy model, enabling it to quickly infer an approximately optimal scheduling strategy based on the input system state data.

[0035] It should be noted that the optimized scheduling model addresses the pain points of traditional microgrid optimization algorithms, such as high dependence on model accuracy, slow online calculation speed, and difficulty in handling the strong uncertainties of wind, solar, and load. Its core logic is to create a proxy model with rapid decision-making capabilities through the construction of a system-level digital twin and deep reinforcement learning training, providing high-precision, low-latency prediction and optimization support for multi-timescale scheduling. The following analysis breaks down its connotation, implementation path, and technical value point by point according to three layers of technical logic: Constructing a system-level digital twin to achieve high-fidelity mapping of the physical microgrid: Traditional wind and solar power and load forecasting models are mostly based on statistical analysis of historical data (such as ARIMA and LSTM), which suffers from incomplete scenario coverage and weak model generalization ability. For example, the prediction error increases sharply under extreme weather conditions, leading to the failure of dispatch commands. This step digitizes the physical characteristics, operating status, and external influencing factors (meteorology, business) of the microgrid by constructing a physical-virtual bidirectional mapping digital twin, providing a high-fidelity virtual test platform for subsequent massive scenario simulation and model training.

[0036] The construction of a digital twin relies on three types of core data to ensure consistency between the virtual image and the physical microgrid: Historical microgrid operation data: including full operational parameters such as wind and solar power generation, energy storage SOC, load power, bus voltage, and equipment start-up / shutdown status, used to calibrate the twin's equipment characteristics and operating patterns; High-precision weather forecast data: integrating numerical weather prediction (NWP) data and local meteorological monitoring data, covering parameters such as light intensity, wind speed, temperature, and humidity, and supporting multiple time scale resolutions (day-ahead hourly, intraday minute, and real-time second), used to simulate wind and solar power output characteristics under different meteorological conditions; Base station service data: including base station user access volume, service type (voice, data, video), and equipment power consumption curves, used to accurately simulate dynamic changes in load power, such as service volume fluctuations during the school opening / summer / winter break and sudden load changes in emergency communication scenarios.

[0037] Massive scenario simulations generate training sample sets covering typical and extreme cases: The performance of deep reinforcement learning models depends entirely on the quantity and diversity of training samples. The actual operating scenarios of physical microgrids are limited, especially low-probability, high-impact scenarios such as extreme weather and equipment failures, making it difficult to accumulate sufficient sample data. This step generates a full-scenario sample set covering normal, extreme steady-state, and transient states through large-scale scenario simulations in a digital twin, ensuring that the surrogate model can stably output high-quality prediction results under any actual operating conditions.

[0038] The generation of the sample set must follow the principles of full coverage and hierarchical classification, specifically including three types of scenarios: Typical operating scenarios: covering the daily operating conditions of the campus microgrid, such as normal weather conditions like sunny, cloudy, and overcast days, and fluctuations in base station traffic on weekdays / weekends; these scenarios account for 70% of the sample set and are used to train the basic predictive capabilities of the model; Extreme weather scenarios: including extreme weather conditions such as heavy rain, typhoons, sandstorms, and continuous overcast and rainy days, as well as sudden increases and decreases in wind and solar power (such as rapid cloud cover blocking photovoltaic panels); these scenarios account for 20% of the sample set and are used to improve the model's anti-interference capabilities; Fault and disturbance scenarios: including equipment failures (abnormal energy storage units, inverter shutdowns), sudden load changes (emergency communication load access), and grid voltage fluctuations; these scenarios account for 10% of the sample set and are used to strengthen the robustness of the model.

[0039] In a digital twin, samples are generated through parameter randomization and combined simulation. For example, the variation range of light intensity and wind speed is randomly adjusted, and the sudden change time and magnitude of base station traffic are randomly set. Finally, sample pairs of input features (historical wind and solar power, SOC, carbon intensity, etc.) and output labels (future wind and solar power, load demand) are generated, forming a training sample set of millions. The sample set generated by the digital twin solves the industry pain point of scarce physical samples and avoids the overfitting problem of traditional models performing well on the training set but having large errors in actual working conditions, laying the foundation for the generalization ability of the surrogate model.

[0040] Deep reinforcement learning training enables the creation of a digital twin agent model with rapid decision-making capabilities: Traditional optimization scheduling models (such as particle swarm optimization and genetic algorithms) require iterative calculations and have long online response times (typically on the order of seconds or even minutes), which cannot meet the scheduling requirements at the real-time scale (on the order of seconds). This step uses deep reinforcement learning training to allow the agent model to learn the optimal scheduling strategy within the digital twin, ultimately achieving rapid decision-making by inputting system state data and outputting prediction and optimization results in milliseconds.

[0041] Training Logic and Key Design: Deep Reinforcement Learning (DRL) is a trial-and-error learning algorithm. Its core is that an agent continuously adjusts its strategy to maximize the reward function by interacting with the environment (digital twin). Compared to the supervised learning model of traditional machine learning, DRL is more suitable for the dynamic optimization scenario of microgrids because the operating state of a microgrid is continuously changing, and the uncertainties of wind, solar, and load cannot be completely covered by fixed labels. DRL can adapt to the dynamically changing environment through continuous learning. This invention preferably uses the Deep Deterministic Policy Gradient (DDPG) algorithm, which is suitable for optimization problems in continuous action spaces and can output continuous wind and solar power predictions and energy storage dispatch commands, meeting the dispatch requirements of microgrids. The design of the reward function is the core of training and directly determines the optimization direction of the agent model. This invention uses a multi-objective optimization function (operational economy, power supply reliability, and carbon emission intensity) as the reward function. The specific design logic is as follows:

[0042] Where: C(t) is the operating cost at time t, This means that the lower the cost, the higher the reward value. LPSP(t) is the load deficit rate at time t, and 1−LPSP(t) indicates that the higher the reliability, the higher the reward value. Let be the carbon emission intensity at time t. The better the low-carbon performance, the higher the reward value. , , This is a weighting coefficient, which can be dynamically adjusted according to the priority of campus zero-carbon construction (e.g., increasing it during the final stage of achieving zero-carbon goals). (weight).

[0043] Training process: The agent inputs a scheduling strategy (such as energy storage charging and discharging power, wind and solar power output commands) into the digital twin; the digital twin simulates the system operation state under the strategy and outputs indicators such as operating cost, power shortage rate, and carbon emission intensity; the agent calculates the reward value according to the reward function and feeds it back to the agent; the agent updates the network parameters through the gradient descent algorithm and adjusts the strategy to pursue higher rewards; the above steps are repeated tens of thousands of times until the reward value no longer increases (the strategy converges), and the trained agent model is obtained.

[0044] The trained digital twin agent model possesses two key capabilities that directly support multi-timescale scheduling: High-precision prediction capability: Inputting real-time system state data, it can output prediction results of wind and solar power generation and load demand for multiple future timescales (24-72 hours before the day, 1-4 hours during the day, and 1-10 minutes in real time), with prediction errors in extreme scenarios reduced by more than 30% compared to traditional models; Low-latency decision-making capability: Without complex iterative calculations, it directly outputs near-optimal scheduling strategies with a response time ≤100ms, meeting the scheduling requirements of real-time scale.

[0045] In this embodiment, based on the prediction results and the optimized scheduling model, the optimized scheduling instruction sequences for photovoltaic power generation units, wind power generation units, and energy storage units at the day-ahead, intraday, and real-time scales are obtained. Specifically: at the day-ahead scale, based on the refined meteorological forecast for the next 24-72 hours, with the goal of minimizing the total daily operating cost and total carbon emissions, the basic charging and discharging plan of the energy storage unit and the day-ahead plan curve of each power generation unit are obtained; at the intraday rolling scale, based on the updated ultra-short-term wind and solar forecasts and load forecasts, with the goal of minimizing adjustment costs and carbon emission fluctuations, the day-ahead plan curves are rolled over and corrected; at the real-time scale, based on the second- to minute-level optimization instructions output by the surrogate model, the output of the energy storage converter and distributed power sources is dynamically adjusted to quickly smooth out power fluctuations and ensure that the real-time carbon intensity meets the constraints.

[0046] It should be noted that day-ahead scheduling covers the next 24-72 hours, with a typical scheduling step size of 1 hour, and belongs to the medium- to long-term energy management scale. Its core purpose is to formulate a globally optimal basic scheduling plan, providing a benchmark framework for subsequent intraday and real-time scheduling, and is a key link in ensuring the economic efficiency and low carbon emissions of the microgrid throughout its entire lifecycle. Core data sources include: refined meteorological forecast data for the next 24-72 hours (hourly forecasts of light intensity, wind speed, and temperature), medium- to long-term forecast data of base station services (such as weekday / weekend service volume patterns and holiday load fluctuation trends), and day-ahead carbon intensity forecast curves for the regional power grid. The optimization objective is to minimize both the total daily operating cost and the total daily carbon emissions. This objective fully accommodates multi-objective optimization functions. Due to the ample decision-making time at the day-ahead scale, computational delays do not need to be considered, and more accurate traditional optimization algorithms (such as genetic algorithms and particle swarm optimization) can be used to solve the problem.

[0047] Scheduling Logic: Combining equipment operation constraints and carbon flow constraints, the day-ahead predicted power of wind and solar power generation and the day-ahead predicted demand of base station load are substituted into the optimization model to solve for the basic charging and discharging plan of the energy storage unit (defining the charging and discharging period and the upper limit of charging and discharging power), as well as the day-ahead planned output curves of photovoltaic and wind power generation units (defining the optimal output value for each period). For example, when the day-ahead prediction indicates sufficient sunlight at noon the next day (peak photovoltaic output) and high grid carbon intensity, the model will formulate a plan for full photovoltaic power generation + energy storage charging; when the prediction indicates no wind or sunlight at night and low grid carbon intensity, a plan for energy storage discharging + moderate electricity purchase is formulated to achieve a global balance between cost and carbon emissions. Output Results: Baseline scheduling instructions are issued to the controllers of each device in the microgrid as a reference for intraday and real-time scheduling. Day-ahead-scale scheduling avoids short-sighted decisions (such as excessive electricity purchase to reduce immediate costs, leading to excessive carbon emissions throughout the day) through long-term global optimization, ensuring that the microgrid operation meets the medium- and long-term planning requirements of a zero-carbon campus.

[0048] Intraday rolling-scale scheduling: The time range is 1-4 hours ahead, with a rolling step size typically set at 15-30 minutes, belonging to the medium-to-short-term error correction scale. Its core positioning is to make rolling corrections to the day-ahead plan based on ultra-short-term forecast data, offsetting the deviation of the day-ahead forecast and reducing the risk of scheduling failure due to sudden changes in wind, solar, and load. Core data basis: Updated ultra-short-term wind and solar forecast data (forecast duration 1-4 hours, step size 5-10 minutes, forecast accuracy significantly higher than day-ahead forecasts), ultra-short-term real-time monitoring data of base station load (such as sudden emergency communication service load), and real-time carbon intensity data. Optimization objectives: The core objectives are to minimize the cost of plan adjustment and the fluctuation of carbon emissions. Adjustment cost refers to the additional cost incurred by modifying the day-ahead plan (such as the life loss of energy storage due to early discharge, and the premium of temporary transactions with the power grid); carbon emission fluctuation refers to the deviation between actual carbon emissions and the carbon emissions of the day-ahead plan, avoiding the runaway situation of low carbon emissions at the beginning and high carbon emissions at the end. Scheduling logic: A rolling optimization strategy is adopted, and the day-ahead plan curve for the next 1-4 hours is corrected every 15-30 minutes based on the latest ultra-short-term forecast data. The digital twin agent model is prioritized here to replace the traditional optimization algorithm because intraday scheduling requires a rapid response (correction instructions need to be generated within minutes), and the agent model, after being trained on a massive number of samples, can directly output an approximately optimal correction scheme without the need for complex iterative calculations.

[0049] For example, if the daily plan predicts stable photovoltaic (PV) output at midday, but a short-term intraday forecast indicates cloud cover in one hour (which will reduce PV output by 30%), the proxy model will immediately output a correction instruction: reduce energy storage charging power 10 minutes in advance to reserve capacity to cope with the subsequent drop in PV output, avoiding power shortages at base stations or excessive carbon emissions due to a sudden drop in PV power. Output: A continuously updated intraday correction scheduling curve covering equipment output instructions for the next 1-4 hours, replacing the corresponding time period content of the original daily plan in real time.

[0050] Real-time scheduling: With a time range from seconds to minutes and a scheduling step size as low as 1 second, this scale focuses on instantaneous power balance and stability control. Its core function is to address high-frequency fluctuations in wind and solar power and load, rapidly adjusting equipment output to ensure microgrid voltage and frequency stability while meeting real-time carbon intensity constraints. It serves as the last line of defense for ensuring high-reliability power supply to base station loads. Key data sources: Real-time system status data collected at the second level (instantaneous output of photovoltaic / wind power, second-level changes in energy storage SOC, instantaneous power of base station loads, bus voltage / frequency), and second-level optimization instructions output by the digital twin proxy model. Optimization objectives: The core objectives are to rapidly smooth power fluctuations, maintain system stability, and meet real-time carbon intensity constraints. Optimization at this scale no longer pursues global cost optimization but prioritizes ensuring power supply reliability and carbon intensity compliance.

[0051] Scheduling Logic: Real-time scheduling deals with high-frequency, small-amplitude power disturbances (such as a 20% drop in photovoltaic power within 10 seconds due to a rapid cloud cover, or a 15% surge in load power due to sudden data traffic from a base station). These disturbances cannot be predicted in advance through day-ahead or intraday scheduling and must rely on the second-level decision-making capabilities of the proxy model. Based on real-time collected system status data, the proxy model directly outputs instantaneous adjustment commands for energy storage converters and distributed power sources: when wind and solar power output drops sharply or load surges, the energy storage converter is instructed to immediately increase its discharge power to mitigate the power deficit; when wind and solar power output surges or load drops sharply, the energy storage converter is instructed to immediately increase its charging power to absorb excess electricity and avoid power curtailment; simultaneously, the carbon intensity of the microgrid is monitored in real time. If the carbon intensity approaches the threshold due to power fluctuations, the power purchased from the main grid is reduced, and zero-carbon energy from energy storage is prioritized to ensure carbon intensity compliance. Furthermore, this scale of scheduling incorporates a power fluctuation spectrum hierarchical management strategy, allocating high-frequency fluctuations to power-type energy storage and low-frequency fluctuations to energy-type energy storage, achieving efficient utilization of energy storage resources. Output: Real-time control commands are issued to each device controller within seconds, directly adjusting the output values ​​of photovoltaic inverters, wind power converters, and energy storage converters, ensuring the system restores power balance within milliseconds. Real-time scheduling addresses the pain point of traditional microgrids' delayed response to high-frequency disturbances, improving power supply reliability to over 99.99% as required by communication base stations, while simultaneously ensuring the real-time achievement of zero-carbon goals.

[0052] In this embodiment, the method further includes: real-time monitoring of the microgrid topology and load demand changes; when a significant change in the load of a communication base station is detected due to service migration or equipment start-up and shutdown, or when some lines of the microgrid need maintenance, the network topology of the microgrid is dynamically reconstructed; after topology reconstruction, distributed collaborative control of multiple parallel-operating energy storage converters is performed based on a consensus algorithm to achieve autonomous power distribution and synchronization, and maintain system stability.

[0053] It should be noted that traditional microgrids use a fixed topology, which can easily lead to localized power outages (e.g., maintenance on a feeder can cause a power outage for base stations in that area) due to sudden changes in communication base station load or line maintenance. Furthermore, they cannot optimize power flow distribution to reduce carbon emissions and operating costs. This step, through real-time status monitoring and dynamic topology reconfiguration, upgrades the microgrid from a rigid network to a flexible network, adapting to the dynamic changes in load and line status in a campus setting.

[0054] Topology reconfiguration is triggered when real-time monitoring of operational condition changes reaches a threshold. The core monitoring involves two dimensions of status data: **Communication base station load changes:** When a base station experiences a sudden increase in load power exceeding a preset threshold (typically 20%-30%) due to service migration (e.g., a large-scale campus event causing a surge in base station traffic) or equipment startup / shutdown (e.g., a backup base station being put into operation), topology reconfiguration is triggered. **Microgrid line status changes:** When some microgrid lines need to be taken out of service due to faults or scheduled maintenance, topology reconfiguration is triggered. This status data is collected in real-time by the load power acquisition unit and line fault monitoring terminal in the microgrid sensing layer, with a collection frequency on the order of seconds, ensuring rapid perception of operational condition changes.

[0055] Topology reconfiguration is not a blind adjustment of network connections, but rather an intelligent networking based on load importance levels and power supply zoning planning. Core actions include: Load priority identification: automatically distinguishing between core loads of campus communication base stations (such as main base stations in the core area of ​​the campus, requiring 99.99% power reliability) and flexible loads (such as backup base stations in edge areas, allowing short-term power reduction); Network connection mode switching: issuing commands through the software-defined networking (SDN) controller to drive the solid-state switch array to adjust the connection paths between power supply zones. For example, switching core loads covered by faulty lines to adjacent healthy feeders; networking zones with sudden load increases with zones with sufficient wind and solar power output to achieve energy mutual support; Power flow re-optimization: after topology reconfiguration, combining a carbon flow tracing model, recalculating the carbon intensity and power flow direction of each node to ensure that the power distribution under the new topology still meets carbon emission intensity constraints and operational economic targets. Campus microgrids are characterized by scattered base station distribution and variable line conditions (e.g., campus infrastructure construction easily leads to temporary line maintenance, and base station traffic fluctuates significantly during the start of the school year / summer and winter / summer vacations). Dynamic topology reconfiguration avoids the cascading effects of a single change in a traditional fixed topology, ensuring uninterrupted power supply to core base stations while maximizing the utilization of zero-carbon wind and solar power within the campus.

[0056] Topology reconfiguration directly alters the parallel operation topology of energy storage converters (e.g., some energy storage units change from parallel networking to independent power supply, or new energy storage units are added to the network). If a traditional centralized control strategy is used, problems such as power distribution imbalance and bus voltage fluctuations can easily occur, potentially leading to unstable power supply to base station loads. This step utilizes a consensus algorithm to achieve distributed collaborative control of the energy storage converters. It eliminates the need for centralized commands from a central controller, achieving autonomous power distribution and synchronization through local information exchange, thus ensuring system stability after topology reconfiguration.

[0057] Consensus algorithms are distributed intelligent control algorithms. Their core principle is to enable multiple parallel-operating energy storage converters to achieve global consistency in output power and bus voltage through local communication and state synchronization. Considering the distributed deployment characteristics of multiple energy storage units in a campus microgrid, the specific execution logic is as follows: Information interaction layer design: A peer-to-peer communication network (such as power line carrier communication or 5G local communication) is established between each energy storage converter. This eliminates the need for a central controller, allowing only the exchange of their own operating status data (output power, bus voltage, SOC); Consensus protocol formulation: A consensus objective of power equalization is set, meaning the output power difference between any two energy storage converters must converge to 0. The mathematical expression is: .

[0058] Where N is the set of parallel energy storage converters. Let be the real-time output power of the i-th energy storage converter.

[0059] When topology reconfiguration causes the power output of a certain energy storage converter to deviate from the target value, that converter will autonomously adjust its own converter control parameters (such as modulation ratio and switching frequency) based on the state data of adjacent converters until the power output of all parallel converters tends to be consistent, while maintaining the bus voltage stable within the rated range. The application of consensus algorithms enables the microgrid to achieve decentralized and stable operation after topology reconfiguration, which not only improves the fault tolerance of the system (avoiding the risk of central controller failure) but also accelerates the response speed of power balancing, ensuring that the communication base station load will not experience voltage fluctuations or power outages due to topology adjustments.

[0060] In this embodiment, the microgrid's topology and load demand changes are monitored in real time. When a significant change in the communication base station load is detected due to service migration or equipment start-up and shutdown, or when some lines of the microgrid need maintenance, the network topology of the microgrid is dynamically reconstructed. Specifically, based on the distribution and importance level of the communication base station load, the microgrid is logically divided into multiple power supply zones, including core load areas and flexible load areas. Through software-defined networking and solid-state switches, the network connection mode is automatically switched in response to load changes or faults, realizing flexible networking and energy sharing between power supply zones.

[0061] It should be noted that traditional microgrid topology adjustments are often extensive whole-network reconfigurations, lacking differentiation of load priorities, which can easily lead to the risk of simultaneous power outages for core base stations and ordinary base stations. This step, through logical partitioning, clarifies the power supply priorities and operating rules of different areas, making topology reconfiguration based on evidence and ensuring that the core services of campus communication base stations are not interrupted.

[0062] The zoning plan needs to consider two dimensions simultaneously to ensure that it aligns with the actual operational needs of the campus microgrid: Dimension 1, Distribution of base station load: According to the geographical layout of the campus, the microgrid is divided into several physically adjacent power supply areas (such as teaching area, dormitory area, playground area, and administrative office area). Each area is equipped with independent feeders and switching equipment to achieve regional management and avoid losses and safety risks caused by large-scale power transmission across regions.

[0063] Dimension 2, Importance Level of Base Station Load: Based on the priority of campus communication services, each zone is further divided into core load zones and flexible load zones. The core differences between the two types of zones are shown in Table 1: Table 1

[0064] Provides priority criteria for topology reconfiguration: During the reconfiguration process, priority is given to ensuring power supply to the core load area, while the flexible load area can participate in system optimization as an adjustable resource; Reduces reconfiguration complexity: After partitioning, topology adjustments only need to be made between adjacent partitions, without the need for replanning the entire network, thus improving reconfiguration efficiency; Optimizes carbon flow distribution: Zero-carbon wind and solar power can be directionally transmitted to the core load area, reducing the carbon emission intensity of core businesses and contributing to the construction of zero-carbon campuses.

[0065] Core Objective: Traditional microgrid topology switching relies on mechanical switches, which suffers from slow switching speeds (on the order of seconds), contact wear, and susceptibility to arcing, failing to meet the millisecond-level uninterrupted power supply requirements of communication base stations. This step utilizes a combination of Software-Defined Networking (SDN) and solid-state switches to achieve fast, uninterrupted, and programmable topology switching, while also supporting inter-grid energy sharing.

[0066] The core of this step is software-defined topology and hardware-executed switching, specifically divided into three layers of execution logic: Perception layer: Load power acquisition devices and line fault monitoring terminals distributed in each partition collect data in real time. When a load change exceeds the threshold or a line fault / maintenance is detected, a trigger signal is sent to the SDN controller.

[0067] Control Layer: The SDN controller is the brain of topology reconfiguration. Its core function is to automatically generate the optimal networking scheme based on partitioning rules and system status. If a feeder in a core load area fails, the controller will automatically search for backup feeder resources in adjacent partitions and formulate a switching scheme from the core load area to the feeder of the adjacent healthy core / flexible load area. If the wind and solar power output of a flexible load area is excessive, the controller will adjust the topology to deliver the excess power to the partition with the load gap, realizing energy mutual assistance between partitions. When generating the networking scheme, the power flow and carbon intensity under the new topology must be verified simultaneously to ensure that the equipment operation constraints and carbon flow constraints are met.

[0068] Execution Layer: Solid-state switches are the actuators for topology reconfiguration. Compared to traditional mechanical switches, they offer advantages such as millisecond-level switching speed, no contacts, and no electric arc, enabling uninterrupted power supply switching for base station loads. The execution process is as follows: The SDN controller issues a switching action command → After receiving the command, the solid-state switch array disconnects the faulty / redundant feeder and closes the backup feeder → The new topology takes effect, and new power transmission paths are formed between sections.

[0069] Rapid switching ensures power continuity: The millisecond-level switching speed is far below the power interruption tolerance threshold of communication base stations (usually 20ms), ensuring uninterrupted base station services; Software-controlled, flexibly adapts to changing operating conditions: No manual on-site operation is required. The SDN controller can automatically adjust the topology according to the system status to adapt to load fluctuations during the start of the school season / winter and summer vacations; Contactless operation reduces maintenance costs: Solid-state switches have no mechanical wear and their service life is 5-10 times that of traditional mechanical switches, meeting the operational needs of campus microgrids with minimal human intervention.

[0070] In this embodiment, the optimized scheduling command sequences of the photovoltaic power generation unit, wind power generation unit, and energy storage unit at the day-ahead, intraday, and real-time scales are obtained by solving the problem. The following steps are also performed: based on the power fluctuation spectrum characteristics in the prediction results, the high-frequency and low-energy power fluctuation components are allocated to power-type electrochemical energy storage for rapid response; the low-frequency and high-energy energy transfer requirements are allocated to energy-type energy storage.

[0071] It should be noted that during microgrid operation, the fluctuations in wind and solar power generation and base station load power are not singular but rather a superposition of fluctuation components of different frequencies and energy levels. Traditional dispatch strategies do not distinguish between these fluctuation components, often resulting in a mismatch where energy-type energy storage handles high-frequency fluctuations while power-type energy storage processes low-frequency energy demands. The former suffers from accelerated battery degradation due to high-frequency charge-discharge cycles, while the latter leads to resource waste due to prolonged idleness. The core objective of this step is to decompose the mixed fluctuation signal into clearly defined components through spectrum analysis, providing a basis for the precise allocation of energy storage units.

[0072] The real-time power fluctuation signal of the microgrid (i.e., the combined difference signal of wind and solar power output fluctuation + load demand fluctuation) is decomposed into components of different frequencies using Fast Fourier Transform (FFT) or wavelet transform, and classified into two categories according to frequency-energy characteristics, as shown in Table 2 below: Table 2

[0073] The essence of spectrum feature analysis is to accurately profile power fluctuations, enabling the dispatch system to identify small disturbances that require rapid response and large fluctuations that require long-term energy support, thus preventing energy storage units from undertaking tasks beyond their own characteristics and providing a quantitative basis for subsequent division of labor.

[0074] Microgrids are equipped with two types of energy storage units: power-type electrochemical energy storage and energy-type energy storage. The core characteristics of these two types differ significantly, which determines the types of fluctuations they are suited for, as shown in Table 3. Table 3

[0075] The core principle is characteristic matching and division of labor, precisely allocating fluctuation components of different frequencies to the corresponding energy storage units. The specific execution strategy is as follows: Response Timing: Primarily plays a role in real-time scheduling, responding to instantaneous power disturbances ranging from seconds to minutes. Execution Actions: When high-frequency fluctuations are detected, power-type energy storage charges and discharges with millisecond-level response speeds. During sudden increases in wind and solar power output, it rapidly charges to absorb excess power; during sudden decreases in wind and solar power output / sudden increases in load, it rapidly discharges to fill the power gap, achieving instantaneous power balance. Advantages: Utilizing its fast response and long lifespan, it avoids energy-type energy storage participating in high-frequency cycles, effectively extending the lifespan of energy-type energy storage.

[0076] Response Timing: Covering day-ahead, intraday, and real-time full-scale dispatching, with a focus on long-cycle energy regulation tasks. Execution Actions: On a day-ahead scale, based on long-term forecasts of wind and solar power output, it stores zero-carbon electricity during off-peak hours in advance; on an intraday scale, it corrects day-ahead planning deviations and performs hourly energy transfers; on a real-time scale, it accommodates low-frequency fluctuations in energy demand, providing stable energy support for power-type energy storage. Advantages: Utilizing its high energy density, it meets the regulation needs of large-scale energy operations, avoiding the inability of power-type energy storage to cope with long-term fluctuations due to insufficient capacity.

[0077] The task allocation strategy is not executed independently, but is deeply embedded in a three-level scheduling framework to achieve precise matching of scale, fluctuation, and energy storage: Day-ahead scale: Based on long-term forecasts of wind, solar and load, low-frequency high-energy fluctuation trends are identified, and charging and discharging plans for energy storage are formulated in advance; Intra-day scale: Based on ultra-short-term forecast corrections, the energy transfer rhythm of energy storage is adjusted, while reserving response margins for power storage; Real-time scale: For instantaneous high-frequency fluctuations, millisecond-level responses of power storage are triggered to smooth disturbances and maintain system stability.

[0078] Achieving refined management and control of energy storage resources: Breaking away from the traditional one-size-fits-all energy storage scheduling model, this approach enables on-demand allocation based on spectrum characteristics, allowing different types of energy storage units to fully utilize their characteristics and avoiding resource mismatch. Extending the overall lifespan of energy storage systems: The core value lies in reducing high-frequency charge-discharge cycles of energy-type energy storage, a major cause of capacity decay in energy storage units such as lithium batteries. This strategy can extend the lifespan of energy-type energy storage by 30%–50%, reducing microgrid operation and maintenance costs. Improving the response accuracy of real-time scheduling: The millisecond-level response characteristics of power-type energy storage can quickly smooth out high-frequency disturbances, controlling the fluctuation rate of microgrid power fluctuations within ±2%, meeting the stringent power stability requirements of communication base stations. Enhancing the execution efficiency of multi-timescale scheduling: Through precise division of labor among energy storage units, day-ahead / intra-day energy planning and real-time power smoothing operate independently, ensuring the efficient implementation of full-scale scheduling goals.

[0079] Example 2, Figure 2 The present invention provides a multi-energy complementary zero-carbon campus AC / DC microgrid configuration system, including a data acquisition module, a model building module, a prediction and decision-making module, a coordination control module, and a topology management module. The data acquisition module is used to collect system status data of the microgrid; The model building module is used to build and update a multi-timescale optimization scheduling model for base station microgrids that integrates dynamic carbon sensing. The prediction and decision-making module has an embedded pre-trained digital twin agent model, which is used to receive system status data and make predictions and fast optimization decisions to generate optimized scheduling instruction sequences. The coordination and control module is used to execute the optimized scheduling command sequence and send control commands to the photovoltaic inverters, wind power converters, energy storage converters and network switches in the microgrid; The topology management module is used to monitor system status and dynamically manage the network topology of the microgrid.

[0080] In this embodiment, the system also includes an energy management platform deployed in the cloud, which is communicatively connected to multiple microgrid systems; The energy management platform aggregates the regulation capabilities of multiple microgrids and, based on blockchain technology, enables point-to-point green electricity trading and carbon emission reduction verification among multiple microgrids.

[0081] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0082] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, in the form of a computer program product.

[0083] Those skilled in the art will recognize that the modules and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0084] In addition, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0085] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0086] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for configuring a zero-carbon AC / DC microgrid on a campus with multiple complementary energy sources, characterized in that: This includes photovoltaic power generation units, wind power generation units, energy storage units, and communication base station loads, specifically comprising the following steps: Collect system status data of the microgrid, including real-time wind and solar power generation, energy storage unit state of charge, base station load power demand, and real-time carbon intensity information. Based on the system status data, a multi-timescale optimization scheduling model for base station microgrids integrating dynamic carbon sensing is constructed. The optimization scheduling model takes operational economy, power supply reliability and carbon emission intensity as the synergistic optimization objectives, and includes equipment operation constraints and carbon flow constraints. The system state data is input into a pre-trained digital twin agent model, which outputs prediction results of wind and solar power generation and load demand at multiple time scales in the future. Based on the prediction results and the optimized scheduling model, the optimized scheduling instruction sequences for the photovoltaic power generation unit, wind power generation unit, and energy storage unit at the day-ahead, intraday, and real-time scales are obtained. Based on the optimized scheduling instruction sequence, the microgrid's units are controlled in real time by a coordination controller to achieve low-carbon and economical operation.

2. The method for configuring a multi-energy complementary zero-carbon AC / DC microgrid on a campus according to claim 1, characterized in that, Based on the system state data, a multi-timescale optimal scheduling model for base station microgrids integrating dynamic carbon sensing is constructed. This optimal scheduling model takes operational economy, power supply reliability, and carbon emission intensity as synergistic optimization objectives, and includes equipment operation constraints and carbon flow constraints, specifically: A carbon flow tracking model for the microgrid is established, and the carbon emission flow within the microgrid and each power supply link is quantified based on the real-time carbon intensity information and power flow calculation. The carbon emission flow is treated as an endogenous variable and, together with system operating cost and load power shortage rate, constitutes a multi-objective optimization function. Based on the physical characteristics and operational constraints of the equipment in the microgrid, a set of equipment operation constraints is constructed, including energy storage charging and discharging status, power ramping, and bus voltage stability. Based on the carbon flow tracking model, a carbon flow constraint set is constructed, which is used to control the total carbon emissions or carbon emission intensity of the microgrid operation within a preset threshold.

3. The method for configuring a multi-energy complementary zero-carbon campus AC / DC microgrid according to claim 2, characterized in that, The system state data is input into a pre-trained digital twin agent model, which outputs predictions of future wind and solar power generation and load demand across multiple time scales, specifically: Based on the historical operation data of the microgrid, high-precision weather forecast data, and base station service data, a system-level digital twin is constructed to perform high-fidelity mapping of the physical microgrid; Massive scenario simulations are performed in the digital twin to generate training sample sets covering a variety of typical and extreme weather conditions and load fluctuations; A deep reinforcement learning algorithm is used to train the agent model with the optimization objective of the optimized scheduling model as the reward function, enabling it to quickly infer an approximately optimal scheduling strategy based on the input system state data.

4. The method for configuring a multi-energy complementary zero-carbon campus AC / DC microgrid according to claim 3, characterized in that, Based on the prediction results and the optimized scheduling model, the optimized scheduling instruction sequences for the photovoltaic power generation unit, wind power generation unit, and energy storage unit at the day-ahead, intraday, and real-time scales are obtained, specifically as follows: At the day-ahead scale, based on refined weather forecasts for the next 24-72 hours, with the goal of minimizing total daily operating costs and total carbon emissions, the basic charge and discharge plans of the energy storage unit and the day-ahead plan curves of each power generation unit are obtained. On an intraday rolling scale, based on updated ultra-short-term wind and solar load forecasts, the day-ahead planning curve is rolled over and corrected with the goal of minimizing adjustment costs and carbon emission fluctuations. At the real-time scale, based on the second- to minute-level optimization instructions output by the proxy model, the output of the energy storage converter and distributed power source is dynamically adjusted to quickly smooth out power fluctuations and ensure that the real-time carbon intensity meets the constraints.

5. The method for configuring a multi-energy complementary zero-carbon campus AC / DC microgrid according to claim 4, characterized in that, The method further includes: The topology and load demand of the microgrid are monitored in real time. When a significant change in the load of the communication base station is detected due to service migration or equipment start-up and shutdown, or when some lines of the microgrid need maintenance, the network topology of the microgrid is dynamically reconstructed. After topology reconfiguration, distributed collaborative control of multiple parallel-operating energy storage converters is performed based on a consensus algorithm to achieve autonomous power sharing and synchronization, thereby maintaining system stability.

6. The method for configuring a multi-energy complementary zero-carbon campus AC / DC microgrid according to claim 5, characterized in that, The microgrid's topology and load demand changes are monitored in real time. When a significant change in the communication base station load is detected due to service migration or equipment startup / shutdown, or when some lines in the microgrid require maintenance, the network topology of the microgrid is dynamically reconstructed. Specifically: Based on the distribution and importance level of the communication base station load, the microgrid is logically divided into multiple power supply zones, including a core load zone and a flexible load zone; By using software-defined networking and solid-state switches, the network connection mode can be automatically switched in response to load changes or faults, enabling flexible networking and energy sharing between the power supply zones.

7. The method for configuring a multi-energy complementary zero-carbon campus AC / DC microgrid according to claim 6, characterized in that, After obtaining the optimized scheduling command sequences for the photovoltaic power generation unit, wind power generation unit, and energy storage unit at the day-ahead, intraday, and real-time scales, the following steps are also performed: Based on the power fluctuation spectrum characteristics in the prediction results, high-frequency, low-energy power fluctuation components are allocated to the power-type electrochemical energy storage for rapid response; The energy transfer needs of low frequency and high energy are allocated to the energy storage.

8. A photovoltaic-wind power complementary microgrid operation optimization system for low-carbon base stations, applied to the multi-energy complementary zero-carbon campus AC / DC microgrid configuration method as described in any one of claims 1-7, characterized in that, The system includes: The data acquisition module is used to collect system status data of the microgrid; The model building module is used to build and update the multi-time-scale optimization scheduling model of the base station microgrid that integrates dynamic carbon sensing. The prediction and decision-making module, which is embedded with the pre-trained digital twin agent model, is used to receive the system state data and make predictions and rapid optimization decisions to generate an optimized scheduling instruction sequence. The coordination and control module is used to execute the optimized scheduling instruction sequence and send control instructions to the photovoltaic inverters, wind power converters, energy storage converters and network switches in the microgrid; The topology management module is used to monitor the system status and dynamically manage the network topology of the microgrid.

9. The photovoltaic-wind power complementary microgrid operation optimization system for low-carbon base stations according to claim 8, characterized in that, The system also includes an energy management platform deployed in the cloud, which is communicatively connected to multiple microgrid systems; The energy management platform is used to aggregate the regulation capabilities of multiple microgrids and, based on blockchain technology, to realize point-to-point green electricity trading and carbon emission reduction verification among multiple microgrids.

Images