Two-stage optimization method for energy management of carbon-constrained microgrid

Abstract

The invention discloses a two-stage optimization method for carbon constraint micro-grid energy management. The method comprises the following steps: establishing power balance constraints of various components in a micro-grid, an economic-low carbon joint objective function, optimization solution of diesel engine power, grid transaction, a load strategy and energy storage scheduling; the system-level CVaR tail risk index is decoupled to a short-time-period CVaR value according to the probability distribution of the system-level CVaR tail risk index and the time window area proportion, and the short-time-period CVaR value serves as a carbon over-quota emission risk constraint boundary and is transmitted to the intra-day level; actually measured wind and light load data are fed back to the prediction model for real-time correction in each period; constructing a robust optimization model to dynamically adjust conservative coefficients; the risk budget is used as a hard constraint, and the expected value of the excess carbon emission condition in the window is calculated in real time; a closed loop is formed. The intra-day rolling optimization provided by the invention is based on an ultra-short-term prediction technology, and the prediction error of the intra-day rolling optimization is only slowly increased in a finite time window, so that the prediction precision at the same time point is obviously superior to that of a day-ahead optimization result.

Description

Technical Field

[0001] The present invention belongs to the technical field of power grid management, and in particular relates to a two-stage optimization method for carbon-constrained microgrid energy management. Background Art

[0002] As a distributed, flexible and controllable new type of efficient energy network, microgrid (MG) can effectively integrate and utilize distributed renewable energy sources (DERs) such as regional wind energy (Wind) and solar energy (PV). It is an important way and key support to meet the challenges of energy transformation and alleviate the pressure of carbon emission reduction.

[0003] However, the large-scale deployment of microgrids and their full potential for carbon reduction face fundamental challenges posed by the significant uncertainty of renewable energy, particularly in the refinement of carbon emission control. Therefore, accurately and in real time quantifying carbon emissions and dynamically controlling tail carbon emission risks during microgrid operation have become key bottlenecks in achieving low-carbon development goals.

[0004] Traditional deterministic optimization methods are insufficient to address the core challenges of wind and solar power generation driven by uncertainty. Robust optimization (RO) and model predictive control (MPC) have become the focus of current research due to their superior uncertainty handling capabilities.

[0005] RO is an optimization method that addresses uncertainty by defining intervals for uncertainty parameters. Existing technologies employ a two-stage robust optimization approach and a column and constraint (C&CG) algorithm to solve the non-convex stability constraint problem under the worst-case scenario of wind and solar output. However, the conservative and static nature of robust optimization has limited its development.

[0006] MPC is a model-based closed-loop control strategy. Its core is to perform rolling time domain optimization and first-step control based on the prediction model, while combining feedback to correct the prediction error to form a closed-loop control process.

[0007] Conditional Value at Risk (CVaR), a risk assessment metric used to measure expected losses under worst-case scenarios exceeding a specified confidence level, offers unique advantages in risk modeling for renewable energy power systems due to its ability to effectively quantify extreme tail risks. However, risk management based on CVaR exhibits coarse-grained characteristics, with fragmented transmission of risk constraints across time and space, making it difficult to effectively support the real-world needs of refined dynamic risk management.

[0008] Based on existing research, the following key limitations remain: RMPC and the carbon quota system lack dynamic quantification tools for tail risks of carbon over-emissions, and the CVaR method has not yet been migrated to the carbon risk domain, resulting in an inability to accurately capture extreme high-carbon emission scenarios. Existing CVaR risk constraints exhibit a coarse-grained, single, all-day risk threshold that is fragmented in time and space, making it difficult to support closed-loop, refined risk control. Therefore, it is urgent to develop new energy management strategies that integrate multi-timescale dynamic optimization with precise short-term risk quantification to systematically address the challenges of strong uncertainty, measure and constrain tail risks of carbon over-emissions in real time, and ultimately ensure the low-carbon, economical, and reliable operation of microgrids.

[0009] In addition, existing research has encountered the problem of time and space separation when using CVaR. A fixed risk preference coefficient is used to regulate the risk level throughout the day, and no time-based CVaR constraint mechanism is established. This makes it impossible to dynamically respond to changes in operating conditions, resulting in over-conservativeness in low-risk periods and insufficient constraints in high-risk periods. CVaR also has the problem of being decoupled from the optimization control framework: if multi-scenario CVaR optimization is adopted, but risk constraints are only used as additional items in the objective function and are not embedded in the rolling optimization kernel, the feedforward risk forecast is missing and the feedback risk adjustment is absent. The more prominent problem lies in the limitation of the field. The existing CVaR model is not adequately adapted to the tail risk of carbon emissions.

[0010] Therefore, building a collaborative framework that integrates dynamic carbon risk refinement quantification, multi-timescale optimization control and carbon quota status feedback to achieve real-time perception, prediction, constraint and suppression of tail carbon over-emission risks has become the key to solving the aforementioned core challenges. Summary of the Invention

[0011] In view of this, this application proposes a two-stage microgrid energy management strategy based on robust model predictive control that integrates carbon quota and CVaR conditional value at risk.

[0012] To achieve the above objectives, the two-stage optimization method for carbon-constrained microgrid energy management disclosed in this application includes the following steps: Day-ahead stage: Establish power balance constraints for various components in the microgrid and a joint economic-low-carbon objective function, optimizing diesel engine power, grid transactions, load strategies, and energy storage scheduling. Decouple the system-level CVaR tail risk indicator into a short-term CVaR value based on its probability distribution and the time window area ratio, and transfer it to the intraday level as the carbon over-quota emission risk constraint boundary. Intraday rolling phase: Feedback the measured wind and solar load data to the forecast model for real-time correction every cycle; Build a robust optimization model to dynamically adjust the conservative coefficient , coordinate electricity price fluctuations, renewable energy deviations and load uncertainty; calculate the excess emission tail risk value within the window in real time , taking the carbon excess quota emission risk constraint boundary allocated in the day-ahead phase as the excess emission tail risk value within the window The upper limit forms a hard constraint; the execution layer adopts the first-stage decision and transmits the diesel engine climbing status and energy storage SOC to the next cycle, forming a "prediction-optimization-execution-feedback" closed loop.

[0013] Preferably, the power balance constraints include constraints on wind and solar power generation, energy storage charging, diesel power generation, total power purchases from the grid, load power consumption, energy storage discharge, and total power sales from the grid: ; ; Among them, P wind It is the wind power generation, P pv It is the photovoltaic power generation, P DG Is the diesel generator, P BESS,dis is the day-ahead energy storage discharge, P grid，buy is the total amount of electricity purchased by the power grid today, P load is the day-ahead load electricity consumption, P BESS,ch For the day before energy storage charging, P grid，sell is the total amount of electricity sold by the power grid on the previous day; The economic-low-carbon joint objective function is composed of the operating cost of each device, the electricity price response cost and the total carbon emission cost: ; in, is the operation and maintenance cost of battery energy storage, is the operation and maintenance cost of the diesel generator, is the grid transaction cost, is the penalty cost of load reduction, is the electricity price response cost, is the total carbon emission cost.

[0014] Preferably, the total carbon emission cost Including conventional carbon emission costs and excess carbon emission costs; conventional carbon emission costs , Excess carbon emission costs and total carbon emission costs The calculation of is as follows: ; ; ; is the carbon price, is the excess penalty factor, is the carbon quota threshold, and T is the number of time granularity windows.

[0015] Preferably, the confidence level Day-ahead forecast conditional value at risk is defined as follows: ; in, Indicates the confidence level The conditional value at risk of the day-ahead forecast under The tail average loss of the quantile, Indicates The value at risk at the level is the kernel density probability estimation function.

[0016] Preferably, the system-level CVaR tail risk indicator calculation includes calculating the day-ahead CVaR value, decoupling the day-ahead CVaR, and calculating the excess emission tail risk value. .

[0017] Preferably, the calculating of the day-ahead CVaR value specifically includes: Based on the day-ahead optimization scheduling results, the carbon emission factor and emission control coefficient are used. Calculating carbon emissions at a time granularity and carbon quota threshold : Carbon quota threshold The calculation formula is as follows: ; t is time, P load is the total actual load demand energy; is the control and discharge coefficient, It is the time granularity; The carbon emissions in the microgrid take into account the direct carbon emissions of diesel generators and the indirect carbon emissions from purchasing electricity from the grid. The carbon emissions at each time granularity are for: ; P DG is the electricity generated by the diesel generator a few days ago, P grid，buy is the electricity purchased by the grid on the previous day, EF grid is the diesel engine carbon emission factor, EF DG is the grid carbon emission factor; Calculating excess carbon emissions : ; The kernel density estimation method is used to construct the carbon emission probability density function to circumvent the limitations of the conventional normal distribution assumption, including: First, determine any point on its estimated domain , a and b are the lower and upper bounds of the domain, and they are extended by 10 units on both sides of the data extreme value to ensure that the kernel density estimation can fully cover the tail characteristics of the carbon emission distribution and avoid the deviation of the probability density estimation caused by boundary truncation: ; ; Calculate the kernel density probability estimation function of excess carbon emissions on the previous day: ; In the formula, x is the function variable, and the kernel density estimation bandwidth is The calculation of is based on the Silverman criterion, which is as follows: ; ; is the variance, is the mean; Then calculate the day before , find the cumulative probability to reach The carbon emission threshold is: ; is the set of real numbers, u is the integration variable; Before the calculation date When , it is approximated by Monte Carlo integration, and the formula is as follows: ; in, For The simulated value of the sample, i is the sampling number, M is the value .

[0018] Preferably, the decoupling of day-ahead CVaR specifically includes: Based on the day-ahead CVaR forecast value and its probability distribution characteristics, the risk period weight division is implemented according to the time window area ratio, and discrete mapping is performed to T time granularity windows to support dynamic closed-loop control of risks during the day. Define the weight coefficient according to the integral area ratio : ; k is the prediction length of the intraday real-time rolling optimization layer; Decouple the day-ahead CVaR to the windowed CVaR: .

[0019] Preferably, the calculation of excess emission tail risk value , specifically including: Calculate the tail risk value of excess emissions within the day according to the method of calculating the day-ahead CVaR value ,The difference is that, in the calculation of the kernel density probability estimation function f(x), the day-ahead prediction domain T is replaced by the intraday prediction domain k; During the intraday rolling optimization process, CVaR(t) is used as the upper limit of the dynamic risk constraint for the intraday rolling optimization, and real-time power adjustment is used to ensure that the risk of carbon over-quota emissions is always under control: .

[0020] Preferably, during the intraday rolling optimization phase, the robust optimization model processes the errors in the forecasts of wind and solar output and key load demands: ; ; ; is the intraday wind speed forecast after robust processing, is the wind speed forecast for the day, is the wind prediction error variable, with a value between [-1, 1], is the intraday forecasted photovoltaic power after robust processing, is the intraday predicted photovoltaic power, is the photovoltaic prediction error variable, with a value between [-1, 1], is the intraday forecast load demand after robust processing, is the intraday forecasted load demand, is the load forecast error variable, with a value between [-1, 1], These are the wind, solar, and load forecast error boundary parameters, which are 15%, 10%, and 5%, respectively; In order to transform the optimization problem from a nonlinear programming problem that is difficult to solve into a linear problem and minimize the influence of uncertainty, auxiliary variables are introduced. , , , to express the upper bound of wind, solar, and load forecast power errors: ; And add a global constraint to limit the total error impact to not exceed a preset threshold: ; ; Where, yes , and The maximum value in is a conservative coefficient that dynamically tightens or relaxes the system tolerance according to the error intensity and is determined using the following formula: ; is the power balance relaxation calculated by the error upper bound auxiliary variable and the dynamic robustness coefficient; In order to fully consider the worst case scenario, slack is added to the power balance, which becomes an inequality constraint: ; in, It is the daily energy storage discharge, It is diesel power generation during the day. is the total amount of electricity purchased by the power grid during the day, It is a daily energy storage charging. is the total amount of electricity sold by the power grid during the day, is the maximum rated power of the load that can be cut, is the daily load shedding rate; Slack is added to the objective function of the intraday rolling optimization. The objective function consists of the deviation cost of the power grid, diesel, and energy storage, the power balance slack cost, and the risk term cost: ; in, is the deviation cost of the grid, is the deviation cost of diesel, is the deviation cost of energy storage, is the power balancing relaxation cost, is the cost of the risk term.

[0021] The beneficial effects achieved by this application are as follows: 1) Introducing a time-varying carbon price mechanism, through the combined guidance of carbon quota constraints and the CVaR tail risk model, significantly reduces systemic carbon emissions and fully measures the tail risk value of excessive carbon emissions; 2) A CVaR risk stratification modeling method based on a rolling time domain is proposed. This method decouples all-weather CVaR values ​​into short-time window granularity thresholds, enabling refined dynamic transmission of risk constraints and building a closed-loop "day-ahead pre-assessment - intraday tracking and correction" chain. This solves the problem of risk fragmentation in time and space and supports refined over-discharge risk management and control. 3) Design a collaborative architecture of "day-ahead planning + intraday RMPC", formulate a 24-hour benchmark scheduling plan in the day-ahead phase, and embed RMPC in the intraday phase to correct wind and solar load fluctuation deviations in real time; jointly minimize equipment operating costs, carbon quota compliance costs, and CVaR tail risk costs to achieve the triple optimization goals of economy, low carbon, and robustness. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1Schematic diagram of the two-stage optimization method of this application; Figure 2 The experimental results are as follows: a comparison chart of wind power forecast data within the day before; Figure 3 The experimental results are as follows: a comparison chart of load demand forecast data within the day before the day. DETAILED DESCRIPTION

[0023] The present invention will be further described below with reference to the accompanying drawings, but the present invention is not limited in any way. Any changes or substitutions made based on the teachings of the present invention fall within the scope of protection of the present invention.

[0024] The grid-connected microgrid (MG) considered in this application consists of an energy management system (EMS), a diesel generator (DG), a battery energy storage system (BESS), renewable energy sources (DERS) such as photovoltaic (PV) and wind power (wind), shedding loads, critical loads and a direct current / alternating current (DC / AC) inverter.

[0025] Advanced metering infrastructure is deployed at key nodes in the microgrid to collect real-time data on renewable energy generation, load consumption, energy storage status, and power interaction with the main grid. The data is then transmitted to the local energy management system (EMS), which then optimizes and formulates energy scheduling strategies based on this data.

[0026] During the energy management process, the following principles must be followed: renewable energy generation within the microgrid system should prioritize meeting local load demand, and surplus power should be stored in battery energy storage (BESS) first, and then sold to the main grid; when the local power generation and energy storage of the microgrid are insufficient to meet load demand, the difference in power is supplemented by the main grid by purchasing or starting diesel generators (DG).

[0027] The mechanism of the two-stage energy scheduling management framework based on model predictive control proposed in this application is as follows: Figure 1 .

[0028] In the day-ahead stage, forward-looking decision-making is achieved through time-decay error prediction and carbon risk decoupling mechanism: based on wind and solar load forecast data, an exponential decay curve is used to dynamically approximate the actual scenario error; an economic-low-carbon joint objective function is established to optimize diesel engine power, grid transactions, load strategy and energy storage scheduling; the system-level CVaR carbon risk indicator is innovatively decoupled into a short-term probability distribution model, generating a dynamic carbon quota constraint boundary and transmitting it to the intraday level.

[0029] During the intraday rolling phase, Robust Model Predictive Control (RMPC) forms a closed-loop optimization: measured wind and solar load data is fed back into the forecasting model for real-time correction each cycle. A robust optimization model is constructed to dynamically adjust the conservative coefficient to coordinate electricity price fluctuations, renewable energy deviations, and load uncertainty. The CVaR risk budget assigned on the day before is used as a hard constraint, and the expected value of excess carbon emissions within the window is calculated in real time. The execution layer adopts the initial decision and transmits the diesel engine ramp status and energy storage SOC to the next cycle, forming a "prediction-optimization-execution-feedback" closed loop. This framework collaboratively optimizes operating costs, carbon emission costs, and risk costs on a dual time scale, significantly improving the system's robustness and low-carbon performance under multiple uncertainties.

[0030] Set the day-ahead prediction time domain length to T and the time granularity to , after optimization, the optimal control sequence at time t=0 is obtained , and pass the scheduling instructions to the intraday real-time layer. The intraday real-time rolling optimization layer has a prediction length of k and a time granularity of , each rolling optimization gets Optimal control sequence at time , , and only issue the first scheduling instruction Give the microgrid execution and then continue The cycle continues until the scheduling period ends.

[0031] Carbon quotas, as tradable and limited carbon emission quotas, are a key market-based tool to incentivize micro-entities to reduce emissions. In order to guide users to deeply participate in demand response, this application introduces a carbon quota mechanism. The current main carbon quota allocation methods include the baseline method, the historical intensity method, and the grandfather method. This application uses the baseline method with more significant energy-saving and emission reduction effects. Carbon emission cost accounting adopts a full carbon pricing mechanism, covering the conventional costs of all emissions and the penalty costs of excess emissions, and setting quota thresholds for users to set independently in the day-ahead stage, thereby constraining users' active emission control behavior.

[0032] In the day-ahead phase, the calculation formula for carbon quota is as follows: (1); The carbon emissions in the microgrid take into account the direct carbon emissions of diesel generators and the indirect carbon emissions from purchasing electricity from the grid. The carbon emissions at each time granularity are for: (2); Carbon emission costs include conventional carbon emission costs and excess carbon emission costs. Conventional carbon emission costs , Excess carbon emission costs and total carbon emission costs The calculation of is as follows: (3); (4); (5); is the carbon price, is the carbon quota threshold, and T is the number of time granularity windows.

[0033] CVaR Conditional Value at Risk Model The CVaR approach addresses a fundamental limitation of the traditional risk measurement method, Value at Risk (VaR). While traditional methods focus on average performance, CVaR specifically addresses tail risk—the most severe scenarios with low probability of occurrence but high impact. In terms of carbon emissions, this means keeping emissions within acceptable limits even under the worst-case scenario for renewable energy generation.

[0034] CvaR addresses a fundamental limitation of traditional risk measurement methods. While conventional approaches focus on average performance, CVaR specifically targets tail risks—the most severe scenarios with low probability but high impact. For carbon emissions, this ensures that even under the worst-case renewal scenario, emissions remain within acceptable limits.

[0035] The definition of CVaR is as follows: (6); in, Indicates the confidence level The conditional value at risk under the day-ahead forecast reflects the The tail average loss of the quantile, Indicates that the level The value at risk, is the kernel density probability estimation function.

[0036] This application calculates the conditional expected value of excess emissions under a specific confidence level and strictly controls the actual cumulative probability of excess emissions within a day to be less than the preset upper limit on the previous day, ensuring that the total carbon emissions in extreme and uncertain scenarios are controllable, thereby significantly improving the robustness of the system's carbon constraints.

[0037] The application of CVaR in this application includes the following three key steps: (1) Calculate the day-ahead CVaR value Based on the day-ahead optimization scheduling results, the carbon emission factor and emission control coefficient are used. Calculating carbon emissions at a time granularity and carbon quota threshold , the calculation follows formula (1) and (2); Recalculate excess carbon emissions : (7); Given that carbon emissions throughout the day may exhibit non-normal distribution characteristics (such as left / right skewness), this application uses kernel density estimation (KDE) to construct the carbon emissions probability density function to avoid the limitations of the conventional normal distribution assumption.

[0038] First, determine any point on its estimated domain The domains of a and b are shown in formula (8). By expanding 10 units on both sides of the data extreme value, it is ensured that the kernel density estimation can fully cover the tail characteristics of the carbon emission distribution and avoid the deviation of the probability density estimation caused by boundary truncation; ; (8); Then calculate the kernel density probability estimation function of excess carbon emissions on the previous day: (9); Where, the kernel density estimation bandwidth is The calculation of is based on the Silverman criterion, see formula (10) ; (10); is the variance, is the mean.

[0039] Then calculate the day before , find the cumulative probability to reach The carbon emission threshold is: (11); Before the calculation date When , it is approximated by Monte Carlo integration, and the formula is as follows: (12); in, For The simulated value sampled in the middle, M takes a maximum value .

[0040] (2) Decoupling day-ahead CVaR Based on the day-ahead CVaR forecast value and its probability distribution characteristics, the risk period weight division is implemented according to the time window area ratio, and the discrete mapping is performed to A time granularity window to support dynamic closed-loop control of risks during the day; Define the weight coefficient according to the integral area ratio : (13); Decouple the day-ahead CVaR to the windowed CVaR: (14).

[0041] (3) Calculate the tail risk value of excess emissions within a day ; Intraday excess emission tail risk value The calculation of the value is the same as that of Equations (7)-(13). In Equation (7), the probability density function is calculated by replacing the day-ahead prediction domain T with the intraday prediction domain k. During the intraday rolling optimization process, As the dynamic risk constraint upper limit for intraday rolling optimization, it ensures that the risk of carbon over-quota emissions is always under control through real-time power adjustment; (15); In the day-ahead planning stage, the optimization solution is performed by setting the constraints of various components, power balance constraints and objective functions, and the carbon quota and CVaR constraints are calculated. The solution results are saved and passed to the intraday rolling optimization stage.

[0042] (1) Power balance constraints The power generation of power balance is the total amount of wind and solar power generation, energy storage charging, diesel power generation, and power purchased from the grid. The power consumption is the total amount of load power consumption, energy storage discharge, and power sold to the grid. ; (16).

[0043] (2) Objective function The objective function consists of the operating cost of each device, the electricity price response cost, and the carbon emission cost; ; ; Intraday rolling optimization: The intraday rolling optimization stage combines robust optimization with model predictive control, incorporates carbon quotas and CVaR constraints to guide carbon emissions, effectively smooths out fluctuations caused by wind and solar uncertainties, significantly reduces carbon emissions, and ensures that the risk of carbon emission over-emissions is within a controllable range.

[0044] (1) Robust processing of wind and solar load forecast In the daily rolling optimization of microgrids, there are inevitable errors in the forecast of wind, solar output and key load demand, so they are robustly processed; ; ; (18).

[0045] (2) Robust processing of power balance: In order to change the optimization problem from a nonlinear programming problem that is difficult to solve to a linear problem and minimize the influence of uncertainty, auxiliary variables are introduced. , , , which represents the upper bound of the wind, solar and load forecast power errors, and its expression is shown in formula (19); (19); And add a global constraint to limit the total error impact to not exceed the preset threshold, as shown in formula (20): ; (20); Where, yes , and The maximum value in is the dynamic robustness coefficient, which dynamically tightens or relaxes the system tolerance according to the error intensity and is determined using the following calculation formula: (twenty one); The power balance slack is calculated by using the error upper bound auxiliary variable and the dynamic robustness coefficient. In order to fully consider the worst case in all possible scenarios, the slack is added to the power balance, which becomes an inequality constraint: (twenty two).

[0046] (3) Objective function Add slack to the objective function, which consists of the deviation cost of the power grid, diesel, and energy storage, the power balance slack cost, and the risk term cost; (twenty three).

[0047] This application conducts experiments and simulates the results of research to evaluate the advantages of the two-stage energy management framework and the performance of the energy management strategy considering carbon quotas and CVaR constraints. The optimized scheduling results of the two-stage energy management framework are demonstrated, and the coordinated scheduling scheme of the power grid purchase and sales plan, diesel generator output plan and energy storage charging and discharging plan under the constraints of time-varying electricity prices and carbon prices is presented. By designing comparative experiments with / without CVaR carbon risk constraints, the effectiveness of the risk guidance mechanism is verified: under conditions without CVaR constraints, the carbon emission tail risk threshold constraint is lifted in the intraday rolling optimization stage, and the active prevention and control of extreme over-emissions are cancelled. Based on the comparison of carbon emission distribution and tail over-emission levels at different confidence levels, especially the 95% confidence level, the superiority of the risk-taking optimization framework proposed in this application is systematically analyzed.

[0048] All simulations in this application were performed on a personal computer with an Intel(R) Core(TM) i9-14900HX processor and 32GB of RAM. The simulation experiments for the two-stage energy management strategy for the microgrid were implemented in MATLAB, and the optimization process was solved using YALMIP and Gurobi solvers.

[0049] The historical data of wind, solar and load in this application are all from Open Power System Data. Due to the large scale of the original data, we scaled it down proportionally based on the original data without changing the rules and characteristics of the original data. Then, we extracted typical scenarios through K-means clustering, and then generated the single-day forecast data of wind, solar and load in four seasons through ARIMA. Finally, we used the probability envelope weighted average method to fuse the multi-season curves into a single probabilistic forecast, and used the safety margin coefficient to perform robust boundary constraints. The wind power and load forecast data after processing are as follows: Figure 2 、 Figure 3 shown.

[0050] Time-of-use electricity price data is generated based on the peak and valley characteristics of the power grid. Carbon price data retains the characteristics of electricity price fluctuations and is taken as 0.5 times the electricity price data.

[0051] During both the day-ahead optimization and intraday rolling optimization phases, a forecast error model with linear growth over time was employed. Specifically, the maximum wind power forecast error was set at 15%, the maximum photovoltaic power forecast error was set at 10%, and the maximum load power forecast error was controlled at 5%. The wind and photovoltaic power forecast errors were stochastically simulated using a uniform distribution, while the load forecast error was characterized by its random fluctuations using a normal distribution.

[0052] The core of error modeling lies in its time-varying nature: forecast error exhibits a linear growth trend as the time period progresses. This mechanism is particularly evident in day-ahead scheduling, where, due to its long timescale (covering a full 24-hour period), forecast error accumulates and amplifies as the scheduling period progresses. In contrast, intraday rolling optimization, based on ultra-short-term forecasting techniques, only allows for a slow growth of forecast error within a limited time window, ensuring significantly better forecast accuracy at the same point in time than day-ahead optimization results.

[0053] As used herein, the word "preferred" is intended to serve as an example, instance, or illustration. Any aspect or design described herein as "preferred" is not necessarily to be construed as advantageous over other aspects or designs. Rather, the use of the word "preferred" is intended to present concepts in a concrete manner. As used in this application, the term "or" is intended to mean an inclusive "or" rather than an exclusive "or." That is, unless otherwise specified or clear from the context, "X employs A or B" is intended to mean any of the naturally inclusive permutations. That is, if X employs A; X employs B; or X employs both A and B, then "X employs A or B" is satisfied in any of the foregoing examples.

[0054] Moreover, although the present disclosure has been shown and described with respect to one or implementations, those skilled in the art will think of equivalent variations and modifications based on reading and understanding of this specification and the accompanying drawings. The present disclosure includes all such modifications and variations and is limited only by the scope of the appended claims. In particular, with respect to the various functions performed by the above-mentioned components (such as elements, etc.), the terms used to describe such components are intended to correspond to any component (unless otherwise indicated) that performs the specified function of the component, even if structurally not equivalent to the disclosed structure that performs the function in the exemplary implementation of the present disclosure shown herein. In addition, although the specific features of the present disclosure have been disclosed with respect to only one of several implementations, such features can be combined with one or other features of other implementations that may be desired and advantageous for a given or specific application. Moreover, insofar as the terms "including", "having", "containing" or their variations are used in specific embodiments or claims, such terms are intended to be included in a manner similar to the term "comprising".

[0055] The functional units in the embodiments of the present invention may be integrated into a single processing module, or each unit may exist physically separately, or multiple or more units may be integrated into a single module. The aforementioned integrated module may be implemented in the form of hardware or in the form of a software functional module. If the integrated module is implemented in the form of a software functional module and sold or used as an independent product, it may also be stored in a computer-readable storage medium. The aforementioned storage medium may be a read-only memory, a magnetic disk, or an optical disk, etc. The aforementioned devices or systems may execute the storage method in the corresponding method embodiment.

[0056] In summary, the above embodiment is one implementation method of the present invention, but the implementation method of the present invention is not limited to the described embodiment. Any other changes, modifications, substitutions, combinations, and simplifications that deviate from the spirit and principles of the present invention should be equivalent replacement methods and are included in the scope of protection of the present invention.

Claims

1. A two-stage optimization method for carbon-constrained microgrid energy management, characterized by: The following steps are involved: Day-ahead stage: Establish power balance constraints for various components in the microgrid and a joint economic-low-carbon objective function, optimizing diesel engine power, grid transactions, load strategies, and energy storage scheduling. Decouple the system-level CVaR tail risk indicator into a short-term CVaR value based on its probability distribution and the time window area ratio, and transfer it to the intraday level as the carbon over-quota emission risk constraint boundary. Intraday rolling phase: Feedback the measured wind and solar load data to the forecast model for real-time correction every cycle; Build a robust optimization model to dynamically adjust the conservative coefficient , coordinate electricity price fluctuations, renewable energy deviations and load uncertainty; calculate the excess emission tail risk value within the window in real time , taking the carbon excess quota emission risk constraint boundary allocated in the day-ahead phase as the excess emission tail risk value within the window The upper limit forms a hard constraint; the execution layer adopts the first-stage decision and transmits the diesel engine climbing status and energy storage SOC to the next cycle, forming a "prediction-optimization-execution-feedback" closed loop.

2. The two-stage optimization method for carbon-constrained microgrid energy management according to claim 1 is characterized in that: The power balance constraints include constraints on wind and solar power generation, energy storage charging, diesel power generation, total power purchases from the grid, load power consumption, energy storage discharge, and total power sales from the grid: ； ； Among them, P wind It is the wind power generation, P pv It is the photovoltaic power generation, P DG Is the diesel generator, P BESS,dis is the day-ahead energy storage discharge, P grid，buy is the total amount of electricity purchased by the power grid today, P load is the day-ahead load electricity consumption, P BESS,ch For the day before energy storage charging, P grid，sell is the total amount of electricity sold by the power grid on the previous day; The economic-low-carbon joint objective function is composed of the operating cost of each device, the electricity price response cost and the total carbon emission cost: ； in, is the operation and maintenance cost of battery energy storage, is the operation and maintenance cost of the diesel generator, is the grid transaction cost, is the penalty cost of load reduction, is the electricity price response cost, is the total carbon emission cost.

3. The two-stage optimization method for carbon-constrained microgrid energy management according to claim 2 is characterized in that: The total carbon emission cost Including conventional carbon emission costs and excess carbon emission costs; conventional carbon emission costs , Excess carbon emission costs and total carbon emission costs The calculation is as follows: ； ； ； is the carbon price, is the excess penalty factor, is the carbon quota threshold, and T is the number of time granularity windows.

4. The two-stage optimization method for carbon-constrained microgrid energy management according to claim 3 is characterized in that: At the confidence level Day-ahead forecast conditional value at risk is defined as follows: ； in, Reflect more than The tail average loss of the quantile, Indicates The value at risk at the level is the kernel density probability estimation function.

5. The two-stage optimization method for carbon-constrained microgrid energy management according to claim 4 is characterized in that: The calculation of system-level CVaR tail risk indicators includes calculating the day-ahead CVaR value, decoupling the day-ahead CVaR, and calculating the excess emission tail risk value. .

6. The two-stage optimization method for carbon-constrained microgrid energy management according to claim 5, characterized in that: The calculation of the day-ahead CVaR value specifically includes: Based on the day-ahead optimization scheduling results, using carbon emission factors and emission control coefficients, Calculating carbon emissions at a time granularity and carbon quota threshold : Carbon quota threshold The calculation formula is as follows: ； t is time, P load is the total actual load demand energy; is the control and discharge coefficient, It is the time granularity; The carbon emissions in the microgrid take into account the direct carbon emissions of diesel generators and the indirect carbon emissions generated by power purchases from the grid. The carbon emissions at each time granularity are for: ； P DG is the electricity generated by the diesel generator a few days ago, P grid，buy is the total amount of electricity purchased by the power grid today, EF grid is the diesel engine carbon emission factor, EF DG is the grid carbon emission factor; Calculating excess carbon emissions : ； The kernel density estimation method is used to construct the carbon emission probability density function to circumvent the limitations of the conventional normal distribution assumption, including: First, determine any point on its estimated domain , a and b are the lower and upper bounds of the domain, and they are extended by 10 units on both sides of the data extreme value to ensure that the kernel density estimation can fully cover the tail characteristics of the carbon emission distribution and avoid the deviation of the probability density estimation caused by boundary truncation: ； ； Calculate the kernel density probability estimation function of excess carbon emissions on the previous day: ； In the formula, x is the function variable, and the kernel density estimation bandwidth is The calculation of is based on the Silverman criterion, which is as follows: ； ； is the variance, is the mean; Then calculate the day before , find the cumulative probability to reach The carbon emission threshold is: ； is the set of real numbers, u is the integration variable; Before the calculation date When , it is approximated by Monte Carlo integration, and the formula is as follows: ； in, For The simulated value of the sample, i is the sampling number, M is the value .

7. The two-stage optimization method for carbon-constrained microgrid energy management according to claim 6, characterized in that: The decoupling of day-ahead CVaR specifically includes: Based on the day-ahead CVaR forecast and its probability distribution characteristics, risk time periods are weighted according to the proportion of time window area and discretely mapped to T time granularity windows to support dynamic closed-loop control of risks during the day. Define the weight coefficient according to the integral area ratio : ； k is the prediction length of the intraday real-time rolling optimization layer; Decouple the day-ahead CVaR to the windowed CVaR: 。 8. The two-stage optimization method for carbon-constrained microgrid energy management according to claim 7, characterized in that: The calculation of excess emission tail risk value , specifically including: The intraday excess emission tail risk value is calculated according to the method for calculating the day-ahead CVaR value. The difference is that in the calculation of the kernel density probability estimation function f(x), the day-ahead prediction domain T is replaced by the intraday prediction domain k. During the intraday rolling optimization process, CVaR(t) is used as the upper limit of the dynamic risk constraint for the intraday rolling optimization, and real-time power adjustment is used to ensure that the risk of carbon over-quota emissions is always under control: 。 9. The two-stage optimization method for carbon-constrained microgrid energy management according to claim 8, characterized in that: During the intraday rolling optimization phase, the robust optimization model processes the errors in the forecasts of wind and solar output and key load demands: ； ； ； is the intraday wind speed forecast after robust processing, is the wind speed forecast for the day, is the wind prediction error variable, with a value between [-1, 1], is the intraday forecasted photovoltaic power after robust processing, is the intraday predicted photovoltaic power, is the photovoltaic prediction error variable, with a value between [-1, 1], is the intraday forecast load demand after robust processing, is the load demand forecasted for the day, is the load forecast error variable, with a value between [-1, 1], These are the wind, solar, and load forecast error boundary parameters, which are 15%, 10%, and 5%, respectively; In order to transform the optimization problem from a nonlinear programming problem that is difficult to solve into a linear problem and minimize the influence of uncertainty, auxiliary variables are introduced. , , , to express the upper bound of wind, solar, and load forecast power errors: ； And add a global constraint to limit the total error impact to not exceed a preset threshold: ； ； Where, yes , and The maximum value in is a conservative coefficient that dynamically tightens or relaxes the system tolerance according to the error intensity and is determined using the following formula: ； is the power balance relaxation calculated by the error upper bound auxiliary variable and the dynamic robustness coefficient; In order to fully consider the worst case scenario, slack is added to the power balance, which becomes an inequality constraint: ； in, It is the daily energy storage discharge, It is diesel power generation during the day. is the total amount of electricity purchased by the power grid during the day, It is a daily energy storage charging. is the total amount of electricity sold by the power grid during the day, is the maximum rated power of the load that can be cut, is the daily load shedding rate; Slack is added to the objective function of the intraday rolling optimization. The objective function consists of the deviation cost of the power grid, diesel, and energy storage, the power balance slack cost, and the risk term cost: ； in, is the deviation cost of the grid, is the deviation cost of diesel, is the deviation cost of energy storage, is the power balancing relaxation cost, is the cost of the risk term.

Images