A Virtual Power Plant Response Scheduling Method and Device Based on Dynamic Bayesian Game Incentives

Abstract

This invention discloses a virtual power plant response scheduling method and apparatus based on dynamic Bayesian game incentives. The method first constructs a three-layer game framework with the power grid as the leader, aggregators and users as followers. Then, it constructs the first objective function for the power grid, the second objective function for the aggregators, and the third objective function for the users. Next, it constructs a game model based on the three-layer game framework and the corresponding objective functions. Finally, it solves the game model using backward induction to obtain the virtual power plant response scheduling result. Specifically, in the third objective function, posterior beliefs are obtained by weighted fusion of rational Bayesian beliefs, private anchor beliefs about aggregators, and public anchor beliefs about the power grid. Prospect theory is also introduced to construct a detailed and complex characterization of user behavior, transforming the existing price-quantity game into a dynamic game of competition between the power grid and aggregators for user dominance, thus improving the accuracy of the game.

Description

Technical Field

[0001] This invention relates to the field of smart grid technology, and more specifically, to a virtual power plant response scheduling method and apparatus based on dynamic Bayesian game incentives. Background Technology

[0002] A virtual power plant (VFP) is a software platform system that integrates digital technology, control technology, Internet of Things (IoT) technology, and information and communication technology. Its core functions are communication and aggregation. VFP response dispatch refers to the use of advanced communication, control, and optimization algorithms to aggregate dispersed distributed energy resources (such as photovoltaic, wind power, energy storage, adjustable loads, and electric vehicles) into a unified, dispatchable virtual entity. Based on grid operation demands or electricity market signals, it dynamically adjusts its output or load behavior to achieve functions such as peak shaving, frequency regulation, voltage regulation, and ancillary services.

[0003] Existing technologies employ a three-tier demand response architecture that includes the power grid, aggregators, and users. In this architecture, the aggregator's operating costs and the user's comfort preferences are their respective private information, and information asymmetry is a key issue affecting market efficiency.

[0004] To address the aforementioned issues, some existing technologies propose a three-layer Bayesian-Stackelberg game model. In this model, the top-level power grid G ​​acts as the dominant player, setting the settlement price; the middle-level aggregator A, as a follower, selects a compensation price after observing the settlement price; and the bottom-level user U selects the optimal load reduction amount after observing the compensation price. The core mechanism of this scheme lies in the fact that the aggregator's compensation price is not only economic compensation but also a signal. User U, as the signal receiver, updates its posterior belief about the aggregator type using Bayes' theorem after observing the compensation price, and designs a user utility function based on this posterior belief, thereby effectively utilizing the technical support or service value provided by the efficient aggregator.

[0005] However, the aforementioned technologies still have the following technical problems. First, the existing models assume that individuals are perfectly rational Bayesians who can strictly update their beliefs according to Bayes' rule and strictly maximize their expected utility. This does not conform to reality. Numerous studies have confirmed that individuals exhibit non-expected utility characteristics such as loss aversion and reference point dependence when making decisions, and also have cognitive biases such as anchoring effect when updating beliefs. As a result, the calculation of posterior beliefs does not conform to reality. In addition, the existing models use static game theory, while the interaction between the power grid, aggregators, and users is dynamic and changes. The model cannot accurately describe or characterize it, resulting in poor response scheduling performance. Summary of the Invention

[0006] The purpose of this invention is to provide a virtual power plant response scheduling method and apparatus based on dynamic Bayesian game incentives, which solves the technical problem of poor response scheduling effect caused by mismatch in posterior belief calculation and static game framework in the prior art.

[0007] To achieve the above objectives, the first aspect of the present invention provides a virtual power plant response scheduling method based on dynamic Bayesian game incentives, comprising:

[0008] Construct a three-tiered game theory framework with the power grid as the leader and aggregators and users as followers;

[0009] Based on the current social value of the power grid and the maximum expected discounted social value that the power grid can achieve based on the current system state, the first objective function of the power grid is constructed with the goal of maximizing the total expected profit of the power grid.

[0010] Based on the aggregator's current profit and the maximum expected discounted social value that the aggregator can achieve based on the current system state, a second objective function for the aggregator is constructed with the goal of maximizing the aggregator's total expected profit.

[0011] Based on posterior beliefs, user's private type parameters, unit load reduction compensation price provided by aggregator to user, and aggregator's additional service value, a third objective function for user is constructed with the goal of maximizing prospect theoretical value. The posterior beliefs are obtained by Softmax weighted fusion of rational Bayesian beliefs, private anchor beliefs of aggregator, and public anchor beliefs of power grid.

[0012] A game model is constructed based on a three-layer game framework, a first objective function, a second objective function, and a third objective function;

[0013] The game model was solved using backward induction to obtain the virtual power plant response scheduling results.

[0014] In one implementation, a first objective function for the power grid is constructed based on the current social value of the power grid and the maximum expected discounted social value that the power grid can achieve based on the current system state, with the goal of maximizing the total expected profit of the power grid. This function includes:

[0015] The current social value is obtained by dividing the social value gained by the system due to the actual load reduction by users by the total cost required to achieve the load reduction.

[0016] Based on the first discount factor and the maximum expected discounted social value function under the current system state, calculate the maximum expected discounted social value that the power grid can achieve under the current system state.

[0017] The first objective function is obtained by combining the current social value with the maximum expected discounted social value that the power grid can achieve based on the current system state.

[0018] In one implementation, the first objective function is:

[0019]

[0020] in, Indicates the power grid in the first Each scheduling cycle Under the given conditions, the maximum expected discounted social value function that can be achieved is taken as the objective function. For the first The system state vector for each scheduling cycle includes user belief anchors, trust strength, historical prices, and the aggregator's belief in user type. For the power grid The settlement price set within each scheduling cycle, For the power grid in the first The guidance price published within each scheduling cycle This is a private type parameter for the aggregator. For the conditional expectation operator, Let the social value function of the power grid be within the current dispatch cycle. As the first discount factor, For the expectation operator to address the uncertainty of system state transitions, This is the system state vector for the next cycle, formed after the current cycle's decision-making and user response are completed. Indicates the power grid in the first Each scheduling cycle Under these conditions, the maximum expected discounted social value function that can be achieved.

[0021] In one implementation, the second objective function is:

[0022]

[0023] in, Based on Expectations regarding user type For current profit, For aggregators in the first Within each scheduling cycle, Conditions and their own type The optimal expected discounted value function at that time, i.e., the second objective function, For the first The system state vector for each scheduling cycle includes user belief anchors, trust strength, historical price information, and the aggregator's belief in user type. This is a private type parameter for the aggregator. For aggregators in the first The unit load reduction compensation price issued to users within each scheduling cycle For aggregators in the first Posterior beliefs about user type over a period of time As the second discount factor, For future value function, For the expectation operator of system state transition uncertainty, In the first The next cycle state vector is obtained after the system evolves from the end of each cycle of decision-making and user response.

[0024] In one implementation, the third objective function is:

[0025]

[0026] in, For the load reduction decision variable selected by the user within the current scheduling cycle, For the user utility function based on prospect theory, For users in the first The final posterior belief formed within a scheduling cycle For user-private type parameters, For aggregators in the first The unit load reduction compensation price provided to users within each scheduling cycle In prospect theory, this is the revenue-value function used to describe a user's nonlinear value perception of positive returns within the revenue domain. In prospect theory, this is the loss value function used to describe a user's nonlinear value perception of negative gains within the gain domain. Additional service value function for aggregators This is a private type parameter for aggregators, used to distinguish between efficient and inefficient aggregators. Based on user posterior beliefs The expected value of the additional services provided by the aggregator.

[0027] In one implementation, posterior beliefs are calculated as follows:

[0028]

[0029] in, Indicates the source of information. , This represents a rational Bayesian source. This indicates the aggregator's private anchor source. Indicates the common anchor point information source of the power grid. Indicates the source of information In the Dynamic weights for each scheduling cycle, Indicates the source of information In the The beliefs for each scheduling cycle are dynamically weighted based on the strength of trust and the valence of evidence.

[0030] In one implementation, when solving the game model using backward induction, the method further includes:

[0031] The dynamic weights are updated based on the counterfactual regret value, which is calculated based on the counterfactual prospect theory utility value that the user can obtain when making decisions entirely based on the source’s beliefs and the maximum value among all counterfactual utilities. The magnitude of the counterfactual regret value is used to characterize the ex-post guidance effect of the corresponding source on the user’s decision-making.

[0032] Trust strength is updated using a memory decay mechanism based on counterfactual regret value.

[0033] Based on the same inventive concept, a second aspect of the present invention provides a virtual power plant response scheduling device based on dynamic Bayesian game incentives, comprising:

[0034] The game framework construction module is used to build a three-layer game framework with the power grid as the leader and aggregators and users as followers.

[0035] The first objective function construction module is used to construct the first objective function of the power grid based on the current social value of the power grid and the maximum expected discounted social value that the power grid can achieve based on the current system state, with the goal of maximizing the total expected profit of the power grid.

[0036] The second objective function construction module is used to construct the aggregator's second objective function based on the aggregator's current profit and the maximum expected discounted social value that the aggregator can achieve based on the current system state, with the goal of maximizing the aggregator's total expected profit.

[0037] The third objective function construction module is used to construct the user's third objective function with the goal of maximizing the prospect theoretical value, based on posterior beliefs, the user's private type parameters, the unit load reduction compensation price provided by the aggregator to the user, and the additional service value of the aggregator. The posterior beliefs are obtained by Softmax weighted fusion of rational Bayesian beliefs, private anchor beliefs of the aggregator, and public anchor beliefs of the power grid.

[0038] The game model construction module is used to construct game models based on a three-layer game framework, a first objective function, a second objective function, and a third objective function.

[0039] The model solving module is used to solve the game model using backward induction to obtain the virtual power plant response scheduling results.

[0040] Based on the same inventive concept, a third aspect of the present invention provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, provides the virtual power plant response scheduling method based on dynamic Bayesian game incentives described in the first aspect.

[0041] Based on the same inventive concept, the fourth aspect of the present invention provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, it implements the virtual power plant response scheduling method based on dynamic Bayesian game incentives described in the first aspect.

[0042] Compared with the prior art, the advantages and beneficial technical effects of the present invention are as follows:

[0043] This invention provides a virtual power plant response scheduling method based on dynamic Bayesian game incentives. It constructs a three-layer game framework, establishing a first objective function for the power grid, a second objective function for aggregators, and a third objective function for users. A game model is then built based on the three-layer framework, the first objective function, the second objective function, and the third objective function. Finally, backward induction is used to solve the game model to obtain the virtual power plant response scheduling result. At the user level, posterior beliefs are obtained by Softmax weighted fusion of rational Bayesian beliefs, private anchor beliefs about aggregators, and public anchor beliefs about the power grid. Furthermore, the third objective function incorporates prospect theory to maximize its value, thus constructing a refined and complex characterization of user behavior. This transforms the price-quantity game in existing technologies into a dynamic game of competition between the power grid and aggregators for user dominance, thereby improving the accuracy of the game.

[0044] Furthermore, the system state vector includes user belief anchors, trust strength, historical prices, and aggregator beliefs about user types. This embeds the trust strength between the power grid and the aggregator, as well as the aggregator's beliefs about user types, into the game process, making them endogenous outcomes of the game and further improving the accuracy of the game.

[0045] Furthermore, by updating dynamic weights and trust strength through regret-reinforcement and memory decay mechanisms, anchors (power grids or aggregators) that continuously provide bad information can be adaptively eliminated, enabling the system to converge to a more realistic market state in the long run, thereby improving information efficiency and system robustness. Attached Figure Description

[0046] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0047] Figure 1 This is a flowchart of the virtual power plant response scheduling method based on dynamic Bayesian game incentives in an embodiment of the present invention;

[0048] Figure 2 This is a structural block diagram of a virtual power plant response scheduling device based on dynamic Bayesian game incentives in an embodiment of the present invention. Detailed Implementation

[0049] Example 1

[0050] This embodiment provides a virtual power plant response scheduling method based on dynamic Bayesian game incentives. Please refer to [link to relevant documentation]. Figure 1 ,include:

[0051] S1: Construct a three-layer game framework with the power grid as the leader and aggregators and users as followers.

[0052] Specifically, in the three-layer game framework, the power grid G ​​is the top-level leader, the aggregator A is the middle-level follower, and the private type is the operating cost. User U is a low-level follower, and their private type is comfort loss. in, This refers to the aggregator, which is a mid-level market player in the virtual power plant. This represents a private type parameter for aggregator A, used to characterize its actual operational characteristics that are not directly observed by other participants. This represents the finite set of types to which the aggregator may belong. Aggregators with low operating costs typically possess high efficiency, strong technical capabilities, and low marginal scheduling costs. Aggregators with high operating costs typically have lower efficiency and higher marginal scheduling costs. This refers to the user, which is the terminal load response entity in the virtual power plant. This represents a private type parameter for user U, used to characterize the user's inherent discomfort or loss of comfort in response to load reduction behavior; This represents the set of comfort loss types that a user may belong to. This indicates users with low comfort loss, meaning they are more flexible in responding to load reduction, have high tolerance, and low response costs. This indicates users with a high level of comfort loss, meaning they are more sensitive to load reduction and have high response costs.

[0053] Comfort loss refers to the degree to which a person's subjective feelings or user experience decline due to deviations from ideal conditions in system design, operating status, or external environment.

[0054] The key technical points of this invention include:

[0055] (1) The prospect theory-based non-expected utility function is used to replace the traditional expected utility function.

[0056] (2) Replace the single Bayesian update with a competitive anchoring and nonlinear belief fusion approach.

[0057] (3) The regret-reinforcement learning mechanism enables the trust or reputation of the game participants to be endogenized and dynamically evolved.

[0058] The core of the three-layer game framework includes the user layer (user utility and user beliefs), the weight evolution layer, and the strategy game layer.

[0059] Among them, the user utility layer is based on prospect theory. In the final stage of the game, users no longer maximize expected utility, but instead maximize their prospect theory value function. This function can more realistically depict the user's asymmetric perception of compensation (gain) and loss of comfort.

[0060] The user belief layer is implemented based on Softmax competitive anchoring. Unlike existing methods, during the user belief update phase, the user's final posterior belief is... It is no longer a pure Bayesian update, but a nonlinear weighted fusion of three information sources. As a rational Bayesian information source, For aggregator's private anchor source, As a common anchor point information source for the power grid, the corresponding weight is , indicating the first The user in the cycle [number] for the [number]th period [number] The degree of subjective trust (trust strength) in each source of information, including rational Bayesian beliefs. (Based on rational Bayesian calculations), aggregator's private anchor beliefs (Based on historical prices) and the belief in the common anchor of the power grid (Based on the guidance price).

[0061] Weighted Evolution Layer (Regret-Reinforcement Learning): The aforementioned fused weights The value representing the user's trust in each information source is no longer a fixed parameter, but is dynamically updated based on counterfactual regret at the end of each cycle. Furthermore, a regret-reinforcement learning (RRL) algorithm is used to assign higher trust weights to information sources with less regret (i.e., more accurate anchors), and a memory decay mechanism is introduced.

[0062] Strategic Game Theory Layer (High-Dimensional Dynamic Programming): The power grid and aggregators, as upper-level participants, can anticipate the complex behaviors of users. Their decision-making is a high-dimensional dynamic programming problem, with state variables... It encompasses all intrinsic belief anchors and trust weights; therefore, the strategies of power grids and aggregators are not merely short-term profit-seeking, but also a matter of trust weights placed on users' future trust. and belief anchors Long-term investment and competitive manipulation.

[0063] S2: Based on the current social value of the power grid and the maximum expected discounted social value that the power grid can achieve based on the current system state, construct the first objective function of the power grid with the goal of maximizing the total expected profit of the power grid.

[0064] S2 includes:

[0065] S2.1: The current social value is obtained by combining the social value gained by the system due to the actual load reduction by users with the total cost required to achieve the load reduction.

[0066] S2.2: Calculate the maximum expected discounted social value that the power grid can achieve based on the current system state, according to the discount factor and the maximum expected discounted social value function under the current system state.

[0067] S2.3: Based on the current social value and the maximum expected discounted social value that the power grid can achieve based on the current system state, the first objective function is obtained.

[0068] The first objective function is as follows:

[0069]

[0070] in, Indicates the power grid in the first Each scheduling cycle Under the given conditions, the maximum expected discounted social value function that can be achieved is taken as the objective function. For the first The system state vector for each scheduling cycle includes user belief anchors, trust strength, historical prices, and the aggregator's belief in user type. For the power grid The settlement price set within each scheduling cycle is used to pay aggregators economic compensation for unit load reduction; it is a hard incentive tool. For the power grid in the first The guidance price released within each scheduling cycle serves as a public reference signal, influencing users' belief formation and the evolution of trust weight. It is a soft intervention tool used to characterize the aggregator's operational efficiency or cost level. This is a private type parameter for aggregators, used to characterize the aggregator's operational efficiency or cost level. This is a conditional expectation operator, used for expectation calculation based on prior or posterior beliefs about the aggregator type of the power grid. This is the social value function of the power grid during the current dispatch cycle, used to measure the net benefits at the system level. This refers to the optimal load reduction response amount determined by the user within the current period based on their posterior beliefs and prospect theory utility function. The social value or system benefits that users gain from the optimal load reduction in the system include, but are not limited to, improved grid operation security, reduced peak-to-valley difference, and enhanced dispatch flexibility. The total cost required to achieve load shedding includes expenses such as settlement fees paid by the grid to aggregators. As the first discount factor, satisfying It is used to weigh the importance of current cycle profits against future cycle profits. This is an expectation operator for addressing the uncertainty of system state transitions, used to describe the change in system state from the current decision action. Evolved to randomness, This is the system state vector for the next cycle, formed after the current cycle decision and user response are completed.

[0071] S3: Based on the aggregator's current profit and the maximum expected discounted social value that the aggregator can achieve based on the current system state, construct the aggregator's second objective function with the goal of maximizing the aggregator's total expected profit.

[0072] Specifically, the second objective function is:

[0073]

[0074] in, For user types Expectations (based on their beliefs) ), It is the profit of the current cycle, that is, the profit of the aggregator in the first cycle. The instantaneous profit function within each period depends on the compensation price, user response behavior, and the aggregator's operating costs. The calculation formula is as follows: , It is a complex response based on prospect theory during the game process. The actual load reduction for users; For aggregators in the first Within each scheduling cycle, Conditions and their own type The optimal expected discounted value function, i.e., the second objective function, is used to characterize the long-term profit objective of the aggregator in a multi-period game. For the first The system state vector for each scheduling cycle includes endogenous state variables such as user belief anchors, trust strength, historical price information, and aggregator beliefs about user types. This is a private type parameter for aggregators, used to distinguish aggregators with different operational efficiencies or cost levels. For aggregators in the first The unit load reduction compensation price issued to users within each scheduling cycle is a decision variable for the aggregator during the decision-making stage. For aggregators in the first Posterior beliefs about user type over a given period are used to calculate expected returns under conditions of incomplete information. As the second discount factor, satisfying This is used to weigh the importance of current gains against future gains; The expectation operator is used to describe the uncertainty of system state transitions given the current decision. Towards The randomness of evolution In the first After each cycle of decision-making and user response, the system evolves to obtain the state vector for the next cycle. It is a future value function.

[0075] S4: Based on posterior beliefs, the user's private type parameters, the unit load reduction compensation price provided by the aggregator to the user, and the additional service value of the aggregator, a third objective function for the user is constructed with the goal of maximizing the prospect theoretical value. The posterior beliefs are obtained by Softmax weighted fusion of rational Bayesian beliefs, private anchor beliefs of the aggregator, and public anchor beliefs of the power grid.

[0076] Specifically, the third objective function is:

[0077]

[0078] in, The load reduction decision variable selected by the user in the current scheduling cycle represents the amount of electricity consumption that the user actively reduces. It is a user utility function based on prospect theory, used to characterize the nonlinear perception of compensation gains and comfort losses by users under reference point dependence and loss aversion conditions. For users in the first The final posterior belief formed within a scheduling cycle represents the subjective probability that the user considers the aggregator to be of an efficient type. This is a user-private parameter used to characterize the user's sensitivity to the loss of comfort caused by load reduction. The larger the value, the less willing the user is to accept load reduction. For aggregators in the first The unit load reduction compensation price provided to users within each scheduling cycle In prospect theory, the benefit-value function is used to describe a user's nonlinear value perception of positive benefits within the benefit domain, and it typically exhibits diminishing marginal utility. In prospect theory, the loss value function is used to describe a user's nonlinear value perception of negative gains within the gain domain, and it typically reflects loss aversion characteristics. The additional service value function for aggregators represents the non-monetary added value that users can obtain when aggregators are of different types, with efficient aggregators providing higher added value. This is a private type parameter for aggregators, used to distinguish between efficient and inefficient aggregators. These represent efficient and inefficient aggregator types, respectively. Based on user posterior beliefs The expected value of the aggregator's additional services calculated. This is the optimal load reduction response strategy determined by maximizing the prospect theory utility function, given the user's current posterior beliefs, aggregator compensation price, and their own type.

[0079] It's important to note that loss aversion refers to the psychological perception of loss being stronger than that of equivalent gains when making decisions, thus leading to a greater tendency to avoid losses. Reference point dependence refers to evaluating outcomes not based on absolute gains or costs, but rather on changes relative to a specific reference point.

[0080] The posterior belief is calculated as follows:

[0081]

[0082] in, Indicates the source of information. , This represents a rational Bayesian source. This indicates the aggregator's private anchor source. Indicates the common anchor point information source of the power grid. Indicates the source of information Dynamic weights, Indicates the source of information The dynamic weights of the beliefs are derived from the strength of trust and the valence of evidence.

[0083] Specifically, dynamic weights Determined by the normalized exponential model (Softmax), this weight is based on the strength of trust. and the valence of evidence The following was decided jointly:

[0084]

[0085] in, Indicates the source of information The strength of trust, Indicates the source of information The valence of evidence is determined by the logarithmic odds of the belief, specifically:

[0086]

[0087] in, Indicates the first Within each scheduling cycle, the information source The valence of evidence is used to characterize the relative persuasiveness or information strength of the belief information currently provided by the source. Its value is expressed as the logarithmic probability of the corresponding belief, as follows:

[0088]

[0089] in, Indicates the source of information In the The probability of the belief that the aggregator is an efficient type given within a period. (Subscript) Indicates a source index for beliefs, used to distinguish different information sources; subscript It represents the scheduling cycle or game round index; the log odds form is used to map probabilistic beliefs to an unbounded real space so that they can be used together with the trust weights to determine the relative influence of each source in the subsequent Softmax competitive belief fusion.

[0090] The parameters in the three objective functions of this application are obtained based on objective power operation data. This embodiment illustrates the relationship between variables and physical quantities in this application, specifically providing a method for mapping objective power operation data to the input parameters of the demand response game model. This method aims to solve the problem that abstract variables (such as social value, comfort loss, etc.) in the original model are difficult to measure directly in engineering practice. It establishes a bridging relationship between physical quantities and abstract quantities in this application through the following steps.

[0091] I. Optimal load reduction for users

[0092] Based on real-time load monitoring data, calculate the user's load in the first... Actual load reduction power within each scheduling cycle :

[0093]

[0094] In the formula: Indicates the first The actual load reduction power of users within each scheduling cycle; Indicates the user's baseline load; Indicates the first Real-time active load of users within a scheduling cycle.

[0095] This leads to the corresponding load reduction in electricity consumption. :

[0096]

[0097] In the formula: Indicates the first The user's load reduction electricity consumption within a scheduling cycle; This indicates the duration of a single scheduling cycle.

[0098] Ultimately, the normalized optimal load reduction for users. The calculation formula is:

[0099]

[0100] In the formula: This represents the optimal load reduction for the user, and is a dimensionless proportion. Indicates user The rated power consumption.

[0101] II. Social Value Function

[0102] The social value consists of three parts: the value of avoiding power shortages, the revenue from peak-valley price differences, and the value of reducing reserve demand. The comprehensive formula is as follows:

[0103]

[0104] In the formula: Represents the social value function; This indicates the value of the load loss caused by a power outage. Indicates the electricity price during peak hours of the system; This indicates the electricity price during off-peak hours. This represents the cost coefficient per unit of standby capacity.

[0105] III. Total Cost Function

[0106] The total cost includes settlement and payment costs, communication and scheduling costs, and energy storage dispatch costs. The calculation formula is as follows:

[0107]

[0108] In the formula: Represents the total cost function; Indicates the power grid in the first Settlement price for each scheduling cycle; This represents the fixed communication cost of a single scheduling operation. This represents the marginal scheduling cost coefficient; This indicates the cost of calling up the energy storage system.

[0109] Among them, the unit depreciation cost in the energy storage system call cost The formula is as follows:

[0110]

[0111] In the formula: This represents the depreciation cost per unit discharge of energy storage. This indicates the total investment cost of the energy storage system; Indicates the rated capacity of the energy storage system; Indicates the rated number of cycles; Indicates the depth of discharge.

[0112] IV. User comfort loss parameters

[0113] Based on the rigid characteristics of user load, comfort loss parameters are obtained using historical load data. :

[0114]

[0115] In the formula: This represents a parameter indicating the loss of user comfort. This represents the user's historical average load. The standard deviation of a user's historical load; It represents the smallest positive number that is not divisible by zero.

[0116] V. Aggregator Operating Cost Parameters

[0117] Aggregator's proprietary cost parameters Determined by the operational efficiency of the resources under its jurisdiction:

[0118]

[0119] In the formula: This indicates the aggregator's operating cost parameters; This represents the unit power generation and maintenance cost of distributed power sources; This indicates the real-time output of the distributed power source; Indicates the energy storage discharge efficiency; This represents the communication control cost for a single user; Indicates the number of users; This indicates the total response power of the aggregator during this period.

[0120] VI. Price Benchmark and

[0121] 1. User-side pricing benchmark (Using an exponentially weighted moving average):

[0122]

[0123] In the formula: Indicates the first The price benchmark for users to aggregators in a given period; Indicates the attenuation factor; This indicates the actual compensation price for a historical period.

[0124] 2. Power Grid Guidance Price Benchmark :

[0125]

[0126] In the formula: Indicates the first The benchmark price for the power grid guidance price in each cycle; This indicates that the spot market is clearing out electricity prices; Indicates the peak-valley adjustment coefficient; Indicates the system's peak load; This indicates a low load on the system.

[0127] VII. Aggregator's Real-Time Profit (Current Profit) Measured correlation

[0128] Using electricity consumption data measured by smart meters, the aggregator's real-time profit is expressed as:

[0129]

[0130] In the formula: Indicates the aggregator in the first Instant profit for each scheduling cycle.

[0131] S5: Construct a game model based on a three-layer game framework, a first objective function, a second objective function, and a third objective function.

[0132] Specifically, the decision-making processes of the three parties in the game theory model are as follows:

[0133] 1. Top-level leader of the power grid

[0134] The power grid's decision-making is driven by market managers who combine hard incentives with soft interventions.

[0135] The dynamic decision-making of the power grid is for each cycle Initially, based on the current global state A strategy package needs to be decided simultaneously. ,in, Let be the settlement price for period t, and let be a hard incentive tool. Its decision objective is to set an optimal price to achieve the desired total reduction at the lowest cost after anticipating that the aggregator will implement a complex anchoring-manipulation strategy.

[0136] It should be noted that the hard incentive strategy refers to the power grid, as the top leader, determining a state-dependent settlement price decision strategy by solving cross-cycle dynamic programming within each scheduling cycle, based on the current system state and anticipating the dynamic pricing manipulation behavior that aggregators may take to influence users' belief anchors and trust weights, in order to minimize the unit load reduction cost while meeting the expected load reduction demand.

[0137] Anchoring-manipulation strategy refers to the strategy by which aggregators, in the process of multi-cycle games, do not only aim to maximize profits in the current cycle, but also consciously adjust their compensation price to form or reinforce a specific price anchor in the user's perception. This influences the user's belief formation, trust weight allocation, and response decisions in subsequent cycles, sacrificing or adjusting short-term gains in exchange for more favorable response behavior and long-term benefits in future cycles.

[0138] It serves as a public anchor price and as a tool for soft or behavioral intervention. It is the power grid in a cycle Published, independent of settlement price This provides a public reference signal, enabling the power grid to directly intervene in user beliefs. Its decision-making objectives are counterbalancing and investment, with counterbalancing including: issuing a public anchor point. To counter the private anchors of mid-level aggregators Excessive manipulation of users, investment includes strategic settings This makes it an accurate anchor point, winning user trust in the long run (through regret-reinforcement learning), i.e., maximizing the credibility weight of the power grid. The power grid's ultimate decision is to maximize its total discounted social value across cycles. .

[0139] 2. Aggregator (A) - Mid-level follower

[0140] Aggregators' decisions include belief manipulation and reputation management.

[0141] The dynamic decision-making in this invention is the decision made by the aggregator based on its observation of the power grid. and current state Then, the compensation price was decided. , It is an extremely complex meta-strategy whose objective is no longer to maximize short-term profits, but rather to maximize its total discounted value function across cycles. This decision The following objectives must be weighed simultaneously:

[0142] Short-term profits: Current cycle The immediate benefits.

[0143] Private Anchoring: Decision Making (Especially high-efficiency aggregators) Should we deliberately set high prices (sacrificing short-term profits) in order to achieve our goals within a cycle? Use a high-value private anchor Imprinted in the minds of users.

[0144] Trust manipulation: decision making How to influence users' actual utility and counterfactual regret In order to period In the evolution of weights, maximizing the user's trust weight. .

[0145] Learning users: Decision-making How to most effectively attract different types of users Respond so that aggregators can learn in reverse (i.e. update) This allows for more accurate identification of user types.

[0146] 3. Users - Bottom-level Followers

[0147] User decision-making includes basic decisions and dynamic decisions.

[0148] (1) Basic decision-making: observation Use Bayes' theorem to update beliefs.

[0149] Based on one's own private comfort type Select a load reduction amount To maximize expected utility (obtained by subtracting the loss of comfort from the compensation benefit).

[0150] (2) Dynamic decision-making:

[0151] Users' decision-making process can be divided into two stages, in terms of time and cognition:

[0152] Decision 1: Action Decision (Corresponds to stage 4 in the reverse solution)

[0153] Belief Fusion: The user first performs Softmax belief fusion based on the current trust weights. Together with three competing anchors, a final irrational posterior belief is formed. .

[0154] In the model, the game is repeated over multiple cycles, and variables such as trust weights, anchor beliefs, and prices evolve with each cycle.

[0155] Irrational optimization: Users will Substituting into its objective function: Prospect Theory Value Function This function reflects that users are more sensitive to losses when making decisions and tend to use a certain reference level as a benchmark to judge gains or losses. User selection To maximize this nonlinearity Rather than the traditional expected utility.

[0156] Decision 2: Trust Renewal (Corresponding to stage 3.3 in the reverse solution)

[0157] Counterfactual regret: in cycles After completion, the user calculates their information for each anchor source. Counterfactual regret .

[0158] Weight update: The user's cognitive system automatically executes the regret-reinforcement learning algorithm, assigning a higher trust strength to the information source with the least regret in the next cycle.

[0159] The user's decision within a cycle (corresponding to stage 4 in the reverse solution) is to maximize the future value, and the decision between cycles (corresponding to stage 3.3 in the reverse solution) is to minimize future regret.

[0160] S6: The game model is solved using backward induction to obtain the virtual power plant response scheduling results.

[0161] When using backward induction to solve the game theory model, the method further includes:

[0162] The dynamic weights are updated based on the counterfactual regret value, which is calculated based on the counterfactual prospect theory utility value that the user can obtain when making decisions entirely based on the source’s beliefs and the maximum value among all counterfactual utilities. The magnitude of the counterfactual regret value is used to characterize the ex-post guidance effect of the corresponding source on the user’s decision-making.

[0163] Trust strength is updated using a memory decay mechanism based on counterfactual regret value.

[0164] Specifically, game theory in The process is repeated within a certain number of cycles, and the solution is obtained using backward induction. The core of this invention lies in defining a complex, high-dimensional state variable. This state variable dynamically shifts between periods.

[0165]

[0166] in, This serves as a reference value for anchoring aggregator prices, representing users' private anchoring beliefs about the aggregator (based on...). (), meaning that the aggregator intelligent agent in the cycle The intrinsic price benchmark upon which decisions are based. The grid settlement price anchoring reference value represents users' public anchoring belief in aggregators (based on...). ), meaning that the power grid-side intelligent agent in the cycle The reference benchmarks used when setting settlement prices Represented as (i=B, A, G): the user's trust strength among the three information sources: Bayesian, aggregator, and power grid. For aggregators in the first At the end of each scheduling cycle, the posterior belief in the user type, also known as the system discount factor, is used to characterize the subjective probability assessment of whether a user belongs to a different type. The formula is:

[0167]

[0168] Specifically, this means that, given the date up to the [number]th [year]... User type for The probability, Indicates the scheduling period or game round index. This represents a user's private type, used to characterize the user's comfort loss characteristics. This indicates a predefined user type (e.g., high flexibility or low comfort loss type). Indicates as of the date The set of historical information observable by the aggregator for each cycle includes at least historical compensation prices, actual user load response, and user behavior information inferred from them.

[0169] Specifically, the physical meaning of each parameter in the system state vector and its role in the model are defined as follows:

[0170] (1) Aggregator price anchoring reference value Defined as an aggregator intelligent agent in a cycle The intrinsic price benchmark upon which decisions are based. This parameter quantifies the "anchoring effect" of historical trading prices or market signals on the aggregator's current pricing strategy. It is a key state variable characterizing the aggregator's imperfectly irrational decision-making behavior and directly affects the direction and magnitude of its pricing strategy adjustments.

[0171] (2) Reference value for grid settlement price anchoring Defined as a power grid-side intelligent agent in a periodic The benchmark upon which settlement prices are based. This parameter reflects the price inertia formed by grid operators based on historical electricity prices, long-term system operating costs, etc., and is used to model the price adjustment stickiness of the grid in the face of real-time supply and demand changes.

[0172] (3) Marginal utility coefficient of user demand response Defined as an end-user intelligent agent in a periodic The marginal utility gained from a unit load reduction. This parameter is the core coefficient connecting the incentive price and the user response, directly determining the elasticity of the user demand response curve and reflecting the user's sensitivity to price signals.

[0173] (4) Aggregator revenue weighting coefficient Defined as an aggregator intelligent agent in a cycle The weight assigned to its own profit objective. In the game theory model of this invention, this parameter is used to balance the weight allocation between the aggregator's pursuit of maximizing its own profit and assisting the power grid in achieving the regulation objective, and is a component of the objective function of its strategy optimization problem.

[0174] (5) Weighting coefficient of power grid regulation revenue Defined as a power grid-side intelligent agent in a periodic The weight assigned to social value or system operational benefits. This parameter is used to quantify the comprehensive consideration of system operating costs, power supply reliability, and social benefits in power grid dispatching decisions, reflecting the public utility nature of the power grid as a system operator.

[0175] (6) System discount factor : Defined as the discount factor that discounts future returns to the current period. This parameter satisfies It is used to construct the intertemporal payoff function in dynamic games, so that the scheduling model of the present invention can simulate and optimize multi-period, forward-looking decision-making processes, and its numerical value represents the system's attention to long-term payoffs.

[0176] In the specific implementation process, the game sequence and implementation are as follows:

[0177] In any period The game unfolds in the following four stages, described using backward induction, from stage 4 to stage 3 to stage 2, and finally to stage 1.

[0178] Phase 4(t): User Response (Based on Prospect Theory)

[0179] At this stage, users have formed the posterior beliefs established in stage 3 of the previous stage. and its own type is known. Quotations from aggregators Users select load reduction amount .

[0180] Step 4.1: Define the reference point and value function

[0181] This invention employs Prospect Theory.

[0182] (1) Set "not to participate" ( (as a reference point for decision-making) .

[0183] (2) Define the nonlinear value function :

[0184] Revenue Domain : , ( (Diminishing marginal utility).

[0185] Loss domain : , ( (loss aversion).

[0186] Step 4.2: User Optimization

[0187] User selection To maximize its total prospect value :

[0188]

[0189] in, that place It is an added value of efficient aggregators. This is the optimal response strategy for users.

[0190] Phase 3(t): User Belief Fusion (Based on Softmax Competitive Anchoring)

[0191] At this stage, type Users observed the aggregator's and the power grid This forms the final belief about the aggregator type. .

[0192] The specific steps of stage 3(t) include:

[0193] Step 3.1: Identify three competing information sources

[0194] (1) Rational Bayes User based And the rational posterior calculated according to Bayes' rule, which is the (equilibrium) expectation of the aggregator's strategy.

[0195] Before the game begins, users have a priori belief about the type of aggregator. :

[0196] (i.e., users assume in advance that the aggregator is efficient) (probability)

[0197] (i.e., users presuppose that aggregators are inefficient) (probability)

[0198] At the same time, what price range will users choose for different types of aggregators? There is an equilibrium expectation (strategy expectation):

[0199] (i.e., users expect an efficient) Aggregator Chamber of Commerce Selection (Probability of this price)

[0200] (i.e., users expect an inefficient) Aggregator Chamber of Commerce Selection (Probability of this price)

[0201] When users are in a cycle Observed specific prices At that time, calculate rational posterior beliefs Bayes' theorem is as follows:

[0202]

[0203] Among them, private anchor points In the corresponding state variables It is based on the price of the previous period. The formed beliefs are calculated as follows:

[0204]

[0205] in, It is the historical average or reference benchmark of the aggregator's prices for users. It is the user's sensitivity to private price signals.

[0206] Public Anchor Point In the corresponding state variables It is based on the power grid announcement for this cycle. The formed beliefs are calculated as follows:

[0207]

[0208] in, It serves as a reference benchmark for the power grid guidance price. It is the user's sensitivity to the public guidance price.

[0209] Step 3.2: Nonlinear Belief Fusion

[0210] In this invention, the ultimate belief It is a softmax weighted fusion of three information sources, calculated as follows:

[0211]

[0212] The meaning of the specific parameters has been explained in the previous text and will not be repeated here.

[0213] Stage 3.3(t): Evolution of Trust Weights (Based on Regret-Reinforcement Learning)

[0214] This stage is in the cycle Executed at the end, used to calculate the state. In .

[0215] Step 3.3.1: Calculate counterfactual regret

[0216] Counterfactual regret Regret is defined as a loss of opportunity. In this invention, we first define the optimal utility that can be realized among all information sources, and then calculate... The regret regarding this optimal value:

[0217]

[0218]

[0219] In this embodiment, , The smaller the value, the better the source. The better the performance.

[0220] Indicates the first After each scheduling cycle ends, the user has access to all information sources. The maximum counterfactual utility value that can be obtained in the middle.

[0221] The formula for calculating counterfactual regret is:

[0222]

[0223] in, Indicates the first Within each scheduling cycle, the user is relative to the information source. The resulting counterfactual regret value; These represent rational Bayesian information sources, aggregator private anchor point information sources, and power grid public anchor point information sources, respectively. This indicates that, assuming the user is in the first... Within each cycle, it is entirely based on the information source. The counterfactual prospect theory utility value that can be obtained when making decisions based on beliefs.

[0224] because It is the maximum of all counterfactual utilities, therefore it is always present in this embodiment. and, The smaller the value, the stronger the corresponding information source. The better the performance during this period, the better the post-event guidance effect on user decision-making, and thus the greater the weight increase in subsequent trust weight reinforcement learning updates.

[0225] Step 3.3.2: Regret - Enhancement Update

[0226] This implementation uses a memory decay factor. and learning rate The regret-reinforcement rule updates the trust strength, specifically in the following way:

[0227]

[0228] In the formula, This reflects forgetting or memory decline. This embodies reinforcement learning. The closer a source is to 0 and the less regret it receives, the greater the increase in trust strength it receives in the next cycle.

[0229] Phase 2(t): Dynamic Strategy of Aggregators

[0230] At this stage, type Aggregator A observed the power grid and system status After that, decision .

[0231] The specific steps in this stage include:

[0232] Step 2.1: Establish the Bellman equation

[0233] Aggregator A is solving a high-dimensional dynamic programming problem to maximize its total expected discounted profit. :

[0234]

[0235] Step 2.2: State Transition

[0236] Manipulating Private Anchors: Aggregator Decisions Will become Periodic Highly efficient aggregators may sacrifice short-term profits. To establish a higher price reference level, thereby obtaining a higher discounted profit in subsequent periods, in exchange for discounted profit. .

[0237] Manipulating trust levels: By influencing users Indirect impact The evolution of.

[0238] Learning User: A Observed at the end of the period This is about The signal. A uses Bayesian update. .

[0239] Phase 1(t): Dynamic Strategy of the Power Grid

[0240] During this phase, the power grid G ​​acts as the dominant player, making decisions. .

[0241] Step 1.1: Establish the top-level Bellman equation

[0242] Solve the dynamic programming problem of power grid G ​​to maximize the total expected social value. :

[0243]

[0244] Step 1.2: Dual strategy tools for the power grid, including hard incentives and soft anchoring.

[0245] (1) Hard incentives This refers to the traditional settlement price, used to incentivize aggregator A.

[0246] (2) Soft anchoring This refers to the behavioral intervention tool proposed in this invention, which the power grid uses to set up... Active investment and This is to counteract the private anchoring manipulation of aggregators, guide users' bounded rationality toward the socially optimal direction, and compete for dominance over users' beliefs.

[0247] Phase 0: Initialization

[0248] exist At that time, the system is in the initialization state. This includes the prior beliefs of all parties and the initial anchor point. (Based on historical averages), and initial trust weights (Uniformly distributed).

[0249] It should be noted that the above process is a reverse solution process. In the actual implementation, the game is executed sequentially in the order of stage 0 → stage 1 → stage 2 → stage 3 → stage 4 within each scheduling cycle.

[0250] Example 2

[0251] Based on the same inventive concept, this embodiment discloses a virtual power plant response scheduling device based on dynamic Bayesian game incentives. Please refer to [link to relevant documentation]. Figure 2 ,include:

[0252] Game framework construction module 101 is used to construct a three-layer game framework with the power grid as the leader and aggregators and users as followers;

[0253] The first objective function construction module 102 is used to construct the first objective function of the power grid based on the current social value of the power grid and the maximum expected discounted social value that the power grid can achieve based on the current system state, with the goal of maximizing the total expected profit of the power grid.

[0254] The second objective function construction module 103 is used to construct the aggregator's second objective function based on the aggregator's current profit and the maximum expected discounted social value that the aggregator can achieve based on the current system state, with the goal of maximizing the aggregator's total expected profit.

[0255] The third objective function construction module 104 is used to construct the user's third objective function with the goal of maximizing the prospect theoretical value, based on the posterior belief, the user's private type parameter, the unit load reduction compensation price provided by the aggregator to the user, and the additional service value of the aggregator. The posterior belief is obtained by Softmax weighted fusion of rational Bayesian belief, private anchor belief of the aggregator, and public anchor belief of the power grid.

[0256] The game model construction module 105 is used to construct a game model based on a three-layer game framework, a first objective function, a second objective function, and a third objective function.

[0257] The model solving module 106 is used to solve the game model using backward induction to obtain the virtual power plant response scheduling results.

[0258] Since the apparatus in Embodiment 2 of this invention is the same apparatus used in the virtual power plant response scheduling method based on dynamic Bayesian game incentives in Embodiment 1, those skilled in the art can understand the specific structure and variations of this apparatus based on the method described in Embodiment 1 of this invention, and therefore will not be repeated here. All apparatuses used in the method of Embodiment 1 of this invention fall within the scope of protection of this invention.

[0259] Example 3

[0260] Based on the same inventive concept, the present invention also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the virtual power plant response scheduling method based on dynamic Bayesian game incentives of Embodiment 1.

[0261] Since the computer-readable storage medium described in Embodiment 3 of this invention is the same computer-readable storage medium used in implementing the virtual power plant response scheduling method based on dynamic Bayesian game incentives in Embodiment 1 of this invention, those skilled in the art can understand the specific structure and variations of this computer-readable storage medium based on the method described in Embodiment 1 of this invention, and therefore will not be repeated here. All computer-readable storage media used in the method of Embodiment 1 of this invention fall within the scope of protection of this invention.

[0262] Example 4

[0263] Based on the same inventive concept, the present invention also provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the method described in Embodiment 1.

[0264] Since the computer device described in Embodiment 4 of this invention is the same computer device used to implement the virtual power plant response scheduling method based on dynamic Bayesian game incentives in Embodiment 1 of this invention, those skilled in the art can understand the specific structure and variations of this computer device based on the method described in Embodiment 1 of this invention, and therefore will not be described again here. All computer devices used in the method of Embodiment 1 of this invention fall within the scope of protection of this invention.

[0265] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0266] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0267] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention. Clearly, those skilled in the art can make various modifications and variations to the embodiments of the invention without departing from the spirit and scope of the invention. Thus, if these modifications and variations of the embodiments of the invention fall within the scope of the claims of the invention and their equivalents, the invention also intends to include these modifications and variations.

Claims

1. A virtual power plant response scheduling method based on dynamic Bayesian game incentives, characterized in that, include: Construct a three-tiered game theory framework with the power grid as the leader and aggregators and users as followers; Based on the current social value of the power grid and the maximum expected discounted social value that the power grid can achieve based on the current system state, the first objective function of the power grid is constructed with the goal of maximizing the total expected profit of the power grid. Based on the aggregator's current profit and the maximum expected discounted social value that the aggregator can achieve based on the current system state, a second objective function for the aggregator is constructed with the goal of maximizing the aggregator's total expected profit. Based on posterior beliefs, user's private type parameters, unit load reduction compensation price provided by aggregators to users, and the additional service value of aggregators, a third objective function for users is constructed with the goal of maximizing the prospect theoretical value. The posterior beliefs are obtained by Softmax weighted fusion of rational Bayesian beliefs, private anchor beliefs about aggregators, and public anchor beliefs about the power grid. A game model is constructed based on a three-layer game framework, a first objective function, a second objective function, and a third objective function; The game model is solved using backward induction to obtain the virtual power plant response scheduling results. When solving the game model using backward induction, the method further includes: The dynamic weights are updated based on the counterfactual regret value. The dynamic weights are the weights used when performing Softmax weighted fusion. The counterfactual regret value is calculated based on the counterfactual prospect theory utility value that the user can obtain when making decisions entirely based on the source's beliefs and the maximum value among all counterfactual utilities. The magnitude of the counterfactual regret value is used to characterize the ex-post guidance effect of the corresponding source on the user's decision. Trust strength is updated using a memory decay mechanism based on counterfactual regret value.

2. The virtual power plant response scheduling method based on dynamic Bayesian game incentives as described in claim 1, characterized in that, Based on the current social value of the power grid and the maximum expected discounted social value that the power grid can achieve under the current system state, a first objective function for the power grid is constructed with the goal of maximizing the total expected profit of the power grid, including: The current social value is derived from the social value gained by the system due to actual load reduction by users and the total cost required to achieve the load reduction. Based on the first discount factor and the maximum expected discounted social value function under the current system state, calculate the maximum expected discounted social value that the power grid can achieve under the current system state. The first objective function is obtained by combining the current social value with the maximum expected discounted social value that the power grid can achieve based on the current system state.

3. The virtual power plant response scheduling method based on dynamic Bayesian game incentives as described in claim 2, characterized in that, The first objective function is: in, Indicates the power grid in the first Each scheduling cycle Under the given conditions, the maximum expected discounted social value function that can be achieved is taken as the objective function. For the first The system state vector for each scheduling cycle includes user belief anchors, trust strength, historical prices, and the aggregator's belief in user type. For the power grid The settlement price set within each scheduling cycle, For the power grid in the first The guidance price published within each scheduling cycle This is a private type parameter for the aggregator. For the conditional expectation operator, Let the social value function of the power grid be within the current dispatch cycle. As the first discount factor, For the expectation operator to address the uncertainty of system state transitions, This is the system state vector for the next cycle, formed after the current cycle's decision-making and user response are completed. Indicates the power grid in the first Each scheduling cycle Under these conditions, the maximum expected discounted social value function that can be achieved.

4. The virtual power plant response scheduling method based on dynamic Bayesian game incentives as described in claim 1, characterized in that, The second objective function is: in, For based on Expectations regarding user type For current profit, For aggregators in the first Within each scheduling cycle, and its own type The optimal expected discounted value function at that time, i.e., the second objective function, For the first The system state vector for each scheduling cycle includes user belief anchors, trust strength, historical price information, and the aggregator's belief in user type. This is a private type parameter for the aggregator. For aggregators in the first The unit load reduction compensation price issued to users within each scheduling cycle For aggregators in the first Posterior beliefs about user type over a period of time As the second discount factor, For future value function, For the expectation operator of the uncertainty of system state transition, In the first The next cycle state vector is obtained after the system evolves from the end of each cycle of decision-making and user response.

5. The virtual power plant response scheduling method based on dynamic Bayesian game incentives as described in claim 1, characterized in that, The third objective function is: in, For the load reduction decision variable selected by the user within the current scheduling cycle, For the user utility function based on prospect theory, For users in the first The final posterior belief formed within a scheduling cycle For user-private type parameters, For aggregators in the first The unit load reduction compensation price provided to users within each scheduling cycle In prospect theory, this is the revenue-value function used to describe a user's nonlinear value perception of positive returns within the revenue domain. In prospect theory, this is the loss value function used to describe a user's nonlinear value perception of negative gains within the gain domain. Additional service value function for aggregators This is a private type parameter for aggregators, used to distinguish between efficient and inefficient aggregators. Based on user posterior beliefs The expected value of the additional services provided by the aggregator.

6. The virtual power plant response scheduling method based on dynamic Bayesian game incentives as described in claim 3, characterized in that, The method for calculating posterior beliefs is as follows: in, Indicates the source of information. , This represents a rational Bayesian source. This indicates the aggregator's private anchor source. Indicates the common anchor point information source of the power grid. Indicates the source of information In the Dynamic weights for each scheduling cycle, Indicates the source of information In the The beliefs for each scheduling cycle are dynamically weighted based on the strength of trust and the valence of evidence.

7. A virtual power plant response scheduling device based on dynamic Bayesian game incentives, characterized in that, Based on the method described in claim 1, it includes: The game framework construction module is used to build a three-layer game framework with the power grid as the leader and aggregators and users as followers. The first objective function construction module is used to construct the first objective function of the power grid based on the current social value of the power grid and the maximum expected discounted social value that the power grid can achieve based on the current system state, with the goal of maximizing the total expected profit of the power grid. The second objective function construction module is used to construct the aggregator's second objective function based on the aggregator's current profit and the maximum expected discounted social value that the aggregator can achieve based on the current system state, with the goal of maximizing the aggregator's total expected profit. The third objective function construction module is used to construct the user's third objective function with the goal of maximizing the prospect theoretical value, based on posterior beliefs, the user's private type parameters, the unit load reduction compensation price provided by the aggregator to the user, and the additional service value of the aggregator. The posterior beliefs are obtained by Softmax weighted fusion of rational Bayesian beliefs, private anchor beliefs about the aggregator, and public anchor beliefs about the power grid. The game model construction module is used to construct game models based on a three-layer game framework, a first objective function, a second objective function, and a third objective function. The model solving module is used to solve the game model using backward induction to obtain the virtual power plant response scheduling results.

8. A computer-readable storage medium, characterized in that, It stores a computer program that, when executed by a processor, implements the virtual power plant response scheduling method based on dynamic Bayesian game incentives as described in any one of claims 1 to 6.

9. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the virtual power plant response scheduling method based on dynamic Bayesian game incentives as described in any one of claims 1 to 6.

Images