High-robustness v2g-vpp multi-time scale closed-loop feedback collaborative scheduling method

Abstract

The application relates to a high-robustness V2G-VPP multi-time-scale closed-loop feedback cooperative scheduling method, which comprises the following steps: acquiring multi-source data of a V2G-VPP system in real time, and performing data cleaning and feature extraction processing; a day-ahead-intra-day-real-time three-level scheduling architecture is constructed, a multi-dimensional uncertainty modeling method is adopted to quantitatively describe the uncertainty of key parameters in the system, and in the day-ahead robustness scheduling, a day-ahead scheduling plan is formulated with the minimum total cost as the target; the day-ahead scheduling plan is used as the basis for intra-day robustness scheduling, and the intra-day corrected scheduling plan is obtained by further correction according to the SOC state and the charging and discharging cost; the intra-day corrected scheduling plan is used as the basis for real-time scheduling, real-time scheduling instructions are formed, feedback signals are formed according to the execution results, demand response incentives are adjusted, and a closed-loop feedback is formed. Compared with the prior art, the application can significantly improve the robustness of the virtual power plant to extreme operating scenarios.

Description

Technical Field

[0001] This invention relates to the field of vehicle-to-grid interactive virtual power plant technology, and in particular to a highly robust V2G-VPP multi-timescale closed-loop feedback collaborative scheduling method. Background Technology

[0002] As the global energy structure transitions towards a low-carbon model, vehicle-to-grid (V2G) technology has become a crucial component of smart grids. V2G technology effectively balances grid load and enhances renewable energy absorption capacity by allowing electric vehicles to supply power back to the grid during peak demand periods. Meanwhile, virtual power plants (VPPs), as aggregation platforms for distributed energy resources, can integrate dispersed EV charging facilities, energy storage systems, and renewable energy generation units into unified and controllable virtual generation units, achieving efficient resource scheduling and optimized allocation.

[0003] The deep integration of V2G and VPP creates a new energy interaction model that integrates "vehicle-charging station-grid". Its core lies in dynamically regulating EV charging and discharging behavior to transform massive distributed EV resources into flexible loads or energy storage units that can be dispatched by the grid. This is of great significance for enhancing grid stability, reducing operating costs, promoting the decarbonization of the energy system, addressing the volatility of new energy sources, and building a new type of power system.

[0004] However, the current V2G-VPP scheduling system still faces multiple shortcomings and problems:

[0005] (1) Insufficient multi-timescale coordination capability

[0006] Existing V2G-VPP scheduling strategies mostly employ single-time-scale optimization, making it difficult to balance long-term resource planning with short-term fluctuation response. Research shows that day-ahead scheduling relies on user-submitted charging plans, but there is a certain degree of deviation in actual execution; intraday scheduling, due to the lack of a real-time feedback mechanism, is unable to adequately address the randomness of EV charging behavior; and real-time scheduling suffers from excessively long response times (often exceeding 5 minutes) due to computational delays, making it difficult to meet grid frequency regulation requirements.

[0007] (2) Uncertainty modeling methods are lagging behind

[0008] EV charging demand is deeply influenced by a combination of factors, including user behavior patterns, battery health, and time-of-use pricing mechanisms. This makes it difficult for traditional deterministic optimization models to accurately depict its stochastic fluctuations, often resulting in significant biases in predictions. While uncertainty models based on probability distributions, such as scenario generation methods, can more comprehensively reflect the fluctuations in actual charging demand, their modeling accuracy still needs further improvement. Furthermore, renewable energy output is highly dependent on weather conditions, especially during cloudy or rainy weather, where photovoltaic power generation often fluctuates dramatically. This results in insufficient fault tolerance for fixed-parameter-based scheduling strategies, making it difficult to meet the real-time grid balance requirements.

[0009] In response, Chinese patent CN113408962A discloses a multi-timescale, multi-objective energy optimization scheduling method for power grids, comprising: S1, collecting historical operating data of the power grid; S2, based on the historical operating data, establishing an uncertain description of the power grid's weakly robust multi-objective optimization model, a day-ahead optimization scheduling objective function, and day-ahead optimization scheduling constraints, and constructing a day-ahead long-term weakly robust multi-objective optimization scheduling model; S3, solving the day-ahead long-term weakly robust multi-objective optimization scheduling model to obtain the day-ahead scheduling plan; S4, based on the actual scheduling plan of the power grid and the day-ahead scheduling plan, establishing intraday control time domain and performance indicators, and constructing an intraday MPC rolling optimization scheduling model; S5, based on the intraday MPC rolling optimization scheduling model, continuously revising the day-ahead scheduling plan in real time. This invention can be well applied to power grid optimization scheduling, and achieves good economic and environmental benefits while ensuring a certain level of robustness. However, this method and similar prior art still have the following drawbacks:

[0010] (1) Lack of closed-loop feedback mechanism

[0011] Current dispatching systems generally employ an open-loop control architecture, resulting in significant information transmission gaps between the "day-ahead, intraday, and real-time" three-tier dispatching system. Specifically, during the intraday dispatching phase, the lack of real-time data feedback channels makes it difficult to dynamically revise the day-ahead charging plans, causing large deviations between the grid load curve and forecasts. During the real-time dispatching phase, the absence of a user-side response mechanism prevents adjustments to control strategies based on the actual charging status of EVs, leading not only to low utilization of charging facilities but also hindering the use of price incentives to increase user participation. This fragmented control model directly weakens the system's responsiveness to sudden load changes, easily triggering local grid power imbalances in scenarios with a high proportion of renewable energy integration, ultimately impacting the overall economic efficiency and stability of the grid operation.

[0012] (2) Insufficient robustness of the system

[0013] Current V2G-VPP collaborative dispatching technology suffers from significant shortcomings in adaptability to extreme scenarios. Most research focuses excessively on optimizing economic indicators, neglecting system resilience under complex conditions such as equipment failures and extreme weather. Existing dispatching strategies are prone to stability failures when responding to emergencies due to a lack of targeted design, such as voltage drops and response delays, directly threatening the safe and stable operation of the power grid. This "economy-first, resilience-light" technological orientation results in weak fault tolerance and recovery capabilities under extreme scenarios, making it difficult to meet the high reliability requirements of modern power systems.

[0014] In summary, the current V2G-VPP scheduling system has significant shortcomings in closed-loop feedback and system robustness, making it difficult to cope with the complex scenarios following large-scale EV grid integration. The existing technology's focus on "economy over resilience" results in weak fault tolerance under extreme conditions, leading to increased grid load fluctuations, limited renewable energy absorption capacity, and a continuous rise in grid security risks. Therefore, there is an urgent need to focus on breakthroughs in key technologies such as multi-scale collaborative optimization, the construction of closed-loop feedback mechanisms, and robustness enhancement, providing a systemic technological innovation to offer a V2G-VPP collaborative scheduling solution that combines economy and security. Summary of the Invention

[0015] The purpose of this invention is to address the multiple challenges faced by power systems under high-proportion renewable energy integration, such as the volatility of renewable energy output, the uncertainty of load demand, and the randomness of electric vehicle (EV) integration. This invention provides a robust V2G-VPP multi-timescale closed-loop feedback collaborative scheduling method. By constructing a "day-ahead-intraday-real-time" multi-timescale closed-loop feedback collaborative scheduling architecture and integrating robust optimization, a V2G-VPP collaborative scheduling strategy capable of adaptive adjustment and dynamic response is formed. This scheme aims to significantly improve the robustness of virtual power plants to extreme operating scenarios, achieving comprehensive optimization of system operating economy, renewable energy absorption capacity, and user participation while ensuring the safe and stable operation of the power grid.

[0016] The objective of this invention can be achieved through the following technical solutions:

[0017] According to a first aspect of the present invention, a robust V2G-VPP multi-timescale closed-loop feedback cooperative scheduling method is provided, the method comprising the following steps:

[0018] Real-time acquisition of multi-source data from the V2G-VPP system, followed by data cleaning and feature extraction processing;

[0019] A three-level scheduling architecture—day-today, intraday, and real-time—is constructed. Based on processed data, a multi-dimensional uncertainty modeling method is used to quantify the uncertainty of key parameters in the V2G-VPP system. In day-to-day robust scheduling, a day-to-day scheduling plan is formulated with the goal of minimizing total cost. Intraday robust scheduling is performed based on the day-to-day scheduling plan, and hourly dynamic adjustments are achieved through rolling optimization. Further adjustments are made based on SOC status and charging / discharging costs to obtain an intraday revised scheduling plan. Real-time scheduling is performed based on the intraday revised scheduling plan, and a model predictive control algorithm is used for minute-level tracking and response to generate real-time scheduling instructions. Feedback signals are generated based on the execution results to adjust demand response stimuli, forming a closed-loop feedback.

[0020] As a preferred technical solution, the method of using multi-dimensional uncertainty modeling to quantitatively describe the uncertainty of key parameters in the V2G-VPP system specifically involves:

[0021] Establish a joint probability distribution model for distributed generation and load demand:

[0022] ,

[0023] in, N ( m , s (This represents a normal distribution of new energy output.) LN ( m , s ) represents the log-normal distribution of load demand. m and s These represent the mean and standard deviation of historical data, respectively. The subscript "gen" indicates distributed generation, and the subscript "load" indicates the load. t Indicates time; Indicates the scheduling time;

[0024] Typical scene set generated based on Monte Carlo simulation Each scenario includes a sequence of new energy power outputs. and load demand sequence The scene weights are determined by normalizing the probability density function.

[0025] As a preferred technical solution, the objective function of the day-ahead robust scheduling is to minimize the sum of total generation cost, EV user charging cost, and discharge revenue, expressed as:

[0026] ,

[0027] Among them, decision variables , , , These represent EV charging power, EV discharging power, and distributed generation power within the V2G-VPP system, respectively; uncertainty parameters. , Indicates load demand; D It is a set of uncertainty parameters generated based on Monte Carlo simulation; t Indicates time; express t The unit generation cost of distributed generation at any time; Representing a scene exist t Distributed generation power at any given time; i The EV number indicates the charging status. j The EV number indicating the discharge state. N This indicates the total number of EVs participating in the charging process. M This indicates the total number of EVs participating in the discharge; and They represent the first i EVs in t The charging power and charging cost at any given time. and They represent the first j EVs in t Discharge power and discharge benefit at any given moment; t Indicates the error weighting coefficient; To determine the day-ahead load forecast deviation, the actual load value of the V2G-VPP system is obtained. And compared with the current forecast load value The difference is obtained by calculation; This represents the prediction error penalty term.

[0028] As a preferred technical solution, the constraints of the day-ahead robust scheduling include power balance constraints, EV battery state of charge constraints, distributed unit ramping constraints, and charging rate constraints, as shown below:

[0029] ,

[0030] ,

[0031] ,

[0032] ,

[0033] ,

[0034] in, , These represent the lower and upper limits of the EV battery's state of charge, respectively. Indicates the initial state of charge. This indicates the charging efficiency of the EV. This indicates the discharge efficiency of the EV. Indicates the length of a period of uncertainty. This represents the ramp rate of the distributed generator set. For time intervals, Indicates the maximum charging power of the EV. Indicates the EV discharge rate. This indicates the maximum discharge power of the EV.

[0035] As a preferred technical solution, the objective function of the intraday robust scheduling is as follows:

[0036] ,

[0037] in, As decision variables, For uncertain parameters, D It is a set of uncertainty parameters generated based on Monte Carlo simulation; t Indicates time; Indicates the scheduling time; express t The unit generation cost of distributed generation at any time; Representing a scene exist t Distributed generation power at any given time; i The EV number indicates the charging status. j The EV number indicating the discharge state. N This indicates the total number of EVs participating in the charging process. M This indicates the total number of EVs participating in the discharge; and They represent the first i EVs in t The charging power and charging cost at any given time. and They represent the first j EVs in t Discharge power and discharge benefit at any given moment; This is the intraday correction factor; The power deviation between intraday scheduling and the previous day's plan, ,in, Indicates EV in t Discharge power at any given time Indicates EV in t Charging power at any time Representing a scene exist t The time slot relative to the revised load of the day-ahead plan is shown in the following formula: ,in, α ( t )for t Error correction coefficient at time, This is the day-ahead load forecast deviation. This represents the day-ahead forecast load value.

[0038] As a preferred technical solution, the constraints of the intraday robust scheduling include power balance constraints, EV battery SOC dynamic constraints, distributed unit ramping constraints, and charging rate constraints, as shown below:

[0039] ,

[0040] ,

[0041] ,

[0042] ,

[0043] ,

[0044] in, express t EV battery state of charge at any given time This indicates the charging efficiency of the EV. This indicates the discharge efficiency of the EV. This represents the ramp rate of the distributed generator set. For time intervals, Indicates the maximum charging power of the EV. Indicates the EV discharge rate. This indicates the maximum discharge power of the EV.

[0045] As a preferred technical solution, the error correction coefficient α ( t )By real-time charging fees Discharge benefits Real-time status of EV battery SOC SOC ( t Collaborative execution is dynamically adjusted through weighted coefficient allocation:

[0046] ,

[0047] in, , and These represent the adjustment weighting coefficients for real-time charging costs, discharge benefits, and EV battery SOC state, respectively, satisfying... ; This indicates the maximum charging cost. This represents the maximum discharge benefit. This indicates the upper limit of the EV battery's state of charge.

[0048] As a preferred technical solution, the objective function of the real-time scheduling is:

[0049] ,

[0050] in, As decision variables, For uncertain parameters, D It is a set of uncertainty parameters generated based on Monte Carlo simulation; k Sampling time; express t The future of the time period k Each sampling time; express t The future of the time period k System power deviation at each sampling time , Representing a scene exist t The future of the time period k Distributed generation power at each sampling time, , They represent in t The future of the time period k EV discharge power and charging power at each sampling time. Representing a scene exist t The future of the time period k Each sampling time corresponds to the revised load of the day-ahead plan; This represents a penalty term for changes in the output of distributed units in a V2G-VPP system. Indicates in t The future of the time period k The dynamic weighting factor at each sampling time. , This serves as the baseline value for the dynamic weighting factor. Indicates the benchmark electricity price. For adjustment coefficients, Indicates the future number k Real-time electricity price at each sampling point. This represents the error correction coefficient for intraday scheduling. Representing a scene exist t The future of the time period k Changes in distributed generation output at each sampling time; W track The coefficient of the load tracking deviation term in the objective function.

[0051] As a preferred technical solution, the constraints of the real-time scheduling include dynamic constraints on EV battery SOC and distributed unit ramping constraints, as shown below:

[0052] ,

[0053] ,

[0054] in, express t The future of the time period k The EV battery state of charge at each sampling time. This indicates the charging efficiency of the EV. This indicates the discharge efficiency of the EV. This represents the ramp rate of the distributed generator set. For time intervals.

[0055] As a preferred technical solution, a feedback signal is generated based on the execution result, and the demand response incentive is adjusted to form a closed-loop feedback, specifically:

[0056] Based on the target value of load adjustment in the real-time dispatch command Calculate the response deviation rate:

[0057] ,

[0058] in, This is the actual load adjustment amount. For response deviation rate;

[0059] Calculate the cost adjustment factor based on the response deviation rate:

[0060] ,

[0061] in, For the next cycle t Cost adjustment factor required at time +1; This is the cost adjustment factor for this period; For the adjustment amplitude factor, 0 < <1, used to control the magnitude and stability of the correction;

[0062] Adjust the demand response incentives based on the cost adjustment factor, and update the objective function value in the day-ahead scheduling:

[0063] ,

[0064] in, Indicates the first i EVs in t The cost of charging at any time, Indicates the first j EVs in t The discharge benefit at any given moment; Indicates the first i EVs in the next cycle t The charging cost at +1 hour; Indicates the first j EVs in the next cycle t The discharge gain at time +1.

[0065] According to a second aspect of the present invention, an electronic device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the program to implement the method described thereon.

[0066] According to a third aspect of the present invention, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the method described thereon.

[0067] Compared with the prior art, the present invention has the following beneficial effects:

[0068] (1) This invention constructs a three-level hierarchical collaborative optimization architecture of "day-to-day-real-time", which integrates robust optimization processing. Day-to-day scheduling generates a baseline plan that can withstand the worst uncertainty scenarios; day-to-day and real-time scheduling are then rolled over and precisely tracked on this basis. This architecture achieves a balance between long-term robustness and short-term accuracy, fundamentally understanding the contradiction between the variability of the decision-making environment and the stability of the scheduling scheme.

[0069] (2) This invention designs a closed-loop feedback mechanism deeply coupled with scheduling. This mechanism monitors user response deviations in real time and dynamically corrects the demand response cost adjustment coefficient based on a quantitative evaluation model, thus realizing an automated closed loop of "evaluation-incentive-optimization". This data-driven adaptive incentive strategy significantly improves user response accuracy and participation, ensuring the sustainability and practical application effect of the V2G-VPP aggregation scheduling strategy.

[0070] (3) This invention proposes a multi-dimensional uncertainty collaborative modeling method for new energy output, load and EV behavior. It uses probability distribution and Monte Carlo scene generation technology to accurately quantify uncertainty and constructs a robust optimization model with the minimum-maximum criterion as the core. This strategy ensures that under extreme conditions of new energy fluctuation ±15% and load deviation ±10%, the system cost fluctuation can still be controlled within 3%, which significantly improves the safe and stable operation capability of V2G-VPP under extreme scenarios. Attached Figure Description

[0071] Figure 1 This is a schematic diagram of the V2G-VPP system architecture;

[0072] Figure 2 This is a schematic diagram of the V2G-VPP multi-timescale scheduling architecture of the present invention;

[0073] Figure 3 This is a schematic diagram of the closed-loop feedback mechanism of the present invention. Detailed Implementation

[0074] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0075] Unless otherwise defined, the technical or scientific terms used in this application shall have the ordinary meaning understood by one of ordinary skill in the art to which this application pertains. The terms “a,” “an,” “an,” “the,” and similar words used in this application do not indicate quantity limitation and may indicate singular or plural. The terms “comprising,” “including,” “having,” and any variations thereof used in this application are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or device that includes a series of steps or modules (units) is not limited to the listed steps or units, but may also include steps or units not listed, or may include other steps or units inherent to these processes, methods, products, or devices. The terms “connected,” “linked,” “coupled,” and similar words used in this application are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. “Multiple” used in this application refers to two or more. “And / or” describes the relationship between related objects, indicating that three relationships may exist; for example, “A and / or B” can represent: A alone, A and B simultaneously, and B alone. The character " / " generally indicates that the preceding and following objects are in an "or" relationship. The terms "first," "second," and "third" used in this application are merely to distinguish similar objects and do not represent a specific ordering of the objects.

[0076] Example 1

[0077] 1. V2G-VPP System Structure and Operation Mechanism

[0078] V2G-VPP (Vehicle-to-Grid Virtual Power Plant) is an intelligent aggregation system that integrates electric vehicles (EVs) and distributed energy resources, such as... Figure 1As shown, the system comprises four core modules: distributed power sources (PV / wind power), energy storage devices, flexible loads, and EV clusters. The system achieves energy interaction between EVs and the grid through vehicle-to-grid (V2G) technology, combining the local absorption of distributed energy sources with the spatiotemporal migration capabilities of energy storage to construct a multi-energy complementary flexible grid unit. Its core principle lies in addressing fluctuations in renewable energy output, the randomness of EV user behavior, and the uncertainty of battery state of charge (SOC) through multi-timescale collaborative optimization. It utilizes the aggregation effect of V2G-VPP to transform dispersed resources into dispatchable virtual power plants. Key functions include: real-time coordination of EV charging and discharging plans to smooth power fluctuations, dynamic matching of distributed energy output with load demand, and improving the system's robustness to extreme scenarios through a closed-loop feedback mechanism, ultimately achieving dual optimization of energy efficiency and grid stability.

[0079] 2. Multi-timescale scheduling architecture

[0080] Based on the V2G-VPP system architecture and operating mechanism, this embodiment designs a V2G-VPP multi-timescale scheduling architecture, the structure of which is as follows: Figure 2 As shown, the architecture consists of three layers: a data acquisition layer, an edge computing layer, and a three-level scheduling layer.

[0081] The "data acquisition layer" collects multi-source data such as distributed photovoltaic, load, EV battery, electricity price, and weather through various IoT devices (smart meters / battery management system (BMS) / weather station), and aggregates and connects to the edge computing unit through an efficient information transmission channel.

[0082] The "edge computing layer" performs a series of data refinement and optimization analyses on the collected multi-source data, including data cleaning, feature extraction, and local optimization, providing the "three-level scheduling layer" with distinctive, refined, and effective data. This layer requires sufficient CPU and memory, deploys algorithm libraries, and communicates with the scheduling center via a 5G network with a latency of ≤30ms.

[0083] After receiving and analyzing the multi-source data from the first two layers, the "three-level scheduling layer" formulates a day-ahead scheduling plan through "cost minimization" to achieve economic incentives for V2G-VPP and EV participating users. Based on the day-ahead scheduling plan, an intraday scheduling plan is formulated through "mixed integer linear programming (MILP) rolling optimization" and further revised. Finally, based on the intraday revised scheduling plan, real-time scheduling instructions are generated through model predictive control (MPC) and the execution results are fed back to the data acquisition layer in real time.

[0084] Based on this architecture, this embodiment provides a robust V2G-VPP multi-timescale closed-loop feedback cooperative scheduling method, which includes the following steps:

[0085] Real-time acquisition of multi-source data from the V2G-VPP system, followed by data cleaning and feature extraction processing;

[0086] A three-level scheduling architecture—day-today, intraday, and real-time—is constructed. Based on processed data, a multi-dimensional uncertainty modeling method is used to quantify the uncertainty of key parameters in the V2G-VPP system. In day-to-day robust scheduling, a day-to-day scheduling plan is formulated with the goal of minimizing total cost. Intraday robust scheduling is performed based on the day-to-day scheduling plan, and hourly dynamic adjustments are achieved through rolling optimization. Further adjustments are made based on SOC status and charging / discharging costs to obtain an intraday revised scheduling plan. Real-time scheduling is performed based on the intraday revised scheduling plan, and a model predictive control algorithm is used for minute-level tracking and response to generate real-time scheduling instructions. Feedback signals are generated based on the execution results to adjust demand response stimuli, forming a closed-loop feedback.

[0087] This method, through a "three-layer architecture - closed-loop operation - multi-algorithm fusion" design, achieves comprehensive optimization of complex V2G-VPP multi-timescale scheduling systems. Its scheduling system mechanism is shown in Table 1, and its core technical features include:

[0088] (1) Vertical time-sharing optimization system

[0089] ① Day-ahead scheduling: Receives ultra-short-term forecast data (such as PV output, load demand, EV access status, and electricity price information), historical and meteorological data, V2G-VPP specific data, market data, and feedback data (mainly correction signals from intraday rolling optimization feedback) from the data acquisition and processing module. Based on this information, this layer constructs a load forecasting model using time series analysis, regression analysis, or deep learning algorithms, and optimizes the hyperparameters of the forecasting model using heuristic algorithms as a baseline plan. With the goal of minimizing the total system operating cost, and introducing a forecast error penalty term, a baseline scheduling plan for the next 24 hours on a 15-minute timescale is formulated and distributed to the intraday scheduling layer.

[0090] ② Intraday Scheduling: Initiated every 15 minutes. Its core inputs include ultra-short-term high-precision forecast data and real-time feedback data from the underlying layers (such as actual EV charging status, real-time grid frequency, etc.). This layer adopts a mixed-integer linear programming (MILP) model to continuously optimize the scheduling scheme for the next few hours, dynamically correcting the day-ahead plan to smooth fluctuations and compensate for forecast errors. Blockchain storage ensures the traceability of adjustment strategies, and the correction coefficient is dynamically correlated with real-time electricity prices and EV battery SOC. The corrected instructions are then sent to the underlying controller of the real-time scheduling layer.

[0091] ③ Real-time scheduling: With the shortest cycle (5 minutes or even seconds), it can respond quickly, directly receive real-time measurement data of the power grid (such as node voltage and frequency deviation), and use the model predictive control (MPC) algorithm to perform fast and accurate power control on distributed units. The weight factor balances the tracking accuracy and control quantity to eliminate real-time disturbances and maintain the instantaneous balance of the power grid.

[0092] (2) Horizontal data closed loop path

[0093] IoT devices (smart meters / BMS / weather stations) collect multi-source data, which is then cleaned and feature extracted by the edge computing unit and input into the three-level scheduling layer to form dynamic instructions. The execution results of the instructions are fed back to the data acquisition layer to form a self-learning closed loop, which meets the data flow coupling requirements of "vehicle-pile-network".

[0094] (3) Deep integration of V2G-VPP architecture

[0095] Data such as EV battery SOC and charging / discharging power are integrated throughout the entire process to achieve VPP aggregation effect. By linking demand response subsidies with carbon trading prices, a dual-drive model of "incentive mechanism + technical constraints" is constructed to achieve the effect of "user-side adjustable resource aggregation".

[0096] This system significantly improves the capacity for renewable energy absorption and the economic efficiency of grid operation through a spatiotemporal coupling optimization strategy and a vehicle-pile-grid collaborative architecture.

[0097] Table 1. V2G-VPP Three-Level Scheduling System Mechanism

[0098]

[0099] 2.1 Robust Optimization Scheduling Method with Multiple Time Scales

[0100] A multi-dimensional uncertainty modeling method is adopted to quantitatively describe the uncertainties of key parameters such as renewable energy output, load demand, and EV access status in the V2G-VPP system.

[0101] Therefore, for multi-timescale scheduling of "day-to-day-real-time", a robust optimization scheduling model based on the min-max criterion with three time scales is constructed in sequence to minimize the total system cost under the worst case of uncertainty parameters.

[0102] 2.1.1 Uncertainty Modeling

[0103] (1) Probability distribution model

[0104] A joint probability distribution model is established for photovoltaic / wind power output and load demand:

[0105] ,

[0106] in, N ( m , s (This represents a normal distribution of new energy output.) LN ( m , s ) represents the log-normal distribution of load demand. m and s These represent the mean and standard deviation of historical data, respectively. The subscript "gen" indicates distributed generation, and the subscript "load" indicates the load. t Indicates time; Indicates the scheduling time.

[0107] (2) Scene generation

[0108] Typical scene set generated based on Monte Carlo simulation Each scenario includes a sequence of new energy power outputs. and load demand sequence The scene weights are determined by normalizing the probability density function.

[0109] 2.1.2 Characteristics and Advantages of Multi-Time-Scale Robust Optimization Scheduling Model

[0110] (1) Model features

[0111] Robust optimization models adjust decision variables (such as charging power) P charge Discharge power P discharge Power purchased by the power grid P grid The robust model addresses uncertainty by addressing it rather than modifying physical constraints. The core of a robust model is ensuring the objective function (total cost) remains acceptable in the worst-case scenario, while constraints guarantee the basic operational feasibility of the system. Uncertainties in the objective function (such as load demand) are addressed by... L ( t , d ), new energy power output P gen ( t , d This directly affects the target cost and needs to be addressed through robust optimization.

[0112] Battery state of charge (SOC) range [SOC] min SOC max Upper limit of charging and discharging power P max These parameters are determined by the battery's chemical characteristics and hardware design, and are inherent properties of the equipment, unrelated to external uncertainties (such as load fluctuations and errors in new energy output prediction). Meanwhile, the unit's ramp-up rate...R ramp Determined by the mechanical characteristics of the generator set, it is a hard constraint at the equipment level and is unrelated to external uncertainties.

[0113] Therefore, the EV battery state constraints and distributed unit ramping constraints do not incorporate uncertainty parameters. This is a result of the combined effects of the deterministic nature of physical limitations, the adjustment space of control variables, the focus on robust optimization, and the requirement for model solvability. This design balances robustness and computational efficiency while ensuring safe system operation, meeting the dual requirements of real-time performance and economy for V2G-VPP systems.

[0114] (2) Core function of the model

[0115] The robust performance optimization strategy aims to systematically solve key technical challenges in V2G-VPP multi-timescale scheduling, such as fluctuations in renewable energy output, uncertainty in load demand, and randomness in EV access, through deep integration of the V2G-VPP three-level scheduling system and the robust core model. Its core role is reflected in three aspects:

[0116] 1. Uncertainty Quantification and Fault Tolerance: Through probability distribution modeling and scenario generation technology, the randomness of parameters such as new energy sources and loads is accurately characterized. Combined with a robust optimization model based on the minimum-maximum criterion, the system can maintain power balance and economic objectives even in extreme scenarios.

[0117] 2. Multi-timescale collaborative optimization: Construct a three-level robust optimization scheduling architecture of "day-to-day-real-time". The day-to-day layer formulates a baseline plan based on robust optimization, and the intra-day / real-time layer combines MILP and MPC algorithms to complete dynamic correction under robust optimization, forming a progressive optimization chain of "robust planning - rapid adjustment - real-time response".

[0118] 3. Closed-loop feedback and adaptive adjustment: By using the deviation rate index and the demand response incentive mechanism, the execution results are fed back to the robust scheduling model to realize dynamic adjustment of scenario weights and self-optimization of parameters, thus avoiding model rigidity.

[0119] The ultimate goal of the model is to maximize the absorption capacity of new energy sources and reduce system operating costs while ensuring the safe and stable operation of the power grid. At the same time, it aims to enhance user participation through a "vehicle-pile-grid" collaborative architecture and provide a replicable and scalable robust dispatch solution for power systems with a high proportion of new energy access.

[0120] 2.2 Robust Optimization of Scheduling (24 Hours)

[0121] Recently, the scheduling system has been constructing load forecasting models using time series methods (such as the ARIMA model), regression analysis, or deep learning algorithms (such as LSTM), and then using heuristic algorithms (such as particle swarm optimization (PSO)) to optimize the hyperparameters of the forecasting models and generate load forecasts. , as a benchmark plan.

[0122] The objective function for robust scheduling is currently to minimize the sum of total generation cost, EV user charging cost, and discharge revenue, expressed as:

[0123] ,

[0124] Among them, decision variables , , , These represent EV charging power, EV discharging power, and distributed generation power within the V2G-VPP system, respectively; uncertainty parameters. , Indicates load demand; D It is a set of uncertainty parameters generated based on Monte Carlo simulation; t Indicates time; express t The unit generation cost of distributed generation at any time; Representing a scene exist t Distributed generation power at any given time; i The EV number indicates the charging status. j The EV number indicating the discharge state. N This indicates the total number of EVs participating in the charging process. M This indicates the total number of EVs participating in the discharge; and They represent the first i EVs in t The charging power and charging cost at any given time. and They represent the first j EVs in t Discharge power and discharge benefit at any given time.

[0125] The objective function includes a prediction error penalty term. This ensures that the optimization model takes into account prediction uncertainties. Among these, t This represents the error weighting coefficient, with a typical value of 0.1 to 0.3. To account for day-ahead load forecast deviation, the actual load value of the V2G-VPP system is obtained through the SCADA system. And compared with the current forecast load value Calculate the difference and establish an error statistical model: .

[0126] The error statistical model uses the posterior error test to quantify the prediction error and generate a probabilistic load prediction range (e.g., the load prediction error range at a 95% confidence level is [-5%, +5%]).

[0127] The current constraints on robust dispatching include power balance constraints, EV battery state of charge (SOC) constraints, distributed unit ramping constraints, and charging rate constraints, as shown below:

[0128] ,

[0129] in, , These represent the lower and upper limits of the EV battery's state of charge, respectively. Indicates the initial state of charge. Indicates the length of a period of uncertainty. This indicates the ramp rate of distributed generator sets (mainly referring to distributed photovoltaic, distributed wind power, etc.). For time intervals, Indicates the maximum charging power of the EV. Indicates the EV discharge rate. Indicates the maximum discharge power of the EV; The charge / discharge efficiency of an EV is expressed by the following formula:

[0130]

[0131] in, Indicates the actual charge / discharge amount. This represents the theoretical maximum charge / discharge capacity.

[0132] The calculation method is as follows:

[0133]

[0134] in, This indicates the current output of the distributed generator set. This indicates the output of the distributed generator set at the previous moment. This refers to the ramp-up time period for the distributed generator set (i.e., the time interval mentioned above).

[0135] 2.3 Robust scheduling optimization within the day (1 hour)

[0136] Intraday scheduling performs MILP optimization every 15 minutes, introducing a feedback correction term to correct daytime plan deviations, and using CPLEX or Gurobi to solve the MILP model. Based on intraday time-of-use electricity prices, the target time period is minimized. t ϵ[1,T The total cost, and its objective function are as follows:

[0137] ,

[0138] in, Indicates the scheduling time; This is the intraday correction factor; The power deviation between intraday scheduling and the previous day's plan, ,in, Indicates EV in t Discharge power at any given time Indicates EV in t Charging power at any time Representing a scene exist t The time slot relative to the revised load of the day-ahead plan is shown in the following formula: ,in, α ( t )for t The error correction factor at any given time is used for system cost adjustment. This is the day-ahead load forecast deviation. This represents the day-ahead forecast load value.

[0139] In a preferred embodiment, the error correction coefficient α ( t )By real-time charging fees Discharge benefits Real-time status of EV battery SOC SOC ( t Collaborative execution is dynamically adjusted through weighted coefficient allocation:

[0140] ,

[0141] in, , and These represent the adjustment weighting coefficients for real-time charging costs, discharge benefits, and EV battery SOC state, respectively, satisfying... Typically, these values ​​are taken as 0.3, 0.3, and 0.4 respectively. This indicates the maximum charging cost. This represents the maximum discharge benefit. This indicates the upper limit of the EV battery's state of charge.

[0142] The larger the error correction coefficient, the greater the correction of the deviation. Fluctuations can be compensated first by energy storage or flexible resources in the V2G-VPP system.

[0143] The constraints for robust intraday scheduling include power balance constraints, dynamic EV battery SOC constraints, distributed unit ramping constraints, and charging rate constraints, as shown below:

[0144] ,

[0145] in, express t The state of charge of the EV battery at any given time.

[0146] 2.4 Real-time robustness optimization scheduling (15 minutes)

[0147] Real-time scheduling employs an MPC controller. The prediction range covers the next 15 minutes (3 sampling periods), and it is integrated with the daily rolling optimization cycle of scheduling, executing control every 5 minutes (1 sampling period) to maintain real-time responsiveness. Weighting factors... l (t) Corrected by real-time electricity price and intraday error factor α ( t Dynamic adjustments are made to balance tracking performance with changes in control parameters, suppressing frequent unit adjustments during peak hours and improving tracking performance during off-peak hours. The initial values ​​of EV battery SOC and unit output are directly taken from the intraday scheduling results, ensuring seamless coordination of multi-timescale scheduling.

[0148] The objective function for real-time scheduling is:

[0149] ,

[0150] in, k Sampling time; express t The future of the time period k Each sampling time; express t The future of the time period k System power deviation at each sampling time , Representing a scene exist t The future of the time period k Distributed generation power at each sampling time, , They represent in t The future of the time period k EV discharge power and charging power at each sampling time. Representing a scene exist t The future of the time period k Each sampling time corresponds to the revised load of the previous day's plan. ,in, , They are respectivelyt The future of the time period k The day-ahead forecast load, the actual load, and the deviation between the day-ahead forecast at each sampling time; W track The coefficient of the load tracking deviation term in the objective function; This represents a penalty for changes in the output of distributed units in the V2G-VPP system. This penalty ensures the system prioritizes control strategies with smaller output changes, thereby reducing frequent start-ups, shutdowns, or output adjustments due to load fluctuations, lowering equipment wear and maintenance costs, and ensuring smooth and economical system control. Indicates in t The future of the time period k The dynamic weighting factor at each sampling time. , This serves as the baseline value for the dynamic weighting factor. Indicates the benchmark electricity price. For adjustment coefficients, Indicates the future number k Real-time electricity price at each sampling point. This represents the error correction coefficient for intraday scheduling. Representing a scene exist t The future of the time period k The change in distributed generation output at each sampling time.

[0151] W track As a coefficient of the load tracking deviation term in the objective function, it reflects the system's requirements for load tracking accuracy. A larger value indicates a stricter penalty for deviation, meaning the model will prioritize ensuring that the actual dispatch closely tracks and corrects the load. However, this may increase the frequency of generator adjustments and battery charge / discharge cycles, leading to increased equipment wear and operating costs. Conversely, a smaller value allows for greater tracking deviations to reduce other costs (such as generator adjustment costs or battery wear) and extend equipment lifespan. However, this may cause load tracking deviations, affecting grid stability and the quality of power supply to users.

[0152] W track The value of and the error correction coefficient of intraday scheduling α ( t Coordination. When α ( t When the load deviation is large (requiring significant correction for intraday scheduling), the load can be appropriately increased. W track This ensures that real-time scheduling closely executes the daily plan. When α ( t When the load deviation is small, it can be reduced. W trackTo reduce the tracking accuracy requirements and save costs.

[0153] W track The typical value range is 10-20 yuan / MW. The specific value needs to be adjusted according to system characteristics. During peak load periods, when electricity prices are high and load fluctuations are large, the value can be appropriately increased. W track (e.g., 15-20 yuan / MW) to ensure power supply reliability. During off-peak hours, low electricity prices and stable loads can reduce... W track (e.g., 10~15 yuan / MW), allowing for moderate deviations to reduce adjustment costs.

[0154] The constraints for real-time scheduling include dynamic constraints on EV battery SOC and distributed unit ramping constraints, as shown below:

[0155] ,

[0156] in, express t The future of the time period k The EV battery state of charge at each sampling time. This represents the ramp rate of the distributed generator set. For time intervals.

[0157] 2.5 Closed-loop feedback mechanism

[0158] Figure 3 The core processes of the closed-loop feedback architecture and the dynamic interactions between its modules are demonstrated. The system forms a complete closed-loop feedback mechanism through a demand response incentive system, EV user-side load, cost adjustment module, benefit accounting, optimization scheduling module, execution control module, physical system, response deviation analysis, and parameter correction module.

[0159] The execution process of a closed-loop feedback system includes:

[0160] (1) Response Deviation Analysis: The VPP's optimization scheduling module generates scheduling instructions based on the current cycle's demand response requirements and sends them to the physical system via the execution control module for equipment adjustment. The physical system receives the scheduling instructions from the execution control module, performs equipment adjustment, and collects equipment operating status data in real time. Simultaneously, based on the target value of the EV user-side load adjustment amount issued by the scheduling instructions, the load response deviation of the VPP scheduling within this cycle is evaluated, and the response deviation rate for this cycle is obtained. E .

[0161] (2) Parameter correction: Based on the deviation rate between the current cost adjustment coefficient and the response of this period, the cost adjustment coefficient required for the next period is obtained by controlling the adjustment magnitude factor of the correction.

[0162] (3) Demand response incentive update: Based on the cost adjustment coefficient of this period, calculate the incentive update for different EV users in the next period. t The charging cost or discharging benefit adjusted at +1 moment optimizes the subsequent V2G response performance.

[0163] (5) Rolling optimization: Solve the above model in each scheduling cycle, respond to changes in cost and EV user behavior in real time, realize the dynamic closed loop of "decision-execution-feedback", and form a strong coupling mechanism of "adaptive adjustment of demand response incentive intensity - rebalancing of optimization target".

[0164] This closed-loop feedback system ensures that the system can be continuously adjusted and optimized according to actual conditions, improving the overall performance and robustness of the V2G-VPP system. This mechanism not only enhances the real-time response capability of the V2G-VPP system, but also improves the system's adaptability and flexibility through continuous data feedback and parameter correction, effectively coping with various uncertainties.

[0165] Specifically, the process includes the following steps:

[0166] 1. Response deviation analysis.

[0167] Based on the target value of load adjustment in the real-time dispatch command Calculate the response deviation rate:

[0168] ,

[0169] in, E The response deviation rate is used to quantitatively assess the gap between the response effect and the expected target. This represents the actual load adjustment.

[0170] 2. Parameter correction.

[0171] Calculate the cost adjustment factor based on the response deviation rate:

[0172] ,

[0173] in, For the next cycle t Cost adjustment factor required at time +1; This is the cost adjustment factor for this period; Adjustment amplitude factor (0 < <1), used to control the magnitude and stability of the correction. This correction method dynamically adjusts the cost coefficient based on the deviation rate: when the response is insufficient ( E When the value is greater than 0, the incentive can be appropriately increased. C ( t When the response is excessive (E When <0), the incentive should be appropriately reduced. C ( t This forms an adaptive optimization closed-loop system designed to reduce future response bias.

[0174] 3. Adjustment of incentives for demand response.

[0175] Adjust the demand response incentives based on the cost adjustment factor, and update the objective function value in the day-ahead scheduling:

[0176] ,

[0177] in, Indicates the first i EVs in t The cost-adjusted charging cost at +1 hour; Indicates the first j EVs in t Cost-adjusted discharge revenue at time +1; Indicates the first i EVs in the next cycle t The charging cost at +1 hour; Indicates the first j EVs in the next cycle t The discharge gain at time +1. This formula reflects the dynamic linkage between cost adjustments and the actual user response.

[0178] Example 2

[0179] This embodiment, based on embodiment 1, provides a specific implementation process for data cleaning and feature extraction in an edge computing layer.

[0180] (1) Data cleaning

[0181] ① Sliding window filtering (noise reduction)

[0182] Sliding window mean filtering is used to suppress high-frequency noise.

[0183] ,

[0184] in, t For system time, y Number the data. H The sliding window length is typically 3 to 5 sampling periods, i.e., 15 to 25 minutes. These are the original measured values. This is the data after filtering and denoising.

[0185] ② Data interpolation (linear interpolation)

[0186] Linear interpolation is used to fill in missing or abnormal data. It is suitable for missing values ​​between two consecutive normal data points.

[0187] ,

[0188] in, The data after linear interpolation. and They are respectively t -1 time and t The original data at time +1.

[0189] (2) Feature extraction

[0190] Key features related to multi-timescale scheduling are extracted from the cleaned data to support the decision-making of the MPC controller. Scheduling-related features and state variable consistency features are extracted. A multi-dimensional adaptive feature extraction (MAFE) method is employed, with core feature vectors... Defined as:

[0191] ,

[0192] in, The mean of the data. For data variance, α ( t ) represents the error correction coefficient for intraday scheduling. To manage load deviations, This refers to the state of charge of the EV battery. This represents the ramp-up coefficient for distributed power generation units.

[0193] ① Data mean

[0194] ,

[0195] It reflects the average load level within the sliding window, supporting the MPC controller's tracking of the baseline values ​​of characteristic parameters. It is compared with the baseline characteristic parameters of the intraday scheduling MILP model. Among them, This represents the data length.

[0196] ② Data variance

[0197] ,

[0198] Quantifying load fluctuations guides the MPC controller to adjust unit output to cope with uncertainties. Among these, This represents the load data after sliding window filtering.

[0199] ② State variable consistency characteristics

[0200] 1) EV charging (discharging) efficiency

[0201] ,

[0202] in, Indicates EV charging (discharging) efficiency. Indicates the actual charge (discharge) capacity. This indicates the theoretical maximum charge (discharge) capacity.

[0203] 2) Battery SOC dynamic equation:

[0204] ,

[0205] in, express t EV battery state of charge at any given time This indicates the charge / discharge efficiency of the EV. , They represent t The charging and discharging power at any given time, For time intervals.

[0206] 3) Unit ramp-up rate:

[0207] ,

[0208] in, Indicates the current output of the generating unit. This indicates the unit's output at the previous moment.

[0209] Example 3

[0210] This embodiment provides a typical test case based on embodiments 1 and 2 to illustrate the effectiveness of the method.

[0211] 3.1 Test Scenario

[0212] ①EV Cluster: 100 EVs participate in the scheduling, with a battery capacity of 60±5kWh (mean ± standard deviation) and an initial SOC randomly distributed between 20% and 80%.

[0213] ② New energy configuration: 2MW distributed photovoltaic (forecasted output 5000kWh), 1.5MW wind power (forecasted output 3000kWh).

[0214] ③ Time-of-use pricing: A five-tier system is adopted (0.3 yuan / kWh for deep valley, 0.5 yuan for valley, 0.8 yuan for flat, 1.2 yuan for peak, and 1.5 yuan for peak).

[0215] ④ Robustness target: Ensure stable system operation under extreme scenarios with fluctuations of ±15% in renewable energy output and ±10% in load demand.

[0216] 3.2 Test Procedures and Results

[0217] Daytime scheduling (robust optimization baseline plan)

[0218] This phase aims to develop a robust benchmark plan that can cope with a variety of uncertainties.

[0219] ① Data preparation: The system first constructs a schedulable capacity model for the EV cluster based on historical 30-day EV travel data (off-grid time, starting SOC of charging), time-of-use electricity price curves, and predicted output of new energy vehicles.

[0220] ② Uncertainty modeling: Generate 5 typical scenarios ( This covers photovoltaic output fluctuations of ±15% and load demand deviations of ±10%, and employs a weight allocation based on posterior error testing to provide accurate input for robust optimization.

[0221] .

[0222] ③ Robust optimization solution: The system adopts a two-stage robust optimization framework and C&CG algorithm (column and constraint generation algorithm) for solution. The first stage determines the EV charging and discharging power benchmark value and the power boundary of VPP and grid interaction. The second stage verifies the feasibility under the worst scenario, thus ensuring the robustness of the benchmark plan.

[0223] ④ Closed-loop feedback: After the baseline plan is implemented, the system collects the actual load value and the actual output of new energy sources in real time, and calculates the deviation Δ from the day-ahead forecast value. L ( t The response deviation rate is assessed based on the discrepancy between actual and predicted values. E ≈−0.05, through the parameter correction module (take) r =0.1) Dynamically reduce the cost adjustment coefficient B This feedback is sent to the cost adjustment module, forming a closed-loop adaptive adjustment mechanism. This provides adaptive capability for subsequent intraday and real-time scheduling cycles, improving the system's fault tolerance to prediction deviations.

[0224] ⑤ Output results:

[0225]

[0226] The results demonstrate that the baseline plan established in this phase not only performs excellently under normal conditions but also lays a solid foundation for robustness. The total cost of the V2G-VPP system is RMB 18,240, establishing a robust economic benchmark for subsequent intraday and real-time dispatch, ensuring that cost fluctuations are strictly controlled within ≤3% under extreme scenarios. Regarding renewable energy consumption, although the actual photovoltaic consumption (4300 kWh) and wind power consumption (2760 kWh) differ from the predicted values ​​by 14% and 8% respectively, this reflects the forward-looking and conservative nature of the robust optimization strategy—it pre-considers the possibility of fluctuations and avoids the actual operational risks that overly aggressive plans might lead to. Simultaneously, the CO2 emission reduction of 6,080 kg and the power balance deviation of ≤5% jointly verify that the baseline plan achieves a good balance between environmental protection and system safety, providing a reliable initial strategy for achieving multi-timescale collaborative optimization.

[0227] Intraday adjustments (15-minute rolling optimization)

[0228] This stage is crucial for addressing real-time operational deviations.

[0229] ① Real-time data correction: When the actual photovoltaic output (4400kWh) is 12% lower than the forecast and the load demand (21600kWh) is 8% higher than the forecast, the system triggers a 15-minute rolling optimization. The core of the optimization lies in the rapid solution and decision adjustment of the MILP model: strategically increasing the charging amount during off-peak hours by 15% (utilizing robust buffer capacity) to utilize low-cost electricity and fill the gap in renewable energy output; at the same time, appropriately reducing the discharge compensation amount during peak hours by 10% to maintain power balance ≤5%.

[0230] ② Robust Optimization Solution: A robust MILP model and C&CG algorithm are used to optimize EV charging and discharging power and distributed unit output, generating rolling dispatch instructions. Extreme scenario weights are fully considered during the solution process to ensure that the dispatch scheme meets economic and safety requirements even within a ±15% fluctuation range in photovoltaic output.

[0231] ③ User response deviation and closed-loop feedback: After executing the MILP scheduling command, the system monitors the deviation between the actual average discharge power (14.2kW) and the target value (15.5kW) and calculates the response deviation rate. E The system adjusts the cost adjustment coefficient by approximately -0.084 through a closed-loop feedback mechanism. B By appropriately reducing the compensation for EV users through a demand response incentive system (robust compensation effect), this closed loop of "response-evaluation-compensation-re-optimization" effectively ensures user participation and system adjustment accuracy.

[0232] ④ Output results:

[0233]

[0234] The results demonstrate that this phase is crucial for transforming the robust plan into resilient execution. Faced with unpredictable real-time deviations, rolling optimization successfully stabilized the total system cost at RMB 18,520, an increase of only 1.5% compared to the day-ahead baseline. This small increase is invaluable, indicating that the system did not incur high additional costs due to uncertainty, but rather effectively absorbed disturbances through flexible adjustments. Although the total revenue for EV users temporarily decreased to RMB 1,657 due to response deviations, the system immediately identified and initiated compensation strategies through a closed-loop feedback mechanism, providing a mechanism to ensure long-term user participation. The most significant achievement is the leap in system dynamic performance: the recovery time was drastically reduced from the expected 150 seconds to 48 seconds, fully validating the decisive role of intraday scheduling (15-minute-level optimization) in improving system resilience and rapid recovery capabilities.

[0235] Real-time execution (5-minute model predictive control)

[0236] This phase aims to cope with sudden disturbances on the order of minutes or even seconds, testing the system's instantaneous response capability.

[0237] ① Disturbance response test: In the extreme test simulating a sudden drop of 20% in photovoltaic output (from 1500kW to 1200kW), the MPC controller played a key role: dynamically adjusting the EV discharge power within seconds (compensating for 180kW, accounting for 90% of the missing output), and supplementing it with grid power purchase (120kWh) during the valley period, quickly smoothing out the power shortage.

[0238] ② Robust optimization solution: Under the severe disturbance of a sudden 20% drop in photovoltaic output, the key operating parameters of the system remained stable within the safe range: the maximum load deviation rate was 4.2% (meeting the target of ≤5%), and the recovery time was 52 seconds (including communication delay), fully meeting the preset targets and confirming that the system still maintains power balance and frequency stability under extreme scenarios.

[0239] ③ Real-time compensation for user revenue: For user revenue deviation caused by disturbance, the system calculates the dynamic compensation amount based on the real-time electricity price (0.3 yuan / kWh) and the response deviation amount, and provides dynamic electricity price subsidies to affected vehicles. Ultimately, the total user revenue is restored from the low point after the disturbance to 1760 yuan, and the EV user revenue volatility is controlled within 12%, ensuring both EV user experience and system reliability.

[0240] ④ Output results:

[0241]

[0242] The output results demonstrate the ultimate control precision of the method during the real-time scheduling execution phase. When dealing with a sudden 20% drop in power, the MPC controller exhibited a second-level response speed and high-precision control capabilities, suppressing the power balance deviation to an extremely low level of 1.8%. The total system cost was ultimately RMB 18,610, meaning that the cumulative cost fluctuation from day-ahead planning to real-time execution was only 2.0%, significantly enhancing the certainty and predictability of the system's economic operation. Simultaneously, through a precise real-time compensation mechanism, EV user revenue was quickly pulled back from the low point of the disturbance to RMB 1,760, ensuring fairness and incentives for users. The power balance deviation was further optimized to a high-precision range of 1.8%, fully validating the powerful capability of the multi-timescale collaborative scheduling method in achieving "second-level response and minute-level recovery."

[0243] Final indicator comparison:

[0244] The final test metrics for the above "Daily Scheduling (Robust Optimized Baseline Plan)", "Intraday Adjustment (15-Minute Rolling Optimization)", and "Real-Time Execution (5-Minute Model Predictive Control)" are summarized to obtain the final test metrics for typical cases.

[0245]

[0246] Through deep integration of closed-loop feedback and robustness optimization, this test successfully and completely verified the superior performance of the multi-timescale closed-loop feedback collaborative scheduling method of the V2G-VPP system in a complex scenario involving hundreds of EV users. It achieved collaborative optimization of multiple objectives, finding the optimal balance between traditionally conflicting goals such as economic cost, user benefits, environmental benefits, and system safety. This provides a reliable technical path and solution for building a highly robust, flexible, and efficient V2G-VPP system. Specific test results and performance indicators include:

[0247] ① Dual improvement in economy and robustness: The total system cost was significantly reduced from RMB 20,100 to RMB 18,610 (a decrease of 7.4%), directly reflecting the economic benefits of the optimization strategy. Simultaneously, the standard deviation of cost fluctuation in extreme scenarios decreased from 12.7% to 8.3%, a reduction of 34.6%. This verifies the synergistic enhancement of the method in terms of economy and robustness.

[0248] ② Reasonable user-side incentives: The total revenue for EV users increased significantly from RMB 1,200 to RMB 1,760 (an increase of 46.7%), which is the core driving force for attracting users to continue participating. At the same time, the volatility of returns (standard deviation) was successfully optimized from 15% to within 8%, which means that users have obtained higher and more stable return expectations, which is crucial for the promotion of the V2G business model.

[0249] ③ Outstanding environmental benefits: Carbon dioxide emission reduction increased by 23.3%, while the prediction error narrowed from 10% to 3%. This indicates that while increasing the consumption of green energy, the system has greatly enhanced the measurability and reportability of environmental contributions, providing accurate evidence for obtaining support from environmental policies.

[0250] ④ The system's safety and resilience have been significantly improved: the power balance deviation has been optimized from 8.5% to 1.8% (a reduction of 66.4%), and the system recovery time has been shortened from 192 seconds to 52 seconds (an improvement of 73%), indicating that the power grid's safe operation level and fault recovery capability have been qualitatively improved.

[0251] Example 4

[0252] The electronic device of this invention includes a central processing unit (CPU), which can perform various appropriate actions and processes according to computer program instructions stored in read-only memory (ROM) or loaded from a storage unit into random access memory (RAM). The RAM may also store various programs and data required for device operation. The CPU, ROM, and RAM are interconnected via a bus. Input / output (I / O) interfaces are also connected to the bus.

[0253] Multiple components in the device are connected to the I / O interface, including: input units such as keyboards and mice; output units such as various types of displays and speakers; storage units such as disks and optical discs; and communication units such as network interface cards (NICs), modems, and wireless transceivers. The communication unit allows the device to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0254] The processing unit executes the various methods and processes described above. For example, in some embodiments, the V2G-VPP multi-timescale closed-loop feedback cooperative scheduling method can be implemented as a computer software program tangibly contained in a machine-readable medium, such as a storage unit. In some embodiments, part or all of the computer program can be loaded and / or installed on the device via ROM and / or a communication unit. When the computer program is loaded into RAM and executed by the CPU, one or more steps of the methods described above can be performed. Alternatively, in other embodiments, the CPU can be configured to execute the V2G-VPP multi-timescale closed-loop feedback cooperative scheduling method by any other suitable means (e.g., by means of firmware).

[0255] The functions described above in this document can be performed, at least in part, by one or more hardware logic components. For example, exemplary types of hardware logic components that can be used, without limitation, include: Field Programmable Gate Arrays (FPGAs), Application-Specific Integrated Circuits (ASICs), Application Standard Products (ASSPs), System-on-Chip (SoCs), Complex Programmable Logic Devices (CPLDs), and so on.

[0256] The program code used to implement the methods of the present invention can be written in any combination of one or more programming languages. This program code can be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing device, such that when executed by the processor or controller, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code can be executed entirely on the machine, partially on the machine, as a standalone software package partially on the machine and partially on a remote machine, or entirely on a remote machine or server.

[0257] In the context of this invention, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. Machine-readable media can include, but are not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0258] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and these modifications or substitutions should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A robust V2G-VPP multi-timescale closed-loop feedback collaborative scheduling method, characterized in that, The method includes the following steps: Real-time acquisition of multi-source data from the V2G-VPP system, followed by data cleaning and feature extraction processing; A three-level scheduling architecture—day-today, intraday, and real-time—is constructed. Based on processed data, a multi-dimensional uncertainty modeling method is used to quantify the uncertainty of key parameters in the V2G-VPP system. In day-to-day robust scheduling, a day-to-day scheduling plan is formulated with the goal of minimizing total cost. Intraday robust scheduling is performed based on the day-to-day scheduling plan, with hourly dynamic adjustments achieved through rolling optimization and corrections based on SOC status and charging / discharging costs to obtain an intraday revised scheduling plan. Real-time scheduling is performed based on the intraday revised scheduling plan, using a model predictive control algorithm for minute-level tracking and response, generating real-time scheduling instructions, and generating feedback signals based on execution results to adjust demand response stimuli, forming a closed-loop feedback. Based on the execution results, feedback signals are generated to adjust the demand response incentives, forming a closed-loop feedback, specifically: Based on the target value of load adjustment in the real-time dispatch command Calculate the response deviation rate: ， in, This is the actual load adjustment amount. For response deviation rate; Calculate the cost adjustment factor based on the response deviation rate: ， in, For the next cycle t Cost adjustment factor required at time +1; This is the cost adjustment factor for this period; For the adjustment amplitude factor, 0 < <1, used to control the magnitude and stability of the correction; Adjust the demand response incentives based on the cost adjustment factor, and update the objective function value in the day-ahead scheduling: ， in, Indicates the first i EVs in t The cost of charging at any time, Indicates the first j EVs in t The discharge benefit at any given moment; Indicates the first i EVs in the next cycle t The charging cost at +1 hour; Indicates the first j EVs in the next cycle t The discharge gain at time +1.

2. The robust V2G-VPP multi-timescale closed-loop feedback cooperative scheduling method according to claim 1, characterized in that, The multi-dimensional uncertainty modeling method is used to quantitatively describe the uncertainty of key parameters in the V2G-VPP system. Establish a joint probability distribution model for distributed generation and load demand: ， in, N ( μ , σ The output of new energy sources follows a normal distribution. LN ( μ , σ ) represents the log-normal distribution of load demand. μ and σ These represent the mean and standard deviation of historical data, respectively. The subscript "gen" indicates distributed generation, and the subscript "load" indicates the load. t Indicates time; Indicates the scheduling time; A set of typical scenes generated based on Monte Carlo simulation. Each scenario includes a sequence of new energy power outputs. and load demand sequence The scene weights are determined by normalizing the probability density function.

3. The robust V2G-VPP multi-timescale closed-loop feedback cooperative scheduling method according to claim 1, characterized in that, The objective function for the day-ahead robust scheduling is to minimize the sum of total generation cost, EV user charging cost, and discharge revenue, expressed as: ， Among them, decision variables , , , These represent EV charging power, EV discharging power, and distributed generation power within the V2G-VPP system, respectively; uncertainty parameters. , Indicates load demand; D It is a set of uncertainty parameters generated based on Monte Carlo simulation; t Indicates time; express t The unit generation cost of distributed generation at any time; Representing a scene exist t Distributed generation power at any given time; i The EV number indicates the charging status. j The EV number indicating the discharge state. N This indicates the total number of EVs participating in the charging process. M This indicates the total number of EVs participating in the discharge; and They represent the first i EVs in t The charging power and charging cost at any given time. and They represent the first j EVs in t Discharge power and discharge benefit at any given moment; τ Indicates the error weighting coefficient; To determine the day-ahead load forecast deviation, the actual load value of the V2G-VPP system is obtained. and the current forecast load value The difference is obtained by calculation; This represents the prediction error penalty term.

4. The robust V2G-VPP multi-timescale closed-loop feedback cooperative scheduling method according to claim 3, characterized in that, The constraints for the day-ahead robust scheduling include power balance constraints, EV battery state of charge constraints, distributed unit ramping constraints, and charging rate constraints, as shown below: ， ， ， ， ， in, , These represent the lower and upper limits of the EV battery's state of charge, respectively. Indicates the initial state of charge. This indicates the charging efficiency of the EV. This indicates the discharge efficiency of the EV. Indicates the length of a period of uncertainty. This represents the ramp rate of the distributed generator set. For time intervals, Indicates the maximum charging power of the EV. Indicates the EV discharge rate. This indicates the maximum discharge power of the EV.

5. The robust V2G-VPP multi-timescale closed-loop feedback cooperative scheduling method according to claim 1, characterized in that, The objective function for the intraday robust scheduling is as follows: ， in, As decision variables, For uncertain parameters, D It is a set of uncertainty parameters generated based on Monte Carlo simulation; t Indicates time; Indicates the scheduling time; express t The unit generation cost of distributed generation at any time; Representing a scene exist t Distributed generation power at any given time; i The EV number indicates the charging status. j The EV number indicating the discharge state. N This indicates the total number of EVs participating in the charging process. M This indicates the total number of EVs participating in the discharge; and They represent the first i EVs in t The charging power and charging cost at any given time. and They represent the first j EVs in t Discharge power and discharge benefit at any given moment; This is the intraday correction factor; The power deviation between intraday scheduling and the previous day's plan, ,in, Indicates EV in t Discharge power at any given time Indicates EV in t Charging power at any time Representing a scene exist t The time slot relative to the revised load of the day-ahead plan is shown in the following formula: ,in, α ( t )for t Error correction coefficient at time, This is the day-ahead load forecast deviation. This represents the day-ahead forecast load value.

6. The robust V2G-VPP multi-timescale closed-loop feedback cooperative scheduling method according to claim 5, characterized in that, The constraints for the intraday robust scheduling include power balance constraints, EV battery SOC dynamic constraints, distributed unit ramping constraints, and charging rate constraints, as shown below: ， ， ， ， ， in, express t EV battery state of charge at any given time This indicates the charging efficiency of the EV. This indicates the discharge efficiency of the EV. This represents the ramp rate of the distributed generator set. For time intervals, Indicates the maximum charging power of the EV. Indicates the EV discharge rate. This indicates the maximum discharge power of the EV.

7. A robust V2G-VPP multi-timescale closed-loop feedback cooperative scheduling method according to claim 5, characterized in that, The error correction coefficient α ( t )By real-time charging fees Discharge benefits Real-time status of EV battery SOC SOC ( t Collaborative execution is dynamically adjusted through weighted coefficient allocation: ， in, , and These represent the adjustment weighting coefficients for real-time charging costs, discharge benefits, and EV battery SOC state, respectively, satisfying... ; This indicates the maximum charging cost. This represents the maximum discharge benefit. This indicates the upper limit of the EV battery's state of charge.

8. The robust V2G-VPP multi-timescale closed-loop feedback cooperative scheduling method according to claim 1, characterized in that, The objective function for the real-time scheduling is: ， in, As decision variables, For uncertain parameters, D It is a set of uncertainty parameters generated based on Monte Carlo simulation; k Sampling time; express t The future of the time period k Each sampling time; express t The future of the time period k System power deviation at each sampling time , Representing a scene exist t The future of the time period k Distributed generation power at each sampling time, , They represent in t The future of the time period k EV discharge power and charging power at each sampling time. Representing a scene exist t The future of the time period k Each sampling time corresponds to the revised load of the day-ahead plan; This represents a penalty term for changes in the output of distributed units in a V2G-VPP system. Indicates in t The future of the time period k The dynamic weighting factor at each sampling time. , This serves as the baseline value for the dynamic weighting factor. Indicates the benchmark electricity price. For adjustment coefficients, Indicates the future number k Real-time electricity price at each sampling point. This represents the error correction coefficient for intraday scheduling. Representing a scene exist t The future of the time period k Changes in distributed generation output at each sampling time; W track The coefficient of the load tracking deviation term in the objective function.

9. A robust V2G-VPP multi-timescale closed-loop feedback cooperative scheduling method according to claim 8, characterized in that, The constraints of the real-time scheduling include dynamic constraints on EV battery SOC and distributed unit ramping constraints, as shown below: ， ， in, express t The future of the time period k EV battery state of charge at each sampling time, This indicates the charging efficiency of the EV. This indicates the discharge efficiency of the EV. This represents the ramp rate of the distributed generator set. For time intervals.

Images