Bayesian optimization-based virtual power plant multi-cell model participation spot market operation method and system

Abstract

The invention discloses a Bayesian optimization-based method and system for virtual power plants to participate in spot market operation. The method comprises the steps of S1, constructing multi-dimensional multi-cell tuple information reflecting parameter characteristics of different virtual power plants participating in spot market operation; s2, proposing a multi-cell space-time associated virtual power plant operation performance evaluation index, and establishing a bidding model of the virtual power plant by taking optimization of the operation performance of the virtual power plant and maximization of an economic benefit index as targets; s3, establishing an operation model of all bidding power plants including the virtual power plant participating in spot market transaction; and S4, calculating the clearing transaction price and clearing transaction electric quantity of the virtual power plant by adopting a Bayesian optimization method, and executing the power operation strategy on the virtual power plant so as to realize flexible adjustment of the power generation and utilization of the virtual power plant, give full play to the efficiency of the virtual power plant participating in the operation of the novel power system, and significantly improve the operation benefits of the power market.

Description

Technical Field

[0001] The present invention relates to the technical field of power grid dispatching automation, and in particular to a method and system for a virtual power plant polyhedral model to participate in spot market operation based on Bayesian optimization. Background Art

[0002] In order to support the realization of the "dual carbon" goals and the construction of a new power system, the main players in the power market have gradually diversified, and their characteristics and market positioning have also changed accordingly. Due to the large differences in output characteristics, operating characteristics, and applicable control methods among various virtual power plants composed of distributed resources such as different types of flexible storage loads, which are also far from the load and power supply characteristics of traditional power systems, it is difficult to directly apply the power market mechanism based on traditional power sources and loads, which greatly affects the flexible regulation of virtual power plants in the power system. In summary, it is necessary to deeply and accurately analyze the operation mode of the power spot market suitable for the participation of virtual power plants, fully stimulate the flexible regulation and control support capabilities of virtual power plants, and thus promote and guarantee the construction and development of new power systems.

[0003] At present, although the distributed resources in virtual power plants have the potential for regulation and support, the power response characteristics after the aggregation of distributed resources are not accurate, and the market incentive analysis of distributed resources participating in the response of virtual power plants is not clear enough. As a result, the regulation potential of virtual power plants participating in dispatching operations is insufficient, and the regulation and support role of virtual power plants in improving new power systems cannot be fully utilized. Therefore, it is urgent to deeply explore the power regulation capabilities of virtual power plants composed of different distributed resources, introduce bidding strategies for virtual power plants to participate in the spot market, and improve the market operation methods for virtual power plants to flexibly support the operation of new power systems.

[0004] In order to ensure that different distributed resources can provide electricity in real time and efficiently as a flexible and adjustable virtual power plant when participating in the operation of the virtual power plant, it is necessary to describe the influencing factors and action behaviors of different distributed resources in the virtual power plant in detail. Therefore, it is proposed to use multidimensional multicellular modeling technology to establish key feature information of different distributed resources to participate in power dispatch. For different virtual power plant participation market bidding strategies, the Bayesian optimization learning algorithm can be used to solve and improve the effectiveness of flexible dispatch strategies for virtual power plants. Summary of the invention

[0005] In order to solve the above-mentioned deficiencies in the prior art, the present invention provides a method and system for virtual power plants to participate in the spot market operation based on Bayesian-polytopic optimization, including:

[0006] Construct multidimensional multicellular group information reflecting the operating parameter characteristics of virtual power plants composed of different flexible load-storage distributed resources, establish the power boundary area of ​​virtual power plants under the influence of different power generation and consumption behaviors, divide the boundary area according to the range of different influencing factors, and establish a multidimensional multicellular boundary model for the operation of virtual power plants to determine the aggregated power of virtual power plants.

[0007] The power changes of the virtual power plant polytope after the distributed resource group is aggregated in different ways under different influencing factors are characterized, and the LP norm is proposed to measure the power changes of the polytope. The information calculation of the virtual power plant multidimensional polytope tuples established satisfies the closure and linear transformation of the Minkowski sum. Virtual power plant polytope The LP norm of is expressed as:

[0008]

[0009] Among them, p is a coefficient reflecting the Minkowski distance.

[0010] The construction of different virtual power plant bidding models includes:

[0011] According to the power regulation performance index requirements of different virtual power plants after the aggregation of distributed resources, the LP norm is reasonably selected to determine the maximum power-price benefit L of different virtual power plant polytopes participating in the bidding. p :

[0012]

[0013] Among them, C z is the price quote vector of each distributed resource, c z,i is the quotation coefficient of the i-th dimension distributed resource.

[0014] The optimization objective function F is constructed to maximize the bidding revenue of multiple virtual power plants and minimize the power generation and consumption price of the system. t :

[0015]

[0016] Where k represents the kth virtual power plant.

[0017] Establishing a bidding constraint equation group based on power plant parameters for the objective function;

[0018] Among them, the power plant parameters include: the regulation capacity constraint of the virtual power plant, the regulation accuracy constraint, the maximum output limit, the minimum output limit and the rising output speed limit of each bidding power plant.

[0019] Preferably, the use of an improved Bayesian optimization learning method to calculate the bidding model to obtain the transaction power includes:

[0020] Bayesian optimization initializes the Bayesian network model by setting the optimization objective function F, and defines two parameters to be optimized ξ1, ξ2 with the goal of maximizing the bidding revenue of multiple virtual power plants and minimizing the system power generation and consumption price. Through multiple iterations, the objective function is evaluated, the optimal parameter combination is selected each time, and the model is updated according to the evaluation results, gradually approaching the optimal solution, and finally determining the optimal parameter combination to maximize the objective function F.

[0021] In the Bayesian optimization calculation process, firstly, the distributed resource related data, including characteristic information such as historical load, ambient temperature and user usage habits, are collected, and these data are cleaned and normalized. Then, key features such as historical load, ambient temperature and time are extracted from the preprocessed data, and feature dimension reduction is performed using principal component analysis to retain the main characteristic information. When establishing the Bayesian optimization model, a suitable Gaussian process prior model is selected. By setting the initial point and evaluating the objective function value, the initial Gaussian process proxy model is constructed. These initial points are selected by random selection or space filling design method. Then, the acquisition function is set and optimized, and the confidence bound acquisition function is used to determine the next evaluation point. During the optimization process, the standard evolutionary optimization algorithm is used to select the optimal sampling position. Whenever a new evaluation point is selected and the objective function is evaluated, the parameters of the Gaussian process model are updated using the Bayesian theorem to obtain the updated proxy model. By repeatedly optimizing the acquisition function and updating the Gaussian process model until the preset convergence condition is reached (number of iterations, budget limit, or acquisition function improvement is less than a threshold), the point with the best objective function value among all evaluation points is selected as the optimal feasible solution. The parameter combination corresponding to the final optimal solution is the optimal solution found by Bayesian optimization.

[0022] Preferably, after obtaining the traded electricity, the clearing results and bidding strategies of each power plant in the spot market are obtained.

[0023] The present invention also provides a method and system for a virtual power plant multi-cell model to participate in the spot market operation based on Bayesian optimization.

[0024] include:

[0025] An optimization module is used to bring the bidding power and electricity prices of all bidding power plants including virtual power plants into the transaction bidding model, and use the improved Bayesian optimization learning algorithm to calculate the transaction bidding model to obtain the transaction power and transaction price;

[0026] Preferably, the transaction operation optimization module is specifically used to:

[0027] Determine whether the current basic feasible solution is the optimal solution. If the basic feasible solution is the optimal solution, obtain the current basic feasible solution; otherwise, select the variable entering the basis and the variable leaving the basis;

[0028] The calculation ends when the Bayesian optimization learning algorithm finds the optimal solution.

[0029] The technical solution provided by the present invention has the following beneficial effects:

[0030] The technical solution provided by the present invention brings the bidding information of all bidding power plants including virtual power plants into the spot market transaction bidding model, uses the improved Bayesian optimization algorithm to calculate the bidding model, and obtains the transaction electricity and transaction price. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 The transaction price of the virtual power plant in the present invention after participating in the spot trading market;

[0032] Figure 2 It is the winning transaction electricity of the units containing the virtual power plant in the present invention. DETAILED DESCRIPTION

[0033] In order to better understand the present invention, the specific implementation content of the present invention is further described below in conjunction with the accompanying drawings and examples.

[0034] Extract the computational analysis of the operating parameter characteristics of virtual power plants that reflect different flexible load-storage distributed resource compositions, establish a virtual power plant that reflects the multidimensional multi-cellular group information of different distributed resources, establish the power boundary area of ​​the virtual power plant under the influence of different power generation and consumption behaviors, divide the boundary area according to the range of different influencing factors, and establish a multidimensional multi-cellular boundary model of the virtual power plant operation to determine the power regulation capability of the virtual power plant.

[0035] Virtual power plants integrate distributed resources such as adjustable loads and energy storage distributed in different locations and effectively participate in spot market transactions. In the process of virtual power plants participating in the spot market, regulation capacity becomes an important variable affecting their revenue settlement. Based on the main characteristics of virtual power plants composed of different distributed resources such as air conditioners, electric vehicles and energy storage, an aggregate power calculation model is established.

[0036] (1) When calculating the aggregate power of the air conditioning load group, the power of N air conditioners is accumulated, and the aggregate power of the air conditioning load group at time t is for:

[0037]

[0038] In the formula, N is the number of air conditioners. is the rated power of the i-th air conditioner at time t (kW), is the aggregate power of the air conditioner group at time t (kW), s i(t) is the operating state variable of the i-th air conditioner (start is 1, stop is 0). The operating state change of the air conditioner is affected by characteristic parameters such as indoor temperature and set temperature.

[0039] (2) When calculating the adjustable power of an electric vehicle group, the aggregate power of charging / discharging of N electric vehicles for:

[0040]

[0041] In the formula, N is the number of electric vehicles aggregated. is the rated power of the i-th electric vehicle at time t (kW), r i (t) is the charge and discharge state variable of the i-th electric vehicle (charging is 1, stopping is 0, and discharging is -1). The charge and discharge state of the electric vehicle is affected by characteristic parameters such as battery capacity, guaranteed power, expected power, current SOC state, and user travel plan.

[0042] (3) When calculating the aggregate adjustable power of the energy storage group, the aggregate power of N energy storage resources charged and discharged is for:

[0043]

[0044] In the formula, N is the number of energy storage aggregation units, is the rated power of the i-th energy storage unit at time t (kW), k i (t) is the charge and discharge state variable of the i-th energy storage (charging is 1, stopping is 0, and discharging is -1). The charge and discharge state variables of energy storage are affected by characteristic parameters such as battery capacity, basic power, expected power, current SOC state, discharge depth, and discharge times.

[0045] To this end, different air conditioners, electric vehicles, and energy storage groups in a virtual power plant are defined as different dimensions, and the state variables and characteristic parameters in different groups are selected as polytope information of different orders, and a virtual power plant is defined as an m-order polytope in n-dimensional space. satisfy:

[0046]

[0047] Among them, q z ∈R m is the reference coordinate position of each order polyhedron of the virtual power plant, G z Generate a matrix for the polytope of the virtual power plant. The composition of different distributed resources determines the shape change of the polytope after aggregation. It is the r-order characteristic parameter matrix, which is affected by the characteristic parameters and determines the aggregation of m-order distributed resources.

[0048] In order to characterize the power change of the virtual power plant polytope after the distributed resource group is aggregated in different ways under different influencing factors, it is proposed to construct the LP norm to measure the power change of the polytope. The information calculation of the virtual power plant multidimensional polytope tuples established satisfies the closedness and linear transformation of the Minkowski sum. Virtual power plant polytope The LP norm of is expressed as:

[0049]

[0050] Among them, p is a coefficient reflecting the Minkowski distance.

[0051] This paper proposes to reasonably select the LP norm according to the power regulation performance index requirements of the virtual power plant after the aggregation of distributed resources. To this end, the maximum power-price benefit L of the virtual power plant polyhedron participating in the bidding can be calculated p :

[0052]

[0053] Among them, C z is the price quote vector of each distributed resource, c z,i is the quotation coefficient of the i-th dimension distributed resource.

[0054] By representing the polyhedral space of the feasible set of distributed resource power in the virtual power plant at time t, the regulation performance index of the virtual power plant at time t+1 can be calculated and updated. The regulation performance indexes in the polyhedral model of the virtual power plant at time t include:

[0055] 1) Maximum adjustable capacity (kW):

[0056]

[0057] in, is the average value of the baseline load power of the virtual power plant before regulation during the regulation period (kW), It is the average aggregate power (kW) of the virtual power plant after being adjusted in the maximum adjustment mode during the adjustment period.

[0058] 2) Adjustable duration is the duration of power regulation of the virtual power plant multi-cell model (min).

[0059] 3) Response time t VPP (min):

[0060] t VPP =t first -t start

[0061] Among them, tfirst is the time (min) when the virtual power plant polyhedron power reaches the regulation target value for the first time, t start is the adjustment start time (min).

[0062] 4) Adjustment rate θ VPP (kW / min):

[0063] θ VPP =ΔP VPP / t VPP

[0064] In the formula, ΔP VPP is the power regulation of the virtual power plant polyhedron (kW), t VPP is the response time (min).

[0065] Finally, by maximizing the power-price benefit and regulation performance index evaluation at time t, the virtual power plant can compare and select the appropriate LP norm at time t+1.

[0066] Virtual power plants declare trading volume and price based on maximizing power-price benefits, and the spot market determines the bidding trading model for all power plants including virtual power plants. Under the condition of meeting the security constraints of the power system, the power generation and consumption price of the system is minimized as the optimization goal, and the output plan of all bidding units in different time periods is calculated and compiled.

[0067] Objective function:

[0068]

[0069] Where P i,t is the declared power of all bidding power plants, C i,t is the bid price of all bidding power plants, N is the number of all bidding power plants, and T is the time period length.

[0070] The improved Bayesian optimization learning method is used to calculate the bidding models of all power plants including virtual power plants to obtain the trading electricity and trading price of each power plant.

[0071] Construct an optimization function F to maximize the bidding revenue of multiple virtual power plants and minimize the power generation and consumption price of the system:

[0072]

[0073] Where k represents the kth virtual power plant.

[0074] The solution process of the Bayesian optimization learning method is improved to maximize the bidding revenue of multiple virtual power plants and minimize the system power generation and consumption price. Two parameters to be optimized are defined, namely the system dispatch cost weight ξ1 and the virtual power plant revenue weight ξ2.

[0075] The Gaussian Naive Bayes model is used to calculate the conditional probability:

[0076]

[0077] Among them, σ i,j represents the variance of the jth feature attribute in the i-th class label, μ i,j Represents the mean of the jth feature attribute in the i-th category label.

[0078] Based on information gain, weights are assigned to different attributes to adjust the contribution of each attribute in flexibility assessment. The weight calculation formula is:

[0079]

[0080] Then according to the existing model F t Predict the distribution of the target function and get the mean μ t (ξ 1,t ,ξ 2,t ) and variance And select the next evaluation parameter (ξ 1,t ,ξ 2,t ):

[0081]

[0082] Among them, α(ξ 1,t ,ξ 2,t |F t ) is the confidence bound acquisition function, expressed as:

[0083] Evaluate the objective function F(ξ 1,t ,ξ 2,t ) and update the dataset, and update the model F with the new dataset t This iteration is repeated until the maximum number of iterations is reached or the objective function value converges, and finally the current optimal parameter combination (ξ1 * ,ξ2 * ) and the optimal objective function.

[0084] The specific Bayesian optimization process is as follows:

[0085] (1) Initialization

[0086] Select some initial points and evaluate the value of the objective function. These initial points are selected randomly or using a space-filling design method.

[0087] (2) Selecting priors and building a surrogate model

[0088] A surrogate model of the objective function is established based on the prior assumption of the Gaussian process. This model provides an estimate of the objective function and the uncertainty of the corresponding prediction.

[0089] (3) Determine the acquisition function

[0090] The acquisition function is key in Bayesian optimization. It determines the location of the next sampling point based on the current proxy model to balance exploring unknown areas and utilizing known information. Common acquisition functions include expected improvement, maximum probability improvement, knowledge gradient, and confidence bound acquisition functions.

[0091] (4) Optimize the acquisition function

[0092] A standard evolutionary optimization algorithm is used to optimize the acquisition function to determine the next point at which to evaluate the objective function.

[0093] (5) Update the proxy model

[0094] The function is evaluated at the new sampling point, and then this data point is used to update the Gaussian process model, that is, to update the prior information.

[0095] (6) Iteration

[0096] Repeat steps (3)-(5) until convergence conditions are met, such as the number of iterations, budget constraints, or the improvement of the acquisition function is less than a certain threshold.

[0097] (7) Output the optimal solution

[0098] The optimal solution ultimately produced by Bayesian optimization is the point where the function value is the largest among all evaluation points.

[0099] The pseudo code for Bayesian optimization is as follows:

[0100]

[0101] According to the Bayesian optimization process, a specific Bayesian network model is established. The model construction process is as follows:

[0102] In the Bayesian optimization calculation process, firstly, the distributed resource related data, including characteristic information such as historical load, ambient temperature and user usage habits, are collected, and these data are cleaned and normalized. Then, key features such as historical load, ambient temperature and time are extracted from the preprocessed data, and feature dimension reduction is performed using principal component analysis to retain the main characteristic information. When establishing the Bayesian optimization model, the Gaussian naive Bayes prior model is selected. By setting the initial point and evaluating the objective function value, the initial Gaussian process proxy model is constructed. These initial points are selected by random selection or space filling design method. Then, the confidence bound acquisition function is set and optimized to determine the next evaluation point. During the optimization process, the standard evolutionary optimization algorithm is used to select the optimal sampling position. Whenever a new evaluation point is selected and the objective function is evaluated, the parameters of the Gaussian process model are updated using the Bayesian theorem to obtain the updated proxy model. By repeatedly optimizing the acquisition function and updating the Gaussian process model until the preset convergence condition is reached (number of iterations, budget limit, or acquisition function improvement is less than a threshold), the point with the best objective function value among all evaluation points is selected as the optimal feasible solution. The parameter combination corresponding to the final optimal solution is the optimal solution found by Bayesian optimization.

[0103] The entire calculation example of the embodiment of the present invention is built on the IEEE-39 node system. The entire system includes 8 thermal power plants G1-G8 and 4 virtual power plants VPP1-VPP4, allowing the entire system to declare, clear, price, and settle, so as to achieve the scenario of the source-grid-load-storage integrated virtual power plant participating in the on-site spot market. The transaction information declared by the 8 thermal power units mainly includes: unit startup and shutdown costs (yuan / time), electric energy costs (yuan / MWh), time-divided operation upper and lower limits, time-divided ramp rate, minimum continuous operation time, minimum continuous shutdown time, etc. As a combination of power generation and power consumption, the "source-grid-load-storage integrated" virtual power plant participates in the market in the form of "quantity quotation", and independently decides to declare the quantity-price curve under load state and the quantity-price curve under power generation state, as well as the upper limit of power generation operation and the upper limit of load operation power consumption, as the basis for clearing in the spot market. Now based on the MATLAB platform, the Gurobi solver is called for solution.

[0104] The virtual power plant can provide 110MW of power generation during peak hours to meet the high demand of the power grid; during off-peak hours, through the charging of the energy storage system and the response of the adjustable load, the virtual power plant can increase the power load by 110MW, balance the power grid load, and achieve peak shaving and valley filling. The energy storage system can store electricity during off-peak hours for use during peak hours. The 110MW power setting is mainly concentrated on industrial and commercial users. By reasonably adjusting the power consumption time, the optimal configuration of the load is achieved. Through the above reasonable quotations and power settings, the virtual power plant can not only optimize the economic benefits in the power market, but also technically guarantee flexibility and stability, provide stable power support and balance the supply and demand of the power grid.

[0105] The winning bid clearance volumes of thermal power units and virtual power plants are shown in the following table.

[0106] Table 1 Winning bids for each unit (MW)

[0107] Time Thermal Power VPP1 VPP2 VPP3 VPP4 00:00-06:00 0 1703 2203.5 2529 2538 06:00-12:00 725 2425 2469 2588 2640 12:00-18:00 2091 2640 2640 2640 2640 18:00-24:00 2969 2640 2640 2599 2640

[0108] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented in one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that include computer-usable program code.

[0109] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0110] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0111] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process in the computer or other programmable device. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0112] The above are merely embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention are included in the scope of the claims of the present invention to be approved.

Claims

1. A method and system for participating in the spot market operation of a virtual power plant multi-cell model based on Bayesian optimization, characterized in that: include: Analyze the parameter requirements of different virtual power plants participating in the spot market operation, construct multi-dimensional and multi-cell group information reflecting the operating parameter characteristics of different virtual power plants, establish a bidding model aimed at optimizing the operating performance and economic benefits of virtual power plants, and promote trading entities to maximize returns through optimization strategies; include all bidding power plants including virtual power plants in the operating model of spot market transactions, propose the use of improved Bayesian optimization method to calculate the transaction price and transaction electricity of virtual power plants, obtain the transaction electricity and transaction price of all power plants, and execute the power operation strategy for virtual power plants to achieve flexible adjustment of the power generation and consumption mode of the virtual power plants.

2. The method according to claim 1, characterized in that The construction of the virtual power plant operation model includes: Construct multidimensional polyhedral group information reflecting the characteristics of virtual power plant operation parameters composed of distributed resources such as different flexible storage loads, establish the power boundary area of ​​virtual power plants under the influence of different power generation and consumption behaviors, divide the boundary area according to the range of different influencing factors, and establish a multidimensional polyhedral boundary model for virtual power plant operation to determine the power of virtual power plants after different distributed resources are aggregated. Set different air conditioners, electric vehicles, and energy storage groups in a virtual power plant to be defined as different dimensions, select state variables and characteristic parameters in different groups as polyhedral group information of different orders, and define a virtual power plant as an m-order polyhedron in n-dimensional space. (m≤M), satisfying: Among them, q z ∈R m is the reference coordinate position of each order polyhedron of the virtual power plant, G z Generate a matrix for the polytope of the virtual power plant. The composition of different distributed resources determines the shape change of the polytope after aggregation. It is the r-order characteristic parameter matrix, which is affected by the characteristic parameters and determines the aggregation of m-order distributed resources.

3. The method according to claim 2, characterized in that The bidding objective function of the virtual power plant is as follows: According to the constructed multidimensional polytope group information of virtual power plants, the LP norm is constructed to characterize the power regulation size of different virtual power plant polytope characteristics under the influence of uncertain factors. Based on the uncertainty gap theory and reachability objectives, an objective function reflecting the performance indicators and economic benefits of different virtual power plants is proposed as the bidding objective function of virtual power plants. In order to characterize the power change of the virtual power plant polytope after the distributed resource group is aggregated in different ways under different influencing factors, it is proposed to construct the LP norm to measure the power change of the polytope. The information calculation of the virtual power plant multidimensional polytope tuples established satisfies the closedness and linear transformation of the Minkowski sum. Virtual power plant polytope The LP norm of is expressed as: Among them, p is a coefficient reflecting the Minkowski distance. It is proposed to reasonably select the LP norm according to the power regulation performance index requirements of the virtual power plant after the aggregation of distributed resources. To this end, the maximum power-price benefit L of the virtual power plant polytope participating in the bidding can be calculated p : Among them, C z is the price quote vector of each distributed resource, c z,i is the quotation coefficient of the i-th dimension distributed resource.

4. The method according to claim 2, characterized in that The Bayesian optimization learning method is used to calculate the bidding models of all power plants including virtual power plants to obtain the transaction power of each power plant, including: Substitute the bidding objective function and constraint equations into the improved Bayesian optimization learning algorithm to obtain the bidding results of all power plants including virtual power plants; Construct an optimization objective function F to maximize the bidding revenue of multiple virtual power plants and minimize the power generation and consumption price of the system: Where k represents the kth virtual power plant. In the bidding constraint adjustment based on power plant parameters, the power plant parameter constraints include: total power adjustment capability constraint, maximum power constraint, minimum power constraint, adjustment time constraint, adjustment rate constraint, adjustment accuracy constraint of each virtual power plant, maximum output constraint, minimum output constraint and rising output speed constraint of the bidding power plant.

5. The method and system according to claim 4, characterized in that: Using the method described above, the Bayesian optimization method is used to optimize the above optimization objective function F. Bayesian optimization initializes the Bayesian network model by setting the optimization objective function F, and defines two parameters to be optimized ξ with the goal of maximizing the bidding revenue of multiple virtual power plants and minimizing the system power generation and consumption price. 1,t ,ξ 2,t Through multiple iterations, the objective function is evaluated, the optimal parameter combination is selected each time, and the model is updated according to the evaluation results, gradually approaching the optimal solution, and finally determining the optimal parameter combination to maximize the objective function F. The Bayesian optimization iterative process is: First, the Gaussian Naive Bayes model is used to calculate the conditional probability; the calculation formula is: Among them, σ i,j represents the variance of the jth feature attribute in the i-th class label, μ i,j Represents the mean of the jth feature attribute in the i-th category label. Then, weights are assigned to different feature attributes based on information gain to adjust the contribution of each feature attribute in flexibility evaluation. The formula for calculating the feature attribute weight is as follows: Then according to the existing model F t Predict the distribution of the target function and get the mean μ t (ξ 1,t ,ξ 2,t ) and variance And select the next evaluation parameter (ξ 1,t ,ξ 2,t ): Among them, α(ξ 1,t ,ξ 2,t |F t ) is the confidence bound acquisition function, expressed as: Finally, the objective function F(ξ 1,t ,ξ 2,t ) and update the dataset, and update the model F with the new dataset t This iteration is repeated until the maximum number of iterations is reached or the objective function value converges, and finally the current optimal parameter combination (ξ1 * ,ξ2 * ) and the optimal objective function. The Bayesian optimization learning algorithm is used to solve and determine whether the current feasible solution is the optimal feasible solution. When the basic feasible solution is the optimal feasible solution, the current basic feasible solution is obtained; and the basic feasible solution is continued to be determined to be the optimal solution, and the calculation is terminated when the basic feasible solution is determined to be the optimal feasible solution.

Images