Virtual power plant scheduling optimization method and system based on schedulable space prediction

Abstract

The invention discloses a virtual power plant scheduling optimization method and system based on schedulable space prediction. The method comprises the following steps: constructing a target virtual power plant schedulable space prediction model based on Gaussian process regression; respectively constructing a virtual power plant bidding model and a coal-fired power plant bidding model based on net income maximization; constructing a day-ahead electricity market clearing model based on market energy consumption cost minimization; constructing a double-layer game bidding model based on the virtual power plant bidding model, the coal-fired power plant bidding model and the day-ahead electricity market clearing model; iteratively solving the double-layer game bidding model at least based on the schedulable space of the virtual power plant until a convergence condition is met, and outputting an equilibrium solution; and adjusting the wind and light output strategies of the virtual power plant in different power demand periods based on the equilibrium solution. Theoretical support is provided for making a bidding strategy for the virtual power plant to participate in the day-ahead electricity market transaction, supply and demand balance of an electric power system can be maintained, and the stability of the system is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of power grids, and in particular to a virtual power plant scheduling optimization method and system based on dispatchable space prediction. Background Art

[0002] Virtual power plants (VPPs) have been widely used as an effective way to aggregate and manage multiple types of renewable energy and energy storage. Virtual power plants have also gradually become an important trading entity in the electricity spot trading market. Therefore, a fair and scientific bidding strategy is an important guarantee for promoting the economic and efficient operation of virtual power plants.

[0003] The existing bidding technology for virtual power plants fails to fully consider the interest game between virtual power plants and coal-fired power plants and other power generators in the process of participating in the electricity spot trading market, and the existing game model solution method is inefficient and takes a long time to solve. Therefore, it is necessary to select reasonable and advanced game optimization methods to study the game bidding technology for virtual power plants to participate in the day-ahead electricity market in order to increase the operating income of virtual power plants.

[0004] In order to maximize the benefits of virtual power plants, many methods have proposed virtual power plant bidding technologies, but they have not studied how to balance the two dimensions of social welfare maximization and virtual power plant profit maximization when virtual power plants participate in market transactions. Therefore, it is urgent to optimize the bidding strategy of virtual power plants with a market-oriented approach and the goal of increasing social welfare.

[0005] The traditional virtual power plant dispatchable space prediction model only fits or predicts one object separately, and does not take into account the dispatchable space prediction of multiple types of flexible resources within the virtual power plant, which makes the prediction accuracy of the dispatchable space of the virtual power plant low, thereby reducing the rationality of the bidding strategy. Moreover, the change process of the dispatchable space of the virtual power plant is a typical time-varying, dynamic and nonlinear process. Therefore, it is urgent to select a suitable method to process and predict the time series. Summary of the invention

[0006] The purpose of the present invention is to formulate a reasonable bidding strategy for virtual power plants, and then adjust the output strategy of virtual power plants in different power demand periods, so as to help the power system maintain supply and demand balance and improve the stability of the system. In order to achieve the above purpose, the present invention provides a virtual power plant scheduling optimization method and system based on dispatchable space prediction.

[0007] In a first aspect, an embodiment of the present invention provides a virtual power plant scheduling optimization method based on dispatchable space prediction, comprising:

[0008] A target virtual power plant dispatchable space prediction model is constructed based on Gaussian process regression; wherein the target virtual power plant dispatchable space prediction model is configured to take any moment in the future day as input and the virtual power plant dispatchable space as output, and the virtual power plant dispatchable space includes the maximum predicted wind power output and the maximum predicted photovoltaic output at the corresponding moment in the virtual power plant;

[0009] Based on the maximization of net revenue, a virtual power plant bidding model and a coal-fired power plant bidding model are constructed respectively;

[0010] Based on minimizing the market energy cost, a day-ahead electricity market clearing model is constructed;

[0011] Based on the virtual power plant bidding model, the coal-fired power plant bidding model and the day-ahead power market clearing model, a two-layer game bidding model is constructed; wherein the two-layer game bidding model includes an upper model and a lower model, the upper model includes the virtual power plant bidding model and the coal-fired power plant bidding model that forms a non-cooperative game relationship with the virtual power plant bidding model, the lower model includes the day-ahead power market clearing model, the upper model is configured to pass the declared electricity price and the declared electricity quantity to the lower model, and the lower model is configured to pass the winning electricity quantity to the upper model;

[0012] The two-layer game bidding model is iteratively solved at least based on the dispatchable space of the virtual power plant until the convergence condition is met, and an equilibrium solution is output; wherein the equilibrium solution includes the declared electricity price and declared electricity quantity of the virtual power plant, the declared electricity price and declared electricity quantity of the coal-fired power plant, and the clearing electricity price and clearing electricity quantity of the day-ahead power market;

[0013] Based on the balancing solution, the wind and solar power output strategy of the virtual power plant is adjusted during different power demand periods.

[0014] Preferably, the constructing of the target virtual power plant dispatchable space prediction model based on Gaussian process regression includes:

[0015] Construct an initial virtual power plant dispatchable space prediction model based on Gaussian process regression;

[0016] Obtain wind and solar power output data at each moment in the virtual power plant within a preset historical time range to form a data set;

[0017] The initial virtual power plant dispatchable space prediction model is trained according to the data set to obtain a target virtual power plant dispatchable space prediction model.

[0018] Preferably, the virtual power plant bidding model and the coal-fired power plant bidding model are constructed based on the maximization of net revenue, respectively, including:

[0019] Constructing a first objective function based on maximizing the net operating income of the virtual power plant, and constructing a virtual power plant bidding model according to the first objective function;

[0020] A second objective function is constructed based on maximizing the net operating profit of the coal-fired power plant, and a bidding model for the coal-fired power plant is constructed according to the second objective function.

[0021] Preferably, a first preset formula is used to characterize the first objective function; wherein the first preset formula is expressed as follows:

[0022] maxF VPP,i =M e,i -C oper,i

[0023] Among them, F VPP，i represents the net operating income of virtual power plant i, M e,i represents the revenue of virtual power plant i participating in the day-ahead electricity market, C oper,i represents the operating cost of virtual power plant i;

[0024] The step of constructing a virtual power plant bidding model according to the first objective function includes:

[0025] The first objective function is constrained by a first preset constraint condition to obtain a virtual power plant bidding model; wherein the first preset constraint condition includes supply and demand balance constraints, renewable energy power constraints and unit upper and lower limit constraints.

[0026] Preferably, a second preset formula is used to characterize the second objective function; wherein the second preset formula is expressed as follows:

[0027] maxF G,m =M e,m -C oper,m

[0028] Among them, F G,m represents the net operating income of the mth coal-fired power plant, M e,m represents the revenue of the mth coal-fired power plant participating in the day-ahead electricity market, C oper,m represents the operating cost of the mth coal-fired power plant;

[0029] The step of constructing a coal-fired power plant bidding model according to the second objective function includes:

[0030] The second objective function is constrained by a second preset constraint condition to obtain a coal-fired power plant bidding model; wherein the second preset constraint condition includes a unit output upper limit constraint.

[0031] Preferably, the method of constructing a day-ahead electricity market clearing model based on minimization of market energy cost includes:

[0032] A third objective function is constructed based on minimizing the market energy cost, and a day-ahead electricity market clearing model is constructed according to the third objective function.

[0033] Preferably, a third preset formula is used to characterize the third objective function; wherein the third preset formula is expressed as follows:

[0034]

[0035] in, represents the electricity price reported by virtual power plant i in the day-ahead power market, represents the declared power of virtual power plant i in the day-ahead power market, represents the electricity price reported by the mth coal-fired power plant in the day-ahead power market, represents the power reported by the mth coal-fired power plant in the day-ahead power market, I represents the number of virtual power plants, and M represents the number of coal-fired power plants;

[0036] The constructing of a day-ahead power market clearing model according to the third objective function comprises:

[0037] The third objective function is constrained by using a third preset constraint condition to obtain a day-ahead electricity market clearing model; wherein the third preset constraint condition includes a constraint on the upper limit of the reported electricity quantity.

[0038] Preferably, the iterative solution of the two-layer game bidding model is performed at least based on the dispatchable space of the virtual power plant until a convergence condition is met and an equilibrium solution is output, including:

[0039] Based on the dispatchable space of the virtual power plant, the unit parameters of the virtual power plant and the unit parameters of the coal-fired power plant, the PR-ADMM algorithm is used to iteratively solve the two-layer game bidding model until the convergence condition is met and the equilibrium solution is output; wherein, the convergence condition is that the dual residual reaches the convergence accuracy.

[0040] Preferably, the adjusting the wind and solar power output strategy of the virtual power plant in different power demand periods based on the balancing solution includes:

[0041] Based on the declared electricity price and declared electricity volume of the virtual power plant, the wind and solar power output of the virtual power plant during the peak period of electricity demand is increased, and the wind and solar power output of the virtual power plant during the low period of electricity demand is reduced.

[0042] In a second aspect, an embodiment of the present invention provides a virtual power plant scheduling optimization system based on dispatchable space prediction, including:

[0043] A dispatchable space prediction model construction module is used to construct a dispatchable space prediction model of a target virtual power plant based on Gaussian process regression; wherein the dispatchable space prediction model of the target virtual power plant is configured to take any moment in the future day as input and the dispatchable space of the virtual power plant as output, and the dispatchable space of the virtual power plant includes the maximum predicted wind power output and the maximum predicted photovoltaic output at the corresponding moment in the virtual power plant;

[0044] A bidding model building module, used to build a virtual power plant bidding model and a coal-fired power plant bidding model based on net profit maximization;

[0045] The clearing model construction module is used to construct a day-ahead power market clearing model based on minimization of market energy costs;

[0046] A two-layer game bidding model construction module, used to construct a two-layer game bidding model based on the virtual power plant bidding model, the coal-fired power plant bidding model and the day-ahead power market clearing model; wherein the two-layer game bidding model includes an upper model and a lower model, the upper model includes the virtual power plant bidding model and the coal-fired power plant bidding model that forms a non-cooperative game relationship with the virtual power plant bidding model, the lower model includes the day-ahead power market clearing model, the upper model is configured to pass the declared electricity price and the declared electricity quantity to the lower model, and the lower model is configured to pass the winning electricity quantity to the upper model;

[0047] An iterative solution module, used for iteratively solving the two-layer game bidding model based on at least the dispatchable space of the virtual power plant until the convergence condition is met, and outputting an equilibrium solution; wherein the equilibrium solution includes the declared electricity price and declared electricity quantity of the virtual power plant, the declared electricity price and declared electricity quantity of the coal-fired power plant, and the clearing electricity price and clearing electricity quantity of the day-ahead power market;

[0048] An output strategy adjustment module is used to adjust the wind and solar output strategy of the virtual power plant in different power demand periods based on the balanced solution.

[0049] Compared with the prior art, the virtual power plant scheduling optimization method and system based on dispatchable space prediction in the embodiment of the present invention has the following beneficial effects: combining with the Gaussian process regression model, a flexible resource dispatchable space prediction theory is proposed, which improves the flexibility and economy of power market transactions; considering the interest game problem between power generators in the market, a non-cooperative game model between virtual power plants and coal-fired power plants is constructed, and the PR-ADMM algorithm is used to obtain the game equilibrium strategy, which promotes the cost reduction and efficiency improvement of virtual power plants and the increase of social welfare; wherein, the PR-ADMM algorithm can increase the number of updates of dual variables, effectively reduce the solution time and the number of iterations, and improve the solution efficiency; in general, the present invention provides theoretical support for the formulation of bidding strategies for virtual power plants to participate in day-ahead power market transactions, which helps the power system to maintain supply and demand balance and improve the stability of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 It is a flow chart of a virtual power plant scheduling optimization method based on scheduling space prediction according to an embodiment of the present invention;

[0051] Figure 2 It is a schematic diagram of a process for constructing a dispatchable space prediction model of a target virtual power plant according to an embodiment of the present invention;

[0052] Figure 3 It is a schematic diagram of a process of constructing a virtual power plant bidding model and a coal-fired power plant bidding model respectively in an embodiment of the present invention;

[0053] Figure 4 It is a structural schematic diagram of a virtual power plant scheduling optimization system based on dispatchable space prediction according to an embodiment of the present invention. DETAILED DESCRIPTION

[0054] The specific implementation of the present invention is further described in detail below in conjunction with the accompanying drawings and examples. The following examples are used to illustrate the present invention, but are not intended to limit the scope of the present invention.

[0055] In the description of the present invention, it should be understood that the terms "first" and "second" etc. are used in the present invention to distinguish different objects rather than to describe a specific order.

[0056] like Figure 1 As shown, an embodiment of the present invention provides a virtual power plant scheduling optimization method based on dispatchable space prediction, comprising the steps of:

[0057] S1. Construct a dispatchable space prediction model of the target virtual power plant based on Gaussian process regression;

[0058] In the virtual power plant constructed by the present invention, wind power, photovoltaic and electric energy storage units have strong rapid response capabilities. By conducting research on the dispatchable space prediction of virtual power plants, a better bidding strategy can be obtained. The present invention establishes a virtual power plant dispatchable space prediction model based on Gaussian process regression, which provides a basis for virtual power plants to participate in power market bidding.

[0059] Specifically, the schedulable space prediction process is as follows:

[0060] 1) Construction of dispatchable space model of virtual power plant

[0061] The dispatchable space of the virtual power plant is jointly determined by distributed photovoltaic, distributed wind power and electric energy storage. The dispatchable space of energy storage is determined by the rated capacity of the energy storage battery. In order to ensure the continuity of the energy storage state, this embodiment sets the energy storage to 30% at the beginning and end of daily scheduling. The energy storage must meet the following constraints:

[0062]

[0063] in, and are the charging and discharging power of energy storage in the virtual power plant at time t, and are the rated charging and discharging power of energy storage, S t,e is the remaining energy storage capacity at time t, δ ess is the energy storage self-loss coefficient, η essch is the energy storage charging efficiency, η essdis is the energy storage discharge efficiency, is the rated capacity of energy storage.

[0064] During the bidding process, the output of energy storage can be flexibly adjusted according to the wind power forecast results, photovoltaic forecast results and electricity price information. Therefore, the dispatchable space C of the virtual power plant is VPP It can be expressed as:

[0065]

[0066] in, is the dispatchable space of photovoltaics, that is, the maximum predicted output of photovoltaics; It is the dispatchable space of wind power, that is, the maximum predicted output of wind power.

[0067] 2) Prediction of virtual power plant dispatchable space based on Gaussian process regression

[0068] Specifically, Figure 2 As shown, step S1 includes:

[0069] S101. Construct an initial virtual power plant dispatchable space prediction model based on Gaussian process regression;

[0070] The change process of the dispatchable space of virtual power plants is a typical time-varying, dynamic nonlinear process, and it is impossible to establish a complex model based on its composition and internal specific parameters. Gaussian process regression (GPR) is a statistical learning method based on the Bayesian principle, which is suitable for nonlinear modeling. The initial virtual power plant dispatchable space prediction model based on Gaussian process regression takes into account the influence of multiple factors such as the internal electricity price of the virtual power plant and photovoltaic uncertainty, which helps to accurately predict the overall dispatchable space of the virtual power plant.

[0071] S102, obtaining wind and solar power output data of the virtual power plant at each moment within a preset historical time range to form a data set;

[0072] This embodiment obtains the wind and solar power output data of the virtual power plant at every moment in the past 30 days to form a data set.

[0073] S103. Train the initial virtual power plant dispatchable space prediction model according to the data set to obtain the target virtual power plant dispatchable space prediction model.

[0074] Specifically, the target virtual power plant dispatchable space prediction model is configured to take any time in the future day as input and the virtual power plant dispatchable space as output. The virtual power plant dispatchable space includes the maximum predicted wind power output and the maximum predicted photovoltaic output at the corresponding time in the virtual power plant.

[0075] In a specific embodiment, the wind and solar power output for the next day is predicted based on the actual wind and solar power output data of the past 30 days. The prediction process is as follows:

[0076] Assume that the observed data in the training set is:

[0077]

[0078] Among them, t represents the time, x u represents the input variable, which is time t; X represents the set of input variables, that is, the main influencing factors affecting the dispatchable space of the virtual power plant. In the prediction model constructed by the present invention, the input variable only represents time t, with a value of 1 to 720, and time t is a positive integer representing every hour in the past 30 days; y u represents the observed output value at time t, U represents the number of training sample points, that is, the number of observation times; y represents the set of observed output values, including the observed output values ​​of wind power and photovoltaic output at each moment; C VPP,u represents the schedulable space of the u-th sample; the sample value y u Usually involves a certain degree of measurement error, usually in the training process can be added Gaussian white noise to calculate the uncertainty error of the sample. In this embodiment, the mean is set to 0 and the variance is Gaussian distributed white noise, denoted by The prior distribution of y has a mean of 0 and a covariance of The Gaussian distribution of is:

[0079]

[0080] Where GP() represents Gaussian distribution; I represents the identity matrix; K(X,X) represents the L-order kernel matrix of the elements in the training set, which represents the similarity between each sample point in the input X. It is calculated by the square exponential covariance function. The input data in the test set is represented by x * , f * Represents x * The schedulable space corresponding to the future moment.

[0081] y and f * The joint prior distribution between can be expressed as:

[0082]

[0083] Where y represents the known input x u The observed output value under f * Then the new input x * The observed output values ​​at future moments are used to predict the wind power and photovoltaic output within 24 hours in the next day. * )、K(x * ,X) and K(x * ,x * ) represent X and x respectively. * 、x * and X and x * and x * The correlation between them.

[0084] The covariance function parameters of the Gaussian process regression model define the correlation between any two points in the Gaussian process, determine the prior distribution, and then combine the historical output data of wind power and photovoltaic power generation, use Bayes' theorem to update the distribution of parameters, and obtain the posterior distribution. The posterior distribution will contain more information about the parameters and can be used for future predictions and decisions.

[0085] According to the analytical characteristics of Gaussian process regression and Bayesian principle, under the condition that y is known, the predicted value f * The posterior distribution of still obeys the multidimensional Gaussian distribution, which is expressed as:

[0086]

[0087] in, Indicates the prediction C VPP,u The mean value of cov(f* ) represents the predicted C VPP The variance of .

[0088] In order to describe the numerical features of the test set and the training set, it is necessary to measure the similarity between the test set and the training set input, determine a similarity weight and compare it with the known C VPP Data superposition is used to characterize the closeness of input and output data in different dimensions. The similarity is measured by the covariance function. The squared exponential (SE) covariance function is infinitely differentiable and can be used to process high-dimensional data. The SE covariance function is selected for training, which is expressed as:

[0089]

[0090] Among them, x r and x w represents the rth and wth input values ​​in the set U; σ f 2 and e represent the hyperparameters in the covariance function, which are the signal variance and variance scale respectively, and B is the dimension of the input time series.

[0091] Based on the above virtual power plant dispatchable space prediction model, the maximum output prediction value of distributed photovoltaic and distributed wind power at each moment in the virtual power plant in the next 24 hours can be obtained.

[0092] S2. Based on the maximization of net revenue, a virtual power plant bidding model and a coal-fired power plant bidding model are constructed respectively;

[0093] Specifically, Figure 3 As shown, step S2 includes:

[0094] S201, constructing a first objective function based on maximizing the net operating income of the virtual power plant, and constructing a virtual power plant bidding model according to the first objective function;

[0095] Taking the maximization of the net profit of virtual power plant operation as the objective function, a virtual power plant bidding model is constructed. The profit of virtual power plant is the profit of day-ahead power market transaction, and the cost of virtual power plant is mainly the operation and maintenance cost of various units and the penalty cost of renewable energy curtailment.

[0096] Specifically, the revenue of virtual power plants participating in the day-ahead electricity market is expressed by the following formula:

[0097]

[0098] The operating cost of a virtual power plant is expressed by the following formula:

[0099]

[0100] Furthermore, a first preset formula is used to characterize the first objective function. The first preset formula is expressed as follows:

[0101] maxF VPP,i =M e,i -C oper,i

[0102] Among them, M e,i , C oper,i , C om,i and C abandon,i They represent the revenue, operating cost, equipment operation and maintenance cost, and penalty cost of new energy curtailment of virtual power plant i participating in the day-ahead power market; represents the electricity price reported by virtual power plant i in the day-ahead power market, represents the declared electricity quantity in the day-ahead electricity market; P i,n,t represents the output of device n in virtual power plant i; om,i,n F represents the operation and maintenance cost coefficient of equipment n in virtual power plant i; VPP，i represents the net operating income of virtual power plant i; e,i,t P represents the penalty cost coefficient of power abandonment of virtual power plant i; abandon,i,t represents the amount of renewable energy curtailment of virtual power plant i; and They represent the output power of the photovoltaic unit and wind turbine unit in the virtual power plant i at time t respectively.

[0103] Constructing a virtual power plant bidding model according to the first objective function includes: using a first preset constraint condition to constrain the first objective function to obtain a virtual power plant bidding model. Specifically, the first preset constraint condition includes a supply and demand balance constraint, a renewable energy power constraint, and a unit upper and lower limit constraint.

[0104] The supply and demand balance constraint is expressed by the following formula:

[0105]

[0106] The renewable energy power constraint is expressed by the following formula:

[0107]

[0108] The upper and lower limit constraints of the unit are expressed by the following formula:

[0109]

[0110] Among them, P CGT,i,t Represents the output power of the photovoltaic unit, wind turbine unit and gas turbine in the virtual power plant i at time t.

[0111] S202. Construct a second objective function based on maximizing the net operating profit of the coal-fired power plant, and construct a bidding model for the coal-fired power plant based on the second objective function.

[0112] Taking the maximization of the net operating income of coal-fired power plants as the objective function, a bidding model for coal-fired power plants is constructed. The income of coal-fired power plants is the income from day-ahead power market transactions, and the cost of coal-fired power plants is mainly the operation and maintenance costs of various units.

[0113] Specifically, the revenue of coal-fired power plants participating in the day-ahead electricity market is expressed by the following formula:

[0114]

[0115] The operating cost of a coal-fired power plant is expressed using the following formula:

[0116]

[0117] Furthermore, a second preset formula is used to characterize the second objective function. The second preset formula is expressed as follows:

[0118] maxF G,m =M e,m -C oper,m

[0119] Among them, M e,m and C oper,m They represent the revenue and operating cost of coal-fired power plants participating in the day-ahead electricity market; represents the electricity price reported by the mth coal-fired power plant in the day-ahead power market, represents the power reported by the mth coal-fired power plant in the day-ahead power market; P G,m,t represents the output of the thermal power unit in the mth coal-fired power plant; a m , b m 、c m represents the coal consumption characteristic coefficient of the thermal power unit in the mth coal-fired power plant; F G,m represents the net operating profit of the mth coal-fired power plant.

[0120] Constructing a coal-fired power plant bidding model according to the second objective function includes: using a second preset constraint condition to constrain the second objective function to obtain a coal-fired power plant bidding model. Specifically, the second preset constraint condition includes a unit output upper limit constraint.

[0121] The upper limit constraint of unit output is expressed by the following formula:

[0122]

[0123] in, It represents the upper limit of the electricity reported by the mth coal-fired power plant, that is, the maximum output of the unit.

[0124] S3. Based on minimizing the market energy cost, a day-ahead electricity market clearing model is constructed;

[0125] Specifically, the third objective function is constructed based on minimizing the market energy cost, and the day-ahead power market clearing model is constructed according to the third objective function. It can be understood that minimizing the market energy cost is maximizing social welfare.

[0126] Furthermore, a third preset formula is used to characterize the third objective function. The third preset formula is expressed as follows:

[0127]

[0128] in, represents the electricity price reported by virtual power plant i in the day-ahead power market, represents the declared power of virtual power plant i in the day-ahead power market, represents the electricity price reported by the mth coal-fired power plant in the day-ahead power market, represents the power reported by the mth coal-fired power plant in the day-ahead power market, I represents the number of virtual power plants, and M represents the number of coal-fired power plants;

[0129] Constructing a day-ahead power market clearing model according to the third objective function includes: using a third preset constraint condition to constrain the third objective function to obtain a day-ahead power market clearing model. Specifically, the third preset constraint condition includes a declared power upper limit constraint.

[0130] The upper limit constraint of reported electricity consumption is expressed by the following formula:

[0131]

[0132] Among them, L e,t Represents the total load in the electricity market.

[0133] S4. Based on the virtual power plant bidding model, coal-fired power plant bidding model and day-ahead power market clearing model, a two-layer game bidding model is constructed;

[0134] The bidding optimization of virtual power plants participating in the day-ahead power market game is divided into two stages. The first stage is for i virtual power plants and m coal-fired power plants in the market to declare electricity prices and electricity. Since virtual power plants and coal-fired power plants compete in the same market, there is a conflict of interest. In other words, virtual power plants and coal-fired power plants participate in market competition in the form of non-cooperative games. Changes in the bidding strategies of virtual power plants and coal-fired power plants will affect the clearing results of the lower-level power market through decision-making interactions, thereby changing the size of the objective function value. The second stage is to optimize the clearing of the day-ahead power market and feed back the winning electricity to the virtual power plants and coal-fired power plants. The bidding units change their bidding strategies and equipment output according to the winning electricity in the day-ahead power market.

[0135] Specifically, the two-layer game bidding model includes an upper model and a lower model. The upper model includes a virtual power plant bidding model and a coal-fired power plant bidding model that forms a non-cooperative game relationship with the virtual power plant bidding model. The lower model includes a day-ahead electricity market clearing model. The upper model is configured to pass the declared electricity price and the declared electricity quantity to the lower model, and the lower model is configured to pass the winning electricity quantity to the upper model.

[0136] S5. Iteratively solve the two-layer game bidding model based on at least the dispatchable space of the virtual power plant until the convergence condition is met and output an equilibrium solution;

[0137] Specifically, based on the dispatchable space of virtual power plants, the unit parameters of virtual power plants and the unit parameters of coal-fired power plants, the PR-ADMM algorithm is used to iteratively solve the double-layer game bidding model until the convergence conditions are met and the equilibrium solution is output. Among them, the convergence condition is that the dual residual reaches the convergence accuracy, and the equilibrium solution includes the declared electricity price and declared electricity quantity of the virtual power plant, the declared electricity price and declared electricity quantity of the coal-fired power plant, and the clearing electricity price and clearing electricity quantity of the day-ahead power market.

[0138] The game bidding model of virtual power plants participating in the electricity spot market based on dispatchable space prediction is a distributed optimization problem. Considering the privacy protection issue between virtual power plants and coal-fired power plants, the alternating direction multiplier method (ADMM) can be used to continuously update and effectively limit the randomness of penalty parameters. However, the Lagrange multiplier is updated slowly, resulting in a large number of solution iterations.

[0139] In order to overcome this problem, the PR-ADMM algorithm is used to solve the bidding model of virtual power plants participating in the electricity spot market. The Peaceman Rachford splitting method based alternating direction multiplier method (PR-ADMM) is based on the PR splitting method and introduces the 1 / 2 power multiplier λ on the basis of the classic ADMM. pq,k+1 / 2 , which is essentially two iterations of the Lagrange multiplier λ. In the bidirectional interaction between agents p and q, when the information of agent p→q is transmitted forward, λ is increased to the power of 1 / 2 on the basis of the original dual variable.pq,k+1 / 2 The multiplier penalty term is used for the iterative update of the current round of reverse information transmission; when the subject q→p information is transmitted backward, a 1 / 2 power multiplier penalty term λ is added to the dual variable pq,k+1 , and use it for the next round of iterative update to increase the convergence speed. Accordingly, a pending relaxation factor α∈(0,1) is added to the penalty parameter ρ. In this embodiment, the penalty parameter ρ=1e-3 and the relaxation factor α=0.5 are set to increase the convergence speed.

[0140] This embodiment uses the Matlab2024a version of the solver Gurobi to calculate the operating model of each subject in the two stages. The programming language is Yalmip, the computer configuration is Win11 system, Intel Core i7-13700H CPU, the main frequency is 2.4GHz, and the memory is 32GB. The upper and lower layer models are interactively iterated, and the PR-ADMM algorithm is used to solve the two-layer game bidding model until the optimal solution is obtained.

[0141] For a virtual power plant, the augmented Lagrangian function of its net profit maximization subproblem can be expressed as:

[0142]

[0143] For a coal-fired power plant, the augmented Lagrangian function of the net revenue maximization subproblem can be expressed as:

[0144]

[0145] For the day-ahead electricity market, the augmented Lagrangian function of the energy cost minimization subproblem can be expressed as:

[0146]

[0147] Where p = 1, 2, ..., 5; when p = 1, 2, 3, p represents a virtual power plant, and when p = 4, 5, p represents a coal-fired power plant; q represents the day-ahead power market, q = 1; k represents the number of iterations of the PR-ADMM algorithm; F VPP,p,k 、F G,p,k and F market,k They represent the objective function values ​​of the virtual power plant, coal-fired power plant and day-ahead power market at the kth iteration respectively; pq,k and χ qp,k It represents the generalized expression of the amount of electricity reported by the virtual power plant p to the day-ahead power market q and the day-ahead power market q feedbacks the winning amount of electricity to the virtual power plant p during the k-th iteration; pq,k represents the Lagrange multiplier in the kth iteration process, representing the electricity price reported by the virtual power plant p to the day-ahead power market q; ρ pq,k Represents the penalty parameters of the three sub-problems during the k-th iteration.

[0148] During the iteration process, the update of the Lagrange multiplier follows the rules shown in the following formula:

[0149]

[0150] The PR-ADMM algorithm determines whether the algorithm iteration ends by whether the dual residual reaches the convergence accuracy, that is:

[0151]

[0152] Among them, s pq,k+1 represents the dual residual obtained at the kth iteration; ε dual Indicates the convergence accuracy of the dual residual.

[0153] For ease of understanding, the following is a brief description of the solution process of the two-layer game bidding model:

[0154] Step 1: Initialize parameters. Input data such as virtual power plant unit parameters, new energy daily forecast curve, and coal-fired power plant unit parameters; it can be understood that the new energy daily forecast curve is the dispatchable space of the virtual power plant;

[0155] Step 2: Solve the sub-problems of maximizing the net revenue of the virtual power plant, maximizing the net revenue of the coal-fired power plant, and minimizing the energy cost of the day-ahead power market. Set the initial Lagrange multiplier λ pq,1 , set the number of iterations k = 1, and solve the virtual power plant bidding model, coal-fired power plant bidding model and day-ahead power market clearing model in turn. Virtual power plants and coal-fired power plants pass the declared electricity prices and declared electricity quantities to the lower-level market clearing model. Based on the declarations of each power generator, the market updates the winning electricity quantities of each power generator and feeds back to the upper-level power generators;

[0156] Step 3: Determine the convergence conditions of the non-cooperative game problem. Determine whether the algorithm has ended iteration by whether the dual residual reaches the convergence accuracy. If the convergence conditions are met, terminate the iteration process and output the equilibrium solution; otherwise, set the number of iterations k = k + 1. Update the multiplier λ according to the Lagrange multiplier update rule pq,k , each power generator re-declares the electricity price and electricity volume, and returns to step 2 until the convergence condition is met;

[0157] Step 4: The model converges and outputs the equilibrium solution. Determine the bidding strategy for virtual power plants, the bidding strategy for coal-fired power plants, and the electricity market clearing plan.

[0158] S6. Adjust the wind and solar output strategy of the virtual power plant in different power demand periods based on the equilibrium solution.

[0159] Specifically, based on the declared electricity price and declared electricity volume of the virtual power plant, the wind and solar output of the virtual power plant during peak periods of electricity demand will be increased, and the wind and solar output of the virtual power plant during low periods of electricity demand will be reduced, so as to achieve scheduling optimization of the virtual power plant, thereby helping the power system maintain a balance between supply and demand and improving the stability of the system.

[0160] The embodiment of the present invention is a virtual power plant scheduling optimization method based on dispatchable space prediction. In combination with the Gaussian process regression model, a flexible resource dispatchable space prediction theory is proposed, which improves the flexibility and economy of power market transactions; the interest game problem between power generators in the market is considered, a non-cooperative game model between virtual power plants and coal-fired power plants is constructed, and the PR-ADMM algorithm is used to obtain the game equilibrium strategy, which promotes the cost reduction and efficiency improvement of virtual power plants and the increase of social welfare; wherein, the PR-ADMM algorithm can increase the number of updates of dual variables, effectively reduce the solution time and the number of iterations, and improve the solution efficiency; in general, the present invention provides theoretical support for the formulation of bidding strategies for virtual power plants to participate in day-ahead power market transactions, which helps the power system to maintain supply and demand balance and improve the stability of the system.

[0161] Based on the above virtual power plant dispatch optimization method, Figure 4 As shown, an embodiment of the present invention provides a virtual power plant scheduling optimization system based on dispatchable space prediction, including:

[0162] The dispatchable space prediction model construction module 1 is used to construct a dispatchable space prediction model of a target virtual power plant based on Gaussian process regression; wherein the dispatchable space prediction model of the target virtual power plant is configured to take any moment in the future day as input and the dispatchable space of the virtual power plant as output, and the dispatchable space of the virtual power plant includes the maximum predicted wind power output and the maximum predicted photovoltaic output at the corresponding moment in the virtual power plant;

[0163] A bidding model building module 2 is used to build a virtual power plant bidding model and a coal-fired power plant bidding model based on maximizing net revenue;

[0164] The clearing model building module 3 is used to build a day-ahead power market clearing model based on minimization of market energy costs;

[0165] A two-layer game bidding model construction module 4 is used to construct a two-layer game bidding model based on a virtual power plant bidding model, a coal-fired power plant bidding model and a day-ahead power market clearing model; wherein the two-layer game bidding model includes an upper model and a lower model, the upper model includes a virtual power plant bidding model and a coal-fired power plant bidding model that forms a non-cooperative game relationship with the virtual power plant bidding model, the lower model includes a day-ahead power market clearing model, the upper model is configured to pass the declared electricity price and the declared electricity quantity to the lower model, and the lower model is configured to pass the winning electricity quantity to the upper model;

[0166] Iterative solution module 5, used for iteratively solving the double-layer game bidding model based on at least the dispatchable space of the virtual power plant until the convergence condition is met, and outputting an equilibrium solution; wherein the equilibrium solution includes the declared electricity price and declared electricity quantity of the virtual power plant, the declared electricity price and declared electricity quantity of the coal-fired power plant, and the clearing electricity price and clearing electricity quantity of the day-ahead power market;

[0167] The output strategy adjustment module 6 is used to adjust the wind and solar output strategies of the virtual power plant in different power demand periods based on the equilibrium solution.

[0168] It should be noted that each module in the above-mentioned virtual power plant dispatch optimization system based on dispatchable space prediction can be fully or partially implemented by software, hardware and a combination thereof. The above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules. For the specific definition of a virtual power plant dispatch optimization system based on dispatchable space prediction, please refer to the above definition of a virtual power plant dispatch optimization method based on dispatchable space prediction. The two have the same functions and effects, which will not be repeated here.

[0169] In summary, the embodiment of the present invention is a virtual power plant scheduling optimization method and system based on dispatchable space prediction, combined with the Gaussian process regression model, and proposes a flexible resource dispatchable space prediction theory, which improves the flexibility and economy of power market transactions; considering the interest game problem between power generators in the market, a non-cooperative game model between virtual power plants and coal-fired power plants is constructed, and the PR-ADMM algorithm is used to obtain the game equilibrium strategy, which promotes the cost reduction and efficiency improvement of virtual power plants and the increase of social welfare; wherein, the PR-ADMM algorithm can increase the number of updates of dual variables, effectively reduce the solution time and the number of iterations, and improve the solution efficiency; in general, the present invention provides theoretical support for the formulation of bidding strategies for virtual power plants to participate in day-ahead power market transactions, which helps the power system to maintain supply and demand balance and improve the stability of the system.

[0170] Each embodiment in this specification is described in a progressive manner, and the same or similar parts of each embodiment can be directly referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment. It should be noted that the technical features of the above-mentioned embodiments can be combined arbitrarily. In order to make the description concise, all possible combinations of the technical features in the above-mentioned embodiments are not described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0171] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and substitutions can be made without departing from the technical principles of the present invention. These improvements and substitutions should also be regarded as the scope of protection of the present invention.

Claims

1. A virtual power plant scheduling optimization method based on dispatchable space prediction, characterized in that: include: A target virtual power plant dispatchable space prediction model is constructed based on Gaussian process regression; wherein the target virtual power plant dispatchable space prediction model is configured to take any moment in the future day as input and the virtual power plant dispatchable space as output, and the virtual power plant dispatchable space includes the maximum predicted wind power output and the maximum predicted photovoltaic output at the corresponding moment in the virtual power plant; Based on the maximization of net revenue, a virtual power plant bidding model and a coal-fired power plant bidding model are constructed respectively; Based on minimizing the market energy cost, a day-ahead electricity market clearing model is constructed; Based on the virtual power plant bidding model, the coal-fired power plant bidding model and the day-ahead power market clearing model, a two-layer game bidding model is constructed; wherein the two-layer game bidding model includes an upper model and a lower model, the upper model includes the virtual power plant bidding model and the coal-fired power plant bidding model that forms a non-cooperative game relationship with the virtual power plant bidding model, the lower model includes the day-ahead power market clearing model, the upper model is configured to pass the declared electricity price and the declared electricity quantity to the lower model, and the lower model is configured to pass the winning electricity quantity to the upper model; The two-layer game bidding model is iteratively solved at least based on the dispatchable space of the virtual power plant until the convergence condition is met, and an equilibrium solution is output; wherein the equilibrium solution includes the declared electricity price and declared electricity quantity of the virtual power plant, the declared electricity price and declared electricity quantity of the coal-fired power plant, and the clearing electricity price and clearing electricity quantity of the day-ahead power market; Based on the balancing solution, the wind and solar power output strategy of the virtual power plant is adjusted during different power demand periods.

2. The virtual power plant scheduling optimization method according to claim 1, characterized in that: The Gaussian process regression-based target virtual power plant dispatchable space prediction model is constructed, including: Construct an initial virtual power plant dispatchable space prediction model based on Gaussian process regression; Obtain wind and solar power output data at each moment in the virtual power plant within a preset historical time range to form a data set; The initial virtual power plant dispatchable space prediction model is trained according to the data set to obtain a target virtual power plant dispatchable space prediction model.

3. The virtual power plant scheduling optimization method according to claim 1, characterized in that: The virtual power plant bidding model and the coal-fired power plant bidding model are constructed based on the maximization of net income, including: Constructing a first objective function based on maximizing the net operating income of the virtual power plant, and constructing a virtual power plant bidding model according to the first objective function; A second objective function is constructed based on maximizing the net operating profit of the coal-fired power plant, and a bidding model for the coal-fired power plant is constructed according to the second objective function.

4. The virtual power plant scheduling optimization method according to claim 3 is characterized in that: The first objective function is characterized by a first preset formula; wherein the first preset formula is expressed as follows: maxF VPP,i =M e,i -C oper,i Among them, F VPP，i represents the net operating income of virtual power plant i, M e,i represents the revenue of virtual power plant i participating in the day-ahead electricity market, C oper,i represents the operating cost of virtual power plant i; The step of constructing a virtual power plant bidding model according to the first objective function includes: The first objective function is constrained by a first preset constraint condition to obtain a virtual power plant bidding model; wherein the first preset constraint condition includes supply and demand balance constraints, renewable energy power constraints and unit upper and lower limit constraints.

5. The virtual power plant scheduling optimization method according to claim 3 is characterized in that: The second objective function is characterized by a second preset formula; wherein the second preset formula is expressed as follows: maxF G,m =M e,m -C oper,m Among them, F G,m represents the net operating income of the mth coal-fired power plant, M e,m represents the revenue of the mth coal-fired power plant participating in the day-ahead electricity market, C oper,m represents the operating cost of the mth coal-fired power plant; The step of constructing a coal-fired power plant bidding model according to the second objective function includes: The second objective function is constrained by a second preset constraint condition to obtain a coal-fired power plant bidding model; wherein the second preset constraint condition includes a unit output upper limit constraint.

6. The virtual power plant scheduling optimization method according to claim 1, characterized in that: The day-ahead electricity market clearing model is constructed based on minimization of market energy cost, including: A third objective function is constructed based on minimizing the market energy cost, and a day-ahead electricity market clearing model is constructed according to the third objective function.

7. The virtual power plant scheduling optimization method according to claim 6, characterized in that: The third preset formula is used to characterize the third objective function; wherein the third preset formula is expressed as follows: in, represents the electricity price reported by virtual power plant i in the day-ahead power market, represents the declared power of virtual power plant i in the day-ahead power market, represents the electricity price reported by the mth coal-fired power plant in the day-ahead power market, represents the power reported by the mth coal-fired power plant in the day-ahead power market, I represents the number of virtual power plants, and M represents the number of coal-fired power plants; The constructing of a day-ahead power market clearing model according to the third objective function comprises: The third objective function is constrained by using a third preset constraint condition to obtain a day-ahead electricity market clearing model; wherein the third preset constraint condition includes a constraint on the upper limit of the reported electricity quantity.

8. The virtual power plant scheduling optimization method according to claim 1, characterized in that: The iteratively solving the two-layer game bidding model at least based on the dispatchable space of the virtual power plant until a convergence condition is met and an equilibrium solution is output, including: Based on the dispatchable space of the virtual power plant, the unit parameters of the virtual power plant and the unit parameters of the coal-fired power plant, the PR-ADMM algorithm is used to iteratively solve the two-layer game bidding model until the convergence condition is met and the equilibrium solution is output; wherein, the convergence condition is that the dual residual reaches the convergence accuracy.

9. The virtual power plant scheduling optimization method according to claim 1, characterized in that: The adjusting the wind and solar power output strategy of the virtual power plant in different power demand periods based on the balancing solution includes: Based on the declared electricity price and declared electricity volume of the virtual power plant, the wind and solar power output of the virtual power plant during the peak period of electricity demand is increased, and the wind and solar power output of the virtual power plant during the low period of electricity demand is reduced.

10. A virtual power plant dispatch optimization system based on dispatchable space prediction, characterized in that: include: A dispatchable space prediction model construction module is used to construct a dispatchable space prediction model of a target virtual power plant based on Gaussian process regression; wherein the dispatchable space prediction model of the target virtual power plant is configured to take any moment in the future day as input and the dispatchable space of the virtual power plant as output, and the dispatchable space of the virtual power plant includes the maximum predicted wind power output and the maximum predicted photovoltaic output at the corresponding moment in the virtual power plant; A bidding model building module, used to build a virtual power plant bidding model and a coal-fired power plant bidding model based on net profit maximization; The clearing model construction module is used to construct a day-ahead power market clearing model based on minimization of market energy costs; A two-layer game bidding model construction module, used to construct a two-layer game bidding model based on the virtual power plant bidding model, the coal-fired power plant bidding model and the day-ahead power market clearing model; wherein the two-layer game bidding model includes an upper model and a lower model, the upper model includes the virtual power plant bidding model and the coal-fired power plant bidding model that forms a non-cooperative game relationship with the virtual power plant bidding model, the lower model includes the day-ahead power market clearing model, the upper model is configured to pass the declared electricity price and the declared electricity quantity to the lower model, and the lower model is configured to pass the winning electricity quantity to the upper model; An iterative solution module, used for iteratively solving the two-layer game bidding model based on at least the dispatchable space of the virtual power plant until the convergence condition is met, and outputting an equilibrium solution; wherein the equilibrium solution includes the declared electricity price and declared electricity quantity of the virtual power plant, the declared electricity price and declared electricity quantity of the coal-fired power plant, and the clearing electricity price and clearing electricity quantity of the day-ahead power market; An output strategy adjustment module is used to adjust the wind and solar output strategy of the virtual power plant in different power demand periods based on the balanced solution.

Images