V2G aggregator bidding decision-making method based on master-slave random game

Abstract

The invention relates to the technical field of electric vehicle and power grid interaction, and particularly discloses a V2G aggregator bidding decision-making method based on a master-slave random game. The relationship between the V2G aggregator and the electric vehicle user cluster is modeled as a master-slave random game, the aggregator is used as a master party to formulate an electricity price strategy, and the user cluster is used as a slave party to respond with a random charging and discharging behavior. A random response model of the slave party is constructed, the uncertainty of the random response model is quantified, and meanwhile a user participation degree evaluation index which is dynamically updated in real time is constructed; establishing a cooperative constraint distribution robust optimization CC-DRO decision model of the master party; and an improved COA-ESN hybrid intelligent algorithm is adopted to solve the CC-DRO model, a global optimal bidding strategy is output, the bidding strategy is executed, and market electric charge settlement is completed. Compared with the prior art, through fusion of a master-slave random game framework and multiple technologies, breakthrough of a V2G bidding strategy in robustness and economy is realized, and real-time performance of decision making is ensured through an efficient hybrid algorithm.

Description

Technical Field

[0001] This invention belongs to the field of electric vehicle-grid interaction and electricity market technology, specifically involving a V2G aggregator bidding decision-making method based on master-slave random game. Background Technology

[0002] With the large-scale development of electric vehicles, vehicle-to-grid (V2G) technology has become a key means of absorbing renewable energy and providing grid ancillary services. V2G aggregators, by integrating dispersed electric vehicle resources and representing user groups in electricity market transactions, represent the core model for realizing the commercial value of V2G. However, existing research on aggregator bidding strategies faces many serious challenges, hindering its large-scale commercial application.

[0003] First, the behavior of electric vehicle users is highly random and uncertain, with their charging and discharging decisions heavily influenced by multiple dynamic factors such as real-time electricity prices, personal travel needs, psychological preferences, and battery status. Traditional deterministic optimization models struggle to accurately characterize this complex randomness, while existing stochastic programming methods heavily rely on precisely known assumptions about the probability distribution of uncertainty, which are often difficult to satisfy in practice. This leads to significant deviation risks in the actual implementation of bidding strategies, or overly conservative and uneconomical approaches. Second, aggregator decision-making is a complex optimization problem across multiple time scales, requiring simultaneous satisfaction of grid-side operational safety constraints and user-side constraints such as dispatchable capacity and battery life. Existing methods struggle to effectively coordinate these constraints under uncertain environments.

[0004] In terms of modeling and solving, existing research has failed to deeply integrate the uncertainty of user behavior into the decision-making framework in a rigorous mathematical form. Although some studies have attempted to apply game theory or intelligent algorithms, they often separate user response prediction from aggregator strategy optimization, lacking a unified modeling framework that can simultaneously handle distributional ambiguity and achieve synergy among multiple parties.

[0005] Existing algorithms such as standard genetic algorithms and particle swarm optimization often suffer from slow convergence and are prone to getting trapped in local optima when faced with such high-dimensional, non-convex optimization problems. They cannot meet the stringent requirements of real-time decision-making and reliability for day-ahead bidding in the electricity market.

[0006] At the incentive mechanism level, existing solutions mostly rely on a single cash reward model, failing to effectively link users' green discharge behavior with broader environmental value such as carbon emission reduction. Revenue distribution is often simply based on the proportion of discharge volume, lacking dynamic evaluation of users' comprehensive contributions such as real-time response quality and long-term reputation. This results in insufficient incentive compatibility, making it difficult to stimulate users' long-term and stable participation and failing to build a sustainable V2G ecosystem loop. Summary of the Invention

[0007] Purpose of the invention: To address the aforementioned technical problems, this invention provides a V2G aggregator bidding decision-making method based on master-slave random game theory, which can generate an optimal bidding strategy for aggregators that maximizes economic benefits and has controllable operational risks under uncertain environments; thus effectively solving the above problems.

[0008] Technical Solution: This invention discloses a V2G aggregator bidding decision-making method based on master-slave random game theory. The relationship between V2G aggregators and electric vehicle user clusters is modeled as a master-slave random game, where the aggregator, as the master, formulates the electricity pricing strategy, and the user cluster, as the slave, responds with random charging and discharging behaviors. The method includes the following steps:

[0009] Step 1: Collect and preprocess data from V2G aggregators, including the access time, disconnection time, initial SOC, expected SOC, battery capacity, and grid-side data for each vehicle.

[0010] Step 2: Construct a random response model from the user, use Echo State Network (ESN) to predict the prediction error sequence of the random response model, construct a data-driven Wasserstein fuzzy set to quantify uncertainty, and simultaneously construct a real-time dynamically updated user engagement evaluation index.

[0011] Step 3: Establish the Collaborative Constraint Decentralized Optimization Decision Model CC-DRO for the main party. The CC-DRO model introduces a "collaborative constraint mechanism", including dynamic schedulable capacity constraints and hierarchical scheduling constraints based on user priority determined by user participation evaluation indicators.

[0012] Step 4: Solve the CC-DRO model using the COA-ESN hybrid intelligent algorithm to output the globally optimal bidding strategy. Complete the market electricity bill settlement according to the optimal bidding strategy. Determine the slave behavior according to the globally optimal bidding strategy. The uncertainty of the slave behavior is fed back to the CC-DRO model through the Echo State Network (ESN) and Wasserstein fuzzy set.

[0013] Furthermore, step 2 involves constructing a data-driven Wasserstein fuzzy set quantification of uncertainty, specifically including:

[0014] ESN-based prediction error sequence Construct Wasserstein fuzzy sets, defined as:

[0015] ;

[0016] in, This represents the prediction error at time t; F represents the actual observed charging and discharging power of the electric vehicle cluster at time t; F represents the constructed Wasserstein fuzzy set; Q represents any probability distribution belonging to this set; P(E) represents the set of all probability distributions on the uncertainty support set E; Represents the Wasserstein distance; Indicates based on N historical error samples The obtained empirical distribution; The radius of the fuzzy set is a hyperparameter that is greater than zero.

[0017] Furthermore, in step 2, the user engagement evaluation index R-UPI is dynamically updated using a recursive formula:

[0018] ;

[0019] in, For short-cycle A comprehensive score calculated from real-time user behavior data; This represents the user's real-time behavior coefficient.

[0020] Furthermore, in step 3, the collaborative constraint split-bar optimization decision model CC-DRO of the principal aims to minimize the principal's total cost, wherein:

[0021] In the first phase, the main party submits the bidding curve to the day-ahead market before the uncertainty of the user cluster response is revealed;

[0022] In the second stage, during the real-time runtime, once the random charging and discharging response behavior of the user cluster is observed and determined, the principal solves the second-stage value function. To formulate real-time scheduling strategies to balance the power deviation between the day-ahead plan and the actual situation, and to minimize the penalty costs incurred in the real-time balancing market;

[0023] The objective function is to minimize the sum of the first-stage cost and the worst-case expected second-stage cost, specifically:

[0024] ;

[0025] in, Indicates the decision variables for the first stage; This indicates the expected cost of the first phase in the current market. Indicates the uncertainty parameter; This represents the value function for the second stage; Fuzzy sets constructed based on Wasserstein distance.

[0026] Furthermore, the second-stage value function The constraints introduce a "cooperative constraint mechanism," including dynamic schedulable capacity constraints and user priority hierarchical scheduling constraints.

[0027] Dynamic schedulable capacity constraint: Total real-time scheduling power of the cluster The schedulable capacity limit must not be exceeded based on dynamic assessment of user engagement and status. ,Right now:

[0028] ;

[0029] ;

[0030] in, This represents the total real-time scheduling power of the cluster. The upper limit of schedulable capacity for dynamic state assessment; This represents the real-time power scheduling command for user i at time t; Let represent the set of electric vehicle users who are online and available for scheduling at time t; This represents the absolute value of the maximum allowable charging and discharging power for user i. This represents the upper limit of the battery state of charge for user i; This represents the actual state of charge of user i at time t; Indicates charge / discharge efficiency;

[0031] User priority hierarchical scheduling constraints: Users are prioritized based on the real-time user engagement index (R-UPI), ensuring that the scheduling needs of high-priority users are met first; the priority weight of user i at time t is defined. for:

[0032] ;

[0033] in, , , Indicates the weighting coefficient; This represents the state of charge that user i expects to have when disconnected from the grid;

[0034] This indicates the planned offline time for user i; This represents the historical command response rate of user i; during scheduling, the following conditions must be met: .

[0035] Furthermore, it also includes multi-timescale power smoothing constraints, where the total power change between adjacent scheduling periods must meet the ramp rate constraint.

[0036] Furthermore, in step 4, the COA-ESN hybrid intelligent algorithm is used to solve the CC-DRO model. The COA-ESN hybrid intelligent algorithm integrates an elite retention strategy and a weight adaptation mechanism. By retaining elite individuals in the population evolution and dynamically adjusting the search weights based on the historical performance of individuals to guide the optimization direction, it obtains the optimal game strategy of the host, i.e., the bidding curve; specifically:

[0037] Step 4.1 Generate M initial candidate bidding curves using Logistic chaotic mapping and initialize dynamic weights for each individual;

[0038] Step 4.2 For each individual in the current population, perform the following operations:

[0039] Weighted ESN prediction: Individuals and their weights are input into a pre-trained Echo State Network (ESN) to predict the response and uncertainty of user clusters.

[0040] Constraint violation calculation: For each candidate policy, predict the user response through the Echo State Network (ESN) and solve the second-stage value function. Then, calculate the total constraint violation degree. :

[0041]

[0042] in, This represents the maximum permissible rate of change in the total power of the cluster between adjacent time periods, i.e., the upper limit of the ramp rate; T represents the total number of time periods in the scheduling cycle; This represents the maximum total power of the electric vehicle cluster during the scheduling period t.

[0043] Fitness evaluation: Based on constraint violation degree, solve the two-stage sub-Bruker optimization model and calculate the total cost of an individual as its fitness value;

[0044] ESN-based prediction Based on the error distribution, solve the two-stage DRO model and calculate the total cost:

[0045] ;

[0046] in, This represents the fitness value of the i-th individual in the k-th generation of the population; The total cost is calculated for the objective function; This is a relatively large penalty factor;

[0047] Weight Update: Weights are dynamically adjusted based on the historical improvement of an individual's fitness; high-performing individuals receive increased weights, while those that deteriorate receive decreased weights. Weights are also dynamically updated by comparing an individual's current fitness with that of the previous generation.

[0048] ;

[0049] in, This represents the weight decay factor, 0 < <1; Indicates the weighted reward factor. >1;

[0050] Step 4.3 Perform selection operations based on weights, retain contemporary elite individuals to directly enter the next generation, and then perform chaotic crossover and mutation on the remaining individuals to generate a new population;

[0051] Step 4.4 Repeat the iteration until the termination condition is met, the number of iterations reaches the preset maximum value, and the individual with the best fitness in all generations is output. This refers to the globally optimal bidding strategy.

[0052] Furthermore, after implementing the bidding strategy and completing the market electricity bill settlement, the aggregator also implements a "cash-carbon credits" dual-factor incentive compatible allocation mechanism, specifically:

[0053] On the one hand, the actual settlement amount is linked to the user's final R-UPI value and priority to achieve incentive compatibility;

[0054] On the other hand, carbon credit rewards are calculated based on the user's total effective discharge and number of participations during the scheduling period, and the credits are deposited into the user's virtual account.

[0055] Cash incentives and carbon credit rewards together constitute the user's final benefit, and carbon credits can be redeemed for physical goods through the merchant network, completing the entire V2G bidding decision, transaction and dual incentive process.

[0056] Beneficial effects:

[0057] (1) This invention constructs a Real-Time User Participation Index (R-UPI) to dynamically assess user participation willingness and dispatchability, and integrates it into the collaborative constraint mechanism in the CC-DRO model to achieve real-time linkage between user behavior, electricity pricing strategy, and grid demand, significantly improving dispatch accuracy and user participation willingness. Conventional DRO models use simple moment uncertainty sets, while this invention creatively utilizes the prediction error sequence of Echo State Network (ESN) to construct a data-driven Wasserstein fuzzy set. This is a key improvement that directly addresses the pain point that the uncertainty of user behavior in V2G scenarios is difficult to describe using traditional probability distributions.

[0058] (2) This invention introduces a collaborative constraint mechanism on the basis of traditional DRO, constructs a collaborative constraint split-bar optimization (CC-DRO) model to enhance system resilience, and combines R-UPI to dynamically adjust schedulable capacity and user priority, so that the bidding strategy not only pursues "worst-case immunity" when dealing with uncertainty, but also achieves "multi-party collaborative adaptation", thereby improving the overall operational resilience and economy of the system. Conventional DRO only handles uncertainty through objective function, while this invention systematically introduces a "collaborative constraint mechanism" in the second stage of constraints, including dynamic schedulable capacity constraints and user priority hierarchical scheduling constraints. This DRO model does not exist in isolation, but is the decision core of the master (aggregator) in a master-slave random game. Its decision result (electricity price) will affect the behavior of the slave, and the uncertainty of the slave's behavior is fed back to the DRO model through ESN and Wasserstein set. This closed-loop, dynamic game modeling idea elevates DRO from a passive optimization tool to an active game strategy generator.

[0059] (3) This invention improves the COA-ESN hybrid intelligent algorithm by introducing a weight adaptive mechanism and an elite retention strategy into the algorithm. The search direction is dynamically adjusted according to the historical performance of individuals to avoid population degradation, significantly improve the convergence speed and global optimization ability, and ensure the efficient solution of complex master-slave game models in real-time decision-making.

[0060] (4) Based on R-UPI and discharge contribution, this invention constructs an incentive system that combines cash income with redeemable physical carbon credits, which not only improves users’ economic returns, but also enhances their sense of participation and green behavior recognition through the carbon credit exchange mechanism, forming a sustainable V2G ecological closed loop. Attached Figure Description

[0061] Figure 1 This is an overall flowchart of the V2G aggregator bidding decision-making method of the present invention;

[0062] Figure 2 This is a schematic diagram of the model framework constructed in steps S2 and S3 of this embodiment of the invention;

[0063] Figure 3 This is a flowchart of the improved COA-ESN hybrid algorithm used in this invention.

[0064] Figure 4 This is a comparison chart of the convergence of the improved COA-ESN hybrid algorithm of this invention;

[0065] Figure 5 This is the Wasserstein fuzzy set distribution boundary map of the present invention;

[0066] Figure 6 This is a comparison chart of the robustness of the DRO method of this invention with other methods;

[0067] Figure 7 This is a comparative analysis of the optimal bidding strategies of this invention. Detailed Implementation

[0068] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments. The described embodiments are merely some embodiments of the present invention, and not all embodiments. Various modifications and improvements made to the technical solutions of the present invention by those skilled in the art without departing from the design concept of the present invention should fall within the protection scope of the present invention.

[0069] Example 1:

[0070] This invention discloses a V2G aggregator bidding decision-making method based on master-slave random game, comprising the following steps:

[0071] Step 1: Data collection and preprocessing for V2G aggregator participants. Collect historical data from the past year for the aggregator's electric vehicles, including: access time, disconnection time, initial SOC, expected SOC, and battery capacity for each vehicle. Simultaneously, collect grid-side data for the same period, including: time-of-use pricing, spot market clearing pricing, regional total load data, and temperature and weather type data. Clean all this data, removing missing and outlier values, and normalize it using the min-max method to form a standardized dataset. This embodiment demonstrates the application of the method of the present invention in a V2G aggregator scenario involving 1000 electric vehicles. The overall process can be found in [link to documentation]. Figure 1 .

[0072] Step 2: Modeling the response characteristics and uncertainties of electric vehicle clusters.

[0073] The data processed in step 1 is input into ESN for training, and its dynamic system is described by the following equation:

[0074] The logical relationship between this step and the subsequent optimization model—that is, the response model provides key inputs to the optimization model—can be found in [reference needed]. Figure 2 The upper part.

[0075] ESN state update equation:

[0076]

[0077] ESN output equation:

[0078]

[0079] in: This represents the state vector of the ESN reservoir at time t, and is the internal memory unit of the ESN; This represents the input energy at time t, including the electricity price. Timestamp t, weather conditions wait; This represents the predicted value of the cluster charging and discharging power at time t, output by the ESN.

[0080] Step 2.2: Based on the prediction error, construct a data-driven Wasserstein fuzzy set to describe the uncertainty of the slave response, such as... Figure 5 As shown:

[0081] ESN-based prediction error sequence Construct Wasserstein fuzzy sets, whose mathematical definition is:

[0082]

[0083] in, This represents the prediction error at time t; Let F represent the actual observed charging and discharging power of the electric vehicle cluster at time t; F represents the constructed Wasserstein fuzzy set; Q represents any probability distribution belonging to this set; and P(E) represents the set of all probability distributions on the uncertainty support set E. Represents the Wasserstein distance; Indicates based on N historical error samples The obtained empirical distribution; The radius of the fuzzy set is a hyperparameter that is greater than zero.

[0084] Step 2.3: Build and update the Real-Time User Engagement Index (R-UPI). The R-UPI is dynamically updated using a recursive formula:

[0085]

[0086] in, For short-cycle A comprehensive score calculated from real-time user behavior data; This represents the user's real-time behavior coefficient.

[0087] Step 3: Establish a two-stage sub-Browser optimization decision model for the main party (i.e., the V2G aggregator).

[0088] The aggregator's decision-making objectives are modeled as the CC-DRO decision-making model. The framework, decision-making stages, and objectives of this model can be found in [reference needed]. Figure 2 .

[0089] In the first phase, the main party submits the bidding curve to the day-ahead market before the uncertainty of the user cluster response is revealed;

[0090] In the second stage, during the real-time runtime, once the random charging and discharging response behavior of the user cluster is observed and determined, the principal solves the second-stage value function. To formulate real-time scheduling strategies to balance the power deviation between the day-ahead plan and the actual situation, and to minimize the penalty costs incurred in the real-time balancing market;

[0091] The model's objective function is to minimize the sum of the first-stage cost and the worst-case expected second-stage cost. The specific process includes:

[0092] Step 3.1: The mathematical expression of the established CC-DRO model is as follows:

[0093]

[0094] in, Indicates a first-stage decision variable; This indicates the expected cost in the current market at the first stage; Indicates the uncertainty parameter; This represents the two-stage value function; Fuzzy sets constructed based on Wasserstein distance.

[0095] Step 3.2: Two-stage value function The constraints introduce a collaborative constraint mechanism, which specifically includes:

[0096] 3.21 Dynamically schedulable capacity constraint: Total real-time scheduling power of the cluster The schedulable capacity limit must not be exceeded based on dynamic assessment of user engagement and status. ,Right now:

[0097]

[0098]

[0099] in, This represents the real-time power scheduling command for user i at time t; Let represent the set of electric vehicle users who are online and available for scheduling at time t; This represents the absolute value of the maximum allowable charging and discharging power for user i. This represents the upper limit of the battery state of charge for user i; This represents the actual state of charge of user i at time t; This indicates the charge / discharge efficiency.

[0100] 3.22 User Priority Hierarchical Scheduling Constraints: Users are prioritized according to R-UPI, ensuring that the scheduling needs of high-priority users are met first. Define the priority weight of user i at time t. for:

[0101]

[0102] in, , , Indicates the weighting coefficient; This represents the state of charge that user i expects to have when disconnected from the grid; This indicates the planned offline time for user i; This represents the historical command response rate of user i; during scheduling, the following conditions must be met: .

[0103] 3.23 Multi-timescale power smoothing constraint: To ensure grid stability, the total power change between adjacent dispatch periods must meet the ramp rate constraint.

[0104]

[0105] in, This indicates the maximum power ramp rate allowed by the system.

[0106] Step 4: Model solving and bidding strategy generation.

[0107] The improved COA-ESN algorithm is used to solve the CC-DRO model, and the solution framework is shown below. Figure 3 The execution flow of this hybrid algorithm is as follows:

[0108] Step 4.1 Generate an initial population using a Logistic chaotic mapping and initialize dynamic weights for each individual;

[0109] Given a population size M, generate M initial candidate bidding curves using a logistic mapping. Initialize weights for each individual :

[0110]

[0111] Where M represents the population size; Let represent the bidding curve of the i-th individual in the k-th generation population; This represents the dynamic weight of the i-th individual in the k-th generation of the population; The parameters represent the Logistic mapping.

[0112] Step 4.2 Iterative Optimization: For each individual in the current population, perform the following operations:

[0113] 1) Weighted ESN Prediction: Individuals and their weights are input into a pre-trained ESN model to predict the response and uncertainty of a user cluster.

[0114] Candidate strategies With its current weight The concatenation is used as an enhanced input to the ESN for forward computation:

[0115]

[0116] 2) Constraint violation calculation: For each candidate policy Predicting user responses and the second-stage value function through ESN Then, calculate its total constraint violation degree. :

[0117]

[0118] in, This represents the maximum allowable power ramp rate of the system; T represents the total number of time periods in the scheduling cycle; This represents the maximum total power of the electric vehicle cluster during the scheduling period t.

[0119] 3) Fitness Assessment: Based on constraint violation, solve the two-stage biblical optimization model and calculate the total cost of each individual as its fitness value. (Based on ESN prediction) Based on the error distribution, solve the two-stage DRO model and calculate the total cost:

[0120]

[0121] in, This represents the fitness value of the i-th individual in the k-th generation of the population; The total cost is calculated for the objective function; This is a relatively large penalty factor.

[0122] 4) Weight Update: Based on the historical improvement of an individual's fitness, its weight is dynamically adjusted, with increased weight for high-performing individuals and decreased weight for those that deteriorate. The individual's current fitness is compared to the previous generation, and its weight is dynamically updated accordingly.

[0123]

[0124] in, This represents the weight decay factor, 0 < <1; Indicates the weighted reward factor. >1.

[0125] Step 4.3 Perform selection operations based on weights, retaining the current elite individuals who directly enter the next generation, and then performing chaotic crossover and mutation on the remaining individuals to generate a new population:

[0126] 1) Elite preservation: Select the E individuals with the best fitness from the current population and directly replicate them to the next generation.

[0127] 2) Weighted selection: Using the roulette wheel selection method, the probability of individual i being selected as the parent is... Select ME parent individuals.

[0128] 3) Chaotic crossover and mutation: Perform chaotic crossover and mutation operations on the selected parent individuals to generate ME child individuals.

[0129] 4) Merging populations: Merging E elite individuals with ME offspring individuals to form a new generation population. .

[0130] Step 4.4 Repeat the iteration until the termination condition is met, and the number of iterations reaches the preset maximum value. Output the individual with the best fitness across all generations. This refers to the globally optimal bidding strategy.

[0131] like Figure 4 As shown, the improved COA-ESN hybrid algorithm used in this invention exhibits faster convergence speed and better global optimization capability compared to traditional genetic algorithms and particle swarm optimization algorithms. This algorithm can stably converge to the optimal solution within 500 iterations, verifying the superiority of hybrid intelligent algorithms in solving complex min-max problems.

[0132] Step 5: Execute the bidding strategy and complete the market electricity bill settlement.

[0133] The optimal bidding curve generated in step 4 Submitted to the power trading center to participate in the centralized clearing of the day-ahead power market, aggregators construct a dual-factor incentive compatible allocation mechanism of "cash-carbon credits".

[0134] On the one hand, the actual settlement amount is linked to the user's final R-UPI value and priority to achieve incentive compatibility.

[0135] On the other hand, carbon credit rewards are calculated based on the user's total effective discharge and number of participations during the scheduling period, and these credits are deposited into the user's virtual account. Cash incentives and carbon credit rewards together constitute the user's final benefit, and carbon credits are redeemed for physical goods through the merchant network, thus completing the entire V2G bidding decision-making, trading, and dual incentive process.

[0136] The specific solution process includes:

[0137] Step 5.1 The user's electricity revenue is:

[0138] Total revenue that aggregators receive from the market:

[0139]

[0140] in, To aggregate the total electricity revenue obtained from the market; The discharge power during time period t; The discharge grid connection price is for time period t. The charging power during time period t; Let t be the electricity price for grid charging during time period t.

[0141] User electricity revenue:

[0142]

[0143] in, This represents the total cash benefit actually received by user i. The average real-time user engagement index for user i over the entire scheduling period; Let be the absolute value of the total effective discharge of user i over all time periods t.

[0144] Step 5.2 The battery degradation cost is:

[0145]

[0146] in, Costs related to battery degradation; This represents the battery degradation coefficient.

[0147] Step 5.3 Government subsidy amount:

[0148]

[0149] in, This refers to the amount of government subsidies. This is the subsidy coefficient for the discharge amount.

[0150] Step 5.4 introduces a compound benefit model for green carbon credits, with the carbon credit benefit function as follows:

[0151]

[0152] Where G represents the total amount of green carbon credits obtained by the user; The carbon integral reward coefficient per unit discharge amount; A fixed carbon credit reward for each participation; Total number of times a user participates in V2G scheduling.

[0153] Final benefits:

[0154]

[0155] in, This is the exchange rate between carbon credits and the monetary value of goods.

[0156] Ultimately, after all V2G discharge transactions are completed, the aggregator settles accounts with the grid based on the actual metered electricity volume and the market clearing price; at the same time, it pays cash revenue to the users who participated in the discharge and issues carbon credits that can be redeemed for physical goods, thus completing the entire V2G bidding decision and trading process.

[0157] Figure 6 The robustness of DRO compared to other methods, with varying uncertainty parameters As uncertainty increases, the expected cost of deterministic optimization methods rises sharply. Under high uncertainty levels, the expected cost of the CC-DRO model method in this invention is reduced by approximately 40%-60% compared to deterministic optimization methods, and its expected cost is consistently lower than that of the SPO method, with an average improvement of approximately 20%-30%. It overcomes the shortcomings of traditional methods in ignoring uncertainty by providing V2G aggregators with an optimal bidding decision-making scheme that combines excellent economic efficiency, strong robustness, and practical operability when facing complex and uncertain environments through a data-driven, multi-layered robust optimization framework.

[0158] Figure 7 The comparative analysis of the optimal bidding strategies shows that the optimal bidding curve of this invention is smoother and more stable than the market-predicted electricity price, and the average bidding price is lower than the market average price, with bidding efficiency improved by 3.9%.

[0159] The above description of the embodiments is only for the purpose of helping to understand the method and core idea of ​​the present invention; at the same time, for those skilled in the art, there will be changes in the specific implementation and application scope based on the idea of ​​the present invention. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A method for V2G aggregator bidding decision based on master-slave stochastic game, characterized in that, Model the relationship between V2G aggregators and clusters of electric vehicle users as a master-slave stochastic game, where the aggregator is the master who sets the electricity price strategy, and the user cluster is the slave who responds with random charging and discharging behavior; including the following steps: Step 1: Collect and preprocess data of V2G aggregator participants, including access time, off-grid time, starting SOC, expected SOC, battery capacity of each vehicle, and grid-side data during the same period; Step 2: Based on the collected data, build a random response model for the slave, use an echo state network (ESN) to predict the prediction error sequence of the random response model, and build a data-driven Wasserstein fuzzy set to quantify uncertainty, while building a real-time dynamically updated user participation evaluation index; Step 3: Establish a collaborative constraint distribution robust optimization decision model (CC-DRO) for the master, which introduces a "collaborative constraint mechanism", including dynamic schedulable capacity constraints and user priority hierarchical scheduling constraints determined by the user participation evaluation index; Step 4: Use a COA-ESN hybrid intelligent algorithm to solve the CC-DRO model, output the globally optimal bidding strategy, complete market electricity settlement according to the optimal bidding strategy, and determine the slave behavior according to the globally optimal bidding strategy. The uncertainty of the slave behavior is fed back to the CC-DRO model through the echo state network (ESN) and the Wasserstein fuzzy set. 2.The V2G aggregator bidding decision method based on master-slave stochastic game according to claim 1, wherein, In step 2, the data-driven Wasserstein fuzzy set is constructed to quantify uncertainty, which includes: Prediction error sequence based on ESN Wasserstein fuzzy sets are constructed and defined as: ； where, denotes the prediction error at time t; denotes the actual observed value of the aggregated power of the EVs at time t; F denotes the constructed Wasserstein fuzzy set; Q denotes an arbitrary probability distribution belonging to the set; P(E) denotes the set of all probability distributions over the uncertainty support set E; denotes the Wasserstein distance; denotes the empirical distribution based on N historical error samples denotes the empirical distribution based on N historical error samples denotes the radius of the fuzzy set, which is a hyperparameter greater than zero. 3.The V2G aggregator bidding decision method based on master-slave stochastic game according to claim 1, wherein, In step 2, the user participation evaluation index R-UPI is dynamically updated through a recursive formula: ； wherein, is a short period a composite score calculated based on short period is a user real-time behavior coefficient. 4.The V2G aggregator bidding decision method based on master-slave stochastic game according to claim 1, wherein, In step 3, the collaborative constraint distribution robust optimization decision model (CC-DRO) for the master aims to minimize the total cost of the master, which includes: The first stage is day-ahead decision-making, where the master submits a bidding curve to the day-ahead market before the user cluster's response uncertainty is revealed; The second stage is to implement the decision, in the real-time running period, when the random charging and discharging response behavior of the user cluster is observed and determined, the host party solves the second stage value function to formulate a real-time scheduling strategy to balance the power deviation between the day-ahead plan and the actual situation, and minimize the penalty cost generated in the real-time balancing market; The objective function is to minimize the sum of the first-stage cost and the expected cost in the worst-case second stage, which is: ； where, denotes the first stage decision variable; denotes the first stage expected cost in the day-ahead market; denotes the uncertainty parameter; denotes the second stage value function; fuzzy sets constructed based on the Wasserstein distance.

5. The V2G aggregator bidding decision method based on master-slave stochastic game according to claim 4, characterized in that, The second stage value function The constraint conditions introduce "coordinated constraint mechanism", including dynamic schedulable capacity constraint, user priority hierarchical scheduling constraint; Dynamic dispatchable capacity constraint: real-time dispatch total power of the cluster Not to exceed the upper limit of dispatchable capacity based on dynamic assessment of user engagement and status , i.e.: ； ； wherein, is the real-time dispatch total power of the cluster; is the state dynamic assessment of the upper limit of the schedulable capacity; represents the real-time dispatch power instruction for user i at time t; represents the set of electric vehicle users in the grid and schedulable at time t; represents the maximum allowed absolute value of the charge and discharge power of user i; represents the upper limit of the state of charge of user i; represents the actual state of charge of user i at time t; represents the charge and discharge efficiency; User priority hierarchical scheduling constraints: Users are prioritized based on the real-time user engagement index (R-UPI), ensuring that the scheduling needs of high-priority users are met first; the priority weight of user i at time t is defined. for: ； wherein , , denotes a weight coefficient; denotes the state of charge at which user i desires to go off-grid; represents the planned off-net time of user i; represents the historical instruction response rate of user i; at the time of scheduling, it needs to meet: .

6. The V2G aggregator bidding decision method based on master-slave stochastic game according to claim 4, characterized in that, It also includes multi-time scale power smoothing constraints, which require the total power change between adjacent scheduling periods to meet the ramp rate constraint.

7. The V2G aggregator bidding decision method based on master-slave stochastic game according to claim 5, characterized in that, In step 4, the COA-ESN hybrid intelligent algorithm is used to solve the CC-DRO model, which combines elite preservation strategy and weight adaptation mechanism. By preserving elite individuals in population evolution and dynamically adjusting search weights based on individual historical performance to guide the optimization direction, the optimal game strategy of the master, i.e. the bidding curve, is obtained. Specifically: Step 4.1 Generate M initial candidate bidding curves through Logistic chaotic mapping, and initialize dynamic weights for each individual; Step 4.2 For each individual in the current population, perform the following operations: Weighted ESN prediction: input the individual and its weight into the pre-trained echo state network (ESN) to predict the response and uncertainty of the user cluster; Constraint violation calculation: for each candidate policy, predict user response through an echo state network (ESN) and solve the second stage value function After, compute total constraint violation : wherein, denotes the maximum allowed rate of change of the total cluster power between adjacent time periods, i.e. the upper ramp rate limit; T denotes the total number of time periods of the scheduling period; denotes the upper limit of the total cluster power of the electric vehicles at scheduling time period t; Fitness evaluation: based on the constraint violation degree, solve the two-stage distribution robust optimization model to calculate the total cost of the individual as the fitness value; Based on ESN prediction and error distribution, solve the two-stage DRO model to compute the total cost: ； wherein, represents the fitness value of the ith individual in the kth generation population; is the total cost calculated for the objective function; is a large penalty factor; Weight update: according to the historical improvement of individual fitness, dynamically adjust the weight, the weight of the excellent performance increases, the weight of the degradation decreases; compare the current fitness of the individual with the last generation, dynamically update the weight: ； wherein, represents a weight decay factor, 0 <1; represents a weight reward factor, >1; Step 4.

3. Selection operation according to weight, and reserve the current elite individuals directly into the next generation, and then perform chaos crossover and mutation on the remaining individuals to generate a new population; Step 4.4 repeats the iteration until the termination condition is met, the number of iterations reaches the preset maximum value, and outputs the individual with the optimal fitness in the history, i.e. the globally optimal bidding strategy. , namely the globally optimal bidding strategy. 8.The V2G aggregator bidding decision method based on master-slave stochastic game according to claim 1, wherein, After executing the bidding strategy and completing the market electricity settlement, the aggregator also carries out a "cash-carbon point" double-factor incentive compatible allocation mechanism, which is as follows: On the one hand, the actual settlement amount is associated with the final R-UPI value and priority of the user, realizing incentive compatibility; On the other hand, according to the total effective discharge capacity and the number of participation of the user in the dispatching period, the carbon point reward is calculated and stored in the virtual account of the user; The cash incentive and carbon point reward together constitute the final income of the user, and the carbon point is exchanged for real objects through the merchant network, completing the entire V2G bidding decision, transaction and double incentive process.

Images