A micro-grid source network load storage rolling optimization economic dispatch method and system

Abstract

The application discloses a micro-grid source network load storage rolling optimization economic dispatching method and belongs to the technical field of intelligent micro-grid economic operation optimization. The application realizes the cooperation of the day-ahead global optimization and the day-in real-time correction, and improves the economy, safety and self-adaptability of the micro-grid operation.

Description

Technical Field

[0001] This application relates to the field of economic operation optimization and energy management technology of smart microgrids, specifically to a method and system for rolling optimization economic dispatch of microgrid sources, grids, loads and storage. Background Technology

[0002] With the large-scale integration of distributed photovoltaic systems, energy storage systems, charging piles, and adjustable loads on the industrial and commercial user side, the operational objectives of microgrids have expanded from simply peak shaving and valley filling to collaborative optimization of multiple entities including power sources, grids, loads, and storage. Existing energy management systems typically need to comprehensively balance factors such as electricity purchase costs, demand charges, energy storage lifespan losses, transformer safety boundaries, continuity of power supply to critical loads, and renewable energy consumption. This places higher demands on the globality and adaptability of dispatch strategies.

[0003] Currently, there are three main implementation methods for existing microgrid dispatching schemes: The first type is the fixed-rule control method. A typical approach is to charge during off-peak hours and discharge during peak hours, prioritizing energy storage charging when photovoltaic power is surplus, or triggering peak-shaving control when grid-connected power exceeds a threshold. This method is simple to implement and has low hardware requirements, but it cannot uniformly incorporate real-time price fluctuations, weather changes, load fluctuations, and equipment availability into the calculation, thus limiting the improvement in economic efficiency.

[0004] The second type is the day-ahead static scheduling method. This method first generates a 96-point daily plan curve based on the day-ahead forecast, and then executes the original plan during operation. This method is an improvement over rule-based control, but it cannot be corrected in time when there are sudden weather changes, fluctuations caused by solar clouds, temporary load changes, equipment failures, or price updates. If the forecast deviates significantly, the plan will quickly become disconnected from the actual operation.

[0005] The third category is local object optimization. Some solutions only optimize energy storage separately, minimizing electricity costs or suppressing peak demand, while dispersing the processing of transformer capacity, critical load power supply constraints, energy storage lifespan reduction, photovoltaic curtailment penalties, and load comfort penalties. This results in inconsistent global constraints and unclear engineering boundaries, often leading to only local optima. It cannot simultaneously address the coupling problem of multiple objects such as grid connection point power, transformer capacity, charging pile group power allocation, critical load priority, and photovoltaic curtailment.

[0006] From the commonalities of existing technical solutions, the data acquisition module typically first reads photovoltaic power, energy storage state of charge, load power, and electricity price. Then, empirical models or simplified prediction methods are used to obtain future load and photovoltaic trends. Subsequently, energy storage charge and discharge curves are generated through fixed rules or one-time optimization, and finally executed by the field controller. Most existing systems do not adequately consider the closed-loop process of "execution deviation - economic loss - parameter correction." There is a lack of continuous feedback and learning mechanisms between the prediction module, optimization module, and execution module. Many solutions separate demand cost, energy storage lifetime cost, curtailment penalty, comfort constraints, and engineering safety boundaries. In actual implementation, parameters are adjusted manually based on experience, making it difficult to form a unified, reusable, and interpretable optimization framework. As a result, the long-term operating performance gradually deviates from the optimal level.

[0007] In summary, existing microgrid dispatching technologies suffer from limited economic improvement, lack of online correction capabilities, insufficient multi-object coupling processing capabilities, lack of a unified optimization framework, and lack of feedback learning mechanisms. There is an urgent need for an economical dispatching scheme that can achieve integrated collaborative optimization of source, grid, load, and storage and possess self-learning capabilities. Summary of the Invention

[0008] In view of this, the main objective of this application is to provide a method and system for rolling optimization of economic dispatch of microgrid sources, grids, loads and storage, in order to at least partially solve the above-mentioned technical problems.

[0009] To address the shortcomings of existing technologies, this application provides an integrated economic dispatch system and method for intelligent microgrid power generation, grid, load, and storage, aiming to solve the following technical problems: the inability of fixed peak-valley rule control to incorporate real-time prices, actual photovoltaic fluctuations, and sudden load changes into unified decision-making; the lack of online correction capability and disconnect between planned and actual operation in day-ahead static scheduling; the inability of single-storage local optimization to handle multi-object coupling; the problems of separating demand cost, energy storage lifetime cost, curtailment penalty, comfort constraints, and engineering safety boundaries, and relying on human experience; and the lack of a post-operation feedback learning mechanism in existing systems, which cannot automatically correct parameters based on prediction errors and actual economic losses.

[0010] This application establishes a unified scheduling framework for objects such as photovoltaics, energy storage, charging piles, general loads, controllable loads, grid connection points, and transformers in industrial and commercial microgrids. It includes day-ahead forecasting, intraday 15-minute rolling optimization, and closed-loop feedback. This application does not simply generate a static schedule once a day, nor does it rely solely on fixed peak-valley charging and discharging rules. Instead, it solves for a 96-point plan for the entire day based on photovoltaics, load, electricity price / time-of-use, and demand risk forecasts during the day-ahead phase. Intraday, it introduces the latest prices, weather conditions, and equipment availability status every 15 minutes, performs re-optimization on the remaining time domain, and only issues the next control step strategy. This continuously approaches the optimal operating economy using model predictive control principles while ensuring that safety boundaries are not exceeded.

[0011] To achieve the above-mentioned objectives, this application provides the following technical solution: A rolling optimization economic dispatch method for integrated generation, grid, load, and storage in microgrids includes the following steps: S1: Collect and preprocess multi-source operation data of the microgrid; S2: Based on historical operational data and the latest external information, call the day-ahead prediction model to output multi-period prediction results covering the entire scheduling cycle of the next day; S3: Discretize the next day's scheduling cycle into multiple time periods, each time period being 15 minutes long. Define system state variables and control variables. The state variables include at least the energy storage charge state, and the control variables include at least the energy storage charging and discharging power, grid-connected power, controllable load adjustment, and power allocated to each charging pile. S4: Construct a day-ahead optimization objective function with the goal of minimizing total operating cost; S5: Construct unified engineering constraints; S6: Solve the day-ahead optimization objective function under the constraints and output the multi-period baseline plan; S7: Introduce the latest real-time information at the current moment of the day, and perform rolling optimization for the remaining time domain using the current measured state as the initial condition; S8: Extract the control quantity for the current time period from the rolling optimization results, issue and execute it; S9: Collect execution results and assess economic losses; S10: Perform closed-loop parameter correction based on prediction error and economic loss, and proceed to the next cycle.

[0012] In one possible implementation, the multi-source operational data includes at least photovoltaic output data, energy storage system state of charge, charging and discharging power, available capacity and charging and discharging efficiency data, charging pile power and connection status data, park basic load and controllable load data, time-of-use electricity price or spot electricity price data, meteorological data, historical demand data, and key equipment availability status data; the preprocessing includes timestamp alignment, missing value completion, outlier removal and unit standardization to form a data sequence with a uniform time interval.

[0013] In one possible implementation, the day-ahead forecasting model includes a photovoltaic output forecasting submodule, a load forecasting submodule, an electricity price forecasting or price mapping submodule, an equipment available capacity forecasting submodule, and a demand risk forecasting submodule; the demand risk forecasting submodule estimates the risk level of triggering demand overruns in future periods based on historical maximum demand, future load levels, and grid connection point constraints.

[0014] In one possible implementation, the total operating cost includes the cost of purchasing and selling electricity for each time period, the demand electricity cost calculated by multiplying the maximum grid-connected power and the demand unit price within the settlement cycle, the energy storage equivalent cycle life loss cost calculated by multiplying the charge and discharge energy flux and the unit cycle life cost, the cost of photovoltaic curtailment or limiting, the cost of controllable load adjustment, and the cost of penalties for violating critical safety boundaries.

[0015] In one possible implementation, the unified engineering constraints include at least the dynamic and range constraints of the energy storage state of charge, the mutual exclusion constraints of energy storage charging and discharging, the power balance constraints, the available output constraints of photovoltaics, the power constraints of the grid connection point, the transformer capacity constraints, the power ramping constraints, the upper limit constraints of demand, the constraints of ensuring the energy supply of critical loads, and the power allocation constraints of the charging pile group.

[0016] In one possible implementation, the rolling optimization is performed for the remaining time domain, using the measured state at the current moment as the initial condition. A tracking penalty term for the day-ahead baseline plan is introduced into the objective function. The tracking penalty term is used to constrain the deviation between the current rolling decision and the baseline plan, so as to avoid large oscillations in the rolling strategy relative to the day-ahead baseline plan.

[0017] In one possible implementation, the step of extracting and executing the control quantity for the current period from the rolling optimization results specifically involves extracting only the control quantity for the current period from the rolling optimization results as an execution instruction and sending it to the energy management system or edge controller. After the period ends, the system re-enters the next round of rolling optimization, rather than executing all remaining period strategies at once.

[0018] In one possible implementation, the acquisition of execution results and assessment of economic losses includes: obtaining the actual execution data of the current control step, calculating the photovoltaic prediction error, the load prediction error, and the deviation in electricity purchase cost caused by the errors, and defining the difference between the actual operating cost and the optimized cost under ideal prediction conditions as the economic loss.

[0019] In one possible implementation, the closed-loop correction of the execution parameters includes: adaptively correcting the energy storage degradation coefficient, demand weight, curtailment penalty coefficient, and controllable load penalty coefficient based on the prediction error and economic loss, and resolving the optimization model with the updated parameters in the next rolling cycle, forming a closed-loop learning mechanism of prediction error-economic loss-parameter correction-re-optimization.

[0020] In one possible implementation, a microgrid integrated power generation, grid, load, and storage economic dispatch system is used to perform the method described in any of the above embodiments, including: The data acquisition layer is used to collect multi-source operational data of the microgrid; The forecasting layer is used to generate photovoltaic output forecasts, load forecasts, electricity price forecasts, equipment availability capacity forecasts, and demand risk forecasts based on multi-source operating data. The optimization decision layer is used to construct a day-ahead optimization model and an intraday rolling optimization model based on the output of the prediction layer, with the goal of minimizing the total operating cost, and solve for the baseline plan and the strategy for the next control step. The intraday rolling optimization model uses the measured state at the current moment as the initial condition, performs rolling optimization on the remaining time domain, and outputs only the strategy for the current control step. The execution and closed-loop layer is used to send the next control step strategy to the field equipment for execution, and to perform feedback evaluation and adaptive parameter correction based on execution deviation and economic loss.

[0021] Based on the above technical solution, it can be seen that the microgrid source-grid-load-storage rolling optimization economic dispatch method and system of this application has at least one of the following beneficial effects compared with the prior art: 1. By incorporating the costs of electricity purchase and sale, demand electricity costs, energy storage equivalent cycle life loss costs, photovoltaic curtailment penalty costs, controllable load adjustment costs, and penalty costs for violating critical safety boundaries into the same objective function, and by introducing the latest real-time information in 15-minute intervals throughout the day to perform rolling optimization on the remaining time domain, the synergy between day-ahead global optimization and intraday real-time correction is achieved. This overcomes the technical defects of existing technologies, such as the limited economic improvement of fixed rule control and the inability to correct day-ahead static scheduling online. It can still maintain a near-optimal comprehensive operating cost under conditions of electricity price fluctuations, photovoltaic fluctuations, and load abrupt changes.

[0022] 2. By constructing unified engineering constraints, including at least dynamic and range constraints on the state of charge of energy storage, mutual exclusion constraints on energy storage charging and discharging, power balance constraints, constraints on the available output of photovoltaics, power constraints at the grid connection point, transformer capacity constraints, power ramping constraints, upper limit constraints on demand, constraints on the supply guarantee of critical loads, and power allocation constraints for charging pile groups, and setting up a data acquisition layer, prediction layer, optimization decision layer, and execution and closed-loop layer in the system, we can achieve multi-object collaborative scheduling for photovoltaics, energy storage, charging piles, controllable loads, and grid connection interfaces. This overcomes the technical defects of existing technologies, such as the inability of local object optimization to handle multi-object coupling simultaneously and the reliance on manual experience for decomposition processing, and effectively avoids risks such as SOC overrun, transformer overload, grid connection power exceeding limits, and critical load power failure.

[0023] 3. By collecting execution results to assess economic losses, the energy storage degradation coefficient, demand weight, curtailment penalty coefficient, and controllable load penalty coefficient are adaptively corrected based on prediction errors and economic losses. This forms a closed-loop learning mechanism of prediction error - economic loss - parameter correction - re-optimization, which overcomes the technical defects of existing technologies, such as the lack of a continuous feedback learning mechanism between the prediction module, optimization module, and execution module, and the gradual deviation of long-term operating results from the optimal. This makes the model closer to the real operating characteristics, and its long-term operating performance is better than that of a purely static model. Attached Figure Description

[0024] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the embodiments will be briefly introduced below. 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.

[0025] Figure 1 This is a flowchart illustrating a microgrid source-grid-load-storage rolling optimization economic dispatch method according to the present invention. Detailed Implementation

[0026] To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below with reference to specific embodiments and accompanying drawings.

[0027] The terminology used in this invention is for the purpose of describing particular embodiments only and is not intended to limit the embodiments of the invention. The singular forms “a,” “the,” and “the” as used in the embodiments of the invention and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise.

[0028] The present invention will be further illustrated below through specific embodiments. It should be noted that the following embodiments are merely illustrative and not intended to limit the scope of the invention. All other embodiments obtained by those skilled in the art based on the embodiments shown below without inventive effort are within the scope of protection of the embodiments of the present invention.

[0029] The microgrid power generation, grid, load and storage integrated economic dispatch system provided by the present invention includes four functional layers: data acquisition layer, prediction layer, optimization decision layer and execution and closed loop layer.

[0030] The data acquisition layer accesses field data through industrial gateways, edge controllers, or energy management system interfaces. The accessed data includes, but is not limited to, the following five categories: (1) Photovoltaic side data: inverter active power, available output, and power limiting status; (2) Energy storage side data: charging power, discharging power, state of charge, available capacity, charging and discharging efficiency, and available status; (3) Load-side data: total building load, sub-circuit load, critical production line load, controllable load, charging pile load; (4) Grid-side data: grid connection point power, purchased electricity, grid-connected electricity, transformer capacity occupancy, and demand value; (5) External information: time-of-use electricity price, peak and off-peak price, demand billing parameters, weather forecast, irradiance, temperature, cloud cover, wind speed, etc.

[0031] To facilitate a unified solution, all data are preferably aligned to the 15-minute boundary; for raw data at the second or minute level, the average, maximum, or cumulative values ​​can be converted according to the sampling window before entering the prediction and optimization module.

[0032] The prediction layer includes at least the following five sub-modules: (1) Photovoltaic prediction submodule: Based on irradiance, cloud cover, weather type, historical photovoltaic power and equipment availability, obtain the photovoltaic power output prediction sequence for the next 96 15-minute periods; (2) Load forecasting submodule: Based on historical load, production / operation patterns, weekday and holiday characteristics, temperature and humidity and operating status data, the load forecasting sequence for the next 96 15-minute periods is obtained; (3) Electricity price prediction or price mapping submodule: When the electricity price is a time-of-use price, it is directly mapped to each 15-minute time period; when there is an intraday price update or market price signal, the price of the remaining time domain is predicted or corrected. (4) Available capacity prediction submodule: Based on the available energy storage capacity, equipment maintenance status, number of charging piles connected and critical load demand, the dispatchable capacity boundary of each device in the future time domain is obtained; (5) Demand risk prediction submodule: Based on the historical maximum demand, future load level and grid connection point constraints, estimate the risk level of demand exceeding limits in future periods.

[0033] The output of the prediction layer is not the final control value, but rather the input parameter for the subsequent optimization decision layer.

[0034] The optimization decision layer includes a day-ahead optimization module and an intraday rolling optimization module. The day-ahead optimization module is used to build a day-ahead optimization model and solve for the baseline plan for the next 96 time periods based on the next day's forecast data output by the forecast layer at a fixed time each day (e.g., 4 pm). The intraday rolling optimization module is used to re-optimize the remaining time domain based on the latest real-time information at the current time in 15-minute cycles and only outputs the strategy for the current control step.

[0035] The execution and closed-loop layer is used to distribute the optimization results to the energy management system and edge controllers, where they are executed by field devices. It also provides feedback evaluation and parameter write-back on execution deviations, economic losses, and constraint violations. This layer includes a feedback learning module to acquire the errors between actual photovoltaic output and predicted values, the errors between actual load and predicted values, and the corresponding deviations in electricity purchase costs. Based on the accumulated errors and economic losses, it corrects parameters in the optimization decision layer, such as the energy storage degradation coefficient, demand weight, and curtailment penalty coefficient.

[0036] Combination Figure 1 As shown, the method of the present invention specifically includes the following steps: S1: Collect and preprocess multi-source operational data The data collected should include at least: photovoltaic power output data, energy storage system status data, charging pile power request and connection status, park basic load data, time-of-use electricity price or spot electricity price data, meteorological data, and key equipment availability status data. Among them, energy storage status data should at least include state of charge, rated capacity, maximum charging and discharging power, and available capacity attenuation coefficient; meteorological data should at least include irradiance, ambient temperature, cloud cover, and weather type.

[0037] All types of data are uniformly processed by timestamp alignment, missing value completion, outlier removal, and unit standardization to form a data sequence with a 15-minute time interval.

[0038] S2: Call the day-ahead forecasting model and output the day-ahead forecasting results. Based on historical operating data and the latest external information, the photovoltaic (PV) forecasting model, load forecasting model, electricity price forecasting model, and equipment availability capacity forecasting model are invoked respectively. Specifically, the PV forecasting model estimates the generating capacity for each time period by combining irradiance and cloud cover; the load forecasting model estimates the base load and controllable load demand; the electricity price forecasting model estimates the electricity purchase cost for each time period; the equipment availability capacity forecasting model estimates the upper limit of available power for energy storage systems and charging piles in future time periods; and the demand risk forecasting model estimates the risk level of triggering demand exceedance based on historical maximum demand, future load levels, and grid connection constraints. The forecast output covers 96 discrete time periods from 00:00 to 24:00 the following day.

[0039] S3: Establish day-ahead scheduling time windows and define state variables. The next day's scheduling cycle is discretized into 96 time periods from t=1 to 96, each time period having a length Δt=15 minutes. The energy storage state of charge is defined as SOC(t), the energy storage charging power as Pch(t), the energy storage discharging power as Pdis(t), the grid-connected power as Pigrid(t), the photovoltaic curtailed power as Pcurt(t), the controllable load adjustment as Pflex(t), and the power allocated to the i-th charging pile in time period t as Pev,i(t). Grid-connected power purchase is positive, and grid-connected power sales are negative.

[0040] S4: Construct the current day-ahead optimization objective function With the goal of minimizing total operating cost, the objective function can be written as: Jday = Jgrid + Jdemand + Jdeg + Jcurt + Jflex + Jsafe Wherein, Jgrid represents the cost of purchasing and selling electricity in each time period; Jdemand represents the demand electricity cost calculated based on the peak grid-connected power of the settlement cycle, which can be calculated by multiplying max(Pgrid+(t)) by the demand unit price; Jdeg represents the equivalent cycle life loss cost of energy storage, which can be calculated by multiplying the charging and discharging energy flux by the unit cycle life cost; Jcurt represents the cost of photovoltaic curtailment or power limiting penalties; Jflex represents the cost of controllable load regulation, comfort impact, or production line disturbance; and Jsafe represents the penalty item imposed when critical safety boundaries are violated.

[0041] S5: Constructing unified engineering constraints Specifically, it includes at least the following constraints: (a) SOC dynamic constraint: SOC(t+1) = SOC(t) + ηch·Pch(t)·Δt / Enom - Pdis(t)·Δt / (ηdis·Enom); (b) SOC range constraint: SOCmin≤SOC(t)≤SOCmax; (c) Charge and discharge mutual exclusion constraints: Pch(t)≤uch(t)·Pch,max, Pdis(t)≤udis(t)·Pdis,max, and uch(t)+udis(t)≤1; (d) Power balance constraints: Pgrid(t)+Ppv,use(t)+Pdis(t)=Pload,base(t)-Pflex(t)+Pev(t)+Pch(t)+Paux(t); (e) Constraints on photovoltaic availability: 0≤Ppv,use(t)≤Ppv,forecast(t), and Pcurt(t)=Ppv,forecast(t)-Ppv,use(t); (f) Power constraint at grid connection point: Pgrid,min≤Pgrid(t)≤Pgrid,max; (g) Transformer capacity constraint: |Pgrid(t)+Pother(t)|≤Ptrans,max; (h) Power ramping constraints: |Pgrid(t)-Pgrid(t-1)|≤Rgrid, |Pch(t)-Pch(t-1)|≤Rbat, |Pdis(t)-Pdis(t-1)|≤Rbat; (i) Demand upper limit constraint: Pgrid+(t)≤Pdem,limit; (j) Critical load guarantee constraints: The preset power supply priority and minimum power supply capacity must be met during critical load periods; (k) Charging pile allocation constraints: Pev(t)=ΣPev,i(t), and 0≤Pev,i(t)≤Pev,i,max, and satisfy the minimum energy replenishment requirement before the vehicle leaves the station.

[0042] S6: Solve the day-ahead optimization model and output a 96-point baseline plan. The objective function of S4 and the constraints of S5 are unified to form a hybrid linear or nonlinear optimization problem. A solver (such as CPLEX, Gurobi, or an open-source solver) is called to calculate the baseline plan for the next scheduling day's 96 time periods. The baseline plan includes at least: energy storage charging and discharging power, grid-connected power, controllable load adjustment, photovoltaic curtailment, and charging pile group allocated power for each time period; it also outputs the expected SOC trajectory, expected peak demand, and expected total operating cost for each time period.

[0043] S7: Introduces the latest real-time information at the current moment of the day. When the system reaches the current k-th 15-minute period, it re-collects the current measured load, measured photovoltaic power, real-time electricity price or updated price forecast, rolling weather forecast, available energy storage capacity, charging pile connection status, and alarm status of key equipment. It then compares these latest data with the planned values ​​for the corresponding period of the previous day's plan to form a rolling correction input set.

[0044] S8: Perform rolling optimization for the remaining time domain The remaining time domain is defined as the period from the current time period k to the last time period 96. While maintaining the current measured SOC(k) as the initial state, the objective function and constraint set are reconstructed to re-optimize energy storage, grid connection, controllable load, photovoltaic curtailment, and charging pile allocation in the remaining time domain. Preferably, a tracking term Jref for the day-ahead baseline plan is introduced into the rolling objective function to avoid unnecessary large oscillations in the strategy. This tracking term can be represented as a weighted penalty for the difference between the current rolling decision and the day-ahead plan, used to constrain the deviation between the current rolling decision and the baseline plan.

[0045] S9: Issue and execute only the next control step. From the rolling optimization results obtained from S8, only the control quantity of the current time period k is extracted as the execution instruction to be issued to the EMS or edge controller, including the energy storage charging and discharging setpoint, grid connection point power setpoint, controllable load adjustment quantity, charging pile power limit value, and photovoltaic power limiting setpoint when necessary. After the time period ends, the next round of rolling optimization is re-entered, instead of fixing and executing the strategy for all remaining time periods at once.

[0046] S10: Collect execution results and assess economic losses. Obtain the actual execution data for the current control step, including actual grid-connected power, actual SOC change, actual charging amount, power supply status of key loads, and actual electricity purchase cost for the corresponding time period; then calculate the photovoltaic forecast error, load forecast error, and the electricity purchase cost deviation caused by these errors. Preferably, the economic loss can be defined as the difference between the actual operating cost and the optimized cost under ideal forecast conditions.

[0047] S11: Perform closed-loop parameter correction and proceed to the next cycle. Based on the prediction error and economic loss obtained from S10, adaptive adjustments are made to parameters such as the energy storage degradation coefficient λdeg, demand weight λdem, curtailment penalty coefficient λcurt, and controllable load penalty coefficient λflex. For example, when the actual energy storage degradation rate is higher than the model estimate, λdeg is increased; when the risk of actual demand exceeding limits increases, λdem is increased; and when unnecessary curtailment occurs frequently, λcurt is adjusted. Parameter adjustments can be implemented using fixed-step adjustment, adaptive-step adjustment, scenario-specific parameter library switching, or online identification based on historical operating data. After parameter adjustments are completed, the updated parameters are called and the solution is recalculated in the next 15-minute rolling cycle to form a closed-loop learning mechanism of "prediction error - economic loss - parameter adjustment - re-optimization".

[0048] Through the above steps, the present invention realizes an integrated scheduling process of day-ahead planning, intraday rolling re-optimization, and post-run parameter learning, which not only retains the globality of day-ahead optimization, but also has the ability to correct deviations in real time to deal with disturbances during the day.

[0049] To enhance long-term adaptability, this invention includes a feedback learning module. This module continuously evaluates the relationship between "prediction error, economic loss, and strategy correction," thus establishing a parameter update mechanism.

[0050] Learnable parameters include, but are not limited to: Energy storage degradation coefficient; Demand weight; Discard penalty coefficient; Comfort or production line disturbance penalty coefficient; Optimize the time domain length by scrolling; Predict confidence intervals and safety margins.

[0051] When the system detects that a certain type of constraint is frequently triggered or that a certain type of prediction error leads to excessive electricity cost losses, it can automatically increase the relevant penalty coefficient or safety margin to form a more robust subsequent scheduling strategy.

[0052] Taking a microgrid in an industrial park as an example: rooftop photovoltaic (1MWp installed capacity), user-side energy storage (1MWh / 500kW capacity), several charging piles (total power 300kW), and general industrial and commercial loads (average daily load of approximately 800kW): In the daytime phase, based on the next day's weather forecast and operating load prediction, the system generates a 96-point plan, which is expected to utilize the surplus photovoltaic period during midday (11:00-13:00) to charge the energy storage, and discharge during the evening peak (17:00-20:00) and peak periods (20:00-22:00), while controlling the grid connection point power to not exceed the transformer's safe operating threshold (1000kVA). At the same time, the energy storage SOC trajectory and total operating cost for each period are predicted.

[0053] During the intraday phase, when actual weather worsens than predicted and photovoltaic (PV) output decreases by approximately 20%, the system incorporates the latest irradiance and measured PV power in the next 15-minute rolling interval to recalculate the remaining time-domain strategy. If a sudden increase of 30kW in charging pile load is detected simultaneously, the new optimization model prioritizes energy storage discharge and appropriately reduces controllable charging power, thereby mitigating the risks of purchasing electricity during high-price periods and exceeding demand limits. If positive deviations in PV forecasts (actual output lower than predicted) occur for several consecutive days, the feedback learning module will increase the curtailment penalty coefficient, enabling subsequent scheduling to more conservatively estimate available PV output and avoid plan disruptions due to forecast deviations.

[0054] Those skilled in the art will understand that, without departing from the core idea of ​​this invention, the following aspects can be substituted: The day-ahead and intraday forecasting models can use statistical regression models, time series models, machine learning models (such as LSTM, XGBoost) or combinations thereof. There are no restrictions on the specific forecasting algorithm, as long as it can output the forecast results of photovoltaic, load, electricity price and available capacity for each future period.

[0055] Current optimization and rolling optimization can be implemented using mixed-integer linear programming, quadratic programming, dynamic programming, heuristic algorithms, or other solvers under the model predictive control framework, and are not limited to a single solution method.

[0056] In addition to electricity costs, demand, lifetime losses, and curtailment penalties, other targets such as demand response revenue, ancillary service revenue, carbon trading costs, or green electricity trading revenue can be added, as long as they still reflect the unified economic optimization concept.

[0057] In addition to charging piles and controllable loads, it can also be extended to access cold and heat energy storage equipment, interruptible loads, backup generator sets or flexible power distribution equipment, thereby forming a more generalized source-grid-load-storage coordinated scheduling scheme.

[0058] Closed-loop learning can employ fixed step size correction, adaptive step size correction, switching of parameter libraries for different scenarios, or online identification based on historical running data, as long as the parameters can be corrected and optimized based on the running results.

[0059] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

[0060] The foregoing has described specific embodiments of the present invention. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps described in the claims may be performed in a different order than that shown in the embodiments and may still achieve the desired results. Furthermore, the processes depicted in the drawings do not necessarily require the specific or sequential order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0061] In the description of the embodiments of the present invention, the terms "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of the present invention. In the embodiments of the present invention, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials, or characteristics described may be combined in a suitable manner in any one or more embodiments or examples. Furthermore, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in the embodiments of the present invention, as well as the features of the different embodiments or examples.

[0062] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of embodiments of the present invention, "multiple" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0063] Any process or method description in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or more executable instructions for implementing custom logic functions or processes, and the scope of preferred embodiments of the invention includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order according to the functions involved, as should be understood by those skilled in the art to which embodiments of the invention pertain.

[0064] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A rolling optimization economic dispatch method for microgrid sources, grid, load, and storage, characterized in that, Includes the following steps: S1: Collect and preprocess multi-source operation data of the microgrid; S2: Based on historical operational data and the latest external information, call the day-ahead prediction model to output multi-period prediction results covering the entire scheduling cycle of the next day; S3: Discretize the next day's scheduling cycle into multiple time periods, each time period being 15 minutes long. Define system state variables and control variables. The state variables include at least the energy storage charge state, and the control variables include at least the energy storage charging and discharging power, grid-connected power, controllable load adjustment, and power allocated to each charging pile. S4: Construct a day-ahead optimization objective function with the goal of minimizing total operating cost; S5: Construct unified engineering constraints; S6: Solve the day-ahead optimization objective function under the constraints and output the multi-period baseline plan; S7: Introduce the latest real-time information at the current moment of the day, and perform rolling optimization for the remaining time domain using the current measured state as the initial condition; S8: Extract the control quantity for the current time period from the rolling optimization results, issue and execute it; S9: Collect execution results and assess economic losses; S10: Perform closed-loop parameter correction based on prediction error and economic loss, and proceed to the next cycle.

2. The method according to claim 1, characterized in that, The multi-source operation data includes at least photovoltaic power output data, energy storage system state of charge, charging and discharging power, available capacity and charging and discharging efficiency data, charging pile power and connection status data, park basic load and controllable load data, time-of-use electricity price or spot electricity price data, meteorological data, historical demand data, and key equipment availability status data; the preprocessing includes timestamp alignment, missing value completion, outlier removal and unit standardization to form a data sequence with a uniform time interval.

3. The method according to claim 1, characterized in that, The day-ahead forecasting model includes a photovoltaic output forecasting submodule, a load forecasting submodule, an electricity price forecasting or price mapping submodule, an equipment available capacity forecasting submodule, and a demand risk forecasting submodule. The demand risk forecasting submodule estimates the risk level of triggering demand over-limit in future periods based on historical maximum demand, future load levels, and grid connection point constraints.

4. The method according to claim 1, characterized in that, The total operating cost includes the cost of purchasing and selling electricity in each time period, the demand electricity cost calculated by multiplying the maximum grid-connected power and the demand unit price within the settlement cycle, the energy storage equivalent cycle life loss cost calculated by multiplying the charging and discharging energy flux and the unit cycle life cost, the cost of photovoltaic curtailment or limiting, the cost of controllable load adjustment, and the cost of penalties for violating critical safety boundaries.

5. The method according to claim 1, characterized in that, The unified engineering constraints include at least the dynamic and range constraints of the energy storage state of charge, the mutual exclusion constraints of energy storage charging and discharging, the power balance constraints, the available output constraints of photovoltaics, the power constraints of the grid connection point, the transformer capacity constraints, the power ramping constraints, the upper limit constraints of demand, the constraints of ensuring the energy supply of critical loads, and the power allocation constraints of the charging pile group.

6. The method according to claim 1, characterized in that, The rolling optimization for the remaining time domain uses the measured state at the current moment as the initial condition and introduces a tracking penalty term for the day-ahead baseline plan into the objective function. The tracking penalty term is used to constrain the deviation between the current rolling decision and the baseline plan in order to avoid large oscillations in the rolling strategy relative to the day-ahead baseline plan.

7. The method according to claim 1, characterized in that, The step of extracting and executing the control quantity for the current period from the rolling optimization results specifically involves extracting only the control quantity for the current period from the rolling optimization results as an execution instruction and sending it to the energy management system or edge controller. After the period ends, the system re-enters the next round of rolling optimization, rather than executing all remaining period strategies at once.

8. The method according to claim 1, characterized in that, The process of collecting execution results and assessing economic losses includes: obtaining the actual execution data of the current control step, calculating the photovoltaic prediction error, load prediction error, and the deviation in electricity purchase cost caused by the error, and defining the difference between the actual operating cost and the optimized cost under ideal prediction conditions as the economic loss.

9. The method according to claim 1, characterized in that, The closed-loop correction of execution parameters includes: adaptively correcting the energy storage degradation coefficient, demand weight, curtailment penalty coefficient, and controllable load penalty coefficient based on prediction error and economic loss, and calling the updated parameters to resolve the optimization model in the next rolling cycle, forming a closed-loop learning mechanism of prediction error-economic loss-parameter correction-re-optimization.

10. A microgrid integrated power generation, grid, load, and storage economic dispatch system, used to execute the method as described in any one of claims 1 to 9, characterized in that, include: The data acquisition layer is used to collect multi-source operational data of the microgrid; The forecasting layer is used to generate photovoltaic output forecasts, load forecasts, electricity price forecasts, equipment availability capacity forecasts, and demand risk forecasts based on multi-source operating data. The optimization decision layer is used to construct a day-ahead optimization model and an intraday rolling optimization model based on the output of the prediction layer, with the goal of minimizing the total operating cost, and solve for the baseline plan and the next control step strategy. The intraday rolling optimization model uses the measured state at the current moment as the initial condition, performs rolling optimization on the remaining time domain, and only outputs the strategy for the current control step; The execution and closed-loop layer is used to send the next control step strategy to the field equipment for execution, and to perform feedback evaluation and adaptive parameter correction based on execution deviation and economic loss.

Images