Scenario-driven multi-stage robust re-optimization scheduling for v2g integrated community microgrid

Abstract

This application relates to scenario-driven multi-stage robust re-optimization scheduling for V2G integrated community microgrids. The method constructs a microgrid model including photovoltaic, energy storage, electric vehicles, and grid interaction units, establishing a multi-source uncertainty set; it constructs a multi-stage robust optimization model based on the principle of unpredictability and affine decision rules, and uses scenario sampling to construct the expected cost optimization objective; it achieves model solution through robust constraint deterministic transformation; and it combines day-ahead constraint relaxation and intraday rolling re-optimization strategies to restore physical feasibility while reducing computational complexity and locking the charge state trajectories of energy storage and electric vehicles. This application satisfies multi-stage robustness and the principle of unpredictability while reducing the over-conservatism caused by traditional worst-case scenario robust optimization, improving the economy, robustness, and real-time adaptability of V2G microgrid scheduling.

Description

Technical Field

[0001] This application relates to the field of smart grids and energy management, and in particular to scenario-driven multi-stage robust re-optimization scheduling for V2G integrated community microgrids. Background Technology

[0002] With the rapid development of distributed energy and electric vehicles (EVs), campus microgrids (such as industrial parks, commercial areas, campuses, hospitals, etc.) have become important platforms for integrating resources such as photovoltaic power generation, distributed generation equipment, energy storage systems, conventional loads, and vehicle-to-grid (V2G) interaction (see...). Figure 1 Vehicle-to-grid (V2G) technology enables electric vehicles to feed power back to the grid as distributed energy storage units when idle, thereby enhancing the operational flexibility and peak-shaving capabilities of microgrids. However, the optimal scheduling of microgrids faces multiple uncertainties: photovoltaic power generation fluctuates with weather conditions, regular loads change randomly, and the behavior of electric vehicles (such as the state-of-charge deviation at the time of grid connection) is difficult to predict accurately. These uncertainties make developing an economical and robust scheduling scheme a technical challenge.

[0003] Currently, there are two main technical approaches in the field of robust optimization for microgrid dispatch: (1) Two-stage robust optimization (2SRO) method Two-stage robust optimization is currently the most widely used robust scheduling technology for microgrids. Its technical framework is as follows: the first stage determines the basic plan (such as generator start-up and shutdown, and energy storage charging and discharging modes) during the day-ahead phase; the second stage reschedules continuous power based on uncertain realized values ​​during the intraday phase. To solve such models, the Column and Constraint Generation (C&CG) algorithm has become the standard solution framework. For example, the existing patent "Source-Load-Storage Microgrid Management Method and System with Clustered Electric Vehicles for the Worst-Case Scenario" (CN202411905383.7) constructs a two-stage, three-layer robust optimization model with a min-max-min structure. It divides the decision variables into first-stage variables and second-stage variables, aiming to minimize the daily operating cost of the microgrid, and uses the C&CG algorithm to iteratively solve the main problem and sub-problems. The existing patent "A Multi-Time-Scale Optimization Scheduling Method and Device for a Distribution Area Microgrid" (CN202510235879.6) further combines two-stage robust optimization with model predictive control, using the day-ahead plan value as a benchmark and employing intraday rolling adjustments to address source-load fluctuations. Another patent, “Two-stage robust optimization scheduling method, device, equipment and medium for microgrids” (CN202511702974.9), adopts a two-stage framework, with minimizing the total operating cost as the first objective function and minimizing the adjustment cost as the second objective function.

[0004] However, the two-stage robust optimization framework has a fundamental flaw: its second-stage decision-making model assumes that it can fully realize uncertainties across the entire scheduling time domain. However, in actual scheduling, decision-makers can only make decisions based on the uncertainties revealed up to the current moment, and cannot predict uncertainties in future periods. This discrepancy between the modeling assumption and actual operation makes 2SRO theoretically violate the principle of unpredictability in operations research—that is, decisions cannot rely on unrevealed future information. From a scheduling practice perspective, this theoretical flaw may lead to the intraday adjustment schemes output by the two-stage robust optimization model facing the dilemma of incomplete information during actual implementation.

[0005] (2) Multistage Robust Optimization (MSRO) method To overcome the theoretical shortcomings of two-stage robust optimization (2SRO) regarding the principle of unpredictability, the academic community has recently turned its attention to multi-stage robust optimization methods. MSRO divides the scheduling time domain into multiple consecutive stages, with each stage's decision relying solely on historical uncertainty information revealed up to that stage, thus enforcing the principle of unpredictability from the model structure. Through this information-progressive decision-making structure, the MSRO decision-making process is completely consistent with the actual scheduling information flow, effectively avoiding the theoretical shortcomings of 2SRO. Currently, MSRO has shown initial application potential in the field of microgrid scheduling. Related research mainly focuses on the following directions: One approach is a multi-stage adaptive stochastic-robust scheduling method for hydrogen-based multi-energy microgrids. (Zhou et al.) [1] This paper proposes using affine strategies to describe the complex relationship between decision variables and the realization of uncertainties. Combining stochastic programming and multi-stage robust optimization, a solvable MILP model with expected operating costs as the objective is established. Based on this, real-time rolling and non-rolling economic scheduling models are further proposed to meet the requirements of economic efficiency and computational efficiency, respectively, while ensuring the unpredictability of the solution. Numerical tests verify that the proposed method can simultaneously guarantee the feasibility and economy of the solution in the actual operation of a hydrogen-based multi-energy microgrid.

[0006] Second, a method for optimizing decision rules in microgrids with energy storage communities. (Tang et al.) [2] The study points out that energy storage scheduling is essentially a multi-stage decision-making problem, which must ensure the unpredictability of decisions and multi-stage robustness (feasibility across all scenarios). Addressing the issue that mainstream affine decision rules may narrow the feasible region, this research proposes a method based on the necessary and sufficient feasibility conditions for scheduling decisions under a polyhedral uncertainty set. This method can find a feasible solution when the affine decision rule fails, and a better solution when both fail.

[0007] Thirdly, a robust multi-stage scheduling framework for distributed energy uncertainties. Recent research. [3]A multi-stage robust optimization model framework was constructed, and the scheduling decision was linearly approximated by linear decision rules. The solution efficiency of the two-stage optimization model was improved by column generation algorithm. The correctness and effectiveness of the proposed model framework were verified in a microgrid distribution system.

[0008] Fourth, the MSRO model based on affine decision rules. Affine decision rules are the mainstream technical means to deal with MSRO. They represent continuous decision variables as linear functions of historical uncertainty parameters, thereby transforming the infinite-dimensional decision space into a finite-dimensional parameterized expression. Zhou et al. proposed a multi-stage adaptive stochastic-robust scheduling method combining affine strategies for hydrogen-based multi-energy microgrids, and established a solvable MILP (Mixed Integer Linear Program) model with expected operating costs as the objective [4]. Tang et al. studied the necessary and sufficient feasibility conditions for energy storage scheduling under affine decision rules for community microgrids with energy storage, ensuring the unpredictability and multi-stage robustness of the decision [5].

[0009] Fifth, a robust multi-stage scheduling framework for distributed energy uncertainty. Recent studies have constructed a multi-stage robust optimization model framework, which uses linear decision rules to linearly approximate scheduling decisions and improves solution efficiency through column generation algorithms. The effectiveness of the model has been verified in a microgrid distribution system [6]. Another study introduces uncertainty set decomposition technology into MSRO to decompose the complex multi-stage scheduling problem into independent sub-problems to improve computational efficiency [7].

[0010] Based on the above-mentioned current state of industry technology, existing technologies mainly suffer from the following technical deficiencies: Question 1: Two-stage robust optimization violates the principle of unpredictability and is disconnected from the actual scheduling process. In the field of V2G integrated microgrid dispatching, existing two-stage robust optimization methods simplify the dispatching process into a two-stage model of "first determining the variables of the first stage, and then determining the variables of the second stage based on the full-time uncertainty." The second-stage decision is assumed to be based on complete information about future uncertainties. This violates the principle of unpredictability in mathematical modeling—the dispatch center can only make decisions based on the photovoltaic output, load demand, and EV connection status observed up to time t, and cannot rely on information not yet revealed in time t+1 and beyond. This deficiency leads to a discrepancy between the model's theoretical assumptions and the information flow of actual operational decisions, potentially requiring frequent manual intervention in actual implementation due to information mismatch.

[0011] Question 2: V2G integrated microgrid dispatch lacks a multi-stage robust optimization framework. Existing MSRO research has failed to incorporate V2G technology and the uncertainties of electric vehicle behavior into the scheduling framework. Currently, MSROs primarily focus on hydrogen microgrids, energy storage, and source / load uncertainty management, without considering V2G as a dispatchable resource. The core of V2G lies in the fact that electric vehicles serve as both loads and energy storage; however, their access / departure times, initial / expected state of charge (SOC), and response randomness are subject to multiple uncertainties, significantly impacting scheduling robustness and economics. Existing MSROs lack system modeling and collaborative optimization for these uncertainties, resulting in the insufficient development of V2G integration scenarios.

[0012] Question 3: Traditional robust optimization targets the "worst-case scenario," which is overly conservative. Existing worst-case robust optimization (MSRO) methods use cost minimization under the worst-case scenario as the objective function (i.e., a min-max structure), ensuring that the scheduling scheme satisfies constraints under all possible uncertainties. However, worst-case scenarios (such as simultaneous minimum photovoltaic output and simultaneous maximum load demand) rarely occur simultaneously in actual operation. This strategy of "paying for extreme cases" leads to the system being in a state of excessive reserve capacity for a long time, resulting in operating costs significantly higher than actual needs. Experimental data shows that traditional worst-case robust optimization has operating costs 54%-62% higher than the proposed scenario-driven method, which directly reflects the degree of economic loss of existing methods.

[0013] Question 4: Binary variables in the MSRO model lead to NP-hard problems with extremely low solution efficiency. In the MSRO model employing affine decision rules, binary variables (such as charging / discharging signals and electricity purchase / sale signals) cannot be incorporated into the affine expression and must be determined as "here-and-now" decisions during the day-ahead phase. This results in the day-ahead optimization model remaining in the form of mixed-integer linear programming (MILP), whose solution complexity increases exponentially with the number of binary variables. For a typical V2G microgrid example involving 100 electric vehicles (EVs) and 24-hour time periods, the number of binary variables can reach over 2400. Directly solving the MILP often takes several hours or even fails to converge, making it difficult to meet the computational efficiency requirements of actual scheduling.

[0014] Question 5: The linear assumption of affine decision rules leads to economic losses. Affine decision rules are essentially linear approximations and cannot be applied to binary decision variables (such as charging / discharging signals or electricity purchase / sale signals). Binary decision-making loses the flexibility to adjust for uncertainty, limits the pursuit decision space, and restricts the optimality of the pursuit decision. Furthermore, the linear assumption of affine rules restricts the scope of the decision space exploration, preventing the scheduling scheme from achieving optimal economic efficiency.

[0015] Question 6: Existing MSROs lack an effective intraday feasibility recovery mechanism. Existing MSRO research mainly focuses on day-ahead modeling and solving, lacking a systematic solution for how to translate affine decisions into physically executable instructions in actual intraday operations. When affine decisions generate instructions that violate physical constraints (such as simultaneous purchase and sale of electricity and charging / discharging), existing methods often employ simple truncation, which undermines the theoretical optimality of affine decisions.

[0016] In summary, existing V2G microgrid scheduling methods based on MSR have significant shortcomings in terms of computational efficiency, unpredictability, physical feasibility, and economy, and there is an urgent need for an improved method that can solve these problems simultaneously. Summary of the Invention

[0017] To address the technical challenges of strictly adhering to the principle of unpredictability, appropriately reducing the conservatism of robust optimization, and organically integrating multi-stage decision-making structures in V2G integrated microgrid scheduling methods, this application provides scenario-driven multi-stage robust re-optimization scheduling for V2G integrated community microgrids.

[0018] The scenario-driven multi-stage robust re-optimization scheduling for V2G integrated community microgrids provided in this application adopts the following technical solution: A scenario-driven, multi-stage robust re-optimization scheduling method for V2G integrated community microgrids includes the following steps: S1. Construct a deterministic scheduling model for V2G integrated community microgrids; Obtain the operating parameters of photovoltaic power generation units, controllable distributed generation units, energy storage systems, electric vehicle aggregation units, and main grid interaction units in the community microgrid, and establish a deterministic scheduling model that includes equipment power constraints, state of charge constraints, power balance constraints, and operating cost models; The electric vehicle aggregation unit includes multiple electric vehicles operating in a vehicle-to-grid (V2G) mode. These electric vehicles participate in microgrid power regulation as rechargeable and discharging mobile energy storage resources based on their access status. S2. Construct a multi-stage uncertainty set that includes multiple sources of uncertainty factors; The prediction errors of photovoltaic power generation, load forecasting, and initial state of charge of electric vehicles are obtained, and corresponding uncertain variables are established respectively. A multi-stage uncertainty set is constructed based on the historical statistical data of each uncertain variable. The multi-stage uncertainty set is used to describe the range of deviation between the actual operating state and the predicted state in different scheduling stages; S3. Establish a multi-stage robust optimization scheduling model based on the principle of unexpectedness; The scheduling cycle is divided into multiple consecutive decision-making stages, so that the scheduling decision in any stage depends only on the uncertainty information already obtained in the current stage and the historical stages, and not on the uncertainty information not yet revealed in the future stages. S4. Construct an optimization objective based on scenario-driven expected cost; Within the uncertainty set, a random scenario sampling method is used to generate multiple operating scenarios, and the expected value of the operating cost under multiple scenarios is used as the optimization objective. The optimization objective does not adopt the worst-case scenario cost minimization method, but a scenario-driven expected operating cost optimization objective in order to reduce the over-conservatism of traditional robust optimization. S5. Perform deterministic transformation on the multi-stage robust optimization model; For power balance constraints, equipment operation constraints, and state of charge constraints that contain uncertain variables, the infinite-dimensional robust constraints are transformed into finite-dimensional deterministic linear constraints by using the linear dual transformation method in robust optimization. By introducing dual variables, the transformed model becomes a solvable mixed-integer linear programming model. S6. Perform day-ahead constraint relaxation and intraday rolling re-optimization scheduling; In the current stage, the binary variables in the mutual exclusion constraints of electric vehicle charging and discharging, energy storage charging and discharging, and main grid power purchase and sale are relaxed, the mixed integer programming model is transformed into a linear programming model, and the affine decision parameters are obtained by solving the model. During the intraday phase, affine scheduling results are calculated based on the real-time acquired uncertainty realization values, and a single-period rolling optimization model is established based on the recovered physical mutual exclusion constraints to re-optimize the scheduling instructions for the current phase. Meanwhile, the energy storage state-of-charge trajectory and the electric vehicle state-of-charge trajectory calculated in the day-ahead phase are locked as global constraints in order to obtain a final dispatch instruction that meets physical constraints and is economical.

[0019] By adopting the above technical solutions, a multi-stage robust re-optimization scheduling system for V2G integrated community microgrids is constructed, covering photovoltaic power generation, energy storage systems, electric vehicle aggregation units, and main grid interaction units. This system achieves unified modeling and dynamic response to multi-source uncertainties. The system utilizes the principle of unpredictability and affine decision rules to dynamically adjust scheduling decisions based on historical information, ensuring that decisions at each stage conform to the actual operation process. By combining scenario-driven expected cost optimization and robust constraint deterministic transformation, the system improves the solution efficiency of complex optimization models. Through a combination of day-ahead constraint relaxation and intraday rolling re-optimization, the system achieves real-time correction and global collaborative optimization of the scheduling scheme. Furthermore, by locking the charge state trajectories of energy storage and electric vehicles, the system ensures physical feasibility and energy continuity, thereby enhancing the economy, robustness, real-time scheduling capabilities, and renewable energy absorption capacity of the V2G integrated community microgrid under multi-source uncertain environments.

[0020] Optionally, the deterministic scheduling model includes a photovoltaic power generation model, a load power model, and an electric vehicle initial state of charge model.

[0021] By adopting the above technical solutions, photovoltaic power generation, load demand, and the initial state of charge of electric vehicles are uniformly established as a mathematical model including predicted values ​​and prediction errors. This enables unified modeling and quantitative description of multi-source uncertainties in the community microgrid, accurately reflecting the impact of renewable energy output fluctuations, load changes, and the randomness of electric vehicle access on system operation. This provides accurate uncertainty input for subsequent multi-stage robust optimization. Simultaneously, by parameterizing the prediction error and integrating it into the deterministic scheduling model, the optimization algorithm can consider both predicted information and actual operational deviations, improving the scheduling model's adaptability to complex operating environments and enhancing the robustness and reliability of scheduling results. This provides a reliable data foundation and model support for efficient renewable energy consumption, energy storage collaborative control, and V2G interactive regulation, further improving the economy, security, and real-time scheduling performance of the community microgrid.

[0022] Optionally, both the energy storage system and the electric vehicle aggregation unit are equipped with charge / discharge mutual exclusion control variables; Wherein: when the charging state variable is in the first state, only charging power is allowed to exist; When the charging state variable is in the second state, only the discharge power is allowed to exist; This is to avoid energy storage systems or electric vehicles performing charging and discharging operations simultaneously.

[0023] By adopting the above technical solutions, a charging and discharging mutual exclusion control mechanism is established for energy storage systems and electric vehicle aggregation units. This ensures that only charging or discharging operations are performed at the same time, effectively avoiding energy conflicts and model distortions caused by simultaneous charging and discharging. It also improves the physical feasibility of scheduling results and the safety of equipment operation, enhances the consistency between the optimization model and the actual operating state, and provides a guarantee for the stable and efficient operation of the community microgrid.

[0024] Optionally, the multi-stage uncertainty set is constructed using a box-type uncertainty set; The length of the historical information window in the affine decision rule is adaptively determined based on the decision dependency coefficient; Specifically, the dependency coefficient of the current stage decision variable on the historical stage uncertain variable is calculated, the effective historical influence range is determined based on the dependency coefficient, and the historical window length is determined.

[0025] By adopting the above technical solution, the prediction errors of photovoltaic power output, load demand and electric vehicle state of charge are uniformly constrained by the box uncertainty set, so as to achieve effective characterization of multi-source uncertainty. At the same time, the length of the historical information window is adaptively determined according to the decision dependency coefficient, so that the affine decision rule can dynamically match the effective historical information. While ensuring the robustness of the model, the computational complexity is reduced, and the solution efficiency, decision accuracy and real-time response capability of multi-stage optimization scheduling are improved.

[0026] Optionally, the scene sampling employs the Monte Carlo sampling method to generate multiple uncertainty realization samples within the uncertainty set; The multiple samples are used only to calculate the expected operating cost in the objective function, without increasing the number of robust constraints; The aforementioned day-ahead constraint relaxation includes: Delete the charging / discharging state variables of electric vehicles; Delete the energy storage charge / discharge state variables; Delete the main grid power purchase and sale status variable; The corresponding constraints are then converted into independent power upper and lower limit constraints.

[0027] By adopting the above technical solutions, Monte Carlo sampling is used to generate various uncertain operating scenarios, which can accurately assess the expected operating cost while avoiding increasing the scale of robust constraints and reducing the solution complexity of the optimization model. By relaxing the charging and discharging and the mutual exclusion constraints of power purchase and sale in the day-ahead stage, the mixed integer programming is transformed into linear programming, which improves the solution efficiency and provides a high-quality initial scheduling scheme for intraday rolling optimization, taking into account both computational efficiency and scheduling economy.

[0028] Optionally, the intraday rolling re-optimization includes: obtaining the actual photovoltaic output, actual load, and actual state of charge of electric vehicles at the current stage; Calculate the initial scheduling power based on the affine parameters obtained in the previous phase; Reintroduce charging and discharging mutual exclusion constraints and buying and selling electricity mutual exclusion constraints; Establish a mixed-integer optimization subproblem at the current time scale; The final execution power of the current stage is obtained by solving the problem.

[0029] By adopting the above technical solutions, combining real-time operation data to perform rolling corrections on the day-ahead scheduling results, reintroducing physical mutual exclusion constraints and establishing a mixed integer optimization subproblem, dynamic optimization and precise control of the current stage scheduling instructions can be achieved, improving the adaptability of scheduling results to the actual operating state, taking into account the economy, physical feasibility and real-time response performance of system operation, and enhancing the safe and stable operation capability of the community microgrid.

[0030] Optionally, the state of charge trajectory locking includes: calculating the SOC change trajectory of the energy storage system and the electric vehicle based on the day-ahead affine scheduling results; During the intraday rolling optimization process, the SOC change trajectory is used as a constraint to limit the cumulative energy shift caused by the local optimization process.

[0031] By adopting the above technical solution, the state-of-charge trajectories of the energy storage system and electric vehicles obtained by day-ahead optimization are used as intraday rolling optimization constraints to limit the cumulative energy shift caused by local optimization, ensure the continuity and global consistency of state-of-charge changes, improve the physical feasibility and execution reliability of scheduling results, and take into account the overall economy and long-term stable operation of the system.

[0032] Secondly, a scenario-driven multi-stage robust re-optimization scheduling system for V2G integrated community microgrids includes a model building module, which is used to establish a microgrid scheduling model that includes photovoltaic, load, energy storage, electric vehicles and grid interaction units. It includes an uncertainty analysis module, which is used to acquire and construct a set of uncertainties related to photovoltaic, load, and electric vehicle behavior; It includes a multi-stage robust optimization module, which is used to establish a multi-stage robust scheduling model based on the principle of unpredictability; It includes a scene optimization module, which is used to construct a desired cost optimization target based on scene sampling; It includes a robust transformation module, which is used to transform robust constraints containing uncertainties into deterministic optimization constraints; It includes a rolling optimization module, which is used to perform day-ahead relaxation optimization and intraday recovery optimization.

[0033] By adopting the above technical solutions, the functional modules work together to complete microgrid modeling, uncertainty analysis, multi-stage robust optimization, scenario-driven optimization, robust constraint transformation, and rolling re-optimization scheduling. This enables intelligent scheduling and control of the entire process of V2G integrated community microgrids under multi-source uncertainty, improving the solution efficiency of the optimization model, the robustness of the scheduling scheme, the real-time response capability, and the economic efficiency of system operation, thus ensuring the safe, stable, and efficient operation of the community microgrid.

[0034] Thirdly, an electronic device includes: a processor; a memory; and a computer program stored in the memory and executable by the processor; When the computer program is executed, it causes the processor to execute the scenario-driven multi-stage robust re-optimization scheduling method for V2G integrated community microgrids as described in any one of claims 1-8.

[0035] By adopting the above technical solutions, electronic devices can execute scenario-driven multi-stage robust re-optimization scheduling methods for V2G integrated community microgrids, realize microgrid operation status analysis, multi-source uncertainty handling, optimized scheduling decision-making and rolling control, improve the automatic execution capability, real-time computing capability and engineering deployment capability of scheduling algorithms, and enhance the economy, robustness, stability and intelligent management level of community microgrid operation.

[0036] Fourthly, a computer-readable storage medium storing a computer program; When the computer program is executed by the processor, it implements the scenario-driven multi-stage robust re-optimization scheduling method for V2G integrated community microgrids as described in any one of claims 1-9.

[0037] By adopting the above technical solutions, the computer-readable storage medium can store and support the processor to execute a scenario-driven multi-stage robust re-optimization scheduler for V2G integrated community microgrids, realize the software deployment and automatic operation of the scheduling method, improve the system's portability, scalability and engineering application capabilities, ensure the stable and efficient execution of the optimization scheduling algorithm, and enhance the economy, robustness and intelligence level of community microgrid operation.

[0038] In summary, this application includes at least one of the following beneficial technical effects: This application constructs a scenario-driven multi-stage robust re-optimization scheduling method for V2G integrated community microgrids. By integrating photovoltaic power generation, energy storage systems, electric vehicle aggregation units and main grid interaction units, it realizes multi-source energy collaborative optimization scheduling, improves the overall operating efficiency and renewable energy consumption capacity of community microgrids in complex operating environments; This application establishes a multi-stage uncertainty set including photovoltaic prediction error, load prediction error and electric vehicle initial state of charge prediction error, and combines the principle of unpredictability and affine decision rules to realize the dynamic adjustment of scheduling decisions with historical information, improve the adaptability and robustness of the scheduling model to multi-source uncertainties, and reduce the impact of prediction errors on system operation. This application adopts scenario-driven expected operating cost as the optimization objective, and combines Monte Carlo scenario sampling and robust constraint deterministic transformation to effectively reduce the over-conservatism brought about by traditional worst-case scenario robust optimization while ensuring system robustness, thereby improving the solution efficiency and operating economy of the optimization model. This application adopts a scheduling strategy that combines day-ahead constraint relaxation with intraday rolling re-optimization. In the day-ahead stage, it reduces the computational complexity of the mixed integer optimization model, and in the intraday stage, it combines real-time running data to restore physical constraints and dynamically correct the scheduling scheme, thereby improving the real-time response capability, execution accuracy and engineering application performance of the scheduling results. This application establishes a state-of-charge trajectory locking mechanism for energy storage systems and electric vehicles to limit the cumulative energy shift caused by intraday local optimization, ensure the continuity, global consistency and physical feasibility of state-of-charge changes, and improve the long-term stable operation capability of the system and the reliability of the scheduling scheme. This application also provides a corresponding scheduling system, electronic equipment, and computer-readable storage medium to realize the software-based, modular, and intelligent deployment of the scheduling method, improve the portability, scalability, and engineering application capabilities of the algorithm, and provide a complete technical implementation solution for the intelligent scheduling and operation management of V2G integrated community microgrids. Attached Figure Description

[0039] Figure 1 This is the system architecture diagram of the V2G integrated community microgrid in this application.

[0040] Figure 2 This is a schematic diagram of sequential decision-making in the multi-stage robust scheduling of this application.

[0041] Figure 3 This is a schematic diagram illustrating the impact of the historical time window Δt on MSRO performance in the affine decision-making process of this application.

[0042] Figure 4 This is a heatmap visualizing the dependency coefficients of this application.

[0043] Figure 5 This is the overall framework diagram of the two-stage collaborative scheduling before and during the application period.

[0044] Figure 6 This is a schematic diagram illustrating the extension of the feasible region by the constraint relaxation in this application.

[0045] Figure 7This is a schematic diagram showing that the net charge and discharge power remains unchanged during the rolling re-optimization of this application.

[0046] Figure 8 This is a comparison chart of the computational efficiency and cost of this application. Detailed Implementation

[0047] The following is in conjunction with the appendix Figure 1-8 This application will be described in further detail.

[0048] This application discloses a scenario-driven multi-stage robust reoptimization scheduling method for V2G integrated community microgrids. The proposed scenario-driven MSRO scheduling method divides the scheduling time domain into multiple continuous stages (e.g., 24 hours, with each hour as a stage). Decisions in each stage rely solely on the uncertainties revealed up to that stage, strictly adhering to the principle of unpredictability. An affine decision rule is used to describe the linear relationship between continuous decision variables and historical uncertainties, ensuring the model's computational tractability. Simultaneously, multiple scenarios are generated within the uncertainty set through Monte Carlo sampling, using the expected operating cost (rather than the worst-case scenario cost) as the optimization objective, significantly reducing the conservatism of the scheduling scheme while maintaining robustness. The established model is transformed into a deterministic mixed-integer linear programming (MILP) problem through linear duality transformation and the strong duality theorem, which can be efficiently solved using commercial solvers.

[0049] This technical solution further proposes a day-ahead relaxation planning and intraday rolling re-optimization strategy. In the day-ahead stage, the binary mutual exclusion constraints are relaxed, transforming the original MSRO model into an LP problem, which efficiently solves the affine decision parameters (baseline and adjustment coefficients). In the intraday stage, based on real-time uncertainty, the decision quantities are initially calculated according to the affine rules, and then a single-period MILP sub-problem with tight constraints is constructed for rolling re-optimization, outputting the final command. The entire process locks the global SOC trajectory of energy storage and EV, avoiding local optimization that damages the overall economy.

[0050] The technical solution of this patent is explained in detail in six steps below.

[0051] 1.1 Step 1: Constructing a deterministic scheduling model for V2G integrated microgrids The mathematical model for V2G integrated microgrids includes the following core elements.

[0052] 1.1.1 System Components and Variable Definitions (1) Photovoltaic power generation (PV): actual output

[0053] equal to the predicted value Adding prediction bias Forecast bias reflects uncertainties caused by factors such as weather.

[0054] (2) Conventional load: Actual load

[0055] equal to the predicted value Adding prediction bias Load fluctuations stem from the randomness of users' electricity consumption behavior.

[0056] (3) Controllable distributed generators (such as diesel generators): output

[0057] Available in range Internal continuous regulation is used to balance the power gap in the microgrid.

[0058] (4) Energy Storage System (ESS): Charging Power Discharge power Scope

[0059] Rated capacity Charge and discharge efficiency SOC operating range:

[0060] Energy storage can be charged when electricity prices are low and discharged when electricity prices are high, thus enabling arbitrage.

[0061] (5) Electric vehicle aggregation: A car. The first... Battery capacity of EVs Charging power and discharge power scope

[0062] Charge and discharge efficiency Access period

[0063] Access SOC prediction value Offline target SOC is EVs can achieve this through V2G technology. As a mobile energy storage unit.

[0064] (6) Interaction with the main grid: Purchased power Electricity sales capacity ,price Power limits The cost of purchasing electricity is higher than the revenue from selling it, which reflects the business model of power grid companies.

[0065] 1.1.2 Binary Mutual Exclusion Variables Three types of binary variables are introduced to enforce physical mutual exclusion constraints, ensuring that devices cannot be in the same state simultaneously. Two opposite working states: : No. EVs during the period The charging and discharging states, Indicates charging. This indicates a discharge. The variable Ensure that EVs cannot be charged and discharged simultaneously.

[0066] Energy storage during the period The charging and discharging states, Indicates charging. This indicates a discharge.

[0067] The interaction status between the micronet and the mainnet. Indicates the purchase of electricity. This indicates electricity sales. Avoid simultaneous buying and selling. Power cycle caused by electricity.

[0068] 1.1.3 Key Constraints (1) EV charging and discharging power constraints (mutual exclusion constraints) The following two formulas jointly guarantee: when When =1, the EV can only be charged (discharge power is forced to 0). when When the power is 0, the EV can only discharge (charging power is forced to 0). (1) (2) Physical meaning: Charging power is non-negative, discharging power is non-positive; binary variable. Control the direction in which power is allowed to pass.

[0069] (2) EVSOC dynamics and boundary constraints SOC (State of Charge) is the percentage of remaining energy in an EV's battery capacity. The following two equations describe the change of SOC over time and ensure that it operates within a safe range.

[0070] (3) Note: Energy increases during charging (multiplied by efficiency). During discharge, energy decreases (divided by the effective energy). Rate ). The length of each time period (e.g., 1 hour). This constraint ensures that the State of Charge (SOC) does not exceed the electrical limit. The physical safety range of the pool (typically 20%~95%).

[0072] (4) Note: Users typically expect their State of Charge (SOC) to remain above a certain target value (e.g., 80%) when offline. This requirement must be enforced to ensure a good user experience.

[0074] (3) ESS charging and discharging power constraints Similar to EVs, the charging and discharging of ESS is also controlled by binary variables. Control mutual exclusion.

[0075] (5) (6) (4) ESS SOC Dynamics and Boundaries (7) illustrate: This is the initial SOC (State of Charge) for energy storage. This constraint prevents overcharging or over-discharging of the energy storage, extending battery life.

[0077] (8) Note: To ensure dispatchability the following day, the final SOC of energy storage is usually required to be no less than a certain set value. .

[0078] (5) Controllable generator output constraint (9) Note: The generator output is limited by its rated capacity and must also meet the minimum stable operating output requirement.

[0079] (6) Main network interaction power constraints (10) (11) illustrate: When the value is 1, only electricity can be purchased. When the value is 0, only electricity can be sold, to avoid simultaneously buying and selling electricity from the grid. This results in unnecessary power cycling.

[0080] (7) Power balance constraints (12) Note: This formula represents the core constraint of energy conservation in microgrids. The left side represents the sum of load and energy storage / EV charging demand (note...). This represents the electricity sold, and is a negative value. (Also a negative value), the right side is the sum of the power generation and power purchase. The equality ensures real-time power balance in the system.

[0081] in This represents the set of EVs that are in the access state during time period t.

[0082] 1.1.4 Definition of Operating Costs (1) Energy storage cost: including the cost of battery aging caused by charge and discharge cycles.

[0083] (13) , Cost coefficient per unit charge / discharge power (yuan / kWh).

[0084] (2) Generator cost: usually fuel cost, which is proportional to the output.

[0085] (14) Cost coefficient per unit of power generation (yuan / kWh).

[0086] (3) Main grid interaction cost: positive cost (expenditure) when purchasing electricity, and negative cost (revenue) when selling electricity.

[0087] (15) Among them, electricity sales price Usually lower than the electricity purchase price .

[0088] (4) Minimize total operating cost: (16) Step 2: Constructing a multi-time-period uncertainty set The prediction errors for photovoltaic output, conventional load, and EV-connected SOC are respectively denoted as... , and The relationship between the actual value and the predicted value is as follows: (17) Note: Forecast error reflects the deviation between day-ahead forecasts and actual operations. These deviations are a major source of challenges to scheduling robustness.

[0089] A box-type uncertainty set is used to describe the fluctuation range of each uncertain parameter. This set is simple in form, easy to process, and guarantees coverage of extreme cases.

[0090] (18) (19) (20) illustrate: , , These represent the maximum possible absolute values ​​of the prediction error. These bounds are obtained from historical data (e.g., taking 3σ or historical extreme values). These bounds define the "worst-case" for uncertain parameters and are the basis for robust optimization.

[0091] Organize all uncertainty parameters of time period t into a vector The set of uncertainties can be compactly represented as: (twenty one) in The coefficient matrix, A constant vector, Let be the number of uncertain parameters for time period t. This compact form facilitates subsequent dual transformation processing.

[0092] 1.3 Step 3: Establishing a multi-stage robust optimization model based on affine decision rules (core innovation) 1.3.1 Mathematical Expression of the Principle of Unexpectedness Let the historical uncertainty sequence observed up to time period t be: Formula 22: (twenty two) The principle of unpredictability: In actual scheduling, when the scheduling center makes a decision at time period t, it can only use historical information observed up to that time period, and cannot predict future uncertainties (see...). Figure 2 Mathematically, it requires: (twenty three) Note: This principle distinguishes between "multi-stage" and "two-stage" robust optimization. The two-stage method assumes that the second-stage decision can rely on full-time uncertainty, which is unrealistic; the multi-stage structure of this patent forces the decision for each time period to rely only on historical information, which is completely consistent with the actual scheduling information flow.

[0093] 1.3.2 Introduction of Affine Decision Rules To convert the infinite-dimensional decision space (all possible function forms) into a finite-dimensional parametric representation, this patent adopts the Affine Decision Rule. This rule assumes a linear relationship between decision variables and uncertainties, and is a mainstream method in operations research for handling multi-stage robust optimization.

[0094] Take the charging power of energy storage as an example: (24) Description: is a constant term, corresponding to the day-ahead baseline plan without considering uncertainties. It represents the "default" scheduling instruction of the system under the prediction scenario.

[0095] is a row vector (or matrix), which reflects the linear adjustment amplitude of the decision at time period t to the historical time periods realized values of uncertainties. When =t, it means making an immediate response to the newly revealed uncertainty in the current time period; when <t, it represents the "memory" or "lag adjustment" of the decision to past uncertainties.

[0096] is the length of the historical window, which represents the most recent time periods of historical information referenced by the decision. This parameter balances the adaptability of the decision and the complexity of the model.

[0097] All continuous decision variables ( etc.) are combined into a vector , and the general formula of its affine expression is: (25) In the formula: --Day-ahead baseline plan vector; --Slope coefficient matrix. Each row of this matrix corresponds to a decision variable, and each column corresponds to an uncertainty parameter. The absolute value of the matrix element reflects the sensitivity of the decision to the corresponding uncertainty parameter.

[0098] 1.3.3 Processing of Binary Variables Since the affine decision rule is essentially a linear function approximation, it is not applicable to binary (0-1) variables -- because the continuous value output by the linear function cannot be guaranteed to be 0 or 1. Therefore, in this patent, the binary variable (including , , The decision is determined as a "here-and-now" fixed decision in the current phase and is not used as an online adjustment variable after the realization of uncertainty. (26) Note: This means that the charging / discharging mode and the power purchase / sale mode are preset before the day, and intraday scheduling only adjusts the power amplitude without switching modes. This design ensures model solvability and conforms to practical engineering practices—mode switching usually requires a long decision lead time.

[0099] 1.3.4 Scenario-Driven Expected Cost Objective Function This patent abandons the traditional robust optimization objective of "minimizing the cost of the worst-case scenario" and instead adopts scenario-driven expected operating cost as the objective function. The motivation is that worst-case scenarios rarely occur simultaneously, and "over-insurance" for extreme cases will lead to a significant decrease in economic efficiency; while the expected cost objective can better reflect the average performance of the system in long-term operation.

[0100] Traditional worst-case objective (as a benchmark): (27) This objective requires the scheduling scheme to be feasible under all possible uncertainties and to optimize the cost in the worst-case scenario. While this guarantees extreme robustness, it is often overly conservative.

[0101] The scenario-driven objective adopted in this patent is: (28) After substituting the affine decision rule, the objective function expands to: (29) in: S -- The total number of scenes generated through Monte Carlo sampling. Scenes should be within the uncertainty set. The inner part is generated according to the actual probability distribution (such as truncated normal distribution).

[0102] --Scenes in time period The uncertainty of the realized value.

[0103] --Cost coefficient vector (composed of coefficients for energy storage cost, generator cost, and main grid interaction cost).

[0104] Key feature: The scenario is only used to construct the expected cost in the objective function and does not enter the constraints. This means: 1) The model size does not increase significantly with the number of scenarios S (the number of constraints is independent of S); 2) Decision variables This is shared across all scenarios, ensuring the robustness of the scheduling strategy—the same affine decision rule must be defined for all possible uncertainties (although the objective function is the expected value, the constraints have been guaranteed for all scenarios through subsequent robust transformations). (Feasibility). Technical results: Compared with the worst-case objective, the expected cost objective can reduce the average operating cost by about 54%~62% (see experimental data); compared with traditional stochastic programming (scenario entry constraint), this method avoids the curse of dimensionality and maintains computational tractability.

[0105] 1.4 Step Four: Deterministic Transformation of Robust Constraints (Ensuring Solution Feasibility) After introducing uncertain parameters and affine decision rules, the constraints of the original deterministic model become semi-infinite constraints—that is, the constraints must be applied to all possible... This is true. Such problems cannot be solved directly and must be transformed into a finite number of deterministic constraints through dual transformations in robust optimization theory.

[0106] 1.4.1 Handling Equality Constraints The power balance equation constraint (Equation (12)) is written in a compact form: (29) illustrate: Let be the row vector of coefficients for the decision variables. Let be the row vector of coefficients for the uncertainty parameters. This is a constant term (such as the difference between the predicted load and the predicted photovoltaic). This equation requires that the power balance must be precisely satisfied for every possible uncertainty.

[0107] Substituting the affine decision rule and rearranging, we get: (30) Derivation logic: The left side of the above equation is A linear combination of , where the right side is a constant. To make this equation true for any If all of these are true, the only way is to make all of them true. The coefficient is zero, and the constant term matches. Otherwise, it can be adjusted... This makes the equation false. From this, we obtain a set of deterministic linear equality constraints: (31) Physical meaning: First principle: Uncertainty in historical periods ( The power balance equation cannot appear because the power balance must be satisfied immediately in each time period and cannot rely on past uncertainties to "compensate" – this is another manifestation of the principle of unpredictability.

[0108] Second principle: Uncertainty in the current time period The coefficient must be with Precise offsetting means that the system responds instantly to current uncertainties by adjusting its decisions in order to maintain power balance.

[0109] Third Form: Baseline Planning Power balance must be satisfied under the predicted scenario.

[0110] 1.4.2 Handling Inequality Constraints Inequality constraints such as charge / discharge power constraints and SOC constraints can be uniformly expressed as: (32) Note: The left side contains decision variables for the current and historical periods (via...). (Weighted), the right side is It is a linear function with a constant term. This includes all inequality constraints, such as the SOC upper limit constraint (energy cannot exceed the rated value), power limit constraint, etc. It requires that the constraints hold for all uncertainties, meaning that the constraints cannot be violated even in the "worst-case" scenario.

[0111] Substituting the affine decision rule and rearranging, we get: (33) The right side is about linear functions .because In the uncertain set For any value to be taken from the left side, the equation must always hold true; the left side must be no less than 1. The maximum value (i.e., the worst-case value). That is: (34) in Let be the coefficient matrix in the above equation. Using the Linear Duality Theorem, the maximization problem on the right-hand side can be transformed into the minimization form of its dual problem. The core idea of ​​the duality theorem is that the maximum value of a linear programming problem is equal to the minimum value of its dual problem, and the dual variables correspond to the "shadow prices" of the constraints of the original problem.

[0112] Introducing dual variables (vector, dimension and) Number of constraints (same), and utilize uncertainty sets get: (35) Simultaneously, the dual feasibility condition must be satisfied (i.e., the relationship between the coefficient matrix of the original problem and the dual variables): (36) Expand After determining the specific form, a set of deterministic linear inequality constraints is finally obtained: (37) in, The vector of nonnegative dual variables has a dimension equal to the set of uncertainties. The number of constraints.

[0113] Physical meaning: First rule: The deterministic part (left side) must be no less than the dual variable and The weighted sum. The "size" of the uncertainty set is defined. This reflects the "tightness" of each constraint under the worst-case condition.

[0114] The second equation, the dual feasibility condition, links the coefficient matrix of the uncertain parameters in the primal problem to the dual variables. This equation ensures that the introduced dual variables do indeed correspond to the correct dual form of the primal problem.

[0115] 1.4.3 Final Model Form After the above transformation, the original semi-infinite robust optimization problem is transformed into a deterministic MILP model: Objective function: Minimize the expected operating cost as shown in equation (29); Equality constraint: The deterministic linear equation shown in equation (31) (from the equation constraint transformation); Inequality constraints: The deterministic linear inequality shown in equation (37) (derived from the transformation of inequality constraints, including dual variables) ); Variables: Continuous variables (Pre-dated baseline plan); Continuous variables (Affine adjustment coefficient matrix); continuous variables (Dual variable); binary variable (Charging and discharging and electricity purchase and sale mode).

[0116] This MILP model can be solved efficiently using commercial optimization solvers such as Gurobi, CPLEX, and COPT. Because the model is deterministic and the variable size is controllable, the solution time is typically in the minutes range (for a 24-hour period, 100 EVs example, it is approximately within 10 minutes).

[0117] Step 5: Adaptive Configuration of History Window Length History window length This is a key parameter of the affine decision rule, controlling the scope of historical information referenced in the decision. The selection of this parameter requires a balance between decision quality and computational complexity (see [reference]). Figure 3 (Experimental results).

[0118] like Too small (e.g.) Decision-making relies solely on uncertain information within the current timeframe and fails to utilize historical trends. For example, the inability to predict current output deviations based on past hours' output trends may lead to poor decision-making quality.

[0119] like Too large: Slope matrix Dimensions As the number of model variables increases linearly, the complexity of the solution also rises. Furthermore, historical information from too far back has limited reference value for current decisions, leading to diminishing marginal improvements.

[0120] This application proposes dependency coefficient analysis as... Selected quantitative tool. Define the time period. Decisions on historical periods The dependency coefficient of the uncertain realized value is: (38) Note: This coefficient was calculated. The sum of the absolute values ​​of all elements of a matrix. If Approaching zero indicates the time period Decision-making is almost independent of time period Uncertainty; if If the value is significantly greater than zero, then there is a strong linear dependency.

[0121] See Figure 4 By analyzing the dependency coefficient matrix Visual analysis can identify the actual historical depth of non-zero dependency coefficients. Typically, the dependency coefficient increases with... It increases and then decays to zero. Therefore, it is desirable for all... ( The smallest As a reasonable historical window length, this method avoids [the limitations of] [the limitations of] [the previous methods]. Through trial and error, an adaptive balance between model complexity and decision performance was achieved.

[0122] See Figure 4 The horizontal axis in the graph represents historical periods. The vertical axis represents the current decision-making period. Color block depth representation The size of . It can be observed that: All non-zero color blocks are located in the area at or below the diagonal. This reflects the principle of unexpectedness; The color blocks on the diagonal are usually the brightest, indicating that the uncertainty of the current period has the greatest impact on the current decision-making. along with keep away The color blocks gradually darken, indicating that the dependency coefficient decreases.

[0123] Engineering practice suggestion: If there is no prior information, you can take... (Full history window) Solve the model once, and then based on the calculated... Analyze the decay of the dependency coefficient, and then set a smaller one. Solve again. Usually... This allows you to capture the main historical dependencies (see...) Figure 4 (Experimental results) 1.6 Step Six: MSRO-based Day-ahead Constraint Relaxation and Intraday Rolling Re-optimization Strategy The two-stage process of the current constraint relaxation and intraday rolling re-optimization strategy is as follows: Figure 5 As shown. In the day-ahead phase, the mutual exclusion constraints of the binary variables are relaxed, transforming the MILP problem of the original V2G integrated microgrid dispatch MSRO model into a linear programming (LP) problem, efficiently solving for the affine decision parameters (baseline plan and adjustment coefficients). In the intraday phase, for each time period, based on realized uncertainties, the decision quantity for the current time period is first calculated according to the affine rule, then a constraint-tightened single-time MILP subproblem is constructed (restoring all binary constraints). Rolling re-optimization is performed with the objective of minimizing the actual cost of the current time period, yielding the final executable dispatch command. During the rolling re-optimization process, the SOC trajectory of energy storage and EVs is locked to the global trajectory calculated by the day-ahead affine rule to ensure that global economic efficiency is not compromised by local optimization.

[0124] 1.6.1 Day-ahead Stage -- Constraint Relaxation and Affine Parameter Solving (1) MILP form of the original MSRO model The MSRO model for the V2G integrated microgrid described above is a deterministic MILP model, which includes: Binary variables : (EV charging / discharging status) (Energy storage charging and discharging status) (Electricity purchase and sale status); Continuous variables: Day-ahead baseline plan Affine adjustment coefficient Dual variables ; Constraints: power balance, SOC dynamics, charge and discharge power limits, etc.: deterministic linear equations shown in equation (31); deterministic linear inequalities shown in equation (37); Objective function: Minimize the expected operating cost as shown in equation (29).

[0125] The solution complexity of this MILP model increases exponentially with the number of binary variables. For models containing... EVs The system over a given period has approximately [number] binary variables. ,when When the number of binary variables exceeds 2400, direct solution becomes extremely difficult.

[0126] (2) Constraint relaxation operation One of the core innovations of this patent is the relaxation of the mutual exclusion constraints related to binary variables at the day-ahead stage, transforming the original MILP problem into an LP problem. The same relaxation operation is performed on the mutual exclusion constraints of EV, ESS, and mainnet interactions (Equations (1)-(2), (5)-(6), (10)-(11)): deleting binary variables. , and Replace the original constraints with independent upper and lower bound constraints: (39) (40) (41) (42) (43) (44) Key point: After relaxation, both charging power and discharging power can be non-zero simultaneously (e.g., While this is physically impossible, it provides a larger search space for optimization algorithms.

[0127] (3) Relaxed LP model After constraint relaxation, all binary variables The inequality constraint (Equation (37)) is removed from the model and relaxed to a linear inequality: (45) The original MILP problem is transformed into an LP problem, which can be solved in polynomial time using the simplex method or the interior-point method. The key parameters obtained from the solution include: · : Day-ahead baseline plan (baseline power of each device at each time period); · : Affine adjustment coefficient matrix (reflects the linear response of a decision to uncertainty).

[0129] (4) Theoretical nature Feasibility: Let the feasible region of the relaxed model be... Since any feasible solution in the original MILP necessarily satisfies the relaxed independent upper and lower bound constraints (because the original constraints are a more stringent special case), we have: That is, the relaxed feasible region is a superset of the original feasible region. This means that any feasible solution of the original MILP is also a feasible solution of the relaxed LP; the relaxed LP may contain infeasible solutions in the original MILP (see...). Figure 6 ,For example and (Simultaneous occurrences). Relaxation operations allow devices to "charge and discharge simultaneously" or "buy and sell electricity simultaneously," which, while physically impossible, mathematically provides a larger search space for optimization algorithms, helping to obtain better affine parameters.

[0130] Optimality: see Figure 6 ,set up This represents the optimal objective value for the original MILP problem. This is the optimal objective value for the relaxed LP problem. Because... And since the objective functions are all in minimization form, therefore: That is, the optimal value of the relaxed LP constitutes the lower bound of the optimal value of the original MILP. The difference between the two is defined as the relaxation gap: (46) The relaxation gap reflects the improved optimality resulting from the removal of binary constraints.

[0131] Computational complexity: The solution complexity of the LP problem is approximately... ,in The number of continuous variables is used; while MILP is NP-hard, with a complexity of... , The number of binary variables. When When the value is large, LP is several orders of magnitude more efficient than MILP.

[0132] 1.6.2 Intraday Phase – Affine Decision Calculation and SOC Trajectory Update In actual intraday operation, each time period Initially, the system executes the following process: (1) Observe the uncertainty realization value of the current period Obtain the actual value of photovoltaic power output Actual load value And the actual initial SOC value of newly connected EVs in the current period. Thus, the uncertainty vector is calculated. .

[0133] (2) Calculate the affine decision value Using the affine parameters obtained from the previous period, calculate the affine values ​​of each continuous decision variable for the current period according to equation (25): (47) in include wait.

[0134] (3) Update SOC trajectory Based on the charge / discharge power calculated using affine decision theory, the State of Charge (SOC) of both the energy storage and the EV is updated using the SOC dynamic equation. Energy storage SOC: (48) EV SOC (for each EV in service): (49) Key mechanism: These SOC values ​​will be "locked" as global constraints in subsequent rolling re-optimization, ensuring that intraday local adjustments do not deviate from the globally optimal energy trajectory planned in the previous day.

[0135] 1.6.3 Intraday Phase -- SOC Trajectory Locking and Scrolling Re-optimization (Constraint Tightening) Relying solely on affine decision-making presents two problems: first, due to day-ahead relaxation, affine decisions may violate binary mutual exclusion constraints (such as simultaneous charging and discharging); second, the linear assumption of the affine rule limits economic efficiency. To address these issues, this patent proposes a rolling re-optimization mechanism.

[0136] For each time interval t, based on the following known quantities, construct a single-time interval MILP subproblem: Realized uncertainty in the current period ; SOC value at the end of the previous period and (Known); The target SOC value for the current period calculated by affine decision-making recently. and (As a lock value).

[0137] Mathematical model of a single-time MILP subproblem: Objective function: Minimize the operating cost for the current time period. (50) Constraints (Restore all original binary constraints): EV charging and discharging mutual exclusion constraints (Equations (1)-(2)):

[0138] ESS charge / discharge mutual exclusion constraints (Equations (5)-(6)):

[0139] Main grid electricity purchase and sale mutual exclusion constraints (Equations (10)-(11)):

[0140] SOC global trajectory tracking constraints (Equations (48)-(49)):

[0141]

[0142] in and This is the target value calculated and locked by the affine decision in step two.

[0143] Power balance constraints (Equation (12)) and other equipment constraints.

[0144] Solution and Execution: This single-period MILP is very small (the number of binary variables is approximately...). (typically less than 50), it can be solved in milliseconds. The final executable instructions are obtained after solving. The command is then distributed to each device in the microgrid for execution.

[0145] Theoretical nature: Feasibility: Since the SOC trajectory calculated by affine decision is feasible in day-ahead relaxed LP, and the single-time MILP subproblem preserves all original binary constraints, it satisfies physical feasibility; by adjusting the power allocation (keeping the net charge and discharge power constant), a feasible solution can always be constructed (see...). Figure 7 Therefore, as long as the current relaxation LP is feasible, the rolling re-optimization subproblem must have a feasible solution.

[0146] Optimality: By locking onto the SOC trajectory given by the day-ahead affine decision, rolling re-optimization will not deviate from the globally optimal energy trajectory during local adjustments, thus ensuring that the overall economy is close to the day-ahead global optimum. Rolling re-optimization further ensures local optima within a single time period.

[0147] Computational efficiency: The solution time of LP is usually in the second range (compared to several minutes of MILP). The scale of MILP is greatly reduced in each time period of the day, and its solution time is in the millisecond range, which meets the real-time scheduling requirements.

[0148] 1.6.4 Overall Process of the Collaboration Mechanism The day-ahead and intraday collaborative scheduling mechanism of this patent consists of two stages (see Algorithm 1): In the day-ahead stage, the LP model after constraint relaxation is solved based on predicted data to efficiently obtain affine decision parameters (baseline plan and adjustment coefficients); in the intraday stage, the following operations are performed cyclically for each time period: observing the realization of current uncertainties, calculating affine decision values ​​based on affine parameters and updating the locked SOC trajectory, and then solving the single-time MILP subproblem to restore all binary mutual exclusion constraints, resulting in physically feasible and more economically efficient executable scheduling instructions. This collaborative mechanism of "day-ahead relaxation for strategy and intraday contraction for feasibility" significantly improves computational efficiency and ensures the physical feasibility of scheduling instructions while guaranteeing global optimality.

[0149] Algorithm 1 demonstrates the complete execution flow of the method in this patent.

[0150] The beneficial effects of the technical solution in this application are as follows: This patent specifically models and verifies V2G-integrated community microgrid scenarios. A case study of an industrial park microgrid containing 100 electric vehicles was set up to verify the robust scheduling performance of the patented method under multi-source uncertainties such as EV access to the SOC.

[0151] 2.1 Verification of Multi-Stage Robust Optimization and Scenario-Driven Objectives Under various uncertainty levels (ρ=0.1~0.3) in all tests, experimental results validated the advantages of multi-stage robust optimization and scenario-driven objectives in this patent in terms of operational cost optimization and system robustness. Specifically, this patent has the following beneficial effects compared to existing technologies: 2.1.1 Optimization and improvement in operating costs (1) Compared with traditional worst-case robust optimization, operating costs are reduced by 54% to 62%; Table 1. Cost comparison between test scenario-driven and worst-case optimization objectives (Δt=10)

[0152] analyze: 1. Traditional robust optimization adopts the goal of "minimizing the cost of the worst-case scenario," resulting in overly conservative scheduling schemes and extremely high long-term operating costs. 2. This patent adopts a scenario-driven expected cost target and incorporates the probability distribution information of uncertainty into the optimization process through Monte Carlo sampling, so that the decision result is closer to the average situation of actual operation.

[0153] Under three different uncertainty intensities (ρ=0.1, 0.2, 0.3), the method of this patent achieves a cost reduction of 54% to 62%, demonstrating its significant economic advantage.

[0154] (2) Compared with static robust optimization (SRO) and two-stage robust optimization (2SRO), the cost is close to the theoretical lower bound and follows the principle of unpredictability; Table 2 Performance comparison of MSRO and 2SRO in test scenario samples

[0155] analyze: While 2SRO can mathematically achieve a slightly lower cost (which can serve as an idealized lower bound), its model assumes that the second-stage decision can rely on all uncertain information about the future, which violates the principle of unpredictability in actual scheduling. Therefore, its results are not feasible in practice.

[0156] Static robust optimization (SRO) has an operating cost that is about 14% higher than MSRO under an uncertainty ρ of 0.02, and therefore it is too conservative.

[0157] The Multi-Stage Robust Optimization (MSRO) proposed in this application strictly follows the principle of unpredictability, while the cost is only 1 to 3 yuan higher than that of 2SRO (relative deviation <0.3%), which almost achieves the theoretically optimal economic performance.

[0158] This demonstrates that the application achieved a perfect balance between compliance (meeting unexpected needs) and economy without sacrificing economic efficiency while ensuring compliance with decision-making requirements.

[0159] 2.1.2 Improved robustness in handling uncertainty (1) It remains feasible under high uncertainty, while traditional methods fail; Table 3. Feasibility comparison of MSRO with other methods.

[0160] analyze: The deterministic model (DM) completely ignores uncertainty and barely obtains an "unsafe" solution, but it cannot safely cope with the fluctuations of source load and EVSOC.

[0161] Static Robust Optimization (SRO) employs a "here and now" single-stage decision-making approach, which becomes infeasible at ρ=0.1, indicating that its rigid strategy cannot adapt to large uncertainty sets.

[0162] The MSRO proposed in this patent achieves a feasible solution under all test uncertainties (ρ = 0.02~0.3). This is due to: 1. A multi-stage information-driven decision-making structure allows for adjustments to decisions based on the latest observations at each time period; 2. Affine decision rules provide linear adaptive capabilities to uncertainty; 3. Robust constraint transformation ensures that constraints are satisfied under all possible uncertainties.

[0163] Compared to 2SRO, which can also handle high uncertainty, this application better reflects actual information flow. While 2SRO can be mathematically feasible, it violates the principle of unpredictability and essentially relies on the assumption of "knowing the future," making it unworkable in practical scheduling. In contrast, the MSRO proposed in this patent provides a truly feasible and robust scheduling scheme with almost identical economic efficiency under the same level of uncertainty (ρ=0.3, cost 1289 vs 1288). This robustness is reflected not only in mathematical feasibility but also in engineering feasibility.

[0164] 2.1.3 Overall Conclusion Based on the experimental data in Table 1-3, the two core beneficial effects of this patent can be clearly verified: (1) Optimization and improvement in operating costs Compared to traditional worst-case robust optimization, operating costs are reduced by 54% to 62%; Compared to traditional static robust optimization, operating costs can be reduced by 14%; Compared to two-stage robust optimization (2SRO), the cost increases by less than 0.3%, while strictly adhering to the principle of unpredictability.

[0165] (2) Improved robustness in handling uncertainty When the uncertainty intensity ρ = 0.1~0.3, both the deterministic model (DM) and static robust optimization (SRO) become infeasible, while the method of this patent always obtains a feasible solution; Compared to 2SRO, the robustness of this patent is truly executable robustness, satisfying the "unexpectedness principle" and not relying on assumptions about future information.

[0166] 2.2 Validation of Constraint Relaxation and Intraday Rolling Re-optimization The commercial optimization solver Gurobi has a maximum solution time of 1 hour. Under high uncertainty (ρ=0.3), experimental results verify the advantages of day-ahead constraint relaxation and intraday rolling re-optimization in terms of operational cost optimization, solution efficiency, and robustness. Specifically, this patent has the following beneficial effects compared to existing technologies (see...). Figure 8 ): (1) Economic optimization This patent significantly improves the economy of the scheduling scheme through a collaborative mechanism of "day-ahead constraint relaxation + intraday rolling re-optimization". Compared with pure affine decision-making and "here-and-now" fixed binary decision-making, it achieves better results with different historical window lengths. This patent significantly reduces operating costs. For example, compared to the fixed binary decision of "here-and-now", the cost reduction of the constraint relaxation strategy is about 6%; compared to pure affine decision, rolling re-optimization further reduces the cost by 12%. Therefore, the complete solution of this patent (relaxation + re-optimization, point A) reduces the cost from about 1501 yuan to 1240 yuan compared to the non-relaxation + affine solution (point C), with an overall economic improvement of about 18%. Since the global SOC trajectory given by the day-ahead affine decision is locked, the rolling re-optimization does not deviate from the global optimal energy transfer strategy during local adjustments, thus ensuring both intraday economy and maintaining the day-ahead global optimum.

[0167] (2) Computational efficiency This patent transforms the NP-hard MILP problem into a polynomial-time solvable LP problem through day-ahead constraint relaxation, significantly improving solution efficiency. The solution time is reduced from hours to seconds: MSRO without constraint relaxation... The solution efficiency is on the order of hours; A feasible solution cannot be obtained within the 1-hour time limit. Figure 8 (Missing points); however, after adopting the constraint relaxation of this patent, the solution time is stabilized at 2~15 seconds, and the computational efficiency is improved by more than 3 orders of magnitude. Millisecond-level response through intraday rolling re-optimization: The size of the MILP sub-problem model and the number of binary variables in a single time period are significantly reduced, and the solution time is in the millisecond range. The total intraday computation time for 24 time periods is less than the second range. The overall scheduling cycle (day-ahead + intraday) is in the second range, improving the solution efficiency by 3 orders of magnitude, fully meeting the real-time scheduling requirements.

[0168] (3) Robustness This patent remains feasible under conditions of high uncertainty and ensures the physical realizability of scheduling instructions. It remains feasible under high uncertainty because it does not relax the MSRO. The model has no solution and cannot be processed. The uncertainty of (high prediction error); while the relaxation strategy of this patent is in all Feasible solutions are obtained under all test conditions, and the scheduling scheme can still be output stably. The physical infeasibility of affine decisions is eliminated: pure affine decisions may produce instructions that violate physical mutual exclusion constraints, such as simultaneous charging and discharging, or simultaneous buying and selling of electricity. This patent's rolling re-optimization restores all original binary constraints and "projects" the affine decision onto the original feasible region through local optimization, ensuring that the final instruction is physically executable.

[0169] In summary, the collaborative strategy of "day-ahead constraint relaxation + intraday rolling re-optimization" proposed in this patent is significantly superior to existing technologies in the three core dimensions of economic optimization, computational efficiency, and robustness, providing an efficient and feasible solution for real-time robust scheduling of V2G integrated microgrids.

[0170] The key advantages of this application's technical solution compared to existing technologies (especially two-stage robust optimization 2SRO and traditional worst-case robust optimization) are: The key advantage of this patent over existing technologies (especially two-stage robust optimization 2SRO and traditional worst-case robust optimization) is: While strictly adhering to the principle of unpredictability (i.e., decisions rely solely on historically revealed information), it simultaneously achieves significant improvements in operational economy (reducing operating costs by 54%–62% compared to traditional worst-case robust optimization) and strong robustness (achieving feasible solutions under all levels of test uncertainty). Furthermore, the operating cost is almost on par with the theoretical lower bound of 2SRO (deviation <0.3%), which violates the principle of unpredictability. In addition, through a two-stage collaborative mechanism of "day-ahead binary constraint relaxation + intraday SOC trajectory locking rolling re-optimization," it simultaneously solves three major challenges in existing V2G integrated microgrid multi-stage robust scheduling (MSRO): low computational efficiency (NP-hard), physically infeasible scheduling instructions (violation of mutual exclusion constraints), and economic losses. While ensuring global optimality, it improves day-ahead solution efficiency by more than three orders of magnitude (from hours to seconds), ensures fully physical executable intraday instructions, and achieves an overall economic improvement of approximately 18%.

[0171] The essence of this advantage is that, for the first time in V2G integrated microgrid dispatching, it organically integrates multi-stage decision-making structure, affine decision-making rules, scenario-driven expected cost targets, day-ahead constraint relaxation, and intraday rolling re-optimization, breaking the rigid trade-off between "conservatism and feasibility" in traditional robust optimization, and achieving a triple unification of "compliance (unexpectedness) + economy (low average cost) + robustness (feasibility in all scenarios)".

[0172] Figure 5 This diagram illustrates the overall framework of the two-stage collaborative scheduling from day-ahead to day-intraday, showcasing the processes of the day-ahead stage (constraint relaxation LP solution) and the day-intraday stage (uncertainty observation → affine decision → rolling re-optimization → instruction execution). Key data flows are marked: forecast data, affine parameters, uncertainty realization, locked SOC, and executable instructions.

[0173] Figure 6 The diagram illustrates the expansion of the feasible region by constraint relaxation, comparing the original feasible region (line segment) with the relaxed feasible region (rectangle), demonstrating the feasibility and optimality of the relaxation strategy.

[0174] Figure 7 This diagram illustrates the constant net charge / discharge power during rolling re-optimization, showing the mapping from the relaxed solution (points A and B) to the feasible solution (point D), demonstrating that with constant net charge / discharge power, the SOC trajectory is maintained.

[0175] Figure 8 The graph compares computational efficiency and cost. The horizontal axis represents the historical window length Δt, and the vertical axis represents cost and solution time, showing a comparison between relaxation vs. non-relaxation and affine vs. re-optimization.

[0176] The technical protection points of the technical solution in this application are: The following three core technical features work together to directly generate the above-mentioned greatest advantages. Among them, (1) and (2) are the key to improving economic efficiency, (1) and (3) are the foundation of compliance and robustness, and (4) is the guarantee of model practicality and adaptability. In addition, (5) is the key to leaping computational efficiency and improving optimization, (6) is the direct source of physical feasibility guarantee and global economic maintenance, and (7) is a complete technical solution formed by organically integrating the first two.

[0177] (1) Multi-stage decision-making structure for V2G integrated microgrid dispatch (1.2, 1.3.1 ~ 1.3.2). Formulas (17)~(21): The uncertainty set comprehensively considers the multi-source uncertainty in the microgrid and the uncertainty of EV behavior in V2G, and is a prerequisite for robust scheduling of V2G integrated microgrids; Formulas (23) to (25) enforce the principle of unpredictability, ensuring that the affine rule-based decision-making for each time period relies solely on the uncertainty information revealed up to that time period, perfectly aligning with the actual scheduling information flow. Uncertainty is addressed promptly in each time period, preventing the long-term accumulation of uncertainty from increasing the conservatism of the decision-making process. This is the direct source of the compliance and optimization advantages of this patent.

[0178] (2) Scenario-driven expected cost objective function (1.3.4) Formulas (28) and (29) abandon the overly conservative objective of "minimizing the worst-case scenario cost" in traditional robust optimization and adopt the expected operating cost under Monte Carlo sampling as the optimization objective. This is the direct reason for the 54%~62% reduction in operating cost in this patent, and it is also the core innovation that distinguishes it from all existing robust optimization methods.

[0179] (3) Deterministic transformation of robust constraints (1.4.1 ~ 1.4.3) Formulas (31) and (37) transform the semi-infinite robust constraints into a finite number of deterministic linear / mixed integer constraints, ensuring that the physical constraints (power balance, SOC limit, etc.) are satisfied for realizations within all uncertainty sets. This is the technical guarantee that this patent can obtain feasible solutions under all uncertainty levels.

[0180] (4) Adaptive configuration of historical window length (1.5) Formula (38) and Figure 4 : By calculating the dependence coefficient of decision variables on the realized value of uncertainties over historical periods Visualize and identify the actual historical depth with non-zero dependency coefficients, thereby quantifying the selection of a reasonable historical window length. This step avoids blind trial and error, controls the scale of model variables and solution complexity while ensuring decision quality, and enhances the adaptability and engineering practicality of this patented technical solution in different calculation scenarios.

[0181] (5) Binary mutual exclusion constraint relaxation operation in the previous stage (1.6.1, Equations (39)~(45)) In day-ahead scheduling, all mutual exclusion constraints related to binary variables (including mutual exclusion for EV charging and discharging (1)-(2), mutual exclusion for energy storage charging and discharging (5)-(6), and mutual exclusion for main grid power purchase and sale (10)-(11)) are deleted, and the original constraints are replaced with independent upper and lower bound constraints (Equations (39)-(44)), so that charging power and discharging power can be non-zero at the same time. Equation (45) constructs a tractable form of MSRO inequality constraints based on relaxation strategy. This operation completely transforms the original NP-hard MILP model into a polynomial-time solvable LP model, which is the direct reason for reducing the day-ahead solution time from hours (or even non-convergence) to seconds (2~15 seconds), and improving computational efficiency by more than 3 orders of magnitude. At the same time, the relaxed feasible region is a superset of the original feasible region, ensuring that the optimal value of the relaxed LP constitutes the lower bound of the optimal value of the original MILP, providing a mathematical basis for intraday recovery feasibility.

[0182] (6) Intraday rolling re-optimization and SOC trajectory locking mechanism (1.6.2 ~ 1.6.3, Equations (48) ~ (50) and complete recovery of the original binary constraints) The target value of energy storage SOC calculated and locked based on day-ahead affine decision for each time period. and EVSOC target value (Equations (48)-(49)) construct and solve a single-time MILP subproblem that restores all original binary mutual exclusion constraints (Equations (1)-(2), (5)-(6), (10)-(11), (50) and SOC tracking constraints). This mechanism has a dual effect: ① It projects the physically infeasible solutions (such as simultaneous charging and discharging) that may be generated by affine decision-making onto the original feasible region through local optimization, ensuring that the final scheduling instructions fully satisfy the physical mutual exclusion constraints and eliminating the absurd instructions under the existing affine rules; ② By forcibly tracking the global SOC trajectory locked a day-ahead, it ensures that local adjustments do not deviate from the global optimal energy strategy a day-ahead, which is a technical guarantee for improving comprehensive economic efficiency and maintaining feasible solutions under high uncertainty. The number of binary variables in this single-time MILP subproblem is extremely small, and the solution time is in the millisecond range, which meets the requirements of real-time scheduling.

[0183] (7) Two-stage collaborative mechanism of “constraint relaxation + SOC trajectory locking and rolling re-optimization” (1.6.3, Algorithm 1 and overall process) The day-ahead constraint relaxation in item (5) and the intraday SOC trajectory locking and rolling re-optimization in item (6) are organically integrated into a complete scheduling framework: in the day-ahead stage, the global affine strategy (baseline plan and adjustment coefficients) and global SOC trajectory are efficiently obtained by relaxing binary constraints; in the intraday stage, the SOC trajectory is used as a rigid constraint, and the relaxed solution is "projected" back to the original feasible region through rolling re-optimization. This two-stage collaborative mechanism realizes the synergistic effect of "day-ahead relaxation to find the strategy and intraday contraction to find the feasibility", so that the three goals of computational efficiency, physical feasibility and global economy, which were originally mutually restrictive, can be achieved simultaneously. This is the complete technical solution that distinguishes this patent from all existing MSRO methods, and it is also the overall guarantee for realizing the greatest advantage.

[0184] This patent aims to address three major technical problems in V2G integrated microgrid scheduling: ① violation of the unexpectedness principle, ② lack of a V2G multi-stage robust framework, ③ excessive conservatism in traditional worst-case robust optimization, ④ conservative traditional affine decision rules that are unsuitable for binary variables, and ⑤ a large number of binary variables that severely limit the efficiency of MSRO solution. To achieve the technical objective of "improving both economy and robustness while strictly adhering to the unexpectedness principle," in addition to the "multi-stage affine decision rules + scenario-driven expected cost + adaptive window for dependency coefficients + day-ahead constraint relaxation + intraday SOC trajectory locking and rolling re-optimization" technical solution proposed in this patent, those skilled in the art may also adopt the following alternative solutions based on existing technologies. It should be noted that these alternative solutions all have varying degrees of defects and cannot simultaneously achieve the beneficial effects described in this patent.

[0185] Alternative Solution 1: Multi-stage stochastic programming based on stochastic dual dynamic programming (SDDP) Solution Description: The photovoltaic output, load, and EV connection to the State of the Oxide (SOC) are modeled as a multi-stage Markov process (such as an autoregressive model or discrete state transition matrix). Stochastic dual dynamic programming (SDDP) is used to solve the multi-stage stochastic programming model. Decision variables are approximated using fully nonlinear value functions, without restriction to affine form. The objective function is the expected total cost.

[0186] The difference between Alternative Solution 1 and the technical solution of this application: Uncertainty modeling: SDDP usually requires uncertainties to have finite scenario trees or Markov properties, while this patent uses box-type uncertainty sets (without requiring probability distribution assumptions). Decision rules: SDDP does not predefine the affine form, but calculates the piecewise linear value function through backward recursion; this patent adopts explicit affine decision rules, which are easy to combine with robust constraints; Robustness: SDDP only guarantees feasibility in the expected sense, and does not guarantee that the constraints will be met for all uncertain implementations (i.e., solutions that may violate the constraints may be generated); this patent guarantees feasibility in all scenarios through robust transformation.

[0187] Technical drawbacks of Alternative Solution 1: The size of the scene tree grows exponentially with the number of stages, resulting in a huge computational burden (for a 24-hour period and 100 EVs example, SDDP may not be able to solve it within an acceptable time). It only provides expected feasibility, not worst-case guarantees. In V2G scheduling, SOC overruns or power imbalances may occur. Binary variables (charge / discharge mode selection) are difficult to handle and usually require relaxation or heuristic processing.

[0188] Alternative Solution 2: Robust Stochastic Hybrid Optimization Based on Conditional Value at Risk (CVaR) Solution Description: The multi-stage decision-making structure is retained, but the objective function is changed to "expected cost + λ·CVaR_α(cost)", where CVaR_α is the conditional risk value, α∈(0,1). The constraints adopt chance constraints (satisfied with a certain probability) instead of the robust constraints of this patent (satisfied for all uncertainties). During the solution process, the chance constraints are transformed into mixed-integer linear programming through scenario sampling.

[0189] The differences between Alternative Solution 2 and the technical solution of this application are as follows: The technical solution in this application adopts strict robust constraints (feasible under all uncertainties), while the alternative solution adopts probabilistic constraints (allowing a small probability of violation); Objective function: This patent uses pure expected cost; alternative solutions introduce a risk penalty term.

[0190] Technical drawbacks of alternative solution two: Opportunity constraints require a pre-set threshold for the probability of violation. The selection of this threshold is subjective and cannot guarantee safety in extreme scenarios (such as insufficient SOC when an EV leaves the network). When the uncertainty set is large, in order to meet the high confidence chance constraint, its conservatism may be close to or even exceed the robust scheme of this patent, while the computational complexity increases significantly. The combination of binary variables and chance constraints often leads to non-convex models, requiring complex Big M methods or combined approximations.

[0191] Alternative Solution 3: Rolling Robust Optimization Based on Model Predictive Control (MPC) Solution Description: In each scheduling period, a robust optimization problem with a finite time domain (e.g., the next 4 hours) is solved. Only the decision for the first period is executed, and then the solution is rolled back upon the arrival of new information. The uncertainty set adopts a box or ellipsoidal form, and the objective function can be the worst case or the nominal scenario cost. This is a rolling implementation of two-stage robust optimization.

[0192] The differences between Alternative Solution 3 and the technical solution of this application are as follows: The technical solution of this application solves the multi-stage affine strategy in the entire time domain in one go, and the decision rule explicitly gives the mapping from uncertainty to decision; MPC requires online iterative solution of the optimization problem, and the computational load is distributed in the running process; The technical solution of this application strictly satisfies the principle of unpredictability (forced through affine rules), while MPC, although unpredictability is satisfied in actual execution, may still rely on the assumption of future uncertainty in solving each subproblem (depending on the modeling method of the subproblem).

[0193] Technical drawbacks of alternative solution three: Rolling solutions require real-time computing capabilities and have high requirements for communication and computing resources; this patent can obtain the closed-loop strategy in one offline solution, and only the uncertain realization value needs to be substituted within a day. Finite-domain truncation of MPC may introduce boundary effects, leading to suboptimality; this patent considers full-time-domain optimization. If the MPC subproblem adopts the worst-case objective, the conservatism problem still exists; if the expected objective is adopted, the feasibility of the constraints cannot be guaranteed.

[0194] Alternative Solution 4: Multi-stage robust optimization based on piecewise affine decision rules Solution Description: The affine decision rule is extended to a piecewise affine form, employing different linear mappings in different uncertainty regions. For example, the affine coefficients are switched based on whether the photovoltaic output falls below a certain threshold. This can better approximate the nonlinear optimal strategy.

[0195] The differences between Alternative Solution 4 and the technical solution of this application are as follows: The technical solution of this application adopts a single affine rule (global linear); the piecewise affine rule introduces additional binary variables to indicate the region division, and the model becomes a mixed integer second-order cone or mixed integer linear programming (larger scale).

[0196] Defects of Alternative Option Four: The number of segments grows exponentially ("curse of dimensionality"), making it almost unsolvable for V2G microgrids (which have high uncertainty dimensions); The partition boundaries need to be defined in advance, and the selection of the boundaries is itself a difficult optimization problem; The experiments in this patent show that the simple affine rule can achieve performance close to the theoretical lower bound (deviation from 2SRO <0.3%). The marginal improvement brought by piecewise affine is minimal, but the computational complexity increases dramatically, making it impractical for engineering applications.

[0197] Alternative Option 5: End-to-End Scheduling Strategy Based on Deep Reinforcement Learning (DRL) Scheme Description: The V2G microgrid scheduling is modeled as a Markov Decision Process (MDP). A neural network policy is trained using DRL algorithms such as Deep Q-Network (DQN) and Proximal Policy Optimization (PPO). The input is the current observation state (SOC, prediction error, etc.), and the output is actions such as charging and discharging power. The objective function is the cumulative expected cost (discounted or averaged).

[0198] The differences between Alternative Solution 5 and the technical solution of this application are as follows: DRL does not explicitly model the uncertainty set, but learns the implicit distribution through a large number of simulations; this patent explicitly uses a box-type uncertainty set to provide robustness guarantees. DRL strategies typically lack feasibility guarantees (neural networks may output actions that violate constraints); this patent ensures that constraints are always satisfied through robust transformation.

[0199] Technical drawbacks of alternative option five: Training requires a large amount of data and time, and convergence lacks theoretical guarantees; The inability to guarantee safety constraints (such as SOC not exceeding limits) may cause equipment damage or violate power grid regulations in a real system. The black-box strategy has poor interpretability and is difficult to pass the power grid audit and acceptance. For situations outside the uncertainty distribution (such as extreme weather), the DRL strategy may fail completely.

[0200] Alternative Solution Six: An Alternative Based on Adaptive Piecewise Affine Decision Making Solution Description: This patent employs a linear affine decision rule (Equation (23)) to represent continuous decision variables as linear functions of historical uncertainty parameters. An alternative approach is to use a piecewise affine decision rule, which divides the uncertainty parameter space into multiple regions and uses different linear affine functions in each region.

[0201] Specific implementation: The uncertainty space of photovoltaic output, load and EV connection to SOC is divided into several intervals (such as low, medium and high intervals). Solve for a set of affine parameters in each interval. ); During intraday operation, the decision value is calculated by selecting the corresponding affine function based on the interval into which the current uncertainty realization value falls.

[0202] Differences from the main solution: The current stage requires solving multiple LP problems (one for each interval), which slightly increases the computational load. It has stronger non-linear expression capabilities, which can further reduce the relaxation gap and improve economy; The two-stage framework of "slack off strategy in the day and re-optimize in the day" is still retained.

[0203] Applicable scenarios: Suitable for real-world scenarios with large ranges of uncertainty and distinct segmentation characteristics (such as microgrids with significant seasonal differences).

[0204] Alternative Solution 7: Day-ahead Integer Constraint Handling Scheme Based on Lagrange Relaxation Solution Description: This patent employs a "hard relaxation" method that directly deletes binary variables. An alternative approach is to use Lagrangian relaxation, which adds binary mutual exclusion constraints as penalty terms to the objective function instead of deleting them directly.

[0205] Specific implementation: Preserve binary variables However, the mutual exclusion constraints (equations (1)-(2), (5)-(6), (10)-(11)) are removed from the constraint set; The degree of violation of these constraints is added to the objective function as a penalty term:

[0206] By adjusting the penalty coefficient This allows optimization to achieve a balance between cost and constraint satisfaction; The intraday phase will still employ the SOC trajectory locking and rolling re-optimization mechanism.

[0207] Differences from the main solution: The current model still contains binary variables, but it is made easier to solve through penalty relaxation (it can be transformed into LP or quadratic programming). The selection of the penalty coefficient requires experience or adaptive adjustment, which increases the work of parameter tuning. An intermediate solution between "fully relaxed" and "fully strict" can be provided.

[0208] Applicable scenarios: Suitable for applications that require high accuracy in day-to-day calculations and can accept a certain amount of parameter tuning work.

[0209] Alternative Option 8: Intraday Full Re-optimization Scheme Based on Model Predictive Control Solution Description: This patent employs a lightweight mechanism of "affine decision pre-computation + single-period MILP rolling re-optimization" during the intraday phase. An alternative approach could be to use a complete Model Predictive Control (MPC) framework, solving a multi-period (e.g., the next 4-6 hours) MILP sub-problem in each intraday period.

[0210] Specific implementation: In the current phase, only the global SOC trajectory is output, and affine parameters are not output. Each time period during the day Using the current SOC as the initial state and the current day's SOC trajectory as the final constraint, solve for a future... Time period ( or MILP subproblems; Only the instructions for the first time period are executed, and the next time period is updated on a rolling basis.

[0211] Differences from the main solution: The computational load increased significantly during the day (multi-period MILP), but it is still much smaller than the global MILP of the previous day; Longer-term economic considerations could be taken into account to further reduce operating costs; The real-time problem of solving MILP in multiple time periods needs to be solved, which requires higher computing power.

[0212] Applicable scenarios: Suitable for scenarios with sufficient computing resources and higher requirements for intraday cost-effectiveness optimization.

[0213] Alternative Solution Nine: Intraday Order Correction Scheme Based on Reinforcement Learning Solution Description: The rolling re-optimization in this patent relies on solving a single-time MILP subproblem. An alternative approach could be to employ a reinforcement learning (RL) model, which uses a pre-trained RL agent to correct physically infeasible instructions generated by affine decisions in real time.

[0214] Specific implementation: Train an RL agent offline, with affine decision values ​​as input. Given the current SOC status, the output is the corrected feasible instructions. ; During training, the reward function is designed to: penalize violations of mutual exclusion constraints (such as simultaneous charging and discharging), and reward low cost and high SOC tracking accuracy; During daytime operation, the RL agent outputs feasible instructions at millisecond speeds, without requiring MILP decomposition.

[0215] Differences from the main solution: Faster computation speed within the day (forward inference vs. solving MILP); It requires offline training data, and the interpretability and feasibility guarantees of RL policies are not as strict as those of MILP. Reliability can be improved through a hybrid mode of "RL pre-selection + MILP verification".

[0216] Applicable scenarios: Suitable for scenarios with extremely high requirements for intraday response speed (below millisecond) and where a certain probability of constraint violation risk is acceptable.

[0217] The implementation principle of scenario-driven multi-stage robust re-optimization scheduling for V2G integrated community microgrids in this application is as follows: Taking the V2G integrated microgrid as the research object, and considering the uncertainties of multiple sources such as photovoltaic output, conventional load, and electric vehicle access status, a multi-stage robust optimization framework is constructed. A scenario-driven expected cost objective function and affine decision rules are introduced to achieve a unified optimization of the scheduling strategy's robustness and economy. The scheduling time domain is divided into multiple continuous stages, ensuring that the decisions at each stage rely only on observed uncertainty information, strictly satisfying the principle of unpredictability. A affine decision rules are used to establish a linear mapping relationship between continuous decision variables and historical uncertainties, enhancing the system's adaptability to disturbances while ensuring model computability. Furthermore, multiple random scenarios are generated within the box uncertainty set using the Monte Carlo method, used only to construct the expected operating cost objective function, avoiding the over-conservatism brought about by traditional worst-case robust optimization. Duality theory is used to perform an equivalent transformation on the constraints containing uncertainties, transforming the original semi-infinite constraint problem into a deterministic mixed-integer linear programming model, achieving feasibility assurance within the entire uncertainty range.

[0218] The above are all preferred embodiments of this application, and are not intended to limit the scope of protection of this application. Therefore, all equivalent changes made in accordance with the structure, shape and principle of this application should be covered within the scope of protection of this application.

Claims

1. A scenario-driven multi-stage robust re-optimization scheduling method for V2G integrated community microgrids, characterized in that, Includes the following steps: S1. Construct a deterministic scheduling model for V2G integrated community microgrids; Obtain the operating parameters of photovoltaic power generation units, controllable distributed generation units, energy storage systems, electric vehicle aggregation units, and main grid interaction units in the community microgrid, and establish a deterministic scheduling model that includes equipment power constraints, state of charge constraints, power balance constraints, and operating cost models; The electric vehicle aggregation unit includes multiple electric vehicles operating in a vehicle-to-grid (V2G) mode. These electric vehicles participate in microgrid power regulation as rechargeable and discharging mobile energy storage resources based on their access status. S2. Construct a multi-stage uncertainty set that includes multiple sources of uncertainty factors; The prediction errors of photovoltaic power generation, load forecasting, and initial state of charge of electric vehicles are obtained, and corresponding uncertain variables are established respectively. A multi-stage uncertainty set is constructed based on the historical statistical data of each uncertain variable. The multi-stage uncertainty set is used to describe the range of deviation between the actual operating state and the predicted state in different scheduling stages; S3. Establish a multi-stage robust optimization scheduling model based on the principle of unexpectedness; The scheduling cycle is divided into multiple consecutive decision-making stages, so that the scheduling decision in any stage depends only on the uncertainty information already obtained in the current stage and the historical stages, and not on the uncertainty information not yet revealed in the future stages. S4. Construct an optimization objective based on scenario-driven expected cost; Within the uncertainty set, a random scenario sampling method is used to generate multiple operating scenarios, and the expected value of the operating cost under multiple scenarios is used as the optimization objective. The optimization objective does not adopt the worst-case scenario cost minimization method, but a scenario-driven expected operating cost optimization objective in order to reduce the over-conservatism of traditional robust optimization. S5. Perform deterministic transformation on the multi-stage robust optimization model; For power balance constraints, equipment operation constraints, and state of charge constraints that contain uncertain variables, the infinite-dimensional robust constraints are transformed into finite-dimensional deterministic linear constraints by using the linear dual transformation method in robust optimization. By introducing dual variables, the transformed model becomes a solvable mixed-integer linear programming model. S6. Perform day-ahead constraint relaxation and intraday rolling re-optimization scheduling; In the current stage, the binary variables in the mutual exclusion constraints of electric vehicle charging and discharging, energy storage charging and discharging, and main grid power purchase and sale are relaxed, the mixed integer programming model is transformed into a linear programming model, and the affine decision parameters are obtained by solving the model. During the intraday phase, affine scheduling results are calculated based on the real-time acquired uncertainty realization values, and a single-period rolling optimization model is established based on the recovered physical mutual exclusion constraints to re-optimize the scheduling instructions for the current phase. Meanwhile, the energy storage state-of-charge trajectory and the electric vehicle state-of-charge trajectory calculated in the day-ahead phase are locked as global constraints in order to obtain a final dispatch instruction that meets physical constraints and is economical.

2. The scenario-driven multi-stage robust re-optimization scheduling method for V2G integrated community microgrids according to claim 1, characterized in that: The deterministic scheduling model includes: a photovoltaic power generation model, a load power model, and an electric vehicle initial state of charge model.

3. The scenario-driven multi-stage robust re-optimization scheduling method for V2G integrated community microgrids according to claim 1, characterized in that: Both the energy storage system and the electric vehicle aggregation unit are equipped with charge and discharge mutual exclusion control variables. Wherein: when the charging state variable is in the first state, only charging power is allowed to exist; When the charging state variable is in the second state, only the discharge power is allowed to exist; This is to avoid energy storage systems or electric vehicles performing charging and discharging operations simultaneously.

4. The scenario-driven multi-stage robust re-optimization scheduling method for V2G integrated community microgrids according to claim 1, characterized in that: The multi-stage uncertainty set is constructed using a box uncertainty set. The length of the historical information window in the affine decision rule is adaptively determined based on the decision dependency coefficient; Specifically, the dependency coefficients of the current stage decision variables on the uncertain variables of the historical stage are calculated: The effective historical influence range is determined based on the dependency coefficient, and the historical window length is determined accordingly.

5. The scenario-driven multi-stage robust re-optimization scheduling method for V2G integrated community microgrids according to claim 1, characterized in that: The scene sampling adopts the Monte Carlo sampling method to generate multiple uncertainty realization samples within the uncertainty set; The multiple samples are used only to calculate the expected operating cost in the objective function, without increasing the number of robust constraints; The aforementioned day-ahead constraint relaxation includes: Delete the charging / discharging state variables of electric vehicles; Delete the energy storage charge / discharge state variables; Delete the main grid power purchase and sale status variable; The corresponding constraints are then converted into independent power upper and lower limit constraints.

6. The scenario-driven multi-stage robust re-optimization scheduling method for V2G integrated community microgrids according to claim 1, characterized in that: The intraday rolling re-optimization includes: Obtain the current actual photovoltaic output, actual load, and actual connected electric vehicle charge status; Calculate the initial scheduling power based on the affine parameters obtained in the previous phase; Reintroduce charging and discharging mutual exclusion constraints and buying and selling electricity mutual exclusion constraints; Establish a mixed-integer optimization subproblem at the current time scale; The final execution power of the current stage is obtained by solving the problem.

7. The scenario-driven multi-stage robust re-optimization scheduling method for V2G integrated community microgrids according to claim 1, characterized in that: The state of charge trajectory locking includes: calculating the SOC change trajectory of the energy storage system and the electric vehicle based on the day-ahead affine scheduling results; During the intraday rolling optimization process, the SOC change trajectory is used as a constraint to limit the cumulative energy shift caused by the local optimization process.

8. A scenario-driven multi-stage robust re-optimization scheduling system for V2G integrated community microgrids, characterized in that, It includes a model building module for establishing a microgrid dispatch model that includes photovoltaic, load, energy storage, electric vehicles and grid interaction units; It includes an uncertainty analysis module, which is used to acquire and construct a set of uncertainties related to photovoltaic, load, and electric vehicle behavior; It includes a multi-stage robust optimization module, which is used to establish a multi-stage robust scheduling model based on the principle of unpredictability; It includes a scene optimization module, which is used to construct a desired cost optimization target based on scene sampling; It includes a robust transformation module, which is used to transform robust constraints with uncertainties into deterministic optimization constraints; It includes a rolling optimization module, which is used to perform day-ahead relaxation optimization and intraday recovery optimization.

9. An electronic device, characterized in that, include: processor; Memory; And computer programs stored in the memory and executable by the processor; When the computer program is executed, it causes the processor to execute the scenario-driven multi-stage robust re-optimization scheduling method for V2G integrated community microgrids as described in any one of claims 1-9.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program; When the computer program is executed by the processor, it implements the scenario-driven multi-stage robust re-optimization scheduling method for V2G integrated community microgrids as described in any one of claims 1-9.

Images