A microgrid energy economic dispatch system and method
By constructing a dynamic uncertainty set through a spatiotemporal hierarchical physical information neural network and a distributed collaborative algorithm, the multi-scale fluctuation characteristics of high-entropy energy are quantified, solving the problem of irregular fluctuations of high-entropy energy in microgrid scheduling, realizing the linkage and complementarity of power and computing resources, and improving the stability and economy of microgrids.
Patent Information
- Application Number
- CN202611491543.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-09-23
- Publication Date
- 2026-11-17
AI Technical Summary
Existing microgrid dispatching schemes cannot adapt to the irregular, transient, and non-stationary fluctuation characteristics of high-entropy energy, resulting in large output prediction deviations, high energy curtailment rates, resource waste, and dispatching failures. Furthermore, traditional dispatching modes fail to effectively utilize computing resources and cannot achieve the linkage and complementarity between power and computing resources.
A spatiotemporal hierarchical physical information neural network is used to model the multi-scale fluctuation characteristics of high-entropy energy, construct a dynamic uncertainty set, and realize dual-domain entropy balance scheduling of power and computing power through distributed cooperative algorithms and adaptive entropy reinforcement learning. Combined with hybrid energy storage and equipment hierarchical control, a robust correction mechanism is triggered to ensure the stable and economical operation of the system.
It achieves accurate quantification of the multi-scale fluctuation characteristics of high-entropy energy, reduces the curtailment rate of new energy, improves the energy utilization efficiency of microgrids, significantly reduces system operating costs, and improves scheduling efficiency and the service life of energy storage equipment.
Smart Images

Figure CN123121850A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of energy economic dispatch technology, and in particular to a microgrid energy economic dispatch system and method. Background Technology
[0002] Microgrid economic dispatch is a core technology for ensuring the safe, stable, and economically efficient operation of distributed renewable energy systems. Current mainstream microgrid dispatch schemes mostly employ deterministic output prediction combined with traditional linear optimization algorithms, focusing only on single-domain power dispatch for stable and controllable renewable energy sources such as wind and solar power.
[0003] Existing technologies have several inherent limitations. First, conventional prediction models are only applicable to stationary time-series data and cannot adapt to the irregular, transient, and non-stationary fluctuations of high-entropy energy sources such as traffic flow energy and turbulent energy. This results in significant output prediction errors, leading to a disconnect between day-ahead scheduling plans and actual operating conditions, high system curtailment rates, and poor scheduling economics. Second, traditional scheduling models focus solely on optimizing and controlling power resources, treating edge computing tasks as fixed loads. They fail to utilize the flexible adjustment capabilities of deferred computing tasks, thus failing to achieve the synergistic complementarity between computing power and power resources. This results in the waste of redundant renewable energy and the idleness of computing resources.
[0004] Meanwhile, existing robust scheduling methods for microgrids often use fixed uncertain sets to describe the power output fluctuations of new energy sources, which cannot adaptively adapt to the multi-scale dynamic fluctuation characteristics of high-entropy energy sources. Under extreme operating conditions, scheduling failures and power instability are prone to occur. Multi-microgrid cluster scheduling often uses centralized solutions or general distributed algorithms, which have poor adaptability to ring microgrid topologies, high communication overhead, and slow solution convergence speed. They are difficult to balance day-ahead global scheduling accuracy and intraday real-time scheduling response speed, and cannot meet the stable and economical operation requirements of microgrid clusters after large-scale access of high-entropy energy sources. Summary of the Invention
[0005] To address the aforementioned technical problems, this invention provides a microgrid energy economic dispatch system and method. The technical solution adopted is as follows: A microgrid energy economic dispatch method includes the following steps: Step 1: Collect microgrid power operation data, high-entropy energy entropy flow data, edge computing task data, and grid electricity price data, and extract power fluctuation entropy and computing task distribution entropy features; Step 2: A spatiotemporal hierarchical physical information neural network is used to model the multi-scale fluctuation characteristics of high-entropy energy and adaptively generate dynamic output prediction intervals and dynamic uncertainty sets. Step 3: Construct a hierarchical Bruker optimization model based on the dynamic uncertainty set, and solve it using a distributed cooperative algorithm to obtain the microgrid group day-ahead power mutual assistance plan and energy storage operation baseline; Step 4: With the dual-domain entropy balance of power and computing power as the core objective, a Markov decision model is constructed, and intraday collaborative real-time scheduling of computing power and power is achieved through adaptive entropy reinforcement learning. Step 5: Issue scheduling instructions to complete the hierarchical control of equipment, and trigger a robust correction mechanism when the output of high-entropy energy exceeds the limit to ensure the stable and economical operation of the system.
[0006] Optionally, the spatiotemporal hierarchical physical information neural network modeling method in step 2 includes the following sub-steps: Step 21, Spatial Stratification: Divide the high-entropy energy capture region into a near-field turbulent region and a far-field stable region, and match the turbulent energy loss equation and the steady-state energy conversion equation as physical constraints respectively; Step 22, Time Stratification: Distinguish between two types of time series features: second-level transient fluctuations and minute-level steady-state offsets, and set differentiated time series weighting coefficients; Step 23, Model Training: Using quantile regression error and hierarchical physical residual as the joint loss function, adaptively adjust the model parameters and output the dynamic output prediction interval and fluctuation entropy value with confidence.
[0007] Optionally, the hierarchical and sub-bar optimization model construction method in step 3 is as follows: the goal is to minimize the sum of the microgrid group's electricity purchase cost, energy storage depreciation cost, and high-entropy energy curtailment cost; a dynamic Wasserstein probabilistic fuzzy set constraint is introduced, and the radius of the fuzzy set is adaptively adjusted based on the fluctuation entropy value in step 2 to replace the fixed uncertain set; a high-entropy energy output opportunity constraint is set to ensure that the system power supply reliability meets the preset threshold under extreme fluctuation conditions.
[0008] Optionally, in step 3, the distributed collaborative solution adopts the lightweight consistency alternating direction multiplier method, which is as follows: the global optimization model is decomposed into independent sub-problems of each micronet, and each micronet only interacts with the boundary power and Lagrange multiplier local information with the adjacent interconnected micronets; the global variables are updated iteratively through weighted consistency of adjacent nodes until the boundary power deviation and dual residual are less than the preset convergence threshold, without the need for global centralized calculation, thus reducing communication and computation overhead.
[0009] Optionally, in step 4, the state space of the Markov decision model includes the real-time output fluctuation entropy of high-entropy energy, the state of charge of hybrid energy storage, the queuing entropy of computing tasks, the real-time electricity price and output prediction deviation; the action space includes the steady-state charging and discharging power of lithium batteries, the transient compensation power of supercapacitors, the fine-tuning power of microgrid interconnects, the start and stop time slots of edge computing tasks and the adjustment amount of node operating frequency.
[0010] Optionally, in step 4, the dual-domain entropy balance optimization objective is the weighted minimum value of system operating cost, power fluctuation entropy, and computing power task congestion entropy; by adaptively adjusting the reinforcement learning temperature parameter, the model's exploration and utilization capabilities are dynamically balanced, the model's exploration degree is improved when high-entropy energy fluctuations are severe, and the model actively adapts to sudden changes in operating conditions, thereby achieving dynamic coordination between power consumption and computing power scheduling.
[0011] Optionally, the equipment hierarchical control method in step 5 is as follows: the supercapacitor responds independently to the high-frequency power fluctuations of high-entropy energy at the second level, suppressing instantaneous power disturbances; the lithium battery undertakes the steady-state power regulation at the minute level, balancing the energy supply and demand of the system; and the high-entropy energy capture device adjusts its operating parameters according to the feedforward of the prediction interval, improving the efficiency of disordered energy capture.
[0012] Optionally, the robust correction mechanism in step 5 is triggered when the real-time output of the high-entropy energy exceeds the boundary of the dynamic prediction interval; the execution strategy after triggering is as follows: Prioritize shielding low-priority, latency-tolerant computing tasks on edge computing nodes to reduce unnecessary power consumption, while controlling the instantaneous output of supercapacitors to compensate for power gaps and ensure uninterrupted power supply to critical loads in the microgrid.
[0013] Optionally, both daytime robust scheduling and intraday real-time scheduling employ differentiated constraints on hybrid energy storage: high-frequency power response constraints and short-term charge-discharge lifetime constraints are set for supercapacitors to quantify transient fluctuation adjustment losses; energy throughput constraints and long-term cycle lifetime constraints are set for lithium batteries to quantify steady-state energy dispatch losses. The cost of both types of energy storage losses is included in the optimization of the total system operating cost.
[0014] A microgrid energy economic dispatch system is used to implement a microgrid energy economic dispatch method. The system is characterized by comprising: a data acquisition module, an entropy feature extraction module, a high-entropy energy hierarchical modeling and prediction module, a day-ahead distributed robust dispatch module, a dual-domain entropy balance real-time dispatch module, and an equipment collaborative control module. The data acquisition module is used to collect heterogeneous data on microgrid power, high-entropy energy, edge computing power, and electricity prices. The entropy feature extraction module is connected to the data acquisition module and is used to extract the core features of power fluctuation entropy and computing power queuing entropy. The high-entropy energy stratification modeling and prediction module embeds a spatiotemporal stratified physical information neural network, which adaptively outputs the dynamic output prediction interval and dynamic uncertainty set of high-entropy energy. The distributed robust scheduling module is used to construct and solve the hierarchical robust optimization model to generate the day-ahead basic scheduling plan for the microgrid cluster. The dual-domain entropy balance real-time scheduling module deploys an adaptive entropy reinforcement learning agent to achieve optimal decision-making between computing power and electricity within the day; the equipment collaborative control and correction module is used to execute hierarchical scheduling instructions and trigger robust correction when the output of high-entropy energy exceeds the limit to ensure stable and economical operation of the system.
[0015] In summary, the present invention has at least one of the following beneficial technical effects: This invention provides a microgrid energy economic dispatch system and method, which introduces a spatiotemporal hierarchical adaptive modeling approach to achieve accurate quantification of the multi-scale fluctuation characteristics of high-entropy energy, constructs a dynamic uncertain set to replace the traditional fixed uncertain set, effectively adapts to the disorderly and abrupt characteristics of high-entropy energy, significantly reduces the curtailment rate of new energy sources, and improves the energy utilization efficiency of microgrids.
[0016] By adopting a dual-domain entropy balance collaborative scheduling mechanism for power and computing power, the technical barriers to independent scheduling of power and computing power resources are broken. Flexible computing power tasks are used to dynamically absorb redundant power from new energy sources. When the output of new energy sources fluctuates drastically, priority is given to ensuring the stability of core power loads, thereby achieving complementary optimization of dual resources and significantly reducing the overall operating cost of the system.
[0017] A lightweight distributed collaborative solution strategy adapted to ring microgrid topologies is adopted, significantly reducing the communication complexity and computational overhead of multi-microgrid cluster scheduling and improving scheduling efficiency. Simultaneously, through hybrid energy storage hierarchical regulation and over-limit robust correction mechanisms, high-frequency power fluctuations of high-entropy energy are precisely smoothed, effectively extending the service life of energy storage equipment. This comprehensively improves the stability, robustness, and economy of microgrid cluster operation, making it suitable for various engineering applications such as municipal smart microgrids and road renewable energy supply. Attached Figure Description
[0018] Figure 1 This is a flowchart illustrating a microgrid energy economic dispatch method according to the present invention. Figure 2 This is a schematic diagram of a typical day-ahead scheduling plan curve for a single microgrid according to a specific embodiment of the present invention; Figure 3 This is a schematic diagram illustrating the quantitative comparison of the full-cycle operation effect of a specific embodiment of the present invention. Detailed Implementation
[0019] The present invention will be further described in detail below with reference to the accompanying drawings.
[0020] This invention discloses a microgrid energy economic dispatch system and method.
[0021] Reference Figures 1-3 Example 1: A microgrid energy economic dispatch method, comprising the following steps: Step 1: Collect microgrid power operation data, high-entropy energy entropy flow data, edge computing task data, and grid electricity price data, and extract power fluctuation entropy and computing task distribution entropy features; Step 2: A spatiotemporal hierarchical physical information neural network is used to model the multi-scale fluctuation characteristics of high-entropy energy and adaptively generate dynamic output prediction intervals and dynamic uncertainty sets. Step 3: Construct a hierarchical Bruker optimization model based on the dynamic uncertainty set, and solve it using a distributed cooperative algorithm to obtain the microgrid group day-ahead power mutual assistance plan and energy storage operation baseline; Step 4: With the dual-domain entropy balance of power and computing power as the core objective, a Markov decision model is constructed, and intraday collaborative real-time scheduling of computing power and power is achieved through adaptive entropy reinforcement learning. Step 5: Issue scheduling instructions to complete the hierarchical control of equipment, and trigger a robust correction mechanism when the output of high-entropy energy exceeds the limit to ensure the stable and economical operation of the system.
[0022] By adopting the above technical solution, a multi-microgrid interconnection system adapting and configuring high-entropy energy capture devices, lithium-ion battery-supercapacitor hybrid energy storage, and edge computing nodes is implemented. The specific steps are as follows: Step 1: Collect power operation data, high-entropy energy entropy flow data, edge computing task data, and real-time grid price data under multi-microgrid operating conditions. Based on the collected raw data, perform standardization and normalization processing, and calculate and extract two core state features: power fluctuation entropy and computing task distribution entropy, providing basic input for subsequent scheduling modeling. Step 2: Build a spatiotemporal hierarchical physical information neural network model to simultaneously characterize the multi-scale fluctuation characteristics of high-entropy energy from both spatial and temporal dimensions. Through a training method that integrates physical constraints and data-driven approaches, adaptively generate a dynamic output prediction interval for high-entropy energy and construct a dynamic uncertainty set matching real-time fluctuation conditions. Step 3: Using the dynamic uncertainty set as the uncertainty constraint, build a microgrid cluster hierarchical Bruker optimization model. Use a distributed collaborative iterative algorithm to decompose and solve the global optimization problem, outputting a 24-hour day-ahead microgrid cluster cross-grid power mutual assistance plan, and simultaneously calibrating the baseline parameters for the all-day charging and discharging operation of the hybrid energy storage. Step four: Establish a dual-domain entropy balance optimization objective for power computing power, construct a Markov decision model adapted to the dynamic operation scenario of the microgrid, and complete real-time iterative solution based on the adaptive entropy reinforcement learning algorithm to achieve collaborative optimization scheduling of computing power resources and power resources at the minute-level scale within the day. Step five: Issue real-time optimization scheduling instructions to execution devices at each level to complete hierarchical collaborative control of energy storage, energy capture devices, and edge computing nodes. When high-entropy energy real-time output is detected to exceed the prediction range, a robust correction mechanism is automatically triggered to ensure the stable and economical operation of the microgrid cluster throughout the entire process.
[0023] Example 2, the spatiotemporal hierarchical physical information neural network modeling method in step 2 includes the following sub-steps: Step 21, Spatial Stratification: Divide the high-entropy energy capture region into a near-field turbulent region and a far-field stable region, and match the turbulent energy loss equation and the steady-state energy conversion equation as physical constraints respectively; Step 22, Time Stratification: Distinguish between two types of time series features: second-level transient fluctuations and minute-level steady-state offsets, and set differentiated time series weighting coefficients; Step 23, Model Training: Using quantile regression error and hierarchical physical residual as the joint loss function, adaptively adjust the model parameters and output the dynamic output prediction interval and fluctuation entropy value with confidence.
[0024] By adopting the above technical solution, step 21, spatial hierarchical modeling, divides the global high-entropy energy capture region into a near-field turbulent region and a far-field stable region. A turbulent energy loss equation is introduced as a physical constraint for the near-field turbulent region, and a steady-state energy conversion equation is introduced as a physical constraint for the far-field stable region, quantifying the differences in high-entropy energy conversion losses in different spatial regions. Step 22, temporal hierarchical modeling, divides high-entropy energy output fluctuations into two types of temporal features: second-level transient fluctuations and minute-level steady-state offsets. Differentiated temporal weight coefficients are set for different temporal features to achieve differentiated fitting of multi-scale temporal features. Step 23, model training and optimization, constructs a joint loss function composed of quantile regression error and hierarchical physical residuals. The core loss function formula is as follows: ;That For quantile regression error, For layered physical residuals, These are the physical constraint weight coefficients. The network model parameters are adaptively corrected through backpropagation. After iterative convergence, the high-entropy energy dynamic output prediction interval corresponding to the specified confidence level is output, while simultaneously quantifying the real-time fluctuation entropy value.
[0025] Example 3: The hierarchical and sub-Bruker optimization model construction method in step 3 is as follows: the goal is to minimize the sum of the microgrid group's electricity purchase cost, energy storage depreciation cost, and high-entropy energy curtailment cost; a dynamic Wasserstein probabilistic fuzzy set constraint is introduced, and the fuzzy set radius is adaptively adjusted based on the fluctuation entropy value in step 2 to replace the fixed uncertain set; a high-entropy energy output opportunity constraint is set to ensure that the system power supply reliability meets the preset threshold under extreme fluctuation conditions.
[0026] By adopting the above technical solution, the model aims to minimize the overall operating cost of the microgrid group. The core formula of the objective function is as follows: ,in For the electricity purchase cost of microgrid clusters, For the depreciation and loss costs of hybrid energy storage equipment, The cost of curtailing high-entropy energy is addressed. The model introduces a dynamic Wasserstein probabilistic fuzzy set as an uncertainty constraint. Based on the high-entropy energy fluctuation entropy value output in Example 2, the fuzzy set radius is adaptively adjusted to replace the traditional fixed uncertainty set, accurately adapting to dynamic fluctuation conditions. A high-entropy energy output opportunity constraint is set, with the core constraint formula as follows: in To actually contribute to high-entropy energy, , These are the upper and lower limits of the power output prediction, To preset power supply risk thresholds, the system power supply reliability is kept within the preset threshold range under extreme fluctuation conditions.
[0027] In Example 4, the distributed collaborative solution in step 3 adopts the lightweight consistency alternating direction multiplier method. Specifically, the global optimization model is decomposed into independent sub-problems of each microgrid. Each microgrid only interacts with the boundary power and Lagrange multiplier local information with the adjacent interconnected microgrids. The global variables are updated iteratively through weighted consistency of adjacent nodes until the boundary power deviation and dual residual are less than the preset convergence threshold. There is no need for global centralized calculation, which reduces communication and computing overhead.
[0028] By adopting the above technical solution, the implementation process of the lightweight consistency alternating direction multiplier method for distributed collaborative solution is refined. The global optimization model of the microgrid group is decomposed into multiple independent local subproblems according to the topological unit of a single microgrid. Each microgrid only interacts with adjacent microgrids in the ring topology, and the interaction variables only include the boundary mutual assistance power and the local Lagrange multiplier. During the iteration process, each microgrid solves the local quadratic programming subproblem in parallel, updates the local original dual residuals, and uses a weighting coefficient positively correlated with the microgrid spacing to complete the consistency fusion of the boundary power of adjacent nodes. The convergence criterion formula is as follows: ;in , The power at the boundary between adjacent microgrids. , These are the original residual and the dual residual, respectively. A preset convergence threshold is set. The solution terminates after the iteration meets the convergence condition, eliminating the need for global centralized computation and effectively reducing system communication overhead and computational resource consumption.
[0029] In Example 5, the state space of the Markov decision model in step 4 includes the real-time output fluctuation entropy of high-entropy energy, the state of charge of hybrid energy storage, the queuing entropy of computing tasks, the real-time electricity price and the output prediction deviation; the action space includes the steady-state charging and discharging power of lithium batteries, the transient compensation power of supercapacitors, the fine-tuning power of microgrid interconnects, the start and stop time slots of edge computing tasks and the adjustment amount of node operating frequency.
[0030] By adopting the above technical solutions, the model's state space is used to comprehensively characterize the real-time operating conditions of the microgrid. Specifically, it includes five types of state variables: real-time output fluctuation entropy of high-entropy energy sources, real-time state of charge of hybrid energy storage, queuing entropy of computing tasks, real-time grid electricity price, and prediction deviation of high-entropy energy output. This fully covers the fluctuation characteristics of the power domain and the load characteristics of the computing domain. The model's action space is a set of controllable variables, specifically including five types of controllable action variables: steady-state charging and discharging power of lithium batteries, transient compensation power of supercapacitors, fine-tuning power of microgrid interconnection lines, start-stop time slots of edge computing tasks, and adjustment of edge node operating frequency. This enables refined and coordinated control of power resources and computing resources.
[0031] In Example 6, the dual-domain entropy balance optimization objective in step 4 is the weighted minimum value of system operating cost, power fluctuation entropy, and computing power task congestion entropy. By adaptively adjusting the reinforcement learning temperature parameter, the model's exploration and utilization capabilities are dynamically balanced. When high-entropy energy fluctuations are severe, the model's exploration degree is improved, actively adapting to sudden changes in operating conditions, and realizing dynamic coordination between power consumption and computing power scheduling.
[0032] By adopting the above technical solution, the model aims to minimize the weighted sum of the overall system operating cost, power fluctuation entropy, and computing power task congestion entropy. The core objective formula is as follows: ;in To accumulate reward value, Incentives for cost savings For power fluctuation entropy, For computing power tasks, congestion entropy, , , The weighting coefficients are fixed. The adaptive entropy reinforcement learning algorithm explores and utilizes the model by dynamically adjusting the temperature parameter balance. When the output of high-entropy energy fluctuates drastically, it automatically increases the strategy exploration degree, captures the optimal scheduling strategy under sudden operating conditions, and strengthens the strategy utilization capability under stable operating conditions, ensuring scheduling stability and economy, and realizing dynamic collaborative matching between power consumption and computing power scheduling.
[0033] In Example 7, the equipment hierarchical control method in step 5 is as follows: the supercapacitor responds independently to the high-frequency power fluctuations of high-entropy energy at the second level, suppressing instantaneous power disturbances; the lithium battery undertakes the steady-state power regulation at the minute level, balancing the energy supply and demand of the system; the high-entropy energy capture device adjusts the operating parameters according to the feedforward of the prediction interval, improving the efficiency of disordered energy capture.
[0034] By adopting the above technical solutions, the system employs a hybrid energy storage strategy with differentiated spatiotemporal scale control. The supercapacitor specifically responds to the second-level high-frequency power fluctuations of high-entropy energy, rapidly suppressing instantaneous power disturbances and offsetting short-term power surges caused by turbulent changes and traffic fluctuations. The lithium battery focuses on minute-level steady-state power regulation, smoothing out long-term power deviations and balancing the overall energy supply and demand of the microgrid. The high-entropy energy capture device relies on real-time dynamic prediction intervals to perform feedforward control, dynamically adjusting the guide vane angle and energy capture operating parameters to adapt to real-time entropy flow fluctuations, maximizing the capture and utilization efficiency of disordered high-entropy energy.
[0035] In Example 8, the robust correction mechanism in step 5 is triggered when the real-time output of the high-entropy energy exceeds the boundary of the dynamic prediction interval; the execution strategy after triggering is as follows: Prioritize shielding low-priority, latency-tolerant computing tasks on edge computing nodes to reduce unnecessary power consumption, while controlling the instantaneous output of supercapacitors to compensate for power gaps and ensure uninterrupted power supply to critical loads in the microgrid.
[0036] By adopting the above technical solution, the robust correction mechanism is triggered when the real-time output value of the high-entropy energy exceeds the upper and lower boundaries of the dynamic prediction range, indicating that the system is experiencing extreme output fluctuations. After the mechanism is triggered, a hierarchical priority control strategy is implemented, prioritizing the blocking of low-priority, latency-tolerant computing tasks on edge computing nodes, reducing ineffective computing power consumption, and simultaneously controlling the instantaneous compensation power output of supercapacitors to quickly fill the system power gap. This ensures uninterrupted and stable power supply to critical power loads of the microgrid throughout the entire process, avoiding power failures and scheduling failures caused by extreme fluctuations.
[0037] Example 9: In both daytime robust scheduling and intraday real-time scheduling, differentiated constraints are applied to hybrid energy storage: high-frequency power response constraints and short-term charge-discharge lifetime constraints are set for supercapacitors to quantify transient fluctuation adjustment losses; energy throughput constraints and long-term cycle lifetime constraints are set for lithium batteries to quantify steady-state energy scheduling losses. The cost of both types of energy storage losses is included in the optimization of the total system operating cost.
[0038] By adopting the above technical solutions, the constraint mechanism is simultaneously applicable to the entire process of day-ahead robust scheduling and intraday real-time scheduling. For supercapacitors, high-frequency power response constraints and short-term charge-discharge lifetime constraints are set, quantifying equipment power loss and lifespan wear during second-level high-frequency fluctuation adjustment, and limiting the maximum charge-discharge frequency and instantaneous power amplitude within a unit time period. For lithium batteries, long-term energy throughput constraints and cycle life constraints are set, quantifying capacity loss and cycle aging loss during steady-state energy scheduling, and limiting the maximum daily energy throughput and cycle count. The dynamic loss costs of both types of energy storage equipment are uniformly included in the overall system operating cost optimization scope, achieving optimal management of energy storage equipment lifespan while ensuring scheduling performance.
[0039] Example 10: A microgrid energy economic dispatch system for implementing a microgrid energy economic dispatch method. The system comprises: a data acquisition module, an entropy feature extraction module, a high-entropy energy hierarchical modeling and prediction module, a day-ahead distributed robust dispatch module, a dual-domain entropy balance real-time dispatch module, and an equipment collaborative control module. The data acquisition module is used to collect heterogeneous data on microgrid power, high-entropy energy, edge computing power, and electricity prices. The entropy feature extraction module is connected to the data acquisition module and is used to extract the core features of power fluctuation entropy and computing power queuing entropy. The high-entropy energy stratification modeling and prediction module embeds a spatiotemporal stratified physical information neural network, which adaptively outputs the dynamic output prediction interval and dynamic uncertainty set of high-entropy energy. The distributed robust scheduling module is used to construct and solve the hierarchical robust optimization model to generate the day-ahead basic scheduling plan for the microgrid cluster. The dual-domain entropy balance real-time scheduling module deploys an adaptive entropy reinforcement learning agent to achieve optimal decision-making between computing power and electricity within the day; the equipment collaborative control and correction module is used to execute hierarchical scheduling instructions and trigger robust correction when the output of high-entropy energy exceeds the limit to ensure stable and economical operation of the system.
[0040] By adopting the above technical solutions, the data acquisition module is responsible for collecting microgrid power operation data, high-entropy energy entropy flow data, edge computing task data, and heterogeneous grid electricity price data across the entire domain, and completing the preprocessing of raw data. The entropy feature extraction module connects to the data acquisition module and calculates and extracts the core feature parameters of power fluctuation entropy and computing task queuing entropy based on the preprocessed data. The high-entropy energy hierarchical modeling and prediction module embeds a spatiotemporal hierarchical physical information neural network model, adaptively outputting the dynamic output prediction interval and dynamic uncertainty set of high-entropy energy. The day-ahead distributed robust scheduling module is responsible for constructing a hierarchical distributed robust optimization model, and generating the microgrid group's day-ahead basic scheduling plan and energy storage operation baseline through a lightweight distributed collaborative algorithm. The dual-domain entropy balance real-time scheduling module deploys an adaptive entropy reinforcement learning agent, and completes the optimal decision output of computing power and power dual-domain collaborative scheduling based on real-time operating conditions. The equipment collaborative control and correction module is responsible for parsing and executing scheduling instructions at all levels, realizing hierarchical and precise control of equipment, and automatically triggering a robust correction mechanism when the output of high-entropy energy exceeds the limit, ensuring the stable, efficient, and economical operation of the microgrid group system throughout the entire process.
[0041] The following specific embodiments illustrate the implementation principle of the present invention: This invention presents a 12-node ring-connected microgrid cluster engineering scenario. Each node microgrid is configured with a photovoltaic power generation unit, an airflow-type high-entropy energy capture device, lithium battery energy storage, supercapacitor energy storage, and an edge computing unit, adapting to the normal traffic flow and airflow disturbance conditions of urban roads. The system scheduling duration is set to 24 hours, with scheduling steps divided into 1-hour pre-scheduling before the day and 5-minute real-time rolling scheduling within the day. All simulation and operating parameters are quantitatively configured. The entire scheduling method of this invention is used to complete the whole process calculation and control, realizing a complete closed loop of data, formulas, decisions, and execution.
[0042] The case study uses the following basic quantitative parameter configurations: a single microgrid with a rated photovoltaic output of 150kW, a high-entropy energy device with a rated output of 50kW, a lithium battery with a rated capacity of 300kWh and a rated charge / discharge power of 80kW, a supercapacitor with a rated power of 50kW, and an effective energy storage capacity of 15kWh. The microgrid's typical base load fluctuates between 60kW and 120kW. The edge computing unit includes three types of deferred tasks: video analysis, data statistics, and model inference. The maximum computing power consumption per node is 30kW. The grid's peak electricity price is 0.85 yuan / kWh, the average price is 0.55 yuan / kWh, and the off-peak price is 0.25 yuan / kWh. The high-entropy energy output has a pre-set confidence interval of 95%, a power supply risk threshold ε is set to 0.05, and the algorithm convergence threshold δ is set to 0.01.
[0043] Data acquisition and entropy feature quantification calculation: The system collects full-domain operational data every 5 seconds. Collected variables include real-time photovoltaic power output, real-time high-entropy energy output, energy storage state of charge, load power, computing task queue length, node power consumption, and real-time electricity price. Power fluctuation entropy is used to quantify the degree of disordered fluctuations in the power supply; the calculation formula is as follows: In the formula, This represents the probability distribution of power fluctuations at each sampling time. The computing power task queuing entropy is used to quantify the degree of computing power congestion; the calculation formula is as follows: In the formula The system uses dual entropy values to quantify the real-time disordered state of each computing task queue length ratio, providing accurate state input for subsequent optimization and scheduling.
[0044] High-entropy energy spatiotemporal hierarchical modeling and interval prediction calculations were performed. The model employed a spatiotemporal hierarchical physical information neural network, with the physical constraint weight coefficient α set to 0.7. The near-field turbulent region was substituted into the turbulent energy loss equation. In the formula This refers to real-time airflow speed. The turbulence loss coefficient is set to 0.08. Substituting this value into the steady-state energy conversion equation in the far-field stable region... In the formula The steady-state transition coefficient is set to 0.42. The time-layered weighting is configured with a 0.6 weight for second-level transients and a 0.4 weight for minute-level steady-states. The joint loss function for model training is... The model converged after 120 rounds of iterative training. Under daytime traffic disturbance conditions, the model outputs a high-entropy energy output prediction range of 12.6kW to 38.4kW, and the radius of the dynamic uncertainty set is adaptively updated to 0.18 based on the real-time fluctuation entropy value, accurately covering the power fluctuations caused by sudden turbulence changes.
[0045] The numerical solution of the hierarchical Bruker bar optimization model was recently obtained, and the objective function of the model was... Complete 24-hour global cost optimization. Combine dynamic Wasserstein fuzzy set constraints with opportunity constraints. To ensure that the probability of exceeding the limit under extreme fluctuations is less than 5%, a lightweight, consistent alternating directional multiplier method is used to complete the 12-node distributed solution. Each microgrid only interacts with the boundary power and Lagrange multipliers. After 18 iterations, the residual converges to within 0.01. The solution yields a total day-ahead optimal electricity purchase of 2186 kWh, an average daily charge-discharge cycle of 4.2 cycles for energy storage, and a theoretical high-entropy energy absorption capacity of 628 kWh. Hourly interconnection power and energy storage charge-discharge baselines are generated for each microgrid. The baseline power range for lithium batteries is -40 kW to 60 kW, and the baseline power range for supercapacitors is -30 kW to 30 kW. Figure 2 This is a 24-hour day-ahead scheduling plan curve for a single microgrid generated based on the method of this invention. The upper subplot shows the photovoltaic output curve, the high-entropy energy prediction range and its median, and the load power curve; the lower subplot shows the charging and discharging power baseline of the lithium battery and supercapacitor. As can be seen from the figure, the high-entropy energy output prediction range widens significantly during daytime traffic peaks, accurately reflecting the power fluctuation range caused by traffic airflow disturbances. The lithium battery mainly performs minute-level steady-state power regulation, undertaking peak shaving and valley filling functions, with a power range of -60kW to 60kW; the supercapacitor mainly performs second-level transient regulation, with a power range of -30kW to 30kW. The two achieve clear decoupling in terms of time scale and power allocation. This plan is obtained by solving a day-ahead hierarchical split-bar optimization model and a lightweight consistent alternating direction multiplier method, providing a reliable baseline for intraday real-time scheduling.
[0046] Intraday dual-domain entropy balance real-time scheduling quantization computation is performed to construct a Markov decision model. The state vector is input in real-time with five types of variables: fluctuation entropy, energy storage state of charge, computing power queuing entropy, electricity price, and prediction bias. A reinforcement learning reward function is then applied. Weighting coefficient The value is 0.8. The value is 0.1. The value is set to 0.1. The system adaptively adjusts the temperature parameter based on real-time operating conditions. During periods of high fluctuation, the temperature parameter is increased to 1.2 to enhance strategy exploration capabilities, while during periods of stability, the temperature parameter is decreased to 0.6 to strengthen strategy reuse. During the day, the actual output of high-entropy energy increases to 42.8kW, exceeding the steady-state baseline. The system increases the edge computing power operating power to 24kW, absorbing 18.2kWh of redundant energy and reducing energy waste. Under sudden gusts of wind, the output of high-entropy energy drops to 10.2kW. The system postpones all low-priority computing tasks, instantly shuts down 16kW of computing power, and simultaneously outputs 28kW of lithium battery compensation power and 12kW of supercapacitor instantaneous compensation power to stabilize the system power balance.
[0047] Hybrid energy storage employs tiered constraints and quantitative loss calculations. Supercapacitors undergo second-level high-frequency regulation, constrained to a maximum charge / discharge frequency of 120 times / hour, with instantaneous power not exceeding 50kW and a single high-frequency loss coefficient of 0.002. Lithium batteries undergo minute-level steady-state regulation, constrained to a maximum daily energy throughput of 240kWh and a cycle loss coefficient of 0.015. The system incorporates the dynamic loss costs of both types of energy storage into the total cost calculation in real time. Under high-frequency fluctuation conditions, the daily average loss cost of supercapacitors is reduced by 12.4%, and under steady-state operation conditions, the cycle life loss of lithium batteries is reduced by 8.7%, achieving optimal synergy between dispatch performance and equipment lifespan.
[0048] The robust correction mechanism under extreme operating conditions was verified during operation. Under a sudden strong turbulence condition, the instantaneous output of high-entropy energy was 9.8kW, lower than the predicted lower limit of 12.6kW. The system determined that the output exceeded the limit and triggered the correction mechanism. All non-urgent computing tasks were immediately disabled, reducing ineffective power consumption by 22kW. The supercapacitor output 14.2kW of instantaneous compensation power, completing the power gap filling within 3 seconds. The microgrid bus voltage fluctuation was controlled within ±2%, and the core lighting and monitoring loads operated uninterruptedly without any scheduling failures.
[0049] A quantitative comparison of the full-cycle operation performance shows that, compared to traditional fixed-uncertain-set scheduling and independent power computing power scheduling schemes, this case study reduces the overall 24-hour operating cost by 18.3%, decreases the high-entropy energy curtailment rate from 11.6% to 2.8%, reduces the average daily loss of energy storage equipment by 10.5%, and improves the effective utilization rate of edge computing resources by 26.7%. The system can adaptively adapt to the irregular and sudden changes in high-entropy energy conditions, reduces the communication interaction overhead of the ring microgrid cluster by 42%, and improves the solution convergence speed by 35%, fully verifying the creativity and engineering practicality of the modeling mechanism, collaborative scheduling logic, and distributed solution algorithm of this invention.
[0050] Figure 3This paper compares the key performance indicators of the proposed solution with those of the traditional fixed scheduling solution under a 24-hour full-cycle operation. The comparison dimensions include the reduction in overall operating cost, high-entropy energy curtailment rate, reduction in daily average energy storage loss, improvement in edge computing resource utilization, reduction in communication overhead, and improvement in solution convergence speed. The results show that the proposed solution reduces the high-entropy energy curtailment rate to 2.8%, a significant decrease compared to the traditional solution's 11.6%; reduces overall operating cost by 18.3%, reduces daily average energy storage loss by 10.5%, and improves edge computing utilization by 26.7%. Furthermore, the distributed solution based on the lightweight, consistent alternating direction multiplier method and the ring microgrid group local communication mechanism reduces communication overhead by 42% and improves solution convergence speed by 35%. These data fully verify the significant technical effects of the proposed solution in improving absorption capacity, reducing operating costs, extending equipment lifespan, and enhancing system scalability.
[0051] The system's overall operating cycle is set to a full calendar day, employing a dual-timescale scheduling mode of pre-scheduling before the day and real-time rolling scheduling within the day. During the data acquisition phase, the system collects basic power data on photovoltaic output, equipment load power, and energy storage charge status across the entire domain. Simultaneously, it collects high-entropy energy entropy flow data such as road turbulence intensity, traffic flow pulsation frequency, and airflow output fluctuation amplitude. It also aggregates edge computing unit task queue length, task processing time, and node operating power consumption data. Combined with regional time-of-use electricity price signals, the system performs unified data cleaning and normalization, extracting power fluctuation entropy and computing power task distribution entropy as the basic input features for all-day scheduling.
[0052] In the high-entropy energy prediction modeling phase, the system employs a spatiotemporally hierarchical physical information neural network to predict energy output across the entire domain. Spatially, the near-field turbulent region is defined as the area with traffic disturbance near the road, while the far-field stable region is defined as the open road section. Corresponding physical constraint equations are matched to quantify energy loss for each region. Temporally, different fluctuation characteristics are distinguished between day and night. During peak traffic periods in the daytime, the weight of transient fluctuations at the second level is strengthened, while during stable airflow periods at night, the weight of steady-state offsets at the minute level is strengthened. The model parameters are iteratively optimized through a joint loss function, outputting dynamic prediction intervals for high-entropy energy output across all time periods of the day. Corresponding dynamic uncertainty sets are adaptively generated, providing precise uncertainty constraints for robust scheduling.
[0053] During the recent scheduling phase, the system aims to minimize the overall daily operating cost of the twelve-node microgrid cluster. A hierarchical, distributed bar optimization model is constructed, incorporating electricity purchase costs, energy storage depreciation costs, and high-entropy energy curtailment costs into a unified optimization framework. Dynamic Wasserstein probabilistic fuzzy sets are used to adaptively adjust the uncertainty set constraint radius, ensuring power supply reliability under extreme turbulence and sudden traffic flow conditions. A lightweight, consistent alternating direction multiplier method is employed to distribute the global model. Each microgrid node interacts only with its ring-shaped adjacent nodes regarding boundary power and Lagrange multiplier information, solving local subproblems in parallel iteratively. Upon convergence, a 24-hour microgrid cluster interconnection and mutual support power plan is output, the charging and discharging operation baselines for hybrid energy storage at different times are calibrated, and the basic high-entropy energy consumption scheme for the entire day is determined.
[0054] During the daytime real-time scheduling phase, the system focuses on balancing the dual-domain entropy of power and computing power, continuously building and updating the Markov decision model. It collects and updates real-time data on high-entropy energy output fluctuation entropy, energy storage state of charge, computing power task queuing entropy, real-time electricity prices, and prediction deviation state variables. It dynamically outputs variables such as lithium battery steady-state charging and discharging power, supercapacitor transient compensation power, microgrid tie-line fine-tuning power, edge task start / stop time slots, and node operating frequency adjustment. Through an adaptive entropy reinforcement learning algorithm, it dynamically adjusts the strategy exploration and utilization coefficients. When high-entropy energy output fluctuates drastically during the day, it enhances the model's exploration capabilities, prioritizing power supply to core road monitoring and lighting loads, and postponing non-urgent data inference tasks. At night, when renewable energy output is stable and computing power task backlog is low, it fully utilizes surplus high-entropy energy output, increases edge computing power operation power, locally absorbs redundant energy, and reduces system energy curtailment rate and grid purchase costs.
[0055] During the equipment execution and robust correction phases, the system strictly adheres to the hybrid energy storage hierarchical control logic. Supercapacitors continuously respond to second-level high-frequency power fluctuations caused by sudden changes in traffic flow and airflow, quickly smoothing out instantaneous power disturbances. Lithium batteries track the intraday scheduling baseline to complete minute-level steady-state energy regulation, balancing power supply and demand across the entire region. The high-entropy energy capture device adaptively adjusts the guide vane angle and operating parameters based on real-time prediction intervals, continuously improving the efficiency of disordered energy capture. When extreme gusts or sudden heavy traffic conditions cause the high-entropy energy output to exceed the dynamic prediction interval, the system immediately triggers a robust correction mechanism, batch-screening non-emergency background computing tasks at edge nodes to reduce ineffective power consumption, while simultaneously controlling the instantaneous output of supercapacitors to compensate for power gaps, ensuring uninterrupted operation of the core road power load throughout the entire process.
[0056] Through engineering testing and operational verification, compared with traditional fixed scheduling and independent computing power control modes, the application of this case study significantly reduces the high-entropy energy curtailment rate of the microgrid cluster, lowers the overall operating cost, greatly reduces high-frequency losses of energy storage equipment, and effectively improves the utilization rate of edge computing resources. Without human intervention, the system can adaptively adapt to complex and fluctuating road conditions in various scenarios, balancing the stability of microgrid power supply with operational economy, fully validating the engineering practicality and inventive advantages of the technical solution of this invention.
[0057] The above are all preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Therefore, all equivalent changes made in accordance with the structure, shape and principle of the present invention should be covered within the scope of protection of the present invention.
Claims
1. A micro-grid energy economic dispatching method, characterized in that, Includes the following steps: Step 1: Collect microgrid power operation data, high-entropy energy entropy flow data, edge computing task data, and grid electricity price data, and extract power fluctuation entropy and computing task distribution entropy features; Step 2: A spatiotemporal hierarchical physical information neural network is used to model the multi-scale fluctuation characteristics of high-entropy energy and adaptively generate dynamic output prediction intervals and dynamic uncertainty sets. Step 3: Construct a hierarchical Bruker optimization model based on the dynamic uncertainty set, and solve it using a distributed cooperative algorithm to obtain the microgrid group day-ahead power mutual assistance plan and energy storage operation baseline; Step 4: With the dual-domain entropy balance of power and computing power as the core objective, a Markov decision model is constructed, and intraday collaborative real-time scheduling of computing power and power is achieved through adaptive entropy reinforcement learning. Step 5: Issue scheduling instructions to complete the hierarchical control of equipment, and trigger a robust correction mechanism when the output of high-entropy energy exceeds the limit to ensure the stable and economical operation of the system.
2. The microgrid energy economic dispatch method of claim 1, wherein, Step 2, the spatiotemporal hierarchical physical information neural network modeling method, includes the following sub-steps: Step 21, Spatial Stratification: Divide the high-entropy energy capture region into a near-field turbulent region and a far-field stable region, and match the turbulent energy loss equation and the steady-state energy conversion equation as physical constraints respectively; Step 22, Time Stratification: Distinguish between two types of time series features: second-level transient fluctuations and minute-level steady-state offsets, and set differentiated time series weighting coefficients; Step 23, Model Training: Using quantile regression error and hierarchical physical residual as the joint loss function, adaptively adjust the model parameters and output the dynamic output prediction interval and fluctuation entropy value with confidence.
3. The microgrid energy economic dispatch method of claim 2, wherein, The hierarchical and sub-Bruker optimization model construction method in step 3 is as follows: the goal is to minimize the sum of the microgrid group's electricity purchase cost, energy storage depreciation cost, and high-entropy energy curtailment cost; a dynamic Wasserstein probabilistic fuzzy set constraint is introduced, and the radius of the fuzzy set is adaptively adjusted based on the fluctuation entropy value in step 2 to replace the fixed uncertain set; a high-entropy energy output opportunity constraint is set to ensure that the system power supply reliability meets the preset threshold under extreme fluctuation conditions.
4. The microgrid energy economic dispatch method of claim 3, wherein, In step 3, the distributed collaborative solution adopts the lightweight consistency alternating direction multiplier method. Specifically, the global optimization model is decomposed into independent sub-problems of each microgrid. Each microgrid only interacts with the boundary power and Lagrange multiplier local information with the adjacent interconnected microgrids. The global variables are updated iteratively through weighted consistency between adjacent nodes until the boundary power deviation and dual residual are less than the preset convergence threshold. There is no need for global centralized calculation, which reduces communication and computation overhead.
5. A microgrid energy economic dispatch method according to claim 4, characterized in that, In step 4, the state space of the Markov decision model includes the real-time output fluctuation entropy of high-entropy energy, the state of charge of hybrid energy storage, the queuing entropy of computing tasks, the real-time electricity price and output prediction deviation; the action space includes the steady-state charging and discharging power of lithium batteries, the transient compensation power of supercapacitors, the fine-tuning power of microgrid interconnects, the start and stop time slots of edge computing tasks and the adjustment amount of node operating frequency.
6. The microgrid energy economic dispatch method according to claim 5, characterized in that, In step 4, the dual-domain entropy balance optimization objective is to minimize the weighted average of system operating cost, power fluctuation entropy, and computing power task congestion entropy. By adaptively adjusting the reinforcement learning temperature parameter, the model's exploration and utilization capabilities are dynamically balanced. When high-entropy energy fluctuations are severe, the model's exploration degree is improved, and it actively adapts to sudden changes in operating conditions, thereby achieving dynamic coordination between power consumption and computing power scheduling.
7. The microgrid energy economic dispatch method of claim 6, wherein, The equipment hierarchical control method in step 5 is as follows: the supercapacitor responds independently to the high-frequency power fluctuations of high-entropy energy at the second level, suppressing instantaneous power disturbances; the lithium battery undertakes the steady-state power regulation at the minute level, balancing the energy supply and demand of the system; and the high-entropy energy capture device adjusts its operating parameters according to the feedforward of the prediction interval, improving the efficiency of disordered energy capture. 8.The micro-grid energy economic dispatching method of claim 7, wherein, In step 5, the robust correction mechanism is triggered when the real-time output of the high-entropy energy source exceeds the boundary of the dynamic prediction interval; the strategy executed after triggering is as follows: Prioritize shielding low-priority, latency-tolerant computing tasks on edge computing nodes to reduce unnecessary power consumption, while controlling the instantaneous output of supercapacitors to compensate for power gaps and ensure uninterrupted power supply to critical loads in the microgrid. 9.The micro-grid energy economic dispatching method of claim 8, wherein, In recent robust scheduling and intraday real-time scheduling, differentiated constraints were adopted for hybrid energy storage: high-frequency power response constraints and short-term charge-discharge lifetime constraints were set for supercapacitors to quantify transient fluctuation adjustment losses; energy throughput constraints and long-term cycle lifetime constraints were set for lithium batteries to quantify steady-state energy dispatch losses. The cost of both types of energy storage losses was included in the optimization of the total system operating cost.
10. A micro-grid energy economic dispatch system, characterized in that, The system is used to implement the scheduling method according to any one of claims 1-9, characterized in that it includes: a data acquisition module, an entropy feature extraction module, a high-entropy energy hierarchical modeling and prediction module, a day-ahead distributed robust scheduling module, a dual-domain entropy balance real-time scheduling module, and an equipment collaborative control module. The data acquisition module is used to collect heterogeneous data on microgrid power, high-entropy energy, edge computing power, and electricity prices. The entropy feature extraction module is connected to the data acquisition module and is used to extract the core features of power fluctuation entropy and computing power queuing entropy. The high-entropy energy stratification modeling and prediction module embeds a spatiotemporal stratified physical information neural network, which adaptively outputs the dynamic output prediction interval and dynamic uncertainty set of high-entropy energy. The distributed robust scheduling module is used to construct and solve the hierarchical robust optimization model to generate the day-ahead basic scheduling plan for the microgrid cluster. The dual-domain entropy balance real-time scheduling module deploys an adaptive entropy reinforcement learning agent to achieve optimal decision-making between computing power and electricity within the day; the equipment collaborative control and correction module is used to execute hierarchical scheduling instructions and trigger robust correction when the output of high-entropy energy exceeds the limit to ensure stable and economical operation of the system.
