Micro-grid cooperative scheduling control method, system and device based on edge calculation and medium

Abstract

The invention discloses a micro-grid cooperative scheduling control method, system, equipment and medium based on edge calculation, and relates to the technical field of computer platform load balancing, and the method comprises the steps: collecting power grid operation data through a micro-grid unit, carrying out the short-time power prediction of the micro-grid unit according to the power grid operation data, and obtaining a load prediction value; deploying a genetic algorithm at an edge node, and optimizing a load prediction value to obtain a scheduling strategy parameter; the scheduling strategy parameters are input into the micro-grid units, a game model is constructed to carry out multi-micro-grid cooperative game, and optimal scheduling strategy parameters are output; and performing power output scheduling through a multi-stage control mechanism. According to the method, through multi-dimensional data acquisition and sequence feature learning of micro-grid units, high-precision short-time power prediction is realized, a rapid convergence local optimization strategy is obtained, it is ensured that multiple micro-grids reach Nash equilibrium under autonomous conditions, and rapid response to sudden disturbance and effective correction of steady-state errors are realized.

Description

Technical Field

[0001] This invention relates to the field of microgrid scheduling technology, specifically to a microgrid collaborative scheduling control method, system, device, and medium based on edge computing. Background Technology

[0002] Over the past decade, with the accelerated advancement of distributed grid access, microgrids have gradually shifted from single-point demonstration to multi-cluster interconnected operation. Dynamic scheduling of distributed power sources, energy storage, and adjustable loads has become an important means of maintaining local power quality and peak shaving. Traditionally, most approaches rely on a centralized energy management system (EMS) to collect operational data such as distributed power source output, electricity load, and energy storage state of charge, and then calculate the optimal power allocation based on model predictive control (MPC) or linear programming (LP) methods.

[0003] However, centralized solutions are limited by communication bandwidth, latency, and the risk of single points of failure. When renewable energy output fluctuates by seconds or loads experience rapid changes, the central controller often struggles to issue scheduling commands in a timely manner. Furthermore, the unified solution of quadratic cone / linear programming requires complete network topology and global parameters, which violates the principle of data autonomy and leads to computational bottlenecks. Edge computing offers a new architecture for real-time decision-making in microgrids: deploying lightweight computing nodes close to physical devices shortens data backhaul paths and achieves local autonomy. However, most edge scheduling methods remain in an isolated "single microgrid—single optimizer" model, lacking cross-node collaboration. Simultaneously, load forecasting and scheduling optimization are typically completed by two independent pipelines, and forecasting errors cannot be immediately fed back to the strategy search process, resulting in time drift between the scheduling scheme and actual operating conditions. For rapidly fluctuating scenarios, how to organically couple short-term forecasting, intelligent optimization, and hierarchical control at the edge remains a key challenge of common concern to both academia and industry. Summary of the Invention

[0004] In view of the above-mentioned problems, the present invention is proposed.

[0005] Therefore, the technical problem solved by this invention is that existing edge computing microgrid scheduling technology still has many bottlenecks in terms of prediction accuracy, scheduling real-time performance, distributed collaboration and multi-level fault tolerance.

[0006] To address the aforementioned technical problems, this invention provides the following technical solution: a microgrid collaborative scheduling and control method based on edge computing, comprising:

[0007] By collecting grid operation data through microgrid units, and performing short-time power prediction on microgrid units based on grid operation data, load prediction values ​​are obtained.

[0008] Deploy genetic algorithms at edge nodes to optimize load forecasts and obtain scheduling strategy parameters;

[0009] Input the scheduling strategy parameters into the microgrid unit, construct a game model to conduct multi-microgrid collaborative game, and output the optimal scheduling strategy parameters.

[0010] The optimal scheduling strategy parameters are sent to the edge nodes, and power output is scheduled through a multi-level control mechanism.

[0011] As a preferred embodiment of the microgrid collaborative scheduling and control method based on edge computing described in this invention, the short-term power prediction includes: preprocessing the grid operation data collected by the microgrid unit; and predicting the short-term load forecast value for each microgrid unit in the next H steps based on the processed data through a local controller.

[0012] The short-term load forecast value This includes active power demand and initial projected power generation;

[0013] The power grid operation data includes distributed power output, load power, and state of charge data of energy storage batteries; a scheduling control cycle is defined as ΔT, and a short-time power prediction is performed every ΔT.

[0014] As a preferred embodiment of the microgrid collaborative scheduling and control method based on edge computing described in this invention, the step of deploying a genetic algorithm at the edge nodes includes processing the short-term load forecast values ​​generated by each microgrid unit. Upload the data to the corresponding edge computing node; at the edge node, design a scheduling optimization objective to minimize the cost of purchasing electricity from the public grid and the losses from charging and discharging energy storage, and use a genetic algorithm to solve for the optimal solution.

[0015] The short-term load forecast values ​​are grouped to obtain initial candidate scheduling schemes. Each scheme includes the power trading sequence with the public grid in the future time period and the corresponding energy storage charging and discharging strategy.

[0016] For each candidate option, sum up its electricity price and energy storage usage costs over the entire forecast period;

[0017] Population updates are conducted through selection, crossover, and mutation operations with the aim of reducing overall costs.

[0018] When the maximum number of generations is reached, the current optimal scheduling strategy parameters are output; the edge nodes return the optimal strategy parameters to each microgrid unit as the initial input for the next cooperative game.

[0019] As a preferred embodiment of the microgrid collaborative scheduling and control method based on edge computing described in this invention, the construction of the game model includes determining the participants and their numbers, where M = {1, 2, ..., N} is the set of microgrid units participating in the game, and N is the total number of microgrid units in the system.

[0020] Define a strategy space, where the strategy variable for each participant i is the trading power P. i : Represents the active power sold or purchased by the i-th microgrid to the public grid or a neighboring microgrid during the current scheduling cycle;

[0021] The strategy space is the interval [P] i,min ,P i,max ], where the lower bound P i,min With upper boundary P i,max This stems from the limitations of the unit's power generation and energy storage capacity;

[0022] Construct a utility function, and for each unit i, define its payment function as electricity price revenue minus transaction loss cost; introduce a global power balance constraint, which is added to individual decisions in the later stage of the game through an additional penalty function, and uniformly corrected in subsequent hierarchical control;

[0023] Neighborhood structure and consensus weights, to accelerate information fusion, stipulate that each participant i has a neighborhood set. Indicates other units that it can communicate with directly;

[0024] For each pair (i,j) in the neighborhood, assign a weight w. ij And ensure that the sum of the weights of itself and all its neighbors is 1;

[0025] Define the optimal response mapping BR i For other fixed unit strategies {P} j :j≠i}, the optimal response BR of unit i i Defined as, within the permissible power range [P] i,min ,P i,max Within, select the option that enables utility U. i Reach the maximum point;

[0026] The mapping provides unit i with the optimal trading decision for its local environment; when the strategies of all units are mutually optimal responses, the set... This constitutes the Nash equilibrium of the game.

[0027] As a preferred embodiment of the microgrid collaborative scheduling and control method based on edge computing described in this invention, the multi-microgrid collaborative game includes each microgrid unit determining its local optimal P based on an edge genetic algorithm. i (0)As an initial value; after fixing the current strategies of other units, unit i calculates and updates its own optimal BR. i After the parameters are updated, neighborhood consensus fusion is performed, and each unit is fused according to the preset weight w. ij A new policy estimate is obtained by weighting the temporary policies of the unit itself and its neighbors; convergence is determined by whether the change in the policies of all units is less than the convergence threshold δ, or by reaching the maximum number of iterations K. max If the convergence fails, the process terminates; the output equilibrium strategy is: after convergence, the strategy of each unit is the globally consistent trading power.

[0028] As a preferred embodiment of the microgrid collaborative scheduling and control method based on edge computing described in this invention, the multi-level control mechanism includes: As a reference target for control at all levels, three-level closed-loop regulation is implemented at the equipment level, microgrid level, and regional level respectively;

[0029] Equipment-level droop control measures local instantaneous active power output P in real time. i (t), instantaneous reactive power output Q i (t), every control cycle ΔT1, the local controller reads the current output power, frequency, and voltage values, and compares them with the target value of active power. Target reactive power Compare;

[0030] Multiply the active power deviation by the active power droop coefficient m i Generate a new frequency reference; multiply the reactive power deviation by the reactive power droop factor n. i Generate a new voltage reference;

[0031] The adjusted frequency and voltage are sent to the inverter or excitation device to guide the equipment output toward the target value.

[0032] For microgrid-level integral correction control, the historical accumulated values ​​of frequency deviation and voltage deviation are collected every ΔT2; the accumulated values ​​are then multiplied by the corresponding integral gain. Generating active power compensation and Generate reactive power compensation; superimpose the two compensation values ​​onto the device-level droop reference, and the output continuously converges to...

[0033] As a preferred embodiment of the microgrid collaborative scheduling and control method based on edge computing described in this invention, the power output scheduling includes, every ΔT3, node i sends a request to all its neighbors. Send the current power target;

[0034] After receiving the neighbor's target, node i, based on the difference between each neighbor's target and the target, assigns a weight w. ijCalculate the weighted average and multiply it by the step size factor β before adding it to the target.

[0035] Compare the maximum change in the target before and after the update, and consider the case where the change is less than the threshold ∈ and the number of execution rounds has reached K. max If the iteration stops, proceed to the next round of communication and update; otherwise, proceed to the next round of communication and update.

[0036] Each node converges to a consistent power target within a region and feeds the target back to the local controller as the basis for the next cycle.

[0037] After completing the three-level control, return to the short-time power prediction step, and use the sampled data at the new time to perform prediction, optimization, game theory and control again, maintaining closed-loop iteration.

[0038] As a preferred embodiment of the microgrid collaborative scheduling and control system based on edge computing described in this invention, it includes: a data acquisition module, a short-time power prediction module, an edge scheduling optimization module, and a multi-level control execution module;

[0039] The data acquisition module is used to collect operational data from each microgrid unit;

[0040] The short-time power prediction module is used to make short-time predictions of the active power demand and initial power generation of the microgrid unit.

[0041] The edge scheduling optimization module is used to obtain locally optimal scheduling parameters through a comprehensive evaluation of electricity cost and energy storage loss; and to construct a transaction game between units using the locally optimal parameters as initial values.

[0042] The multi-level control execution module is used to distribute the global optimal scheduling strategy to the edge nodes and execute it sequentially in a three-level closed-loop control within the microgrid unit.

[0043] A computer device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement a microgrid collaborative scheduling and control method based on edge computing.

[0044] A computer-readable storage medium having a computer program stored thereon, the computer program being executed by a processor to implement the steps of a microgrid collaborative scheduling and control method based on edge computing.

[0045] The beneficial effects of this invention are as follows: The microgrid collaborative scheduling and control method based on edge computing provided by this invention achieves high-precision short-term power prediction of less than 5% through multi-dimensional data acquisition and sequence feature learning of microgrid units; it deploys a genetic algorithm oriented towards power purchase cost and energy storage loss at edge nodes to obtain a fast-converging local optimization strategy; it constructs a non-cooperative game model with active power trading as the strategy variable, and supplements it with neighborhood consensus fusion to ensure that multiple microgrids reach Nash equilibrium under autonomous conditions; finally, it adopts a three-level closed-loop control of device-level droop, microgrid-level integration, and regional consensus to achieve rapid response to sudden disturbances and effective correction of steady-state errors. Attached Figure Description

[0046] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0047] Figure 1 The first embodiment of the present invention provides an overall flowchart of a microgrid collaborative scheduling and control method based on edge computing. Detailed Implementation

[0048] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.

[0049] Example 1, referring to Figure 1 As one embodiment of the present invention, a microgrid collaborative scheduling and control method based on edge computing is provided, comprising:

[0050] S1: Collect grid operation data through microgrid units, and perform short-time power prediction on microgrid units based on grid operation data to obtain load prediction values.

[0051] Furthermore, the LSTM algorithm is used to predict the generation load sequence for each microgrid unit in the next H steps.

[0052] Based on the processed data, the local controller predicts the short-term load forecast for each microgrid unit for the next H steps. The local edge node has performed short-term forecasts of active power demand and power generation within a future step of H, obtaining an initial predicted power vector:

[0053]

[0054] The control cycle is denoted as ΔT, and a short-time prediction is performed every ΔT. Hereinafter, the values ​​of each parameter in the k-th iteration are denoted by the superscript (k), and the real-time measurement value at time t is denoted by (t).

[0055] The short-term load forecast value This includes active power demand and initial power generation forecasts. The grid operation data includes distributed generation output, load power, and state of charge data of energy storage batteries; a scheduling and control cycle is defined as ΔT, and a short-time power forecast is performed every ΔT.

[0056] Sampling time t, sampling period Δt, filter window length N, prediction window length τ (number of historical input points).

[0057] Prediction duration H, local input vector:

[0058] X i (t)=[P DG,i (t),P L,i (t),SOC i (t)]

[0059] Predicted output Collect data from the i-th unit and record the output P of the distributed power source every Δt. DG,i (t), Load power P L,i (t), Energy Storage SCO i (t).

[0060] The moving average is applied to each sequence over a window N to obtain a smoothed sequence. To eliminate high-frequency jitter.

[0061] The smoothed data is linearly mapped to [0,1] to eliminate dimensional differences. Short-term prediction: using the past τ normalized input points as features, a trained LSTM model is used. Calculate the load forecast for the next H steps.

[0062] It should be noted that during the short-term power prediction phase, multi-dimensional operational data such as distributed power output, load power, and energy storage SOC are collected in real time by the microgrid unit. The LSTM algorithm is then used to learn the sequence features of data within a historical window length τ, accurately predicting the active power demand and initial power generation for the next H steps. This design can capture both rapid fluctuations in photovoltaic and wind power and reflect sudden changes in load trends, with the average prediction error controlled within 5%. Through periodic Δt-triggered moving averages (window N) and normalization processing, measurement noise and dimensional differences are further eliminated, providing high-quality input for subsequent scheduling and significantly improving the system's predictability and stability.

[0063] S2: Deploy a genetic algorithm on edge nodes to optimize load prediction values ​​and obtain scheduling strategy parameters.

[0064] Furthermore, the deployment of the genetic algorithm at the edge nodes includes processing the short-term load forecasts generated by each microgrid unit. Upload the data to the corresponding edge computing node; at the edge node, design a scheduling optimization objective to minimize the cost of purchasing electricity from the public grid and the losses from charging and discharging energy storage, and use a genetic algorithm to solve for the optimal solution.

[0065] Energy storage charge-discharge sequence ΔSOC i =[ΔSOC i (t+1),…,ΔSOC i (t+H)] Electricity price series [c buy (t+1),…,c buy (t+H)].

[0066] Energy storage cost factor γ, population size P, maximum number of generations G max Crossover probability P c Probability of mutation P m .

[0067] For each candidate solution x i =(P grid,i ,ΔSOC i ), calculate its total operating cost; public grid transaction cost, multiply the electricity price sequence by the transaction power and sum them up.

[0068] The lower the overall cost, the better the solution, as the energy storage loss γ is added together with the absolute values ​​of each charge and discharge step.

[0069] Initialize and randomly generate P vectors.

[0070] Fitness evaluation maps the total cost of the j-th vector to a fitness score for selection.

[0071] Selection is based on fitness, retaining the best individuals.

[0072] Crossover & mutation, according to probability P c ,P m Generate a new generation.

[0073] Repeat the evaluation-selection-crossover-mutation process until G is reached. max The process converges either in terms of generation or cost. The optimal scheduling parameters are then obtained. Including optimal and

[0074] The short-term load forecast values ​​are grouped to obtain initial candidate scheduling schemes. Each scheme includes the power trading sequence with the public grid in the future time period and the corresponding energy storage charging and discharging strategy.

[0075] For each candidate option, sum up its electricity price and energy storage usage costs over the entire forecast period.

[0076] Population updates are conducted through selection, crossover, and mutation operations with the aim of reducing overall costs.

[0077] When the maximum number of generations is reached, the current optimal scheduling strategy parameters are output. The edge nodes return the optimal strategy parameters to each microgrid unit as the initial input for the next cooperative game.

[0078] It should be noted that after deploying the genetic algorithm at the edge nodes, the system packages the short-term load forecasts of each microgrid unit into P candidate scheduling schemes. Each scheme includes a power trading sequence for the next H steps and an energy storage charging and discharging strategy. The algorithm aims to minimize the cost of purchasing electricity from the public grid and the energy storage loss (weighting factor γ), and optimizes the population size, maximum number of generations, crossover probability, and mutation probability P. m The synergistic effect of these factors enables the search for a near-global optimal solution within a dozen generations. Compared with traditional convex optimization, this method can reduce online computation time by about 30% and total scheduling cost by about 10%, while achieving distributed parallel computing without global information, thus meeting real-time requirements.

[0079] S3: Input the scheduling strategy parameters into the microgrid unit, construct a game model to conduct multi-microgrid collaborative game, and output the optimal scheduling strategy parameters.

[0080] Furthermore, the construction of the game model includes determining the participants and their numbers, where M = {1,2,…,N} is the set of microgrid units participating in the game, and N is the total number of microgrid units in the system.

[0081] Define a strategy space, where the strategy variable for each participant i is the trading power P. i : Represents the active power sold or purchased by the i-th microgrid to the public grid or a neighboring microgrid during the current scheduling cycle.

[0082] The strategy space is the interval [P] i,min ,P i,max ], where the lower bound P i,min With upper boundary P i,max This stems from the limitations of the unit's power generation and energy storage capacity.

[0083] A utility function is constructed, and for each unit i, its payoff function is defined as the electricity price revenue minus the transaction loss cost. A global power balance constraint is introduced and added to individual decisions in the later stages of the game through an additional penalty function, which is then uniformly corrected in subsequent hierarchical control.

[0084] Neighborhood structure and consensus weights, to accelerate information fusion, stipulate that each participant i has a neighborhood set. This indicates other units that can communicate directly with it.

[0085] For each pair (i,j) in the neighborhood, assign a weight w. ij And ensure that the sum of the weights of itself and all its neighbors is 1.

[0086] Define the optimal response mapping BR i For other fixed unit strategies {P} j :j≠i}, the optimal response BR of unit i i Defined as, within the permissible power range [P] i,min ,P i,max Within, select the option that enables utility U. i Reach the maximum point.

[0087] The mapping provides unit i with the optimal trading decision for its local environment. The set is optimal when all units' strategies are mutually optimal responses. This constitutes the Nash equilibrium of the game.

[0088] Benefit assessment, unit i in strategy P i The net income is determined by the real-time electricity price and output power, minus losses.

[0089] The updates are performed in turn. In the k-th round, the i-th unit has a fixed strategy for the other units. Adjust P automatically i To maximize net profit.

[0090] After this round of updates is completed, all units will be processed according to the following formula:

[0091]

[0092] Mixing neighborhood information makes the iteration converge faster.

[0093] when or k≥K max The timer ends, and the globally optimal trading power is output.

[0094] The multi-microgrid collaborative game includes each microgrid unit determining its local optimal P based on a local genetic algorithm. i (0) As initial values, after fixing the current strategies of other units, unit i calculates and updates its own optimal BR. i After the parameters are updated, neighborhood consensus fusion is performed, with each unit merging according to a preset weight w. ij A new policy estimate is obtained by weighting the temporary policies of the unit itself and its neighbors. Convergence is determined when the change in policy of all units is less than the convergence threshold δ, or when the maximum number of iterations K is reached. max If the condition is met, the process terminates. The output equilibrium strategy is then applied; after convergence, the strategy of each unit becomes the globally consistent trading power.

[0095] It should be noted that the local optimum output by the edge genetic algorithm is used as the initial strategy for each participant, constructing a non-cooperative game model with traded active power as the decision variable. The system defines the revenue as the electricity price revenue minus the loss cost represented by α, and incorporates global power balance constraints. Through alternating execution of round-based optimal response updates and neighborhood-weighted consensus, each unit can quickly approach Nash equilibrium while maintaining its own economic viability. This collaborative process can take into account both generation complementarity and load coordination between microgrids, further reducing regional electricity purchase and sale costs by 5%–8% compared to single-point optimization, and convergence is achieved in less than 15 game iterations.

[0096] S4: Distribute the optimal scheduling strategy parameters to the edge nodes and schedule power output through a multi-level control mechanism.

[0097] Furthermore, the multi-level control mechanism includes, as follows: As a reference target for control at all levels, three-level closed-loop regulation is implemented at the equipment level, microgrid level, and regional level.

[0098] Equipment-level droop control measures local instantaneous active power output P in real time. i (t), instantaneous reactive power output Q i (t), every control cycle ΔT1, the local controller reads the current output power, frequency, and voltage values, and compares them with the target value of active power. Target reactive power Compare them.

[0099] Global optimal active power target Global optimal reactive power target Droop coefficient (active) m i droop coefficient (reactive power) n i Reference frequency and voltage ω * V * .

[0100] The local controller periodically (period ΔT1) reads the current instantaneous active power P. i (t), reactive power Q i (t) and frequency ω i (t), voltage V i (t).

[0101] Calculate power deviation:

[0102]

[0103] Based on the droop control principle, fine adjustments are made to both the frequency and voltage:

[0104]

[0105] Will With V i cmd The data is sent to the inverter or excitation system, driving the equipment to output... Move closer. Note: Sag coefficient m i ,n i The larger the value, the more sensitive the response, but too large a value may cause oscillations; it is usually set to 0.01-0.05 times the rated power of the system.

[0106] Multiply the active power deviation by the active power droop coefficient m i Generate a new frequency reference. Multiply the reactive power deviation by the reactive power droop factor n. i This generates a new voltage reference.

[0107] The adjusted frequency and voltage are sent to the inverter or excitation device to guide the equipment output toward the target value.

[0108] Microgrid-level integral correction control is performed, and the historical accumulated values ​​of frequency deviation and voltage deviation are recorded every ΔT2. These accumulated values ​​are then multiplied by the corresponding integral gain. Generating active power compensation and Generate reactive power compensation. The two compensation values ​​are superimposed on the device-level droop reference, and the output continues to converge to...

[0109] Frequency deviation Δω i (t)=ω i (t)-ω * Voltage deviation ΔV i (t)=V i (t)-V * Integral gain (active) Integral gain (reactive power) The integral update period is ΔT2.

[0110] Within each secondary control cycle ΔT2, the frequency deviation and voltage deviation are accumulated respectively:

[0111] I ω,i (t+ΔT2)=I ω,i (t)+Δω i (t)ΔT2

[0112] I V,i (t+ΔT2)=I V,i (t)+ΔV i (t)ΔT2

[0113] Generate active / reactive power compensation increments:

[0114]

[0115] Add the compensation increment to the droop command:

[0116]

[0117] The internal target of the local controller is updated to gradually eliminate steady-state bias. Second-order integration can eliminate steady-state bias caused by device-level droop; the integral gain is generally small to avoid integral storms.

[0118] The power output scheduling includes, every ΔT3, node i sends a message to all its neighbors. Send the current power target.

[0119] After receiving the neighbor's target, node i, based on the difference between each neighbor's target and the target, assigns a weight w. ij Calculate the weighted average and multiply it by the step size factor β before adding it to the target.

[0120] Compare the maximum change in the target before and after the update, and consider the case where the change is less than the threshold ∈ and the number of execution rounds has reached K. max If the iteration stops, proceed to the next round of communication and update.

[0121] Each node converges to a consistent power target within a region and feeds the target back to the local controller as the basis for the next cycle.

[0122] Local current power target Regional Neighbor Sets Consensus weight {w ij},satisfy Consensus step size β, update period ΔT3, convergence threshold ∈, maximum number of iterations K max .

[0123] During each consensus cycle ΔT3, each edge node i sends a consensus message to all its neighbors within its region. Send current power target

[0124] After collecting neighbor targets, perform a weighted update:

[0125]

[0126] Check for convergence; if the maximum change... Or the number of iterations k reaches K max If the iteration stops, then stop.

[0127] The final output is a region-wide consistent power target. And distribute it to the local controller as a new Input, proceed to the next control cycle.

[0128] Regional consensus can eliminate local inconsistencies among multiple nodes. Too large a beta can cause oscillations, while too small a beta can lead to slow convergence; it is generally taken as 0.1-0.5.

[0129] After completing the three-level control, return to the short-time power prediction step, and use the sampled data at the new time to perform prediction, optimization, game theory and control again, maintaining closed-loop iteration.

[0130] It should be noted that the globally optimal trading power and the corresponding reactive power target are used as closed-loop references at the device level, microgrid level, and regional level, respectively. Device-level droop control employs active power droop coefficients and reactive power droop coefficients to fine-tune frequency and voltage in real time, enabling the instantaneous output to quickly approach the target. Microgrid-level integral correction uses integral gain to eliminate steady-state deviations. The regional-level consensus mechanism, in conjunction with the step size β and the convergence threshold ε, achieves consistency of multi-node targets. The three-layer control cycle satisfies ΔT1 < ΔT2 < ΔT3, enabling the system to respond to most disturbances within 50ms, maintaining frequency / voltage deviations within ±0.02 pu, significantly improving the microgrid's disturbance rejection capability and operational reliability.

[0131] Example 2 is an embodiment of the present invention, which provides a microgrid collaborative scheduling and control method based on edge computing. In order to verify the beneficial effects of the present invention, scientific demonstration is carried out through economic benefit calculation and simulation experiments.

[0132] First, five independent microgrid units (denoted as MGA to MGE) were deployed within the "Harbor New Energy Experimental Park." Each unit consists of a 250kW photovoltaic array, a 120kW wind turbine, a 200kWh lithium battery energy storage cluster, and a 150kW adjustable load. The experimental platform was equipped with an IEC61850MMS data acquisition device, with a sampling period Δt set to 1 minute; a moving average window N=5, a historical window length τ=30 (i.e., a 30-minute sequence as LSTM input), and a prediction step size H=10 (corresponding to a 10-minute short-term scheduling window). NVIDIA Jetson OrinNX was deployed on the edge nodes; the local genetic algorithm population size P=40, with a maximum generation G. max =15, crossover probability P c =0.85, mutation probability P m =0.1, energy storage cost weight γ = 0.025¥ / (kWh·cycle). To fully evaluate the economics and control performance of the proposed solution, a control object MGRef was set up, adopting the traditional "central EMS + static MPC" scheme: short-term load is predicted through single-point ARIMA, and scheduling parameters are issued by the central server every 15 minutes, without local game theory and multi-level closed loops. All units were continuously tested from 08:00 to 18:00 on April 1, 2025. The system recorded seven key indicators: ① short-term prediction error, ② unit electricity purchase cost, ③ energy storage cycle expenditure, ④ renewable energy abandonment rate, ⑤ frequency / voltage deviation, ⑥ number of game iteration rounds, and ⑦ online solution delay. Equipment-level droop coefficient m i =0.03, n i =0.04; Microgrid-level integral gain The regional consensus step size β = 0.25, and the convergence threshold ε = 0.01 pu. This process follows a fast-medium-slow rhythm: ΔT1 = 100 ms (device level), ΔT2 = 2 s (microgrid level), and ΔT3 = 6 s (regional level) to ensure disturbance response and steady-state accuracy. Before the experiment, an LSTM network (2 × 128 hidden layers, 40 Adam iterations, learning rate 0.001) was trained using the measured load curve from the same period of 202503, and the fitness function of the genetic algorithm was calibrated offline. Then, online closed-loop operation began. The entire process was recorded by the OPCUA data bus and the Prometheus monitoring system, ensuring consistent statistical standards and traceability.

[0133] The table data shows that the proposed solution demonstrated comparative advantages in all five experimental microgrid units. Firstly, in terms of prediction accuracy, the LSTMΔtτ structure, combined with moving average and normalization, controlled the mean square percentage error of short-term loads within the range of 3.8%–4.5%, a significant reduction of approximately 56% compared to the 9.6% error of the control ARIMA solution. This improved prediction accuracy allows the genetic algorithm to search the strategy space with more accurate load priors, reducing scheduling uncertainty caused by prediction biases and thus lowering subsequent electricity purchase costs.

[0134] In terms of economic indicators, the average electricity purchase price per unit of this invention is lower than ¥0.533 / kWh, while the control scheme is ¥0.591 / kWh, a reduction of approximately 9.9%. This result stems from the localized execution of the genetic algorithm at the edge nodes, which dynamically adjusts the three methods of "electricity purchase, energy storage, and curtailment of solar / wind power" through joint optimization of real-time electricity price curves and energy storage health factors, smoothing out peak electricity purchases. Simultaneously, the introduction of an energy storage cycle cost weight γ suppresses frequent switching between overcharging and discharging, protecting battery life while achieving economic benefits. The "Energy Storage Cycle Expenditure" column in the table further verifies this design: the annualized cost per unit of this invention is approximately ¥48, 14% lower than the control scheme.

[0135] The significant reduction in renewable energy curtailment rate (1.8%–2.2% vs 6.4%) reflects the synergistic gains of the game-theoretic consensus mechanism. The twin game model, while maintaining the local economic viability of each unit, utilizes neighborhood weights w... ij The step size β facilitates "mutual support between positive and negative power"—units with surplus output reduce wind and solar curtailment at the local level, while units with high load purchase surplus power at a lower marginal electricity price. The regional power balance constraint penalty function is automatically satisfied, achieving Pareto improvement from both economic and physical perspectives.

[0136] Example 3, an embodiment of the present invention, provides a microgrid collaborative scheduling and control system based on edge computing, including a data acquisition module, a short-time power prediction module, an edge scheduling optimization module, and a multi-level control execution module.

[0137] The data acquisition module is used to collect operational data from each microgrid unit.

[0138] The short-time power prediction module is used to make short-time predictions of the active power demand and initial power generation of the microgrid unit.

[0139] The edge scheduling optimization module is used to obtain locally optimal scheduling parameters through a comprehensive evaluation of electricity cost and energy storage loss; and to construct a transaction game between units using the locally optimal parameters as initial values.

[0140] The multi-level control execution module is used to distribute the global optimal scheduling strategy to the edge nodes and execute it sequentially in a three-level closed-loop control within the microgrid unit.

[0141] If a function is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0142] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0143] More specific examples (a non-exhaustive list) of computer-readable media include: electrical connections (electronic devices) having one or more wires, portable computer disk drives (magnetic devices), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Furthermore, computer-readable media can even be paper or other suitable media on which programs can be printed, because programs can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in computer memory.

[0144] It should be understood that various parts of the present invention can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented using software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using a combination of any of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc. It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

[0145] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A microgrid collaborative scheduling and control method based on edge computing, characterized in that, include: By collecting grid operation data through microgrid units, and performing short-time power prediction on microgrid units based on grid operation data, load prediction values ​​are obtained. Deploy genetic algorithms at edge nodes to optimize load forecasts and obtain scheduling strategy parameters; Input the scheduling strategy parameters into the microgrid unit, construct a game model to conduct multi-microgrid collaborative game, and output the optimal scheduling strategy parameters. The optimal scheduling strategy parameters are sent to the edge nodes, and power output is scheduled through a multi-level control mechanism.

2. The microgrid collaborative scheduling and control method based on edge computing as described in claim 1, characterized in that: The short-term power forecasting includes preprocessing grid operation data collected by the microgrid units, and based on the processed data, forecasting the short-term load for each microgrid unit in the next H steps via a local controller. Let a scheduling control cycle be ΔT, and perform a short-time power prediction once every ΔT; The short-term load forecast value This includes active power demand and initial forecasted power generation; grid operation data includes distributed generation output, load power, and state of charge data of energy storage batteries.

3. The microgrid collaborative scheduling and control method based on edge computing as described in claim 2, characterized in that: The deployment of genetic algorithms at edge nodes includes processing short-term load forecasts generated by each microgrid unit. Upload the data to the corresponding edge computing node; at the edge node, design a scheduling optimization objective to minimize the cost of purchasing electricity from the public grid and the losses from charging and discharging energy storage, and use a genetic algorithm to solve for the optimal solution. The short-term load forecast values ​​are grouped to obtain initial candidate scheduling schemes. Each scheme includes the power trading sequence with the public grid in the future time period and the corresponding energy storage charging and discharging strategy. For each candidate option, sum up its electricity price and energy storage usage costs over the entire forecast period; Population updates are conducted through selection, crossover, and mutation operations with the aim of reducing overall costs. When the maximum number of generations is reached, the current optimal scheduling strategy parameters are output; the edge nodes return the optimal strategy parameters to each microgrid unit as the initial input for the next cooperative game.

4. The microgrid collaborative scheduling and control method based on edge computing as described in claim 3, characterized in that: The construction of the game model includes determining the participants and their number, where M = {1,2,…,N} is the set of microgrid units participating in the game, and N is the total number of microgrid units in the system. Define a strategy space, where the strategy variable for each participant i is the trading power P. i : Represents the active power sold or purchased by the i-th microgrid to the public grid or a neighboring microgrid during the current scheduling cycle; Strategy space interval [P] i,min ,P i,max ], where the lower bound P i,min With upper boundary P i,max The limitations stem from the unit's power generation and energy storage capacity; Construct a utility function, and for each unit i, define its payment function as electricity price revenue minus transaction loss cost; introduce a global power balance constraint, which is added to individual decisions in the later stage of the game through an additional penalty function, and uniformly corrected in subsequent hierarchical control; Neighborhood structure and consensus weights, to accelerate information fusion, stipulate that each participant i has a neighborhood set. Other units that communicate directly; For each pair (i,j) in the neighborhood, assign a weight w. ij And ensure that the sum of the weights of itself and all its neighbors is 1; Define the optimal response mapping BR i For other fixed unit strategies {P} j :j≠i}, the optimal response BR of unit i i Defined as, within the permissible power range [P] i,min ,P i,max Within, select the option that enables utility U. i Reaching the maximum point; The mapping provides unit i with the optimal trading decision for its local environment; when the strategies of all units are mutually optimal responses, the set... This constitutes the Nash equilibrium of the game.

5. The microgrid collaborative scheduling and control method based on edge computing as described in claim 4, characterized in that: The multi-microgrid collaborative game includes each microgrid unit determining its local optimal value based on an edge genetic algorithm. As an initial value; after fixing the current strategies of other units, unit i calculates and updates its own optimal BR. i After the parameters are updated, neighborhood consensus fusion is performed, and each unit is fused according to the preset weight w. ij A new policy estimate is obtained by weighting the temporary policies of the unit itself and its neighbors; convergence is determined by whether the change in the policies of all units is less than the convergence threshold δ, or by reaching the maximum number of iterations K. max If the convergence fails, the process terminates; the output equilibrium strategy is: after convergence, the strategy of each unit is the globally consistent trading power.

6. The microgrid collaborative scheduling and control method based on edge computing as described in claim 5, characterized in that: The multi-level control mechanism includes, with As a reference target for control at all levels, three-level closed-loop regulation is implemented at the equipment level, microgrid level, and regional level respectively; Equipment-level droop control measures local instantaneous active power output P in real time. i (t), instantaneous reactive power output Q i (t), every control cycle ΔT1, the local controller reads the current output power, frequency, and voltage values, and compares them with the target value of active power. Target reactive power Compare; Multiply the active power deviation by the active power droop coefficient m i Generate a new frequency reference; multiply the reactive power deviation by the reactive power droop factor n. i Generate a new voltage reference; The adjusted frequency and voltage are sent to the inverter or excitation device to guide the equipment output toward the target value. For microgrid-level integral correction control, the historical accumulated values ​​of frequency deviation and voltage deviation are collected every ΔT2; the accumulated values ​​are then multiplied by the corresponding integral gain. Generating active power compensation and Generate reactive power compensation; superimpose the two compensation values ​​onto the device-level droop reference, and the output continuously converges to...

7. The microgrid collaborative scheduling and control method based on edge computing as described in claim 6, characterized in that: The power output scheduling includes, every ΔT3, node i sends a message to all its neighbors. Send the current power target; After receiving the neighbor's target, node i, based on the difference between each neighbor's target and the target, assigns a weight w. ij Calculate the weighted average and multiply it by the step size factor β before adding it to the target. Compare the maximum change in the target before and after the update, and consider the case where the change is less than the threshold ∈ and the number of execution rounds has reached K. max Stop iterating when the time comes; Otherwise, proceed to the next round of communication and updates; Each node converges to a consistent power target within a region and feeds the target back to the local controller as the basis for the next cycle. After completing the three-level control, return to the short-time power prediction step, and use the sampled data at the new time to perform prediction, optimization, game theory and control again, maintaining closed-loop iteration.

8. A system employing the microgrid collaborative scheduling and control method based on edge computing as described in any one of claims 1 to 7, characterized in that: It includes a data acquisition module, a short-term power prediction module, an edge scheduling optimization module, and a multi-level control execution module; The data acquisition module is used to collect operational data from each microgrid unit; The short-time power prediction module is used to make short-time predictions of the active power demand and initial power generation of the microgrid unit. The edge scheduling optimization module is used to obtain locally optimal scheduling parameters through a comprehensive evaluation of electricity cost and energy storage loss; and to construct a transaction game between units using the locally optimal parameters as initial values. The multi-level control execution module is used to distribute the global optimal scheduling strategy to the edge nodes and execute it sequentially in a three-level closed-loop control within the microgrid unit.

9. A computer device, characterized in that, The system includes a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the microgrid collaborative scheduling and control method based on edge computing as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, It stores a computer program, characterized in that when the computer program is executed by a processor, it implements the steps of the microgrid collaborative scheduling and control method based on edge computing as described in any one of claims 1 to 7.

Images